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.ppat.1002420 | Polar Flagellar Biosynthesis and a Regulator of Flagellar Number Influence Spatial Parameters of Cell Division in Campylobacter jejuni | Spatial and numerical regulation of flagellar biosynthesis results in different flagellation patterns specific for each bacterial species. Campylobacter jejuni produces amphitrichous (bipolar) flagella to result in a single flagellum at both poles. These flagella confer swimming motility and a distinctive darting motility necessary for infection of humans to cause diarrheal disease and animals to promote commensalism. In addition to flagellation, symmetrical cell division is spatially regulated so that the divisome forms near the cellular midpoint. We have identified an unprecedented system for spatially regulating cell division in C. jejuni composed by FlhG, a regulator of flagellar number in polar flagellates, and components of amphitrichous flagella. Similar to its role in other polarly-flagellated bacteria, we found that FlhG regulates flagellar biosynthesis to limit poles of C. jejuni to one flagellum. Furthermore, we discovered that FlhG negatively influences the ability of FtsZ to initiate cell division. Through analysis of specific flagellar mutants, we discovered that components of the motor and switch complex of amphitrichous flagella are required with FlhG to specifically inhibit division at poles. Without FlhG or specific motor and switch complex proteins, cell division occurs more often at polar regions to form minicells. Our findings suggest a new understanding for the biological requirement of the amphitrichous flagellation pattern in bacteria that extend beyond motility, virulence, and colonization. We propose that amphitrichous bacteria such as Campylobacter species advantageously exploit placement of flagella at both poles to spatially regulate an FlhG-dependent mechanism to inhibit polar cell division, thereby encouraging symmetrical cell division to generate the greatest number of viable offspring. Furthermore, we found that other polarly-flagellated bacteria produce FlhG proteins that influence cell division, suggesting that FlhG and polar flagella may function together in a broad range of bacteria to spatially regulate division.
| Campylobacter jejuni is a leading cause of gastroenteritis in humans and requires amphitrichous (bipolar) flagella to promote infection of hosts. This pattern of flagellation results in a single flagellum at both poles, which is characteristic of many Campylobacter species, but fairly unusual amongst other motile bacteria. In this work, we discovered an unprecedented system to spatially regulate cell division that relies on the FlhG ATPase and amphitrichous flagellar biosynthesis. In addition to its role in other polar flagellates in controlling flagellar number, we discovered that FlhG influences spatial regulation of cell division in C. jejuni. Further analysis revealed that components of the flagellar motor and switch are required with FlhG to inhibit cell division specifically at the poles of the bacterium. These findings indicate that flagella have an additional function in C. jejuni beyond promoting motility, virulence, and colonization in functioning in a mechanism with FlhG to inhibit cell division specifically at poles. Furthermore, our findings suggest that the specific amphitrichous pattern of flagellar biosynthesis in this pathogen is an essential determinant for inhibiting cell division at both bacterial poles so that symmetrical cell division occurs and generates viable progenitors.
| Due to spatial and numerical regulation of flagellar biosynthesis, bacterial species produce distinct patterns of flagellation. These regulatory mechanisms are especially evident in polarly-flagellated bacteria, which synthesize a defined number of flagella and form flagella only at bacterial poles. For many bacterial pathogens, strict spatial and numerical control of flagellar biosynthesis is essential for promoting proper motility and infection of hosts [1]–[4].
The FlhF and FlhG proteins have been implicated in spatial or numerical control of flagellar biosynthesis in polar flagellates such as Vibrio and Pseudomonas species and Campylobacter jejuni. Whereas Vibrio and Pseudomonas species each produce a monotrichous flagellum (a single flagellum only at one pole of a bacterial cell), C. jejuni and many other Campylobacter species produce amphitrichous (bipolar) flagella to result in a single flagellum at each pole. Although a defined mechanism has remained elusive, the FlhF GTPase appears to be required at an early step in flagellar biosynthesis to specifically influence formation of flagella at bacterial poles [4]–[7]. FlhG, a member of the ParA superfamily of ATPases, is involved in numerical regulation of monotrichous flagellar biosynthesis in Vibrio and Pseudomonas species, presumably by a mechanism that limits flagellar gene expression so that only one flagellum is produced per bacterial cell [1], [2].
Symmetrical cell division in bacteria also must be spatially regulated so that the divisome forms specifically at the cellular midpoint to result in two viable daughter cells of similar lengths (reviewed in [8], [9]). Without spatial regulation, the divisome may form anywhere in a bacterial cell and not always generate viable progenitors as products of cell division. Many commonly studied bacteria encode a Min system, which inhibits divisome formation at poles in Escherichia coli and Bacillus subtilis. Components of the Min system include MinD, another member of the ParA superfamily of ATPases, and MinC, the inhibitor of FtsZ polymerization into the Z-ring [10]–[15]. In E. coli, MinE is a topological specificity factor that spatially restricts MinCD complexes to primarily polar regions so that Z-ring formation is inhibited at poles [16], [17]. MinE stimulates the ATPase activity of MinD, which causes dissociation of MinCD complexes sequentially at each pole, resulting in polar oscillation of MinCD [14], [18]–[25]. As a result, the cellular midpoint remains relatively free of MinCD so that the Z-ring forms at the middle to promote symmetrical cell division. In B. subtilis, DivIVA functions as the topological specificity factor by first recruiting MinJ, which then recruits MinCD to the division site [15], [17], [26]–[28]. MinCD localizes to the divisome after a step when the Z-ring is no longer sensitive to MinC-mediated depolymerization, which likely prevents a second cell division event from occurring at the new pole of the newly formed daughter cells [15], [26], [29]. A second mechanism, termed nucleoid occlusion, also functions in E. coli and B. subtilis to inhibit Z-ring formation at the cellular midpoint [30]–[35]. Specific DNA-bound proteins inhibit Z-ring formation when the chromosomal DNA occupies the midregion of the cell. This inhibitory mechanism is relieved once chromosomal DNA is replicated and segregated to poles. Both Min and nucleoid occlusion systems may cooperatively function in many bacteria to influence formation of the divisome precisely at the midpoint at the appropriate time in a dividing cell.
The MipZ ATPase, a MinD ortholog and another ParA ATPase family member, spatially regulates Z-ring formation in Caulobacter crescentus [36]. Unlike MinD, MipZ itself directly dissociates FtsZ polymers and inhibits Z-ring formation. MipZ interacts with ParB, which is bound to DNA near the chromosomal origin of replication, and moves with the replicated chromosome as it segregates to the opposite pole before cell division. MipZ depolymerizes polar FtsZ polymers present from the last round of cell division, causing reorganization of FtsZ into the Z-ring near the midpoint. Interaction of MipZ with ParB-bound DNA spatially restricts MipZ to inhibit cell division primarily at poles.
Most Campylobacter species are amphitrichous organisms, a fairly unusual pattern of flagellation amongst polar flagellates. Flagellar motility of C. jejuni is an essential virulence and colonization factor required for infection of humans to result in diarrheal disease and many animals to promote commensalism [37]–[40]. Upon analysis of factors that regulate amphitrichous flagellar biosynthesis, we identified an unprecedented system to spatially regulate symmetrical cell division that involves FlhG, an ortholog of the MinD and MipZ ATPases, and components of amphitrichous flagella. We discovered that FlhG not only regulates flagellar number, but FlhG also influences where cell division occurs in C. jejuni. We found that deletion of flhG in C. jejuni resulted in a minicell phenotype, which is an indication of cell division occurring at polar regions rather than strictly at the cellular midpoint. Unexpectedly, mutants lacking components of the flagellar MS and C rings, which have established motor, switch, and secretory functions for the flagellum, also possessed a minicell phenotype. We propose that due to the lack of a Min system in C. jejuni, the flagellar MS ring and switch complex may serve as a topological specificity factor to modulate or localize a FlhG-dependent mechanism to inhibit cell division specifically at polar regions so that symmetrical division occurs to generate viable progenitors. Furthermore, our results demonstrate that amphitrichous flagellation in C. jejuni is not only essential for conferring motility required for infection of hosts, but also significantly influences symmetrical cell division to generate viable daughter cells. Our study also reveals that FlhG proteins of other polarly-flagellated bacteria influence placement of division sites in C. jejuni, suggesting that FlhG and polar flagellar biosynthesis may spatially influence cell division in a broad range of motile bacteria.
Members of the ParA ATPase superfamily are involved in process such as numerical regulation of flagellar biosynthesis and spatial regulation of cell division. Many polarly-flagellated bacteria appear to encode FlhG/FleN orthologs and a complete Min system including MinD (for a sequence alignment of FlhG and MinD proteins, see Figure S1). Although Min systems have not been analyzed in polar flagellates, FlhG/FleN numerically regulate flagellar biosynthesis in the monotrichous species, V. cholerae and P. aeruginosa [1], [2]. In contrast, C. crescentus produces the MipZ ATPase to spatially regulate cell division and does not appear to encode MinD or FlhG [36]. Completed genomes of all Campylobacter species encode the putative FlhG ATPase, but not MinD or any other Min proteins. Therefore, we analyzed C. jejuni 81–176 with an in-frame deletion within flhG (Cjj81176_0101) to ascertain a role for FlhG in flagellar biosynthesis and other processes such as cell division.
We first observed that FlhG numerically controls amphitrichous flagellation by examining flagellar biosynthesis of populations of wild-type C. jejuni and ΔflhG mutant strains. Over 90% of individual wild-type cells produced a single flagellum at one or both poles (62% were amphitrichous, 29% were monotrichous), which together were classified as the normal flagellar number phenotype (Figure 1A and 1C). Only about 1% of wild-type C. jejuni produced more than one flagellum at least at one pole. In contrast, approximately 40% of C. jejuni ΔflhG cells produced extra flagella at least at one pole, with a correlative decrease in the population producing wild-type flagellar numbers (Figure 1B and 1C). As a population, the ΔflhG mutant was less motile than wild-type C. jejuni (Figure S2A). Both motility and wild-type flagellar numbers were restored to C. jejuni ΔflhG by expressing flhG in trans (Figure 1C and Figure S2A).
In V. cholerae or P. aeruginosa, FlhG/FleN negatively regulates flagellar gene expression. As such, flhG/fleN mutants demonstrate increased expression of almost all classes of flagellar genes, which is believed to contribute to extra polar flagella [1], [2], [41]. Unlike these mutants, deletion of flhG in C. jejuni did not result in augmented expression of all classes of flagellar genes. Instead, we observed less than a 2.5-fold increase in expression of only σ54-dependent flagellar genes (encoding primarily rod and hook proteins), but not for other classes of flagellar genes, such as the early class encoding the flagellar type III secretion system (T3SS) or the late σ28-dependent flaA gene encoding the major flagellin (Figure S2B). These results suggest that FlhG is involved in numerical control of amphitrichous flagellation by a process different from monotrichous bacteria.
Since deletion of flhG resulted in an increase in the bacterial population that were hyperflagellated at least at one pole, we hypothesized that increasing the levels of FlhG in wild-type C. jejuni may suppress flagellation and increase the population of aflagellated bacteria. Systems to induce protein production are lacking in C. jejuni. Therefore, to increase FlhG levels in C. jejuni, flhG was overexpressed in wild-type C. jejuni by using the plasmid that complemented the C. jejuni ΔflhG mutant to restore proper flagellar numbers. However, overproduction of FlhG did not increase the aflagellated population compared to wild-type C. jejuni (Figure 1C).
Upon closer examination of the C. jejuni ΔflhG mutant by electron microscopy, we observed a change in the distribution of lengths of the bacterial cell bodies. In addition to bacteria of normal size, minicells were abundant in the C. jejuni ΔflhG population (Figure 2A-D). The minicells were normally 0.2–0.4 µm in diameter and originated from the poles of ΔflhG cells (Figure 2A and 2B). Many minicells were flagellated, and some were multiply flagellated due to the increased flagellation phenotype of the ΔflhG mutant (Figure 2C and 2D). We were unable to determine by phase-contrast or darkfield microscopy if the flagella rotated on minicells to confer motility (data not shown). These findings indicate that minicells are most likely generated due to division occurring at poles of C. jejuni ΔflhG, similar to observations of E. coli and B. subtilis minD mutants and a C. crescentus mipZ mutant [16], [36], [42].
We analyzed the lengths of individual C. jejuni cells to assess the abundance of minicells and distribution of cell lengths within a population. Wild-type C. jejuni cells averaged 1.41 µm in length with approximately 88% of the population between 1–2 µm (Figure 2E). More in depth analysis revealed about 54% of the bacteria were within a narrow range of 1.1–1.5 µm (Figure S3). Only 1.6% of wild-type C. jejuni were minicells, which were classified as bacterial-derived, spherical particles under 0.5 µm in diameter (Figure 2E). In contrast, the minicell phenotype of C. jejuni ΔflhG was pronounced, composing 28% of the cellular population (Figure 2E). In E. coli min mutants, both elongated cells and minicells are present in the bacterial populations [43], [44]. However, our analysis of the C. jejuni ΔflhG mutant, did not reveal an increase in the elongated cell population. Instead, the number of bacterial cells between 1–2 µm in the ΔflhG population was reduced, with the average cell length of the bacterial population consequently decreased compared to wild-type C. jejuni to 1.12 µm (Figure 2E and Figure S3). The reason for these differences in the population composition of E. coli and C. jejuni mutants producing minicell phenotypes is currently unknown.
Complementation of C. jejuni ΔflhG in trans with wild-type flhG greatly reduced the minicell population to less than 9%, demonstrating that the minicell phenotype of the ΔflhG mutant was due to loss of flhG (Figure 2E). We also noticed upon addition of flhG in trans to either wild-type C. jejuni or C. jejuni ΔflhG, the elongated cell population (>2 µm) increased to 24–26%, an approximately four-fold increase relative to wild-type C. jejuni (Figure 2E). This elongated cell phenotype is reminiscent of E. coli or B. subtilis strains upon minD overexpression in the presence of the FtsZ polymerization inhibitor, MinC, or mipZ overexpression in C. crescentus [15], [16], [36]. The occurrence of minicells upon elimination of flhG and the elongated cell phenotype upon increased FlhG production suggest that FlhG is involved in a process to inhibit cell division in C. jejuni.
We next verified that minicell production in C. jejuni ΔflhG was the result of a cell division event. Cephalexin is a late-stage cell division inhibitor that targets FtsI, which is required for peptidoglycan production at a septum during the final stages of cell division [45], [46]. We monitored minicell production in wild-type C. jejuni and ΔflhG mutant strains before and after a 6-hour incubation with the highest concentration of cephalexin that caused a noticeable cell division defect without killing the bacteria. Like untreated cells, the generation time of wild-type C. jejuni treated with 15 µg/ml cephalexin for six hours progressed normally through 1.5–2 doublings (data not shown). However, the cephalexin-treated bacteria displayed an increase in the population of elongated cells relative to untreated bacteria (Table 1).
Upon examination of the C. jejuni ΔflhG mutant, we noticed that the mutant was more sensitive to 15 µg/ml cephalexin that wild-type cells as indicated by cell lysis that obscured confident identification of minicell production by electron microscopy. Therefore, we treated the C. jejuni ΔflhG mutant with 12.5 µg/ml cephalexin. At this concentration, the ΔflhG mutant progressed through two doublings similar to untreated C. jejuni ΔflhG (data not shown). Without cephalexin treatment, the minicell population only slightly increased with time compared to the ΔflhG culture at the beginning of the experiment (Table 1), suggesting that minicell production is fairly consistent over time. However, six hours after cephalexin treatment, the minicell population was reduced about 56% compared to untreated ΔflhG cells (Table 1). Furthermore, an increased number of elongated cells was observed in the cephalexin-treated ΔflhG cells, indicating that cephalexin influenced cell division. We interpret the data as suggesting that minicells present after cephalexin treatment were likely those present at the start of the experiment and that cephalexin largely inhibited formation of new minicells. Therefore, we conclude that minicells are formed by a process that involves cell division in C. jejuni ΔflhG.
Members of the ParA family of ATPases contain a conserved aspartic acid residue that has been proposed to be required for ATP hydrolysis, but not for ATP binding [47], [48] (Figure S1). E. coli MinD and C. crescentus MipZ mutant proteins lacking this aspartic acid are thought to be locked into an ATP-bound state that caused cell elongation due to increased inhibition of cell division [36], [49]. Analysis of this type of mutation in E. coli MinD, revealed an increased association of the mutant protein with phospholipids even in the presence of MinE, and a peripheral distribution of the protein along the cytoplasmic membrane [49]. This distribution allowed the protein to function with MinC to inhibit cell division throughout the cell to result in elongation. In C. crescentus, production of the MipZD42A mutant protein resulted in a hyperactive form of the protein that was dominant over wild-type MipZ to result in cell elongation [36]. Considering these findings, we examined a role the putative ATPase domain of FlhG in influencing cell division by mutating the similarly conserved aspartic acid residue (D61) in FlhG. To perform these experiments, we replaced chromosomal wild-type flhG with flhGD61A, which we predicted would encode an FlhG mutant protein locked into an ATP-bound state, which may cause increased inhibition of cell division.
In the C. jejuni flhGD61A mutant population, we noticed a mixed population of cells, which contained cells of normal length and cells with elongated bodies (compare wild-type cells in Figure 3A with the flhGD61A mutant in Figure 3B and 3C). By analyzing the distribution of the lengths of cell bodies of the C. jejuni flhGD61A population, we found that approximately 24% of cells were elongated (>2 µm in length), compared to only 4% of wild-type C. jejuni (Table 2 and Figures 3D-F). Whereas elongated wild-type cells were largely confined to 2–3 µm in length, elongated C. jejuni flhGD61A cells of up to 9 µm in length were observed (Figure S4). A significant proportion of the flhGD61A population continued to produce cells of wild-type length between 1–2 µm (73.9% for C. jejuni flhGD61A vs 92.4% for wild-type C. jejuni; Table 2), suggesting that symmetrical cell division to produce daughter cells of normal lengths occurs with some frequency in many of these mutant cells. Of note, many flhGD61A cells produced wild-type flagella with a single flagellum at the poles and were motile as observed by darkfield microscopy (Figure 3D and 3F; data not shown). These observations suggest that the cells were metabolically active and viable.
As observed by electron microscopy, elongated bodies of the C. jejuni flhGD61A mutant often appeared to lack septa, indicating that FlhG may function in a mechanism to inhibit divisome formation (Figure 3D-F). One of the first steps in initiating cell division is formation of FtsZ into the Z-ring. Therefore, we tested if increasing the levels of FtsZ in C. jejuni flhGD61A could overcome the apparent cell division block in these cells and reduce the cell elongation phenotype of this mutant. Similar to E. coli [50], overexpression of ftsZ in trans in wild-type C. jejuni caused a 5-fold increase in minicell production (Table 2), indicating that FtsZ functions in cell division. Overexpression of ftsZ in C. jejuni flhGD61A reduced the elongated cell phenotype of this mutant from 24% to 9% (Table 2). These results suggest a regulatory link between FlhG and FtsZ, with Z-ring formation as a potential target of cell inhibition mediated by an FlhG-dependent mechanism.
Factors that spatially regulate formation of cell division sites have not been examined in other polar flagellates. Because other polarly-flagellated bacteria produce FlhG homologs that control flagellar number, we reasoned that these FlhG proteins may have an additional capacity like C. jejuni FlhG to influence cell division. Therefore, we analyzed if either FlhG or MinD proteins from other polarly-flagellated bacteria or E. coli MinD could functionally complement C. jejuni ΔflhG for numerical control of flagellar biosynthesis or spatial control of cell division to reduce the minicell phenotype. For these experiments, heterologous flhG or minD genes were cloned into a plasmid downstream of a constitutive promoter to ensure expression. These plasmids were then used to complement in trans C. jejuni ΔflhG. We found that H. pylori FlhG was just as competent as C. jejuni FlhG in reducing extra polar flagella and restoring wild-type flagellar numbers to C. jejuni ΔflhG (Figure 4A). Furthermore, H. pylori FlhG dramatically reduced the minicell population of the C. jejuni ΔflhG mutant and even caused an increase in elongated cells (Figure 4B), suggesting that H. pylori FlhG can function in a mechanism to inhibit cell division. In addition, V. cholerae FlhG partially restored both wild-type flagellar numbers and normal cell division to C. jejuni ΔflhG (Figure 4A and 4B). In contrast, all MinD proteins failed to complement C. jejuni ΔflhG for either phenotype (Figure 4A and 4B). These results indicate that H. pylori FlhG, and to a lesser extent V. cholerae FlhG, have the ability to modulate cell division. Secondly, these findings suggest that C. jejuni has evolved to preferentially use FlhG to regulate cell division and numerically control flagellar biosynthesis.
Our data suggest that the lack of FlhG results in the production of minicells due to polar cell division. Therefore, if FlhG is involved in a mechanism to prevent cell division at poles, it may be expected that FlhG localizes to polar regions to mediate this division inhibitory effect. We attempted to analyze the ability of FlhG to localize to poles of C. jejuni ΔflhG by using a plasmid to express flhG-gfp, which produces wild-type FlhG with a C-terminal GFP. The use of fluorescent protein technology to analyze cellular localization of proteins has typically been challenging in C. jejuni and other ε-proteobacteria. However, we were able to observe fluorescence due to FlhG-GFP in a small population of cells. In these bacteria, fluorescence was observed often at both poles, with some diffuse cellular fluorescence also visible (Figure 5). In contrast, C. jejuni ΔflhG producing GFP alone demonstrated diffuse fluorescence throughout the cell (Figure 5). These results suggest that FlhG is likely polarly localized and possibly available at poles to function in a mechanism to prevent cell division.
Because FlhG-GFP was detected primarily at poles in bacteria producing FlhG-GFP, we considered if other factors present at poles of C. jejuni may function with FlhG to limit cell division at poles. A leading candidate for a polarly-localized protein that may interact with FlhG is the FlhF GTPase. Previous studies suggested that the FlhF and FlhG proteins of C. jejuni and Vibrio alginolyticus interact, which may influence their respective ability to spatially and numerically control flagellar biosynthesis [4], [51]. FlhF of C. jejuni and other polar flagellates has been implicated as a regulatory factor required for expression of flagellar genes and to properly localize flagellar biosynthesis to poles [1], [4], [5], [7], [52]. Of note, FlhF polar localization has been observed in C. jejuni [53]. A current hypothesis for a role of C. jejuni FlhF in polar flagellar placement suggest that the GTPase activity of FlhF may influence its positioning to the new pole after cell division. After polar localization, FlhF may promote organization of the initial flagellar proteins, such as the motor, switch, and secretory components at the pole [6].
Considering the potential interactions between FlhF and FlhG, we examined a C. jejuni ΔflhF mutant and observed a minicell population that was at least as prevalent as that of C. jejuni ΔflhG (Figure 6A and Figure S5). To determine if the minicell phenotype of the ΔflhF mutant was linked to FlhG, we expressed flhG in trans in the ΔflhF mutant to result in flhG overexpression due to the presence of both chromosomal- and plasmid-encoded copies of the gene. In this strain, the minicell phenotype caused by deletion of flhF was suppressed (Figure 6C). These results suggest that FlhF and FlhG are linked in a mechanism to influence cell division in C. jejuni. Furthermore, by overexpressing FlhG in the ΔflhF background, FlhG overcomes an apparent dependency on FlhF to modulate cell division.
Two hypotheses were developed for how FlhF may influence an FlhG-dependent mechanism to inhibit cell division at poles. First, we considered if either σ54-dependent flagellar gene expression or flagellar rod biosynthesis, which are both dependent on FlhF [6], [54], are required for FlhG to function in a mechanism to inhibit cell division at poles. If lack of σ 54-dependent flagellar gene expression or rod biosynthesis caused the minicell phenotype in the ΔflhF mutant, minicell production in a ΔrpoN mutant (which lacks σ 54) or a ΔfliE mutant (which lacks rod biosynthesis) would be expected. However, neither mutant demonstrated a significant minicell phenotype compared to the ΔflhG or ΔflhF mutants (Figure 6A).
For the second hypothesis, we considered if flagellar components, which are likely dependent on FlhF for polar formation, are required to inhibit minicell formation. Although the very initial steps in flagellar biosynthesis are unknown in C. jejuni, it is likely that the first components of a flagellum that are constructed include: FliF (which forms the inner membrane MS ring); FliG, FliM, and FliN (the motor/switch components of the cytoplasmic C ring); and the flagellar T3SS (which is located within the MS ring) [55]. These components mediate motor, switch, and secretory functions for a flagellum. Of note, the MS ring of V. cholerae appears to be dependent on FlhF for polar localization [5].
We analyzed a panel of C. jejuni mutants lacking these flagellar components for a defect in cell division that results in production of minicells. Inactivation of fliF, fliM, and fliN resulted in strong minicell phenotypes comparable to C. jejuni ΔflhF and ΔflhG mutants (Figure 6A and Figure S5). In contrast, a fliG mutant or mutants lacking a single component of the flagellar T3SS either did not produce a significant level of minicells or showed a significantly reduced minicell phenotype relative to C. jejuni ΔflhG (Figure 6A). These findings suggest that the MS ring and switch complex (made up of FliM and FliN) are involved in a mechanism to influence cell division in C. jejuni.
To verify that minicells are products of cell division in the flhF, fliF, fliM, and fliN mutants, minicell production was monitored in mutants after exposure to cephalexin. In each case, the minicell population was reduced in cephalexin-treated cells (Figure 6B). In the fliF, fliM, and fliN mutants, the minicell population was reduced over 50%. We next determined if the minicell phenotypes of the fliF, fliM, and fliN mutants are linked to FlhG, similar to what we observed with C. jejuni ΔflhF. When we overexpressed flhG in trans in each of these mutants, the minicell phenotype decreased (Figure 6C). Furthermore, deletion of flhG in the fliF, fliM, or fliN mutants did not augment minicell production in the mutants compared to single deletions of each gene (Figure 6D), suggesting that the MS and switch complex of flagella function in the same pathway as FlhG to influence a mechanism to inhibit cell division. Considering our findings, we surmise that polar flagellar biosynthesis influences formation of cell division sites via FlhG to ultimately result in symmetrical cell division in C. jejuni.
Elegant studies of E. coli, B. subtilis, and C. crescentus have elucidated highly refined mechanisms by which bacteria regulate precise placement of the divisome to promote cell division and generate viable progenitors. Although the mechanisms vary, a common theme in regulating divisome placement is that bacterial poles are usually protected from cell division so that the divisome more likely forms near the cellular midpoint for symmetrical division. We have identified a new collection of factors that compose a system to regulate formation of cell division sites in C. jejuni. This novel system is composed of the ParA ATPase family member FlhG and the MS ring and switch complex of polar flagella. Our work indicates that FlhG numerically controls amphitrichous flagellation and spatially regulates cell division. Furthermore, components of polar flagella are required for FlhG produced at normal levels to inhibit cell division specifically at polar regions. In the absence of FlhG, the MS ring, or switch components, cell division at poles more freely occurs to generate minicells. Thus, FlhG and polar flagellar biosynthesis block cell division from occurring at poles so that symmetrical division predominates to ensure generation of viable progenitors.
Other ParA ATPase family members, such as E. coli and B. subtilis MinD or C. crescentus MipZ, regulate formation of cell division sites by protecting poles from cell division, but the mechanisms by which these proteins function differ. MinD proteins do not directly inhibit cell division. Instead, these proteins localize MinC, the FtsZ inhibitor, to poles or maturing septa to inhibit divisome formation [10]–[13], [15], [56]. In contrast, C. crescentus MipZ directly mediates FtsZ depolymerization at polar sites [36]. C. jejuni FlhG is more homologous to MinDs than MipZ, with approximately 55% similarity and 35% identity between a 167-amino acid region of FlhG and E. coli MinD. This region includes the ATPase domains and some surrounding residues. In addition, a C-terminal amphipathic helix present in MinDs, but absent in MipZ, that is required for membrane interactions at poles to mediate inhibition of cell division is predicted at the C-terminus of FlhG [10], [57]–[59]. In contrast, FlhG and MipZ only share homology that is limited to a 40-amino acid region within the ATPase domains.
Considering how MinD and MipZ spatially mediate inhibition of cell division, it is currently unclear how FlhG may modulate cell division in C. jejuni. Production of FlhGD61A, which is predicted to be locked into an ATP-bound state, resulted in an elongated phenotype likely due to the mutant FlhG protein conferring a heightened cell division inhibitory activity. These results suggest that cycles of ATP binding and hydrolysis by FlhG are likely important for normal spatial regulation of Z-ring formation, similar to what has been observed with MinD and MipZ [18]–[20], [36], [60]. Due to the lack of genes encoding MinC and MinE in C. jejuni, it is unlikely that FlhG functions in an identical mechanism as MinD to inhibit cell division. However, FlhG may interact with and polarly localize other proteins with MinC-like functions in directly inhibiting Z-ring formation. Alternatively, FlhG may function similarly to C. crescentus MipZ to directly interact with FtsZ and inhibit Z-ring formation at poles. However, preliminary experiments failed to demonstrate that purified FlhG stimulated the GTPase activity of C. jejuni FtsZ in vitro (MB and DRH, unpublished observations), which would promote depolymerization of Z-rings into FtsZ monomers and inhibit cell division in the bacterium [61]–[63]. In addition, we were unable to detect an in vitro ATPase activity for FlhG (MB and DRH, unpublished observations), yet in vivo analysis of the elongated phenotype promoted by the flhGD61A mutant suggested that FlhG binds and hydrolyzes ATP for normal spatial regulation of Z-ring formation. These results indicate either that our in vitro conditions are not optimal for detecting a direct inhibitory activity of FlhG for FtsZ polymerization into Z-rings or that other components are required to activate or function with FlhG to inhibit FtsZ polymerization. Due to the requirement of flagellar components to limit cell division at poles, FlhG may require intact flagellar MS ring and switch structures to inhibit Z-ring formation and cell division.
As with the Min and MipZ division-site determination systems, we expect that an FlhG-dependent mechanism to protect poles from cell division likely requires some sort of topological specificity factor that either spatially confines or specifically activates this system at polar regions. Due to the minicell phenotype of MS ring and switch complex mutants, it is tempting to speculate that formation of a MS ring and C-ring switch complex at a pole, which is numerically regulated by FlhG, forms a topological specificity factor that assists FlhG and possibly other associated factors to facilitate inhibition of cell division specifically at poles. Without fliF, fliM, and fliN, a mechanism involving FlhG to inhibit cell division at poles is inoperable. Furthermore, the lack of an elongated cell phenotype in these mutants also suggests that this mechanism is not simply spatially deregulated and blocking divisome formation throughout the cell. Hence, specific flagellar components function with FlhG to inhibit cell division when the protein is produced at normal levels in the bacterium.
The minicell phenotype of fliF, fliM, and fliN mutants could be suppressed by increasing expression of flhG. Furthermore, analysis of double mutants lacking flhG and either fliF, fliM, or fliN did not reveal an increase in minicell production compared to the deletion of flhG alone. These findings together suggest that FlhG functions downstream of these flagellar components to influence cell division, rather than FlhG, the MS ring, and switch complex functioning in two separate pathways to influence cell division. Curiously, the base of the MS ring (composed by FliF) and the FliM and FliN structures in the switch complex of the C ring are cytoplasmic-accessible portions of the flagellar organelle, which may be available to interact with FlhG. In contrast, the flagellar T3SS and FliG, which are internal components of the MS and C rings, respectively, did not appear to be required to inhibit cell division at poles. If the MS ring and switch complex compose a topological specificity factor, FlhG may require a superficial domain of the flagellar motor and switch complex to initiate a mechanism to spatially inhibit cell division specifically at polar regions. Possible hypotheses for a mechanism by which flagellar components assist in modulating cell division include: 1) the flagellar structures may interact with FlhG to accumulate the protein to a critical concentration necessary to specifically inhibit Z-ring formation and cell division at poles; or 2) specific flagellar proteins or flagellum-associated components may activate a FlhG-dependent mechanism to inhibit cell division that is spatially confined to poles. Currently, our data do not allow for discerning which hypothesis may be true. The observation that FlhG alone did not stimulate the GTPase activity of C. jejuni FtsZ in preliminary assays suggests that FlhG may require intact flagellar components or other factors to promote FtsZ depolymerizaton (MB and DRH, unpublished observations). On the other hand, the fact that flhG overexpression suppressed the minicell phenotype of MS ring and switch complex mutants suggest that a FlhG-dependent mechanism to inhibit cell division is functional without flagella if FlhG levels are artificially high, which may add more strength to the hypothesis that polar flagellar structures promote polar accumulation of FlhG when produced at normal levels to inhibit cell division at poles. We attempted to determine if polar localization of FlhG-GFP was dependent on the flagellar motor and switch complex. Due to the low number of cells producing fluorescence and the low level of fluorescence of the fusion protein in these flagellar mutants, we were unable to confidently conclude that polar flagellar structures are required to localize FlhG to poles (MB and DRH, unpublished observations). Improved fluorescence microscopic procedures will be required to identify factors required for polar localization of FlhG. Although much more is to be learned about this system, our data suggest that polar flagellar biosynthesis functions with FlhG to inhibit cell division at poles.
Our findings have also revealed a previously unrecognized biological advantage for amphitrichous flagellation of C. jejuni that extends beyond an obvious role in promoting motility. Amphitrichous flagellation confers a characteristic darting motility for C. jejuni that assists in colonization of the intestinal mucosa in hosts [64], [65]. However, our studies have found that C. jejuni possesses a flagellum-influenced cell division inhibition system. The construction of such a system appears to allow amphitrichous flagellar biosynthesis, which is numerically controlled by FlhG, to influence an FlhG-dependent mechanism to prevent cell division at both poles. We propose that immediately after cell division, two daughter cells lack a flagellum at the new pole. Initiation of a single round of flagellar biosynthesis at this pole to complete the amphitrichous flagellation pattern would result in a MS ring and switch complex structure that an FlhG-dependent mechanism requires to inhibit cell division specifically at the new pole. A strictly monotrichous flagellar pattern in C. jejuni may inhibit cell division only at the flagellated pole with the aflagellated pole susceptible to divisome formation. A peritrichous flagellation pattern in C. jejuni may inhibit division throughout the cell. As such, amphitrichous flagella facilitate a mechanism so that FlhG-dependent cell division inhibition primarily occurs at the poles, which encourages symmetrical cell division to generate the highest number of viable C. jejuni daughter cells.
The polarly-flagellated bacteria commonly studied for motility, such as Vibrio, Pseudomonas, and Helicobacter species, encode FlhG and all Min components, except for Campylobacter species. A likely hypothesis for most polar flagellates is that FlhG controls numerical parameters of polar flagellar biosynthesis, whereas the Min system influences division-site placement. However, we observed that H. pylori FlhG, and to a lesser extent V. cholerae FlhG, resolved the minicell phenotype of C. jejuni ΔflhG, indicating that these proteins have the ability to influence cell division and possibly spatially regulate divisome formation. Therefore, the ability of FlhG to influence cell division may extend to other polarly-flagellated bacteria and form a broad mechanism used by many other motile bacteria to regulate cell division processes.
In this work, we have established a new paradigm that links polar flagellar biosynthesis to cell division in bacteria. Furthermore, we showed how amphitrichous flagellation is beneficial for influencing symmetrical cell division in Campylobacter so that two normal daughter cells are generated during each round of division. In addition, we provided a new function for the flagellar MS ring and switch complex in functioning with FlhG to prevent cell division from occurring at polar sites. Further exploration of this system will undoubtedly lead to a new understanding of the process of cell division that may occur in a broad range of polar flagellates.
All C. jejuni 81–176 SmR strains and procedures for generating mutants are described in Text S1 and Tables S1 and S2. For all experiments, C. jejuni strains were grown from freezer stocks on Mueller-Hinton (MH) agar containing 10 µg/ml trimethoprim for 48 h under microaerobic conditions at 37°C. Strains with plasmids for complementation analyses were grown with 50 µg/ml kanamycin. Strains were then restreaked onto MH agar containing appropriate antibiotics and grown for an additional 16 h and then used in experiments accordingly as described.
Introduction of flhG or minD alleles on plasmids into wild-type C. jejuni 81–176 SmR or C. jejuni 81–176 SmR ΔflhG for overexpression or complementation was accomplished by amplifying the alleles with primers containing 5' BamHI restriction sites immediately upstream of the start and stop codons of the respective genes. flhG alleles were amplified from the chromosomal DNA of C. jejuni 81–176, H. pylori J99, and V. cholerae O395. minD alleles were amplified from the chromosomal DNA of H. pylori J99, V. cholerae O395, and E. coli MG1655. The alleles were cloned into BamHI-digested pCE107, an E. coli-C. jejuni shuttle vector containing the σ28-dependent flaA promoter of C. jejuni followed by a BamHI site fused in-frame to a gene for the Zoanthus species green-fluorescent protein [66]. Insertion of flhG or minD alleles in the correct orientation placed a stop codon between the allele and the gene for GFP, preventing the formation of a fusion protein. All plasmids were sequenced and then transformed into DH5α/RK212.1 [67]. The plasmids were then conjugated into the appropriate C. jejuni 81–176 SmR strains by a previously published method [68].
To increase expression of ftsZ in C. jejuni SmR flhGD61A (MB1054), ftsZ was amplified from chromosomal DNA of C. jejuni 81–176 using primers containing 5' BamHI restriction sites immediately upstream of the start and stop codons of the gene. ftsZ was cloned into BamHI-digested pCE107 and then conjugated into wild-type C. jejuni or the flhGD61A mutant as described above. Expression of ftsZ from this plasmid, along with constitutive expression of ftsZ from the native chromosomal locus allowed for expression of ftsZ at increased levels relative to wild-type C. jejuni.
To increase expression of flhG in the C. jejuni ΔflhF (DRH1056), ΔfliF (DRH2074), fliM (DRH3304), and fliN (DRH1407) mutants, the coding sequence of flhG was amplified from chromosomal DNA from C. jejuni 81–176 with primers containing BamHI restriction sites in-frame to codon 2 and the stop codon. flhG was then cloned into BamHI-digested pECO101, an E. coli-C. jejuni shuttle vector containing the promoter of the chloramphenicol-acetyltransferase (cat) gene. After screening for correct orientation of flhG and sequencing, one plasmid (pMB1230) was transformed into DH5α/RK212.1 for conjugation into C. jejuni mutants as described above. Constitutive expression of flhG from the cat promoter on the plasmid, along with constitutive expression of flhG from the native chromosomal locus allowed for expression of flhG at increased levels relative to wild-type C. jejuni.
Construction of a plasmid to produce a FlhG-GFP fusion protein was accomplished by amplifying the coding sequence of flhG from C. jejuni 81–176 chromosomal DNA with primers containing BamHI restriction sites in frame the start and penultimate codons. This fragment was then cloned into BamHI-digested pCE107 [53]. One plasmid, pMB722, contained flhG in the correct orientation to produce a FlhG-GFP fusion protein. pMB722 and pCE107, which expressed GFP alone, was transformed into DH5α/pRK212.1 for conjugation into C. jejuni 81–176 SmR ΔflhG (MB770) as described above.
Strains were grown for 16 h on agar plates and then resuspended in PBS, pelleted for 3 min at full speed in a microcentrifuge, resuspended in 2% gluteraldehyde in 0.1 M cacodylate, and then incubated on ice for 1 h for fixation. Samples were then stained with 2% uranyl acetate and visualized with a FEI Technai G2 Spirit BioTWIN transmission electron microscope.
For analysis of the effect of cephalexin treatment on cell length and minicell production, strains were suspended from agar plates after growth for 16 h on agar plates and diluted to an OD600 1.0. Each strain was diluted 1∶10 in MH broth in two separate flasks. Cephalexin was added at a final concentration of 15 or 12.5 µg/ml for wild-type C. jejuni or the C. jejuni ΔflhG mutant, respectively. Cultures were then incubated under microaerobic conditions at 37°C for 6 h. Samples were taken at time 0 h and 6 h to determine the number of viable bacteria in each culture and for analysis by electron microscopy. Samples were then fixed and stained as described above for visualization by transmission electron microscopy.
Data from two separate experiments were combined and averaged to determine the proportion of bacterial populations producing different numbers of polar flagella or cell bodies of different lengths as visualized by transmission electron microscopy. In total, over 210 individual bacteria were analyzed for each strain. For analysis of flagellar numbers, bacteria were placed into one of three categories: >2 flagella, producing at least two flagella at one or both poles; wild-type flagella, producing a single flagellum at one or both poles; or 0 flagella, aflagellated bacteria. For analysis of cell body lengths, bacteria were placed into one of four categories:<0.5 µm, minicells; 0.5–1 µm; 1–2 µm; and >2 µm. After averaging, the standard error for each population was calculated.
After 16 h growth, bacteria were suspended from agar plates in MH broth and then diluted to OD600 1.0. Approximately 1.5 ml of culture was pelleted in a microcentrifuge, followed by fixation with 4% formalin. Then, 350 µl of fixed cells were stained with 20 µl of 1 mg/ml FM4–64 for 15 min. Samples were added to poly-L-lysine-coated chamber slides. After 5 min, excess liquid was removed with a vacuum. ProLong Gold antifade reagent was applied to the chamber slides. After 24 h, fluorescent images were obtained with an Applied Precision PersonalDV deconvolution microscope with an Olympus 100x objective lens and a CoolSNAP_HQ2 camera. Images were processed using the ImageJ program.
The following GenBank accession numbers identify all previously uncharacterized proteins that were analyzed in this work: Campylobacter jejuni 81–176 FlhG, EAQ71939; Campylobacter jejuni 81–176 FliE, EAQ73125; Campylobacter jejuni 81–176 FliQ, EAQ72806; Campylobacter jejuni 81–176 FliM, EAQ71948; Campylobacter jejuni 81–176 FliN, 73197; Campylobacter jejuni 81–176 FliG, EAQ73253; Helicobacter pylori 26695 FlhG, AAD08077; Helicobacter pylori 26695 MinD, AAD07400; Vibrio cholerae O395 MinD, ACP10067.
|
10.1371/journal.pbio.0050046 | Pre-Steady-State Decoding of the Bicoid Morphogen Gradient | Morphogen gradients are established by the localized production and subsequent diffusion of signaling molecules. It is generally assumed that cell fates are induced only after morphogen profiles have reached their steady state. Yet, patterning processes during early development occur rapidly, and tissue patterning may precede the convergence of the gradient to its steady state. Here we consider the implications of pre-steady-state decoding of the Bicoid morphogen gradient for patterning of the anterior–posterior axis of the Drosophila embryo. Quantitative analysis of the shift in the expression domains of several Bicoid targets (gap genes) upon alteration of bcd dosage, as well as a temporal analysis of a reporter for Bicoid activity, suggest that a transient decoding mechanism is employed in this setting. We show that decoding the pre-steady-state morphogen profile can reduce patterning errors caused by fluctuations in the rate of morphogen production. This can explain the surprisingly small shifts in gap and pair-rule gene expression domains observed in response to alterations in bcd dosage.
| Subdivision of naive fields of cells into separate cell populations, distinguished by the unique combinations of genes they express, constitutes a major aspect of organism development. Classically, this involves the generation of gradients of signaling molecules (morphogens), which induce distinct cell fates in a concentration-dependent manner. It has been generally assumed that morphogen gradients are interpreted only after they reach a spatially fixed, steady-state profile. Our study re-examines this assumption for the classical case of the Bicoid morphogen, a transcription factor that is distributed as a gradient in the early Drosophila embryo. We propose and present evidence for dynamic, pre-steady-state decoding of the Bicoid profile. We further show that such dynamic decoding can directly account for the surprisingly small shifts in the expression domains of target genes, observed in response to altered Bicoid dosage, without invoking additional mechanisms or contributing factors. Pre-steady-state decoding can thus provide a simple explanation for the relative robustness of this classical morphogen system, which has been a long-standing problem.
| Developmental patterning requires the translation of cell position into cell fate. In most prevalent models, positional information is provided by gradients of signaling molecules, called morphogens, which induce several cell fates in a concentration-dependent manner [1]. Prominent examples of such morphogens include members of the BMP, Wnt and Hh families of signaling molecules, which play a crucial role in patterning a broad spectrum of tissues and organisms [2–8]. While a variety of molecular mechanisms involved in the establishment of morphogen gradients have been described, the means by which these gradients are decoded are not well understood. In particular, little is known about the time at which the morphogen signal is being interpreted by its downstream targets. Most studies assume that the eventual pattern is defined according to the steady-state morphogen profile. Relying on the steady-state profile provides two obvious advantages. First, it allows for a temporal integration of a stable gradient, and as such may increase the readout fidelity. Second, it is relatively insensitive to the precise readout time and may thus compensate for perturbations that alter developmental timing.
Recent theoretical studies in several systems, however, predicted that the underlying cells respond to the pre-steady-state morphogen profile. For example, numerical simulations of Shh morphogen formation in the neural tube suggested that as soon as the morphogen signal increases above some threshold value, it can induce a given cell fate, implying that tissue patterning occurs before the morphogen concentration has reached its steady state [9]. Similarly, based on numerical simulations of mutant data, other authors argued in favor of pre-steady-state readouts in the gap gene interaction network [10] and of the BMP gradient [11] during early patterning of the Drosophila embryo. However, qualitative differences between pre-steady-state versus steady-state patterning, and their biological implications, have not been addressed.
A key aspect in developmental patterning is robustness: patterning is remarkably insensitive to fluctuations in the external environment or the precise genetic makeup. In fact, most genetic polymorphisms, or heterozygous mutations in developmentally related genes, have no detectable effect on patterning. Recent studies characterized feedback mechanisms that can be employed for shaping morphogen gradients and buffering their profile against fluctuations in gene dosage or environmental perturbations [12–21]. Most of the feedback mechanisms described require some time delay, and are most effective in steady state. In contrast, the robustness of decoding the pre-steady-state profile has not yet been examined.
The early patterning of the Drosophila embryo along its anterior–posterior axis serves as a classic example of morphogen-based patterning (see [22] for recent review). A principle morphogen in this system is Bicoid (Bcd), a transcription factor that is translated from maternally provided mRNA localized to the anterior pole of the embryo. The graded distribution of Bcd was visualized directly, providing the first molecular demonstration of a gradient of patterning molecules [23,24]. Bcd binds to the promoters of zygotic downstream genes (gap genes), and induces their expression in a concentration-dependent manner [25–30]. This Bcd-dependent induction, together with cross-interactions between the gap genes themselves, governs the patterning of the early embryo into distinct domains of gene expression [30]. Importantly, this patterning proceeds rapidly, with gap gene expression observed in less than 90 min after the onset of embryonic development [31].
Consistent with its presumed function as a morphogen, changes in bcd gene dosage shift the expression domains of gap genes such as hunchback (hb) [28,32] as well as the embryonic fate map [23]. Quantitative measurements, however, revealed that these shifts are significantly smaller than expected from a simple morphogen model [23,32]. For example, in embryos derived from mothers bearing only one functional allele of bcd, the Hb expression domain shifts by only ∼7% in embryo length (EL), about half of what is expected theoretically (see Equation 4 and [32,33]). Based on these apparent inconsistencies, it has been argued that the Bcd gradient is not sufficient for defining gap gene expression domains, and that an additional, and yet unknown, molecular mechanism is required to complement the Bcd gradient [32,33–36].
Here we report an analysis of the dynamics of gap gene determination. We found that the apparent anomalous shifts in gap gene expression domains can be readily explained if the gap genes start being expressed before the Bcd profile has reached its steady state. The interactions between gap genes can stabilize the initial gap expression domains, while the Bcd profile continues to expand. We show analytically that decoding the pre-steady-state profile enhances the robustness to changes in morphogen production rate. Two predictions of the pre-steady-state decoding were examined experimentally. First, we find that the observed shifts in gap gene expression domains depend on their position along the anterior–posterior axis, with more posterior positions being less sensitive to Bcd dosage. This result can be readily explained by a dynamic readout, but is inconsistent with decoding the steady state of the Bcd profile (compare with [15]). Second, using a reporter gene driven by a promoter containing Bcd-binding sites, we provide evidence that the Bcd profile still spreads out posteriorly at the times relevant for gap gene induction. Taken together, our results suggest that gap gene expression domains are defined by a transient, pre-steady-state Bcd profile. This pre-steady-state decoding reduces the sensitivity of the resulting pattern to changes in bcd dosage.
Previous attempts to explain the anomalous shifts in Hb expression domains invoked maternally expressed Hb (mat-Hb) [37], predicting a significant contribution of mat-Hb to embryonic patterning. Indeed, translational repression of mat-Hb by the Nanos protein in the posterior part of the embryo establishes an anterior–posterior gradient of Hb protein [38–40], and Hb protein was shown to synergize with Bcd in patterning the embryo [41–44]. Yet, this proposal was found to be inconsistent, since gap gene expression domains in embryos derived from mat-Hb–deficient females are indistinguishable from wild-type embryos [32,38–40], reflecting the dominance of Bcd-dependent zygotic Hb expression over the contribution of mat-Hb (compare with Protocol S1). Alternative explanations invoked the existence of a secondary, yet to be identified morphogen gradient that is linked to Bcd (e.g., through the use of a common degradation machinery [33–36]).
Recently, a quantitative model of the gene network controlling gap gene expression was reported [45,46]. This model successfully reproduced the gap gene expression domains under wild-type conditions, and showed that gap gene patterning is a dynamic process to which the Bcd gradient only contributes the initial cue. Nevertheless, our reanalysis of this model for altered bcd dosage failed to reproduce the observed shifts of the gap gene expression domains (see Protocol S1). Since most model parameters, such as diffusion constants, transcription rates, or degradation rates, are not firmly established, we asked whether under a different set of assumptions, the gap gene network could still explain the lower than expected sensitivity of the gap gene expression domains to an altered bcd dosage.
Previous dynamic models considered a steady-state Bcd profile [33–37,45,46]. A recent review, however, emphasized the importance of assessing the dynamics of Bcd explicitly [47]. We have thus included these dynamics in our simulations. We considered the known interactions between Bcd and the gap genes, as well as the cross-interactions between the gap genes [30,45,46] (see Figure 1A and Protocol S1 for details of our in silico model). We assumed that translation of bcd mRNA is localized at the anterior pole of the embryo, and is initiated upon egg laying (defined as time t = 0). Zygotic gap gene expression begins at a later time (tgap > 0), taken here as 90 min (corresponding to cycle 10 of the synchronous nuclear divisions of the early embryo, at 25 °C [31]. To characterize the model we first calculated the pattern of gap gene expression in wild-type embryos (with two bcd alleles), and compared these predictions with the experimentally determined patterns. We then measured the shift in the Hb expression domain upon a two-fold change in bcd gene dosage (corresponding to embryos whose mothers had either one or four functional bcd alleles).
The behavior of the model was analyzed for a range of realistic Bcd diffusion coefficients, as well as for different values of the parameters describing the gap gene network. A parameter regime was readily identified which reproduced both the wild-type pattern, as well as the experimentally observed shifts in Hb expression upon changes in bcd dosage (Figure 1B and 1C). Surprisingly, analysis of the relevant parameters suggested that reduced sensitivity to alterations in bcd gene dosage is achieved when pre-steady-state decoding is used: the Bcd profile at the onset of zygotic expression was still far from steady state (Figure 1D). Indeed, increasing the Bcd diffusion constant significantly enhances the sensitivity of Hb expression domains to bcd dosage (Figure 1E). Notably, the Hb expression domain, as well as those of the other gap genes, displayed only a small drift following their initial determination, although the Bcd profile continued to evolve (Figure 1D). This temporal stabilization of gap gene expression pattern is due to the mutual repression between adjacent gap genes and their limited diffusibility. Proper patterning can be achieved for a wide range of parameters, and the exact choice has only a marginal effect on the level of robustness (Figure 1F).
Our results thus indicate that the known interactions between the gap genes are sufficient to account for the phenotypes of embryos derived from mothers with altered bcd dosage. Previous studies, which concluded that the experimentally observed shifts in the Hb expression domain are inconsistent with a simple threshold model, calculated the expected shifts based on the assumption that the Bcd profile has reached a steady state [23,24,32]. In contrast, a significantly smaller shift is expected if decoding is executed before steady state has been reached.
Our numerical simulations indicate that the sensitivity to changes in Bcd dosage is lower when decoding is performed early, before steady state is reached. To better understand this result, we studied analytically the properties of the time-dependent morphogen profile. (Readers less interested in the mathematical details are encouraged to move directly to the next section.) We considered the canonical model of a morphogen system, applicable in the absence of feedback mechanisms affecting morphogen diffusion or degradation. The model postulates a single morphogen that diffuses in a naive field, where it is subject to uniform degradation. The time-dependent morphogen profile M (x,t) is obtained by solving the reaction-diffusion equation
where D and τ denote the morphogen diffusion coefficient and degradation time, respectively. We assume that morphogen is produced at x = 0 at a constant rate s0. Equation 1 can be solved analytically (see Protocol S1 for derivation), giving
As shown in Figure 2A, morphogen spreads away from its source and assumes a more graded spatial distribution with time. At steady state, morphogen is distributed exponentially, M(x) = M0 exp(−x/λ), decaying over a typical length-scale
. Note that the time to reach steady state is controlled by the typical degradation time τ. At early times, t ≪ τ, the system is still far from steady state, whereas for t ≫ τ, the morphogen gradient is close to steady state. Moreover, closer to the source the morphogen gradient approaches its steady state faster (Figure 2B).
To examine the robustness of the profile, we considered gene expression boundaries defined according to particular threshold levels of morphogen concentration. We then determined the position at which the morphogen level equals to that threshold. The exact shift in boundary position caused by a change in the morphogen production rate can be calculated numerically using Equation 2 (Figure 2C–2F; see also Protocol S1). To obtain analytical insight, however, it is instructive to consider the following phenomenological approximation for the time-dependent morphogen profile
where M0(t) is proportional to the morphogen production rate. In this approximation, the exponent p(t) decreases monotonically with time. For a pulse-like morphogen production, the morphogen distribution at short times (t ≪ τ) resembles a Gaussian distribution, corresponding to p(t) = 2 and
(see Protocol S1). When production is continuous, the short-time distribution is better approximated by a smaller exponent, p(t) ≈ 1.6 (Figure 2G). At longer times (t ≫ τ), the distribution approaches an exponential profile, corresponding to p(t) = 1 and λ(t) = λ.
Within this approximation, the computation of the shift in the boundary position is straightforward. Suppose that the morphogen-production rate is altered by a factor γ. We can approximate the position-dependent shift in morphogen profile as (see Protocol S1)
with p = p(t) and λ = λ(t). Clearly, for x ≥ λ, the magnitude of the shift Δx increases with decreasing p > 1. Equation 4 demonstrates that in most of the field (x > λ), the shift in boundary position increases with decreasing p. Since p decreases in time toward its minimal steady-state value (p = 1), a greater degree of robustness is achieved at earlier times, before steady state is reached. This result also holds for the exact solution in Equation 2 (see Figure 2F for numerical analysis). In fact, the system is most sensitive when cell fate boundaries are defined according to the steady-state morphogen profile. Thus, the capacity to buffer fluctuations in morphogen production rate is enhanced if decoding is executed early, when the gradient is still far from steady state.
Notably, at steady state (p = 1), the shift Δx is predicted to be independent of the position x. Thus, a uniform shift independent of the distance is a hallmark for the decoding of a steady-state profile. Indeed, as we have shown previously [15], this result holds not only for the canonical model studied here, but for any decoding of a single morphogen steady-state gradient also in the presence of arbitrary feedback mechanisms affecting morphogen diffusion or degradation. In contrast, when decoding is based on the transient, pre-steady-state morphogen levels (p > 1), the magnitude of the induced shift is position dependent, and decreases with increasing distance from the source. Again, this effect is seen also in the numerical analysis of the full solution in Equation 2 (Figure 2D–2F).
To further examine the possibility that gap gene expression domains are defined by the pre-steady-state Bcd profile, we searched for properties that distinguish between steady-state and pre-steady-state decoding strategies. This search was guided by our mathematical analysis of the canonical morphogen model that takes into account the spatiotemporal formation of the morphogen gradient (see above). Briefly, we considered perturbations to the morphogen production rate, and analyzed the resulting shifts in expression patterns predicted by this model when decoding is performed at different timepoints following the initiation of morphogen production. The most prominent distinction between steady-state versus pre-steady-state decoding that we observed was that in the case of steady-state decoding, the extent of the shift in an expression domain was independent of the spatial position of this domain (Figure 2C). (This result is valid for the decoding of any steady-state profile of a single diffusing morphogen even in the presence of feedback, cooperativity, or other form of nonlinearity; compare with [15].) In contrast, when decoding was based on transient, pre-steady-state morphogen levels, the magnitude of the induced shift was position dependent, decreasing with increasing distance from the source (Figure 2D and 2E).
To examine which of these two behaviors is observed in early Drosophila embryos, we measured the shifts of different gap gene expression domains induced by altered bcd gene dosage. We examined embryos derived from mothers carrying only one functional bcd allele, as well as embryos derived from wild-type females (bearing two functional alleles) and from females that carry two additional (total of four) functional bcd alleles. Using existing antibodies [48], we stained these embryos for the protein products of several gap genes (hb, Kruppel [Kr], or giant [gt]), and for the downstream pair-rule gene even-skipped (eve), whose expression is gap gene dependent. Automated image processing was used to determine expression domain boundaries (Figure 3A–3D). As reported previously, when the maternal bcd dosage was reduced to one copy, all bcd-dependent expression domains were shifted towards the anterior part of the embryo, while increasing maternal bcd dosage to four copies resulted in shifting of all domains towards the posterior end [23,32].
Yet, the extent of the shifts in gap gene expression domains was not uniform, but decreased towards the posterior pole, such that expression domains closer to the source were more strongly affected by changes in bcd dosage (Figure 3E and 3F). Moreover, the measured shifts in midembryo positions were generally consistent with those obtained in numerical simulations based on pre-steady-state decoding and a Bcd diffusion constant D ∼ 1 μm2/s (gray lines in Figure 3E and 3F; compare also Figure 1G and 1H). Deviations from the predicted shifts were observed for posterior expression domains (e.g., posterior Hb), probably reflecting their dependence on the terminal gap genes tailless and huckebein [49–51], and on maternally provided caudal [30,52,53].
As a more direct test of the pre-steady-state decoding strategy, we wanted to follow temporal changes in the Bcd profile itself, at a time when gap gene expression domains are first defined. Gap gene expression is clearly observed at division cycle 10 (∼90 min after egg lay at 25 °C), with some reports suggesting that it is initiated as early as cycle 8 (∼20 min earlier; see [31] and references therein). Recent analyses have identified a similar time window (65–100 min after fertilization) as the critical time for perturbing gap gene expression domains [54]. Moreover, degradation of bcd mRNA is initiated at cycle 12 [55], further pointing to cycles 10–11 as relevant for gap gene determination. Examining anti-Bcd staining images from the FlyEx database [56] we observed that Bcd profiles in cycles 10–12 appear to have not yet reached an exponential shape (see Figure 2I).
Direct immunological quantification of the Bcd profile at early stages is difficult, however, since existing antibodies provide low and variable staining intensity. To overcome this limitation, we resorted to a functional assay using a Bcd-responsive reporter. For this assay we chose to use hb123x3-lacZ, in which lacZ reporter expression is under the control of a triplicated fragment of 123 bp derived from the hb promoter, containing multiple, functional Bcd-binding sites [28]. An added benefit of this approach is the sharp transition of reporter expression, which facilitates the determination of the transcription domain regardless of the absolute level of expression. Transgenic flies bearing multiple copies of this reporter were generated as a means to increase the signal. The expression pattern of this reporter was previously shown to faithfully reflect Bcd activity, and to bind Bcd directly [28]. Although we cannot completely rule out binding of additional factors to this short element, we note that this reporter also maintains its precise expression domain in the absence of zygotic Hb [28]. Since the removal of zygotic Hb was shown to shift the adjacent Kruppel and Knirps expression domains [41], it is unlikely that the expression domain of the reporter element we are using is influenced by cardinal gap genes.
In situ hybridization to mRNA provides a sensitive readout of reporter expression. Expression of lacZ mRNA could be observed in embryos bearing the reporter beginning at cycle 11. The total level of expression increased with time due to the elevated efficiency of zygotic gene expression in subsequent cycles and the increase in the number of nuclei. A significant shift in the posterior boundary of the lacZ mRNA expression domain was observed between cycles 11 and 12 (Figure 4), with the lacZ expression domain in cycle 12 embryos positioned ∼10% more posteriorly, on average, compared to cycle 11 embryos. Posterior progression was observed only until cycle 13, probably due to degradation of bcd mRNA. However, the clear shift between cycles 11 and 12 is consistent with the proposal that the Bcd profile is still far from its steady state at the relevant timeframe for decoding.
Subdivision of the early Drosophila embryo into distinct domains of gap gene expression is arguably the best-studied paradigm of morphogen-induced patterning. Despite extensive investigation, however, quantitative properties of this system have proven difficult to explain, prompting the proposition that additional yet unknown molecules or mechanisms are yet to be identified. The lower-than-expected sensitivity of the pattern to bcd gene dosage is one such mystery, noted repeatedly in studies characterizing the Bcd gradient [23,32,37]. Our study shows, however, that this property can be readily explained within the known framework of the gap gene expression network, by assuming that gap genes begin to be expressed before the Bcd profile has reached its steady state.
More generally, we have shown that pre-steady-state decoding of morphogen gradients can enhance robustness to changes in the rate of morphogen production. This result was derived within the canonical model of morphogen gradient formation, assuming that no feedback mechanisms exist that alter the diffusion or degradation of the morphogen molecules. In previous attempts to explain robustness of patterning, we and others have focused on the steady-state distribution, describing feedback mechanisms that reduce gene dosage sensitivity [12–21]. For example, we have shown that self-enhanced degradation can ensure high robustness with respect to fluctuations in morphogen production rate [15]. However, such feedback mechanisms typically rely on lengthy transcription or translation, and their applicability to early development, where patterning is rapid, is questionable. Pre-steady-state decoding may provide a compelling alternative for increasing robustness, without the need for any explicit feedbacks.
Pre-steady-state decoding assumes that cell fates are defined before the morphogen profile has reached its steady state. The time to reach steady state is controlled by the morphogen decay time τ. In particular, enhanced robustness will be apparent for times that are lower, or comparable to τ. For example, the experimentally measured shifts of gap gene expression domains are consistent with a decoding time tgap that is equal to this decay time, tgap = τ. Interestingly, since the rate of convergence to steady state at a particular position x increases with the distance to the source, anterior regions will appear rather close to their steady state even at t = τ. For example, within our simple model (Equation 2) for t = τ the profile at x = λ /2 (∼50 μm) has already exceeded 75% of its steady-state value, whereas at x = 2λ (∼200 mm) it is still below 40% of its final value (see Figure S1).
Thus, pre-steady-state decoding will be valid if the Bcd degradation time is not shorter than ∼60–90 min, the time when gap gene expression is first observed [31]. Although the Bcd degradation time was not yet measured, a lower bound for this time can be estimated using the measured Bcd profile at cycle 14, which has a decay length of λ ∼ 100 μm. A steady-state profile that extends to this range requires a decay time τ = λ2/D ∼ 170/D minutes, where D is the Bcd diffusion constant (in units of μm2/s). It was recently reported that biologically inert Dextran molecules with a molecular mass comparable to that of Bcd, diffuse quite rapidly in the early Drosophila embryo, with D ∼ 17 μm2/s [57]. This value, if applicable also to Bcd, would imply a short decay time of ∼10 min, such that a steady-state profile is reached before decoding. However, biologically active molecules are known to diffuse at significantly slower rates within cellular environments, with measurements consistently finding diffusion constants in the range of 0.3–3 μm2/s [58–63]. Such values are not consistent with steady-state decoding. They provide a lower bound of 1–10 h for the Bcd decay time, and strongly support the notion of pre-steady-state decoding.
A key issue in pre-steady-state decoding is the definition and execution of a distinct time of decoding. Crucial questions at the mechanistic level are how the decoding time is defined and whether decoding is executed simultaneously in all parts of the embryo. One possibility is that the profile simply does not reach a steady state during the relevant developmental window of morphogen production and signaling. Since development is an ongoing dynamic process, the response to any given morphogen signal is limited to a specific time window, independently of whether the profile has reached its steady state or not. Accordingly, cell fate determination often involves an irreversible transition (commitment), rendering gene expression independent of the inducing signal. A second, conceptually related possibility is that gene expression is determined during the expansion of the morphogen profile, and loses its sensitivity to further changes in this profile. A recent model of neural tube patterning in vertebrates described such a mechanism, showing that a sharp boundary of gene expression is elicited early on in the evolution of the Shh gradient. Self-reinforcing interactions maintain the boundary spatial position even as the Shh gradient itself evolves, moving past the location where the boundary was initially specified [9].
In the case of the gap gene expression domains, the repression between adjacent gap genes can function to stabilize the pattern once it is formed. The gap genes first begin to be expressed at cycles 9–10, when all nuclei reach the periphery, capturing the early Bcd profile. Once gap genes expression is initiated, however, mutual repression between adjacent gap genes stabilizes their spatial expression pattern, and it remains fixed despite further evolution of the Bcd profile. Consistent with this scenario, Yucel and Small have recently argued that measurements of the Bcd gradient during late blastoderm stage (cycle 14) may not accurately reflect the shape of the gradient that defines the gap gene expression domains [47]. Indeed, bcd mRNA starts to degrade at cycle 12 [55], and its protein levels begin to fade [24,55]. Moreover, at cycle 14, gene expression becomes a combination of Bcd-dependent and Bcd-independent activation [47].
The increasing number of nuclei could also function to slow down Bcd diffusion and maintain its pre-steady-state profile. Bcd synthesis is initiated at egg lay, when the embryo consists of a single cell with a single nuclei, and continues throughout the initial set of 14 syncytial nuclear divisions, which increase the number of nuclei to ∼6,000. Interaction of Bcd with the nuclei, or with the cytoplasm surrounding the nuclei, can function to slow down its kinetics, effectively scaling the time to reach steady-state with the number of nuclei (see Protocol S1). Importantly, this apparent stabilization changes the effective time scale of the profile evolution, but does not alter its pre-steady-state characteristics.
By showing that the relative insensitivity of the gap gene expression domains to bcd dosage can be readily accounted for by pre-steady-state decoding, our study provides a simple and parsimonious explanation of one of the long-standing mysteries of the Bcd morphogen system. Additional quantitative issues such as the robust scaling with EL [32,64] and the apparent insensitivity to temperature [54] still remain unexplained. These unresolved issues are similarly likely to reveal new features of the Bcd patterning system.
Females bearing the normal two copies of the bcd gene were used as the wild-type strain. The bcd gene dosage was doubled (4 × bcd) in female progeny of a cross between yw females and P(bcd+5+8) males, a strain which harbors two additional, transgenic copies of the bcd gene on the X chromosome [19]. Females heterozygous for a chromosomal deficiency encompassing the bcd locus, and thus bearing only a single copy of the bcd gene (1 × bcd), were derived from a cross between Df(3R)MAP117/TM3 and yw.
Eggs 2–4 h old laid by females bearing various dosages of bcd and crossed to yw males were collected on agar plates at 25 °C. Egg fixation (3.5% formaldehyde) and preparation for immunostaining were according to standard protocols [65]. Immunostaining of fixed embryos was carried out for 16 h at 4 °C using the following primary antibodies: guinea pig anti-Hb (diluted 1:300), guinea pig anti-Kr (1:300), rabbit anti-Gt (1:500), and rabbit anti-Eve (1:500). Secondary antibodies (Cy3-anti-guinea pig, Cy2-anti-rabbit, and Cy3-anti-rabbit; Jackson Laboratories, http://www.jax.org) were applied for 2 h at room temperature. Double stainings were performed simultaneously, except for Gt/Eve, which was performed sequentially, since both primary antibodies are derived from rabbits.
Images were obtained using a Bio-Rad Laboratories MRC-1024 confocal system (http://www.bio-rad.com), utilizing an argon-krypton mixed-gas laser and mounted on a Zeiss Axiovert microscope (http://www.zeiss.com). Analysis was restricted to cellularizing embryos displaying a distinct seven-stripe Eve pattern. Images were imported and analyzed using software programmed in Matlab (MathWorks, http://www.mathworks.com).
A fragment containing three copies of a 123 bp Bcd-responsive element from the hb promoter (hb123x3-lacZ; [28]) was inserted into the pH-Pelican lacZ reporter [66] and used to generate transgenic lines. To maximize the signal, a fly line homozygous for a chromosome carrying two insertions of the hb123x3-lacZ construct was used. To detect expression of the reporter construct, RNA in situ hybridization was performed with a lacZ RNA probe according to the protocol specified in http://superfly.ucsd.edu/∼davek/intro.html, with hybridization temperature of 65 °C. In addition, a 1:50 dilution of NBT/BCIP (catalog number 1,681,451; Roche, http://www.roche.com) was used as a substrate for alkaline phosphatase. To detect the nuclei, embryos were then incubated for 30 min with a 1:200 dilution of the nuclear dye TOPRO (Molecular Probes, http://probes.invitrogen.com) following treatment with RNAse (10 μg/ml for 30 min). To dynamically monitor the Bcd gradient, the location of the transition point from the signal zone to nonsignal zone of the lacZ gradient was measured in embryos of different ages with different nuclear densities. Microscopic images were obtained using a Bio-Rad Radiance 2100 confocal system. A transmitted light image of the lacZ gradient was obtained in the midfocal plane of a given embryo, and nuclei were viewed by fluorescence of the TOPRO dye. The two corresponding sets of images were analyzed with automated image processing tools developed in MatLab in order to measure the lacZ transition point and the nuclear density as proxy for the embryo's age (Protocol S1). Staining was first detected in embryos at cycle 11, and its intensity increased with embryo stage. We were worried that the increase in staining intensity could lead to the apparent shift in the transition point. This could be the case, for example, if the reporter expression boundaries remained in fact unchanged between different stages, but the minimal expression level cannot be detected at early stages due to low staining intensity. In this case the transition point (determined at half value between minimal and maximal intensity) would actually move posteriorly with increasing total intensity. To control for this possibility, we measured also the spatial distance over which the profile decayed from 80% to 20%. If the changes in transition points are due to changes in staining intensity, the distance over which the profile decays is also expected to increase with staining intensity. In contrast, we observed that the distance over which the profile decays in fact decreases with increasing staining intensity. Thus, it is unlikely that the shift in transition point is due to lower staining intensity at earlier times.
|
10.1371/journal.ppat.1003547 | Local CD4 and CD8 T-Cell Reactivity to HSV-1 Antigens Documents Broad Viral Protein Expression and Immune Competence in Latently Infected Human Trigeminal Ganglia | Herpes simplex virus type 1 (HSV-1) infection results in lifelong chronic infection of trigeminal ganglion (TG) neurons, also referred to as neuronal HSV-1 latency, with periodic reactivation leading to recrudescent herpetic disease in some persons. HSV-1 proteins are expressed in a temporally coordinated fashion during lytic infection, but their expression pattern during latent infection is largely unknown. Selective retention of HSV-1 reactive T-cells in human TG suggests their role in controlling reactivation by recognizing locally expressed HSV-1 proteins. We characterized the HSV-1 proteins recognized by virus-specific CD4 and CD8 T-cells recovered from human HSV-1–infected TG. T-cell clusters, consisting of both CD4 and CD8 T-cells, surrounded neurons and expressed mRNAs and proteins consistent with in situ antigen recognition and antiviral function. HSV-1 proteome-wide scans revealed that intra-TG T-cell responses included both CD4 and CD8 T-cells directed to one to three HSV-1 proteins per person. HSV-1 protein ICP6 was targeted by CD8 T-cells in 4 of 8 HLA-discordant donors. In situ tetramer staining demonstrated HSV-1-specific CD8 T-cells juxtaposed to TG neurons. Intra-TG retention of virus-specific CD4 T-cells, validated to the HSV-1 peptide level, implies trafficking of viral proteins from neurons to HLA class II-expressing non-neuronal cells for antigen presentation. The diversity of viral proteins targeted by TG T-cells across all kinetic and functional classes of viral proteins suggests broad HSV-1 protein expression, and viral antigen processing and presentation, in latently infected human TG. Collectively, the human TG represents an immunocompetent environment for both CD4 and CD8 T-cell recognition of HSV-1 proteins expressed during latent infection. HSV-1 proteins recognized by TG-resident T-cells, particularly ICP6 and VP16, are potential HSV-1 vaccine candidates.
| HSV-1 is an endemic human herpesvirus worldwide that establishes a lifelong latent infection of neurons in the trigeminal ganglion (TG), allowing intermittent reactivation resulting in recurrent disease in some persons. Studies in HSV-1 models suggest a central role of TG-infiltrating virus-specific CD8 T-cells to control reactivation. In humans, however, the functional properties and fine specificity of intra-TG T-cell responses remain enigmatic. The current study used molecular, immunological and in situ analysis platforms on human cadaveric TG obtained within hours after death to characterize the local HSV-1 specific T-cell response in latently infected human TG in detail. We identified that CD4 and CD8 T-cells were juxtaposed to TG neurons and expressed host transcripts and proteins consistent with in situ antigen recognition and antiviral function. The intra-TG T-cell response, involving both CD4 and CD8 T-cells, was directed to a limited set of HSV-1 proteins per person, which was not limited to a specific kinetic or structural class of viral proteins. Collectively, the data indicate that the human TG is an immunocompetent environment for CD4 and CD8 T-cell recognition of diverse HSV-1 proteins expressed during latent infection and that the viral antigens identified herein are rational candidates for HSV-1 subunit vaccines.
| The neurotropic human alphaherpesvirus herpes simplex virus type 1 (HSV-1) is endemic worldwide. It is acquired during early childhood via the orofacial route resulting in a lifelong chronic infection of neurons, also referred to as neuronal HSV-1 latency, in the bilateral trigeminal ganglia (TG) [1]. During latency no infectious virus is produced, virus transcription is mainly directed to latency-associated transcripts (LATs) and microRNAs, and HSV-1 proteins are undetectable using standard methods [2]–[7]. Latent HSV-1 periodically reactivates, producing infectious virus that may lead to recrudescent lesions in some persons. Both primary and recurrent disease can result in clinical disorders of variable severity or even death, emphasizing the unmet need for preventive and therapeutic vaccines [1]. The candidate HSV subunit vaccines, based on the HSV glycoproteins B (gB) and gD were tested in human phase III trials, but were not effective [8]–[10]. Vaccines induced antigen-specific antibodies and CD4 T-cells, but not CD8 T-cells, arguing for novel vaccine formulations that include specific HSV-1 antigens targeted by both antibodies as well as CD4 and CD8 T-cells [11].
Studies in humans and HSV-1 mouse models suggest a pivotal role for virus-specific CD8 T-cells in the control HSV-1 reactivation. Virus-specific CD8 T-cells, expressing an activated effector memory T-cell phenotype, are selectively retained in HSV-1–infected ganglia [5], [12]–[15]. In the HSV-1 mouse model with a C57BL/6 background, the HSV-specific intra-TG CD8 T-cells inhibit HSV-1 reactivation by secreting interferon-γ (IFN-γ) and granzyme B (grB), and are mainly directed against an immunodominant HSV-1 gB epitope [16]–[19]. In nature, however, HSV-1 only infects humans. Because HSV-1 infections in mice mimic but are not equivalent to human disease, it is important that findings from mouse models are confirmed and extended to humans [20]–[22]. Moreover, the HSV-1 antigens recognized by human TG-infiltrating T-cells are rational candidates for HSV-1 subunit vaccines.
HSV-1 encodes at least 77 proteins that during lytic infection are sequentially expressed in a coordinated fashion as immediate early (α), early (β), leaky late (γ1) and true late proteins (γ2) [1], [23]. Expression of γ2 proteins depends on viral DNA replication. While infectious virions eventually assemble in distal axonal structures after reactivation, the temporal expression and trafficking of HSV-1 proteins in human neurons during latency is largely unknown. We previously showed reactivity of human TG-derived CD4 and CD8 T-cells to whole HSV-1 [15], but not which proteins were susceptible to local immune recognition. The detection of transcripts encoding the HSV-1 α proteins infecting cell polypeptide 0 (ICP0) and ICP4 in human TG suggests that this kinetic class of proteins is expressed during latency or early after reactivation [5], [7]. However, their accessibility to antigen processing and presentation within TG-resident cells for local T-cell surveillance is unclear. The aims of this study were to characterize the functional properties and HSV-1 antigens recognized by T-cells in HSV-1–infected human TG.
In contrast to the HSV-1 mouse models, human TG are commonly co-infected with HSV-1 and the closely related neurotropic human alphaherpesvirus varicella-zoster virus (VZV) [13], [15], [20], [24]. To gain insight into the functional properties of human TG-residing T-cells in relation to latent HSV-1 and VZV we determined the transcript levels of the T-cell cytolytic effector molecules perforin and grB, and the cytokines IFN-γ and tumor necrosis factor-α (TNF-α) in 26 TG of 16 donors by reverse transcriptase real-time PCR [25]. We first determined the prevalence and viral load of latent HSV-1 and VZV by real-time PCR. HSV-1 and VZV DNA was detectable in 17 (65%) and 23 (89%) of the 26 TG analyzed, respectively. Sixteen TG contained DNA of both viruses, seven TG had only VZV, one TG had only HSV-1 and two TG had no detectable DNA of either virus. The presence of virus-specific DNA correlated with the donor's HSV-1 and VZV serostatus and was commonly detected in both TG of each individual (data not shown). Consistent with previous reports, the mean number ± standard error of the mean (SEM) of HSV-1 genome equivalent copies per 105 TG cells (1,850±427) was significantly higher compared to VZV (693±184) (p = 0.015) (Figure 1A) [15], [24]. The intra-TG HSV-1 and VZV DNA load did not correlate in paired analysis (Figure 1B). Transcription of the CD8 T-cell–specific gene CD8β correlated weakly with the intra-TG HSV-1 (p = 0.005), but not the VZV DNA load (p = 0.19) (Figure 1C). Next, the expression levels of the T cell transcripts perforin, grB, IFN-γ and TNF-α were compared to the intra-TG HSV-1 and VZV burden (Fig. 1E and F). A weak correlation was observed between both perforin (p = 0.04) and grB (p = 0.004) mRNA levels and the intra-TG HSV-1 DNA load, but not with the latent VZV burden (Figure 1D and 1E). Furthermore, the perforin (p<0.0001), grB (p<0.0001), IFN-γ (p<0.0001) and TNF-α mRNA levels (p = 0.003) correlated strongly with CD8β mRNA levels (Figure 1F) [5]. The data suggest that the extent of CD8 T-cell infiltration in human TG is not only specifically correlated with the latent HSV-1 burden, but is also transcriptionally active to orchestrate an anti-viral function in situ. Finally, the mRNA levels of CD8β (p = 0.003), perforin (p = 0.04), grB (p = 0.01), IFN-γ (p<0.0001) and TNF-α (p<0.0005) correlated strongly between the paired left and right TG indicating that the intra-TG T-cell responses are symmetric intra-individually (Figure 1G) [15].
Compared to HSV-1 negative human TG (Figure S1A), HSV-1 DNA positive human TG are more densely infiltrated with T-cells (Figures S1B), express significantly higher CD8β mRNA levels (Figure S1C) and contain T-cells that on occasion cluster near sensory neuron cell bodies (Figure 2A) [7], [13], [15]. Studies on HSV-1 latently infected sensory ganglia in humans, and particularly in experimentally infected mice, suggest the active role of neuron-interacting T-cells to control neuronal HSV-1 latency [5], [7], [12]–[17]. We first analyzed the presence of TG-infiltrating CD4 and CD8 T-cells by flow cytometry on single cell suspensions of 15 TG of 8 HSV-1 IgG seropositive donors. The data demonstrated infiltration of equivalent numbers of CD4 and CD8 T-cells, with a median ratio of CD4 and CD8 T-cells of 0.99 (range 0.01 to 9.32), which also correlated between the paired left and right TG (p<0.0002). Analogous flow cytometric analysis of a TG of one HSV-1 seronegative donor demonstrated a CD4/CD8 T-cell ratio of 1.1, which resembled that of the HSV-1 seropositive donors. However, the limited number of HSV-1 seronegative donors subjected to ex vivo flow cytometry analyses withhold conclusions to be drawn on the virus-specific role of retention of either T-cell subtype in human alphaherpesvirus latently infected TG.
Subsequently, we aimed to corroborate the potential protective role of neuron-interacting T-cell clusters by performing detailed in situ analyses on HSV-1 latently infected human TG. Neuron-interacting T-cell clusters consisted of both CD4 and CD8 T-cells (Figure 2B). CD8 T-cells expressed both grB and the T-cell intercellular antigen-1 (TIA-1) consistent with their cytotoxic potential (Fig. 2B) [5], [15]. We have recently shown that CD137, a TNF receptor family member [26], is induced on HSV-1 reactive human CD4 and CD8 T-cells shortly after recognition of HSV in vitro [27], [28]. Here, we demonstrated that neuron-interacting T-cells in HSV-1 latently infected human TG express CD137 in situ, implying that they have encountered their cognate antigen locally (Fig. 2B).
To identify the viral proteins recognized by human TG residing T-cells, T-cell lines (TCL) were generated by mitogenic stimulation of TG-derived T-cells from twelve HSV-1 IgG seropositive donors. The HSV-1–specific T-cells were phenotyped and enumerated by a flow cytometric intra-cellular IFN-γ (IFN-γ ICC) assay using mock– and HSV-1–infected autologous Epstein-Barr virus transformed B-cell lines (BLCL) as antigen-presenting cells (APC). The median percentage of HSV-1–specific CD8 T-cells in the TG-TCL was 10% (range 2 to 37%) and 4 of 12 TG-TCL also contained HSV-1-specific CD4 T-cells (range 0.3 to 10%) (Table 1).
The HSV-1 proteins recognized by human TG-derived CD8 T-cells were determined using transfected Cos-7 cells as artificial APC that expressed one of the donor's HLA-A and –B alleles in combination with 74 separate HSV-1 open reading frames (ORFs) [27]. First, we used a set of partially HLA–A or –B allele matched HSV-1-infected BLCL as APC to uncover both the diversity and identity of HSV-1 peptide-presenting HLA class I (HLA-I) alleles used by the CD8 T-cells. The data demonstrated that the virus-specific intra-TG CD8 T-cell response is mediated by 1 to 4 different HLA–A and –B alleles per person (Table 1). Next, we used the implicated HLA-A and -B allele in HSV-1 ORFeome-wide screen [27]. We observed definitive HSV-1 ORFeome screen hits only when the net proportion of CD8 T-cells reactive with HLA-matched HSV-1-infected BLCL was >4% (data not shown). For some donors, we enriched and expanded the TG-TCL using CD137 selection [27], [28]. For this, TG-TCL were incubated with HSV-1–infected autologous BLCL and CD137+ CD8 T-cells were selected after 18 hours of incubation and expanded with a T-cell mitogen to generate a second generation TG-TCL. This led to an approximately 3-fold enrichment of HSV-1 reactive CD8 T-cells for donors TG1, TG4 and TG12. The TG-TCL of donors TG8, TG9 and TG11 yielded insufficient enrichment or ample T-cell numbers to perform HSV-1 ORFeome screens (data not shown).
In total, 8 of 12 HSV-1 reactive TG-TCL revealed reproducible specific HSV-1 CD8 T-cell antigen hits (Figure 3). Thirteen different CD8 T-cell viral targets were identified with 1 to 3 viral proteins per TG-TCL (Figure 3). Virion protein 16 (VP16) and particularly ICP6 were recognized by multiple TG-TCL in the context of diverse HLA-A and –B alleles. In case of ICP6, 4 of 8 TG-TCL were positive and the protein was recognized via HLA-A*3101 (donor TG7), –B*1501 (donor TG2) and in 3 different TG donors via HLA-B*4001 (donors TG5, TG7 and TG12) (Figure 3). Finally, candidate CD8 T-cell epitopes within several HSV-1 screen hit proteins were predicted by in silico algorithms [29]. Epitopes were subsequently validated by IFN-γ ICC using corresponding synthetic peptides and HLA-matched BLCL as APC (Figures S2 and S3 and Table 2).
We recently studied antigenic targets of blood-derived HSV-1 specific CD8 T-cells in HSV-1 IgG seropositive healthy subjects using related methodologies [27]. Systemic HLA–A and –B restricted CD8 T-cell responses were directed to 14 HSV-1 ORFs on average per person and 45 HLA–A and –B allele restricted HSV-1 epitopes were identified [27]. To discern potential similarities between systemic and intra-TG HSV-1 peptide specific CD8 T-cell responses we tested TG-TCL of seven HLA-A and -B allele matched TG donors for responses to the HLA-appropriate HSV-1 peptides from our previous work on blood-derived T-cells [27] (Table S1). Peptide-specific CD8 T-cell responses were detected in two TG-TCL. The TG-TCL of donor TG3 recognized four HLA-A*0101–restricted peptides: gL66–74, gK201–209, and two VP16 peptides VP1690–99 and VP16479–488. The HLA-A*2902–restricted VP13/14508–516 peptide was recognized by the TG-TCL of donor TG6 (Table 2 and Table S1) [27].
Collectively, the data demonstrated that human intra-TG HSV-1–specific CD8 T-cell responses were directed to a relatively restricted number of viral proteins per person. However, even within the small population studied, we detected CD8 T-cell responses to HSV-1 proteins in diverse kinetic and structural classes (Table S2). Notably, the HSV-1 β protein ICP6 was a prominent CD8 T-cell target in TG-TCL of 4 of 8 HLA-discordant TG donors involving 3 different HLA-I alleles.
The intra-TG CD4 T-cell responses were analyzed in detail for donors TG2 and TG3 (Table 1) [28]. CD4 T-cells of donor TG2 responded to the HSV-1 α protein ICP47 and subsequent assays using whole ICP47-spanning peptides defined the antigenic region at residues 57–75 (Figure 4A and 4B). For donor TG3, CD4 T-cell reactivity was directed to the HSV-1 γ1 protein VP16 (Figure 4A). Application of truncated recombinant VP16 proteins and subsequently overlapping peptides identified two distinct antigenic regions located between residues 187–203 and 215–238 (Figure 4C). Besides being a structural viral protein, VP16 has also been implicated as a master initiator protein for HSV-1 neuronal reactivation in mice [30].
The symmetry of the virus and T-cell parameters between paired TG (Figure 1) [15], facilitated studies on the spatial orientation of HSV-1 reactive CD8 T-cells in the contralateral snap-frozen TG specimen of the same donor by in situ tetramer staining (Figure 5) [31]. HSV-1 CD8 T-cell epitopes and corresponding snap-frozen contralateral TG specimens were available for donors TG2 and TG3. HLA-A*0201 tetramers conjugated with identified ICP0642–651 and ICP81096–1105 epitopes and HLA-A*0101 reagents with gL66–74, gK201–209, VP1690–99 and VP16479–488 epitopes (Table 2), were validated on the corresponding TG-TCL (Figures S2 and S3). Whereas control TG of HLA-A mismatched HSV-1 seropositive donors did not reveal tetramer-positive CD8 T-cells (data not shown), HSV-1 tetramer-positive CD8 T-cells were found juxtaposed to neuronal cell bodies in TG of the respective donor (Figures 5 and S4, and Movie S1).
The host-pathogen standoff in human latent HSV-1 infection permits periodic epithelial shedding of infectious virus, and potential transmission, without overt host damage. In contrast, HSV-1 mouse models are either acutely fatal or demonstrate tight neuronal latency in which neither spontaneous nor stress-induced ganglionic reactivation leads to peripheral release of infectious virus [17], [20], [32]. In the current study, we demonstrated that the human TG is an immunocompetent organ capable of presenting viral antigens to both CD4 and CD8 T-cells, presumably over long periods of time, to maintain local enrichment of HSV-1–specific T-cells. The data imply that the HSV-1 proteins expressed in latently infected human TG are not limited to a specific class of kinetic or structural viral proteins and that the viral antigens identified herein are rational candidates for HSV-1 subunit vaccines.
In contrast to viral DNA and transcripts, viral proteins have not been detected in HSV-1 latently infected human ganglia [5], [7], [13]. Viral protein synthesis may be shutdown or occur at low levels or at low frequency. T-cells are activated by only a few MHC/peptide-complexes making them highly sensitive and specific biosensors to detect extremely low-level expression of their cognate antigens [33]. The recognition of diverse HSV-1 proteins by human TG infiltrating T-cells implies their cognate antigen expression in situ. Moreover, the HSV-1 targets identified did not group to a specific kinetic or functional class of viral proteins suggesting that HSV-1 protein synthesis is not limited to early viral proteins in latently infected human TG [5], [7]. Alternatively, the T-cell response reported herein may be directed to local reactivating HSV-1, which has overcome cellular and viral control including HSV-1 micro RNAs [6], [7]. In contrast to humans, HSV-1 mouse models are either fatal or have tight neuronal latency in which spontaneous reactivation does not lead to peripheral release of infectious virus [20], [21]. Nevertheless, latently infected TG of C57BL/6 mice contain neuron-interacting CD8 T-cells, directed to 11 viral proteins including late structural HSV-1 proteins like gC and gK [12], [18], [19]. These data imply that full HSV-1 reactivation is not a prerequisite to retain virus-specific T-cells in ganglia with diverse viral protein reactivity. The combined human and mouse data argue that this process involves recognition of the T-cells' cognate viral antigens produced locally in HSV-1 latently infected ganglia [5], [15], [18], [20].
Activation of HSV-1-specific CD8 T-cells in latently infected murine ganglia is dependent on local CD4 T-cells, MHC class II expression and recruited blood-derived APC [14], [34]. The current study documents inclusion of CD4 T-cells in neuron-interacting T-cell clusters and proves peptide-level recognition of HSV-1 by TG-resident CD4 T-cells in the natural host. Intriguingly, VP16 was targeted by both CD8 and CD4 T-cells in the same TG specimen (i.e., TG3) indicating local expression and presentation of this viral protein by TG cells in the context of both HLA-I and HLA-II molecules. Although local human APC driving the HSV-1-specific CD4 T-cell responses could be blood-derived or ganglion-resident [14], [34], it is unlikely that HLA class II negative neurons are directly involved. The data suggest of HSV-1 proteins or remnants thereof from neurons to secondary APC. We have recently shown that satellite glial cells (SGC), which tightly envelop neuronal cell bodies in ganglia, are most likely of myeloid origin [35]. Human TG-resident SGC are related to macrophages and myeloid dendritic cells with regards to their phagocytic capacity and expression of CD45, co-stimulatory and HLA class II molecules [23]. Given their localization and phenotype, SGC are candidate APC to create an immunocompetent but not overtly inflammatory environment to support HSV-specific CD4 and CD8 T-cell responses within latently infected human TG.
An important and still unanswered question is the functional role of the HSV-specific T-cells documented in this report. In the absence of tools to selectively interrupt or bolster T-cells at specific anatomic sites in humans, this question is difficult to address. Surrogate data can be obtained by examining the phenotype and activation status of human TG-resident CD8 T-cells. Integrating human TG ex vivo flow cytometry and in situ data, it is evident that TG-resident CD8 T-cells express CD137 and grB, and low levels of CD27 and CD28, indicative of recent antigen encounter locally [5], [12], [14], [15], [36]. Messenger RNA expression of the T-cell cytolytic molecules perforin and grB directly correlated with HSV-1 DNA levels (Figure 1D) and grB and TIA-1 protein expression co-localized with neuron-surrounding CD8 T-cells (Figure 2B). The fact that human TG-resident HSV-1-specific CD8 T-cells can massively expand in vitro and then display brisk virus-specific IFN-γ responses argues against an exhausted phenotype [15], [37]. Taken together with the remarkable localization of HSV-1-specific CD8 T-cells juxtaposed to TG neurons (Figures 5 and S4 and Movie S1), our data argue for a functional role for these cells in non-lytic control of HSV-1 infection in human TG in cooperation with local CD4 T-cells.
If this interpretation is correct, elicitation of T-cells with anti-viral activity in the latently infected TG is a rational goal for preventative and therapeutic HSV-1 vaccines. Our findings have several implications for HSV-1 vaccine design. First, ICP6 was recognized by 4 of 8 TG donors in diverse HLA-I contexts (Table 2). ICP6 is a ribonucleotide reductase subunit expressed prior to viral DNA replication [1]. Because ICP6 was also a dominant target for the systemic CD8 T-cell response in HSV-1 seropositive subjects [27], this protein is an attractive vaccine candidate. Second, HSV-1 proteins of diverse kinetic and structural classes were recognized by TG CD8 T-cells (Table 2). These range from nonstructural α (ICP0 and ICP4) and β proteins (ICP6 and thymidine kinase) to the late structural tegument (VP11/12 and VP13/14) and envelope glycoproteins (gB, gK and gL). Tegument protein VP16, recognized by both CD4 and CD8 TG-resident T-cells, is possibly a chameleon with both a hyper-early role in neuronal reactivation and a structural role in tegument assembly [30]. The cell biology implication of this finding appears that diverse HSV-1 proteins are diverted from viral assembly and access the HLA class I pathway in neurons, or possibly surrounding APC after handover. Third, the apparent diversity of recognized HSV-1 antigens is lower in TG than in blood, where we detected a mean of 14 reactive HSV-1 ORFs per person using similar technology [27]. The restricted clonality of the human intra-TG T-cell response is consistent with T-cell receptor spectratyping [5], [15], [36]. Surveys of more participants, ideally with parallel studies on blood-derived T-cells, are mandatory to determine if the breadth or fine specificity of the paired TG and systemic HSV-1 T-cell responses differ and to pick the best antigens for possible subunit approaches that target sensory ganglia as a site of viral control. Manipulation of T-cell priming or boosting to imprint a TG-homing program via vaccination, without imparting an overly aggressive phenotype, is an equally important and challenging task that must be overcome to target the TG as an immunocompetent site for the purpose of HSV-1 latency control.
Heparinized peripheral blood and paired TG specimens were obtained from 39 individuals (median age 71 yrs, range 49–98 yrs) at autopsy with a median post-mortem interval of 6.3 hrs (range 2.3–11.3 hrs). Specimens were collected by the Netherlands Brain Bank (Netherlands Institute for Neuroscience; Amsterdam, the Netherlands) from donors from whom a written informed consent for brain autopsy and the use of the material and clinical information for research purposes had been obtained. All study procedures were performed in compliance with relevant Dutch laws and institutional guidelines, approved by the local ethical committee (VU University Medical Center; Amsterdam, project number 2009/148) and was performed in accordance with the ethical standards of the Declaration of Helsinki. The majority of the TG donors (n = 31) had a neurologic disease history affecting the central nervous system (mainly Alzheimer's and Parkinson's disease). Causes of death were not related to herpesvirus infections. Blood was used to generate BLCL and for HLA typing as described [15], [27]. Plasma HSV-1 and VZV IgG levels were determined by ELISA (Focus Diagnostics).
TG-TCL were generated by phytohemagglutinin (PHA) stimulation of TG cell suspensions, or of CD137-enriched TG-TCL, using γ-irradiated allogeneic PBMC and recombinant human IL-2 as described [15]. Antigen-specificity of TG-TCL was determined by IFN-γ ICC using the following APC: autologous or partially HLA class I-matched BLCL infected overnight with HSV-1 with a multiplicity of infection (MOI) of 10, or BLCL pulsed with 2 µM of HSV-1 peptides [27]. Mock-infected BLCL were used as negative controls. Cells were stained for CD4, CD8, CD3 and IFN-γ (all from Becton Dickinson; BD) and analyzed by multicolor flow cytometry with Diva software (BD) as described [15]. Ex vivo flow cytometry analyses for CD3, CD4, CD8 expression was performed on single TG cell suspensions of a subset of 16 TG specimens as described [15].
One-fifth of a dispersed TG cell suspension was used for RNA and DNA isolation [15]. RNA was reverse transcribed using an oligo-dT primer and used for quantitative real-time PCR (qPCR) on an ABI Prism 7700 with Taqman Universal Master Mix and commercial intron-spanning primer/probe-pairs specific for human perforin, grB, CD8β, TNF-α, IFN-γ and β-actin (Applied Biosystems) per manufacturer. The relative transcript levels were determined by the formula 1,000×2(−delCt), where delCt equals Ct [(target gene) - Ct (β-actin)]. Intra-TG HSV-1 and VZV DNA load were determined by qPCR as described [25].
To enrich HSV-1 reactive CD8 T-cells, autologous BLCL were infected overnight with HSV-1 with a MOI of 10. TG-TCL were added to the APC at a ratio of 1∶1 for the next 24 hrs. Cells were harvested, stained for CD3 (BD), CD8 (BD) and CD137 (Miltenyi). Cells that co-expressed CD3, CD8, and CD137 were enriched with a BD FACS Aria cell sorter, expanded by PHA stimulation and used in HSV-1 ORFeome screens as described [27], [28].
The generation and validation of the HSV-1 ORFeome, covering a total of 74 HSV-1 ORFs, for functional T-cell assays has been detailed elsewhere [27]. In short, each HSV-1 ORF was amplified and cloned into a custom-made eukaryotic expression vector fused to the gene encoding enhanced green fluorescent protein (eGFP). Donor-matched HLA-I specific cDNA (in pcDNA3) and HSV-1 ORFs were expressed in Cos-7 cells (ATCC CRL-1651) by transfection [27]. All HSV-1 ORFs were transfected in duplicate and appropriate mock- or HSV-1 infection controls were included. After 48 hrs, ORF expression was confirmed by eGFP fluorescence and TG-TCL (5×104/well) were added to 104 transfected Cos-7 cells/well. After 24 h, supernatants were collected for IFN-γ ELISA [27].
Whole HSV-1 ORFeome screens to identify CD4 T-cell target antigens were performed in duplicate as described [28], [38]. Gamma-irradiated HLA-DQ/DR-matched allogeneic PBMC were pulsed overnight with predefined dilutions of protein lysates of HSV-1 ORF-transfected Cos-7 cells, HSV-1 proteins made with bacterial lysates or peptides at 2 µM [28], [38]. Ultraviolet light treated mock- and HSV-1-infected Vero cell lysates were used as negative and positive controls, respectively. After 48 hrs, [3H]-thymidine was added and cells harvested to measure [3H]-thymidine incorporation as marker for T-cell proliferation [27], [38].
In situ immunofluorescence was performed using allophycocyanin (APC)-labeled CD4 (clone RPA-T4; BD) and FITC-labeled CD8 (1A5; Monosan) monocloncal antibodies (mAbs). The APC signal was enhanced by the FASER system per manufacturer (Miltenyi). Sections were post-fixed with 4% (w/v) formaldehyde, counterstained for DNA with DAPI (Invitrogen) and mounted with ProLong Gold Antifade Reagent (Invitrogen). For immunohistochemistry, paraffin sections and cryosections of human TG were stained as described [35]. The mAbs used were directed to CD8 (1A5; Monosan), grB (GrB-7; Dako), TIA-1 (2G9; Immunotech), CD3 (UCHT1; Dako) and CD137 (4B4-1; BD). Sections were counterstained with hematoxylin and mounted with glycerol gelatin. The ratio of CD3+ cells per neuron in TG of HSV-1 serotyped donors was determined by counting all sensory neuronal cell bodies and CD3+ cells in multiple sections (n = 3–5), cut at different anatomic levels of the same TG specimen, under the microscope as described previously [39]. The average number CD3+ cells/neuron per TG are presented.
In situ tetramer stainings were performed as described previously [31]. In brief, TG cryosections (8 µm) were fixed with 4% (w/v) formaldehyde and incubated with 2–4 µg of the respective APC-conjugated HSV-1 peptide/HLA-I tetramers at 4°C for 20 hrs. Next, slides were washed and post-fixed in 4% formaldehyde. Slides were counterstained with anti-CD8 (3B5; Invitrogen) and DAPI (Invitrogen), and mounted with ProLong Gold Antifade Reagent (Invitrogen). Fluorescent images were acquired on a Zeiss LSM700 confocal laser scanning microscope.
Statistical differences between were determined by the Mann-Whitney test, paired T-test, Spearman correlation test and Wilcoxon matched-pairs signed-rank test. P<0.05 were considered significant.
|
10.1371/journal.ppat.1006504 | The RhlR quorum-sensing receptor controls Pseudomonas aeruginosa pathogenesis and biofilm development independently of its canonical homoserine lactone autoinducer | Quorum sensing (QS) is a bacterial cell-to-cell communication process that relies on the production, release, and response to extracellular signaling molecules called autoinducers. QS controls virulence and biofilm formation in the human pathogen Pseudomonas aeruginosa. P. aeruginosa possesses two canonical LuxI/R-type QS systems, LasI/R and RhlI/R, which produce and detect 3OC12-homoserine lactone and C4-homoserine lactone, respectively. Here, we use biofilm analyses, reporter assays, RNA-seq studies, and animal infection assays to show that RhlR directs both RhlI-dependent and RhlI-independent regulons. In the absence of RhlI, RhlR controls the expression of genes required for biofilm formation as well as genes encoding virulence factors. Consistent with these findings, ΔrhlR and ΔrhlI mutants have radically different biofilm phenotypes and the ΔrhlI mutant displays full virulence in animals whereas the ΔrhlR mutant is attenuated. The ΔrhlI mutant cell-free culture fluids contain an activity that stimulates RhlR-dependent gene expression. We propose a model in which RhlR responds to an alternative ligand, in addition to its canonical C4-homoserine lactone autoinducer. This alternate ligand promotes a RhlR-dependent transcriptional program in the absence of RhlI.
| Quorum sensing (QS) is a cell-to-cell communication process that bacteria use to coordinate group behaviors. QS is essential for virulence and biofilm formation in many bacteria including the human pathogen Pseudomonas aeruginosa. P. aeruginosa has high clinical relevance because it has acquired resistance to commonly used antibiotics, and is a priority pathogen on the CDC ESKAPE pathogen list. The urgent need for new antimicrobials to combat P. aeruginosa infections makes targeting QS for interference an attractive approach. Here, we investigate P. aeruginosa under biofilm conditions that mimic authentic P. aeruginosa lifestyles in environmental and medical contexts rather than in traditional laboratory conditions. This strategy enabled us to find that P. aeruginosa uses a novel QS signal molecule that controls biofilm formation and virulence. The new signal molecule acts together with the long-known QS receptor RhlR. Using physiologic, genetic, and molecular studies, combined with animal models of infection, we characterize the roles of QS components in biofilm formation and virulence. We find that RhlR and the putative new signal molecule are crucial for both traits. Our work suggests that targeting RhlR with small molecule inhibitors could provide an exciting path forward for the development of novel antimicrobials.
| Quorum sensing (QS) is a process of bacterial cell-to-cell communication that relies on the production, detection, and response to extracellular signaling molecules called autoinducers [1]. QS allows groups of bacteria to synchronously alter behavior in response to changes in the population density and species composition of the surrounding bacterial community [2,3]. In Gram-negative bacteria, acylated homoserine lactones (AHLs) are common QS autoinducers (reviewed in [4]). Typically, an AHL synthase, usually a LuxI homolog, produces an autoinducer that is bound by a partner transcriptional activator, usually a LuxR homolog. LuxR-AHL complexes regulate expression of genes that underpin group behaviors [5]. LuxR-type proteins contain two domains: an amino-terminal AHL-binding domain and a carboxy-terminal helix-turn-helix (HTH) DNA-binding domain [6,7]. Most LuxR-type receptors require their cognate AHLs to be bound to function, and in some cases, AHL binding is necessary for LuxR-type proteins to fold and thus resist proteolysis [8–10].
Bacterial pathogens often require QS to establish or to promote infection (reviewed in [11]). One such QS bacterium, Pseudomonas aeruginosa, is a human pathogen that is frequently responsible for hospital-acquired infections and is the main cause of morbidity and mortality in cystic fibrosis patients [12,13]. The P. aeruginosa QS circuit consists of two canonical LuxI/R pairs: LasI/R and RhlI/R (Fig 1) [14–17]. LasI produces and LasR responds to the autoinducer N-(3-oxododecanoyl)-L-homoserine lactone (3OC12-HSL). The LasR:3OC12–HSL complex activates transcription of many genes including rhlR [18]. RhlR binds to the autoinducer N-butanoyl-L-homoserine lactone (C4-HSL), the product of the RhlI synthase [19]. RhlR:C4-HSL directs a large regulon of genes including those encoding virulence factors such as pyocyanin, elastases, and rhamnolipids, some of which are also members of the LasR:3OC12-HSL regulon (Fig 1) [20,21]. P. aeruginosa strains harboring mutations in QS regulatory components have been reported to be attenuated for virulence, and thus, interfering with QS holds promise for the development of novel anti-microbial therapies [22–26].
Beyond controlling virulence, QS controls biofilm formation in P. aeruginosa [26,27]. Biofilm formation is crucial for P. aeruginosa acute and chronic infections [29]. QS-activated genes encoding exoproducts such as the Pel and Psl exopolysaccharides, rhamnolipids, and phenazines are key for biofilms because these products drive the architecture of the developing communities [30–33]. In keeping with this overarching role for QS in biofilm formation, in the laboratory, P. aeruginosa lasR and lasI mutants form defective biofilms that are thin, undifferentiated, and easily eradicated by SDS and antimicrobial treatments [27].
Most research examining the role of QS in P. aeruginosa virulence and biofilm formation has focused on the LasI/R system because of its location at the top of the QS signal transduction cascade (Fig 1). Curiously, however, lasR loss of function mutants arise in P. aeruginosa chronic cystic fibrosis infections [34–36]. Furthermore, QS-controlled virulence traits are expressed in these lasR mutants. A possible clue to this conundrum comes from recent work showing that RhlR, not LasR, is the primary QS regulator during host infection, using Drosophila as a model [37]. Given this ambiguity, we wanted to define the role of RhlR in virulence and biofilm formation in P. aeruginosa. In this study, we show that RhlR, previously reported to have an obligate dependence on its canonical AHL autoinducer C4-HSL, can function as a transcriptional regulator in the absence of C4-HSL. Indeed, we show that ΔrhlR and ΔrhlI mutants have dramatically different biofilm phenotypes. Genome-wide transcriptomics analyses show that RhlR and RhlI likewise control distinct regulons with little overlap under biofilm-forming conditions. We also find that crucial RhlR-regulated virulence factors are expressed in the absence of RhlI. Consistent with this result, the ΔrhlI mutant infects nematode and mouse animal hosts as effectively as wildtype P. aeruginosa, in contrast to the ΔrhlR mutant that is attenuated for virulence. Finally, cell-free culture fluids prepared from the ΔrhlI mutant possess an activity that stimulates RhlR-dependent gene expression. These findings support the hypothesis that RhlR responds to an alternative ligand, in addition to C4-HSL, and this alternative ligand promotes RhlR-dependent transcriptional activation in the absence of RhlI.
To explore the role of the RhlI/R QS system in P. aeruginosa virulence and biofilm formation, we generated in-frame marker-less deletions of the rhlR and rhlI genes in the UCBPP-PA14 strain of P. aeruginosa (hereafter referred to as PA14). We also made the double ΔrhlR ΔrhlI mutant. As expected, deletion of rhlR or rhlI abolished pyocyanin production in planktonic cultures (Fig 2A) [16]. Introduction of a plasmid expressing rhlR under its native promoter complemented pyocyanin production in the ΔrhlR mutant (Fig 2A). Likewise, exogenous addition of 10 μM C4-HSL to the ΔrhlI mutant restored pyocyanin production to wildtype (WT) levels (Fig 2A). Similar results with pyocyanin and other quorum-sensing-controlled phenotypes have been reported previously, including with mutant strains used in the present work [16,38,39,40,41]. These analyses confirm that every component of our system is functional. We note, and we will return to this point later, that similar to other AHL-dependent LuxR-type transcriptional activators, the ΔrhlR and ΔrhlI mutants phenocopy each other with respect to pyocyanin production in planktonic culture.
P. aeruginosa PA14 forms biofilms in submerged systems (flow-cell biofilms), at liquid-air interfaces (pellicles), and on solid-air interfaces (colony biofilms) [30,42]. Colony biofilm formation can be studied using agar plates containing Congo red, a dye that binds to extracellular matrix components [30]. On colony biofilm medium, WT P. aeruginosa PA14 exhibited the characteristic, and previously reported, rugose-center/smooth-periphery colony biofilm phenotype after 5 days of growth (Fig 2B) [31]. To our knowledge, the roles of RhlR and RhlI have not previously been investigated in P. aeruginosa PA14 colony biofilms. We found that the ΔrhlR mutant was hyper-rugose (Fig 2B and S1A Fig) and introduction of rhlR on a plasmid restored the WT morphology (Fig 2B). We conclude that RhlR controls colony biofilm development. Surprisingly, the ΔrhlI mutant had a phenotype that was strikingly different from the WT and the ΔrhlR mutant. The ΔrhlI mutant was completely smooth, indeed, even more so than the WT (Fig 2B; S1A Fig). Exogenous addition of 10 μM C4-HSL to the agar medium restored the WT biofilm phenotype to the ΔrhlI mutant (Fig 2B). The ΔrhlR phenotype is epistatic to the ΔrhlI phenotype because the ΔrhlR ΔrhlI double mutant phenocopies the ΔrhlR single mutant (Fig 2B; S1B Fig). This phenotypic difference also occurred on agar plates lacking any dyes, confirming that the distinct morphologies of the ΔrhlR and ΔrhlI mutants are not a consequence of the Congo red medium (S1C Fig). The ΔrhlR mutant colony biofilms expanded to cover more surface area than did those of the WT and the ΔrhlI mutant (Fig 2C; S1A Fig). These results are in stark contrast to those for the LasI/R system: we constructed marker-less in-frame ΔlasR and ΔlasI mutants and found that both mutants have identical pyocyanin, colony biofilm, and surface coverage phenotypes (S2A, S2B and S2C Fig).
A previous genome-wide small RNA-seq study identified a putative trans-acting sRNA called SPA0104 in P. aeruginosa PA14 [43]. The SPA0104 gene is located between the rhlR and rhlI genes, co-oriented and overlapping with the rhlI promoter [43]. To determine if SPA0104 is involved in the different ΔrhlR and ΔrhlI phenotypes we observed above, we engineered stop codons into the rhlR and rhlI genes: rhlRW11STOP and rhlIF50STOP. Our rationale was that insertion of a stop codon would prevent translation of the full-length RhlR or full length RhlI protein without affecting transcription of SPA0104, enabling us to determine if the SPA0104 sRNA contributed to the biofilm phenotypes. Neither of the mutants produced pyocyanin showing that, in the case of RhlR and RhlI, introduction of the stop codon eliminated function (S2A Fig). Nonetheless, the rhlRW11STOP mutant was hyper-rugose and the rhlIF50STOP mutant was completely smooth in the colony biofilm assay (S2D Fig). We therefore conclude that the ΔrhlR and ΔrhlI mutants have distinct colony biofilm phenotypes and the SPA0104 sRNA plays no role in conferring these phenotypes.
Rhamnolipids, RhlR-activated exoproducts, have been reported to disperse biofilms [44,45]. Thus, it was possible that, in the ΔrhlR mutant, the absence of rhamnolipids decreased dispersal, and this defect in the process conferred the hyper-rugose phenotype. If so, disabling rhamnolipid production should cause a hyper-rugose phenotype irrespective of the presence or absence of RhlR. We investigated this possibility by inactivating the rhamnolipid biosynthetic gene rhlA [46,47] via introduction of a stop codon (rhlAC11STOP). The rhlAC11STOP mutant has a colony biofilm phenotype that is indistinguishable from the WT, so it is not hyper-rugose (S3 Fig). By contrast, the ΔrhlR rhlAC11STOP double mutant is hyper-rugose, and the ΔrhlI rhlAC11STOP double mutant is completely smooth (S3 Fig). We therefore conclude that rhamnolipids are not involved in the distinct colony biofilm phenotypes we have discovered for the ΔrhlR and ΔrhlI mutants.
To define the molecular basis underpinning the different ΔrhlR and ΔrhlI colony biofilm phenotypes, we used RNA-seq to compare the genomic transcriptional profiles of the WT, ΔrhlR, and ΔrhlI strains. We performed the experiment under two conditions: on mRNA harvested from high cell density (HCD) planktonic cultures and on mature colony biofilms. To our knowledge, this is the first transcriptional profiling study performed on the ΔrhlI mutant. The results from HCD planktonic cultures match previously published studies for the WT and the ΔrhlR mutant [18,26]. Roughly 127 genes constitute the RhlR regulon, defined as greater than two-fold changes in expression in the ΔrhlR mutant compared to the WT in HCD planktonic cultures (Fig 3A, S1 Table). Seventy-three of those genes were also regulated by RhlI. Under colony biofilm forming conditions, the ΔrhlR mutant exhibited differences in expression of 137 genes compared to the WT. However, only 18 of those genes showed more than two-fold changes in the ΔrhlI mutant relative to the WT (Fig 3A; S2 Table). Thus, of the RhlR-regulated genes, ~54% were in common between the ΔrhlR and ΔrhlI mutants in HCD planktonic cultures. By contrast, there was only ~13% overlap in the RhlR-directed gene expression profiles in these two mutants in colony biofilms (Fig 3A; S1 and S2 Tables). We note that 9 and 6 genes were uniquely regulated by RhlI under planktonic and colony biofilm formation conditions, respectively. A few autoinducer-dependent, receptor-independent QS-regulated genes have been reported previously for the P. aeruginosa Las system [40,48]. In summary, RhlR regulates the expression of numerous genes independently of its canonical AHL autoinducer C4-HSL.
The most striking difference between the transcriptional profiles of the ΔrhlR and ΔrhlI mutants in planktonic and colony biofilm culture conditions concerned the phenazine (phz) biosynthesis genes (S1 and S2 Tables). Specifically, each phz gene exhibited 10-fold lower expression in both the ΔrhlR and ΔrhlI mutants compared to the WT in HCD planktonic culture (Fig 3B). In colony biofilms, the ΔrhlR mutant continued to exhibit 10-fold lower expression of phz genes compared to WT. However, the ΔrhlI mutant expressed the phz genes at WT levels (Fig 3B).
Endogenously produced phenazines act as redox-active small molecules to modulate P. aeruginosa colony biofilm morphogenesis. Specifically, mutants that overproduce phenazines have smooth morphologies, whereas mutants that are unable to produce phenazines are hyper-rugose compared to the WT [30,31,33]. We suggest that the ΔrhlI mutant has a smooth colony biofilm phenotype due to overproduction of phenazines while the ΔrhlR mutant forms a hyper-rugose biofilm colony due to the lack of phenazines. We verified this idea using two different approaches. First, we deleted the two phenazine biosynthesis operons (Δphz1 and Δphz2; we call this double mutant Δphz) in the WT and in the ΔrhlR and ΔrhlI mutant backgrounds. All of the Δphz mutants were hyper-rugose under colony biofilm growth conditions (Fig 3C). Second, we quantified pyocyanin production from the WT, the ΔrhlR and ΔrhlI single mutants, and the ΔrhlI Δphz double mutant when grown as colony biofilms. The wild-type colony biofilm produced ~ 2 μg /105 CFU pyocyanin, the ΔrhlI colony biofilm produced four-fold more pyocyanin, while the ΔrhlR mutant and the ΔrhlI Δphz double mutant produced none (S4 Fig). Together, the deletion analysis and pyocyanin production results show that the ΔrhlI smooth colony biofilm phenotype is caused by overproduction of phenazines.
In P. aeruginosa PA14, the hyper-rugosity conferred by the absence of phenazines requires Pel, the primary biofilm matrix exopolysaccharide (note: P. aeruginosa PA14 does not produce the Psl exopolysaccharide) [28,30,32]. We examined the colony biofilm phenotypes of the ΔpelA single and the ΔrhlRΔpelA and ΔrhlIΔpelA double mutants. All of these mutants had completely smooth colony morphologies (S5 Fig). Thus, the Pel exopolysaccharide is required for the ΔrhlR mutant to exhibit the hyper-rugose biofilm phenotype. We note that the mechanism by which the overproduction of phenazines downregulates Pel to cause the smooth colony biofilm phenotype is unknown and beyond the scope of this study. What is crucial for our work is that the ΔrhlI mutant retains the ability to produce phenazines in colony biofilms while the ΔrhlR mutant does not.
Close inspection of the RNA-seq data from the HCD planktonic cultures (S1 Table) revealed three classes of RhlR-regulated genes based on their dependence on RhlI: genes which we call Class I, exemplified by chiC, that require RhlI for RhlR-mediated activation; Class II genes, represented by rhlA, that require RhlR for activation but are only partially RhlI dependent, and Class III genes, such as hcnA, that require RhlR for activation but are RhlI independent. Fig 4 shows quantitative RT-PCR results for the representative genes and complementation assays are shown in Fig 2 and S6 Fig. Based on this classification scheme, the phenazine biosynthesis genes exhibit Class I behavior in HCD planktonic cultures but they behave as Class III genes in colony biofilms. We infer that an as yet unknown ligand(s) exists that allows RhlR to function independently of RhlI in colony biofilms (Fig 1). We speculate that this ligand increases during biofilm formation and promotes RhlR-dependent activation of phz transcription in the absence of the canonical, RhlI-produced autoinducer C4-HSL (Fig 3B).
To garner evidence for a new ligand that acts in conjunction with RhlR, we made a fluorescent transcriptional reporter fusion to the rhlA promoter (PrhlA-mNeonGreen). We chose to follow rhlA because it is a Class II gene (Fig 4), and thus, is expressed in a RhlR-dependent manner both in the presence and absence of RhlI. We incorporated the reporter fusion into an intergenic region on the chromosomes of WT P. aeruginosa and the ΔrhlR and ΔrhlI mutants. The PrhlA-mNeonGreen reporter exhibited 10-fold lower expression in the ΔrhlR mutant than the WT (set to 100%) in HCD planktonic culture (S7A Fig). Expression of the reporter in the ΔrhlI mutant was reproducibly 30% of that in the WT (S7A Fig). These results show that, first, RhlR is absolutely required for expression of PrhlA-mNeonGreen and, second, the reporter fusion continues to be expressed when RhlR is present but RhlI (i.e., C4-HSL) is not. By contrast, the ΔlasR and ΔlasI mutants both exhibit low, but identical, reporter activity (S7A Fig). Thus, LasR does not control rhlA in the absence of LasI (i.e., 3OC12-HSL). Next, we built a strain in which we deleted the lasR, lasI, rhlR, and rhlI genes (called Δ4) and inserted an arabinose-inducible rhlR gene at the glmS locus. We call this strain Δ4 PBAD-rhlR. We inserted the PrhlA-mNeonGreen transcriptional reporter fusion onto the chromosomes of the Δ4 and the Δ4 PBAD-rhlR strains. Both the Δ4 mutant and the Δ4 PBAD-rhlR strain showed background level reporter activity similar to the ΔrhlR mutant (S7A Fig). Exogenous addition of DMSO to the Δ4 PBAD-rhlR strain in conjunction with L-arabinose induction of rhlR also resulted in background activity, which indicates that unliganded RhlR is not capable of activating gene expression (Fig 5, gray bars). Addition of 10 μM synthetic C4-HSL, along with L-arabinose induction of rhlR, restored the WT level of reporter activity. We set this level of activity to 100%. Addition of 10 μM synthetic 3OC12-HSL or 50 μM synthetic PQS (2-heptyl-3-hydroxy-4-quinolone, a third P. aeruginosa QS autoinducer that functions together with a receptor called PqsR [49]) to the Δ4 PBAD-rhlR strain, along with L-arabinose, did not activate reporter activity above background levels (Fig 5, gray bars). Thus, neither 3OC12-HSL nor PQS activate RhlR-driven transcription of rhlA, and so those QS molecules cannot be responsible for the activity present in the cell-free culture fluids prepared from the ΔrhlI mutant.
We next determined whether the activity driving RhlR-dependent expression of the PrhlA-mNeonGreen reporter fusion in the ΔrhlI mutant is present in cell-free culture fluids. Exogenous addition of 20% (v/v) WT cell-free culture fluids to the Δ4 PBAD-rhlR strain in conjunction with L-arabinose induction of rhlR resulted in 10-fold higher reporter activity compared to the background activity obtained following the addition of medium (Fig 5, black bars). Addition of cell-free culture fluids from the ΔrhlI mutant stimulated 55% of the activity stimulated by WT cell-free culture fluids, consistent with the idea that a ligand is indeed present in ΔrhlI culture fluids that can activate RhlR (Fig 5). We considered the possibility that the ΔrhlI mutant produced C4-HSL by a non-RhlI-mediated mechanism. However, LC-MS analyses showed that C4-HSL was present in WT cell-free cultures fluids whereas none could be detected in fluids from the ΔrhlI mutant (S8 Fig), eliminating this formal possibility. We undertook an initial characterization of the activity. It is stable to high temperature, acidic and basic conditions, and it passes through a 3 kDa size exclusion filter (Fig 5). These results preliminarily suggest that the alternative ligand is a small molecule that is not a homoserine lactone (homoserine lactones are not base or heat stable, [50]). We also tested whether phenazines could be the alternate ligand. To do this, we supplied cell-free culture fluids from the ΔrhlI Δphz mutant to the PrhlA-mNeonGreen reporter strain and compared the fluorescence output to that when cell-free culture fluids from the ΔrhlI mutant were supplied. Both preparations elicited the same amount of reporter activity (S7B Fig) showing that phenazines are not capable of activating RhlR-driven gene expression. We conclude that the ΔrhlI mutant cell-free culture fluids contain an as-yet-unidentified ligand(s) that is capable of activating RhlR-dependent target gene expression.
Our finding that RhlR is active in the absence of RhlI suggested to us that there could be RhlI-independent RhlR function under non-laboratory conditions such as during host infection. To probe this possibility, we assessed the relative pathogenicity of WT, ΔrhlR, and ΔrhlI P. aeruginosa PA14 strains in a Caenorhabditis elegans fast-kill infection assay [26,51]. The ΔrhlR mutant was completely avirulent in this assay, while the ΔrhlI mutant remained fully virulent, killing worms as proficiently as WT P. aeruginosa PA14 (Fig 6A). By contrast, both the ΔlasR, and ΔlasI mutants were as virulent as the WT (S9A Fig). These results demonstrate that RhlR does indeed function in the absence of the canonical C4-HSL autoinducer to control virulence in nematodes. Phenazines mediate C. elegans killing in the fast-kill infection assay. To determine if the overproduction of phenazines in the ΔrhlI strain was responsible for the virulence phenotype, we deleted the two phenazine biosynthetic gene clusters, phzA1-G1 and phzA2-G2, in the ΔrhlI mutant (called ΔrhlI Δphz). The ΔrhlI Δphz mutant was attenuated for virulence similar to the Δphz mutant (S10 Fig) showing that the production of phenazines underpins pathogenicity in the ΔrhlI strain.
We also examined the roles of the different QS regulators in a Balb/c murine model of acute P. aeruginosa lung infection. The 50% lethal doses (LD50) for the WT, ΔrhlR, ΔrhlI, ΔlasR, and ΔlasI P. aeruginosa PA14 strains were 1.90 x 106 CFU, 2.6 x 106 CFU, 1.1 x 106 CFU, 3.0 x 106 CFU, and 4.8x106 CFU, respectively. To assess the effects of these mutations on virulence in the murine model, we monitored the time-dependence of infection following intratracheal challenge with sublethal doses (~0.5 LD50) of the strains under study. All of the strains carried a constitutively-expressed luxCDABE operon that had been inserted in the chromosome at the glmS locus, which enabled real-time monitoring of bacterial colonization of the mouse during the course of infection using an IVIS Imaging System. At 24 h, comparable levels of bioluminescence were detected, primarily in the lungs, in all of the infected mice. At 48 h, however, the signal was approximately 20-fold stronger in mice infected with the WT and the ΔrhlI strain than in mice infected with the ΔrhlR, ΔlasI, and ΔlasR strains (Fig 6B; S9B Fig). The imaging data were validated by determining the viable P. aeruginosa CFU per gram of lung homogenate. At 24 h post-infection, all of the infected mice had similar bacterial burdens. The average CFU/gram were 2.1 X 107 for WT, 1.2 X 107 for the ΔrhlR strain, and 5.2 X 107 for the ΔrhlI strain (Fig 6C). At 48 h post-infection, the bacterial load in the lungs of mice infected with the WT and ΔrhlI mutant had increased to 1.6 X 1011 and 3.1 X 1011 CFU/gram, respectively (Fig 6C). By contrast, the bacterial load in lungs of mice infected with the ΔrhlR strain did not increase significantly between 24 and 48 h, maintaining an average of 4.1 X 107 CFU/ gram. Thus, the WT and ΔrhlI mutant produce a four-order of magnitude larger burden of bacteria in the murine host than does the ΔrhlR strain. Unlike in the C. elegans infection assay, in the murine model of acute lung infection, the ΔlasI and ΔlasR mutants did not display the same level of virulence as the WT, but, rather, were attenuated. Specifically, at 48 h, mice infected with the ΔlasI and ΔlasR strains maintained the same bacterial load as at 24 h (S9C Fig). We discuss these differences below. In summary, our results show that RhlR is active in the absence of the RhlI-produced autoinducer in both the C. elegans and the murine animal models. Presumably, a ligand, other than C4-HSL, promotes RhlR function in these two animal assays.
P. aeruginosa uses its two major LuxI/R QS systems, the Las and Rhl systems, to orchestrate the synchronous production of virulence factors and to form biofilms [52]. Both LasR and RhlR regulate gene expression at high cell density when bound to their cognate AHLs. Both receptors recognize DNA motifs possessing dyad symmetry called las-rhl boxes, and both receptors are global transcriptional regulators [20,21]. Here we show, for the first time, that unlike LasR that requires its canonical 3OC12-HSL autoinducer to regulate transcription, RhlR, controls gene expression in both a C4-HSL-dependent and C4-HSL-independent manner. Importantly, C4-HSL-independent regulation by RhlR appears to be highly relevant in biofilms and is critical for pathogenicity in animal models of P. aeruginosa infection.
One mechanism by which RhlR could function in the absence of RhlI is if RhlR activates transcription of target genes without bound ligand. Indeed, LuxR-type regulators with such capability exist. Unliganded EsaR from Pantoea stewartii, ExpR from Erwinia chrysanthemi, and VirR from E. carotovora adopt active dimeric conformations and repress transcription [53–55]. However, RhlR is insoluble, and thus must be inactive, when not bound to a ligand [26]. Consistent with this notion, RhlR displays no basal activity in the absence of exogenously supplied C4-HSL in an E. coli overexpression assay [41]. While it remains a formal possibility that RhlR functions in the absence of a ligand, another possibility that is more consistent with our data and with what is known about RhlR is that an alternative ligand exists that enables RhlR to act as a transcription factor. Our finding that cell-free culture fluids from the ΔrhlI mutant activate RhlR-dependent gene expression strongly supports the alternative ligand hypothesis. We are currently working to purify and identify this putative ligand. The alternative ligand does not appear to be an AHL: the P. aeruginosa genome contains only two LuxI homologs, LasI and RhlI, and the LasI product 3OC12-HSL does not stimulate RhlR-dependent gene expression (Fig 5). Furthermore, 3OC12-HSL is known to inhibit RhlR activity [38]. The alternative ligand is stable in alkaline pH and at high temperatures while canonical AHLs are not [49]. A non-AHL ligand made by a non-LuxI-type synthase is also a possibility, and consistent with this notion, the LuxR homologs PluR and PauR in Photorhabdus spp. bind to pyrones and dialkylresorcinols, respectively, to regulate target gene expression [56,57]. As far as is known, PluR and PauR do not bind AHLs.
LuxR-type proteins exhibit a spectrum of specificities for ligands. Some LuxR-type proteins, including LasR and TraR, are exquisitely specific for their cognate autoinducers [6,58]. Others, including SdiA and CviR, are promiscuous and bind to a variety of AHLs [59,60]. Ligand recognition promiscuity could be a mechanism that enhances the versatility of QS systems in different niches. By expanding the repertoire of stimuli detected, and linking the binding of particular ligands by a LuxR-type receptor to expression of particular subsets of target genes, bacteria could reuse their LuxR-type receptors to tune into and successfully colonize different habitats. Such a scenario would endow P. aeruginosa with the plasticity to diversify its QS outputs, while also being especially economical because it does not necessitate the evolution of a new transcription factor for every small molecule stimulus that is detected. Indeed, P. aeruginosa occupies diverse environments including soil, marshes, plants and animals, and QS is required for success in all of these niches [61,62].
It is possible that additional ligands exist for RhlR, and perhaps for LasR, beyond the new activity we have uncovered here for RhlR. We suggest that each such a ligand or sets of ligands are vital in particular niches making their discovery difficult under laboratory conditions. Indeed, our experiments revealed the activity of one putative alternative RhlR ligand only because we examined colony biofilm conditions, rather than traditional liquid growth. Consistent with this assertion, RNA-seq profiles from P. aeruginosa HCD planktonic cultures and colony biofilms show that RhlR is dramatically less dependent on RhlI, and thus the C4-HSL ligand, in colony biofilms than in planktonic culture. We speculate that the concentration of our proposed alternative ligand increases in colony biofilms enabling the transition of RhlR from being predominantly bound to C4-HSL to being predominantly bound to the alternative ligand. This change, in turn, promotes the transition from expression of RhlR-directed Class I to RhlR-directed Class II and Class III target genes. In this respect, RhlR resembles the multiple antibiotic resistance regulator MarR and homologs from diverse bacteria and nuclear receptors from higher eukaryotes that bind multiple ligands, and depending on which ligand is bound, control expression of discrete subsets of genes [63,64].
Phenazine production contributes to virulence in diverse P. aeruginosa infection models [33,52,65]. Consistent with this finding, we showed that the ΔrhlI mutant that produces phenazines is virulent in both a nematode and a murine infection model while the ΔrhIR mutant, which is defective for phenazine production, is attenuated (Fig 6). Our finding that the ΔlasR and ΔlasI mutants produce phenazines in nematode assay medium provides one condition in which the Rhl system bypasses the requirement for the Las system to promote downstream QS target gene expression (S11 Fig). Phenazine production presumably enables the ΔlasR and ΔlasI mutants to exhibit WT level virulence in the C. elegans fast-kill infection assay. We do note, however, that the ΔlasR and ΔlasI mutants fail to establish a WT level of infection in the murine acute pneumonia model suggesting that a factor(s) other than phenazines plays a crucial role in virulence in mice.
Previous studies have examined QS control of P. aeruginosa virulence in a mouse burn model [13], a corneal infection model [66], a murine neonatal model of acute pneumonia [67], and mouse and rat models of chronic lung infection [68,69]. The majority of these earlier studies were performed using las mutants or strains containing both las and rhl mutations. Rumbaugh et al [13] reported that the P. aeruginosa PA14 ΔrhlI mutant was significantly less virulent than the WT in a mouse burn model. Our results show that the P. aeruginosa PA14 ΔrhlI single mutant is as virulent as WT in a murine acute pneumonia model. Tissue specific responses to different P. aeruginosa PA14 virulence factors could explain these different results.
Our in vivo imaging of P. aeruginosa following intratracheal infection of BALB/c mice revealed infection beyond the respiratory tract (Fig 6B; S9B Fig). The components driving dissemination are not currently known. Earlier rodent studies examining respiratory infection by P. aeruginosa PA14 did not monitor infection in other organs [70,71]. P. aeruginosa PA14 produces the cytotoxin ExoU [72], potentially enabling endothelial permeability and systemic escape of bacteria from the mouse lung into the periphery. However, whether ExoU is responsible for the dissemination from the mouse lung has not been examined. Going forward, the experimental system we developed in this work provides the opportunity to test ExoU as well as other components for roles in P. aeruginosa PA14 dissemination and systemic infection burden in a mammalian context.
P. aeruginosa is a pathogen of high clinical relevance that has acquired resistance to commonly used antibiotics, and is now a priority pathogen on the Centers for Disease Control and Prevention ESKAPE pathogen list (a set of bacteria including Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa and Enterobacter spp. designated as multi-drug resistant pathogens requiring new antimicrobials for treatment) [73–75] and a critical pathogen on the World Health Organization’s priority list [76]. Furthermore, it is now well recognized that lasR mutants arise during adaptation of P. aeruginosa to the cystic fibrosis lung environment [34–36]. The LasI/R system is at the top of the QS hierarchy and LasR:3OC12-HSL is required for rhlI expression, activating rhlI transcription 20-fold (Fig 1, [17]). Thus, a longstanding mystery of urgent clinical importance has been to understand how QS-regulated virulence factors continue to be expressed in the lasR mutants obtained from patients. Our work provides insight into one possible mechanism: the alternative RhlR ligand stabilizes the basal level of RhlR protein that is produced in the absence of LasR, enabling RhlR-dependent virulence gene expression. We propose that targeting RhlR with small molecule inhibitors could provide an exciting path forward for the development of novel antimicrobials.
P. aeruginosa UCBPP-PA14 strain was grown in lysogeny broth (LB) (10 g tryptone, 5 g yeast extract, 5 g NaCl per L), in 1% Tryptone broth (TB) (10 g tryptone per L) and on LB plates fortified with 1.5% Bacto agar at 37°C. When appropriate, antimicrobials were included at the following concentrations: 400 μg/mL carbenicillin, 30 μg/mL gentamycin, 100 μg/mL irgasan, 750 μg/mL trimethoprim.
To construct marker-less in-frame chromosomal deletions in P. aeruginosa, DNA fragments flanking the gene of interest were amplified, assembled by the Gibson method, and cloned into pEXG2 (a generous gift from Dr. Joseph Mougous) [77,78]. The resulting plasmids were used to transform Escherichia coli SM10λpir, and subsequently, mobilized into P. aeruginosa PA14 via biparental mating. Exconjugants were selected on LB containing gentamicin and irgasan, followed by recovery of deletion mutants on LB medium containing 5% sucrose. Candidate mutants were confirmed by PCR. The rhlRW11STOP, rhlIF50STOP and rhlAC11STOP mutants were generated by the above method using overlapping Gibson assembly primers containing the mutations. The rhlR complementation plasmid was constructed by inserting DNA containing ~500 bp upstream of the rhlR gene and the entire rhlR open-reading frame using HindIII and SalI, followed by cloning into similarly digested pUCP18 [79].
To construct the PrhlA-mNeonGreen transcriptional reporter fusion, 500 bp of DNA upstream of the rhlA gene and the DNA encoding the mNeonGreen open-reading frame were amplified using P. aeruginosa PA14 genomic DNA and the plasmid pmNeonGreen-N1 (licensed from Allele Biotech) as templates, respectively [80]. Next, two DNA fragments of ~730 bp, one corresponding to the intergenic region ~700 bp downstream of the P. aeruginosa PA14_20500 gene and the other corresponding to ~1000 bp upstream of the P. aeruginosa PA14_20510 gene, were amplified from P. aeruginosa PA14 genomic DNA. The four DNA fragments were assembled by the Gibson method and cloned into pEXG2. The resulting plasmid was used to transform E. coli SM10λpir, and subsequently mobilized into P. aeruginosa PA14 WT and the ΔrhlR and ΔrhlI mutants via biparental mating as described above. The PchiC-mNeonGreen transcriptional reporter fusion was generated analogously to the PrhlA-mNeonGreen reporter fusion and integrated on the chromosome at the identical ectopic locus.
The L-arabinose inducible PBAD-rhlR construct was generated by inserting the DNA encoding the RhlR open reading frame between the NcoI and EcoRI sites on the pTJ1 plasmid [81]. The construct was mobilized into the ΔlasR ΔlasI ΔrhlR ΔrhlI quadruple mutant (called Δ4) as described previously [78]. Next, the PrhlA-mNeonGreen transcriptional reporter plasmid was conjugated into the Δ4 PBAD-rhlR strain as described above. To construct constitutively bioluminescent strains for mouse infections, the plasmid pUC18T-miniTn7T-lux-Tp was mobilized into the WT, ΔlasR, ΔlasI, ΔrhlR, and ΔrhlI mutants as described previously [82,83]. The strains and plasmids used in this study are listed in S3 Table.
P. aeruginosa strains were grown overnight in LB liquid medium at 37°C with shaking. Cultures were back diluted 1:1000 into fresh medium and grown for 18 h. The cells were pelleted by centrifugation, and the culture fluids were passed through 0.22 μm filters into clear plastic cuvettes. The OD695 of each sample was measured on a spectrophotometer (Beckman Coulter DV 730).
To quantify pyocyanin production from colony biofilms, 10 μL of culture was spotted onto 60 x 15 mm Petri plates containing 10 mL 1% Tryptone medium solidified with 1% agar. The plates were incubated at 25°C for 5 days. Pyocyanin was extracted by the addition of 5 ml chloroform to the plate, followed by a second extraction with 0.2 N HCl. The absorbance of this solution was measured on a spectrophotometer at 530 nm (Beckman Coulter DV 730). The OD520 of each sample was multiplied by 17.072 to obtain the concentration of pyocyanin (μg per CFU) [84].
One microliter of overnight P. aeruginosa cultures grown at 37°C in 1% Tryptone broth was spotted onto 60 x 15 mm Petri plates containing 10 mL 1% Tryptone medium fortified with 40 mg per L Congo red and 20 mg per L Coomassie brilliant blue dyes, and solidified with 1% agar. Colonies were grown at 25°C and images were acquired after 120 h using a Leica stereomicroscope M125 mounted with a Leica MC170 HD camera at 7.78x zoom.
P. aeruginosa strains were harvested from HCD planktonic cultures (OD600 = 2.0) and from mature (5-day old) colony biofilms. RNA was purified using Trizol (Ambion). 0.5 μg of total RNA was subjected to rRNA-depletion using the Ribo-Zero rRNA Removal Kit (Bacteria) from Illumina, followed by library preparation using the PrepX RNA-Seq Illumina Library Kit (WaferGen Biosystems). Unique barcodes were added to each sample pool to enable multiplexing. Libraries were sequenced as single end 75 bp reads on an Illumina HiSeq 2500 instrument. Data analysis was performed on a local Galaxy platform. Reads (~10 million reads per replicate) were mapped to the P. aeruginosa UCBPP-PA14 genome (www.pseudomonas.com, [85]) using TopHat. Differentially expressed genes were identified using DESeq2. Genes showing log2 fold-change > or = 2 and adjusted P-value < or = 0.001 in the mutant strains compared to WT, under the corresponding condition, are reported in this study.
P. aeruginosa strains were harvested from HCD planktonic cultures (OD600 = 2.0) and from mature (5-day old) colony biofilms. RNA was purified using Trizol (Ambion), and subsequently, DNAse treated (TURBO DNA-free, Thermo Fisher). cDNA was synthesized using SuperScript III Reverse Transcriptase (Invitrogen) and quantified using PerfeCTa SYBR Green FastMix Low ROX (Quanta BioSciences).
WT and mutant P. aeruginosa strains harboring the chromosomally encoded PrhlA-mNeonGreen fusion were grown overnight at 37°C. The cultures were diluted 1:1000 in 3 mL of TB medium. When required, DMSO solvent or 20% (v/v) cell-free culture fluids were added and the cultures were incubated at 37°C until the OD600 reached 2.0. 1 mL of culture was harvested, the supernatant was removed, and the cells were resuspended in PBS. 200 μL of culture suspension was transferred to wells of 96 well plates and fluorescence was measured using an Envision 2103 Multilabel Reader (Perkin Elmer) using the FITC filter with an excitation of 485 nM and emission of 535 nM.
WT N2 worms were propagated on E. coli OP50 lawns on Nematode Growth Media (NGM) plates [86]. Gravid adults were allowed to lay eggs on lawns of fresh E. coli OP50 after which the adults were removed and the eggs were allowed to grow for 48 h (to reach the L4 stage) at 20°C prior to transfer to lawns of P. aeruginosa strains at 25°C on PGS plates (1% Bacto-Peptone, 1% NaCl, 1% glucose, 0.15M sorbitol, 1.7% Bacto-Agar). Nematodes were scored for survival at 4, 16, and 24 h time points (30 worms per replicate, three replicates performed). Data were plotted and SEM determined using GraphPad Prism software.
P. aeruginosa strains were grown on Pseudomonas Isolation Agar (PIA) for 16–18 h at 37°C and suspended in PBS to an OD600 of 0.5, corresponding to ~109 CFU/mL. Inocula were adjusted spectrophotometrically to obtain the desired challenge dose in a volume of 50 μL. Six-week old female Balb/c mice (Jackson Laboratories, Bar Harbor, ME) were anesthetized by i.p. injection of 0.2 mL of a mixture of ketamine (25 mg/mL) and xylaxine (12 mg/mL). Mice were infected by non-invasive intratracheal instillation of dilutions of P. aeruginosa PA14 P1-lux or isogenic QS mutants as previously described [87]. Mice were observed over 5 days, and animals that succumbed to infection or appeared to be under acute distress were humanely euthanized and were included in the experiment results. To determine each LD50, groups of mice were challenged with different doses of P. aeruginosa PA14 or isogenic QS mutants. Four mice were tested with each dose of each strain. The percent lethality corresponding to each dose was assessed. The LD50 was calculated using the method of Reed and Muench [88]. Data were analyzed by the log-rank test. All survival experiments were repeated at least three times.
For colonization of mice, P. aeruginosa strains were grown and prepared as described above. Six-week-old female Balb/c mice were anesthetized and infected with sublethal doses (~0.5 LD50) of P. aeruginosa PA14 P1-lux or isogenic QS mutants as described in the preceding section. Mice were euthanized at 24 and 48 h post-infection and whole lungs were collected aseptically, weighed, and homogenized in 1 mL of PBS. Tissue homogenates were serially diluted and plated on PIA and CFU determination was made 16–18 h later. Comparison of the numbers of viable bacteria obtained in lung homogenates relied on the Kruskal-Wallis test for three group analyses or the Mann-Whitney U test for two group analyses.
An additional group of animals was included for each P. aeruginosa strain examined for real time monitoring of colonization and localization of bioluminescent bacteria using an IVIS–Lumina LT III imaging system (PerkinElmer). Briefly, each group of mice was anesthetized with 3% isoflurane using an XGI-8 Gas Anesthesia System (Caliper Life Sciences), and imaged using medium binning, f/stop 1, subject height 1.5 cm. Images were acquired with up to 5 min exposure. Total photon emission from the ventral and dorsal sides of imaged mice was quantified using Living Image Software v4.0x (Xenogen Corp.). Due to the differences in the virulence level and resulting bacterial load, mice infected with P. aeruginosa PA14 P1-lux or isogenic QS mutants were monitored separately, but individual mice were compared at 24 and 48 h under the same conditions. All correlations were done as average radiance of photons emitted per second, area, and steradian (p/s/cm2/sr) under the chosen experimental conditions. All colonization and imaging experiments were repeated at least twice. Statistical analyses were performed using GraphPad Prism software.
Samples were diluted 1:1 with methanol (MeOH). Synthetic C4-HSL (Sigma-Aldrich) standards were prepared in 50% MeOH. Samples and standards were loaded onto a 1 mm x 150 mm C12 column (Phenomenex, Jupitor 4 μm Proteo 90A) using a Shimadzu HPLC system and PAL auto-sampler (20 μL/injection) at a flowrate of 70 μL/min. The column was maintained at 35°C using a column oven. The column was connected inline to an electrospray source coupled to an LTQ-Orbitrap XL mass spectrometer (ThermoFisher). Caffeine (5 pM/μL in 50% Acetonitrile with 0.1% Formic Acid) was injected as a lock mass through a tee at the column outlet using a syringe pump at 10 μL/min (Harvard PHD 2000). Chromatographic separation was achieved with a linear gradient from 1% to 99% B in A in 8.5 min (A: 0.1% Formic Acid, B: 0.1% Formic Acid in Acetonitrile) with an initial 1 min hold at 1% B and followed by 5 min wash at 100% B and equilibration for 10 min with 1% B (total program was 20 min). Electrospray ionization was achieved using a spray voltage of 4.5 kV aided by sheath gas (Nitrogen) flow rate of 12 (arbitrary units) and auxiliary gas (Nitrogen) flow rate of 1 (arbitrary units). Full scan MS data were acquired in the Orbitrap at a resolution of 60,000 in profile mode from the m/z range of 160–320. LC-MS data were manually interpreted using the Xcalibur Qual browser (Thermo, Version 2.1) to visualize C4-HSL mass spectra and to generate extracted ion chromatograms using the theoretical [M+H] within a range of ±2 ppm.
All animal procedures were conducted according to the guidelines of the Emory University Institutional Animal Care and Use Committee (IACUC), under approved protocol number DAR-2003421-042219BN. The study was carried out in strict accordance with established guidelines and policies at Emory University School of Medicine, and recommendations in the Guide for Care and Use of Laboratory Animals of the National Institute of Health, as well as local, state, and federal laws.
|
10.1371/journal.pntd.0004873 | Heat Sensitivity of wMel Wolbachia during Aedes aegypti Development | The wMel strain of Wolbachia bacteria is known to prevent dengue and Zika virus transmission in the mosquito vector Aedes aegypti. Accordingly, the release of wMel-infected A. aegypti in endemic regions has been recommended by the World Health Organization as a potential strategy for controlling dengue and Zika outbreaks. However, the utility of this approach could be limited if high temperatures in the aquatic habitats where A. aegypti develop are detrimental to Wolbachia. We exposed wMel-infected A. aegypti eggs and larvae to fluctuating daily temperatures of 30–40°C for three, five, or seven days during their development. We found that Wolbachia levels in females emerging from heat treatments were significantly lower than in the controls that had developed at 20–30°C. Notably, seven days of high temperatures starting at the egg stage reduced Wolbachia levels in emerging females to less than 0.1% of the wMel control levels. However, after adult females returned to 20–30°C for 4–7 days, they experienced differing degrees of Wolbachia recovery. Our findings suggest that the spread of Wolbachia in wild A. aegypti populations and any consequent protection from dengue and Zika viruses might be limited in ecosystems that experience periods of extreme heat, but Wolbachia levels recover partially after temperatures return to normal.
| The proposed arbovirus biocontrol strategy of releasing mosquitoes infected with the wMel strain of Wolbachia bacteria promises to reduce the transmission of dengue and Zika viruses, but its utility in the field may be limited by the local environment. We show that when Aedes aegypti infected with wMel experience high temperatures during egg and larval development, they have lower Wolbachia levels as emerging adults. High temperatures starting at the egg stage and lasting for seven days reduce Wolbachia levels in emerging females to less than 0.1% of control levels. However, partial recovery of Wolbachia occurs by 4–7 days of age. The spread of Wolbachia in wild A. aegypti populations and any resulting impacts on dengue and Zika transmission could be limited by periods of extreme heat, but Wolbachia levels may subsequently recover.
| Mosquito-borne arboviruses are a growing public health threat. The alarming geographic spread and costly health burden of dengue fever have led the World Health Organization (WHO) to deem it “the most important mosquito-borne viral disease in the world.” Over the last 50 years, the incidence of dengue cases has increased 30-fold [1]. Now more than 100 countries have endemic dengue and over 40% of the world's population is at risk [2]. Zika virus, which recently caused a surge of children born with microcephaly and other neurological disorders, was declared a Public Health Emergency of International Concern (PHEIC) by the WHO after it spread from Brazil to 26 other countries or territories in the Americas within one year [3, 4]. With no effective antiviral treatments in the arsenal and just one licensed dengue vaccine that is 65.6% effective for those 9 years or older, control of the mosquito vectors, Aedes aegypti and Aedes albopictus, is the most viable option for curbing transmission of these arboviruses [5–7]. However, in resource-limited cities with endemic dengue, vector control efforts are often only pursued in response to explosive epidemics [8, 9]. Failure to control these vectors in tropical urban environments is one of the major drivers of the growing incidence and geographic expansion of dengue and other mosquito-borne arboviruses [10].
Alarmingly, existing control options for A. aegypti are of little use in urban areas [9, 11]. Space spraying with ultra-low volume insecticides, including organophosphates and pyrethroids, has been used by many countries in the face of dengue outbreaks for the past 40 years despite limited evidence of its epidemiological benefits [10, 12, 13]. Vector densities inevitably recover after space spraying because ideal larval habitats for A. aegypti abound in cities—exposed water sources for drinking or washing and non-biodegradable trash that collects water [8]. Targeted spraying of potential larval development containers with residual insecticides [14, 15] and indoor residual spraying targeting adult mosquitoes [16] in combination can substantially reduce local dengue incidence, but only if high coverage is achieved [17, 18]. For countries faced with nearly ubiquitous breeding of A. aegypti in their sprawling cities, the comprehensive spraying required to stop transmission is unrealistic [19]. Consequently, there is no urban setting in which vector control has completely eliminated dengue virus (DENV) transmission or prevented dengue epidemics [13, 20].
A potential solution for DENV and Zika virus (ZIKV) transmission involves releasing A. aegypti infected with Wolbachia, a common bacterium infecting the reproductive systems of many insects [21–24]. The fitness effects of Wolbachia on insect hosts are strain specific, ranging from life-shortening to pathogen-blocking phenotypes [25, 26]. The pathogen-blocking properties of some strains of Wolbachia have led researchers to characterize them further and transfer them into vector species for potential use in vector-borne disease control. The wMel strain of Wolbachia shows the most promise currently, as it blocks DENV and ZIKV transmission by the mosquito, raising the possibility of disrupting dengue and Zika transmission cycles [27–30]. The wMel strain was transinfected from Drosophila melanogaster into A. aegypti, and wMel-infected A. aegypti have been released at sites in Australia, Vietnam, Brazil, Indonesia, and Colombia [31]. The success of each Wolbachia strain in invading insect populations is determined by the net fitness effect of the strain coupled with the extent to which it manipulates host reproduction [32]. One key mechanism of reproductive manipulation is cytoplasmic incompatibility (CI). When a strain causes complete CI, Wolbachia-infected females can mate successfully with Wolbachia-infected males, while uninfected females cannot [32]. The wMel Wolbachia strain causes complete CI and has had some success in invading wild A. aegypti populations [28, 33]. However, the prevalence of wMel Wolbachia must remain high in the A. aegypti population in order for wMel to reliably and substantially reduce the capacity of the mosquito population to transmit pathogens [32, 34]. Protection against DENV in field-collected wMel A. aegypti is similar to that observed in the original transinfected wMel line [35], indicating that this strategy might be used to reduce dengue transmission in endemic areas [28, 29, 36]. Recently the WHO recommended the use of Wolbachia for dengue and Zika control [4], although there is currently insufficient epidemiological evidence to know if the approach is effective. It is also unknown whether the prevalence of wMel-infected A. aegypti and the wMel Wolbachia levels within individual mosquitoes will remain high enough to prevent DENV and ZIKV transmission in all environments.
The levels of wMel Wolbachia load throughout the various stages of the A. aegypti lifespan have not been described, as most studies have focused on population dynamics and fitness effects of wMel Wolbachia after adult emergence [28, 33, 35, 37–40]. The early stages of development comprise a sensitive period during the A. aegypti lifespan; immature forms are confined to their aquatic habitats, whereas adults can seek out favorable microclimates to increase their chances of survival [41–43]. Immature A. aegypti develop in containers in the domestic environment that hold water, including flower pots, tanks, and drums as well as bottles, cans, and automobile tires [8, 44]. These containers sometimes hold as little as 5 mL of water [45]. Female A. aegypti preferentially lay their eggs in shaded containers, but it is not uncommon to find immatures in containers fully exposed to the sun [46, 47]. Although comprehensive temperature measurements in sun-exposed containers have not been carried out, lab-reared A. aegypti larvae can tolerate aquatic temperatures as high as 43°C if they are pre-exposed to high but sublethal temperatures [48]. The ability of wMel Wolbachia to tolerate the same elevated temperatures as immature A. aegypti has not been investigated.
The heat sensitivity of Wolbachia with respect to its hosts has been characterized in other arthropods. Exposure to high temperatures during development cured the Wolbachia infections of two-spotted spider mites Tetranychus urticae [49], Tribolium flour beetles [50], and Drosophila spp. [51–54]. In the mosquito Aedes scutellaris, the reproductive effect of CI caused by Wolbachia was lost when larvae were reared at 32.5°C, but it was unknown whether the loss of Wolbachia or host expression of heat-shock proteins was responsible [55–57]. In A. albopictus all life stages maintained at 37°C had a lower levels of Wolbachia than those reared at 25°C, indicating that high temperatures may reduce Wolbachia levels in mosquito hosts [58].
Reduced Wolbachia levels in response to high temperatures during larval development could represent a barrier to the spread of wMel Wolbachia in A. aegypti populations if fundamental drive mechanisms such as maternal transmission and CI are affected. Because only Wolbachia-infected females produce viable offspring with Wolbachia-infected males, CI creates a selective pressure for the spread of Wolbachia [32]. The spread of Wolbachia in mosquito populations is crucial, because incomplete wMel Wolbachia coverage in the A. aegypti population leaves the potential for DENV and ZIKV transmission. A recent study found geographical clusters of uninfected mosquitoes in a wMel-infected A. aegypti release suburb of Cairns, Far North Queensland, Australia [59]. The incomplete Wolbachia coverage was suggested to be due to immigration of uninfected mosquitoes from outside the release area, cryptic breeding sites, or other environmental phenomena such as “larval curing” (loss of Wolbachia infection during larval development) [59]. However, the occurrence of larval curing in mosquitoes has been poorly defined to date. Specifically, little is known about the temperature thresholds for Wolbachia during mosquito development or whether any potential curing persists after temperatures return to normal. Understanding larval curing in wMel-infected A. aegypti has important applications, as lower Wolbachia levels in adults might have downstream impacts on cytoplasmic incompatibility [60–67] (although in D. simulans between-strain differences in CI are not explained by Wolbachia density [68]), maternal transmission [69, 70], and pathogen inhibition [29, 68, 71–73].
We investigated the effects of high temperatures during egg and larval development on laboratory-reared wMel-infected A. aegypti using fluctuating daily temperatures that simulate the real-world conditions of a heatwave in Cairns, Australia. Our results have implications for the projected spread of wMel Wolbachia through A. aegypti populations and for the vector competence of wMel-infected A. aegypti under different environmental conditions.
Blood feeding of mosquito colonies using human volunteers was performed in accordance to the QIMR Berghofer Human Research Ethics Committee permit QIMR HREC361. Written informed consent was obtained from all volunteers who participated in the study.
Mosquitoes were taken from a Wolbachia-free A. aegypti colony (“Cairns” line) started from eggs collected in Cairns, Australia, in January 2015 and from a colony of wMel-infected A. aegypti (“wMel” line) started from eggs collected in suburbs of Cairns in April 2015. The colonies were maintained in separate, identical climate-controlled rooms at 27 ± 1°C and 70 ± 10% relative humidity with a 12:12 hour light:dark cycle and crepuscular periods. Eggs were flooded in aged (≥ 48 h) tap water and allowed to hatch naturally. Larval stages were reared under a controlled density (< 200 larvae per tray) in trays with 3 L of aged tap water. Larvae were fed on ground TetraMin tropical fish food (Tetra, Germany). Pupae were transferred into cages measuring 40 × 40 × 30 cm for adult emergence. Colonies were maintained with a population size of > 500 individuals per generation. Adult mosquitoes received 10% sucrose solution ad libitum, and females were blood-fed on a human volunteer for 15 min every 7 d. The wMel-infected A. aegypti colony was regularly screened for Wolbachia using PCR of the wsp gene from the time of establishment [74]. Prior to the start of the experiments, our screening showed that the colony was completely infected with Wolbachia.
For the experiments, eggs were collected from A. aegypti wMel (F16 and F17 generations used) and A. aegypti Cairns (F18 and F19 generations) colonies at 8:30 A.M. following the first night of oviposition. Eggs were counted under a stereomicroscope at 23°C and were separated into batches of approximately 600 eggs. Each batch was placed inside a dry paper towel, which was folded and placed next to a damp paper towel inside an open plastic bag. Egg bags were placed inside their corresponding environmental chambers at the coldest point of the temperature cycles, which was 20°C for the control condition and 30°C for the treatment condition. Eggs were left to mature for 48 h, and then batches of approximately 150 eggs were flooded in 500 mL aged tap water in plastic trays (183 × 152 × 65 mm). Four replicate trays were used per treatment group. From the day of hatching until pupation, ground TetraMin tropical fish food (Tetra, Germany) was administered daily at the coldest point of the temperature cycles using the “medium” diet described by Hugo et al. [75]. Pupae were transferred into 1-L plastic containers with mesh tops, and emerging adults were given 10% sucrose solution ad libitum. Adult females were aspirated out at 0–2 days post-emergence and at 4–7 days post-emergence. They were frozen at -20°C until processing.
We tested the effect of high temperatures during egg and larval development on Wolbachia levels in A. aegypti wMel adult females in two replicate experiments: Each replicate experiment compared various heatwave temperature regimes applied during particular periods of immature mosquito development that varied in duration and stage of onset. The temperature profiles we used simulated observed temperatures during average and extreme conditions in Cairns, Queensland. The Australian Bureau of Meteorology defines a heatwave as “a period of at least three days where the combined effect of excess heat and heat stress is unusual with respect to the local climate” [76]. We designed our treatment temperature profile to surpass the severe daily mean temperature threshold of 30.4°C for Cairns, which is based on temperature data from 1958 to 2011 [76]. Both treatment and control temperature profiles followed a truncated sinusoidal progression during the day and exponential decrease at night, representing a profile of daily temperature variation [77]. The shapes of the profiles were the same for each condition, but the profile was raised or lowered to adjust the mean temperature (S1 Fig). Experiments were conducted in two environmental chambers (294-L Panasonic MLR-352H-PE and MLR-351H, Gunma, Japan). Nine treatment groups were exposed to fluctuating heatwave temperatures between 30°C and 40°C for varying durations beginning at various life stages. Controls consisted of wMel A. aegypti and wildtype Cairns A. aegypti exposed to diurnal temperature fluctuations between 20°C and 30°C. Transfers between environmental chambers were made at the coldest point of the temperature cycles (20°C for the control condition and 30°C for the treatment condition) in order to minimize the likelihood of heat shock. As illustrated in Fig 1, treatment groups exposed to high temperatures beginning from early embryogenesis (eggs at ≤ 15 hours post-oviposition) lasting three, five, or seven days are denoted by “E3,” “E5,” and “E7.” Groups exposed to high temperatures beginning at the immature larval stages (1st/2nd instars) lasting three, five, or seven days are denoted by “I3,” “I5,” and “I7.” Groups exposed to high temperatures beginning at more mature larval stages (3rd/4th instars) lasting three, five, or seven days are denoted by “M3,” “M5,” and “M7.” Prior to the two studies, a pilot study was conducted to determine differences in means for a range of onsets and durations (S2 Fig).
Data loggers, both factory installed and independent HOBO data loggers (Onset, Cape Cod, MA), recorded light intensity and temperature variation. Actual water temperatures in the control chamber were within 1.00°C of the programmed air temperature throughout the duration of the experiments. This was also the case in the treatment chamber, except during the coldest periods, when water temperature was as much as 2.93°C lower than the programmed air temperature.
Wolbachia densities within individual adult females were determined by quantitative PCR. The head was removed from each frozen adult female before DNA extraction. Genomic DNA was extracted using QuickExtract DNA Extraction Solution (Epicentre Technologies Corporation) as per the manufacturer’s instructions and was diluted 1:10 in purified water. Multiplex qPCR was performed, amplifying the target Wolbachia-specific wsp gene and the somatic Actin5c gene, which acted as a reference gene to standardize for mosquito body size (wsp F: 5´–CATTGGTGTTGGTGTTGGTG–3´, R: 5´–ACACCAGCTTTTACTTGACCAG–3´, Actin5c F: 5´–GACGAAGAAGTTGCTGCTCTGGTTG–3´, R: 5´–TGAGGATACCACGCTTGCTCTGC–3´) (full methods in S1 Appendix) [73, 78, 79]. Quantification cycles (Cq) were normalized by taking into consideration the different amplification efficiencies of the wsp and Actin5c genes, and Wolbachia to host genome ratios were calculated using Q-Gene [80].
Fluorescence in situ hybridization (FISH) was carried out using a Wolbachia-specific 16S rRNA probe [29]. Three freshly collected adult females (legs and wings removed) from each treatment group were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer overnight and were transferred to 70% ethanol. Bodies were embedded in paraffin wax and sectioned with a microtome. Slides were dewaxed with two successive xylene washes for 10 min, two successive 5-min washes with 100% ethanol, and two successive 5-min washes in 95% ethanol. Slides were hybridized with the Wolbachia-specific W2 probe (5´–CTTCTGTGAGTACCGTCATTATC–3´) [29] conjugated on the 5´ end to the fluorescent probe Alexa Fluor 488 (Molecular Probes, Inc). Slides were left in a dark humidity chamber at 37°C overnight and washed briefly in 1× saline sodium citrate (SSC) buffer + 10 mM dithiothreitol (DTT) at room temperature, then two 15-min washes in 1× SSC + 10 mM DTT at 55°C, two 15-min washes in 0.5× SSC at 55°C, a 10-min wash in 0.5× SSC + 10 mM DTT + 4',6-Diamidino-2-phenylindole (DAPI) (0.01 mg/50 mL) at room temperature, and then a final 10-min wash in 0.5× SSC + 10mM DTT at room temperature. Slides were washed briefly with distilled water and mounted with Vectashield Hard Set mounting medium (Vector Laboratories, Burlingame, CA). Slides were allowed to dry in a refrigerator overnight. Images from all sections were captured with a DeltaVision Core Deconvolution Microscope (GE) using identical acquisition settings (S2 Appendix). Images were reformatted using SoftWorx (Enterprise Softworks (Pty) Ltd.) and were cropped and standardized for contrast using Adobe Photoshop CS6 (Adobe Systems, Inc.).
To determine the effect of the heat treatments on adult body size, the left wing of six females from each treatment group was removed and dry mounted on a slide. The distance from the axial notch to the wing tip, excluding the fringe scales, was used as a proxy for body size [75, 81].
All analyses were performed in R [82] and GraphPad Prism v. 6 (GraphPad Software, San Diego, California, USA). Normality and homogeneity of variances within treatments were tested using Shapiro-Wilk and Bartlett’s tests, respectively. Log10-transformed Wolbachia densities were used for all analyses. A two-way blocked analysis of variance (ANOVA) was performed to determine the effects of treatment and collection time point and their interaction on Wolbachia density. Replicate was included as a blocking factor to account for any variation between the two experiments. An analogous two-way blocked ANOVA was performed to determine the effects of treatment group and collection time point and their interaction on body size. Pair-wise post-hoc comparisons between treatments and controls and between collection time points were made for both ANOVAs, and P values were adjusted for multiple comparisons using Tukey’s honest significant difference test. Differences were considered significant if adjusted P values were < 0.05. A nonlinear regression was performed using ordinary least squares fit for each stage of onset at the two collection time points to determine relationships between the heat treatment duration and Wolbachia density. Sum of squares F-tests were used to determine significant differences in slopes and y-intercepts.
We found significantly lower Wolbachia densities relative to wMel controls in 0–2 d-old females emerging from eight of the nine treatments (Fig 2), with only the mature instar treatment lasting three days (M3) showing no significant reduction. Wolbachia levels in the 0–2 d-old females that were exposed to 30–40°C for seven days starting at the egg stage (E7) were less than 0.1% of wMel control densities (Fig 2). Both treatment group and collection time point were significant predictors of Wolbachia density (F(10, 362) = 197.34, MSE = 55.24, P < 0.001 and F(1, 362) = 397.21, MSE = 111.20, P < 0.001, respectively). Compared with the 0–2 d adult collection time point, 4–7 d-old adult females in all treatment groups except the wMel control group and the M3 group had higher Wolbachia levels, with adults from three-day treatments (E3, I3, and M3) showing Wolbachia densities that were not significantly different from wMel-infected controls (Fig 2). The Wolbachia levels in 4–7 d-old adults from the six other treatments remained significantly lower than in wMel-infected controls. There were inverse relationships between the duration of heat treatment and Wolbachia density and for all stages of onset; however, the relationships differed significantly both in their slopes and y-intercepts (F(5, 305) = 3.68, P = 0.003 and F(5, 305) = 2.79, P = 0.02, respectively). Duration of heat exposure had the greatest impact on Wolbachia density in emerging females when high temperatures began in the 3rd/4th instar stages. At 4–7 days of age the impact of heat duration on density was most pronounced when high temperatures began at the egg stage.
We also investigated whether we could visualize reductions in Wolbachia levels in the ovaries of adult mosquitoes after exposure to high temperatures during development. Using FISH we visualized very low levels of Wolbachia in the ovaries of 0–2 d-old E7 females (Fig 3B). We also noticed that the E7 ovaries were much less developed than in controls, a possible consequence of the heat exposure. In 4–7 d-old E7 females (Fig 3D), Wolbachia remained at very low levels compared with 4–7 d-old wMel-infected controls (Fig 3C).
We found a significant effect of treatment group on wing length (F(10,71) = 13.70, MSE = 0.32, P < 0.001) and of the treatment group–collection time point interaction (F(9,71) = 2.81, MSE = 0.07, P = 0.007). Collection time point and replicate were not significant predictors (F(1, 71) = 0.22, MSE = 0.005, P = 0.64 and F(1, 71) = 1.21, MSE = 0.03, P = 0.27, respectively). Treatment groups E7, I5, I7, M3, M5, and M7 were all significantly smaller than wMel controls (S3 Fig). There was no significant difference in wing length between wMel controls and Cairns controls.
We found that when A. aegypti infected with the wMel strain of Wolbachia were exposed to daily fluctuating temperatures of 30–40°C during early development, the emerging females had reduced Wolbachia levels compared with controls. The most affected group consisted of mosquitoes exposed to high temperatures starting at the egg stage and lasting for seven days (E7). In E7 emerging females, mean Wolbachia levels were less than 0.1% of the levels of wMel controls. Loss of Wolbachia density from a subset of the mosquito population may be a concern for Wolbachia-based dengue and Zika control efforts in regions where the aquatic habitats of juvenile A. aegypti can reach extremely high temperatures. It has previously been shown that different Wolbachia strains attain different infection densities and that density is correlated with the level of virus inhibition [68, 71, 73, 78]. The relationship between wMel density and DENV and ZIKV inhibition can be assumed from near complete blockage of these viruses in Ae. aegypti harboring dense wMel infections [28, 30, 35], but the relationship has not been specifically defined. A recent study found that exposure of adult wMel-infected A. aegypti to 28°C ± 4°C beginning at 5–8 d of adult age was associated with reduced Wolbachia densities; however, there was no interaction between the reduced densities and DENV infection, dissemination, or transmission [40]. Eggs and larvae exposed to high temperatures in our study produced adult A. aegypti with very low Wolbachia densities; therefore, the level of pathogen inhibition in adult mosquitoes that were subject to impacts of heat exposure during early development deserves investigation. The partial recovery of Wolbachia density by 4–7 days of age suggests that any impacts of heat exposure during mosquito development on subsequent virus inhibition may be attenuated with age.
This study is the first to investigate the duration and timing of heatwave conditions in relation to immature development of mosquitoes infected with Wolbachia. To achieve this we simulated normal and heatwave conditions based on temperature data from a city selected for Wolbachia biocontrol. We found an inverse relationship between the duration of heat exposure and Wolbachia density in adult females, raising the possibility that longer periods of heat might be capable of clearing Wolbachia. The slope of this relationship varied by the stage of heat onset and by the age of adult females collected. Duration of heat exposure had the greatest impact on Wolbachia density in emerging females when high temperatures began in the 3rd/4th instar stages; however, the impact of heat duration on density at 4–7 days of age was most pronounced when high temperatures began at the egg stage. In addition to reducing bacterial densities, high temperatures resulted in smaller adult body sizes, with more prominent effects in the later stages of heat onset and the longer durations. This is likely due to the known inverse relationship between larval rearing temperature and adult body size [83]. We controlled for the effect of body size by standardizing Wolbachia density measurements with the host gene Actin5c.
Loss of Wolbachia density in response to heat has also been reported in T. urticae [49] O. scapulalis [84], D. simulans [54], D. bifasciata [53], A. albopictus [58], the predatory mite Metaseiulus occidentalis [85], and the wasp Leptopilina heterotoma [86]. The mechanism behind the loss of Wolbachia in response to high temperatures is not fully understood, but deformation of the Wolbachia cellular membrane could be a contributing factor [87]. Our FISH visualization confirms the loss of Wolbachia from the ovaries of mosquitoes exposed to high temperatures. Partial recovery of Wolbachia in the ovaries after the mosquito returns to normal temperatures suggests that Wolbachia replication continues even after the ovaries are fully developed. It is uncertain whether replication continues throughout the female lifespan and at what age Wolbachia densities would be restored to control levels in heat-exposed females.
Our results support the notion that wMel has a more restricted thermotolerance than its mosquito host A. aegypti. Loss of thermotolerance in insect symbionts can be due to point mutations that occur as the symbiont co-evolves with the host [88]. In the case of the obligate symbiont of aphids Buchnera aphidicola, a point mutation affecting heat-shock protein transcription leads to death of the symbiont following a heat treatment [89]. Compared with other symbionts of insects, wMel has experienced far less reductive evolution, as evidenced by its large genome with very high levels of repetitive DNA and mobile DNA elements [90]. Because of the low mutation rate of wMel [90], loss of thermotolerance is less likely than for other symbionts [91]. If reductive evolution of wMel does occur, then rearing wMel-infected A. aegypti under constant temperatures in the lab might accelerate loss of wMel thermotolerance. More studies are needed to understand the co-evolution of wMel and A. aegypti.
Wolbachia may hold the potential to reduce and even eliminate dengue and Zika transmission in endemic areas. The advent of a promising control tool for dengue fever and Zika could not have come at a better time, as currently many tropical countries have no options to control the massive arbovirus outbreaks they experience. The strategy of releasing wMel-infected A. aegypti is being tested in dengue-endemic regions around the globe, including Australia, Vietnam, Brazil, Indonesia, and Colombia [31], although substantial epidemiological data is still needed to assess the impacts on dengue and Zika transmission. The importance of measuring Wolbachia density in field trials, as opposed to presence or absence of Wolbachia, is highlighted by our results and other investigations [61, 64, 78]. We found that the high temperatures that A. aegypti may experience during early development can attenuate wMel Wolbachia levels. Consequently, wMel Wolbachia might be less effective as a dengue or Zika control strategy in regions experiencing periods of extreme heat. If the effectiveness is compromised, increased surveillance and supplementary mosquito control may be required in these regions. Further estimates of Wolbachia recovery rates after heat exposure are needed to understand the impacts on DENV and ZIKV inhibition and the spread of wMel through naïve A. aegypti populations.
In summary, we showed that fluctuating daily temperatures of 30–40°C experienced during wMel-infected A. aegypti egg and larval development significantly reduced Wolbachia levels in emerging adult females. However, Wolbachia recovered to differing degrees after adults returned to 20–30°C. These findings suggest that the effectiveness of Wolbachia-based arbovirus control might be compromised in ecosystems that experience periods of extreme heat, but given that Wolbachia levels partially recover after temperatures return to normal, any effects may be temporary. Greater understanding of environmental variables that affect Wolbachia can inform release site selection and help to better predict the impacts of Wolbachia on arbovirus transmission.
|
10.1371/journal.pgen.1005163 | Genetic Architecture of Abdominal Pigmentation in Drosophila melanogaster | Pigmentation varies within and between species and is often adaptive. The amount of pigmentation on the abdomen of Drosophila melanogaster is a relatively simple morphological trait, which serves as a model for mapping the genetic basis of variation in complex phenotypes. Here, we assessed natural variation in female abdominal pigmentation in 175 sequenced inbred lines of the Drosophila melanogaster Genetic Reference Panel, derived from the Raleigh, NC population. We quantified the proportion of melanization on the two most posterior abdominal segments, tergites 5 and 6 (T5, T6). We found significant genetic variation in the proportion of melanization and high broad-sense heritabilities for each tergite. Genome-wide association studies identified over 150 DNA variants associated with the proportion of melanization on T5 (84), T6 (34), and the difference between T5 and T6 (35). Several of the top variants associated with variation in pigmentation are in tan, ebony, and bric-a-brac1, genes known to affect D. melanogaster abdominal pigmentation. Mutational analyses and targeted RNAi-knockdown showed that 17 out of 28 (61%) novel candidate genes implicated by the genome-wide association study affected abdominal pigmentation. Several of these genes are involved in developmental and regulatory pathways, chitin production, cuticle structure, and vesicle formation and transport. These findings show that genetic variation may affect multiple steps in pathways involved in tergite development and melanization. Variation in these novel candidates may serve as targets for adaptive evolution and sexual selection in D. melanogaster.
| Body pigmentation contributes to the spectacular biodiversity present in nature and mediates mate choice, mimicry, and physiological functions such as thermoregulation and UV resistance. Thus, pigmentation is a significant contributor to fitness. In order to understand how complex traits such as pigmentation evolve, we must first identify the genetic variants underlying phenotypic variation. We used the Drosophila melanogaster Genetic Reference Panel, a wild derived population of fully sequenced inbred fly lines, to identify the contributions of both known and novel genetic variants to natural variation in abdominal pigmentation in female flies. Our results show that genetic variation within many biological pathways contributes to variation in D. melanogaster pigmentation.
| Body pigmentation is a conspicuous trait that is variable within species, giving rise to natural variation, polyphenism and sexual dimorphism [1–4]. It also varies between species, contributing to species recognition, mate choice, thermoregulation, protection (warning signals), mimicry, and crypsis [5–7]. Changes in pigmentation are often adaptive and vital to the fitness of the organism [5,6].
Not only is body pigmentation ecologically relevant, in Drosophila it is a relatively simple and easily measured phenotype to study the genetic architecture of natural variation in complex traits [2,7–10]. Each tergite of female D. melanogaster generally has a stripe of dark coloration (melanin) on a lighter tan background (sclerotin). During pre- and post-ecdysis, the epidermal cells underlying the cuticle secrete tyrosine-derived catecholamines into the cuticle for sclerotinization and melanization [11,12]. The melanin/sclerotin biosynthetic pathway and its underlying genetic basis have been well studied. However, many of the genes known to affect D. melanogaster pigmentation do not form part of this pathway or any parallel pathway [5,13]. Furthermore, the genes that lead to natural variation in body pigmentation are not necessarily the same genes that are directly involved in the biosynthesis of melanin and sclerotin. By mapping the genetic basis of natural variation in body pigmentation, we may discover new genes affecting pigment biosynthesis as well as regulatory regions that determine when and where pigmentation will develop [3,13].
We used the D. melanogaster Genetic Reference Panel (DGRP) to perform a genome-wide association (GWA) study of natural variation in the proportion of melanization on female abdominal tergites 5 and 6. The DGRP consists of 205 sequenced inbred lines derived from a single North American population, facilitating GWA analyses for quantitative traits when all genetic variants are known. Local linkage disequilibrium (LD) in the DGRP is low and thus favorable for identifying candidate genes and even causal polymorphisms [14,15]. We identified single nucleotide polymorphisms (SNPs) affecting three genes previously known to contribute to variation in abdominal pigmentation, bric-à-brac 1 (bab1), tan (t), and ebony (e). However, we also identified novel candidate genes and showed that these contribute to abdominal pigmentation using mutations and RNAi knock-down constructs. Many of these novel genes affect other well-studied pathways and phenotypes, such as wing and bristle development, providing evidence for widespread pleiotropy. Four of the novel genes affecting pigmentation are computationally predicted genes with previously unknown functions. Based on their mutant or RNAi knockdown phenotypes, we have named them pinstripe (pns, CG7852), triforce (tfc, CG9134), plush (ph, CG1887), and farmer (frm, CG10625).
We characterized natural variation in the proportion of melanization of tergites 5 (T5) and 6 (T6) in females for 175 DGRP lines (Figs 1 and 2 and S1 Table). Averaged across all lines, the mean pigmentation scores are 1.44 for T5 and 2.55 for T6 (Fig 2A). There is significant genetic variation in pigmentation among lines for both tergites (PT5 = 4.68 x 10–48 and PT6 = 6.65 x 10–96; S2 Table), with broad sense heritabilities (H2) of H2T5 = 0.66 and H2T6 = 0.88. The phenotypic (rP(T5,T6) = 0.63 ± 0.059) and genetic (rG(T5,T6) = 0.72 ± 0.053) correlations (± standard error) between the tergites for proportion of pigmentation are high but significantly different from unity, suggesting they have different genetic bases (Fig 2B). The high broad sense heritabilities for abdominal pigmentation traits provide a favorable scenario for GWA studies.
We performed genome-wide association analyses on the proportion of T5 and T6 melanization to identify genomic regions harboring variants contributing to natural variation in female abdominal pigmentation. The DGRP lines vary in Wolbachia infection status and karyotype for several common polymorphic inversions. We did not find significant associations of Wolbachia infection (PT5 = 0.58 and PT6 = 0.92) nor inversion karyotype on T5 or T6 pigmentation; however, the difference in pigmentation between T5 and T6 was significantly affected by ln(2L)t (P = 0.04) and In(2R)NS (P = 0.01) (S3 Table). For each GWA analysis, we used both a mixed model that accounted for any effects of Wolbachia, inversions, and cryptic relatedness and a regression model that corrected for all of the aforementioned effects except for cryptic relatedness [15]. Combining all of these models, we identified a total of 155 variants associated with pigmentation for any trait at a nominal reporting threshold of P < 10–5 (S4 Table). Of these, 84 were associated with T5 pigmentation, 34 with T6 pigmentation, 28 with the average of T5 and T6, and 35 with the difference in pigmentation between T5 and T6. A total of 84 candidate genes were implicated by these associated variants. Since variants associated with the average of the two posterior tergites were largely the same as those associated with either T5 or T6 alone, we focus our subsequent analyses on T5, T6 and the difference between them (Fig 3 and S4 and S5 Tables). Among the genes harboring SNPs associated with variation in abdominal pigmentation, we find genes with well documented effects on pigmentation (t, e, bab1); osa, a transcription factor recently shown to affect pigmentation; and a large group of novel candidate genes [2,9,16]. The identification of t, e, and bab1 as prominent contributors to variation in abdominal pigmentation instills confidence in the efficacy of our GWA analyses, as described below.
Only a few variants exceeded a strict Bonferroni correction for multiple tests (P = 2.64 × 10–8): a SNP 41 bp upstream of Gr8a and 528 bp downstream of CG15370—the cis-regulatory region of t—in the T6 and average of T5 and T6 analyses (X_9121129_SNP); and two SNPs in the first intron of bab1 in the analysis of the difference between T5 and T6 (3L_1084990_SNP and 3L_1084199_SNP; S4 Table). The three SNPs that achieved Bonferroni significance levels were all at intermediate frequency and had moderately large effects. The minor allele of the polymorphism in the t cis-regulatory element (CRE) was associated with reduced pigmentation, while the minor alleles of the bab1 intronic polymorphisms were both associated with increased pigmentation in T6.
Although the other variants do not reach individual Bonferroni-corrected significance levels, quantile-quantile plots (S1 Fig) indicate a systematic departure from random expectation below P < 10–5, justifying our choice of this reporting threshold and suggesting that the top associations are enriched for true positives. Indeed, the SNP in the t CRE that reached Bonferroni significance in the T6 analysis was also significant in the T5 analysis at the more lenient reporting threshold, and two additional polymorphisms in the t CRE were significant at P < 10–5: X_9121177_SNP in the T5 and T6 analyses, and X_9121094_SNP in the T6 analysis.
These data also highlight the importance of bab1 with respect to female abdominal pigmentation: we found a total of 21 polymorphisms (20 SNPs, one indel) in the first intron of this gene that are associated with natural variation in pigmentation in one or more analyses (Fig 4 and S4 Table). One bab1 SNP is unique to the T5 analysis, six bab1 SNPs are common between the T6 and T5—T6 difference analysis, and the remaining bab1 variants are unique to the difference in pigmentation between T5 and T6. Twelve of the bab1 variants are located within the minimal functional cis-regulatory regions as reported by REDfly or within other transcription factor binding sites (S4 and S6 Tables) [17]. Three SNPs (3L_1084990_SNP, 3L_1085137_SNP, and 3L_1085230_SNP) are located in the bab1 middle dimorphic element which contain binding sites for the transcription factors caudal (cad) and dl (dorsal) [3]. All of the polymorphisms segregating in bab1 associated with pigmentation have minor allele frequencies ranging from 0.22 to 0.49 and moderate effects. Interestingly, the direction of the effects is both positive (the minor allele is associated with reduced pigmentation) and negative (the minor allele is associated with increased pigmentation), such that variants in the dl and cad cis-regulatory modules have positive effects while those in the latter regions of the intron have largely negative effects (the exceptions are 3L_1093297_SNP in the latter region of the intron and 3L_1099962_SNP in the T5 GWAS, which have positive effects). The functionality of these bab1 CREs has been thoroughly investigated [3]. However, similar to the results of Bickel et al. [18], nine of the bab1 variants from this study are in regions outside of the known cis-regulatory regions. These variants may indicate the presence of a not-yet-described regulatory element, or the structure of the regulatory elements in this region may be more complex than previously thought (Fig 4).
A majority of the variants associated with variation in pigmentation are located within intronic or intergenic regions, suggesting they could affect gene regulation. In support of this hypothesis, we found many of these variants are located in annotated regulatory sites (S4 and S6 Tables). In total, variants associated with pigmentation were located in 24 different transcription factor binding sites (TFBS, each of which contain numerous variants), 17 cis-regulatory modules (CRM), 1 polycomb response element (PRE), and 31 hot spot analysis sites (HSA; where one or more 41 tested TFs bind to a given site) (S6 Table). TFBS for dl and cad are the most frequent of all TFBS, containing 28 and 22 associated variants, respectively. Two intergenic TFBS for bab1 (FBsf0000214860 and FBsf0000214320) were tagged by 3R_25139342_SNP, 3R_25139132_SNP, and 2R_16793853_SNP. A few variants are located in more than one regulatory site (S4 Table).
We asked what fraction of the total broad sense heritability was explained by variants in bab1, t and e using stepwise regression to select the top associations for pigmentation genes. The R2 from these models for each trait gives the heritability explained by the known genes. These loci explain 25.62%, 37.55%, 31.17% and 36.58% of the heritability for T5, T6, and the average and difference of T5 and T6, respectively; consistent with the intermediate allele frequencies and large effects of their top associated variants. Next, we used genomic best linear unbiased prediction (GBLUP) to estimate the total variance explained by all top variants. All variants explain 59.77%, 34.32%, 47.44% and 51.61% of the heritability for T5, T6, and the average and difference of T5 and T6, respectively. With the exception of T6, for which most of the variance is explained by the known pigmentation genes, substantial additional variance in proportion of pigmented cuticle is contributed by variants in novel genes. Finally, we estimated the faction of heritability explained for each variant as well as the fraction of heritability explained after accounting for the variance explained by the pigmentation genes. On average, the variants in novel candidate genes explained an additional 7.3% (T5), 4.5% (T6), 5.8% (average of T5 and T6) and 2.8% (difference between T5 and T6) of the heritability (S2 Fig and S4 Table)
e, t, and bab1 have been associated with variation in D. melanogaster female pigmentation in other populations [2,9,16]. We compared the top variants in these genes in our analyses with those from prior studies [9,16,19]. Three of the four SNPs identified by Bastide et al. [9] are identical to the three t CRE SNPs associated with our T5 and T6 analyses (X_9121094_SNP, X_9121129_SNP, and X_9121177_SNP). The bab1 SNP identified in the Bastide et al. [9] study did not overlap with our results nor those of Bickel et al. [18]. None of the top bab1 variants in this study were significant in the study of Bickel et al. [18], although three of our significant variants were also polymorphic in the Bickel data set (3L_1085788_SNP, 3L_1086799_SNP, and 3L_1086802_INS). Both the e CRE SNPs associated with pigmentation in the DGRP T6 analysis (3R_17063120_SNP) and the Bastide et al. study [9] (3R_17064232_SNP) are located within the CRE regulating e expression in the haltere (e_C [19]; S4 Table). The SNP at 3R_17064232 was also reported in the Pool and Aquadro study of light and dark African D. melanogaster [16]. The concordance among these datasets indicates that the haltere regulatory element may also control expression in the abdomen and warrants further investigation.
We selected 30 novel candidate genes based on the GWA results for functional validation using mutant alleles and RNAi knockdown (S7 Table). We phenotyped Exelixis insertion lines [20] and RNAi knockdown lines [21] with their appropriate controls for the proportion of melanization on T5 and T6 as done for the DGRP (S8 and S9 Tables). Wherever possible we tested both mutant and RNAi lines for the same gene as independent forms of validation. We used three GAL4 drivers for the UAS-RNAi lines. tubulin-GAL4/TM3, Sb (tub-GAL4) and ubiquitin-GAL4/CyO (ubi-GAL4) are ubiquitously expressed, while the pannier driver, y1 w1118; P{w[+mW.hs] = GawB}pnrMD237/TM3, P{w[+mC] = UAS-y.C}MC2, Ser1 (pnr-GAL4), has restricted expression in the midline [22]. The use of the pnr-GAL4 driver adds a spatial component to the validation experiments and allows for more precise testing of the candidate genes (S3 Fig). As positive controls, we also tested RNAi constructs for e and t (S3 Fig and S8 and S9 Tables).
We evaluated 15 Exelixis transposon insertions in candidate genes for effects on pigmentation (See Methods; Fig 5A and S7 Table). Six of these mutations affected the proportion of melanization on T5 (P < 0.0001 for all significant mutations): CG9134e00088, CG7852c04511, Exchange factor for arf6 (Efa6f03476), Fish-lips (Filif04573), and Glucose transporter 1 (Glut1d05758) showed increased melanization; and krotzkopf verkehrt (kkvc06225) showed decreased melanization (Figs 5A and 6). Twelve of the mutations affected the proportion of melanization on T6 (P < 0.0001 for all significant mutations). CG9134e0008, Efa6f03476, Filif04573, and Glut1d05758 showed increased melanization; and CG10625e01211, CG7852c04511, division abnormally delayed (dallyf01097), kayak (kayf02002), kkvc06225, klarsicht (klard05910), locomotion defects (locod09879), and Sucbe01940 showed decreased melanization (Figs 5A and 6). CG7852c04511 had increased pigmentation on T5 and decreased pigmentation on T6 (Fig 6C). CG33298d10678a, multiple wing hairs (mwhd01620), and Kinesin-like protein at 61F (Klp61Ff02870) were not significantly different from the control. Efa6f03476 also has a light and somewhat elongated thoracic trident; this thoracic pigmentation is completely absent in the control flies (S4 Fig).
Of the 28 candidate genes, 26 were available as RNAi knockdown constructs (S7 and S8 Tables). We crossed all of these constructs to the pnr-GAL4 driver, and obtained viable female progeny from all crosses except for kkv and Fili. We found that seven of these knockdown mutations affected the proportion of melanization of T5 and/or T6 (Figs 5B and 7A–7I, P < 0.0001 in all cases). Efa6, klar, and Klp61F knockdowns increased the proportion of dark melanin on T6; buttonless (btn) and CG7852 decreased it. Knockdown of roughoid (ru) and sinuous (sinu) showed decreases in pigmentation for both T5 and T6.
We also crossed the 26 UAS-RNAi constructs to an ubiquitously expressed tub-GAL4 driver, and found that 11 (42%) were lethal in both sexes, suggesting pleiotropic effects on vital functions: Vesicular monoamine transporter (Vmat), Klp61F, CG7852, CG1887, klar, ru, sinu, Nedd2-like caspase (Nc), kkv, kay, and CG42340 (Table 1). Of the 15 UAS-RNAi/tub-GAL4 knockdown alleles available for testing, six had significant (P < 0.0001) effects on pigmentation. Knockdowns of btn, CG10625, and CG9134 had decreased proportions of dark melanin on T5 and T6; Efa6 and loco knockdowns showed increases in pigmentation on both tergites (Figs 5C and 7).
Next, we crossed the 11 UAS-RNAi constructs that were lethal when crossed to the tub-GAL4 driver to another ubiquitously expressed GAL4 driver, ubi-GAL4. Only two RNAi constructs were viable when driven by ubi-GAL4, CG1887 and klar, and both had significant (P < 0.0001) effects on abdominal pigmentation (Figs 5D and 7 and Table 1). The CG1887 knockdown showed a decrease in T5 melanization. Although T6 did not show a significant difference in the proportion of melanin in the CG1887 knockdown (P = 0.71), the dark melanin that is present is a light brown melanin that is only slightly darker than the adjacent sclerotin (S5 Fig). The CG1887 knockdown flies have obvious qualitative differences in overall body coloration from the control. The cuticle as a whole is semi-transparent and its strength is compromised as it ruptures easily when probed with forceps. The third thoracic leg of these progeny is also malformed and bristle number and patterning is highly disrupted (S5 Fig). All progeny of the cross die within 24 hours of eclosion; thus, pigmentation scoring was performed 8 hours after eclosion. The klar knockdown shows an increase of melanization on both tergites. Similar to the Efa6 mutant, this cross leads to a relative darkening of the thoracic trident compared to the surrounding cuticle (S4 Fig). In summary, we found that 17 of the 28 candidate genes tested affected female abdominal pigmentation and that for 12 of these genes, both tergites are affected (Table 1 and S8 and S9 Tables).
The DGRP lines show substantial natural variation in female abdominal pigmentation, ranging from lines with no dark melanin to complete melanization on tergites 5 and 6. Despite being sampled from a single population, the lines span the range of pigmentation difference between the well-studied sister species D. yakuba and D. santomea. D. santomea is the lightest member of the D. melanogaster species subgroup; however, several of the DGRP lines are lighter than D. santomea (e.g., DGRP_441, Fig 1C). Utilizing the substantial genetic variation and a mapping population that is powerful to detect common variants associated with the variation, especially those with moderate to large effects (S6 Fig), we identified a total of 155 genetic variants associated with variation in female abdominal pigmentation using GWA analyses.
We identified variants in four genes previously shown to affect adult D. melanogaster pigmentation: bab1, t, e, and osa [2,16,23,24]. Most of the bab1 SNPs were associated only with the difference between T5 and T6 pigmentation, suggesting that variation in bab1 may underlie the genetic and phenotypic correlations between the traits. Most of the bab1 and t minor alleles are at moderate frequencies in the DGRP. These SNPs could be neutral or could be maintained segregating by a balance of unknown selective forces. We identified three SNPs in the CRE of t that were also found in the European populations studied by Bastide et al. [9], indicating that these SNPs are maintained in distant populations. Of note, our study is the first to associate natural variation in pigmentation with genetic variation in the transcription factor osa.
A majority of the variants identified in this study are in intronic or intergenic regions. Among the total regulatory elements identified, dl and cad binding sites were the most highly represented, suggesting a role for these two TFs on pigmentation patterning. Over half of the genetic variants located within bab1 were in known regulatory regions including some for dl and cad binding. We also identified a SNP upstream of e that is located within a cis-regulatory region consistent with the studies of African and European D. melanogaster [9,16,19]. These results implicate cis-regulatory evolution, which likely limits negative pleiotropic effects, as a major contributor to phenotypic variation within the DGRP population.
In addition to genes previously known to affect Drosophila pigmentation, we identified many novel candidate genes. We showed that 61% of the candidate genes affect the proportion of dark pigmentation on tergites 5 and 6 using mutant and RNAi knockdown experiments. These genes are known to be involved in processes such as sugar binding and transport, vesicle formation and transport, and cuticle formation. We summarize what is known from the literature about these candidate genes below and speculate about their roles in pigmentation and phenotypic evolution.
Prior to molting and eclosion, insects accumulate tyrosine-derivatives conjugated with hydrophilic molecules such as glucose, phosophate, and sulfate in the hemolymph. This keeps the reactive pigment precursors in an inert state until the organism is ready to molt or eclose [13,25–29]. We identified two genes, triforce (tfc, CG9134) and Glut1, which may facilitate the transport of glucose or glucose conjugates from the hemolymph to the epidermal cells. tfc is a C-type lectin with a carbohydrate binding domain and Glut1 is a membrane bound glucose transporter [30].
Several of our new pigmentation genes have roles in relatively well-described developmental pathways. These include kay, dally, Fili, and ru. kay is a transcription factor in the JNK signal transduction pathway [31]. It is required for decapentaplegic (dpp) expression, wing and leg development, and the elongation of the cells in the epidermis [32]. dpp expression in the tergite corresponds to the midline pigment stripe, and ectopic expression of dpp in pupae expands this stripe [32]. Furthermore, dpp signal transduction is potentiated by dally [33]. Additionally, dpp and Epidermal growth factor receptor (Egfr) signaling work synergistically to specify the lateral tergite cell fate [32. ru, also known as Rhomboid-3 (Rho-3), activates Egfr signaling and thus may help determine epidermal cell fate in the developing tergites; however, there are several Rho proteins capable of this activation [34]. Fili is a transmembrane protein that is involved with apoptosis in the wing imaginal disc and we speculate may facilitate proper tergite differentiation during metamorphosis [35].
Four validated candidate genes are involved with vesicle formation and transport: pinstripe (pns, CG7852, which describes the vertical stripe of pigment remaining on T6 in the RNAi knockdown), Efa6, Klp61F, and klar. pns is predicted to have Rab guanyl-nucleotide exchange factor activity, which facilitates vesicle transport by activating Rab proteins [36,37]. Rab5 works in conjunction with Megalin to remove the Yellow protein from the tanning D. melanogaster wing [38]. Efa6 is also involved with Rab signaling and vesicle mediated transport [36]. Klp61F is a kinesin motor, and klarsicht (klar) regulates microtubule-dependent vesicle transport along microtubules. Both could be involved in transporting vesicles of enzymes and/or dopamine-derivatives to and from the cuticle. Together these genes may represent components of the little-known transport mechanism for cuticle tanning in D. melanogaster.
farmer (frm, CG10625—the not quite fully developed stripes of pigment on tergites 5 and 6 are similar to the tan lines on the arms of a farmer after much time spent in the sun), kkv, and sinu, and loco may affect cuticle development and structure. frm is a cuticle structural protein [36]. kkv is one of two chitin synthase genes in D. melanogaster [39]. sinu is a claudin required for septate junction organization, cell-cell adhesion, and proper localization of proteins involved in chitin filament assembly in D. melanogaster [40]. loco regulates G protein signaling, and Gγ1 signaling is required for proper septate junction protein localization including sinu [41,42]. These proteins may help maintain the structural integrity of the adult cuticle and our study shows that when perturbed, they affect pigmentation.
Sucb and ph may affect the organism more broadly. Sucb is a succinate-CoA synthetase in the Krebs cycle [36]. It is plausible that variation in energy production due to genetic variation in key enzymes could indirectly affect variation in adult pigmentation by altering resources available for cuticle development. ph is a CD36 homologue, and dipteran CD36 family members may have roles in transport of cholesterol and steroids during ecdysone synthesis [43]. Since ecdysone is required for proper insect development and molting, disrupted transport of ecdysone precursors may explain the severe RNAi phenotype for this gene.
In summary, we provide evidence that genetic variation at a number of steps in regulatory, developmental, and transport may pathways contribute to natural variation in abdominal pigmentation. These findings exemplify the pleiotropic nature of these genes which may limit their potential as adaptive targets [44–46]. Several of the mutant and RNAi knockdown lines were lethal or had strong debilitating effects providing some support of this. It is known that the large-effect genes, t and e, are pleiotropic as well [23]. The difference may be that not all pigmentation genes have the necessary regulatory modules to alleviate pleiotropic effects. However, these candidate genes may contain tissue- or stage-specific gene regulatory architectures since most of the GWAS associated SNPs are in intronic and intergenic regions. Furthermore, a distinction should be made between pleiotropic genes and pleiotropic variants. A given gene may be pleiotropic, while particular variants within that gene may not be [47]. Additionally, in the DGRP lines allelic variants are homozygous. In nature, alleles with detrimental effects may be tempered in the heterozygous state or epistatic interactions may arise with differing combinations of polymorphic loci.
Consistent with other DGRP studies, we identified many more genetic variants associated with pigmentation than previous studies [48–51]. We suspect earlier studies may have only had the ability to identify the major effect loci and missed the more polygenic variation at these other loci. Most used only a small number of fly lines and thus interrogated a smaller sample of allelic variation, analyzed only known pigmentation genes, or the limited sample size of the mapping population gave reduced statistical power to detect variants with smaller effects. For example, the Winter's Lines, a panel of 144 recombinant inbred lines used to map the effect of bab1 on pigmentation in female D. melanogaster, were generated from only two gravid wild caught females [2]. The study of Bastide et al. [9] pooled individuals with extreme phenotypes for sequencing. This approach may filter out variants that lead to intermediate phenotypes and select for large effect variants. Pool and Aquadro focused only on e sequences among the 25 African lines eliminating any possibility of identifying additional loci [16]. The DGRP is more representative of natural populations and harbors many more polymorphic loci that may contribute to phenotypic variation and evolution [52]. Given both the population sample and the genome-wide coverage of polymorphisms, this study is perhaps the most comprehensive analysis of variation in Drosophila pigmentation to date.
It is important to acknowledge that gene expression knockdown and mutant analyses are only an approximate confirmation of causative SNPs. Genes implicated by the GWA analyses that do not confirm in these functional tests may be true positives and contribute to variation in pigmentation, but they do not change pigmentation when gene expression is reduced. Future studies should test effects of individual SNPs and further characterize the mechanisms though which the candidate genes affect variation in pigmentation; their potential interactions with variants in the candidate genes with major effects; and their allele frequency distributions in different populations. These studies will help elucidate the contribution of these novel variants to adaptive phenotypic evolution or whether they are population-specific deleterious variants that are maintained segregating by mutation-selection balance.
Our results open the door for new hypotheses to be tested on the transport of dopamine derivatives and conjugates from the hemolymph to the cuticle, the formation and movement of vesicles within epidermal cells, the mechanisms of regulatory and developmental pathways during tergite differentiation, the interactions of chitin filaments with cell adhesion and cuticle proteins, and how metabolic and hormonal regulation could lead to variation in pigmentation. Genetic variants that affect these processes could potentially serve as targets of adaptive evolution or sexual selection in natural populations. This study is a start. However, much more work is needed to draw mechanistic inferences about these novel candidate genes and their contributions to the evolution of pigmentation.
The DGRP consists of 205 inbred lines with complete genome sequences. We scored female flies of 175 DGRP lines—aged 5 to 8 days—for the proportion of melanization on abdominal tergites 5 and 6. Two independent replicates for each DGRP line were reared and five individuals were scored from each replicate vial (N = 10 flies per line). The flies were reared in vials at a controlled adult density (CAD) of 10 males and 10 females on cornmeal-molasses-agar medium at 25°C, 75% relative humidity, and a 12-h light-dark cycle. The parental generation was allowed to lay eggs for 3 days. Each fly was visually assessed by a single observer for the percentage of brown/black melanin covering each tergite; the scores ranged from 0 for no dark pigmentation to 4 for 100% dark pigmentation in increments of 0.5.
We partitioned variation in pigmentation into genetic and environmental components using an ANOVA model of form Y = μ + L + T + L×T + R(L×T) + ε, where Y is phenotype, μ is the overall mean, L is the random effect of line, T is the fixed effect of tergite, R is the random effect of replicate vial, and ε is the residual. We also performed reduced ANOVAs separately for each tergite of form Y = μ + L + R (L) + ε. We estimated variance components for the random effects using REML. We computed the broad-sense heritability (H2) of pigmentation for each tergite separately as H2 = σ2L/ (σ2L + σ2ε), where σ2L is the among-line variance component and σ2ε is the error variance. We computed the genetic correlation between the tergites (rT5,T6) as CovT5,T6/ σLT5σLT6, where CovT5,T6 is the covariance in pigmentation score between tergites 5 and 6. All analyses were performed with version 9.3 of the SAS System for Windows (2013 SAS Institute Inc.).
To identify genomic regions harboring variants contributing to natural variation in the proportion of tergite melanization, we conducted a GWA study for each tergite. The DGRP lines are also segregating for Wolbachia infection and for the following common inversions: In(2L)t, In(2R)NS, In(3R)P, In(3R)K, and In(3R)Mo. We performed GWA studies in two stages. In the first stage, we adjusted the line means for the effects of Wolbachia infection and major inversions. We then used the adjusted line means to fit a linear mixed model in the form of Y = Xb + Zu + e, where Y is the adjusted phenotypic values, X is the design matrix for the fixed SNP effect b, Z is the incidence matrix for the random polygenic effect u, and e is the residual. The vector of polygenic effects u has a covariance matrix in the form of Aσ2, where σ2 is the polygenic variance component. We fitted this linear mixed model using the FastLMM program (version 1.09) [53]. We performed these single marker analyses for the 1,897,337 biallelic variants (SNPs and indels) with minor allele frequencies ≥ 0.05 whose Phred scale quality scores were at least 500 and genotypes whose sequencing depths were at least one and genotype quality scores at least 20 [15]. All segregating sites within lines were treated as missing data. Additionally, we performed single marker tests for association on line means that were adjusted for the effects of Wolbachia infection and major inversions but not corrected for the relationship matrix. Significant variants were annotated using the 5.49 Release of the Flybase annotations.
For each variant, we calculated two variance components. First, to calculate the variance explained by a variant without adjusting for variants in known pigmentation genes, we fitted a linear model for the adjusted line means for only the focal variant and used the R2 of the model to represent the variance explained by it. Second, to calculate additional variance explained by a focal variant after accounting for variants in known pigmentation genes, we first used stepwise selection to select the top associations for each pigmentation gene (tan, ebony, or bab1), requiring P-values to be smaller than 10–5 if more than one variant entered the model, and no P-value requirement if there was only one variant. The R2 of this baseline model (different for each of the four traits) is the variance explained by the pigmentation genes. We added each focal variant to the baseline model and calculated the difference between the R2 of the new model and the R2 of the baseline model, which represented the additional variance explained by the variant after accounting for the pigmentation genes. To calculate total variance explained by all significant variants, we used a mixed model approach because of the large number of variants. We computed the variance/covariance matrix based on the genotype matrix and estimated the variance components using the rrBLUP R package.
We tested 12 of the 13 genes implicated by the T6 pigmentation GWA analysis, none of which were previously known to affect variation in pigmentation in D. melanogaster: CG33298, Fili, Vmat, mwh, Klp61F, CG9134, CG7852, CG1887, klar, Glut1, Efa6, and btn. From the T5 pigmentation GWA analysis, we selected candidate genes that (1) had an Exelixis mutant [20] or VDRC RNAi [21] line available at the time of the study; (2) are involved in development, especially of the cuticle or epidermis, or pigmentation according to FlyBase and the available literature; and (3) show mRNA expression patterns similar to the regulatory genes, bab1 and Dsx, and genes in the pigmentation biosynthesis pathway (such as t and e), a peak of expression at 24 hr after puparium formation and 2–4 days after puparium formation, respectively, according to the modENCODE tissue and temporal expression data [27,52]. This resulted in 16 additional candidate genes: ru, CG10625, sinu, Sucb, dally, CG32052, Nc, Cerk, kkv, CG15803, loco, TwdlC, kay, CG1340, CG42594, and CG42340. For each candidate gene, we tested either an Exelixis transposon insertion line [20], a VDRC RNAi line [21], or when possible, both a mutation and RNAi construct. We assessed the proportion of melanization for both T5 and T6 for all candidate genes.
We evaluated 15 Exelixis transposon insertion lines: CG33298d10678a, Filif04573, mwhd01620, Klp61Ff02870, CG9134e00088, CG7852c04511, klard05910, Glut1d05758, Efa6f03476, CG10625e01211, Sucbe01940, dallyf01097, kkvc06225, locod09879, and kayf02002. The Exelixis progenitor w1118 line was used as a control. The KK and GD library progenitor lines were used to make control crosses for the RNAi knockdown experiments. Males from the GAL4 driver line were crossed to virgin females of the VDRC UAS line for all crosses. Three GAL4 driver lines were used for the RNAi crosses. All VDRC UAS lines were crossed with the full-body tubulin-GAL4/Sb driver and a pannier-GAL4 driver (y1 w1118; P{w+mW.hs = GawB}pnrMD237/TM3, P{w+mC = UAS-y.C}MC2, Ser1). In instances of lethality with the tubulin-GAL4/Sb driver, the lines were crossed to another full-body ubiquitin-GAL4/Cy driver. All GAL4-driver lines were obtained from the Bloomington, Indiana Drosophila Stock Center. We tested in total 26 RNAi knockdown constructs: CG33298, Fili, Vmat, mwh, Klp61F, CG9134, CG7852, CG1887, klar, Efa6, btn, ru, CG10625, sinu, dally, CG32052, Nc, Cerk, kkv, CG15803, loco, TwdlC, kay, CG1340, CG42594, and CG42340.
We reared three independent replicates for each Exelixis transposon insertion line, for each RNAi cross and for the appropriate controls under the same conditions as the DGRP lines, but in 8 oz. bottles with a controlled adult density of 20 males and 20 virgin females. We scored the proportion of melanization on T5 and T6 for 50 5–8 day old female progeny per replicate (N = 150 flies per genotype) from each Exelixis line or RNAi cross. In a few instances where viability was low fewer than 50 individuals per replicate were scored: pnr-GAL4 x sinu (N = 23), ubi-GAL4 x CG1887 (N = 90), and ubi-GAL4 x klar (N = 95). Means of test lines were compared to those of the appropriate controls with a Dunnett's test, which corrects for multiple testing, using JMP Pro 10.0.0 (2012 SAS Institute Inc.)
After mutant lines and RNAi knockdown progeny were scored for pigmentation, they were preserved in a 3:1 ethanol/glycerol solution and stored at 4°C until dissection for imaging. The fly cuticles were dissected from the abdomen and mounted to a glass slide using Permount and a glass cover slip. All photographs were taken with an Olympus DP25 microscope camera on an Olympus SZ61 stereo microscope.
|
10.1371/journal.pgen.1006441 | The Gene Regulatory Network of Lens Induction Is Wired through Meis-Dependent Shadow Enhancers of Pax6 | Lens induction is a classical developmental model allowing investigation of cell specification, spatiotemporal control of gene expression, as well as how transcription factors are integrated into highly complex gene regulatory networks (GRNs). Pax6 represents a key node in the gene regulatory network governing mammalian lens induction. Meis1 and Meis2 homeoproteins are considered as essential upstream regulators of Pax6 during lens morphogenesis based on their interaction with the ectoderm enhancer (EE) located upstream of Pax6 transcription start site. Despite this generally accepted regulatory pathway, Meis1-, Meis2- and EE-deficient mice have surprisingly mild eye phenotypes at placodal stage of lens development. Here, we show that simultaneous deletion of Meis1 and Meis2 in presumptive lens ectoderm results in arrested lens development in the pre-placodal stage, and neither lens placode nor lens is formed. We found that in the presumptive lens ectoderm of Meis1/Meis2 deficient embryos Pax6 expression is absent. We demonstrate using chromatin immunoprecipitation (ChIP) that in addition to EE, Meis homeoproteins bind to a remote, ultraconserved SIMO enhancer of Pax6. We further show, using in vivo gene reporter analyses, that the lens-specific activity of SIMO enhancer is dependent on the presence of three Meis binding sites, phylogenetically conserved from man to zebrafish. Genetic ablation of EE and SIMO enhancers demostrates their requirement for lens induction and uncovers an apparent redundancy at early stages of lens development. These findings identify a genetic requirement for Meis1 and Meis2 during the early steps of mammalian eye development. Moreover, they reveal an apparent robustness in the gene regulatory mechanism whereby two independent "shadow enhancers" maintain critical levels of a dosage-sensitive gene, Pax6, during lens induction.
| While significant insights into the functional role of some transcription factors during lens formation have been accomplished, much less is known about the intricate wiring of the gene regulatory network (GRN) that controls the earliest stages of lens development. Our genetic experiments presented here demonstrate a fundamental and redundant role of Meis1 and Meis2 genes in the process of lens induction. Furthermore, we present evidence that the robustness and dose-dependent regulation of Pax6, a key node of lens GRN, occurs via employment of "shadow enhancers" powered by Meis transcription factors. Combined, this study significantly extends knowledge about the genetic wiring of the earliest stages of eye development.
| Cellular and molecular mechanisms of vertebrate lens development are objects of intense studies for many decades, reviewed in [1]. In particular, lens induction represents a classical developmental model allowing investigation of cell specification, spatiotemporal control of gene expression, as well as the integration of signaling pathways and transcription factors into highly complex gene regulatory network (GRN). At the end of neural plate formation, the vertebrate lens originates from the multipotent pre-placodal ectoderm [2, 3] through a series of cell-type specifications, governed by DNA-binding transcription factors Pax6, Six3 and Sox2, and including another transitional population of cells, the presumptive lens ectoderm (PLE). The PLE gives rise to the lens placode, readily observed as a thickening of the head surface ectoderm (SE) that is in close contact with the underlying optic vesicle, an evaginating part of the future diencephalon. Genetic dissection of lens induction has mainly focused on the function of Pax6, Six3 and Sox2, coupled with studies of BMP, retinoic acid and Wnt signaling in the surface ectoderm, neuroectoderm, and surrounding periocular mesenchyme, reviewed in [1]. Pax6-deficient (Pax6 Sey/Sey) mice are anophthalmic with eye development arrested at the optic vesicle stage [4–6]. Numerous studies have shown that Pax6 is essential for lens formation through its expression in the SE and PLE, and in the subsequent stages of lens placode formation [7–9]. In contrast, the role of Six3 and Sox2 are less clear, although it is known these factors play major roles in anterior forebrain development and optic cup formation [10–12], further enforcing Pax6 as an ideal node to decipher genetic wiring of lens induction. Despite a well-established genetic role, much less is known about the factors operating upstream of Pax6 and their interaction with cis-regulatory elements that direct Pax6 expression to the lens ectoderm. Since lens development is sensitive to Pax6 dosage [4] accurate regulation of Pax6 expression level during lens development is therefore of great importance.
Transcriptional control of Pax6 gene expression is very complex and different cells and tissues choose specific promoters and distal regulatory regions from an archipelago of enhancers scattered within the large Pax6 genomic region [13, 14]. The expression of Pax6 in lens ectoderm was initially shown to be driven by an ectoderm enhancer (EE) located approximately 4kb upstream of the Pax6 P0 promotor [15, 16]. However, genetic studies in which EE was inactivated provided strong evidence that EE is not the only regulatory element responsible for Pax6 expression in the lens placode [17]. Surprisingly, detectable expression of Pax6 in lens placode of EE mutants remains. In fact, the relatively small reduction of Pax6 levels in EE mutants leads to only mild lens defects (such as a lens placode of reduced thickness and a small lens pit/vesicle) that do not phenocopy Pax6 deficiency in the PLE [7, 17] raising the possibility that additional regions compensate for the loss of EE. Genetic analysis of human aniridia patients has identified a highly conserved long-range cis-regulatory element called SIMO, located 150 kb downstream of Pax6 [18] that can also direct transgene expression to the developing lens [19, 20] suggesting a role as a tissue-specific enhancer. Mouse-human sequence conservation around the SIMO breakpoint revealed 85% nucleotide identity over a 1400 bp fragment with 500 bp core region showing 96% identity [20]. Recently, de novo point mutation within the SIMO region has been identified in patient suffering aniridia. This mutation disrupts an autoregulatory PAX6 binding site in SIMO, causing defective maintenance of PAX6 expression [19]. Remarkably, a Pax6 autoregulatory loop has also been described in the case of the EE [21]. While autoregulation of Pax6 is critical for lens cell-type identity, and represents a key mechanistic property of both Pax6 lens enhancers, such a mechanism does not address the critical issue, namely the identification of upstream regulators of Pax6. To date, functional interactions of Meis1/2, Prep1, Six3, Sox2 and Oct1 have only been demonstrated at the EE [22–25].
Three amino acid loop extension (TALE) homeobox genes are evolutionarily highly conserved developmental regulators present in both vertebrate and invertebrate genomes. In vertebrates, TALE homeoproteins are represented by the Pbx and Meis/Prep subfamilies. Pbx proteins interact with Prep and Meis through a conserved amino-terminal domain while an independent protein surfaces allow Pbx to form trimeric complexes with Prep or Meis and Hox, reviewed in [26]. Prep and Meis alone preferentially bind DNA motifs with the sequence TGACAG/A, whereas Prep-Pbx and Meis-Pbx dimers bind the sequence TGATTGACAG. In mouse and human, three Meis homologs (Meis1, Meis2 and Meis3) and two homologues of Prep (Prep1 and Prep2) have been identified. Genome-wide analysis of Meis and Prep binding sites using a ChIP-seq approach have revealed their substantial specialization as well as significant regulatory coordination between these factors [27]. Biochemical and transgenic reporter studies have implicated Meis1 and Meis2 in the regulation of the EE of Pax6 [22]. In addition, binding of Prep1 to the EE has been shown to control Pax6 levels and the timing of Pax6 activation in the developing lens [25]. However, Meis1 knockout mice exhibit only a mild lens phenotype at later developmental stages [28]. As Meis1 and Meis2 exhibit similar expression patterns during the early stages of lens development (detailed in this study) we hypothesized that they are genetically redundant. To test this hypothesis, we have generated a Meis2 floxed allele and subsequently investigated the effect of Meis2 and Meis1/Meis2 defficiency on lens development using a lens-specific deleter Le-Cre recombinase [7]. We provide genetic evidence that Meis2 alone is not essential for lens development, however combined depletion of both Meis1 and Meis2 proteins at the early stages of lens development demonstrate that Meis1/2 are redundantly required for lens placode formation. Chromatin immunoprecipitation and transgenic reporter studies further dissect the molecular mechanism of Meis-dependent regulation of Pax6 gene expression. Deletion of SIMO region by genomic engineering in vivo suggests its redundancy with EE and uncovers SIMO function in lens development. Moreover, simultaneous deletion of EE and SIMO in vivo resulting in loss of lens formation confirms the essential role of the two Pax6 enhancers for lens induction. Remarkably, our data demonstrate the existence of two independent and partially redundant Meis-dependent enhancers, with similar molecular architecture, involved in the regulation of Pax6 expression during lens placode formation, thereby providing an unexpected level of robustness to the system.
In this study, we sought to determine the genetic hierarchy during early lens development by investigating the role of Meis1 and Meis2 homeoproteins using knockout mice. In addition, we wanted to examine the extent of Meis-mediated regulation of the critical eye specification gene Pax6 during lens induction. It was previously shown that specific deletion of Pax6 in the PLE resulted in a failure of lens development from the lens placode stage onward [7]. The main prerequisite for transcriptional regulation of placodal Pax6 expression by Meis proteins is their co-expression in the same tissue. Immunoflourescence using specific antibodies against Meis1 and Meis2 [22, 29, 30] revealed that both proteins were expressed in developing lens: in the PLE, lens placode and later in the lens pit (S1A–S1F Fig). Moreover the expression pattern of both Meis1 and Meis2 were overlapping with Pax6 expression in the PLE [31].
Meis1 mutants (Meis1-/-) do not present with arrested lens development [28]. We therefore questioned whether deletion of Meis2 may affect lens development. Accordingly, mice containing a Meis2 floxed allele (Meis2f/f) were generated (S1G Fig) and [32], and subsequently zygotic Hprt1-Cre mice were employed to create whole-body knockout of Meis2 (Meis2-/-). Meis2-/- embryos displayed strong hemorrhage and other developmental defects and died by E14.5 [32]. However, lens development was not affected in these mutants (S2 Fig). To overcome the embryonic lethality of Meis2 whole-body knockout and to conditionally inactivate Meis2 specifically in PLE from E9.0, Le-Cre mice [7], (S1H and S1I Fig) were crossed with Meis2f/f mice. In Le-Cre;Meis2f/f embryos Meis2 protein was efficiently deleted in the lens placode and surface ectoderm at E9.5 (S1J Fig). We accordingly analyzed lens development in the absence of Meis1, Meis2 or both factors. The morphology of lens development was examined at stages E10.0 and E12.5 on tissue sections stained with hematoxylin-eosin. As shown in Fig 1, both Meis1 and Meis2 deficient embryos developed beyond the lens placode stage and subsequently and invariantly formed a lens. Therefore, we decided to generate embryos simultaneously deficient for both Meis1 and Meis2 in PLE; Le-Cre;Meis1-/-;Meis2f/f (referred thereafter as Meis1/Meis2 double mutant). Deletion of Meis1 and Meis2 in the PLE of Le-Cre;Meis1-/-;Meis2f/f embryos resulted in arrested lens development, characterized by a failure of the PLE to thicken and form the lens placode (Fig 1). Histological analysis at E12.5 confirmed an absence of lens tissue on a morphological level in all analyzed Meis1/Meis2 double mutants, where only folded retina was present (Fig 1O). Interestingly, one functional allele of Meis1 in Le-Cre;Meis1+/-;Meis2f/f embryos was sufficient to ensure lens placode and later lens formation, although the lenses were typically smaller (Fig 1N). These results demonstrate a requirement for Meis proteins during lens placode and subsequent lens formation.
To determine, whether the morphological arrest of lens development was accompanied by a loss of Pax6 expression and other lens placode markers, we performed immunofluorescent marker analyses at E10.0. Strikingly, we discovered a dramatic decrease in Pax6 expression in the PLE of Meis1/Meis2 double mutants (Fig 2A–2B’). In addition, the expression of the lens differentiating gene Foxe3, which is known to be highly Pax6-sensitive [33], was also not initiated (Fig 2C–2D’). Conversely, Sox2 expression persisted in the PLE of E10.0 Meis1/Meis2 double mutants (Fig 2E–2F’), which is consistent with Pax6-independent regulation of Sox2 at the lens placode stage [34]. Finally, Six3 expression that is mutually dependent on Pax6 expression in the PLE [23, 35], was also decreased in Meis1/Meis2 double mutants (Fig 2G–2H’). Immunofluorescent analysis of E12.5 Meis1/Meis2 double mutant embryos also confirmed the loss of α-crystallin-positive lens tissue, Prox1-positive differentiating lens fiber cells, Foxe3-positive lens epithelial cells and γ-crystallin-positive lens fiber cells (S3 Fig). Nevertheless, the presence of Pax6 and Sox2 proteins in the neural retina, and Otx2 in the retinal pigmented epithelium suggested that the specification of these tissues was not affected by the arrest of lens development (S3 Fig). Taken together, these results demonstrate that simultaneous inactivation of Meis1 and Meis2 results in early arrest of lens development and phenocopies Pax6 deficiency in the PLE [7].
A previous study has shown that Meis1 and Meis2 directly bind to the Pax6 ectoderm enhancer (EE) and thus control Pax6 expression during early vertebrate lens induction [22]. Here we show that the simultaneous inactivation of Meis1 and Meis2 leads to the dramatic downregulation of Pax6 in PLE and arrested lens development, in a manner reminiscent of that observed in Pax6 mutants [7]. However, as deletion of the EE does not phenocopy Pax6 loss [17], we hypothesized that Meis proteins might, in addition to the EE, interact with another enhancer such as the SIMO to drive appropriate levels of Pax6 expression in the developing lens. Thus, we focused on a 1400 bp evolutionarily conserved fragment of SIMO and used chromatin immunoprecipitation (ChIP) to analyse whether Meis proteins bound the SIMO element in vivo (Fig 3). We initially screened the 1400 bp fragment for the presence of Meis consensus binding site sequence motif, 5’ TGACAG/A 3’ [36], (Fig 3B). In the most conserved core region of the SIMO, we identified five Meis binding sites named SIMO_A, SIMO_B, SIMO_C, SIMO_D, SIMO_E with SIMO_B/C/D clustered in DNA region of 77 bp (Fig 3A). As a positive control for Meis binding ChIP analyses, we used the EE as it has been previously described to be bound by Meis [22] and as negative controls, the Axin2 promoter and Neurod1 coding sequences were used. Chromatin immunoprecipitation was performed on wild-type E10.5 embryos and the αTN-4 cell line [37] representing a model of mouse lens epithelial cells. qRT-PCR analysis of DNA fragments immunoprecipitated with mixture of Meis1+Meis2 specific antibodies from in E10.5 embryos showed significant enrichment at the EE as well as at the SIMO_B/C/D putative Meis-interacting sites (Fig 3C). No enrichment was observed at the negative controls regions or at the predicted Meis binding site SIMO_A. Similar results were also obtained when αTN-4 cells were used for immunoprecipitation (Fig 3D). Taken together these data show that Meis proteins bind the SIMO element in vivo and suggest that simultaneous binding of both the EE and SIMO may be required for appropriate Pax6 expression in the early lens.
To test the functional significance of identified Meis interactions with the SIMO enhancer we prepared reporter gene constructs expressing lacZ gene under the control of a minimal hsp68 promoter fused to the mouse SIMO enhancer (Fig 4A and 4B). To determine the specificity of any interactions, a single point mutation was introduced into Meis binding site that changed the recognition sequence from TGACAG/A into TcACAG/A. The same G to C mutation has previously been shown to abbrogate Meis binding and has been used in functional characterization of the EE and pancreatic enhancer in transgenic mouse models [22, 38]. In accordance with previous studies, FLAG-tagged Meis2 was able to specifically bind double-stranded oligonucleotides ancompassing wild-type Meis binding site but not its mutated version (S4 Fig). DNA constructs containing either the wild-type SIMO enhancer (SIMO WT) or the enhancer simultaneously mutated in conserved Meis binding sites SIMO_B, SIMO_C and SIMO_D (SIMO MUT), respectively, were introduced into the chick eye forming region by in ovo electroporation at embryonic stage HH10-11. The electroporated embryos were collected at stage HH20-21 and tested for β-galactosidase activity. As shown in Fig 4C and 4E and S5 Fig, wild-type SIMO enhancer mediated efficient expression of the lacZ reporter gene in the developing chick lens. In contrast, when all three Meis binding sites were mutated in SIMO, the lens-specific activity of the resulting reporter gene construct was abbrogated (Fig 4D and 4F and S5 Fig).
Next, we wanted to determine a possible contribution of individual Meis binding sites to SIMO enhancer activity. Mutation of SIMO_B Meis binding site alone resulted in decreased expression of reporter gene in lens as compared to wild-type SIMO, whereas simultaneous mutation of both SIMO_B and SIMO_C binding sites led to a complete loss of lens-specific expression of reporter gene (S6A Fig). These data suggest additive effect of three Meis binding sites on SIMO enhancer activity.
We noticed that Meis binding sites (sequence TGACAA in SIMO_B, SIMO_C and SIMO_D) in wild-type SIMO enhancer do not constitute the perfect match to the optimal Meis DNA-binding site motif TGACAG (http://jaspar.genereg.net/) indicating that they might represent a medium affinity sites.
In order to evaluate the functional significance of these non-optimal Meis binding sites for expression in lens we prepared reporter gene constructs expressing lacZ gene under the control of a minimal hsp68 promoter fused to the most conserved region of mouse SIMO enhancer (hereinafter referred to as minSIMO) containing either wild-type or optimized Meis binding sites. As shown in S6B Fig, substitution of wild-type Meis binding sequence in SIMO_B, SIMO_C and SIMO_D for optimal Meis binding sequence motif resulted in higher level of reporter activity in the developing lens. These data are in accord with the key functional role of Meis proteins in SIMO regulation and indicate that strong but restricted SIMO enhancer activity relies on a cluster of three medium affinity non-optimal Meis binding sites. Notably, recent systematic study of a model enhancer shows that enhancer specificity depends on a combination of suboptimal recognition motifs having reduced binding affinities. Conversion of suboptimal binding sites to perfect matches to consensus mediates robust but ectopic patterns of gene expression [39].
Finally, in order to gain further insight into enhancer architecture we used JASPAR database (http://jaspar.genereg.net/) to screen throughout the most evolutionarily conserved core region of SIMO (minSIMO region) for consensus binding sites of additional transcription factors. We identified potential binding sites for Six3, Ets/Tead, Maf and homeodomain-containing transcription factors (S6C Fig). We performed site-directed mutagenesis of SIMO introducing dinucleotide changes in the conserved residues of the consensus binding sites (LOGOs in JASPAR database). In addition, we mutagenized an evolutionarily conserved GCTC box present in SIMO of all species analyzed in Fig 3A. Reporter gene constructs expressing lacZ gene under the control of minimal hsp68 promoter fused either to the wild-type SIMO enhancer, or to the enhancer mutated in binding site for each particular transcription factor, were introduced into the chick eye forming region by in ovo electroporation at embryonic stage HH10-11. As shown in S6C Fig, none of the mutations resulted in a complete abbrogation of lens-specific reporter gene activity as did mutations in Meis binding sites SIMO_B and SIMO_C (S6A Fig). Notably, mutation of Six3 binding site resulted in decreased expression of reporter gene (S6C Fig), suggesting the requirement of similar Six3 input in SIMO enhancer as in EE [23]. Mutations in homeodomain binding sites HD1 and HD2 but not in HD3 lead to a subtle decrease of reporter activity (S6C Fig). Taken together, reporter gene assays in chick demonstrated an essential role of Meis transcription factors for SIMO enhancer activity.
Intrigued by the fact that Meis binding sites SIMO_B, SIMO_C and SIMO_D were phylogenetically conserved between mouse and zebrafish we next examined the functional significance of these sites in the context of zebrafish SIMO element. It was previously shown that the region encompassing zebrafish SIMO was able to drive expression to the lens of 48 hpf zebrafish [19]. We made a zebrafish EGFP reporter gene transgenic using wild-type and Meis-mutated versions of zebrafish SIMO element fused to minimal gata2a promoter (Fig 4G and 4H). In order to control for successful transgenesis and to quantitate results between the two constructs, ZED vector containing surrogate muscle-specific DsRed marker gene separated from EGFP reporter gene by an insulator was used [40]. In accordance with a previous study [19], transgenic fish carrying wild-type SIMO enhancer exhibited high level of EGFP in the lens at 48hpf (Fig 4I and 4J). In contrast, mutation of the phylogenetically conserved Meis binding sites resulted in the loss of EGFP due to the loss of lens-specific enhancer activity of SIMO while the muscle-specific surrogate reporter gene was still active (Fig 4K and 4L). These results suggest an evolutionarily conserved role of Meis proteins in the regulation of the Pax6 SIMO enhancer. Combined, our data establish that the SIMO enhancer is a natural target of Meis1 and Meis2 and that this physical interaction conveys expression of Pax6 in developing vertebrate lens.
In order to get an insight into SIMO function in vivo we generated mice carrying deletion of its evolutionarily conserved central core. Targeted engineering of genomic DNA in Pax6 locus was achieved using a pair of transcription activator-like effector nucleases (TALENs) designed to delete approximately 200 bp of the most evolutionarily conserved core region of SIMO (S7A Fig). Several lines of mice were established (S7B Fig) from which the line #710 designated Pax6SIMOdel710/+ was used for most of further studies. Enhancer region deleted in line #710 encompass Pax6 autoregulatory element and Meis1/2 binding sites SIMO_B, SIMO_C and SIMO_D, respectively, and is absolutely required for lens-specific activity based on transgenic reporter assay in chick (S7C Fig). To our surprize, mice carrying a homozygous deletion of SIMO (Pax6SIMOdel710/ SIMOdel710) did not manifest a major lens developmental phenotype (S7D Fig). To test whether lowering the dose of Pax6 may phenotypically uncover SIMO function during early lens development, we combined Pax6SIMOdel710/+ allele with Sey allele (Pax6 loss-of-function), (Fig 5). Under these conditions, only one allele of Pax6 carries SIMO enhancer deletion, while the second is genetically inactive in Sey. Although there are several lens phenotypes associated with the complete inactivation of one Pax6 allele in Sey mice, lens is always formed [5, 6], (Fig 5B). Remarkably, when the function of the second allele of Pax6 in Sey mice is compromised by SIMO deletion, lens development is arrested prior to lens pit stage (Fig 5B, the bottom panel) and no lens is detected in compound Pax6 heterozygote embryos at E13.5 (Fig 5B, the middle panel).
Finally, to demonstrate redundant role of Pax6 enhancers EE and SIMO for lens induction, we generated mice carrying deletion of both enhancers SIMO and EE simultaneously. For that purpose, we used CRISPR/Cas9 system to delete approximately 500 bp long critical region of EE [15, 16] on the Pax6SIMOdel710/SIMOdel710 genetic background. Several transgenic lines of Pax6ΔEE;ΔSIMO/ ΔEE;ΔSIMO mice were estabilished (hereinafter referred to as Pax6 EE/SIMO double mutant), from which line containing 477bp deletion of EE simultaneously with SIMO deletion was used for further analysis (Fig 6A). Histological analysis of mice lacking all four copies of lens enhancers at E11.0 revealed arrest of lens development prior to lens pit formation (Fig 6B). Immunofluorescent staining for lens marker Prox1 at E12.5 confirmed the absence of lens tissue in Pax6 EE/SIMO double mutant embryos (Fig 6B, the bottom panel). Remarkably, a single copy of a functional enhancer in Pax6ΔEE;ΔSIMO/ EE+;ΔSIMO embryo was sufficient for lens induction albeit the resulting lens was much smaller at E11.0 as compared to control and lens stalk was apparent in Pax6ΔEE;ΔSIMO/ EE+;ΔSIMO mice at E12.5 indicating delayed development (Fig 6B).
Genetic data indicated redundancy as well as potential additive activity of EE and SIMO. To provide further evidence that both EE and SIMO might be additively required for high level of Pax6 expression during lens induction we tested synergistic role of SIMO and EE on strength and specificity of expression of reporter genes in the developing chick lens. For that purpose we used reporter gene constructs expressing lacZ gene under the control of a minimal hsp68 promoter fused to either SIMO alone, EE alone, or combination of both enhancers (S8 Fig). As expected, combination of full-length EE [16] with SIMO elicited stronger expression of lacZ reporter gene than did SIMO alone (S8B Fig). Similarly, combination of minimal functional EE [15] with the most conserved region of SIMO (minSIMO) ensured stronger expression than did either of the minimal enhancers alone (S8C Fig). Strong and specific reporter gene activity may also be achieved by duplication of the same type of enhancer (S8C Fig). Reporter gene assays suggest that simultaneous use of both EE and SIMO enhancers may be beneficial for achieving high-level tissue-specific Pax6 gene expression during lens induction.
Combined, our data demonstrate simultaneous requirement of EE and SIMO Pax6 enhancers for normal lens development and provide evidence of their apparent redundancy and synergistic activity at early stages of lens induction.
GRNs provide a system level explanation of development in terms of the genomic regulatory code [41, 42]. While significant insights into the functional role of many transcription factors during the lens placode formation have been realised, much less is known about the upstream regulation of these critical factors and the intricate wiring of the GRN that controls the earliest stages of lens development. Previous studies have shown that the GRN of mammalian lens induction is governed by a multitude of mutual cross-regulations, including the transcription factors Pax6, Six3 and Sox2 (summarized in the BioTapestry visualization Fig 7). Six3 appears to regulate the onset of Pax6 expression in the PLE while Pax6 subsequently maintains Six3 levels [23, 35, 43]. Only a small fraction of Six3 f/del;Le-Cre embryos, type III in [23], exhibit a complete arrest of lens development prior to the lens pit stage, a phenotype comparable to Pax6 knockout phenotype, although this might be due to the inefficient deletion of Six3. Consequently, the level of Six3 ablation in lens-derived tissue correlates well with the grade of phenotype and Pax6 and Sox2 downregulation [23]. Epistasis of Pax6 and Sox2 is stage-dependent. In pre-placodal ectoderm, Pax6 and Sox2 are regulated independently. By contrast, after the lens placode has formed, Sox2 expression is dependent on Pax6 [34]. Genetic data presented here reveal a fundamental and redundant role of Meis1 and Meis2 homeoproteins in the regulation of lens induction. Meis1 and Meis2 transcription factors have previously been identified as upstream regulators of the Pax6 EE [22]. However, Meis1- and EE-deficient mice surprisingly do not display eye phenotypes at placodal stage of lens development [17, 28] and therefore are not comparable to that of the lens-specific ablation of Pax6 [7]. This indicates that (i) Meis2 may compensate for the loss of Meis1, and that (ii) another Pax6 enhancer driving expression to lens may substitute for missing EE [17, 44]. Until recently, interrogation of the combined role of Meis1/2 proteins on lens induction and Pax6 expression in vivo has been hampered by the lack of suitable Meis2 knockout allele. Herein, we have conducted a comprehensive genetic analysis of Meis1 and Meis2 function in mouse to show that simultaneous depletion of Meis1 and Meis2 in the presumptive lens ectoderm results in the failure of lens placode formation and a marked reduction of Pax6 and Six3 expression in the presumptive lens areas. In contrast, expression of Sox2 is maintained in the Meis1/Meis2 mutated ectoderm. The Meis-related TALE homeodomain protein Prep1 (also known as Pknox1) apears to control the timing of Pax6 activation and its expression level in the developing lens via direct binding to the EE [25]. The available data regarding the genetic requirement for Prep1 suggest it has a cell-nonautonomous function in lens induction. Prep1 trans-heterozygotes composed of a germline knockout and retroviral insertion allele (a hypomorph), respectively, demonstrate defects at the lens induction step [25]. In contrast, conditional gene targeting of Prep1 at pre-placodal and placodal phases of lens induction using Ap2alpha-Cre and Le-Cre did not reveal any developmental phenotype [45]. We were unable to detect any changes in Prep expression using imunohistochemistry (S9 Fig), making it unlikely that the observed phenotype in Meis1/2 double knockout mice is due to Prep1 deficiency.
Our data are consistent with the scenario in which Meis1/2 function as regulators of lens placode development primarily via activation of Pax6 enhancers. However, it is likely that Meis1 and Meis2 regulate other factors contributing to early lens development such as the ones identified for Meis1 [46]. It was recently shown that Meis1 regulates either directly or indirectly the expression of genes involved in patterning, proliferation and differentiation of the neural retina, and that haploinsufficiency of Meis1 causes micropthalmic traits and visual impairment in adult mice [46]. Based on the fact that Marcos et al. could not detect Meis2 expression at early stages of eye development, authors considered only Meis1 function to be critical for early mouse eye development [46]. In contrast, in this study we detected Meis2 expression in early stages of lens development (S1 Fig). Furthermore, Meis2 expression is lost upon genetic ablation of Meis2 gene (S1J Fig). This data together with the fact that only simultaneous deletion of Meis1 and Meis2 in PLE leads to an arrest of lens development in pre-placodal stage strongly suggests that both Meis1 and Meis2 are expressed and essential for early eye development. Nevertheless, it is very likely that Meis1 and Meis2 fulfill the redundant function only in specific developmental stages and processes (our data and [46]), while having many discrete functions in the embryo even within the eye development.
Mammalian eye development is highly sensitive to the levels of Pax6 as haploinsufficiency causes aniridia in humans and multiple ocular defects in mice [4, 47–50]. In contrast, increased levels of Pax6 result in various ocular abnormalities [51]. In the mammalian lens, Pax6 controls all known steps of tissue morphogenesis [7, 34, 52] but its dosage appears to be especially critical during the earliest developmental stages. The data presented here show that the molecular mechanisms of Meis1/2 regulation of Pax6 are mediated by at least two "shadow enhancers" (Fig 7): a 3‘-located ultraconserved SIMO identified as a Meis target here, and a 5‘-located ectoderm enhancer (EE), identified as a target of TALE proteins earlier [22, 25]. The concept of the seemingly redundant "shadow enhancers" driving expression of a given gene to overlapping or identical patterns has been pioneered in Drosophila as a potential source of evolutionary novelty [53]. It was hypothesized that "shadow enhancers" may evolve novel binding sites and achieve new regulatory activities without disrupting the core patterning function of a developmental control gene. As cis-regulatory mutations are the main driving force of animal evolution [54, 55] buffering loss-of-function situations during enhancer evolution may be critical. "Shadow enhancers" analyzed in detail in Drosophila to date provide robustness and precision to the system [56–58]. A remote "shadow enhancer" identified in the human ATOH7 gene, by virtue of its deletion in patients suffering with nonsyndromic congenital retinal nonattachment, displays identical spatiotemporal activity to the primary enhancer when tested by transgenesis [59]. Although the function of the primary and "shadow enhancer" are not firmly established, dual enhancers may reinforce Atoh7 expression during early critical stages of eye development when retinal neurogenesis is initiated. It is tempting to speculate that the two apparently redundant distal "shadow enhancers" (EE, SIMO) ensure robust and tight regulation of Pax6 gene expression during mammalian lens induction. In our view robustness of Pax6 "shadow enhancer" system provides stable high level of Pax6 gene expression and confers compensation for deleterious effects and protection to expression level fluctuations due to environmental influences. Recent systematic analysis of "shadow enhancers" during Drosophila mesoderm development revealed that their spatio-temporal redundancy is often partial in nature, while the non-overlapping function may explain why these enhancers are maintained within a population [60]. Reporter gene assays and genetic ablation experiments shown here provide evidence for redundant ("shadow") enhancer function of SIMO and EE selectively during early stages of lens induction. Later on the two enhancers may indeed act more independently with some overlap of transcription factor use while their distinctness is likely elicited by different sets of transcription factors co-expressed and co-bound at different times and in different combinations and stoichiometry. It is nevertheless intriguing that the two enhancers responsible for lens placode expression of Pax6 utilize similar molecular logic, namely Meis1/2-dependency ([22] and this study), Six3 regulatory input ([23] and this study) and autoregulatory function [19, 21]. Furthermore, two Meis/Prep binding sites, L1 and L2, were identified in the EE [22, 25] while at least three evolutionarily conserved Meis binding sites are present in SIMO (this study). In theory, the accumulation of homotypic binding sites may aid the enhancer robustness and may protect the enhancer from vulnerable mutations leading to the loss of responsivness to a particular transcriptional regulator. Phylogenetic footprinting and reporter gene transgenics indicate that SIMO enhancer activity in zebrafish not only depends upon Pax6 autoregulation [19] but also on functional Meis binding sites (this study). Given the profound difference in the early stages of lens development in mice (lens formed by invagination) and fish (lens arises by delamination) it is remarkable that the SIMO enhancer maintains its Meis-dependent regulation albeit not for the comparable developmental stage. In fact, SIMO enhancer becomes active in zebrafish only at 48 hours post fertilization when the lens is already formed [19]. This illustrates that species-specific adaptation of enhancer function is combined with a developmental change. It will be interesting to see if other features of SIMO regulation, such as Six3 interaction, are maintained in zebrafish. No functional data exist for the zebrafish EE, although at the sequence level this regulatory element is evolutionarily conserved from human to fish [13, 15, 25]. It remains to be seen if the evolutionary strategy of maintaining lens "shadow enhancers" in the Pax6 locus is utilized in zebrafish, or the developmental robustness is achieved via Pax6 gene duplication giving rise to Pax6.1a and Pax6.1b paralogues [61].
Pax6 is considered as an extreme case of an evolutionarily conserved developmental regulator promoting eye formation in vertebrates and Drosophila [62]. Meis genes belong to the TALE homeobox family found in genomes across all Metazoa [63]. In contrast to Pax6, Homothorax, a Drosophila orthologue of vertebrate Meis/Prep genes, suppresses eye development rather than promoting it [64]. Homothorax together with the Cut homeoprotein supresses expression of Pax6 orthologue Eyeless in the antenna disc [65]. Conversely, Sine oculis, a downstream target of Eyeless, supresses Homothorax and Cut in the eye disc thus allowing eye development to proceed [65]. The different genetic wiring of Pax6/Eyless and Meis/Homothorax in vertebrate and Drosophila eye developmental programs may merely reflect the vast evolutionary distance between the respective species, morphological differences in the eye types being built and a general strategy of re-purposing individual components from the common genetic toolkit during the course of evolution.
In conclusion, this study identifies a genetic requirement for Meis1 and Meis2 for early steps of mammalian eye development and reveals an apparent robustness of the gene regulatory mechanism whereby two independent "shadow enhancers" of similar molecular architecture maintain critical levels of a dosage-sensitive gene, Pax6, during lens induction. These results allow us to establish a genetic hierarchy during early vertebrate eye development and provide novel mechanistic insights into the regulatory logic of this process.
Housing of mice and in vivo experiments were performed in compliance with the European Communities Council Directive of 24 November 1986 (86/609/EEC) and national and institutional guidelines. Animal care and experimental procedures were approved by the Animal Care Committee of the Institute of Molecular Genetics (study #174/2010). Mice were sacrificed by cervical dislocation.
To inactivate Meis1, Meis1+/- [28] mice were used. A conditional mutant allele of the Meis2 gene (Meis2f/f) was generated by inserting loxP sites in the introns 2 and 6, flanking exons 3 and 6 in the Meis2 gene (S1G Fig) at the Gene Targeting & Transgenic Facility, University of Connecticut, USA [32]. To generate whole-body knockout of Meis2, Meis2f/f mice were crossed with Hprt-Cre mice (strain 129S1/Sv-Hprttm1(cre)Mnn /J, stock 004302, The Jackson Laboratory) that display the zygotic Cre recombinase activity. For specific deletion of Meis2 in presumptive lens ectoderm, Le-Cre [7] mice were used. ROSA26R [66] and Pax6Sey-1Neu[4] mice (herein designated as Pax6Sey/+) have been described previously. SIMO enhancer was deleted using a pair of TALENs targeting sequences TCAGCCCCCACCCATACTCtcaaaaggaatgtcgTCGAGCGTCAGTGCCTGAA and TGCACTTGTCACTCAGCATTAtccatcctcattaaTGACAATGGGAAAGTTTA (recognition sequence shown in capital letters). TALENs were designed using TAL Effector Nucleotide Targeter 2.0 (https://tale-nt.cac.cornell.edu/), assembled using the Golden Gate Cloning system [67], and cloned into the ELD-KKR backbone plasmid [68]. Polyadenylated TALEN mRNAs were prepared using mMESSAGE mMACHINE T7 ULTRA Kit (Ambion) and were injected into the cytoplasm of fertilized mouse oocytes. EE [16] was deleted using CRISPR/Cas9 system. A sequence containing EE region was submitted to CRISPR Design Tool (http://crispr.mit.edu/) to select for a set of sgRNAs‘. Oligonucleotides used to make sgRNA constructs are listed in S1 Table and were cloned into pT7-gRNA (pT7-gRNA was a gift from Wenbiao Chen, Addgene plasmid # 46759). Cas9 mRNA was prepared using mMESSAGE mMACHINE T7 ULTRA Kit (Ambion) using plasmid pCS2-nCas9n (pCS2-nCas9n was a gift from Wenbiao Chen, Addgene plasmid # 47929). The sgRNAs were transcribed using MEGAshortscript kit (Ambion). A mixture of Cas9 mRNA (100ng/μl) and specific sgRNAs (25ng/μl each) was injected into the cytoplasm of fertilized mouse oocytes with homozygous or heterozygous deletion of SIMO enhancer (genetic background Pax6SIMOdel710/SIMOdel710 or Pax6SIMOdel710/+). Multiple independent lines were estabilished and the extent of EE deletion was analysed in F1 animals by DNA sequencing.
Mouse embryos were staged by designation the noon of the day when the vaginal plug was observed as embryonic day 0.5 (E0.5). Embryos of desired age were disected, fixed in 4% paraformaldehyde (PFA) from 45 minutes up to 4 hours at 4°C, washed with PBS, cryopreserved in 30% sucrose and frozen in OCT (Sakura). The cryosections (10–12 μm) were permeabilized with PBT (PBS with 0.1% Tween), blocked with 10% BSA in PBT and incubated with primary antibody (1% BSA in PBT) overnight at 4°C. Sections were washed with PBS, incubated with fluorescent secondary antibody (Life Technologies, 1:500) for one hour at room temperature, washed with PBS, counterstained with DAPI and mounted in Mowiol. The images were taken on Leica SP5 confocal microscope and were processed (contrast and brightness) with Adobe Photoshop. For hematoxylin-eosin staining, embryos were fixed in 8% PFA overnight, processed, embedded in paraffin, sectioned (8 μm), deparaffinized and stained. For β-galactosidase staining, embryos were fixed in 2% PFA, washed with rinse buffer (0.1 M phosphate buffer pH 7.3, 2 mM MgCl2, 20 mM Tris pH 7.3, 0.01% sodium deoxycholate, and 0.02% Nonidet P-40) and incubated in X-Gal staining solution (rinse buffer supplemented with 5 mM potasium ferricyanide, 5 mM potassium ferrocyanide, 20 mM Tris pH 7.3, and 1 mg/ml X-gal) at 37°C for 2 hours and at room temperature overnight shaking.
For chromatin immunoprecipitation whole E10.5 embryos or murine lens epithelial cells αTN4 [37] were used. A chromatin immunoprecipitation assay was performed according to manufacturer’s protocol (Upstate Biotech) with slight modifications as previously described [69]. The assay was repeated twice for both embryonic and tissue culture samples. The immunoprecipitated DNA was analyzed by qRT-PCR.
In silico analysis to identify putative Meis binding sites in SIMO was performed using high-quality transcription factor binding profile database JASPAR [70]. Electrophoretic mobility shift assays (EMSAs) was performed using double-stranded oligonucleotides comprising binding sites SIMO_B. A single point mutation was introduced into binding site changing Meis recognition sequence TGACAG/A into TcACAG/A. 32P-labeled oligonucleotides were incubated with in vitro-synthesized FLAG-Meis2 (TNT Quick, Promega) in binding buffer (10 mM HEPES pH 7.9, 100 mM KCl, 1mM EDTA, 4% Ficoll, 0.05mg/mL poly-dIdC) at room temperature for 15 minutes. For supershift experiment, anti-FLAG M2 antibody was included in the binding reaction. Samples were analysed by 6% polyacrylamide gel electrophoresis and autoradiography.
The wild-type mouse SIMO enhancer was amplified from genomic DNA using primers shown in S1 Table and introduced into the electroporation vector containing hsp68-lacZ reporter cassette [20]. Transcription factor binding sites within SIMO were mutagenized using QuickChange mutagenesis kit (Stratagene). Constructs carrying minimal EE and minimal SIMO enhancers were generated using synthetic double stranded oligonucleotides shown in S1 Table. All reporter gene constructs were verified by DNA sequencing. Brown Leghorn eggs were incubated until reaching HH10–11 stages and electroporation was performed as described [71]. The DNA mixture was injected outside of the right developing optic cup and electroporated using voltage of 12 V, length of pulse 20 ms, interval length 100 ms. The embryos were collected in stage HH20-HH21, fixed for 15 minutes in 2% formaldehyde and proceeded to X-gal staining.
The wild-type zebrafish SIMO enhancer was introduced into ZED vector upstream of minimal gata2a promoter [40]. Meis binding sites within SIMO were mutagenized using QuickChange mutagenesis kit (Stratagene). For transgenesis, the Tol2 transposon/transposase method [72] was used with minor modifications. A mixture containing 30 ng/μl of transposase mRNA, 30 ng/μl of Qiagen column purified DNA, and 0.05% phenol red was injected in the cell of one-cell stage embryos. Embryos were raised at 28.5 oC and staged by hours post fertilization (hpf). Embryos selected for imaging were anaesthetised with tricaine and mounted in low-melting agarose. Images were taken on Leica SP5 confocal microscope.
All used oligonucleotides are listed in S1 Table. All used primary antibodies are listed in S2 Table.
|
10.1371/journal.pntd.0004832 | Activation, Impaired Tumor Necrosis Factor-α Production, and Deficiency of Circulating Mucosal-Associated Invariant T Cells in Patients with Scrub Typhus | Mucosal-associated invariant T (MAIT) cells contribute to protection against certain microorganism infections. However, little is known about the role of MAIT cells in Orientia tsutsugamushi infection. Hence, the aims of this study were to examine the level and function of MAIT cells in patients with scrub typhus and to evaluate the clinical relevance of MAIT cell levels.
Thirty-eight patients with scrub typhus and 53 health control subjects were enrolled in the study. The patients were further divided into subgroups according to disease severity. MAIT cell level and function in the peripheral blood were measured by flow cytometry. Circulating MAIT cell levels were found to be significantly reduced in scrub typhus patients. MAIT cell deficiency reflects a variety of clinical conditions. In particular, MAT cell levels reflect disease severity. MAIT cells in scrub typhus patients displayed impaired tumor necrosis factor (TNF)-α production, which was restored during the remission phase. In addition, the impaired production of TNF-α by MAIT cells was associated with elevated CD69 expression.
This study shows that circulating MAIT cells are activated, numerically deficient, and functionally impaired in TNF-α production in patients with scrub typhus. These abnormalities possibly contribute to immune system dysregulation in scrub typhus infection.
| Scrub typhus is a mite-borne bacterial infection in humans caused by Orientia tsutsugamushi, an obligate intracellular bacterium, prevalent in Asia, Northern Australia, and the Indian subcontinent. The pathogenesis of O. tsutsugamushi infection is known to be not only related to the virulence of O. tsutsugamushi, but also to the host immune response. Mucosal-associated invariant T (MAIT) cells are an evolutionarily conserved antimicrobial MR1-restricted T cell subset. Upon antigen recognition, MAIT cells rapidly produce proinflammatory cytokines, maintain an activated phenotype throughout the course of an infection, and have the potential to directly kill infected cells; thus, playing an important role in controlling the host response. However, little is known about the role of MAIT cells in Orientia tsutsugamushi infection. To the best of our knowledge, this is the first study to measure the levels and functions of circulating MAIT cells in scrub typhus patients and to examine the clinical relevance of MAIT cell levels. The present study demonstrates that circulating MAIT cells are activated and numerically deficient in patients with scrub typhus. Notably, impairment of TNF-α production represents the susceptibility of individuals to O. tsutsugamushi infection. These findings provide important information for predicting the prognosis of scrub typhus infection.
| Scrub typhus is a mite-borne bacterial infection in humans caused by Orientia tsutsugamushi, an obligate intracellular bacterium, prevalent in Asia, Northern Australia, and the Indian subcontinent. With early diagnosis and management, most patients with scrub typhus are able to recover without complications [1]. However, some patients develop serious and potentially fatal complications such as interstitial pneumonia, acute renal failure, myocarditis, meningoencephalitis, gastrointestinal bleeding, acute hearing loss, and multiple organ failure [2–4]. As the primary targets of O. tsutsugamushi are endothelial cells, the variable extent of vasculitis in each individuals helps in part to explain the different levels of severity [5]. However, a previous study has shown that diffuse alveolar damage could present without evidence of vasculitis, suggesting that the immunologic response plays a significant role in development of the disease and determination of the severity of illness [3].
The immune response induced by O. tsutsugamushi is a combination of innate and adaptive immunity and the proper response of macrophages and T lymphocytes may be the driving factor in immunity in patients with scrub typhus [5]. Furthermore, several studies reported dysfunction of the immunologic response of the host to O. tsutsugamushi; dysregulated levels of certain cytokines, imbalance of Th1/Th2 cytokines and apoptosis of T lymphocytes during acute infection [6–9]. These findings suggested that the pathogenesis of O. tsutsugamushi infection is not only related to the virulence of O. tsutsugamushi, but also to the host immune response.
Mucosal-associated invariant T (MAIT) cells are a relatively newly recognized T cell subset that expresses a conserved invariant T cell receptor (TCR) α-chain (Vα7.2-Jα33 in humans and Vα19-Jα33 in mice) paired with a limited set of Vβ chains [10]. Human MAIT cells are defined as CD3+TCRδγ-Vα7.2+CD161high or CD3+TCRδγ-Vα7.2+IL-18Rα+ cells [11,12]. MAIT cells recognize bacteria-derived riboflavin (vitamin B2) metabolites presented by the MHC class 1b-like related protein (MR1) [10,13]. Upon antigen recognition, MAIT cells rapidly produce proinflammatory cytokines, such as interferon (IFN)-γ, tumor necrosis factor (TNF)-α, and interleukin (IL)-17, in an innate-like manner [14]. MAIT cells maintain an activated phenotype throughout the course of an infection, secrete inflammatory cytokines, and have the potential to directly kill infected cells; thus, playing an important role in controlling the host response [11,12,15–19]. However, little is known about the role of MAIT cells in O. tsutsugamushi infection. Accordingly, the aims of this study were to examine the level and function of MAIT cells in patients with scrub typhus and to evaluate the clinical relevance of MAIT cell levels.
The study cohort comprised 38 patients with scrub typhus (25 women and 13 men; mean age ± SD, 64.3 ± 15.6 years) and 53 healthy controls (HCs; 30 women and 23 men; mean age ± SD, 63.6 ± 12.2 years). All patients were confirmed as having scrub typhus by the serologic test using a passive hemagglutination assay (PHA) to detect O. tsutsugamushi antigen. A positive result was defined as a titer of ≥ 1:80 in a single serum sample or at least a fourfold rise in antibody titer at follow-up examination. PHA was performed using Genedia Tsutsu PHA II test kits (GreenCross SangA, Yongin, Korea). Scrub typhus patients were further divided into subgroups according to disease severity as previously described [7]. Patients with no organ dysfunction were considered to have mild disease, those with one organ dysfunction were considered to have moderate disease, while those with dysfunction of two or more organ systems were defined as having severe disease. Organ dysfunction was defined as follows: (1) renal dysfunction, creatinine ≥ 2.5 mg/dL; (2) hepatic dysfunction, total bilirubin ≥ 2.5 mg/dL; (3) pulmonary dysfunction, bilateral pulmonary infiltration on chest X-rays with moderate to severe hypoxia (PaO2/FiO2 < 300 mmHg or PaO2 < 60 mmHg or SpO2 < 90%); (4) cardiovascular dysfunction, systolic blood pressure < 80 mmHg despite fluid resuscitation; and (5) central nervous system dysfunction, significantly altered sensorium with Glasgow Coma Scale (GCS) ≤ 8/15. None of the controls had a history of autoimmune disease, infectious disease, malignancy, chronic liver or renal disease, diabetes mellitus, immunosuppressive therapy, or fever within 72 hours prior to enrollment.
The study protocol was approved by the Institutional Review Board of Chonnam National University Hospital, and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.
The following mAbs and reagents were used in this study: Allophycocyanin (APC)-Cy7-conjugated anti-CD3, phycoerythrin (PE)-Cy5-conjugated anti-CD161 and fluorescein isothiocyanate (FITC)-conjugated anti-TCR γδ, FITC-conjugated anti-CD3, FITC-conjugated anti-IFN-γ, FITC-conjugated annexin V, PE-conjugated anti-CD3, PE-conjugated anti-IL-17, PE-Cy7-conjugated anti-TNF-α, PE-conjugated anti-CD69, FITC-conjugated mouse IgG isotype, PE-conjugated mouse IgG isotype and PE-Cy7-conjugated mouse IgG isotype control (all from Becton Dickinson, San Diego, CA); PE-conjugated anti-programmed death-1 (anti-PD-1; eBioscience, San Diego, CA) and APC-conjugated anti-TCR Vα7.2 (BioLegend, San Diego, CA). Cells were stained with combinations of appropriate mAbs for 20 minutes at 4°C. Stained cells were analyzed on a Navios flow cytometer using Kaluza software (version 1.1; Beckman Coulter, Brea, CA).
Peripheral venous blood samples were collected in heparin-containing tubes, and PBMCs were isolated by density-gradient centrifugation using Lymphoprep (Axis-Shield PoC AS, Oslo, Norway). MAIT cells were identified phenotypically as CD3+TCRγδ-Vα7.2+CD161high cells, by flow cytometry as previously described [20,21]. Total lymphocyte numbers were measured by Coulter LH750 automatic hematology analyzer (Beckman Coulter, Miami, FL). Absolute numbers of MAIT cells were calculated by multiplying the MAIT cell percentages by the CD3+γδ- T cell percentages and the total lymphocyte numbers (per microliter) in peripheral blood.
IFN-γ, IL-17, and TNF-α expression in MAIT cells was detected by intracellular cytokine flow cytometry as previously described [11,12,21]. Briefly, freshly isolated PBMCs (1 × 106/well) were incubated in 1 mL complete media, consisting of RPMI 1640, 2 mM L-glutamine, 100 units/mL of penicillin, and 100 μg/mL of streptomycin, and supplemented with 10% fetal bovine serum (FBS; Welgene, Gyeongsan, Korea) for 4 hours in the presence of phorbol myristate acetate (PMA) (100 ng/mL; Sigma, St Louis, MO) and ionomycin (IM) (1 μM; Sigma). For intracellular cytokine staining, 10 μL of brefeldin A (GolgiPlug; BD Biosciences, San Diego, CA) was added, and the final concentration of brefeldin A was 10 μg/mL. After incubation for additional 4 hours, cells were stained with APC-Cy7-conjugated anti-CD3, PE-Cy5-conjugated anti-CD161, and APC-conjugated anti-TCR Vα7.2 mAbs for 20 minutes at 4°C, fixed in 4% paraformaldehyde for 15 minutes at room temperature, and permeabilized using Perm/Wash solution (BD Biosciences) for 10 minutes. Cells were then stained with FITC-conjugated anti-IFN-γ, PE-conjugated anti-IL-17 and PE-Cy7-conjugated anti-TNF-α mAbs for 30 minutes at 4°C and analyzed by flow cytometry.
To determine changes in expression levels of CD69, annexin V and PD-1 in MAIT cells after stimulation with IL-12 and IL-18, freshly isolated PBMCs (1 × 106/well) were incubated in 1 mL complete media for 24 hours in the presence of IL-12 (50 ng/mL; Miltenyi biotec, Bergisch Gladbach, Germany) and IL-18 (50 ng/mL; Medical and Biological Laboratories, Woburn, MA). Cells were then stained with FITC-conjugated anti-CD3, FITC-conjugated annexin V, APC-conjugated anti-TCR Vα7.2, PE-conjugated anti-CD3, PE-conjugated anti-CD69, PE-conjugated anti-PD-1 and PE-Cy5-conjugated anti-CD161 mAbs for 20 minutes at 4°C. CD69+, annexin V+, and PD-1+ MAIT cell levels were determined by flow cytometry.
All comparisons of percentages and absolute numbers of MAIT cells, their cytokine levels, and expression levels of CD69, PD-1 and annexin V were analyzed using the Mann-Whitney U tests or paired t test. Linear regression analysis was used to test associations between MAIT cell levels and clinical or laboratory parameters. The Wilcoxon matched-pairs signed rank test was used to compare changes in MAIT cell levels and functions according to disease activity. P values less than 0.05 were considered statistically significant. Statistical analyses and graphic works were performed using SPSS version 18.0 software (SPSS, Chicago, IL) and GraphPad Prism version 5.03 software (GraphPad Software, San Diego, CA), respectively.
The clinical and laboratory characteristics of scrub typhus patients are summarized in Table 1. Thirty-eight patients were included in this study. Of the 38 scrub typhus patients, 26 patients (68.4%) had mild disease; 9 patients (23.7%) had moderate disease; and 3 patients (7.9%) had severe disease. Eighteen patients were monitored longitudinally from the active state (before antibiotic therapy) to the remitted state (defined as resolution of all presenting symptoms after antibiotic therapy).
The percentages and absolute numbers of MAIT cells in the peripheral blood samples of 38 patients with scrub typhus and 53 HCs were determined by flow cytometry. MAIT cells were defined as CD3+TCRγδ- cells expressing TCR Vα7.2 and CD161high (Fig 1A). Percentages of MAIT cells were significantly lower in scrub typhus patients than in HCs (median 0.69% versus 1.37% [p < 0.05]) (Fig 1B). Absolute numbers of MAIT cells were calculated by multiplying MAIT cell percentages by the CD3+TCRγδ- T cell percentages and the total lymphocyte numbers (per microliter of peripheral blood). Scrub typhus patients had significantly lower absolute numbers of MAIT cells than HCs (median 1.89 cells/μL versus 11.0 cells/μL [p < 0.001]) (Fig 1C).
To evaluate the clinical relevance of MAIT cell levels in 38 patients with scrub typhus, we investigated the correlation between MAIT cell percentages in the peripheral blood and clinical parameters by regression analysis (Table 2). The univariate linear regression analysis showed that circulating MAIT cell percentages were significantly correlated with age, alanine aminotransferase level, alkaline phosphatase level, and severity (p = 0.001, p = 0.043, p = 0.009, and p = 0.047, respectively). However, no significant correlation was observed between MAIT cell percentages and leukocyte count, lymphocyte count, hemoglobin level, neutrophil count, platelet count, total bilirubin level, total protein level, albumin level, aspartate aminotransferase level, lactate dehydrogenase level, C-reactive protein level and erythrocyte sedimentation rate (Table 2).
Hepatocytes and endothelial cells stimulated by proinflammatory cytokines (e.g., IFN-γ and TNF-α) have been known to kill intracellular rickettsiae via inducible nitric oxide synthase expression and nitric oxide-dependent mechanism [22,23]. To investigate the expression of these cytokines in MAIT cells, we incubated PBMCs from 23 patients with scrub typhus and 14 HCs for 4 hours in the presence of PMA and IM; then the expressions of IFN-γ, IL-17A, TNF-α in the MAIT cell populations were examined at the single-cell level by intracellular flow cytometry (Fig 2A). Percentages of TNF-α+ MAIT cells were found to be significantly lower in scrub typhus patients than in HCs (median 15.6% versus 45.0% [p < 0.05]). However, IFN-γ+ or IL-17A+ MAIT cell levels were comparable between scrub typhus patients and HCs (Fig 2B).
The relationship between the expression of immune activation markers and the loss of circulating MAIT cells has been reported in HIV-infected patients [24]. To determine whether MAIT cell deficiency in scrub typhus patients is correlated with activation-induced cell death, CD69+ and annexin V+ cell levels in circulating MAIT cells were determined by flow cytometry. Percentages of CD69+ MAIT cells were found to be significantly higher in scrub typhus patients than in HCs (median 36.8% versus 6.0% [p < 0.0001]) (Fig 3A and 3B). However, annexin V+ cell levels were comparable between scrub typhus patients and HCs (Fig 3C and 3D). PD-1 and its ligands, PD-1 L1 and PD-L2, are known to deliver inhibitory signals that regulate the balance among T cell activation, tolerance and immunopathology [25]. To determine whether impaired TNF-α production by MAIT cells is related to PD-1, we examined the expression levels of PD-1 in the peripheral blood samples of 17 scrub typhus patients and 14 HCs. The expression levels of PD-1 in MAIT cells were similar between scrub typhus patients and HCs (Fig 3E and 3F).
To determine whether MAIT cells can be activated by IL-12 and IL-18, PBMCs from HCs were cultured with IL-12 and IL-18 for 24 hours and then CD69+, annexin V+, and PD-1+ cell levels in MAIT cells were determined by flow cytometry. Percentages of CD69+ MAIT cells were found to be significantly higher in IL-12- and IL-18-treated cultures compared with untreated cultures (mean ± SEM 59.9 ± 10.29% versus 6.5 ± 0.48% [p < 0.005]) (Fig 4A and 4B). However, annexin V+ and PD-1+ MAIT cell levels were comparable between the treated and untreated cultures (Fig 4C and 4F).
Based on our observation that circulating MAIT cell levels and TNF-α production are reduced in scrub typhus patients, we investigated the changes in circulating MAIT cell levels and functions in relation to disease activity. Eighteen and eleven patients were available for follow-up examination of MAIT cell levels and functions, respectively. As shown in Fig 5A, no significant changes in circulating MAIT cell levels were found according to disease activity. However, TNF-α+ MAIT cell levels were found to be greater when the disease was in remission than when it was active (median 56.8% versus 34.1% [p < 0.05]) (Fig 5B).
Although the importance of an immune response of MAIT cells in host protection against certain mycobacterial and enterobacterial infections is well established [11,12,15–19], the role of MAIT cells in O. tsutsugamushi infection remains unclear. To the best of our knowledge, this is the first study to measure the levels and functions of circulating MAIT cells in scrub typhus patients and to examine the clinical relevance of MAIT cell levels. The present study showed that percentages and numbers of circulating MAIT cells were lower in scrub typhus patients than in HCs. We also demonstrated that MAIT cell deficiency reflects disease severity. In particular, in vitro experiments showed poor production of TNF-α by MAIT cells during active disease, but TNF-α production was found to be increased during remission. Finally, our study showed that the impaired production of TNF-α by MAIT cells was associated with elevated CD69 expression in scrub typhus patients. These findings possibly contribute to immune system dysregulation in scrub typhus infection and have important implications for the development of cell-based immunotherapy.
In this study, circulating MAIT cell levels were found to be reduced in scrub typhus patients. In contrast, frequencies of other T cell subsets (e.g., CD3 T cells, αβ T cells and γδ T cells) were found to be similar between the patients and HCs (S1 Fig), suggesting that the decline in cell levels is specific to MAIT cells. MAIT cell deficiency in peripheral blood has also been reported in several other infectious diseases, including Vibrio cholerae O1 infection, Pseudomonas aeruginosa infection in cystic fibrosis, and severe bacterial sepsis, in particular non-streptococcal infection [26–28]. Our previous study revealed that MAIT cells were numerically and functionally deficient in patients with pulmonary tuberculosis and nontuberculous mycobacteria lung disease [19]. In the present study, circulating MAIT cells were more reduced in patients with organ dysfunction than in patients without organ dysufunction (S2 Fig). It is well known that scrub typhus occurs when the host is bitten by an O. tsutsugamushi-infected trombiculid mite and it can spread to other organs via the bloodstream (e.g., kidney, liver, lung and brain) [2]. Thus, the decline in circulating MAIT cell numbers might be due to migration of MAIT cells from the peripheral circulation to infected tissues and organs, for defending against O. tsutsugamushi infection.
The present study revealed that MAIT cell percentages in the peripheral blood were significantly correlated with age, alanine aminotransferase level, alkaline phosphatase level, and disease severity. Consistent with our data, a recent study demonstrated that MAIT cell deficiency was associated with increased disease severity in cystic fibrosis patients with P. aeruginosa infection [27]. Furthermore, these findings are supported by our previous studies which showed that MAIT cell deficiency is related to the disease severity or extent in patients with chronic obstructive pulmonary disease and mycobacterial infection [19,29]. Collectively, circulating MAIT cell levels may reflect the inflammatory activity or severity of infectious diseases. However, after multivariate analysis, only age was found to be a significant independent predictors of MAIT cell deficiency in scrub typhus patients (Table 2). Similarly, circulating MAIT cell levels were found to be considerably affected by age, irrespective of the healthy or disease state [21,30,31]. Therefore, age-dependent changes in MAIT cell levels could be explained partially by the fact that old age is one of the risk factors affecting the complication and mortality in scrub typhus or other infectious diseases [32,33]. Interestingly, we also found that liver function abnormalities, especially in alanine aminotransferase and alkaline phosphatase, were positively correlated with the MAIT cell levels. Liver is known to be a site of accumulation of MAIT cells and abnormal liver function is frequently observed in scrub typhus patients [34,35]. Therefore, further studies are needed to investigate the integrated role of MAIT cells in hepatic injury.
The fundamental role of TNF-α, one of the cytokines secreted by the Th1 subset, in control of intracellular growth of O. tsutsugamushi is well established [36,37]. In the present study, the production of TNF-α by MAIT cells was found to be diminished in scrub typhus patients, whereas no significant differences were found between scrub typhus patients and HCs in the production of IFN-γ or IL-17. These findings suggest that the dysregulation of TNF-α production may have a certain role in the pathogenesis of scrub typhus. However, it is not clear why TNF-α secretion by MAIT cells was reduced in response to PMA/IM stimulation. One possible explanation is that impaired TNF-α production by MAIT cells may be due to anergy or exhaustion of these cells during infection. These findings have also been previously observed in other infectious diseases that showed upregulation of coinhibitory receptors [19,24,38]. Another possibility is that the impairment of TNF-α production represents the susceptibility of individuals to O. tsutsugamushi infection. Interestingly, a previous study revealed that O. tsutsugamushi inhibited TNF-α production by inducing IL-10 secretion in murine macrophages [39]. Third possibility is that high levels of proinflammatory cytokines can induce such apparent reduced responsiveness of MAIT cells. This hypothesis is supported by our supplementary data showing that plasma levels of proinflammatory cytokines, such as IFN-γ and TNF-α, were found to be significantly higher in scrub typhus patients than in HCs (S3 Fig), which was consistent with the results of previous study [40]. Moreover, these similar findings have also been observed in our previous study treating another invariant T cells (e.g., natural killer T cells) with IL-1β, IL-6, IL-8, IL-18, IFN-γ and TNF-α [41]. Further studies are needed to prove this hypothesis in MAIT cells. Altogether, these results indicate that impaired TNF-α production by MAIT cells is due to a negative-feedback mechanism that allows the pathogen to survive in the hostile environment.
MAIT cells have been known to be stimulated either in an MR1-dependent manner or in an MR1-independent manner [42]. Our study showed that MAIT cells were activated in scrub typhus patients, indicated by CD69 up-regulation. According to the Kyoto Encyclopedia of Genes and Genomes (KEGG) at http://www.kegg.jp/kegg/pathway.html, none of O. tsutsugamushi Boryong and Ikeda strains harbor the riboflavin biosynthesis pathway. Until proven otherwise, this means that this bacterial strain (especially, Boryong, most common strain in South Korea) does not provide the MAIT cell MR1 ligand, suggesting that MAIT cells can be activated in MR1-independent manner. It has been previously reported that MAIT cells highly express IL-18R and IL-12R [43,44]. In accordance with the results of previous studies [40,43], the present study demonstrated that MAIT cells were activated after stimulation by IL-12 and IL-18. Furthermore, a previous study showed that plasma IL-12p40 and IL-18 levels were elevated in scrub typhus patients [40]. Collectively, these findings suggest that MAIT cells can be activated by IL-12 and IL-18, in an MR1-independent manner, in scrub typhus.
CD69 is known to be early activation marker, whereas PD-1 is considered as relative late activation marker. Following TCR activation, CD69 is upregulated within ~4 hours [45], whereas PD-1 is upregulated within 24–72 hours [46]. PD-1 expression in circulating MAIT cells is also known to be upregulated in chronic infectious diseases, such as chronic human immunodeficiency virus, tuberculosis, and chronic hepatitis C virus infections [19,47,48]. In addition, recent studies reported that the PD-1 plays a complex role during acute infection [46]. However, little is known about the role of PD-1 in acute bacterial infection, particularly in scrub typhus. Investigation of CD69 and PD-1 expression on MAIT cells can provide information on the time course of disease activation during O. tsutsugamushi infection. In the current study, circulating MAIT cells were found to be markedly activated, as indicated by the upregulation of CD69 expression, while marginal expression of PD-1 was observed in a subset of acute scrub typhus patients, indicating that MAIT cells may play an important role in early phase during scrub typhus infection. In line with our data, Doe et al recently reported that CD4+ T cells from mice infected with Plasmodium parasites expressed PD-1 as early as 6 days after infection, while Listeria monocytogenes induced marginal expression of PD-1 [49]. These results imply that the mode and function of PD-1 expression might differ between acute infection and chronic infection, i.e., PD-1 expression may occur in the late stage of infection or may depend on the microbes. In addition, our data showed that PD-1+ MAIT cell levels were inversely correlated with the corresponding TNF-α production by MAIT cells from the patients (S4 Fig), suggesting that functional impairment of MAIT cells may partially be due to the negative regulation including PD-1. Another interesting finding from our study is that the levels of apoptotic T cells, defined as annexin V+ cells, were comparable between scrub typhus patients and HCs. These results indicate that reduced MAIT cells in peripheral blood might be due to their migration to the infected tissue or organs, rather than MAIT cell death.
In the present study, impairment in TNF-α production of MAIT cells in the acute phase of scrub typhus infection was found to be restored during the remission phase, while numerical deficiency of MAIT cells did not change over time. Consistent with our data, a previous study demonstrated that the function of MAIT cells was restored partially after effective antiretroviral therapy, although levels of MAIT cells in peripheral blood were not restored [24]. A similar finding about MAIT cell levels in peripheral blood has also been reported in pediatric patients with Vibrio cholerae O1 infection [26]. Circulating MAIT cell levels were significantly reduced at day 7 and this deficiency persisted up to 90 days after onset of cholera. In a recent study, circulating MAIT cell numbers started to increase as early as 4 days after severe sepsis, but persistent depletion of MAIT cells was found in some patients, depending on their clinical condition [28]. Thus, further long-term follow-up studies are needed to determine when circulating MAIT cell levels are restored after scrub typhus infection.
In summary, the present study demonstrates that circulating MAIT cells are activated, numerically deficient, and functionally impaired in TNF-α production in patients with scrub typhus. In addition, MAIT cell deficiency reflects disease severity. These findings provide important information for predicting the prognosis of scrub typhus infection.
|
10.1371/journal.pcbi.1005557 | Predicting cryptic links in host-parasite networks | Networks are a way to represent interactions among one (e.g., social networks) or more (e.g., plant-pollinator networks) classes of nodes. The ability to predict likely, but unobserved, interactions has generated a great deal of interest, and is sometimes referred to as the link prediction problem. However, most studies of link prediction have focused on social networks, and have assumed a completely censused network. In biological networks, it is unlikely that all interactions are censused, and ignoring incomplete detection of interactions may lead to biased or incorrect conclusions. Previous attempts to predict network interactions have relied on known properties of network structure, making the approach sensitive to observation errors. This is an obvious shortcoming, as networks are dynamic, and sometimes not well sampled, leading to incomplete detection of links. Here, we develop an algorithm to predict missing links based on conditional probability estimation and associated, node-level features. We validate this algorithm on simulated data, and then apply it to a desert small mammal host-parasite network. Our approach achieves high accuracy on simulated and observed data, providing a simple method to accurately predict missing links in networks without relying on prior knowledge about network structure.
| The majority of host-parasite associations are poorly understood or not known at all because the number of associations is so vast. Further, interactions may shift seasonally, or as a function of changing host densities. Consequently, host-parasite networks may be poorly characterized since effects of cryptic host-parasite associations on network structure are unknown. To address this, we developed theory and applied it to empirical data to test the ability of a simple algorithm to predict interactions between hosts and parasites. The algorithm uses host and parasite trait data to train predictive probabilistic models of host-parasite interaction. We tested the accuracy of our approach using simulated networks that vary greatly in their properties, demonstrating high accuracy and robustness. We then applied this algorithm to data on a small mammal host-parasite network, estimated model accuracy, identified host and parasite traits important to prediction, and quantified expected changes to structural properties of the network as a result of link relabeling.
| Complex interactions between host and parasite species can be described as a network, with host and parasite species as two distinct node types connected by links that represent associations between a given parasite and host species. Understanding the structure [1] and stability [2, 3] of host-parasite networks is important for establishing drivers of host-parasite interactions, parasite specificity, and the consequences of host extinctions on parasite diversity. Recently, authors have applied concepts and tools from community ecology and graph theory to host-parasite interactions [4–7] in an effort to understand how host and parasite communities interact, including investigations into how host community diversity influences disease transmission [8], how parasites interact within infected hosts [9], and how host functional and phylogenetic similarity promote parasite sharing [10, 11]. Additional research has focused on topological measures of host-parasite networks—such as nestedness [12] and modularity [13]—which attempt to quantify the formation of patterns of interactions between host and parasite species. These patterns may influence network stability [2] and resilience [3]. Identifying the factors influencing the formation of these patterns is an important nascent area of research.
There is little consensus about whether various reported topological patterns are common [14–16], which may be a result of the influence of sampling effort and the effect of incomplete detection on measures of topological network structure [17]. Specifically, the detection of patterns in most studies is predicated on having completely sampled the network of host-parasite interactions. That is, all interactions between host and parasite species are assumed to have been documented in the course of the study. However, such exhaustive sampling is rare at best, as logistical constraints often limit detection of all interactions. Moreover, the total number of potential host-parasite interactions increases as a product of the number of host and parasite species, creating a large number of opportunities for a missed detection of a host-parasite interaction. It is unlikely that studies of ecological networks are recording all of the potential interactions between species, as even long term data have been unable to detect a large number (nearly 50% of plant-pollinator interactions) of species interactions [18]. Incomplete sampling compromises inference of network structure and stability, and may undermine studies of parasite specificity and measures of parasite species richness for a given host species.
Despite this complication, there is a body of research aimed at predicting host-parasite interactions. This work is of clear importance to wildlife and human health—as the it is possible to identify potential spillover events [19–21]—and to a general understanding of the traits associated with parasite specialization. To this end, current approaches examine parasite species independent of the network within which they are embedded, using host traits to predict likely interactions. Two such efforts attempted to predict the fish host community parasitized by helminth parasites [22, 23]. However, approaches to date have not explicitly considered how the distribution of host and parasite traits, or the complex interactions at the host-parasite network level could influence predictability of host-parasite interactions. By considering all potential interactions simultaneously, it is possible to find the most probable interactions given the entire network, rooting the problem of predicting likely host-parasite interactions within a body of theory from the study of complex networks [24, 25].
Here, we address this problem by developing and testing a method capable of determining the number of likely unobserved host-parasite interactions, and accurately predicting the most likely, but undetected, host-parasite interactions in the network. This is not a new problem, as computer scientists have struggled with the link prediction problem for decades, most notably in studies of social networks [26–28]. We focus on link prediction in bipartite networks, with a specific application to ecological networks. Previous work in link prediction for bipartite networks has required information on traits of both node classes (e.g., host and parasite species), as well as knowledge of network topology (e.g., degree distribution) [29]. Here, we develop a highly accurate link prediction method based on trait matching between host and parasite species. That is, we make no assumption about network topology, but predict bipartite interactions using only trait data on host and parasite species. We examine the performance of our algorithm on simulated data extensively, and then test the algorithm on an ecological host-parasite network of small mammals and their resident parasite communities in a New Mexican desert ecosystem.
We propose an approach to identifying cryptic associations in host-parasite networks based on numerical estimation of conditional density functions. We represent the connections between hosts and parasites as a sparse bipartite graph (H, P, E) with vertex sets H (host species) and P (parasite species) and edges E, such that an edge connects Hi and Pj if species j parasitizes species i. If there is an edge between Hi and Pj, we write yi,j = 1 whether the edge has been observed or not; otherwise yi,j = 0. Not all edges have been observed and not all possible edges exist. Thus, E consists of both observed edges Eo and unobserved edges Eu = E∖Eo and is itself a subset of the possible edges E ˜ = H × P. Attached to each host and parasite species are vectors of features h and p, respectively. Thus, edge (Hi, Pj) has the combined feature set xi,j = (hi, pj).
To identify cryptic links in Eu, we seek a ranking of edges according to their probability. The probability that there is an edge between two vertices given its feature set is written P(y = 1|x). From Bayes’ theorem, we have
P ( y = 1 | x ) = f 1 ( x ) P ( y = 1 ) f ( x )
where f1 is the conditional probability of feature set xi,j given that yi,j = 1, P(y = 1) is the connectance of the graph, and f is the density of all possible combined feature sets. That is, f1 is the probability density of features when a link exists between host and parasite, and f is the density of features for all possible host-parasite combinations. The model assumes that the observation process (probability of detection) is either constant or random with respect to host and parasite features. Extensions of this model could address this assumption through the incorporation of features related to sampling probabilities or the use of model simulations directly incorporating the observation process. Since we seek only a rank ordering, we ignore P(y = 1) which is simply a normalizing constant, and estimate q = f1/f.
Estimating q is a density-ratio estimation problem [30]. The plug-in approach we propose, which we call plug-and-play, is to separately estimate f1 and f from the features of Eo and E ˜ and to take the quotient as required for evaluating any given host-parasite pair, i.e., q ^ = f 1 ^ / f ^. In practice, we use the kernel density estimator npudens in the np package [31] and the “normal-reference” bandwidth. This nonparametric approach to density-ratio estimation was chosen because it generally performs very well, particularly when the feature set contains a combination of binary and continuous features [32].
The estimated probabilities of all edges in E ˜ \ E o are then evaluated and ordered. That is, the model outputs the probability of each edge E ˜ \ E o, which can then be ranked by the most probable undetected edge in the set of cryptic links Eu. The AUC (area under the receiver operating characteristic) statistic can be calculated by comparing the observed labels and the estimated probabilities. If probabilities need to be translated into binary states, we begin with the most likely cryptic link, and re-label unobserved edges in order until a stopping criterion is met.
Host-parasite networks were simulated as follows. First, we generated a number (typically n = 5) trait values for both host and parasite species by drawing random numbers from a beta distribution, with the two shape parameters (α and β) drawn from a uniform distribution bounded between 0.5 and 1.5. The beta distribution was chosen for its flexibility and generality to many ecological and epidemiological problems [33, 34], as it is bound between 0 and 1, can take a variety of shapes, and is easily extensible (e.g., beta-binomial modeling; [35]). Then, the probability that host i interacts with parasite j was given as the outer product of host h and parasite p trait vectors, calculated as the row-wise product of host and parasite trait matrices, where rows correspond to either host or parasite species and columns are traits. This forms a matrix of h rows and p columns. This matrix (M) was scaled to the unit interval by dividing each value by the maximum value observed. Interactions were assigned probabilistically by conducting single binomial trials with probability Mi,j. This process was performed iteratively until a specified connectance value was reached (c = c*).
h = [ h 1 , h 2 , … h i ]
p = [ p 1 , p 2 , … p j]
M = h × p
while(c < c*)
M i , j = 1 if M i , j > U ( 0 , 1 ) - if M i , j < U ( 0 , 1 )
To determine how well the plug-and-play model performed, we tested the predictive accuracy of the model on simulated data. We trained models on 80% of the simulated data, and predicted on the remaining 20% test set, i.e., a setup that assumes only 80% of host-parasite associations to have been sampled. (This criterion is relaxed in the Supplemental Materials where we show how the fraction of the network used for model training influenced predictive accuracy; S1 Fig). The AUC statistic was uesd as a measure of predictive accuracy, and examined how model performance was influenced by interaction matrix size, the fraction of realized links (i.e., connectance), the number of traits used to predict species interactions, and the inclusion of binary (e.g., thresholded at the mean) and uninformative (e.g., standard normal variates) traits (see Supplemental Materials for more information). Unless otherwise stated, species interaction matrices were created and predicted using five host and parasite traits each, and a connectance (c) of 0.2, which reflects observations of empirical host-parasite networks [36].
First, we determined the predictive accuracy of our model on 1000 randomly generated species interaction networks. To examine the influence of interaction matrix size, we varied host and parasite species richness from 10 to 30, and simulated 50 networks for each host and parasite richness combination. The influence of connectance was examined by creating 1000 species interaction networks with 30 host species and 20 parasite species for each value along a gradient of connectance values from 0.05 to 0.35. To examine the influence of host and parasite trait number, we simulated 1000 species interaction networks for each host and parasite trait number combination between 1 and 20 (total of 20,000 networks). The influence of training the model on binary trait data was examined by creating 1000 species interaction networks created using 20 host and parasite traits, and varying the fraction of those 20 traits that were binary from 5% (1 trait was binary) to 100% (all traits were binary). To determine if the inclusion of random, uninformative traits influenced predictive power, we simulated 1000 species interactions networks with 10 host and parasite traits, and included between 1 and 50 random host and parasite traits (50,000 total species interaction networks). Lastly, we tested predictive accuracy when the model was trained only on random traits by creating species interaction matrices (1000 per treatment) and then shuffling trait values.
The plug-and-play model was able to predict links on simulated bipartite networks with high accuracy (S2 Fig). Further, accuracy was not appreciably reduced by matrix size (S3 Fig), incorporation of binary variables (S4 Fig), number of host and parasite traits (S5 Fig), connectance (S6 Fig), or the incorporation of random variables (S7 and S8 Figs). Specifically, we found that more than three host and parasite traits were needed to have a mean AUC value of 0.9, and training on only a single host and parasite trait resulted in moderate predictive accuracy (A U C ¯ = 0.72).
We applied the plug-and-play algorithm to data on parasites of small mammals sampled as part of the Sevilleta Long-Term Ecological Research project. We aggregated data from 1992 to 1997 from six sites in three nearby habitats into one interaction matrix. Details of animal sampling and processing are reported elsewhere [4, 37]. Hosts with fewer than five captures over the six year sampling effort were excluded from analysis, resulting in a total of 22 small mammal host species and 87 parasite species, including both macroparasites (e.g., helminths) and microparasites (e.g., coccidians).
Host trait data were obtained from Pantheria [38], supplemented with published literature sources (see Supplemental Table A1 of [4] for more information). Host trait data included life history traits (Table 1), and phylogenetic information. Phylogenetic relationships were estimated using the first five axes of a principal coordinates analysis (PCoA) on the phylogenetic distance matrix obtained using the mammal supertree [39] and the ape R package [40]. Together, these first five PCoA vectors captured 95% of the variance in the eigenvalues, suggesting that most of the information in the phylogeny was captured in these five vectors.
Host life history traits included host diet breadth, body mass, home range size, maximum age, and species abundance (Table 1). Parasite trait data included three variables representing the life history and transmission modes of parasites; parasite type (arthropod, protozoan, or helminth), parasite genus (genus), and location (intracellular or extracellular). Some host trait data was unavailable, and we imputed the unavailable data using the randomForest R package [41]. This procedure imputes missing data by first replacing missing values with column averages, and then iteratively updating imputed values based on proximity of observations to one another in the random forest model. Variable importance was determined by permuting each predictor variable 500 times, and determining the reduction in model performance as a result for each permutation. Model accuracy (AUC) was determined through 5-fold cross validation. The final model was trained on all available data.
We then determined the number of likely missing links from the host-parasite network, and sequentially added the most likely links as predicted by our trained model. We used the Abundance-based Coverage Estimator (ACE; [42]), commonly used for species richness estimation, to estimate the number of missing links. ACE is a non-parametric species richness estimator typically applied to communities of free-living organisms ([43, 44]) and has previously been demonstrated to perform well for many different coverage levels and survey designs ([45]). We treat links between known hosts and parasites to be equivalent to organisms in the traditional context, which allows us to estimate the likely number of links missing from the network.
At each link addition, we calculated properties of the network to observe how network structure changed with link addition. Some stuctural properties change obviously and deterministically with link addition (e.g., mean degree and connectance), which we ignore. Rather, we focused on stochastic aspects of network structure, including measures previously related to network stability (nestedness; [3, 14]), aggegration of parasite species among host species (togetherness and variance-to-mean ratio; [46]), and measures of interaction clustering or host-parasite co-occurrence (C-score; [47]). The resulting changes to network metrics with model-predicted link addition were compared with changes in network metrics if links were added randomly.
Nestedness, quantified as the NODF metric [48], measures the tendency of hosts with few parasites to harbor nested subsets of the parasite communities of parasite species-rich hosts, and has previously been related to network structural stability [3]. Nestedness was quantified relative a null model, as aspects of matrix size and fill alter the raw measure. Further, the use of the standard score (z-score) allows a quantification of the magnitude of divergence from a null expectation, which is commonly used for significance testing. Thus, this approach allows us to determine changes in the magnitude of nestedness with link addition relative to a null expectation. We used the sequential swap algorithm to randomize matrix interactions [49], and compared the empirical network to 1000 null networks after each link addition.
Togetherness measures the tendency of host species to share parasites, with large values suggesting ecological similarity between hosts may be more important than competition in driving community structure, and small values suggesting the opposite ([12, 50]). The variance-to-mean ratio is an index of aggregation traditionally used in studies of single species parasite distributions [46, 51], where larger values indicate more skewed or aggregated parasite burdens. Here, we use it to express the skew in parasite species richness for a range of host species.
Originally used to infer interspecific competition, the C-score (or checkerboard score; calculated here as the mean pairwise score for all host species) is more generally a measure of non-independence in species interaction patterns, with large values indicating that species occupy different habitats ([47]). These interaction differences could be a result of interspecific competition, dispersal limitation, or differences in host habitat utilization. In terms of host-parasite networks, this would correspond to parasite communities with little overlap in host use, such that parasite communities are clumped across the range of potential host species.
The algorithm we develop here was able to accurately predict missing links in bipartite networks based solely on host and parasite traits, both in simulated networks (see Methods paragraph “Model validation on simulated data”), and an empirical network of small mammal host-parasite interactions sampled as part of the Sevilleta LTER.
The plug-and-play algorithm recovered the Sevilleta small mammal-parasite interaction network structure with high accuracy (AUC = 0.82) when trained on all available data, and performed fairly well during 5-fold cross validation, with a mean AUC from 500 training/test data splits of 0.63, and a maximum observed AUC of 0.81. We permuted predictor variables to obtain measures of variable importance, which suggested that host litter size, parasite genus, and host diet breadth were the most important variables to model performance (Fig 1). Meanwhile, some covariates had a negative effect on the model, resulting in improvement in predictive accuracy with randomization. These included coarse, low-variance variables such as habitat breadth and trophic status, as well as potentially important variables such as parasite type (e.g., helminths), and host body mass. Predictive model accuracy is predicated on the network being fully sampled, such that predicted links that are not observed in the empirical network are treated as errors, and reduce accuracy. We predicted that between 110 and 157 links were missing from the empirical network, changing the connectance from 0.12 to between 0.18 and 0.21.
We then sequentially added the most probable links, based on model-predicted suitability scores (Fig 2), plug-and-play examine how network properties changed. Measures of network structure fluctuated with link additions (Fig 3). Specifically, nestedness, quantified as the z-score in NODF values relative to null models, fluctuated from -4.6 to -0.6. Since these z-scores can be used for significance testing, this suggests that the addition of missing links can change the ability to detect fundamental network properties. Further, togetherness, variance-to-mean ratio, and C-score all declined more strongly with the addition of predicted missing links compared to the addition of random links. Further, togetherness actually increased when link addition was random.
Here, we present, validate, and test a link prediction algorithm that does not require information on network structure for training, extending the problem of link prediction in social networks to bipartite networks. This is important, as network structure is often dynamic, and generalizing link prediction to novel or changing networks is necessary for some applications (e.g., forecasting the most probable prey items or parasites of a novel host species to the network). Our approach allows for the ranking of node characteristics, which can enhance our understanding of what determines the likelihood of species interactions, and for the prediction of cryptic interactions, which can influence network structure.
In our small mammal-parasite network, we determined that host litter size, parasite genus, and host diet breadth were the top three most important predictors of host-parasite interactions. Host litter size was the most important interaction predictor, suggesting the importance of host life history traits. Because host litter size is linked to other aspects of host biology known to alter parasite burdens, such as host metabolic rate [52], we suspect that the importance of litter size in this analysis may reflect an aspect of the host species’ pace of life [53, 54]. The second most important variable to our predictive model was parasite type (i.e., arthropod, helminth, or protozoa), which accounts for unmeasured differences among parasite species in their transmission or host preferences. Lastly, host habitat breadth, which can influence contact rates with parasites was an important variable in our model. Interestingly, despite the previously documented importance of host phylogenetic distance in predicting parasite community similarity [10], we found no evidence that host phylogeny improved predictive accuracy in this system. The inclusion of some covariates actively detracted from model performance, a phenomenon not observed in simulated data. This is likely a result of the low information content of these variables, or could signal the influence of variable interactions on model predictive accuracy.
Our algorithm predicted that between 110 and 157 links were missing from the network. When these links were added based on their suitability score, several network properties changed, including nestedness, togetherness, variance-to-mean ratio, and checkerboard score. While the ability to detect nestedness fluctuated with link addition, the other three metrics of network interaction patterns demonstrated consistent declining trends. This suggests that the interaction patterns became less clumped (as indicated by the checkerboard score), parasite communities became less dissimilar (as indicated by togetherness), and less aggregated (as indicated by variance-to-mean ratio). Taken together, these findings suggest link addition was not confined to species that already had many links, otherwise the variance-to-mean ratio wouldn’t have been reduced. Instead, the addition of missing links reduced overdispersion commonly observed in many host-parasite networks (including in Fig 2).
Ecologists have long recognized the issue of incomplete sampling leading to imperfect detection [55], but only recently have studies of ecological networks addressed this issue [2, 17, 56]. Here, we present an algorithm capable of accurately reconstructing a network using information on interactor traits, and predicting interaction likelihoods. This overcomes the problem of imperfect detection, and allows for the forecasting of the most probable links in ecological networks. Other approaches for the link prediction problem in bipartite networks exist. For instance, recent Bayesian approaches have used occupancy models [17] and Dirichlet network distributions [57]. However, these approaches are largely used to address slightly different problems. The first is an attempt to combine occupancy models with metacommunity analysis, predicting missing links as a means to correct error, and not for the sake of predicting unknown links. The second was developed to predict links in integer-based directed networks, and was developed under the assumption that nodes have repeated and directed interactions, such as a network of email correspondence among a group of people. Extensions of this approach could potentially support binary bipartite networks as we have examined. Another approach, the matching-centrality method [29], allows for the accurate forecasting of unobserved links in both unipartite and bipartite networks. Our approach differs in that we rely solely on trait matching between bipartite interactors to predict interaction probability, meaning that the algorithm is insensitive to network structure (allowing for increased flexibility). Lastly, by relying on host and parasite traits, our approach may provide insight into what host traits, parasite traits, or trait combinations promote the likelihood of a host-parasite interaction, and further provides a way to quantify the relative importance of host and parasite traits to interaction patterns.
Extensions of our current approach could disentangle the effect of disproportionate sampling effort, as well as other host and parasite traits, to provide a more complete understanding of what controls host-parasite interactions. This trait-based approach can be applied to other bipartite networks (e.g., plant-pollinator), as well as to spatial networks (e.g., metapopulations). The incorporation of missing links into networks that change seasonally or are logistically difficult to sample provides a more accurate description of network interactions. Further, the incorporation of these interactions may change basic network properties in non-random ways. The functional consequences for revising our understanding of ecological networks are not currently known.
|
10.1371/journal.pcbi.1004139 | PRIMAL: Fast and Accurate Pedigree-based Imputation from Sequence Data in a Founder Population | Founder populations and large pedigrees offer many well-known advantages for genetic mapping studies, including cost-efficient study designs. Here, we describe PRIMAL (PedigRee IMputation ALgorithm), a fast and accurate pedigree-based phasing and imputation algorithm for founder populations. PRIMAL incorporates both existing and original ideas, such as a novel indexing strategy of Identity-By-Descent (IBD) segments based on clique graphs. We were able to impute the genomes of 1,317 South Dakota Hutterites, who had genome-wide genotypes for ~300,000 common single nucleotide variants (SNVs), from 98 whole genome sequences. Using a combination of pedigree-based and LD-based imputation, we were able to assign 87% of genotypes with >99% accuracy over the full range of allele frequencies. Using the IBD cliques we were also able to infer the parental origin of 83% of alleles, and genotypes of deceased recent ancestors for whom no genotype information was available. This imputed data set will enable us to better study the relative contribution of rare and common variants on human phenotypes, as well as parental origin effect of disease risk alleles in >1,000 individuals at minimal cost.
| The recent availability of whole genome and whole exome sequencing allows genetic studies of human diseases and traits at an unprecedented resolution, although their cost limits the size of the studied sample. To overcome this limitation and design cost-efficient studies, we developed a two step method: sequencing of relatively few members of a well-characterized founder population followed by pedigree-based whole genome imputation of many other individuals with genome-wide genotype data. We show that by sequencing only 98 Hutterites, we can impute 7 million variants in an additional 1,317 Hutterites with >99% accuracy and an average call rate of 87%. Furthermore, parental origin was assigned to 83% of the alleles. Such studies in the Hutterites and other founder populations should yield new insights into the genetic architecture of common diseases, gene expression traits, and clinically relevant biomarkers of disease, and ultimately provide outstanding opportunities for personalized medicine in these well-characterized populations.
| Despite decreasing costs of whole exome and whole genome sequencing, the role of rare genetic variants in common disease risk remains hard to assess due to the very large sample sizes required for such studies [1,2]. Therefore, approaches that allow accurate imputation of rare variants to large numbers of individuals based on the sequences of relatively few individuals could address this important question at minimal cost. Founder populations are particularly suitable to this strategy because pedigree relationships are either known or can be inferred from genotypes, facilitating imputation approaches that incorporate identity by descent (IBD) relationships between chromosomal segments and improving imputation accuracy. Moreover, variants that occur at low frequency (<5%) or are rare (<1%) in large outbred populations, may occur at common frequencies (>5%) in founder populations due to the bottleneck at the time of their founding followed by random genetic drift effects in subsequent generations. Similar to mutations for rare monogenic disorders reaching relatively common frequencies in founder populations [3–6], subsets of the rare variants contributing to common complex diseases are also expected to occur at higher frequencies in these populations. This provides a unique opportunity to study the relative roles of rare and common variants on common disease risk in individuals exposed to similar environments, which further minimizes the contribution of non-genetic factors to inter-individual variation in disease risk and facilitates identification of disease-associated alleles.
Methodological approaches to genotype imputation fall into two general categories depending on whether they are based on linkage disequilibrium (LD) or on genetic relationships (i.e., pedigrees) [7]. LD-based imputation methods require a reference panel of genotype training data, usually from unrelated individuals, to infer local haplotype structure, and sharing of haplotype stretches are used for filling in missing genotypes [8–11]. These approaches typically result in high call rates at the expense of lower accuracy, especially for rare alleles [12]. In contrast, pedigree-based imputation approaches are more accurate because they rely on identifying regions of IBD sharing among the study subjects [13,14]. However, call rates are typically lower than from LD-based methods, and pedigree-based imputation can be significantly slower to implement due to complex pedigree structures, which often pose limitations on maximum family sizes and minimum relatedness of individuals [15].
To address the limitations of LD- and pedigree-based imputation methods, we developed PRIMAL (PedigRee IMputation ALgorithm), a fast phasing and imputation algorithm, to assign genotypes at 7 million bi-allelic variants that were discovered in the whole genome sequences of 98 Hutterites to an additional set of 1,317 Hutterites who had genome-wide genotypes for ~300,000 common single nucleotide variants (SNVs). We first phased the SNV genotypes using pedigree-based phasing algorithms [16,17] and determined IBD segments between each pair of haplotypes using a Hidden-Markov Model [18]. We then organized IBD segments into an IBD clique dictionary, a novel data structure for efficient IBD lookup queries that enables fast pedigree-based imputation of the variants identified in the 98 genomes. We demonstrate that the accuracy of the algorithm is above 99% regardless of minor allele frequency, with a call rate of approximately 77%. To improve the call rate, the missing genotypes were imputed using the LD-based IMPUTE2 program [11], with the phased haplotypes of the 98 whole genome sequenced Hutterites as the reference panel. The result is a hybrid method that combines the benefits of pedigree- and LD-based strategies to obtain similar accuracy (> 99%), and higher call rates (87.3%). Moreover, using the IBD clique dictionary implemented in PRIMAL, we can infer the parental origin of 83% of alleles. We are also able to impute whole genome genotypes to recent ancestors with no available DNA. The PRIMAL algorithm and software will facilitate genetic studies of rare variants and parent-of-origin effects in the Hutterites and in other founder populations with similar data.
This study was conducted according to the principles expressed in the Declaration of Helsinki. All participants in the experiment provided written informed consent in approval with the University of Chicago Institutional Review Board.
The Hutterites originated in central Europe in the 1500s. After a series of migrations and population bottlenecks, they settled in what is now South Dakota in the 1870s, and currently live on communal farms in the northern U.S. plains states and western Canadian provinces [19]. At present, there are over 14,000 Hutterites living in South Dakota, all of whom are descendants of just 64 founders and related to each other with a mean kinship coefficient of 3.4% [20]. This study includes 1,415 Hutterites who previously participated in one or more of our studies of Mendelian and common diseases and associated phenotypes (e.g., [5,21]). These individuals are related to each other through multiple lines of descent in a 3,671-person minimum pedigree.
We genotyped DNA from Hutterite individuals using one of three Affymetrix arrays (500k, 5.0 and 6.0), as previously described [21,22,23]. As part of our quality control (QC) process, we removed SNVs with five or more Mendelian errors, Hardy-Weinberg p-values < 0.001, or call rates <95%, resulting in 332,242 SNVs present on all three platforms. The final sample included 1,415 Hutterites with genotype call rate > 95%. We used the subset of 271,486 SNVs with minor allele frequency (MAF) ≥ 5% for phasing and imputation in this study. These SNVs are referred to as the “framework SNVs”, and genotyped individuals for whom both parents were not genotyped are referred to as the “quasi-founders” of this sub-pedigree.
Ninety-eight Hutterites were selected from the 1,415 for whole genome sequencing (WGS) to maximize their relatedness to the other 1,317 Hutterites (and thus leading to high pedigree-based imputation call rates), while minimizing the pairwise relatedness among the 98 (to reduce the amount of redundant sequencing). To achieve this we used a greedy algorithm described elsewhere [16] where subjects were selected sequentially to maximize the average kinship to the non-sequenced individuals, while imposing a kinship smaller than 0.1 with the sequenced individuals. Sequencing was performed by Complete Genomics, Inc. (Mountain View, CA). A total of 18.2 million variants (14.0M SNVs, 2.7M insertions, 1.4M deletions; Table 1) were discovered in the 98 WGS, including 11.6 million variants (9.2M SNVs, 1.3M insertions and 1.1M deletions) for which both alleles were called as high quality by Complete Genomics. Using the 332,242 SNVs, the concordance between the genotypes from the whole genome sequences and those determined by genotyping with the Affymetrix arrays was 99.8%.
To investigate the quality of sequencing-based genotypes for classes of variants (for example, all novel singletons), we developed a pedigree-based method to assess genotyping errors. The method is an extension of the classical Mendelian error checking in families. However, in contrast to Mendelian checks that use parents and their offspring, our approach includes all pairs of related individuals, regardless of the distance of the relationship, using their IBD segments. High confidence IBD2 segments (i.e., IBD = 2, or regions where two individuals inherited the same chromosomal segments from a common ancestor) were previously calculated between each pair of individuals among the 98 Hutterites using the 332,242 framework SNVs [24]. Next, for each sequenced variant, we determined the number of IBD2 segments shared between pairs of individuals that contain the variant and counted the number of discordances (the number of pairs of IBD2 segments in which the genotypes for the variant under investigation did not match). We then estimated the variant calling discordant rate (the proportion of discordances) for each class of variants as the total number of discordances divided by the total number of pairs of IBD2 segments in that category. Discordant rates increased with decreasing call rate, suggesting poorer quality of genotype calls for variants with more missing data. Thus, we determined call rate cut-offs for each variant class to maintain a less than 0.5% discordant rate. This resulted in a final set of variants that included all non-singletons (i.e., variants in which the rare allele occurred at least twice) with rs numbers (in dbSNP135) with call rates > 90% and novel variants (no rs number in dbSNP135) with call rates > 99% (i.e., at most one missing call). Among singletons (variants with one copy of the rare allele in the sequenced subjects), we retained novel insertions with call rates > 90% and all other variant types with call rates > 99% (Table 1, and Fig S2 in S1 Text). The allele frequency distribution and functional annotation of the final set of 7,008,666 variants in the 98 Hutterites with WGS are shown in Fig S2 in S1 Text.
The quality of imputed genotypes was assessed by comparing them to data from a different whole-genome sequencing study in five parent-offspring trios who were among the 1,317 Hutterites in our study [25]. These 15 individuals were sequenced on the Illumina platform at a 10–17x coverage. High quality (as determined by Illumina) genotypes were extracted for all the SNVs imputed using PRIMAL and that passed QC. One of the 15 subjects was sequenced on both platforms and this allowed us to estimate the joint sequencing error rate. Discordance rates between the Illumina sequence-based and PRIMAL-imputed genotypes were calculated as the proportion of differences in genotypes in each of the remaining 14 individuals using these two methods.
The algorithm described in Results is implemented in software, PRIMAL v1, that is freely available for academic use from the website: https://github.com/orenlivne/ober
Our imputation algorithm consists of five main stages (Fig. 1, steps 4–8). The first four require only the framework SNVs: (i) phasing; (ii) identifying IBD segments among all haplotype pairs; (iii) indexing IBD segments into a dictionary of IBD cliques; and (iv) assigning parental origin to haplotypes. In the fifth step, we phase the WGS-derived genotypes, and then perform fast pedigree-based imputation of all variants present in the WGS using the IBD clique dictionary.
Our phasing method is similar to the long-range phasing algorithms described by Kong et al. [17] and Glodzik et al. [13] and to our earlier phasing algorithm for Hutterite genotype data [16], but introduces two key improvements that boost its quality. First, we use a phased proband as a template to phase siblings in nuclear families as in Coop et al. [26] (Supplementary Materials S1 Text), and second, we employ a Hidden Markov Model (HMM) similar to the IBDLD model [24] to identify IBD segments between a proband and his/her surrogate parents (Fig S3 in S1 Text). The phasing workflow is outlined in Fig S3 and described in detail in S1 Text. Using this approach, only 0.5% of the framework genotypes remained unphased, 99.2% of the genotypes were correctly phased, and the remaining 0.3% of the framework genotypes were discordant.
During the phasing step, IBD segments are identified, but only between the individual being phased and his/her surrogate parents. Therefore, we created a complete IBD dictionary by identifying IBD segments between each pair of the 2×1,415 = 2,830 haplotypes in the sample (S1 Text). Computational complexity prevented us from using available software to estimate IBD segments in related individuals [27–29]. Our HMM is the haplotype analogue of the genotype HMM used for phasing, and is similar to the HBD-HMM developed previously [18]. However, only kinship coefficients are used instead of condensed identity coefficients. The complexity is quadratic in the number of samples, but the hidden constant is small because only two states (IBD or not IBD) are possible instead of the nine in the genotype HMM (Table S1 and S1 Text).
A total of 97,821,947 IBD segments were identified among the 1,415 Hutterites (~1.1 segment per haplotype pair on average, because there are 2830×2829/2 = 4,003,035 individual pairs and 22 chromosomes). To verify the overall quality of the detected IBD segments, our fraction of the genome covered by IBD segments was compared to the fraction calculated by IBDLD [24]. The methods were concordant (correlation coefficient r = 0.96 with a slope of β = 1.01) and the length distribution followed an exponential distribution, in accordance with theory [30].
We organize IBD segments in an IBD segment index data structure, which consists of a set of IBD cliques at each SNV and allows a quick O(1)-time queries of whether a pair of haplotypes is IBD at a certain SNV.
At each SNV, we build a weighted, undirected pairwise IBD graph G (Fig. 2) whose nodes are the 2,830 haplotypes of the 1,415 Hutterites, an edge indicates the two haplotypes are IBD, and the edge weight is the HMM posterior probability of IBD (S1 Text, Eq. (19c)). Large weights are thus given to haplotype pairs that have a higher probability of being IBD.
Because IBD is a transitive relation, G must be a union of disjoint cliques (fully connected sub-graphs), one for each ancestral haplotype present in the population. In practice, G is a perturbation of a clique union due to very low HMM certainty near segment ends and genotyping errors, and we would like to recover a “reasonable'' set of cliques from it. Cluster editing methods (see for example [31]) find the minimum number of edges (or total edge weight) that need to be added or removed to transform G to a clique union. This is an NP-hard problem, and practical heuristic-based algorithms run in superlinear time in the number of edges. We chose a different heuristic inspired by the graph algebraic multigrid literature [32–34] that resulted in good imputation cross-validation accuracy and has linear complexity (S1 Text). First, we calculate new edge weights called affinities that measure the connectedness or affinity between the graph neighborhoods of the nodes (Fig. 2). A large affinity means that the nodes share many common neighbors, i.e., they are connected via many short paths. Next, we removed graph edges with weight < 0.85 or affinity < 0.9. These thresholds were chosen to minimize imputation errors in a cross-validation of several framework SNVs representing the entire MAF spectrum. Finally, each of the resulting graph’s connected components is transformed to a clique by adding links between all nodes that are not yet connected (Fig. 2 and Fig S4 in S1 Text). This method worked well for our data set, and these thresholds should be good default values for other data sets. However, threshold determination and a comparison with other clique-generation methods undoubtedly need to be further investigated in a future research.
The use of cliques significantly speeds up imputation because all haplotypes in a clique are imputed simultaneously. In addition, cliques allow the derivation of the maximum call rate obtainable per SNV from imputation, which is the ratio of the number of haplotypes in cliques containing haplotypes of sequenced individuals to the total number of haplotypes. The predicted imputation rate was 85% ± 9% for the framework SNVs. Note that, using pedigree-based imputation, the accuracy approaches 100% (because we rely on Mendelian rules).
Genotyped individuals are considered “quasi-founders” if either of their parents were not genotyped. Haplotypes of non-quasi-founders can be automatically labeled as paternal and maternal because their parents are included in our sample and haplotypes are assigned using Mendelian rules. However, because the quasi-founders do not have genotyped parents, the parental origin of the quasi-founder haplotypes is assigned in two stages.
First, during phasing, we do not determine which haplotypes are paternal and maternal, but we ensure that the first haplotype of every child comes from the same parent (arbitrarily denoted A), and the second haplotype from the other parent (arbitrarily denoted B). This is achieved using the following steps:
a)Regions of the children’s haplotypes are assigned to four different “bins” (illustrated as four colors in Fig S5 in S1 Text) that represent the four parental haplotypes. Regions that are IBD are in the same bin, under the constraint that the number of recombinations be minimized.
b)There are three possible assignments of four parental haplotypes to parents A and B. For each assignment, we calculate for each child C with haplotypes C1, C2 a separation measure as follows: let F1 be the fraction of C1 covered by A’s haplotypes plus the fraction of C2 covered by B’s haplotypes, and F2 be the fraction of C1 covered by B’s haplotypes plus the fraction of C2 covered by A’s haplotypes. The separation is the ratio max(F1,F2), which measures how decisively C’s haplotypes can be identified as paternal or maternal haplotypes.
c)We pick the parental assignment that maximizes the minimum child separation, and order C1, C2 in all children so that the first always corresponds to parent A and the second to parent B. The separation measure is defined in S1 Text.
Next, after parental origin is assigned to haplotypes within each nuclear family (both parents and their children), we calculate a different separation measure at each SNV for each quasi-founder C. Let 1 and 2 denote the child’s haplotypes, C1 and C2 the corresponding IBD cliques, and A and B representing C’s untyped parents. For each parent and each clique, we calculate the median of the set of kinship coefficients between the parent and all quasi-founders in the clique that are not siblings of the proband (the quasi-founder in question), resulting in a 2×2 matrix (Fig. 3; siblings and non-quasi-founders are excluded to minimize bias). For each SNV, indexed by s, we define a separation measure m(C, s) (precisely defined in the Supplementary S1 Text, Eq. (4)) such that-1 ≤ m(C, s) ≤ 1. The measure approaches-1 when the off-diagonal matrix elements are much larger than the diagonal elements, and approaches 1 when the diagonal elements dominate. If the proband is properly phased, m(C, s) must be consistently positive or negative across the chromosome. We consider only “informative variants” as those where |m(C, s)| > 0.25 is separated from 0. Suppose there are n+ informative variants with m(C, s) > 0 and n- with m(C, s) < 0; the sample separation measure M(C) is defined as max(n+, n-) /(n+ + n-). That is, the fraction of variants exhibiting the “majority sign”. We assign parental origin when M(C) > 0.75. Using this approach we were able to assign parental origin to 76% (313 out of 411) of the quasi-founders’ chromosomes, with 279 having M(C) > 0.99 (Fig S6 in S1 Text). Including non-quasi-founders, we were able to assign parental origin to 93% of the sample.
Once the IBD clique dictionary is constructed, imputation is performed separately and in parallel for each variant present in one or more of the 98 whole genome sequences. The main idea behind the approach is that each sequencing-based allele that is phased on a particular haplotype can be imputed to all the haplotypes in its IBD clique. First, homozygous genotypes are phased, and the alleles and indices of the two haplotypes are placed into a queue. We remove the first haplotype from the queue, and impute all haplotypes in its IBD clique with the same allele. If these include haplotypes of heterozygous genotypes in the 98 sequenced individuals, they can now be phased. For each such individual, we add its other haplotype index and allele to the end of the queue. The next entry in the queue is then similarly processed, except that, when there is conflicting allele information within a clique (when a two-third majority vote does not exist), no haplotype is imputed. We process queue entries one by one until the queue becomes empty.
Using this approach, we imputed 7M variants (Table 2, columns 4–5) in about 75,000 CPU on Beagle, a 150 teraflops, 18,000-core Cray XE6 supercomputer at the Computation Institute, a joint initiative between The University of Chicago and Argonne National Laboratory [35]. Finding and indexing IBD segments into cliques takes the majority of computing time in the PRIMAL pipeline. The dominant complexity term is O(n2s), where n = 1415 is the number of genotyped individuals and s = 271,486 is the number of framework markers (S1 Table in S1 Text, columns 2–3).
The overall genotype call rate was 76.2%. The mean individual call rate was 75.5%; 547 out of 1317 individuals (41%) had call rate ≥ 80%. Call rates were higher in regions with higher framework SNV density, lower recombination rate and farther from the telomeres (Fig S7a in S1 Text). Fig S8a in S1 Text shows that the MAF distributions of European ultra-rare SNVs (MAF = 0 in the 1000 genomes CEU database) are comparable in both the 98 sequenced Hutterite sample set and the 98 sequenced + 1,317 imputed Hutterites (n = 1,415). Furthermore, we compared the Alternative Allele Frequency (AAF) in the Hutterites and CEU sample set. The Hutterite and CEU AAF were highly correlated (Fig S8b-d in S1 Text). Out of 6,715,275 variants that were not A/T or C/G SNVs, 5,299,330 had similar CEU and Hutterite AAFs (absolute difference < 0.1); there were more variants with larger AAF in the Hutterites than in CEU compared to the opposite case (880,912 vs. 534,012 variants).
To check the accuracy of PRIMAL imputed genotypes, their concordance with the framework genotypes was assessed. First, we phased the framework (Affymetrix) genotypes, identified IBD segments and indexed them into cliques. We then masked the framework genotypes of the 1,317 individuals whose genomes were not sequenced, imputed the framework genotypes, and calculated the concordance between the imputed and true genotypes over a sample of 53,861 framework SNVs (sorted by base-pair position, every 5th framework SNV was picked instead of using all SNVs to save computing time). The concordance was close to a 100% regardless of MAF (Fig S7c in S1 Text). In addition, we also tested for heterozygote concordance rate within the variants with MAF < 5% because the concordance over all genotypes would be high even if they were randomly imputed. The heterozygous concordance also approached 100%.
We also calculated concordance rates between imputed genotypes based on the 98 Hutterites sequenced by Complete Genomics and genotype calls for 14 Hutterites who were sequenced on the Illumina platform as part of a separate study [25]. The concordance rate for each subject was larger than 99% (the concordance rates ranged from 99.3% to 99.8%) with an overall average of 99.7%. This overall rate is very similar to the rate of concordance obtained from the subject sequenced on both platforms.
The use of cliques significantly speeds up imputation and also allowed us to determine that the maximum predicted imputation rate is 85% for the framework SNVs. However, while genotypes imputed by PRIMAL had high accuracy, the call rate (77%) is lower than the maximum predicted rate, most likely due to imperfect phasing of variants without a consensus allele. To mitigate this problem, we filled in as many genotypes as possible for the remaining 23% of variants using LD-based imputation. We chose IMPUTE2 [11] because of its ease of use, high speed and high imputation accuracy. Importantly, we used the high quality pedigree-based phased haplotypes from the 98 whole genome sequenced individuals as the reference panel. This boosted the IMPUTE2 accuracy (evidenced by the measures described below) and reflects the accuracy of our phasing. To obtain data that are consistent in format and accuracy to those generated by PRIMAL, IMPUTE2 genotype probabilities were converted to hard genotype calls only if the maximum probability among the three possible genotypes was > 99%; otherwise, they were not called. When using this criterion, the concordance rates between IMPUTE2 genotypes and those based on sequencing in the 14 individuals range between 99.5 and 99.8% with an overall average of 99.7% (identical to PRIMAL).
As a QC check on this second round of imputation, we calculated overall as well as heterozygote concordance rates between PRIMAL and IMPUTE2 imputed SNVs. All genotypes called by both methods and called as heterozygous by at least one of them were included. IMPUTE2-imputed genotypes were retained only if the heterozygous concordance rate was ≥ 99% and the MAF ≥ 1% (heterozygous concordance rate drops significantly for variants with MAF <1%—Fig S9 in S1 Text). Finally, the PRIMAL+IMPUTE2 combined method yielded an overall call rate of 87.3% with > 99% estimated accuracy (Table 2).
We also used LD-based imputation to increase the parent of origin (PO) assignment for each allele. First, we created a data set with twice the number of samples (2N). For each subject, we created “paternal haploid” and “maternal haploid” sets. For unphased genotypes, the haploid entries were set to missing. We ran IMPUTE2 on the haploid data set. We then assigned parental origin to each genotype called by IMPUTE2 in the original data set only if both the PO of the paternal and maternal haplotypes were imputed with maximum probability > 99% and were compatible with the genotype. PRIMAL alone assigned PO to 80% of alleles, but with IMPUTE2 directly imputing from PO-assigned haplotypes, we increase the PO call rate to 83%.
Despite trends over the past nearly 20 years toward genetic association studies in large case-control samples [36], there have been strong arguments for, and a recent re-appreciation of the advantages of family studies for understanding the genetic architecture of complex phenotypes [37–39]. For example, family-based studies are particularly well suited for discovery of rare disease-associated variants and revealing parent-of-origin effects while minimizing potential confounding due to population substructure and genetic and environmental heterogeneity. Moreover, the family structure itself allows more extensive quality control checks of genotype data and ultimately more accurate genotype calls. Now, in the era of whole exome and whole genome sequencing, studies in families and founder populations offer a new, powerful framework for mapping studies because the genome or exome sequences of relatively few ‘founders’ are needed to impute highly accurate whole genome genotypes to other members of the pedigree with only framework genotypes.
We describe in this paper a fast phasing and computationally efficient imputation method (PRIMAL) that combines the advantages of pedigree-based and LD-based methods and obtains accurate genotypes (>99%) and high (87%) call rates in 1,317 related Hutterites using whole genome sequencing data on only 98 related individuals, providing unprecedented coverage of genetic variation in a population sample with extensive phenotyping and demographic data. The call rates and, to a lesser degree the concordance rates, are correlated to the degree of relatedness between the imputed individuals and the sequenced subjects. Fig S16 in S1 Text illustrates these relationships, and suggest that the rates are mostly influenced by the few sequenced subjects who are most related to the imputed individual. Note that similar accuracy can be achieved using IMPUTE2 (as detailed above), with a call rate of 84% when restricting to the high quality called genotypes.
In addition, PRIMAL allows accurate parent-of-origin assignments for each allele as well as imputed genotypes of recent ancestors (or other members of the pedigree) with no DNA or available genotype information. This additional information is unique to this approach, and is crucial for many analyses, such as those looking for parent-of-origin effects in associated variants, and imprinting. PRIMAL can be applied to other founder populations or to large families to provide accurate and nearly complete genotype coverage for relatively very small cost and minimal computation time. The quantity and quality of the genotypes generated using PRIMAL will depend on several factors including the family structures, the extent of IBD sharing between the reference and target subjects, and the quality of framework genotypes that are used for inferring the IBD cliques.
In addition to comprehensive surveys of the effects of all variants present in the Hutterite genomes on risk for common and Mendelian diseases and on disease-associated quantitative phenotypes, these data will facilitate association studies with the > 460,000 variants that are rare (<1%) in European populations but have risen to common (>5%) frequencies in the Hutterites and investigations of the effects of maternally-inherited versus paternally-inherited alleles on disease risks and quantitative trait values, and will allow the incorporation of the additional information from IBD sharing in more efficient genetic association studies. Such studies in the Hutterites and other founder populations should yield new insights into the genetic architecture of common diseases, gene expression traits, and clinically relevant biomarkers of disease, and ultimately provide outstanding opportunities for personalized medicine in these well-characterized populations.
|
10.1371/journal.pgen.1003112 | A Genome-Wide RNAi Screen Reveals MAP Kinase Phosphatases as Key ERK Pathway Regulators during Embryonic Stem Cell Differentiation | Embryonic stem cells and induced pluripotent stem cells represent potentially important therapeutic agents in regenerative medicine. Complex interlinked transcriptional and signaling networks control the fate of these cells towards maintenance of pluripotency or differentiation. In this study we have focused on how mouse embryonic stem cells begin to differentiate and lose pluripotency and, in particular, the role that the ERK MAP kinase and GSK3 signaling pathways play in this process. Through a genome-wide siRNA screen we have identified more than 400 genes involved in loss of pluripotency and promoting the onset of differentiation. These genes were functionally associated with the ERK and/or GSK3 pathways, providing an important resource for studying the roles of these pathways in controlling escape from the pluripotent ground state. More detailed analysis identified MAP kinase phosphatases as a focal point of regulation and demonstrated an important role for these enzymes in controlling ERK activation kinetics and subsequently determining early embryonic stem cell fate decisions.
| Embryonic stem cells and induced pluripotent stem cells represent potentially important therapeutic agents in regenerative medicine. Manipulation of these cell types could allow us to replace dead or diseased cells in our bodies and hence potentially provide a solution to a wide range of medical problems. However, before we can perform such manipulations, we need to understand how the stem cells are wired so that we are able to re-wire them in a logical way to produce the desired cell types. Here we have attempted to understand this wiring by using an RNAi screen in which each individual component of the cell is systematically removed and the consequences on cellular fate determined. We have identified hundreds of genes that are required for efficient loss of stem cell characteristics and hence conversion into other cell types. By studying a subset of these genes, we have been able to show that many converge on two related negative regulators of one of the key pathways that act to promote loss of stem cell identity. These negative regulators, Dusps, normally limit the ability of stem cells to change their function and hence be converted to different cell types.
| Embryonic stem cells and induced pluripotent stem cells (iPS cells) are currently generating intense interest due to their potential therapeutic roles in regenerative medicine (reviewed in [1]). We are beginning to understand the rules governing the establishment and maintenance of the pluripotent state and, in particular, the signaling and transcriptional networks which define this state (reviewed in [2]–[3]). A number of genome-wide si/shRNA screens have been instrumental in deciphering these networks [4]–[6]. In contrast, less attention has been directed towards understanding how embryonic stem cells lose their pluripotency and begin to differentiate.
Mouse embryonic stem cells can be maintained in a pluripotent state by culturing under a variety of defined conditions (reviewed in [7]). Traditionally, these cells are cultured in medium containing serum and the cytokine leukaemia inhibitory factor (LIF) [8]–[9]. However, more recently, it was demonstrated that mouse embryonic stem cells can be maintained in a pluripotent ground state by using two specific protein kinase inhibitors (known as “2i” conditions) which target the ERK pathway component MEK and glycogen synthase kinase (GSK3) ([10]; reviewed in [11]). Removal of these two inhibitors promotes exit from the naïve ground state. These studies therefore revealed an important role for the ERK and GSK3 pathways to enter into lineage commitment (reviewed in [12]). Moreover, the suppression of ERK signalling in the mouse embryo is sufficient to expand the pluripotent compartment in the early mouse embryo [13] and can enhance the efficiency of iPS cell generation by promoting completion of reprogramming [14]–[15]. Importantly, the same pathways may operate in a functionally analogous manner in human pluripotent stem cells that have been genetically manipulated [16]–[17]. The ERK pathway has previously been shown to trigger mouse ES cell differentiation [18]–[19] and is implicated in numerous developmental processes (reviewed in [20]) in addition to playing an important role in a variety of different stem cell types (reviewed in [21]). Less is known about GSK3 function in development and stem cell biology and the role for GSK3 is usually attributed to its ability to regulate β-catenin stability and hence limit the responses to Wnt pathway signalling (reviewed in [11], [22]). Recently, a β-catenin-dependent mode of action has been demonstrated for GSK3 in the context of mouse embryonic stem cells, although this mode of action is not sufficient to explain all the effects of GSK3 signalling in this context ([23]–[24]; reviewed in [25]).
One major function of ERK MAP kinase signalling, is to orchestrate gene expression programmes in the cell. In particular, this pathway directly targets a number of transcription and chromatin regulators and thereby controls their activities (reviewed in [26]–[27]). However, which of the ERK targets are important in embryonic stem cell differentiation are unknown. It is also unclear how the canonical ERK pathway is controlled in these cells. In this study, we took advantage of the fact that the combinatorial use of ERK pathway and GSK3 inhibitors maintains mouse embryonic stem cell pluripotency [10] and carried out a genome-wide siRNA screen to identify regulators and mediators of these pathways that influence the exit from pluripotency. This has led to the identification of over 400 genes whose functions are required for efficient embryonic stem cell differentiation away from the pluripotent ground state. The vast majority of these genes have not previously been implicated in this process; therefore our study provides an important new resource for the community. Moreover, further downstream analysis has partitioned these genes into classes that functionally interact with the ERK and/or GSK3 pathways and has revealed an important role for MAP kinase phosphatases in controlling embryonic stem cell fate.
To identify the programme of genes involved in the loss of pluripotency and subsequent differentiation of embryonic stem cells, a genome-wide RNAi screen was performed using E14Tg2a mouse ES cells which are engineered to express an unstable version of GFP from the endogenous rex1 (also known as zfp42) locus. This reporter gene is regulated in an analogous manner to endogenous rex1 [23] and provides a convenient readout for the loss of a naieve pluripotent stem cell marker Rex1 [28] (reviewed in [11]). Rex1GFPd2 ES cells were maintained in media containing MEK and GSK inhibitors (2i) to maintain their ES cell status and treated with siRNAs pools targeting ∼17,000 individual genes. After 24 hrs, cells were exchanged into fresh media lacking these inhibitors and the levels of GFP in each cell were assessed over time (Figure 1A). A gradual loss of GFP expression occurred upon inhibitor withdrawal over a ∼2 day time period, with conversion of the majority of cells to low expression (1A). We wanted to conduct the screen at the earliest possible time point to maximise the chances of detecting genes directly involved in the exit from pluripotency rather than secondary effectors. The control siRNAs for fgf4 and gsk3β both significantly reduced GFP loss at 27–30 hrs (Figure S1B). Therefore we monitored the ratio of cells expressing high and low levels of GFP at this time point. siRNAs were scored as positive hits when this ratio increased by more than two standard deviations (SD) above the mean of all siRNAs on each plate. A conservative threshold was selected at this stage to be more inclusive before further downstream validation was performed. This led to the identification of 792 siRNAs that delayed the loss of GFP expression, and hence target genes potentially involved in promoting pluripotency loss and/or cell differentiation (Figure 1B; Table S1A). Examples, include 2400001e08rik, raf1 and jarid2 (Figure 1C; Figure S2A–S2D, left panels). Importantly, this primary screen identified RNAi pools targeting nras, raf1 and gsk3β, as would be expected due to their known roles in the ERK and GSK3 pathways. Moreover, further validation of the efficacy of our screen was demonstrated by the identification of a large number of siRNAs targeting genes encoding proteosomal proteins, as would be expected due to the subsequent increased half-life of the unstable GFP protein used as a readout in these assays. In addition, this primary screen also revealed 130 siRNAs that accelerate the loss of GFP expression and hence target genes that function to maintain pluripotency and/or inhibit cell differentiation including known effectors such as esrrb, stat3, and ctr9 [4], [29]–[30](Figure 1B; Figure S2E; Table S2). Furthermore, several of genes identified in our screen in this category were also identified in other screens designed to identify genes required for pluripotency [4]–[5], [31]–[33], including stat3 and smc1a (both identified in 2 and 3 additional screens, respectively) (Table S3). As our primary interest was on the mechanisms of escape from the pluripotent ground state rather than the maintenance of pluripotency, we subsequently focussed on genes that were required for modulating the onset of differentiation. Two secondary screens were performed with a different set of siRNA pools targeting the genes identified in the primary screen and either the same reporter cells (ie Rex1GFPd2) or ES cells containing an alternative reporter gene, where GFP is instead driven by the oct4 (also known as pou5fl) promoter, thereby providing an independent readout for the loss of pluripotency (Figure 1A; Figure S1C). These screens gave rise to 398 and 420 positive hits respectively, and 316 of these siRNAs scored positive in both secondary screens (Figure 1A, 1D and 1E; Figure S2A–S2D, Table S1B and S1C). These 316 siRNAs therefore define a high confidence dataset of genes that are required for the efficient loss of pluripotency and/or promoting the onset of differentiation of ES cells. A number of these genes have already been implicated in embryonic stem cell differentiation control including tcf7l1(tcf3), jarid2, and dpy30 [34]–[37] (Table S4) further supporting the quality of our dataset. Moreover, comparisons to other RNAi and overexpression screens performed on mouse ES cells [4], [31]–[33], [38] identified several genes in common, including jun and mbd3 which were both identified in two of these screens in addition to our own (Table S3). However, the vast majority of genes we have identified here, have not been previously implicated in controlling the escape from the pluripotent ground state. To assess the types of biological processes and potential mechanisms of actions of these 316 genes, gene ontology (GO) analysis was performed and prominent terms identified included a number of signalling pathways and also genes encoding transcriptional regulators (Figure 1F and 1G; Figure S3). Thus cellular signalling events and subsequent gene expression control appear to play prominent roles in the early events associated with ES cell differentiation.
Having established the core network of genes working in concert with the GSK3 and ERK pathways we wanted to discover the relative contributions of these genes to the actions of the individual pathways. First we performed a counter screen in the presence of both pathway inhibitors (“+2i”) to eliminate siRNAs which promoted accumulation of GFP in the cells irrespective of the activity of the ERK and GSK3 pathways (Figure 2A). This eliminated a further 42 siRNAs, including 14 that targeted proteosomal components and hence stabilised the GFP (Table S5). This left 274 siRNAs which define genes required for efficient signal-dependent loss of pluripotency and the onset of differentiation. The differentiation of ES cells away from pluripotency is maximally promoted by removing inhibitors of both GSK3 and the ERK pathway. However, the removal of a single inhibitor permits ES cell differentiation and loss of Rex1-GFP signal, albeit with delayed kinetics (Figure S4). We took advantage of this to partition our dataset and identify genes whose functions are specifically required for differentiation driven by either the ERK pathway or the GSK3 pathway alone. siRNAs targeting the genes constituting the high confidence data set from the “2i” withdrawal screens were tested for their effect on Rex1-GFP loss upon single inhibitor withdrawal (ie “1i” withdrawal screens; Figure 2A). Of the 274 siRNAs tested, 133 delayed GFP loss upon withdrawal of the MEK inhibitor and 168 upon withdrawal of the GSK3 inhibitor. Amongst these, 106 were in common. A further 79 siRNAs had no effect on Rex1-GFP expression under either condition (Figure 2A and 2B; Figure S5). Thus there are four functionally distinct classes of hits identified that are involved in promoting the onset of differentiation: (i) in the context of the ERK pathway (“ERK only hits” eg nras, Figure S2A; identified upon MEK inhibitor withdrawal only); (ii) in the context of the GSK3 pathway (“GSK only hits” eg dmbx1, Figure S2B; identified upon GSK3 inhibitor withdrawal only); (iii) in the context of either pathway (“ERK/GSK hits” eg jun, Figure S2C; identified upon GSK3 or MEK inhibitor withdrawal); and (iv) in the context of both pathways together (“ERK and GSK hits” eg gli3, Figure S2D; no effect when either inhibitor is withdrawn). Next to gain an insight into how the ERK and GSK3 pathways might function in the context of embryonic stem cells, we used gene ontology analysis to determine whether different groups of genes identified from the single inhibitor (“1i”) screens are associated with different biological processes. Generally, the enriched GO terms for the genes from the initial 2i screen closely resemble those enriched in the “GSK” dataset (Figure S6). However, closer inspection of the data revealed enriched GO terms that are more specific for genes which were associated with either the ERK or the GSK3 pathway, thereby revealing functionally distinct contributions of these pathways to the exit from pluripotency (Figure 2C and 2D; Figure S7A–S7D). For example, genes associated with either the ERK or GSK3 pathways are enriched in different signalling pathways (Figure 2C) and a number of terms associated with mitochondrial function are preferentially enriched in the genes associated with the GSK pathway (Figure S7C). However, other groups of GO terms were identified with generally high enrichment for genes associated with both the GSK and the ERK pathways. This is typified by a large number of GO terms associated with transcriptional control (Figure 2D). Weaker enrichment of specific terms could be discerned for genes functionally associated with either the ERK or the GSK pathways (Figure 2D).
We then created a network out of the genes from the high confidence dataset identified in the “2i” screen based on previous knowledge of physical and functional interactions. Functionally related subnetworks could be identified, two of the most prominent of which are composed of genes encoding proteins associated with regulating chromatin modifications and sequence-specific DNA binding transcription factors (Figure 2E; Figure S8A). These genes showed strong interconnectivities with the rest of the network as might be expected from their regulatory functions. Although only a limited number of connections between ERK and GSK3 signalling pathway components identified in the screen were revealed during network construction, these connections are made to transcription and chromatin regulators associated with the correct respective pathways (eg Jun is connected to the Ras pathway and Gli3 is connected to Gsk3β; Figure 2E; Figure S8B).
In summary, by comparing single inhibitor assays, we have been able to subcategorise the genes required for embryonic stem cell differentiation and tentatively assign them to mediating or regulating the effects of either the ERK pathway, or the GSK3 pathway or both. Each pathway appears to require genes associated with overlapping and yet distinct biological processes.
Our RNAi screen identified genes belonging to many functionally related categories and they are potentially involved in many biological processes. However, to begin to understand the roles of the genes we have identified in controlling the loss of pluripotency and subsequent differentiation, we decided to focus mainly on the genes which were required for ERK-mediated differentiation as this pathway has a well established role in triggering mouse ES cell differentiation [18]–[19]. The majority of “ERK only” genes and a subset of “ERK/GSK” genes were taken for further investigation alongside several control genes from the “GSK only” hits (Figure 3A). The relative strength of the effect of the knockdown of each gene in the context of the “1i” screens is illustrated in Figure 3A. First we validated the roles of these genes by using RT-qPCR to monitor the loss of the pluripotency markers rex1 and nanog and the appearance of the early differentiation marker fgf5. The majority of the siRNAs tested showed increased rex1 and nanog expression relative to control siRNAs upon “2i” withdrawal (Figure 3B; Figure S9B). Importantly, an excellent correlation was observed between effects on rex1 and nanog expression (Figure 3B; R2 = 0.81). Conversely, more than half of the siRNAs tested reduced the accumulation of fgf5 mRNA (Figure S9C). However, there was generally reduced concordance between the severity of the effects on fgf5 and rex1 (Figure 3C) or fgf5 and nanog (Figure S9D) expression. For example depletion of jarid2 and pabpc1 causes some of the largest effects in maintaining rex1 expression but has no effect on reducing fgf5 accumulation. Conversely, reductions in ets1 and dmbx1 limit fgf5 expression while having only a small effect on rex1 expression. Nevertheless, a group of siRNAs can be identified that limit the loss of rex1 expression and show reduced accumulation of fgf5 (Figure 3C; quadrant 1) and hence have effects on both loss of naive pluripotency and the onset of differentiation. In contrast, there is another large group of genes that appear to affect pluripotency status but have little effect on the onset of early differentiation (Figure 3C, quadrant 2). It is unclear why this occurs but it might reflect that although individual siRNAs promote retention of pluripotency, they might also trigger the activation of subsets differentiation markers, thus the two processes need not be tightly linked. To extend the analysis of differentiation events, we focused on the two of the top hits attributed to ERK signalling, gmnn and 3830406c13rik, and also asked whether the appearance of markers of the three embryonic cell lineages was affected. First we determined whether pluripotent cells remained in the population by alkaline phosphatase staining. Increased numbers of alkaline phosphatase stained cells were identified 5 days after “2i” withdrawal upon depletion of either gene, confirming their importance for escape from the pluripotent ground state (Figure 3D). Depletion of gmnn caused reductions in the expression of all three lineage markers at both 3 and 5 days following “2i” withdrawal, consistent with a general role in regulating the escape from the pluripotent ground state (Figure 3E). Similarly, depletion of 3830406c13rik, caused reduced expression of all three markers at day 3 (albeit only marginally for tbx6), and reduced levels of nestin after 5 days (Figure 3E). However, increased expression of gata4 and tbx6 was observed at this later timepoint, suggesting a lineage specific role for this gene. Thus, the contributions of individual genes identified in our screen towards individual lineage commitment are likely complex. In summary, the use of marker genes allows us to further validate the hits in our screen, although the effects of depleting individual genes on the loss of naive pluripotency and/or differentiation vary according to the gene involved.
Next, to further investigate the function of the hits identified in our screen, we investigated how this subset of genes impacted on ERK pathway regulation and function. In theory, genes might act to control ERK pathway activity or alternatively might mediate the effects of ERK pathway signaling. Therefore as a first step to partition genes as acting up or downstream of ERK, we used western blotting to monitor the active phosphorylated form of ERK (Figure 4A, Figure S10). Using this assay, siRNAs targeting 21 different genes were identified as upstream regulators of ERK. Importantly, none of the “GSK3 only” hits affected ERK activation, further validating our partitioning of the data (Figure 4A; Figure S10). Furthermore, while “ERK only” hits are partitioned evenly as acting up and downstream of ERK activation, the “ERK/GSK” hits are more prominent downstream of ERK (Figure 4B), as might be expected for genes which are important for GSK-mediated differentiation when ERK signaling is inhibited. To further delineate their point of action, we then tested the subset of siRNAs which acted upstream of ERK for their effects on Ras activation by an ELISA-based assay (Figure 4C). Eleven genes were identified whose point of action is upstream of both Ras and ERK (Figure 4C; Figure S11). Importantly, one of these genes was nras itself. These assays therefore enabled us to position genes from the “ERK only”, and “ERK/GSK” datasets at different points in the ERK pathway, either acting upstream of Ras eg plekh1 or on the core pathway downstream from Ras (Figure 4D). The rest of the genes analysed appear to act downstream from ERK and hence are likely mediators of ERK pathway function. Interestingly, transcription factors are over-represented in the subgroup of genes which act downstream of ERK (Figure 4E), in keeping with the known major role of ERK signalling in controlling gene expression programmes (reviewed in [26]–[27]). Together, these findings indicate that we have identified groups of genes which affect either signalling through the ERK pathway and/or the downstream consequences of ERK activation.
While in this study we have focussed on studying genes which affect escape from pluripotency, and are associated with the ERK and GSK3 pathways, it is likely that many of the genes we have identified might also play a more general role in controlling stem cell pluripotency. Indeed, several genes identified in our study were also identified previously in other siRNA screens conducted in cells maintained in the presence of serum and LIF rather than the “2i” conditions we used (Table S3). To investigate this further, we tested 8 genes for their role in escape from pluripotency in Rex1GFPd2 ES cells maintained in serum and LIF and induced to differentiate by withdrawal of LIF. Depletion of three of these genes, otx2, etv5 and mbd3, caused an increased retention of rex1 promoter-driven GFP expression, consistent with a disruption in escape from pluripotency (Figure S12). Thus, it is likely that many of the genes we have identified in this screen will play a more general role in controlling cell fate decisions in ES cells maintained in serum plus LIF or “2i” conditions.
A group of 10 genes was identified which acted downstream of Ras but affected ERK phosphorylation levels and hence ERK activity (Figure 4D). To further probe the point of action of these genes, we tested MEK activation levels following their depletion but saw little difference (data not shown). Next, we therefore focussed on MAP kinase phosphatases (also known as dual specificity phosphatases [DUSPs]), and hypothesised that increases in the levels and/or activity of these enzymes might be responsible for the reduced ERK activation that we observed and consequent effects on embryonic stem cell differentiation.
First we examined the set of genes we identified which accelerated differentiation in our primary siRNA screen for candidate dusp genes as we expected the loss of DUSPs would be predicted to enhance ERK phosphorylation and promote exit from pluripotency. Dusp1, dusp3 and dusp15 were amongst this category of genes (Table S2). We therefore determined the expression of these genes and a range of additional phosphatases in embryonic stem cells before and after “2i” removal. Amongst the genes tested, dusp1, dusp5 and dusp6 levels all increased following “2i” withdrawal while dusp14 levels were fairly constant (Figure 5A; Figure S13A). The increased expression of all these phosphatases was dependent on active ERK pathway signalling as expected from other cellular systems (reviewed in [39]) but in the case of dusp1 combinatorial inhibition of ERK and GSK signalling was required for maximal inhibition (Figure 5B; Figure S13B). However, at the protein level, Dusp1 levels gradually declined following “2i” withdrawal while Dusp6 levels increased in line with the increases in their mRNA levels (Figure S13C). Due to their dynamic expression, we focussed on Dusp1, Dusp5 and Dusp6 as these have the potential for controlling ERK pathway activity during embryonic stem cell differentiation. We therefore asked whether depletion of any of the genes identified in our screen would affect Dusp1, Dusp5 and Dusp6 expression at the mRNA or protein levels. Almost all the siRNAs tested (9/10) caused an increase in basal dusp1 mRNA levels and the same was observed on dusp6 levels for 4/10 genes (Figure 5C). In contrast, none of the siRNAs caused increases in dusp5 levels under these conditions (Figure S13D). Similarly, the levels of these dusps followed a similar pattern in response to siRNA treatment after release from “2i” for 40 mins (Figure S13E). Importantly, increases in Dusp1 and Dusp6 at the protein level were also observed which generally correlated with the effects of these siRNAs on mRNA levels (Figure S13F) although there were exceptions typified by Rab24 whose depletion does not affect dusp1 mRNA levels but instead appears to act post-transcriptionally to cause increased levels of Dusp1 protein.
An increase in the basal levels of MAP kinase phosphatases would likely lead to changes in the ERK activation kinetics, leading to the decreases in phosphorylated ERK levels we observed previously (Figure 4A). Indeed all of the siRNAs tested which promote increases in Dusp levels also cause a delay in peak activation of ERK and a subsequent reduction in the magnitude of this activation (Figure 5D; Figure S13G). Importantly other control siRNAs do not elicit this effect (Figure S14A). This suggests a causative link between the genes we identified in our screen, their effects on dusp gene expression and subsequent changes in ERK activity and downstream differentiation. Two key predictions of this model are that reductions in Dusp levels should first increase the rate and level of ERK pathway activation, and secondly, promote differentiation of embryonic stem cells. Indeed, depletion of Dusp6 and Dusp1 levels caused premature and higher amplitude activation of ERK whereas depletion of Dusp3 and a range of other Dusps had little effect on ERK activity levels (Figure 5E; Figure S14A). Importantly, while depletion of dusp1 and dusp6 caused increased levels of ERK activation, no increases could be detected on the low levels of Jnk and p38 phosphorylation, demonstrating a specific effect on the ERK pathway (Figure S14B). In our primary siRNA screen, we found that dusp1 depletion enhanced the loss of rex1 promoter-driven GFP expression (Figure 5F). We therefore depleted other Dusps to examine whether might function in an analogous manner and found that amongst these, only reductions in dusp6 levels triggered more efficient inactivation of the rex1-GFP reporter gene (Figure 5G). Similarly, dusp1 and dusp6 depletion caused increased loss of mRNA expression of the pluripotency marker nanog whereas dusp3 depletion had little effect (Figure 5H). Thus Dusp1 and Dusp6 appear to play an important role in maintaining pluripotency.
We extended this analysis to examine whether depletion of dusp1 or dusp6 affected lineage commitment by examining the expression of different marker genes 5 days after “2i” withdrawal. The depletion of dusp1 caused increased expression of all three lineage markers, consistent with a general role for this gene in inhibiting loss of pluripotency (Figure 5I). In contrast, depletion of dusp6 only caused increased levels of the ectoderm marker nestin, suggesting a more specific role in controlling differentiation into this lineage (Figure 5I).
Together, these results therefore demonstrate that our RNAi screen has enabled us to identify an important role for a subset of MAP kinase phosphatases in determining the rate and efficiency of ERK pathway activation in embryonic stem cells, and hence influence their ability to escape from pluripotency and begin to differentiate.
The derivation of pluripotent iPS cells and the controlled differentiation of embryonic stem cells into defined cell fates are two of the most important areas of research in the area of regenerative medicine. Numerous studies have helped build up a view of the complex signaling and transcriptional networks involved in maintaining the pluripotent state of embryonic stem cells (reviewed in [2]–[3]) but in contrast, much less is known about the pathways leading to the loss of pluripotency. Here we have conducted a genome-wide siRNA screen and identified over 400 genes which play a role in the onset of differentiation which allows ES cells to initiate escape from pluripotency. The vast majority of these genes have not previously been implicated in this process. This dataset therefore provides an important resource for the community and is a rich source of information for further investigating this phenomenon and also for a more basic understanding of the mechanisms governing the regulation and action of the core ERK and GSK3 signaling pathways.
Due to the controlled conditions used in our screen, we were able to link the genes which we identified to either the ERK and/or the GSK3 pathways as potential regulators or mediators of pathway functions. Importantly, it appears likely that many genes we have identified might also be important in the context of different culture conditions such as the commonly used serum and LIF-containing media (see Figure S12). However, further analysis on a case by case basis is required to substantiate a role for individual genes under these conditions. It is important to emphasise that ES cells grown in LIF and “2i” conditions exhibit very different epigenetic landscapes, so only a partial overlap in regulatory factors is expected when comparing these conditions [40]. Indeed, this is not unexpected considering that RNAi screens, including our own, commonly identify chromatin and transcriptional regulators as major important functionally enriched categories (see Figure 2E). Here we focused on genetic interactions with the ERK pathway, and we were able to place a large number of genes as acting upstream or downstream from ERK (Figure 4). Further subpartitioning of the dataset enabled us to identify genes which functioned upstream of Ras or between Ras and ERK (Figure 4D; Table S6). A surprising finding was that all of the genes which acted downstream of Ras, controlled ERK activation levels through controlling the levels of the MAP kinase phosphatases Dusp1 and/or Dusp6. The major point of control was at the transcriptional level. MAP kinase phosphatases are known regulators of MAP kinases activity in different cellular contexts, and Dusp6 in particular operates as part of a feedback loop in response to ERK activation (reviewed in [39]). While Dusp6 is able to specifically dephosphorylate and inactivate ERK in vitro, Dusp1 can also target the stress activated MAP kinases, JNK and p38 (reviewed in [39]). However, we saw no evidence for elevated levels of phosphorylated Jnk and p38 in mouse embryonic stem cells upon depletion of Dusp1, indicating its effects are likely via ERK. Fluctuations in both Dusp1 and Dusp6 levels occur upon ERK pathway activation in ES cells, suggesting that they play an important feedback regulatory role in this system. It appears likely that the combined amounts of these phosphatases helps set the threshold for ERK activation and hence ERK-mediated loss of pluripotency (Figure 5J). Indeed, tampering with this threshold control switch, either by depleting genes that control Dusp levels, or by directly depleting dusp1 or dusp6, alters this threshold and changes the activation kinetics of the ERK pathway. This in turn accelerates the loss of pluripotency and increases the expression of lineage-specific markers, indicating that Dusps help control the equilibrium between pluripotency and differentiation by maintaining the correct levels of ERK activity. Our demonstration of a key role for Dusps in early ES cell differentiation, adds to the literature demonstrating the role of these enzymes in controlling developmental processes (reviewed in [41]) and illustrates the importance of establishing signaling thresholds by balancing activating and inactivating mechanisms which converge on ERK pathway signaling. Indeed, a recent study demonstrated a role for a different phosphatase, Dusp9, in maintaining pluripotency in mouse embryonic stem cells maintained in the presence of LIF and BMP4 [42]. In this study, BMP4 was implicated in upregulating dusp9 expression through Smad pathway activation and hence leading to a dampening down of ERK activity. Importantly, they demonstrated that Dusp9 was not relevant to ERK control in stem cells maintained in 2i conditions, and rather as we have demonstrated, Dusp1 and Dusp6 are more important under these conditions. Reciprocally, we have shown that depletion of either dusp1 or dusp6 does not affect escape from pluripotency in ES cells released from maintenance in serum plus LIF conditions (Figure S12). Together these studies emphasise the critical importance of Dusps in controlling ERK signaling levels in stem cells to regulate the decisions about escape from pluripotency. Importantly, two of the top hits we identified in our screen, gmnn and 3830406c13rik, which act to control Dusp levels and hence ERK activation kinetics, are not only involved in the loss of pluripotency but also in the appearance of differentiation markers for all three lineages (Figure 3). At this stage, it is unclear how these proteins impact on ERK pathway regulation at the molecular level but it points to a pivotal role of these proteins in controlling this key cellular fate decision.
In addition to regulating ERK activation, it is clear that many of the genes identified in our screen contribute to other molecular and biological processes. For example, there are a large numbers of genes encoding transcription and chromatin regulators identified (Figure 2E; Figure S6). This is not unexpected as cells must make wholesale changes in their gene expression programmes as they lose pluripotency and begin to differentiate (reviewed in [43]). There is also enrichment in our screen of functional categories of genes associated with core cellular metabolism and cell cycle control, which presumably reflects the changing anabolic, catabolic and proliferative requirements of the cells as they receive altered signaling input which might contribute to their change in identity (Figure 2E). In addition to enrichments of specific functional categories of genes, many of the genes show strong interconnectivities, implying that we have also uncovered functionally interdependent networks of genes which are important in specifying stem cell fate. This is particularly apparent amongst cell cycle regulators, transcription factors and chromatin modifiers where functionally distinct subnetworks can be observed but also clear interactions between the different subnetworks are apparent. Future studies are required to probe the functional relevance of the networks we have uncovered.
One of the future challenges will be to connect the genes identified in our screen with the ERK and GSK3 signalling pathways. We have begun to do this by focusing on a subset of genes associated with the ERK pathway. However, even though we have implicated many genes in controlling ERK activity, the only information we have for 35 of these genes, is that their point of action is downstream from ERK activation. ERK signaling might be needed to activate their expression (either directly or indirectly) or alternatively the genes might encode proteins which are directly phosphorylated by ERK. For example, it is known that transcriptional regulators such as Ets1, Jun and FoxO1 can all be phosphorylated by ERK in other situations [44]–[46]. More complicated mechanisms can also be envisaged where, for example, ERK and/or GSK3 signalling might converge on the activation of a key target gene, in parallel to one of the other regulators identified in this screen. Additional methodologies will need to be applied to help provide these links.
Another key issue to address is whether the ERK and GSK3 pathways work together or in parallel manner, to target different substrates and ultimately control different gene expression programmes and biological functions in ES cells. The two pathway inhibitors have both distinct and overlapping affects on ES cells (reviewed in [11]). Consistent with this, our study suggests that there may well be specific biological functions associated with GSK3 and ERK pathway signaling as different GO terms are enriched in hits from our screen. However, for the most part, the GO terms are often shared by genes associated with both pathways (see Figure 2 and Figure S6), suggesting that there might be a high degree of cooperativity. Indeed, it is well established that ERK-dependent phosphorylation often acts as a priming event for GSK3-mediated phosphorylation of substrates as exemplified by Smad1 [47]. Thus it appears likely that the pathways might act more generally in a combinatorial manner, either at the level of phosphorylation of common substrates or through convergence in activating gene expression through targeting distinct regulatory factors.
In summary, this study has identified an important role for the precise modulation of ERK MAP kinase signaling levels in the ability of a cell to exit the pluripotent ground state. Furthermore, we have identified a large number of genes that potentially impact on the function of the ERK pathway and GSK3 function in embryonic stem cells. It is becoming increasingly obvious that modulating these pathways has a potential impact on the reprogramming of somatic cells to the iPS cell state (reviewed in [3]) and reciprocally in promoting the differentiation of ES and iPS cells down defined lineages. Thus, the resource we have generated has paved the way for designing alternative strategies to either promote pluripotency or the subsequent generation of new cell identities for therapeutic purposes.
ES cells were generally maintained in NDiff N2B27 media (Stem Cells, Inc.; scs-sf-nb-02) in the presence of the GSK3 inhibitor CHIR99021 (Stemgent, 04-0004; 3 µM) and MEK inhibitor PD0325901 (Stemgent, 04-0006; 1 µM) (“+2i” media) and were routinely passaged using Accutase (Sigma, A6964) every other day. For differentiation, the media containing inhibitor was removed and replaced with NDiff N2B27 media. Where indicated, ES cells were maintained in serum/LIF conditions in media containing knockout DMEM (Invitrogen 10829-018), 15% heat inactivated FBS (Invitrogen 10082-147), 2 mM of Glutamax-1 supplement (Invitrogen 35050-038), 1% non-essential amino acids (Invitrogen 11140-035), 50 µM 2-mercaptoethanol (Invitrogen 31350-010) and 5×105 U of LIF (Millipore 103 U/ml). The ES cells cultured under “+2i” conditions were adapted in serum/LIF culture conditions for at least 8 passages before the experiments were performed. Cells were stained for alkaline phosphatase expression using an alkaline phosphatase detection kit as described by the manufacturer's (Millipore).
For RNAi, 4×104/cm2 cells (ie 1.28×104 cells/well of 96 well plate) were plated out into a mixture of 0.3 µl of RNAi Max (Invitrogen) and 100 nM siRNA in 100 µl of “+2i” media for 24 hrs. All validation experiments used ON-TARGETplus siRNA SMART pools from Dharmacon.
Real time RT-qPCR was carried out as described previously [48]. For assays in 96 well plate format, the same basic protocol was followed except the RNA was obtained using a Fastlane cell RT-PCR kit (QIAGEN). Data were normalized for the average expression of the control genes gapdh, hmbs and tbp. The primer-pairs used for RT-PCR experiments are listed in Table S7.
Western blotting was carried out with the primary antibodies; Erk2 (137F5; Cell Signalling, 4695), phospho-ERK (E10; Cell Signalling, 9106), Dusp1 (MKP-1; Upstate, 07535), Dusp6 (MKP-3; Epitomics, 2138-1) and Pou5f1 (Oct-3/4; Santa Cruz, sc-8628). All experiments were carried out in 96-well plates. The lysates were directly harvested in the 2×SDS sample buffer followed by sonication (Bioruptor, Diagenode). The proteins were detected using infrared dye-conjugated secondary antibodies (LI-COR Bioscience, IRDye 800CW [1 in 10,000] and IRDye 680LT [1 in 20,000]), and the signal was collected with a LI-COR Odyssey Infrared Imager and quantified using Odyssey software (LI-COR Bioscience, Odyssey Infrared Imaging system application software version 3.0.25).
The Ras activities were examined using Ras activation ELISA assay kit (Millipore) as described in the manufacturers' instructions. The total lysates used in the ELISA assay was normalised with the quantity of the proteins assayed by the BCA protein assay kit (ThermoScientific).
Flow cytometric analysis was carried out using a LSRII flow cytometer and samples were loaded using HTS loader (BD Biosciences). For sampling, media was removed from each well. Single cell suspensions were generated by treating cells with accutase at 37°C for 7 mins followed by resuspendion in 0.03% BSA/PBS. Dead cells were stained by Sytox Red dead cell stain (Invitrogen, 5 nM). The cells were analysed immediately after sampling. Each sample was analysed with 10,000 event counts with the flow rate at 1 µl/s. The resulting GFP profile (green channel) was created by gating with the right ranges of cell sizes based on forward and reverse scatter plot (ssc vs fsc; blue channel) and dead cells were gated away based on the Sytox Red stain profile (red [APC] channel).
All liquid handling processes were performed using Biomek robotic system (Beckman Coulter). For the primary screen, Rex1GFPd2 ES were grown in 96 well plates in the presence of “2i” and reverse transfection was performed using siGENOME siRNA pools (Dharmacon; mouse protein kinase [G-013500], GPCR [G-013600], druggable [G-014600] and genome [G-015000] libraries). 24 hrs later, the “2i” media was removed and replaced with fresh NDiff N2 B27 media. After 28 hrs, the levels of GFP in the cells were determined by flow cytometry as described above. Each plate contained 8 control non-targeting siRNAs, and the positive control siRNAs against gsk3β and fgf4. To take into account slight variations in the timing of pluripotency loss, the ratio of high GFP to low GFP expressing cells was established on each plate based on the non-targeting controls, allowing a threshold to be set as 1 (ie 50% high GFP and 50% low GFP). This threshold was used to determine the ratio of high to low GFP expressing cells in the other wells. The mean plus/minus standard deviation (SD) was calculated for each plate, and individual wells were scored positive if they exceeded 2×SD above or below this mean. The screen was performed in duplicate, with duplicate plates being analysed on different days. A final list of positive hits was determined by taking siRNAs which scored an average of 2×SD across both plates (or on a single plate where the duplicate well was defective in the case of 30 siRNAs), generally with both plates scoring >1.5×SD above the mean. However, an additional small number of siRNAs were scored as positive where the average score was >2.5×SD above the mean where only one plate had to score >1.5×SD above the mean, and also for seven siRNAs where the average score was >1.9×SD above the mean and both plates scored >1.9×SD above the mean.
For the validation screens, either Rex1GFPd2 or Oct4GFP ES cells were used and screens were performed as above except that ON-TARGETplus siRNA duplexes were used and GFP levels in Oct4GFP ES cells were determined 72 hrs after release from “2i”. Individual wells in each screen were scored as positive if the average GFP(+)/GFP(−) ratio exceeded 1.25×SD above the non-targeting controls across both duplicate plates. Additional hits were considered as positive if they scored >0.8×SD above the mean in one validation screen and also scored >1.0×SD above the mean in the other.
For the “1i” screens, Rex1GFPd2 were used as for the validation screens but only one inhibitor (ie either CHIR99021 or PD0325901) was withdrawn. Wells were scored as positive if the average GFP(+)/GFP(−) ratio exceeded 1.5×SD above the mean of the non-targeting controls.
For constructing networks, lists of gene names were uploaded into STRING [49] with the confidence score set high (0.40). The resulting networks were saved as *.txt files and then uploaded into Cytoscape (v. 2.7.0) choosing coexpression, textmining, knowledge and experimental data as proximity criteria. yFiles→organic network layouts were applied and the positioning and graphic representation of nodes were adjusted manually for increased clarity.
GO term analysis was carried out using DAVID Bioinformatics Resources 6.7 (NIH) [50]. The enriched terms from the functional annotation chart were extracted and manually clustered. Heat maps of GO terms were generated by MultiexperimentViewer (MeV 4_7_4). GO term summary and visualization was carried out by REVIGO [51].
|
10.1371/journal.pcbi.1002847 | Calcium Wave Propagation in Networks of Endothelial Cells: Model-based Theoretical and Experimental Study | In this paper, we present a combined theoretical and experimental study of the propagation of calcium signals in multicellular structures composed of human endothelial cells. We consider multicellular structures composed of a single chain of cells as well as a chain of cells with a side branch, namely a “T” structure. In the experiments, we investigate the result of applying mechano-stimulation to induce signaling in the form of calcium waves along the chain and the effect of single and dual stimulation of the multicellular structure. The experimental results provide evidence of an effect of architecture on the propagation of calcium waves. Simulations based on a model of calcium-induced calcium release and cell-to-cell diffusion through gap junctions shows that the propagation of calcium waves is dependent upon the competition between intracellular calcium regulation and architecture-dependent intercellular diffusion.
| Calcium wave signal has been found in a wide variety of cell types. Over the last years, a large number of calcium experiments have shown that calcium signal is not only an intracellular regulator but is also able to be transmitted to surrounding cells as intercellular signal. This paper focuses on the development of an approach with complementary integration of theoretical and experimental methods for studying the multi-level interactions in multicellular architectures and their effect on collective cell dynamic behavior. We describe new types of higher-order (across structure) behaviors arising from lower-order (within cells) phenomena, and make predictions concerning the mechanisms underlying the dynamics of multicellular biological systems. The theoretical approach describes numerically the dynamics of non-linear behavior of calcium-based signaling in model networks of cells. Microengineered, geometrically constrained networks of human umbilical vein endothelial cells (HUVEC) serve as platforms to arbitrate the theoretical predictions in terms of the effect of network topology on the spatiotemporal characteristics of emerging calcium signals.
| Multi-level organization and dynamics is a hallmark of most biological systems. This is particularly true in tissues in which single cells are organized into multicellular structures, which are further assembled into complex tissue and organs. For example, endothelial cells are assembled into multicellular tubes (i.e. vessels) which are connected to each other to form a branched vascular tree system. Molecular signals are initiated and/or processed at the endothelial cell level yet influence overall tree behavior and vice-versa [1]. Central to the proper behavior in these biological systems is cross-level interdependence. To date, limited studies of signaling in multicellular networks have demonstrated that the architecture of multi-cellular systems have a significant impact on the behavior of individual cells as well as their emerging collective behavior.
Over the past decade, questions concerning the system behavior of cellular structures have received increasing attention. For instance, there is strong evidence that the branching architecture of the mammary gland is a major regulator of normal epithelial cell signaling and function [2], [3]. Normal organ architecture can suppress tumor formation and prevent malignant phenotypes even in grossly abnormal cells [4]. Tissue engineering in its attempt to construct functional tissues faces the challenge of arranging cells (e.g. scaffolding via decellularization of allograph tissue) in a three-dimensional configuration with architecture analogous to the native tissue to support proper spatial and temporal molecular signaling necessary to sustain appropriate development and function [5]. Also, downstream and upstream signal conduction between endothelial cells along the walls of vessels plays an important role in microcirculatory function, vascular network remodeling, vasculogenesis, and neovascularization [6].
A particularly relevant aspect to tissue engineering is the emerging behavior of a multicellular architecture in which cell-level functions, such as intracellular communication, integrate with multicellular architectures through local cell-to-cell interactions. Central to this problem is that cellular networks inherently combine dynamical and structural complexity. Early progress on modeling coupled dynamical systems was limited to space-independent coupling or regular network topologies. Further progress to circumvent the difficulty of modeling associated with the combined complexity of the dynamics and of the architecture was achieved by taking a complementary approach where the dynamics of the network nodes is set aside and the emphasis is placed on the complexity of the network architecture [7]. Accordingly, linear solutions of calcium reaction/diffusion models of multicellular architectures composed of networks of chains of cells with grafted side branches have shown that calcium wave propagation differs in ordered or disordered architectures [8], [9]. Similar effects have also been encountered in chains of endothelial cells with non-linear intracellular calcium dynamics [10].
To evaluate the effects of multilevel architectures on biological signal behavior, we modeled calcium-signal propagation in networks of endothelial cells experimentally and computationally. The vasculature is an ideal system for evaluating multi-scale behavior given the relatively simple but multi-ordered organization of the cells and tissues. Here, the behavior of a calcium wave moving along branched chains of endothelial cells was simulated using experimentally observed parameters in the computation. While there are numerous stimuli that can initiate calcium waves in endothelial cells, we utilized the mechanical stimulation of a single endothelial cell as the wave initiator to minimize confounding issues related to multiple upstream and downstream effects intrinsic to diffusible (i.e. pharmacological) signals. Furthermore, mechanical forces play important roles in endothelial function in vivo [11]. The theoretical aspect leverages progress in modeling of the dynamics of complex networks and in microengineering of multicellular structures to generate new knowledge concerning multicellular architectures.
Our study is based on networks of human umbilical vein endothelial cells (HUVEC) (ATCC CRL-1730) in which intercellular calcium wave propagation is primarily dominated by gap junction [12].
Since we are interested in the behavior of networks of endothelial cells composed of one-dimensional chains of cells and networks of chains of cells, a reaction/diffusion model is developed to gain insight into the architecture-dependence of calcium wave propagation. For the sake of simplicity, we only consider the dynamic of intracellular calcium and assume the intercellular Ca2+ is transported between cells by diffusion through gap junctions.
We investigate the behavior of several types of multicellular structures, namely single chains of endothelial cells and “T” structures subjected to either single or simultaneous double mechano-stimulation at different locations in the structures.
In this section we consider the behavior of a finite chain of endothelial cells among which a single cell in the chain is subjected to mechanical stimulation to initiate a calcium impulse, due to the intracellular increase in calcium concentration.
We now consider the behavior of a chain of cells subjected to dual mechano-stimulation. The stimulations are applied simultaneously on two cells separated by a short distance. In light of an average distance of propagation of a calcium pulse of approximately 4.7 cells, this distance is chosen so that one could expect possible overlap of the signals emanating from the two stimulated cells in the region separating them.
The growth of “T” structures formed by surface-patterning perpendicular single chains of cells does not permit the formation of cellular junctions composed of a single cell. Typically, many cells aggregate at the junction of the three branches forming a cell cluster (see Figure 8 and Figure 9).
In section “Single Chain-Dual Stimulation: Experiments” we have demonstrated that two calcium waves cannot cross when propagating toward each other in a chain of endothelial cells. We consider, here, the dual-stimulation of a “T” structure with stimulations located in two separate branches. We address the question of the interaction of the two calcium pulses in the junction area.
In this paper, we study experimentally the propagation of calcium waves in different multicellular structures composed of human umbilical vein endothelial cells (HUVEC). The fabrication of cell-chain based multicellular chain structures relies on organizing multiple cells into specific configurations via selective plasma surface functionalization, which guides cellular attachment. Calcium waves are actuated via mechano-stimulation of selected cells. Calcium wave propagation is characterized by time-resolved fluorescence microscopy. The experimental observations are complemented by modeling and simulation of calcium wave propagation using a diffusion/reaction model. The model of intracellular calcium dynamics is non-linear and mimics the IP3-induced calcium release and calcium induced calcium release (CICR). In order to capture the essence of cross-level interactions in calcium signal propagation in multicellular architectures, we only consider a single component model of CICR. This model is different from previous CICR models, which consisted of multiple coupled non-linear differential equations describing the kinetics of IP3/Ca2+ pumping, release and activation [25], [26]. Nevertheless, the model is capable of capturing most essential features of calcium wave propagation in HUVEC observed in the experiment.
Cell-to-cell interactions are described in this paper via intercellular diffusion through gap junctions. Experimental observation of calcium waves induced by a single mechano-stimulation and propagating along a chain of endothelial cells is used to calibrate the model. Experiments and simulations of chains of cells subjected to dual stimulation (i.e. simultaneous stimulation of two different cells) show that two calcium waves cannot cross each other due to the refractory stage of endothelial cells. The study of more complex multicellular structures utilized “T” structures, which are composed of three side branches joining at a junction. The junction is comprised of cell clusters. In this case, we observe experimentally that when a single cell in one of the side braches is stimulated, the calcium signal does not propagate beyond the junction area. However, when two mechano-stimulations are simultaneously applied on separate branches the calcium signal can propagate through the junction area and beyond well into the third unstimulated side branch of the “T” structure. A computational model of a “T” structure, which includes a cell cluster at the junction, shows the importance of intracellular calcium dynamics and intercellular diffusion in determining the propagation behavior of calcium waves. In particular, the organization of cells in the junction determines the existence of multiple paths for intercellular diffusion, which may affect the accumulation of cytosolic calcium and subsequently the ability of cells to undergo CICR.
In summary, this work demonstrates that the propagation of calcium waves is dependent upon the architecture of multicellular structures. This dependence is due to the competition between intracellular calcium reaction and diffusion, which is affected by the topology through cell connectivity via gap junctions.
|
10.1371/journal.pbio.1000521 | WRAP53 Is Essential for Cajal Body Formation and for Targeting the Survival of Motor Neuron Complex to Cajal Bodies | The WRAP53 gene gives rise to a p53 antisense transcript that regulates p53. This gene also encodes a protein that directs small Cajal body–specific RNAs to Cajal bodies. Cajal bodies are nuclear organelles involved in diverse functions such as processing ribonucleoproteins important for splicing. Here we identify the WRAP53 protein as an essential factor for Cajal body maintenance and for directing the survival of motor neuron (SMN) complex to Cajal bodies. By RNA interference and immunofluorescence we show that Cajal bodies collapse without WRAP53 and that new Cajal bodies cannot be formed. By immunoprecipitation we find that WRAP53 associates with the Cajal body marker coilin, the splicing regulatory protein SMN, and the nuclear import receptor importinβ, and that WRAP53 is essential for complex formation between SMN–coilin and SMN–importinβ. Furthermore, depletion of WRAP53 leads to accumulation of SMN in the cytoplasm and prevents the SMN complex from reaching Cajal bodies. Thus, WRAP53 mediates the interaction between SMN and associated proteins, which is important for nuclear targeting of SMN and the subsequent localization of the SMN complex to Cajal bodies. Moreover, we detect reduced WRAP53–SMN binding in patients with spinal muscular atrophy, which is the leading genetic cause of infant mortality worldwide, caused by mutations in SMN1. This suggests that loss of WRAP53-mediated SMN trafficking contributes to spinal muscular atrophy.
| Cajal bodies, discovered more than 100 years ago by Santiago Ramón y Cajal, are sub-organelles found in the nucleus of proliferative cells and neurons. They have been implicated in a variety of nuclear functions including ribonucleoprotein maturation, spliceosome formation, histone mRNA processing, RNA polymerase assembly, telomerase biogenesis, and histone gene transcription. Concentrating relevant molecules within Cajal bodies may serve to increase the efficiency of specific nuclear functions. Here we identify the WRAP53 protein as an essential factor for Cajal body maintenance and for directing the splicing regulatory protein “survival of motor neuron” (SMN) complex to Cajal bodies. We show that WRAP53 is a constitutive component of Cajal bodies, and that knockdown of WRAP53 disrupts existing Cajal bodies and prevents formation of new Cajal bodies. Mechanistically, we find that WRAP53 recruits the SMN complex from the cytoplasm to Cajal bodies by mediating interactions between SMN, importinβ, and coilin. Finally, we report deficient WRAP53–SMN binding in patients with spinal muscular atrophy, suggesting a role in this pathology. This study not only reveals new functions of the WRAP53 protein, but also increases our understanding of the molecular mechanism behind Cajal body formation and recruitment of factors to Cajal bodies.
| We previously discovered WRAP53 as an antisense gene to the p53 tumor suppressor gene [1]. WRAP53 gives rise to a regulatory antisense transcript with a critical role for p53 function [1] and was recently approved as the official name of this gene (for “WD40 encoding RNA antisense to p53”; also denoted TCAB1 or WDR79). This gene also encodes a protein that directs small Cajal body–specific RNAs (scaRNAs), including the telomerase RNA, to Cajal bodies [2],[3]. Cajal bodies are nuclear organelles containing factors involved in ribonucleoprotein (RNP) maturation, spliceosome formation, histone mRNA processing, RNA polymerase assembly, telomerase biogenesis, and histone gene transcription [4]–[6]. The Cajal body was discovered more than 100 years ago by Santiago Ramón y Cajal, as a spherical structure often located in close proximity to the nucleolus (formerly called “nucleolar accessory body” or “coiled body”). Cajal bodies are dynamic structures that move within the nucleoplasm, move to and from nucleoli, join each other to form larger structures, and separate from larger into smaller bodies [7]. Nuclei contain 0–10 Cajal bodies, depending on cell cycle stage and cell type. Although Cajal bodies per se are not essential for cell survival, defects in Cajal body formation have been linked to impaired cell proliferation and splicing rates [8]–[10]. The reason why cells survive without Cajal bodies even though many processes in this organelle are essential for survival is probably that these processes can also occur in the nucleoplasm in the absence of Cajal bodies [11]. Thus, collecting enzymes and substrates in Cajal bodies may rather be a way to increase the efficiency of these processes by concentrating all factors at one site.
Cajal bodies are molecularly defined by the presence of the marker protein coilin. Coilin is essential for Cajal body integrity and function, and loss of coilin disrupts Cajal bodies. It has been proposed that coilin, upon oligomerization, provides a scaffold for the assembly of the different types of Cajal body components [12],[13] and that interaction with coilin mediates recruitment of proteins to Cajal bodies [14]. Formation of Cajal bodies also depends on spliceosomal small nuclear RNPs (snRNPs) that are rate-limiting factors for the assembly of additional Cajal bodies [10],[15]. Proteins involved in snRNP biogenesis, such as the survival of motor neuron (SMN) protein, are also important but not essential for Cajal body structure [10].
The SMN protein is part of a large complex essential for the assembly of snRNPs in the cytoplasm [16]. The SMN complex enables nuclear import of the snRNPs by binding to the nuclear import receptor importinβ [17],[18] and further transports the snRNPs to Cajal bodies for additional modification and maturation. Interaction between SMN and importinβ is required for SMN nuclear import, while SMN–coilin interaction is believed to mediate SMN complex localization to Cajal bodies [14]. Reduced levels of SMN due to mutations or deletions of the SMN1 gene cause the common neurodegenerative disorder spinal muscular atrophy (SMA), the leading genetic cause of infant mortality worldwide, which affects approximately one in 6,000 infants. A second copy of the SMN1 gene, SMN2, partially compensates for SMN1 loss. However, because of a single nucleotide change, most SMN2 transcripts lack exon 7, resulting in the production of the C-terminally truncated and unstable protein SMNΔC15 [19],[20]. The reason why the motor neurons in the spinal cord are selectively degenerated in SMN deficiency is still unknown. The clinical severity of this disease is correlated with low copy number of SMN2 and reduced number of nuclear structures containing the SMN protein (encoded by SMN2) [21]–[23]. The latter suggests that targeting defects of SMN to nuclear structures contribute to SMA type I. In the present study, we have identified and characterized WRAP53 as a new critical player in Cajal body formation and for recruiting the SMN complex to Cajal bodies by mediating interactions between SMN, importinβ, and coilin. Moreover, WRAP53 and SMN association is disrupted in SMA patients, suggesting a role of WRAP53 in SMA pathogenesis.
The WRAP53 protein has been found highly enriched in nuclear Cajal bodies in HeLa cells [2],[3]. To further examine the presence of WRAP53 in Cajal bodies, a panel of cancer cell lines and primary cells including U2OS, H1299, HCT116, HEK293, MCF-7, HeLa-PV, and HDF were stained using a polyclonal antibody against WRAP53 and a monoclonal antibody against the Cajal body marker coilin. WRAP53 localized to Cajal bodies in all cell types analyzed (Figure 1A). Importantly, complete overlap between WRAP53 and coilin was observed in 100% of Cajal bodies in all cells (n>300), clearly indicating that WRAP53 is a constitutive component of Cajal bodies (Figure 1A).
To investigate whether WRAP53 plays a role in the formation or maintenance of Cajal bodies, WRAP53 was depleted in U2OS and HeLa cells, and the effects on Cajal bodies, i.e., coilin, was analyzed by immunoflourescence (IF) microscopy and Western blotting (WB). Two different small interfering RNA (siRNA) oligos targeting WRAP53 were used (siWRAP53#1 and siWRAP53#2), both knocking down WRAP53 mRNA with 90% efficiency (Figure S1A and S1B). In control cells, treated with a scramble siRNA with no homology to any gene (siControl), coilin displayed the characteristic Cajal body localization and co-localized with WRAP53 in all Cajal bodies (Figures 1B and S1C). A weak staining of coilin was also seen in nucleoli, consistent with previous findings that coilin transits through the nucleolus during the normal life cycle of the protein [13]. Strikingly, no Cajal bodies were found in WRAP53-depleted cells (Figures 1B and S1C). Instead, coilin accumulated in the nucleoli. Other Cajal body proteins, such as SMN, also showed absence of Cajal body accumulation and increased nucleolar staining in WRAP53-depleted cells (Figures 1B and S1C). Staining with the nucleolar marker fibrillarin confirmed nucleolar accumulation of coilin and SMN upon WRAP53 depletion (Figure 1C). Thus, WRAP53 is required for Cajal body maintenance. WRAP53-depleted cells were also analyzed for changes in other nuclear structures, such as nucleoli (fibrillarin), gems (SMN), and promyelocytic leukemia (PML) bodies. No effects on these structures were observed (Figure S1D), demonstrating that WRAP53 is an essential component for Cajal bodies but is not essential for other nuclear structures.
Loss of Cajal bodies was strictly associated with the degree of WRAP53 knockdown, where complete knockdown of WRAP53 led to the disappearance of all Cajal bodies, and cells still expressing low levels of nuclear WRAP53 showed Cajal body staining (Figure S1C). We also knocked down coilin and SMN in U2OS and HeLa cells. Depletion of coilin resulted in the disappearance of all Cajal bodies, leaving WRAP53 and SMN dispersed throughout the nucleoplasm (Figure S1E). Depletion of SMN significantly reduced the number and size of Cajal bodies, but some cells still had Cajal bodies left. Both WRAP53 and coilin were present in the remaining Cajal bodies and accumulated in nucleoli (Figure S1F). Thus, WRAP53 and coilin are essential for Cajal body structure, whereas SMN is not.
We next examined the effects on Cajal bodies in cells overexpressing WRAP53. Flag-tagged WRAP53 expressed at lower levels showed Cajal body accumulation (Figure S2A). In contrast, Flag-WRAP53 expressed at higher levels gave rise to a different nuclear expression pattern, with a more even distribution throughout the nucleoplasm (Figures 1D and S2A). Interestingly, no Cajal bodies were detected in these cells, and coilin and SMN were, like WRAP53, distributed throughout the nucleoplasm. Similar phenomena were observed using enhanced green fluorescent protein (EGFP)–tagged WRAP53 (Figure S2B). WB analysis of WRAP53 knockdown and WRAP53-overexpressing cells showed no difference in coilin or SMN protein levels (Figures S1B and S2C), and immunostaining of WRAP53-overexpressing cells showed no change in other nuclear structures, including PML bodies (Figure S2D). Thus, aberrant overexpression of WRAP53 prevents Cajal body formation and causes significant mislocalization of the Cajal body proteins coilin and SMN to the nucleoplasm. This finding confirms the notion that WRAP53 is an essential component of Cajal body structure and that proper localization of WRAP53 is required for its role in Cajal body formation.
snRNPs are known to be rate-limiting for Cajal body formation [8],[24]. The SMN complex transports snRNPs into the nucleus, and overexpressing the SMN protein induces formation of additional Cajal bodies. In light of this knowledge, we examined the influence of WRAP53 on de novo formation of Cajal bodies. Flag-tagged SMN was transiently transfected into U2OS cells, which increased the number of Cajal bodies per cell from 2–3 in control cells up to ten in Flag-SMN cells (Figure 2A). All Cajal bodies were positive for WRAP53 and coilin (Figure 2A). Cytosolic accumulations of SMN were observed in Flag-SMN cells; however, neither coilin nor WRAP53 were present in these structures (Figure 2A). Interestingly, both knockdown and aberrant overexpression of WRAP53 repressed generation of Cajal bodies induced by SMN overexpression (Figure 2B and 2C). Instead, Flag-SMN mislocalized to the nucleoli in WRAP53-depleted cells and to the nucleoplasm in WRAP53-overexpressing cells. These results show that WRAP53 is required for formation of new Cajal bodies induced by SMN overexpression, which further supports the idea that WRAP53 is essential for Cajal body assembly.
The finding that exogenous WRAP53 alters the localization of endogenous WRAP53, SMN, and coilin suggests a dominant negative effect of overexpressed WRAP53 that could be caused by WRAP53 self-interaction. Previous reports demonstrate such phenomena for coilin, where overexpressed coilin mislocalizes to nucleoli and disrupts Cajal bodies through dominant negative interference between exogenous and endogenous coilin [13]. To investigate if WRAP53 self-associates, U2OS cells were co-transfected with Flag-WRAP53 and EGFP-WRAP53 constructs. Immunoprecipitation (IP) with anti–green fluorescent protein (GFP) or anti-Flag antibodies showed that Flag-WRAP53 protein co-precipitated EGFP-WRAP53 and vice versa (Figure 2D). This indicates that exogenous WRAP53 self-associates in vivo. Furthermore, IP of EGFP-WRAP53 in U2OS cells co-precipitated endogenous WRAP53, whereas IP of EGFP alone did not (Figure 2E). This suggests that overexpressed WRAP53 interacts with endogenous WRAP53 in vivo, which also strengthens our hypothesis that overexpressed EGFP-WRAP53 or Flag-WRAP53 can cause mislocalization of endogenous WRAP53 by self-association.
IP of endogenous WRAP53 furthermore revealed that WRAP53 associates with coilin and SMN (Figure 3A). Reciprocal IP of coilin and SMN verified the interactions with WRAP53. To assess which region of WRAP53 interacts with coilin and SMN, we generated and transiently overexpressed a series of EGFP-tagged WRAP53 deletion constructs in U2OS cells (Figure 3B). Each construct expressed a protein of the expected size, as demonstrated by immunoblotting using both GFP and WRAP53 antibodies (Figure 3C and data not shown). IP of EGFP-WRAP53 using GFP antibody showed that WRAP53 constructs containing the WD40 domain plus the C-terminal region containing amino acids (aa) 456–533 associated with coilin and SMN. Constructs lacking these two domains or only expressing one of them co-precipitated neither coilin nor SMN (Figure 3C). Hence, WRAP53 associates with both coilin and SMN, and the same sequence in WRAP53 is important for interaction with both these proteins.
To investigate which region of WRAP53 mediates its localization to Cajal bodies, the panel of EGFP-WRAP53 deletion constructs was transiently transfected into U2OS cells, and protein localization was analyzed by IF. The cells were also stained for coilin to visualize Cajal bodies. Interestingly, only the WRAP53 constructs that bind coilin and SMN (EGFP-WRAP53FL, EGFP-WRAP53ΔN149, and EGFP-WRAP53ΔC15) accumulated in Cajal bodies (Figure 4A–4C). In contrast, the constructs unable to bind coilin or SMN failed to localize to Cajal bodies (Figure 4D–4H). This suggests that interaction with coilin and/or SMN is necessary for WRAP53 localization to Cajal bodies. No change in Cajal body number was observed in cells overexpressing the different WRAP53 constructs (data not shown). These observations were made in cells expressing low to moderate levels of WRAP53. In cells with high WRAP53 expression, nuclear mislocalization of WRAP53, SMN, and coilin was observed. Most likely this is due to sequestering of coilin and SMN in the nucleoplasm by EGFP-WRAP53, since high expression of WRAP53 deletion mutants unable to bind coilin did not mislocalize coilin/SMN and had no effects on Cajal body appearance (data not shown). Thus, the WD40 domain and the C-terminal region of WRAP53 target WRAP53 to Cajal bodies.
We also observed that all WRAP53 constructs showed cytoplasmic localization (Figure 4A–4G) and that WRAP53 constructs lacking the C-terminal region (EGFP-WRAP53ΔC93 and EGFP-WRAP53ΔC15) demonstrated a more pronounced cytoplasmic staining (Figure 4C and 4D). This was most apparent with the EGFP-WRAP53ΔC93 construct. In contrast, N-terminally deleted constructs exhibited the opposite distribution, i.e., a stronger nuclear staining (Figure 4B, 4E, and 4F), which was most apparent with the EGFP-WRAP53ΔN149 construct. These results indicate that the C- and N-terminal regions of WRAP53 contain elements important for the subcellular distribution of WRAP53.
Previous studies failed to detect any WRAP53 protein in the cytoplasm [2],[3]. To investigate this further we performed IF staining of endogenous WRAP53 with three different WRAP53 antibodies. Interestingly, all three antibodies show cytoplasmic localization of WRAP53, in addition to accumulation in Cajal bodies, using both methanol and paraformaldehyde fixation (Figures 1A, S3B, and S3C). A clear reduction in cytoplasmic and nuclear WRAP53 staining was observed after WRAP53 depletion, confirming the specificity of the WRAP53 staining in both compartments (Figure 1B). We also performed cell fractionation followed by WB of the WRAP53 protein. This confirmed that WRAP53 is present both in cytoplasmic and nuclear fractions, and quantification of the blots revealed that 83%–88% of the WRAP53 protein is localized to the cytoplasm (Figure 4I). Thus, endogenous and exogenous WRAP53 (Flag-WRAP53 and EGFP-WRAP53) show a clear cytoplasmic localization. We conclude that WRAP53 localizes to the cytoplasm in addition to nuclear Cajal bodies.
To understand the organization of interaction between WRAP53, SMN, and coilin, we separately knocked down each of these proteins, and then performed IP analysis. This showed that WRAP53 co-precipitates coilin in SMN-depleted cells and co-precipitates SMN in coilin-depleted cells (Figure 5A). Thus, WRAP53 interacts with coilin independently of SMN and with SMN independently of coilin. In contrast, IP of coilin and SMN in WRAP53-depleted cells showed a significant reduction in coilin–SMN interaction (Figure 5B and 5C). Loss of coilin–SMN complex formation was also found in U2OS cells stably overexpressing Flag-WRAP53 at high levels (Figure 5D) and displaying nucleoplasmic mislocalization of WRAP53 and coilin (Figure S4A). These results demonstrate that proper expression of WRAP53 is required for coilin–SMN complex formation in vivo.
Previous studies have described a direct binding between SMN and coilin [14], suggesting that WRAP53 is not required for the actual interaction between these proteins but rather brings them in close proximity to allow their interaction. Since WRAP53 localizes to the cytoplasm, we hypothesized that WRAP53 recruits SMN from the cytoplasm to nucleus, thus enabling interaction between SMN and coilin. Using in situ proximity ligation assay (in situ PLA) [25], we found that WRAP53 and SMN associate both in the cytoplasm and in Cajal bodies (Figures 5E and S4B). In situ PLA is a sensitive method that detects endogenous protein–protein interactions (visualized as red dots) in fixed cells and allows identification of the precise subcellular localization of the interaction. Cell fractionation followed by IP of endogenous WRAP53 furthermore confirmed that WRAP53 associates with SMN both in the cytoplasm and in the nucleus (Figure 5F). Thus, WRAP53 and SMN interact both in the cytoplasm and nucleus.
We next analyzed whether knockdown of WRAP53 affects the cytoplasmic and nuclear distribution of SMN. Interestingly, a significant increase of SMN was found in the cytoplasm in WRAP53-depleted cells, which coincided with a decrease of SMN in the nucleus (Figure 6A). This was demonstrated by cell fractionation and WB, and quantification of the Western blots showed a 60% increase of SMN in the cytoplasm and 40%–50% decrease of SMN in the nucleus upon WRAP53 knockdown (Figure 6B). Knockdown of coilin did not affect the intracellular distribution of SMN (Figure S5A), indicating that the function of WRAP53 in Cajal body assembly is not underlying the changed SMN distribution. The subcellular distribution of WRAP53 was not affected by SMN depletion, indicating that WRAP53 controls SMN but not vice versa (Figure S5A). Thus, knockdown of WRAP53 results in cytoplasmic accumulation and nuclear decrease of SMN, supporting our hypothesis that WRAP53 is involved in the recruitment of SMN from the cytoplasm to the nucleus.
Interaction with the nuclear import receptor importinβ is required for SMN nuclear import [18]. Co-IPs of importinβ and WRAP53 showed that the two proteins interact (Figure 6C). To test whether WRAP53 influences interaction between SMN and importinβ, cells were depleted for WRAP53 and immunoprecipitates of importinβ were assayed for SMN. Importinβ efficiently associated with SMN in siControl-treated cells (Figure 6D). In contrast, a significantly smaller amount of SMN protein was co-precipitated with importinβ antibody in siWRAP53-treated cells (Figure 6D), showing that WRAP53 is important for SMN–importinβ association. Thus, knockdown of WRAP53 reduces importinβ and SMN binding, diminishes the nuclear localization of SMN, and causes SMN to accumulate in the cytoplasm. It appears as if WRAP53 recruits SMN from the cytoplasm to the nucleus by facilitating SMN–importinβ interaction and subsequently mediates interaction between SMN and coilin in the nucleus by bringing the proteins in close proximity. This may either occur within an already existing Cajal body or catalyze the formation of a new Cajal body (Figure 6E).
We next examined whether WRAP53 is required for directing other components of the SMN complex to Cajal bodies, and Gemin3 was chosen as a representative component. IP analysis showed that WRAP53 and Gemin3 indeed associate (Figure S5B). Depletion of SMN disrupted this interaction, showing that WRAP53 and Gemin3 interact through SMN (Figure S5B). Knockdown of WRAP53 did not affect SMN–Gemin3 binding (Figure S5C). Like SMN, Gemin3 also showed absence of Cajal body accumulation in WRAP53-depleted or -overexpressed cells, but were still localized in gems (). In WRAP53-depleted cells, Gemin3 localized to the nucleoplasm, to gems, and partially to nucleoli (Figure S5D and S5F). Altered localization of Gemin2 was also observed in WRAP53-depleted cells (Figure S5F). Like SMN, Gemin2 localized to nucleoli and gems upon WRAP53 knockdown. Taken together, WRAP53 is important for localizing the entire SMN complex to Cajal bodies but not to gems.
We also analyzed whether WRAP53 targets snRNPs to Cajal bodies. snRNPs consist of Sm proteins in complex with small nuclear RNAs and are carried by the SMN complex from the cytoplasm to Cajal bodies, where final maturation of the snRNPs takes place. In control cells the Sm proteins were enriched in Cajal bodies and distributed throughout the nucleus in speckles (Figure S6A). In WRAP53-depleted cells, no Sm accumulation in Cajal bodies was observed, but Sm was still present in speckles (Figure S6A). Similary, upon SMN knockdown, Sm did not localize to Cajal bodies but to speckles (data not shown). Thus, WRAP53 is required for Sm localization to Cajal bodies. Knockdown of WRAP53 did not affect SMN–Sm binding nor the cytoplasmic and nuclear distribution of Sm (Figures S5C and S6B), indicating that loss of Sm accumulation in Cajal bodies is not caused by snRNP assembly defects or altered intracellular distribution of Sm.
The most severe form SMA, SMA type I, correlates with a reduced number of SMN-containing nuclear bodies [21]–[23]. The role of WRAP53 in Cajal body formation and nuclear localization of SMN encouraged us to investigate the interplay of WRAP53 and SMN in vivo in SMA disease. We first analyzed SMN localization in nuclear bodies in fibroblasts derived from an unaffected mother (GM03814, serving as control) and her two children with SMA type I (GM03815 and GM03813). Co-staining of SMN and coilin showed that SMN accumulated in nuclear bodies (gems and Cajal bodies) in 67% of control fibroblasts (GM03814), compared to only 13% (GM03815) and 16% (GM03813) of SMA fibroblasts. Cajal bodies (coilin accumulation) were detected in 43% of control fibroblasts (GM03814), but in only 25% (GM03815) and 15% (GM03813) of the SMA fibroblasts. WRAP53 was present in all Cajal bodies. The absence of SMN in nuclear bodies coincides with lack of Cajal bodies in the same cells (Figure 7A). Thus, both gem and Cajal body number are decreased in SMA fibroblasts. This observation could not be explained by the difference in WRAP53 levels in SMA fibroblasts compared to control (Figure 7B). However, WRAP53 showed reduced binding to SMN protein in the SMA fibroblasts (Figure 7C). To investigate whether the lack of binding between WRAP53 and SMN was just a reflection of decreased levels of SMN in the SMA fibroblasts, or whether WRAP53 binding to the SMN protein derived from the SMN2 allele is in fact weaker, we quantified the relative amount of SMN that interacts with WRAP53 in SMA patients and in normal cells. This revealed a lack of binding between SMN and WRAP53 in cells from SMA patients that cannot be explained by the lower amounts of SMN, and that the relative binding between WRAP53 and SMN in these cells was reduced by 83% (Figure 7D). We did not observe altered binding between WRAP53 and coilin in the SMA cells (Figure 7C), demonstrating that the lost interaction between WRAP53 and SMN is specific and not a secondary effect of disrupted WRAP53–coilin interaction. We thus conclude that the interaction between WRAP53 and SMN is disrupted in SMA type I patients, which further relates to a failure of SMN accumulation in nuclear bodies.
Here we identify WRAP53 as an essential factor for Cajal body maintenance and for directing the SMN complex to Cajal bodies. We show that WRAP53 is a constitutive component of Cajal bodies that overlaps coilin in 100% of Cajal bodies in a variety of cell lines. Knockdown of WRAP53 disrupts Cajal bodies, prevents formation of new Cajal bodies, and relocates Cajal body proteins coilin and SMN from Cajal bodies to nucleoli. WRAP53 seems specifically important for Cajal body integrity, since depletion of WRAP53 does not affect gems or other nuclear structures, including nucleoli and PML bodies. We show that WRAP53 separately associates with coilin and SMN and is required for their complex formation. Previous studies demonstrated a direct interaction between SMN and coilin, suggesting that WRAP53 is not important for their binding but rather mediates interaction by bringing the proteins in close proximity. This may either occur within an already existing Cajal body or result in the formation of a new Cajal body. Importantly, WRAP53's role in Cajal body formation goes beyond bringing SMN and coilin together, since knockdown of SMN does not abolish all Cajal body structures. Residual Cajal bodies containing WRAP53 and coilin still remain. Moreover, in HeLa-PV cells, WRAP53 and coilin localize to all Cajal bodies and SMN to only 40% of them (n = 100) (data not shown). These observations, together with the fact that knockdown of WRAP53 or coilin disrupts all Cajal body structures, point to a more general function of WRAP53 in assisting coilin as a scaffold protein in Cajal body formation.
Cajal bodies have been suggested to have separate compartments containing snRNP, snoRNP/scaRNP, or basal transcription factors [10]. Depletion of proteins involved in snRNP maturation, such as SMN, TGS1, and PHAX, disrupts canonical Cajal bodies containing snRNP, whereas residual Cajal bodies lacking snRNPs but containing coilin and snoRNP/scaRNP components still remain. Without WRAP53, both canonical and residual Cajal bodies collapse, suggesting that WRAP53 is important for processes in addition to snRNP maturation. WRAP53 has been shown to be essential for scaRNA, including telomerase RNA, localization to Cajal bodies [2],[3], which could account for some of the observed defects in Cajal body formation upon WRAP53 perturbation.
We also observe that high overexpression of WRAP53 disassembles Cajal bodies and results in nucleoplasmic mislocalization of WRAP53, coilin, and SMN. This indicates that overexpressed WRAP53 has a dominant negative effect on WRAP53 function and that exogenous and endogenous WRAP53 may compete for factors important for WRAP53 localization to Cajal bodies and Cajal body formation. The fact that endogenous WRAP53 co-precipitates with exogenous WRAP53 indicates that WRAP53 can self-associate, which can also explain the observed effect. Indeed, self-oligomerization appears to be a general feature of nuclear body marker proteins including coilin, SMN, and PML [26]–[28], which is consistent with our findings that WRAP53 is a signature protein for Cajal bodies. In line with this notion, overexpression of coilin also disrupts Cajal bodies and results in coilin mislocalization [13]. Hypothetically, WRAP53 self-association could be the event that brings coilin and SMN together and facilitates Cajal body formation.
The effects on Cajal bodies of depletion and overexpression of WRAP53 are highly similar to those of loss or overexpression of coilin. Coilin mutants have been described in human, mouse [8], Arabidopsis [29], and Drosophila [30]. In all of these species, loss of coilin produces defects in Cajal body formation. Overexpression of coilin, on the other hand, produces slightly different effects in the different organisms. In Drosophila and Arabidopsis, enhanced coilin levels result in normal Cajal body formation or formation of larger Cajal bodies, whereas overexpression of coilin in mice and human cells disrupts Cajal bodies, as previously described. It would be interesting to investigate the effects of WRAP53 depletion and overexpression in other organisms as well.
Deletion mapping of the WRAP53 protein demonstrates that the WD40 domain plus C-terminal aa 456–533 are required for interaction with coilin and SMN. The same domains also target WRAP53 to Cajal bodies, suggesting that interaction with coilin and/or SMN mediates this localization. The apparent lack of SMN in a fraction of WRAP53- and coilin-associated Cajal bodies, however, indicates that coilin is the important factor for WRAP53's nuclear localization. However, we cannot exclude that yet unidentified interaction partners of WRAP53, binding via the same regions, may also be important for localization of WRAP53 to Cajal bodies.
Interestingly, we found that WRAP53 also localizes to the cytoplasm, which is in contrast to two previous reports [2],[3]. Our conclusion is based on the following findings: (1) three different WRAP53 antibodies all demonstrate WRAP53 localization to the cytoplasm, (2) overexpressed Flag- and EGFP-tagged WRAP53 shows cytoplasmic localization, (3) WRAP53 knockdown efficiently depletes WRAP53 staining in both locations, (4) cytoplasmic localization of WRAP53 is shown in seven different cell lines and with different fixation protocols, and (5) cell fractionation followed by WB confirms presence of WRAP53 in the cytoplasm.
We find that WRAP53 associates with SMN both in the cytoplasm and in the nucleus and influences the intracellular distribution of SMN. Knockdown of WRAP53 results in SMN accumulation in the cytoplasm and decreased SMN in the nucleus. Thus, cytoplasmic WRAP53 seems to recruit SMN from the cytoplasm to the nucleus. This role of WRAP53 is separate from WRAP53's function in Cajal body formation, since knockdown of coilin does not affect the intracellular distribution of SMN. Moreover, knockdown of WRAP53 abrogates interaction between SMN and the nuclear pore receptor importinβ, which could explain the skewed intracellular distribution of SMN observed in WRAP53-depleted cells. Nuclear localization of SMN has been shown to depend on SMN–importinβ complex formation but also on other factors such as proper snRNP assembly [18]. Importantly, WRAP53 depletion does not affect the interaction between SMN, Gemin3, and Sm. This suggests that WRAP53 does not promote SMN complex formation nor snRNP assembly but rather is important for SMN-associated interactions occurring after these events. It appears as if WRAP53 recruits the SMN complex from the cytoplasm by facilitating SMN–importinβ complex formation and further mediates interaction between SMN and coilin in the nucleus by bringing the proteins in close proximity. Our finding that Sm is not retained in the cytoplasm upon WRAP53 depletion is in line with several previous studies where no cytosolic snRNP accumulations were observed upon silencing of SMN, Gemin3, or Gemin4 [10],[31],[32]. The very stable nature of mature snRNPs [33] and additional import mechanisms for snRNPs independent of SMN–importinβ interaction have been suggested as plausible reasons.
We also find that the interaction between SMN and WRAP53 is disrupted in fibroblasts from SMA patients and that this correlates with a reduced number of SMN foci in the nuclei of these cells. The reason for the loss of binding between WRAP53 and SMN in SMA is currently unknown but opens the possibility that WRAP53 could contribute to SMN dysfunction in SMA. In summary, we have demonstrated two important functions of the WRAP53 protein. First, it acts as a critical scaffold protein for Cajal body formation, along with coilin. Second, it recruits the SMN complex from the cytoplasm to nuclear Cajal bodies by mediating interaction between SMN, importinβ, and coilin. This discovery not only reveals new functions of the WRAP53 protein but also increases our understanding of the molecular mechanism behind Cajal body formation and recruitment of factors to Cajal bodies.
U2OS, H1299, HCT116, HEK293, MCF-7, HeLa-PV, HeLa-ATCC, and HDF cells were maintained in Dulbecco's modified medium supplemented with 10% fetal bovine serum (Invitrogen), 2 mM L-glutamine (Invitrogen), and 2.5 µg/ml Plasmocin (InvivoGen) at 37°C in 5% CO2 humidified incubators. Primary fibroblast cell lines from two patients with SMA type I (GM03813 and GM03815) and one heterozygous carrier (GM03814, clinically unaffected mother of GM03813 and GM03815) were obtained from Coriell Cell Repository and maintained in MEM supplemented with 10% fetal bovine serum. The GM03815 cells were initially described in the Coriell Cell Repository as derived from a heterozygous carrier (clinically unaffected father of GM03813). However, snRNP deficiency and genetic linkage analysis of these cells showed that they in fact are derived from a homozygous carrier with SMA type I (a male sibling of GM03813) [34].
Total RNA was extracted from tumor cell lines using the Trizol reagent (Invitrogen). The RNA was reverse transcribed with First-Strand cDNA synthesis using Superscript II (Invitrogen). Quantitative real-time PCR was carried out in the Applied Biosystems 7500 Real-Time PCR using transcript-specific TaqMan Gene Expression Assays (Applied Biosystems). The following probes were used: Hs01126636_g1 for detection of WRAP53 and Hs00167441_m1 for detection of ALAS1 as endogenous control.
For IF experiments, cells were grown on sterilized cover slips, fixed with 100% MeOH at −20°C, permeabilized with 0.1% Triton X-100 for 5 min, and then blocked in blocking buffer (2% BSA, 5% glycerol, 0.2% Tween 20, and 0.1% NaN3). Cover slips were subsequently incubated for 1 h in primary antibody and 40 min in secondary antibody diluted in blocking buffer. The cover slips were mounted with Vectorshield mounting medium with DAPI (Vector Laboratories). Images were acquired with a Zeiss Axioplan 2 microscope, equipped with an AxioCam HRm camera using 43× or 60× oil immersion lenses, and processed using Axiovision Release 4.7.
For IP, cells were lysed in NP40 buffer (150 mM NaCl, 50 mM Tris-HCL [pH 8.0], 1% NP40, 1% PMSF, and 1% protease inhibitor cocktail) for 15 min on ice, followed by sonication 2× for 10 s. Extracts were spun down at 6,000 rpm for 5 min at 4°C and then quantified by Bradford assay (Bio-Rad). Endogenous proteins were immunoprecipitated with 1 µg of affinity-purified antibody per 1 mg extract supplemented with 10 µl of Dynabeads Protein G (Invitrogen) overnight at 4°C. The beads were washed 4× for 15 min with 1 ml of NP40 buffer and prepared for WB.
Cell extracts for WB analysis were prepared as previously described [1]. WB was performed according to standard procedures.
Cell fractionations were performed using a nuclear extract kit according to the manufacturer's instructions (Nuclear Extract Kit, Active Motif). For fractionation followed by IP, cells were lysed in hypotonic buffer (10 mM NaCl, 20 mM Tris-HCL [pH 7.5], 1% PMSF, and 1% protease inhibitor cocktail) for 20 min on ice. Samples were spun down at 3,600 rpm for 10 min at 4°C, and the supernatant (cytoplasmic fraction) was converted into NP40 buffer. The remaining pellet was lysed in NP40 buffer. The IP was performed as described earlier. In Figure 1B, equal volumes of each sample were loaded on an SDS-PAGE gel, whereas in Figure 3A, the maximal volume of each sample was loaded on the gel.
Four different WRAP53 antibodies were used: rabbit α-WRAP53-C1 [1] (used for IP and WB), rabbit α-WRAP53-C2 (used for WB, IP, and IF), rabbit α-WRAP53 (Wdr79, A301-442A-1, Bethyl Laboratories; used for WB, IP, IF, and in situ PLA), and mouse polyclonal α-WRAP53 full-length (H00055135-B01, Abnova; used only in Figure S3A and S3B). To generate α-WRAP53-C2, rabbits were immunized with a KLH-conjugated WRAP53 peptide that maps to a region between aa 498–548 of full-length WRAP53 protein (Innovagen AB). The following antibodies were used in IF, IP, and WB: mouse α-coilin (ab11822, Abcam), rabbit α-coilin (sc-32860, Santa Cruz Biotechnology), mouse α-SMN (610647, BD Biosciences), mouse α-SMN (sc-32313, Santa Cruz Biotechnology), rabbit α-SMN (sc-15320, Santa Cruz Biotechnology), mouse α-Gemin3 (ab10305, Abcam), mouse α-Gemin3 (sc-57007, Santa Cruz Biotechnology), mouse α-Gemin2 (sc-57006, Santa Cruz Biotechnology), mouse α-importinβ (035K4852, Sigma), mouse α-importinβ (sc-137016, Santa Cruz Biotechnology), rabbit α-fibrillarin (ab5821, Abcam), mouse α-Sm (ab3138, Abcam), mouse α-SmB/B′ (sc-271094, Santa Cruz Biotechnology), mouse α-Hsp90α/β (sc-13119, Santa Cruz Biotechnology), rabbit α-Lamin A/C (sc-20681, Santa Cruz Biotechnology), rabbit IgG (sc-2027, Santa Cruz Biotechnology), mouse IgG (sc-2025, Santa Cruz Biotechnology), mouse α-Flag M2 (200472-21, Stratagene), rabbit α-GFP (ab290, Abcam), and mouse α-β-actin (Sigma). The following secondary antibodies were used: sheep α-mouse HRP (NA931V, GE Healthcare), donkey α-rabbit HRP (NA934V, GE Healthcare), swine α-rabbit FITC (F0054, Dako Cytomation), and horse α-mouse Texas Red (TI-2000, Vector).
The following siRNAs from Qiagen were used: siWRAP53#1 (SI00388941), siWRAP53#2 (SI00388948), siSMN1#7 (SI03108084), siSMN1#11 (SI04950932), siSMN1#12 (SI04950939), siCoilin#3 (SI00350343), siCoilin#7 (SI04330830), and a control siRNA (1027280). siRNA (10–20 nM) was transfected into cells using either Oligofectamine (Invitrogen) or HiPerfect (Qiagen) transfection reagents, in accordance with the supplier's recommendations. Flag-WRAP53 was cloned as described in Mahmoudi et al. [1]. To generate EGFP-tagged WRAP53 constructs, full-length or deletion mutants were amplified by PCR (Advantage-HF 2 polymerase, Clontech) and subcloned into pEGFP-C1 vector (Clontech). Flag-SMN was cloned by PCR (Advantage-HF 2 polymerase, Clontech) from SMN cDNA and subcloned into pCMV-Tag2 vector (Invitrogen). All primers used for PCR amplifications are listed in Table S1. Plasmid transfections were performed using Lipofectamine 2000 Reagent (Invitrogen).
For the generation of cells with stable overexpression of WRAP53, Flag-tagged full-length WRAP53 cDNA was cloned into the pLenti6/V5-D-TOPO vector (Invitrogen). The pLenti6-Flag-WRAP53 plasmid along with the pMDLg/RRE, pCMV-VSVG, and pRSV-Rev plasmids required for viral production were transfected into HEK293FT cells using Lipofectamine 2000 (Invitrogen). U20S cells were infected with viruses containing either pLenti6-Flag-WRAP53 or empty pLenti6/V5-D-TOPO vector, and positive cells were selected 48 h after infection using 10 µg/ml Blasticidin (Invitrogen).
All analyses were performed using Microsoft Office Excel 2003. Two-tailed Student's t test was used to determine statistical significance.
In situ PLA experiments were performed as described previously [35]. Incubation with primary antibodies (0.4 ng/µl rabbit α-WRAP53 and 1 ng/µl mouse α-SMN in blocking solution) was performed at room temperature for 1 h. Cells were washed 3× for 5 min in PBS plus 0.1% Tween 20, with the first wash at 37°C. Secondary proximity probes (Rabbit-PLUS and Mouse-MINUS, Duolink kit, Olink Biosciences AB) were incubated for 2 h at 37°C. Cells were washed 1× for 5 min in 10 mM Tris-HCl (pH 7.5) plus 0.1% Tween 20 at 37°C, then 2× for 5 min in PBS plus 0.1% Tween 20. All subsequent steps were done according to the Duolink kit protocol (Olink Biosciences AB). FITC-labeled donkey α-rabbit F(ab′)2 fragment (Jackson ImmunoResearch) was added in order to counterstain for WRAP53. Images were acquired using an epifluorescent microscope (Axioplan 2, Zeiss) equipped with a 100-W mercury lamp, a CCD camera (C4742-95, Hamamatsu), emission filters for visualization of DAPI, FITC, and Cy3.5, and a 63× objective (plan-neofluar). WRAP53 staining was used to select image position.
|
10.1371/journal.pntd.0005901 | Secondary bacterial infections and antibiotic resistance among tungiasis patients in Western, Kenya | Tungiasis or jigger infestation is a parasitic disease caused by the female sand flea Tunga penetrans. Secondary infection of the lesions caused by this flea is common in endemic communities. This study sought to shed light on the bacterial pathogens causing secondary infections in tungiasis lesions and their susceptibility profiles to commonly prescribed antibiotics. Participants were recruited with the help of Community Health Workers. Swabs were taken from lesions which showed signs of secondary infection. Identification of suspected bacteria colonies was done by colony morphology, Gram staining, and biochemical tests. The Kirby Bauer disc diffusion test was used to determine the drug susceptibility profiles. Out of 37 participants, from whom swabs were collected, specimen were positive in 29 and 8 had no growth. From these, 10 different strains of bacteria were isolated. Two were Gram positive bacteria and they were, Staphylococcus epidermidis (38.3%) and Staphylococcus aureus (21.3%). Eight were Gram negative namely Enterobacter cloacae (8.5%), Proteus species (8.5%), Klebsiellla species (6.4%), Aeromonas sobria (4.3%), Citrobacter species (4.3%), Proteus mirabillis(4.3%), Enterobacter amnigenus (2.1%) and Klebsiella pneumoniae (2.1%). The methicillin resistant S. aureus (MRSA) isolated were also resistant to clindamycin, kanamycin, erythromycin, nalidixic acid, trimethorprim sulfamethoxazole and tetracycline. All the Gram negative and Gram positive bacteria isolates were sensitive to gentamicin and norfloxacin drugs. Results from this study confirms the presence of resistant bacteria in tungiasis lesions hence highlighting the significance of secondary infection of the lesions in endemic communties. This therefore suggests that antimicrobial susceptibility testing may be considered to guide in identification of appropriate antibiotics and treatment therapy among tungiasis patients.
| Secondary bacterial infection of tungiaisis lesions is a threat to jigger infested patients. Once the flea penetrates the skin, it leaves an opening on the skin through which it lays eggs, defecates and breathes throughout it’s life cycle on the host. At the same time, it feeds on the host’s blood hence a direct connection of the environment to the blood stream is established. Chances of bacteria getting into the blood stream is therefore greatly enhanced. Once bacteria gets into the blood, it can lead to life threatening conditions like septicemia, meningitis, pneumonia and toxic shock syndrome. Affected communities are oblivious to this danger and they neither seek nor get treated for this parasitosis. The health officials and scientific community also continue to ignore this disease hence it’s extremely neglected. Consequently, not much is understood concerning the disease dynamics, distribution and pathogenesis. This study therefore is a deliberate effort to bring attention to one aspect of the disease dynamics of this menace.
| Tungiasis is a parasitic disease caused by the sand flea Tunga penetrans [1].The fleas can infest any part of the body. However majority of the cases occur on the feet [2].Children and the elderly bear the brunt of the infection in endemic areas [3], [4]. During the transmission period, a study that followed up individuals entering an endemic area was able to demonstrate that by the third week all the participants were infested by T. penetrans [5]. The ectoparasites can cause more than 50 lesions in a single individual in some cases [1]. This leads to severe inflammation and ulceration which is associated with intense pain. Walking, working or going to school becomes a problem. In endemic communities it’s not uncommon to find children who have dropped out of school due to the pain and stigma brough about by this condition. In some cases there is loss of toe nails and deformation of digits [6].
Secondary infection of the lesions caused by Tunga species occurs in endemic areas. A bacteriological investigation of the lesions from human tungiasis in Brazil reported isolation of various pathogenic bacteria [7]. Untreated Tungiasis is a risk factor in acquiring blood stream bacterial infections (bacteremia) due to broken skin. Once the jigger flea penetrates the skin, it maintains an opening (250 to 500μm) in the epidermis through which it defecates, breathes and lays eggs, consequently connecting the outer surface of the skin and the blood stream as it feeds [7], [8]). The exposed skin tissue is a perfect environment for bacteria to thrive. It provides warmth, moisture and nutrients, factors that are essential for microbial growth [9].
Sepsis in Tungiasis patients has been elucidated hence illustrating the medical significance of systemic infections caused by secondary bacteria infection in these patients [10], [11]. Systemic conditions like pneumonia, meningitis, osteomyelitis, endocarditis, septicemia and Toxic shock syndrome (TSS) can be fatal [12].
Antibiotic resistance compounds the problem, as treatment options are greatly diminished due to bacteria developing mechanisms that neutralize available antibiotics [13]. The selection of appropriate antibiotics for treatment of severe tungiasis is critical for proper management of this parasitosis. This study therefore sought to shed light on the bacterial pathogens causing secondary infections in tungiasis lesions and their susceptibility profiles to commonly used antibiotics.
The study area was Vihiga County in Western Kenya. It is one of the most densely populated rural areas in Kenya.The population density as of 2009, was 1,045 persons per square kilometre. This figure is projected to rise to 1231 persons per square kilometer in 2017. The poverty levels are consequently very high due to the population pressure on land and other resources. The GDP per capita income was reported as US $ 1,103 in 2013. Majority of the inhabitants own small uneconomical pieces of land as a result of increased subdivision occassioned by the high population and cultural practice of land inheritance.
The area has two rainy seasons. Long rainy season in April, May and June and the short rains in September, October and November. The study took place during the dry and hot season from January to March, 2016. Tungiasis peaks during the dry season.
Participants were recruited with the help of Community Health Workers. Swabs were taken from lesions which clinically showed signs of secondary infection like swelling, erythema and pus.
The flea was extracted with a sterile needle after disinfection of the surrounding skin with 70% alcohol for 1 minute [7]. Swabs were then collected from the surgical lesions by use of sterile cotton swabs. The cotton swabs were moistened in sterile physiological saline and gently moved in and out of the remaining cavity. One swab was used for each lesion and labeled accordingly. The swab was then transferred to a sterile transportation tube, briefly stored in an ice box and transported to the laboratory. Upon arrival at the laboratory, the swabs were cultured separately and directly onto Mannitol salt and MacConkey agar (both from Oxoid Ltd, Basingstoke, United Kingdom). They were also inoculated on Brain heart infusion agar (Oxoid Ltd, Basingstoke, United Kingdom) supplemented with the commercially available 5% sheep blood (TCS Biosciences, Botolph Claydon, Buckingham, United Kingdom) and incubated aerobically at 35°C for 24 hours.
Identification of suspected bacteria colonies was done by colony morphology, Gram staining, catalase, coagulase tests and biochemical tests as described previously [14]. Anaerobic bacteria were not isolated due to limited laboratory facilities at the time.
The bacterial isolates were subjected to antimicrobial sensitivity testing by disk diffusion method as described by [15]. Briefly, test organisms were suspended in sterile normal saline to conform to 0.5 McFarland turbidity standard. With the aid of sterile cotton swab the suspended organisms were spread on Mueller-Hinton (Oxoid Ltd, Basingstoke, United Kingdom) Agar plate and the antibiotic disks dispensed. The plates were incubated at 37°C for 16-18h. Inhibition zone diameters were determined and recorded in Excel sheets and interpreted according to the Clinical and Laboratory Standards Institute (CLSI) guidelines [16].
The following panel of antibiotics (all from Oxoid) and their concentrations were used. For Gram negatives cefuroxime sodium 30 μg, amoxycillin\clavulanic acid 2:1 30 μg, chloramphenicol 30 μg, tetracycline 30μg, co-trimoxazole sxt 25 μg, nalidixic acid 30 μg, ampicillin 10 μg, ceftazidime 30 μg, cefotaxime 30 μg, ciprofloxacin 5 μg, norfloxacin 10μg, gentamycin 10μg, were tested. For Gram positives meropenem 10 μg, gentamycin 10μg, kanamycin 30μg, clindamycin 2μg, norfloxacin 10μg, ofloxacin 5μg, oxacillin 5μg, erythromycin 10μg, nalidixic acid 30 μg, trimethoprim-sulfamethoxazole 25μg, chloramphenicol 30 μg, tetracycline 30μg were tested.
The study got approval from the KEMRI Scientific and Ethics Review Unit (SERU). Approval number KEMRI/SERU/CTMDR/015/3116. Informed written consent was obtained from all participants including children, where the guardian provided an informed consent on their behalf. All data analyzed was coded and identity of participants kept confidential. All participants were treated for tungiasis according to the National Policy Guidelines on Prevention and Control of Jigger Infestations in Kenya [17]. This was by (removing the embedded flea with a sterile needle and disinfection of the skin lesion) or bathing the affected area in 0.05% potassium permanganate for 10 minutes. A nurse who was part of the team also vaccinated the participants against tetanus. Severe cases were referred to their health center by Community Health Extension Workers from the study area, who have a functioning referral system in place.
In the three sublocations from Vihiga County, 103 people were identified as having tungiasis. Swabs were taken from 37 patients who had lesions with clinical signs of secondary infection (tenderness, oedema, erythema with or without pus) Figs 1 and 2. Out of the 37 patients, from whom swabs were collected, specimen were positive in 29 and 8 had no growth. The proportion of male to female was 23 to 14 respectively. The median age was 12 years with a range of 5–80 years.
From the 29 Tungiasis patients up to 10 strains of bacteria were isolated. The most common being S. epidermidis (38.3%) followed by S. aureus(21.3%) and the least bacteria isolated being K. pneumoniae (2.1%). The other bacteria isolated are shown in Fig 3.
Out of the ten strains of bacteria isolated two were Gram positive [S. epidermidis (38.3%) and S. aureus (21.3%)]. Eight were Gram negative namely Enterobacter cloacae (8.5%), Proteus species (8.5%), Klebsiella species (6.4%), Aeromonas sobria (4.3%), Citrobacter species (4.3%), Proteus mirabillis(4.3%), Enterobacter amnigenus (2.1%) and K. pneumoniae (2.1%).
Out of the 29 patients, 20 (69%) had a single strain of bacteria whereas nine patients (31%), had more than one strain of bacteria. Most of the single strain infection were Gram positive, S. epidermidis and S. aureus; 60% and 20% respectively. The rest were Proteus species (10%), Klebsiella species (5%) and E. cloacae (5%).
In the polymicrobial infection, majority 66.7% (6/9) were S.aureus and S. epidermidis (2/9) (22.2%) and their combinations as shown in Table 1. Up to 17.2% had two different types of bacteria, three and four polymicrobial infections had 3.4% each and two patients (6.9%) had five different bacteria.
Once identified, bacterial isolates were further tested for drug sensitivity to commonly prescribed drugs using the Kirby Bauer disk diffusion method. Eleven drugs were used to test susceptibility of the Gram negative isolates.All the isolates were sensitive to ciprofloxacin, cefotaxime, norfloxacin, gentamicin, nalidixic acid, chloramphenicol (Table 2). Ampicillin had the highest resistance of 52.6%. All the Gram negative bacterial isolates were resistant to atleast one or more drugs except E. amnigenus.
Enterobacter cloacae showed resistance to ampicillin (75.0%), amoxicillin clavulanic acid (25.0%) tetracycline (50.0%), ceftazidime (25.0%) and cefuroxime (25.0%). Citrobacter species showed resistance to ampicillin (100.0%), amoxicillin clavulanic acid (100.0%) and cefuroxime (50.0%). The two proteus species had resistance to tetracycline; 50.0% and 75.0% respectively (Table 2).
All the Gram positive were sensitive to norfloxacin, ofloxacin, meropenem and gentamicin drugs. However they were resistant to clindamycin, kanamycin, oxacillin, erythromycin, nalidixic acid, trimethorprim sulfamethoxazole, chloramphenicol and tetracycline (Table 3). nalidixic acid had the highest bacterial resistance (31.0%) followed by clindamycin (20.7%).Three patients (10.3%) had S. aureus isolates that were methicillin resistant (MRSA).
Both Gram negative and Gram positive bacteria isolates were sensitive to Gentamicin and Norfloxacin drugs.
Secondary bacteria infection of the lesions caused by jiggers remains a commmon occurrence among communities affected by this parasite [18]. This may be attributed to the fact that the flea interferes with the integrity of the skin which is the first defence of the body against microbes [9]. There is paucity in the number of bacteriological studies that have been done to investigate secondary infection of the lesions in tungiasis patients [7]. In this study, ten different strains of aerobic bacteria were isolated, eight of which were Gram negative and two were Gram positive. These findings corroborate a previous study carried out in Brazil Feldmeier et al., 2002 [7] which reported isolation of aerobic bacteria from tungiasis lesions. However in that study Streptococcus pyogenes, Streptococci serogroup G, Enterococcus faecalis, Morganella morganii, Pseudomonas species and Bacillus species were among the aerobic bacteria species isolated but missing in the present study. In this study, Aeromonas sobria, Citrobacter species and Enterobacter amnigenus were isolated for the first time. This discrepancy may be attributed to the different geographical regions and environmental factors.
Other than aerobic bacteria, anaerobic bacteria like Peptostreptococcus species, Clostridium bifermentans and Clostridium sordelli have also been reported [7]. Clostridium tetani bacteria which causes tetanus has been implicated in some patients presenting with tungiasis [19, 11].
Lesions caused by Tunga penetrans can be infected with several bacteria strains at the same time. In the present study, polymicrobial infection was observed in 31% of the patients assessed with 6.9% of these having up to 5 different bacteria strains. These results are consistent with a previous study which reported a single patient having up to 5 different pathogens [7]. Implying that antimicrobial sussceptibility testing on bacterial isolates from tungiasis patients would guide in identification of appropriate antibiotics and treatment therapy in these patients.
The Gram negative bacteria isolated in this study were found to be sensitive to ciprofloxacin, cefotaxime, norfloxacin, gentamicin and nalidixic acid drugs while the Gram positive were sensitive to norfloxacin, ofloxacin, meropenem and gentamicin. Norfloxacin is a broad spectrum fluoroquinoline that is active against both Gram negative and Gram positive bacteria. Gentamicin is an aminoglycoside that is effective against a wide range of pathogens as well. Several studies including the present study confirms their considerable anti—bacterial activities [20,21,22].
In endemic communities, tungiasis is not recognised as a disease that warrants medical attention. In other communities it’s thought to be a curse or witchcraft hence the affected people rarely seek or get treatment instead they wait to die [23]. The public Should be advised that it’s a parasitic infection that can be managed and if left unmanaged can be life threathening due to secondary infections. In the event that bacteria gets into the blood stream, it could lead to systemic infections which can be fatal [24,25].
Previously, some of the bacteria isolated in this study were considered to be of low pathogenicity. However with increase in knowledge and updated technology, this school of thoght is rapidly being set aside due to the new evidence available. For instance, coagulase negative Staphylococci which were thought to be harmless commensals, are now more than ever considered pathogens of medical importance causing considerable infections of the blood stream and other internal organs once they become invasive [26, 27].
The challenges posed by drug resistance compounds the problem even more. Treatment outcomes of resistant bacteria like methicillin resistant S. aureus (MRSA) infections are worse compared to the sensitive strains as it’s associated with increased morbidity and mortality [28]. MRSA used to be associated with hospital acquired infections but is now being isolated in a broad spectrum of community acquired diseases [29, 30]. Further, most MRSA strains are also found to be multi drug resistant [28]. This was observed in this study, since the MRSA isolated were also resistant to clindamycin, kanamycin, erythromycin, nalidixic acid, trimethorprim sulfamethoxazole and tetracycline.
Several studies have shown that there is an association between the use of antibiotics in livestock reared for food production and emergence of antibiotic resistance in human beings [31, 32, 33, 34]. Other aspects implicated for spread of antibiotic drug resistance are, poor hygiene, contaminated food, polluted water, overcrowding and compromised immunity due to malnutrition or HIV. In addition, we also have misuse and consumption of sub optimal doses of antibiotics hence inducing selection pressure for antibiotic resistance [35].
One limitation of the study is that we were not able to determine if the tungiaisis patients were taking any antibiotic medication prior to participation in the study or their immunity status. These aspects may have further explained whether they had any role in the observed antibiotic resistance. Therefore there is a need to carry out a follow up study to focus on the cause of antimicrobial drug resistance among tungiasis patients. The study was not able to isolate anaerobic bacteria due to lack of equipment at the time.
The findings from this study confirm the presence of resistant bacteria in tungiasis lesions hence highlighting the significance of secondary infection of the lesions in endemic communties. This therefore implies that the treatment regimen for tungiasis especially in severe cases should be expanded to include antibiotics. Antimicrobial susceptibility testing may be considered to guide in identification of appropriate antibiotics. Norfloxacin and gentamicin have shown to be very effective against both Gram negative and Gram positive bacteria. In severe tungiasis where sepsis is observed, a broad spectrum drug may be considered at the onset to avoid delay in starting treatment as results from cultures are awaited.
|
10.1371/journal.pcbi.1002850 | Interconnected Cavernous Structure of Bacterial Fruiting Bodies | The formation of spore-filled fruiting bodies by myxobacteria is a fascinating case of multicellular self-organization by bacteria. The organization of Myxococcus xanthus into fruiting bodies has long been studied not only as an important example of collective motion of bacteria, but also as a simplified model for developmental morphogenesis. Sporulation within the nascent fruiting body requires signaling between moving cells in order that the rod-shaped self-propelled cells differentiate into spores at the appropriate time. Probing the three-dimensional structure of myxobacteria fruiting bodies has previously presented a challenge due to limitations of different imaging methods. A new technique using Infrared Optical Coherence Tomography (OCT) revealed previously unknown details of the internal structure of M. xanthus fruiting bodies consisting of interconnected pockets of relative high and low spore density regions. To make sense of the experimentally observed structure, modeling and computer simulations were used to test a hypothesized mechanism that could produce high-density pockets of spores. The mechanism consists of self-propelled cells aligning with each other and signaling by end-to-end contact to coordinate the process of differentiation resulting in a pattern of clusters observed in the experiment. The integration of novel OCT experimental techniques with computational simulations can provide new insight into the mechanisms that can give rise to the pattern formation seen in other biological systems such as dictyostelids, social amoeba known to form multicellular aggregates observed as slugs under starvation conditions.
| Understanding bacteria self-organization is an active area of research with broad implications in both microbiology and developmental biology. Myxococcus xanthus undergoes multicellular aggregation and differentiation under starvation and is widely used as a model organism for studying bacteria self-organization. In this paper, we present the findings of an innovative non-invasive experimental technique that reveals a heterogeneous structure of the fruiting body not seen in earlier studies. Insight into the biological mechanism for these observed patterns is gained by integrating experiments with biologically relevant computational simulations. The simulations show that a novel mechanism requiring cell alignment, signaling and steric interactions can explain the pockets of spore clusters observed experimentally in the fruiting bodies of M. xanthus.
| The organization of Myxococcus xanthus, the most studied species of the myxobacteria, into structures known as fruiting bodies has long been studied not only as an example of collective motion of bacteria, but also as a simplified model for developmental morphogenesis [1], [2]. Individual M. xanthus cells do not have flagella and move on a substrate using gliding motility [3], [4]. The fruiting body process begins when myxobacteria are starved for nutrients and, in response, the population of cells gather into large aggregates containing hundreds of thousands of cells that continue to move around within the aggregate. Eventually, the cells differentiate from motile rod shaped cells to non-motile spherical spores that can wait out the harsh conditions. During this process, a 17 kD protein known as C-signal is transferred between cells and critical to the differentiation process [5], [6]. It has been shown that C-signal requires end-to-end alignment [5], that C-signaling requires cells to move[7], and that C-signal accumulates on cells throughout development process and peaks when spores form [8].
Although the nascent fruiting body contains on the order of cells, only 1% of the cells in a fruiting body become viable spores [9]. The remaining cells, which constitute the bulk volume of the fruiting body, fail to become spores, lyse, and their extracellular material, polysaccharides in particular, is somehow integrated into the internal structure of the fruiting body. Part of the cell debris would serve as a source of nutrients for cells moving in the mound. Despite the fact that Scanning Electron Microscope (SEM) images showed what appeared as a dense homogeneous packing of spores [9], it is difficult to resolve such a homogeneous distribution of spores with the fact that a bulk of the cells never become spores.
We present, in this paper, an integrative approach that combines a new experimental technique using infra-red optical coherence tomography (OCT) with computational models to study the patterns of spores as they form within a fruiting body. Viewing fruiting bodies by this tomography method revealed that regions of high spore concentrations in the fruiting body were surrounded by less dense regions. Based upon the experimental findings, we developed a hypothesis based on the the underlying biology of M. xanthus that can explain the pattern without chemotaxis or long-range diffusive chemicals which have been used to explain other types of biological patterns. Our hypothesis is that the basic mechanism behind this patterning is that cells move along slime trails and reverse to improve alignment so they can C-signal. The increase of C-signal is done locally by cells which coordinates the differentiation process in order for spores to form in pockets of clusters throughout the mound. We present an extended description of the hypothesis from the biological viewpoint in the Results section.
To test if the hypothesis is plausible, we developed two separate models that use different degrees of biological detail. In the two models that we present, we focus on the later stage of the fruiting body process when cells have already aggregated in some domain and the sporulation is beginning. The general modeling approach studies how the coordinated self-propelled cell movement and C-signaling can give rise to the spatial patterns of spore clusters observed in experiments.
We begin with a one dimensional (1D) model that tests how jamming and C-signaling generate clustering on a circular track. The second two dimensional (2D) model implements cell shape and movement and utilizes C-signaling that requires end-to-end cell alignment. The simplicity of the 1D model allows us to study a very wide range of parameter values in order to gain insight into the relative importance of two specific aspects of cell clustering — jamming by cells encountering spores and impact of C-signaling. The 2D model is more computationally demanding and cannot explore the same range of parameter values, but instead focuses on adding more biological details such as connecting cell shape and movement with C-signaling that requires alignment. Model simulation results were compared with the experimentally observed clustering of spores. The hypothesized mechanism based on cells aligning and signaling by contact to coordinate sporulation was able to recover the structure of the fruiting body observed in the experiments. In addition to gaining new insight into bacterial fruiting body formation, better understanding of cell self-organization based on cell-cell signaling and interaction is of real importance for developmental biology.
In order to carry out these studies, we developed an apparatus that integrated the motorized stage of a microscope with a stand alone Optical Coherence Tomography (OCT) device and ran fully computerized scans using LabView to control timing of scans and probe position. Then, we developed preprocessing routines in ImageJ to extract planar cross-sections of data that could then be analyzed by additional programs written in Matlab. The analysis programs extracted statistical properties from the three-dimensional (3D) intensity data and also rendered 3D images of the mound.
We used CTT agar plates for both normal and starved growing conditions of M. xanthus. CTT-agar plates are made by adding 1.5% agar by weight to a TPM (10 mM Tris pH 8.0, 1 mM , 8 mM ) buffer which has 1.0% Casitone by weight. The fruiting body plates are prepared by reducing the amount of Casitone from 1.0% to 0.1%. This results in a starvation condition for the cells growing on the surface and the fruiting body process is carried out. Scans of mound were made 1–2 weeks after inoculating the starvation plates. This is well beyond the 2–3 days needed to form the fruiting body mounds and ensured that the mounds were no longer actively forming.
In order to examine 3D bacterial density distribution, we employed non-invasive high-resolution infrared optical coherence tomography (OCT) [10]. OCT is an interferometric technique for imaging in scattering media which measures an in-depth profile of optical scattering using light of low coherence. Consequently, a cross-sectional image is created by scanning the beam position laterally over the sample. The fundamentals of OCT relies on the fact that in a scattering medium only the reflected (non-scattered) light is coherent. Correspondingly, an optical interferometer is used to separate scattered light and detect coherent light. A commercially available imaging system (Niris OCT, Imalux Corporation, Cleveland, OH) was used in our work. The time domain OCT system [11] uses common path optical topology, a 1310 nm central wavelength with 55 nm bandwidth, with in-depth resolution of in air and in water. Acquisition time for an image of maximal resolution up to is 1.5 seconds. The OCT probe, mounted on flexible cable, has a diameter of 2.7 mm and a 2 mm lateral field of view with lateral resolution 25 . It can be easily mounted in close proximity to the sample to image its full depth (about 1–2 mm above the sample). While the OCT depth-scan is performed by the piezofiber delay line, the lateral OCT scan is performed either by moving the sample or the probe beam illuminating the sample. The OCT was recently used for the analysis of collective motion in suspensions of swimming bacteria [12]. Since the resolution of the OCT is of the order , it can only distinguish the large scale structure of the fruiting body, such as cavities and clusters of spores, and not individual bacteria cells.
Scanning electron microscopy (SEM) is a technique that has been used previously to show that spores within a mound are tightly packed [9]. However, preparation a fruiting body for SEM is invasive and requires dehydration in alcohol and other drying agents which likely compressed the structure, removing regions that contain significant amounts of hydrated polysaccharide and extracellular material.
Other researchers concerned with these limitations used laser scanning confocal microscopy (LSCM) with fluorescently labeled bacteria to probe the internal structure of mounds [13]. Preparation for LSCM, like OCT, does not require dehydration so the fruiting bodies can be grown and imaged on agar plates without additional processing. However, LSCM also faces limitations concerning the excitation and emission wavelength of Green Fluorescent Protein (GFP). LSCM typically uses optical wavelength light to excitation a sample and capture the emitted light from a particular focal plain by blocking the out-of-focus light. The use of LSCM to explore fruiting bodies mounds raises the following concerns. Researchers observed that GFP expression appeared to form an outer shell for the dome-like mound. They concluded that the core was likely to be a hollowed out region supported by the extracellular polysaccharide. While they did argue against a differential in GFP expression by cells in the shell and cell in the core, there is another possible explanation for the shell-like pattern.
GFP fluorescence uses 480 nm excitation light and emits at a wavelength of 510 nm [14]. As light travels through any media, it undergoes both Raleigh scattering by particles smaller than the wavelength of the light and Mie scattering by particles that larger than the wavelength of light. Raleigh scattering is inversely proportional to the wavelength raised to the fourth power. This means that excitation light of 480 nm scatters 60× as much as IR light with wavelength equal to 1310 nm, which allow IR light to probe more deeply than visible light. A fruiting body mound is composed of micron-sized spores as well as countless molecules ranging widely over the nanometer scale (most importantly smaller than visible wavelength of GFP). The infrared light used by the OCT provides a better probe for the internal structure of the fruiting body. The trade-off for better scattering depth is the reduce resolution.
In addition to the scattering, M. xanthus is known to produce carotenoids designed to absorb visible light [15]. Production of carotenoids is often avoided in lab conditions by growing plates in the dark. Scattering and absorption will determine the maximum depth at which GFP is visible. It was found that the maximum depth for GFP in lung tissue was [14]. The average height of mounds in [13] was found to be 27 microns with some mounds reaching heights of 45 microns. It is quite possible that visible light of GFP cannot be detected from the core of the fruiting body. This problem is overcome by using longer wavelength light, like the infrared (IR) light used by OCT.
Microscopy was performed on an inverted Olympus microscope and images were taken with a Spot Boost EMCCD 2100 (Diagnostic Instruments Inc.) high sensitivity camera. The camera was still sensitive to the IR probe from the OCT device which appeared as a small white dot in the field of view. This is what enabled the accurate positioning of the probe over specific mounds.
In order to obtain 3D OCT scans of fruiting bodies, we search for a desired region using bright field microscopy at low magnification. Then, using a three-axis micrometer driven translational stage, we position the probe head over the site. The inverted microscope allows for accurately positioning the probe because the mounds can still be seen while the probe head is in the optical path of the microscope (see figure 1b). Once positioned, a scan is made by making 2D slices of the mound while the stage is being moved perpendicular to the lateral scan of the probe. The automation was controlled by a LabView program which moved the stage and triggered the OCT for a single slice. The scan parameters could be varied in order to scan a large area of the swarm plate to see many mounds or centered on one particular mound. The distance between slices was usually . The Imalux Imaging system can be set so that the detected signal for a particular bin is averaged over multiple cycles. This is analogous to a longer dwell time per pixel in Laser scanning microscopy. Averaging was typically done for 20 cycles. There is a trade off between resolution and scan period. The depth of scan also affects the scan period.
In order to analyze the 3D OCT intensity scan, the 2D slices are loaded into Matlab as a 3D matrix. The raw data is a Red-Green-Blue (RGB)-value image that is converted to an 8-bit grayscale image with an intensity range between 0 and 255. For the OCT scans, the largest intensity values observed were between 180 and 190. Towards the perimeter of each cross section, the values drop to below 10 corresponding to the surface of the mound. There are interior regions where the intensity values reach as low as 80. In-plane cross-sections were extracted from the 3D data by fixing the z-value to obtain a 2D image in the xy-plane parallel to the agar surface. For each in-plane cross section, the image moments and central image moments are given bywhere is the grayscale intensity data for the cross section that has (columnrow) pixels. The centroid for the cross-section is given by . From the centroid, the second central image moments can be calculated as , , and . The covariance matrix for the cross section is given by . Finally, the eigenvalues and eigenvectors for are used to calculate the orientation, eccentricity, and major and minor axes of the in-plane cross section of a mound.
The 3D renderings of the mound are made using the isosurface function in Matlab. The outer shell is an isosurface using an isovalue of 10 and given a large transparency. For the multi-layer isovolume rendering, the highest isovalues which are largely in the interior were rendered with higher opacity. Subsequent lower isovalues were drawn with decreasing opacity so that the internal structure could be visualized.
To study how the motion of cells and cell-contact signaling within a developing fruiting body could give rise to the patterns characterized by dense pockets showing up as a kind of bumpiness in the OCT, we use computational models that captures the movement of cells in a fruiting body environment. Previously, a 3D Lattice Gas Cellular Automata model was used to study cell aggregation and fruiting body formation as well as spore transport and spatial organization [16], [17].
In both models, we begin simulations with cells in an aggregate and accumulating C-signal. While the vast majority of cells die during the fruiting body process, as evidenced by the fact that 1.0% or less become spores [9], a key importance of cell death is that nutrients are made available to the surviving cells. How the cells die is less pertinent to the current study. The nutrients that come from the dead cells are what allow cells to continue moving and C-signaling in order to reach the level needed for sporulation. We make the assumption that the cells in the models have sufficient energy to maintain their movement. While cell death is not explicit in the model, by enforcing that cells continually move in the aggregate we assume that a source of energy is available. Without enough energy to move, the lack of motility would prevent cells from being able to C-signal [7]. In reality, these energy levels would be maintained by nutrients from the cells in the aggregate that lack sufficient energy and lyse. This approach is used to specifically study how the coordinated cell movement and C-signaling can give rise to the spatial patterns of spore clusters.
The use of the OCT method to scan fruiting bodies was expected to accurately reveal the internal structure due to the improved scanning depth of IR light (technical details for this reasoning are made in the Materials and Methods subsection ‘Comparison of OCT Method with SEM and LSCM’). In fact, we suspect the core of the fruiting body mound cannot be probed with visible light. Evidence for this is seen in the dark appearance of mounds in bright field images (see figure 3). It was discovered that microscopy images of mound structure could be made by using the IR probe as a transmitted light source for the inverted microscope. We centered the infrared light from the probe over a mound to obtain the images which revealed details not visible in the bright field images of mounds. A side by side comparison of bright field images using optical light and IR light can be seen in figure 3. While optical light does not transmit through the mound, the IR light passes through the mound and reveals contours and structure not seen in the bright field images. The IR transmitted light image showed structure that is similar to the structure we find in the 3D renderings of the OCT tomograms (see ‘Large scale inhomogeneous internal structure of fruiting bodies’ below).
Individual slices made by the OCT can be seen in figure 4. Early scans suffered from light reflecting back to the probe from the top of the mound. This reflection is problematic because the probe detects an increased signal directly above the mound. Additionally, the reflected light cuts down on the amount of light moving into the mound which reduces the noise to signal ratio. It was also found that dry scans also suffered from a lensing effect due to the change of index of refraction from air to mound. This lensing effect resulted in the OCT instrument detecting higher levels of backscattering underneath the mounds below the surface of the agar. (See bright area below the mounds in figure 4A and C).
To improve the quality of image, we place a drop of microscopy oil on the surface of the agar plate and submerged the probe head into the oil. Figure 4 demonstrates the difference between imaging in air and in oil. We also performed tests with glycerol that showed similar improvement. However, because the oil is immiscible with the water in the agar and does not evaporate, it provided a better submersion media for the scans. The submerged scans produced a better contrast for signal to noise within the mound and cut down on the reflected light. It has been shown that the refractive index for bacteria is [21]. This explains why microscopy oil, with a refractive index of 1.53, is an ideal submersion media for the bacterial mounds. While scanning in oil, the OCT device corrected for the index of refraction by adjusting the number of pixels used in a scan. This required rescaling the image in the vertical direction to recover the 1∶1 aspect ratio of the scan. Finally, a ratio of 3.3 microns/pixel was adopted.
The detailed analysis of an OCT scan of a fruiting body mound was carried out to study its internal structure. The image analysis (described in Materials and Methods) was able to quantify both the internal structure and the external shape of the mound.
Scans show no indications of a shell and core structure that was suggested by the LSCM study [13]. Instead, they reveal a continuous inhomogeneous density structure containing intensity patches that could reflect variations in the spore concentrations. These domains can be seen in both the three-dimensional renderings (figure 5) as well as individual in-plane cross-sections (figure 6A). In these images, the optical density, measured by intensity, depends on the density of the scatters in the media, i.e. the concentration of spores. Hence, the intensity levels of the OCT scans are proportional to the density of the mound. Each in-plane cross section is analyzed as an elliptical domain whose major and minor axis are obtained from the covariance matrix. The linear decrease of the axes as the height increases is consistent with a cone-like mound (see figure 7). The average intensity was obtained for each cross-sectional elliptical domain as well as the radial density. Figure 6D shows an example of the intensity distribution for a domain as well as the mean value of the intensity for the cross-section. Radial density plots were obtained for each in-plane cross section by averaging the intensity values for all pixels with in an elliptical annulus (figures 6B and C). The standard deviation from the mean value provides a measure for the variation of the density within a particular ring of a particular domain. The radial density for multiple domains from the mound shown in figure 5 can be seen in figure 6E. The regions of reduced density may reflect cavities or lower spore concentrations, while regions of high density are suggestive of closely packed clusters of spores. The graphs in figure 6D show the highest density at the base of the mound and a consistent average density up the mound until it begins to taper off towards the top. The distribution of intensity values is shown for bottom 12 layers of the mound (i.e. the in-plane cross sections). The graph in figure 6E shows that the average radial density ranges between the values of 120 and 140 for distances up to approximately 15 pixels () before tapering off. The radial density plots for individual cross-sections reveal the variation that exists within a given elliptical ring (shown as error bars) as well as intensity variations moving radially outwards from the centroid. For the in-plane cross-section and , we observed the region of increased intensity of approximately 6 pixels () at the distance of 12 pixels.
In addition to the radial distribution, we performed measurements of the angular distribution of intensity. This was done by dividing the domain into sectors (i.e. pie slices) and averaging the intensity within each sector. The results for one cross-section are shown in figure 8. The zones of lower concentration are spread out over the domain and is characterized by peaks and valleys in 2D plots of the distribution (figure 8B) and a smooth undulation in the polar plot (figure 8C) of the distribution. This measurement is repeated in simulations and provides a metric for comparison.
The more striking features that are revealed by the 3D OCT scan are the large interconnected structure of caverns with in the mound. Movie S1 in supplemental material provides a more complete view of the internal structure. Scans show what appear to be regions on the scale of that are more dense than surrounding regions suggesting that large regions of highly packed clusters of spores are spread throughout the mound (see figures 5 and 6A). The regions surrounding these clusters contain material that is less optically dense than the clusters of spores. This could be regions of polysaccharide and extra-cellular material or simply a region where the spores concentration is reduced.
The findings from the experiments were used to contemplate the bigger picture of fruiting body formation, which is presented here. In the fruiting body process, a mound of constantly moving cells gives rise, after several days, to a mature, spore-filled fruiting body [22], [23] (see figure 1). Inasmuch as growing cells have little ATP or other energy reserve, starvation liberates a tiny minority – 1% or fewer of the rod-shaped cells – to cannibalize the other 99% of the developing cells to harvest enough metabolic energy for developmental protein synthesis and for keeping the cells in constant motion as they develop further towards becoming spores. Collisions between moving cells eventually raises the morphogenetic C-signal to a threshold level that is able to trigger differentiation of rod-shaped motile cells into spherical spores [8], [24]. However, before they sporulate, the developing cells aggregate by moving back and forth in a system of traveling waves that surrounds the swarm edge, a process that has been captured in a time-lapse movie [25]. Like the traveling waves assembled for fruiting body aggregation, very similar waves are observed when M. xanthus feeds and grows on E. coli prey cells [26], [27]. In fruiting body development, the vast majority of myxobacterial cells are being eaten by their few siblings that are destined to become spores. Although prey protein, nucleic acid, and lipid are consumed for their calories, the polysaccharides are indigestible. The myxobacterial lytic enzymes include no polysaccharide hydrolases except lysozyme [28]. Polysaccharide fibrils [29]–[31] survive in the waves as an extensive elastic meshwork that surrounds and bundles the cells as well as their undigested bits of cellular debris. The few prespore cells that survive are found to be moving on trails of their polysaccharide slime that also is not digested. Trails remain intact and provide a surface favorable for gliding [32].
Departing from the traveling waves, the surviving cells are seen to migrate to the outer edge of each wave crest where they become one of the small motile aggregates [25]. Initially the motile aggregates are spaced one wavelength apart, and waves are thus the first step in fruiting body aggregation. Next, pairs of adjacent small aggregates fuse with each other to form a larger spherical aggregate of the same cell density but twice the volume. The larger aggregates fuse repeatedly with their neighbors until all the motile aggregates have assembled in a single very large aggregate. The diameter of the final aggregate, which shows signs of cell movement inside [25], is constant from experiment to experiment and is characteristic of mature M. xanthus fruiting bodies. (See [33], for example.) In this way, spores are expected to be formed on debris-laden slime trails that are suspended within a spherical motile aggregate by polysaccharide fibrils. Each slime trail would be expected to trace a truncated arc within the motile aggregate because each cell reverses direction of its motion at regular intervals [19], [34].
Based on the foregoing description of slime trails in a dynamic motile aggregate, it follows that cells would be clustered and aligned on the many trails that would branch from each other. Since individual cells are eating each other as they move, they are also racing to be one of the predators that survive rather than one of the prey that expire. In such a race, long chains of rod-shaped cells, moving on the same trail, would break into shorter segments of fewer and fewer cells until only 1% — to take some definite number since the number depends on residual nutrient — of starting cells remain on the trails and able to transmit C-signal. When two counter-migrating cells on the same trail collide end-to-end, they exchange C-signal with each other. C-signal transmission continuously raises the signal level in each cells outer membrane through positive feedback and the Act system [8]. Eventually, positive feed-back raises the level of C-signal in each cell to the threshold required to differentiate a rod-shaped cell into a spherical, non-motile, dormant spore [8]. Depending on each cells unique history of C-signaling, individual cells will reach the threshold at different moments. Nevertheless, the closer two cells are found to each other on the same trail, the more correlated their time to reach threshold will be. When a rod-cell becomes a spore, it remains on its debris-laden slime trail, and each trail would form some arc within the aggregate mound. Because most cells are destroyed, there will be many trail arcs each of whose spores will have formed at the same time, while different arcs will have sporulated at different times. Finally, the slime trails collapse around their own cluster of spores. As this nascent fruiting body dries out, the polysaccharides will also lose water and the aggregate will shrink. Within the fruiting body, the spores are likely to be clustered in space on their own arc-shaped trail that collapses into a ball of spores and polysaccharides.
To test the hypothesis described in the previous section, we ran simulations with the two models we developed.
A novel imaging technique – infrared optical coherence tomography – revealed that hundreds of thousands of spores in a mature fruiting body of M. xanthus were not packed uniformly, as was surmised previously. Rather, the spores are found clustered in high density pockets, which are no larger than 25 m in diameter, that are separated from each other by domains that have reduced concentration of spores. Why should the fruiting bodies have numerous cavities with relatively fewer spores? Detailed analysis of the way that fruiting bodies form based on the experimental observation of dense pockets yielded a biological hypothesis of how the movement, alignment and C-signaling of self-propelled rod-shaped cells could coordinate the differentiation process (presented above in “Biological Hypothesis”). In what follows, we discuss the outcome of the computer simulations designed to test the mechanism of the fruiting body formation proposed in the biological hypothesis.
First, our 1D track model provided two possible explanations for the formation of the cavernous structures of the fruiting bodies. One explanation was that early sites of spore formation act as focal regions for spore clusters due to jamming of the motile rod-shaped cells that continued to move around the track. This explanation suggested that high levels of clustering could result from spores strongly inhibiting the motility of cells. The highest level of clustering was observed when cells had the smallest passing probability and no C-signal transfer. However, in simulations with higher passing probabilities, (i.e. motile cells were not strongly inhibited by spores), more clustering was seen when C-signal exchange by local cells was present than when only jamming was considered. Experimental movies from our previous study on cell-cell collisions [35] demonstrate the flexibility of myxo cells which would allow easy resolution of collisions with spores. This indicates that having a higher passing probability in the 1D model is more biologically realistic. A mechanism of only low passing probability when spores strongly inhibit cell motility but cells do not signal is not strongly supported by the experimental observations. Rather, the 1D model shows that C-signaling can increase the level of clustering in simulations with higher passing probabilities.
The 1D model simulations initially confirmed that contact-based C-signaling would generate spore clusters when the cell-spore interaction was not characterized by strong spatial jamming. These findings were the motivation for focusing on the movement and alignment of cells in a more detailed model. Thus, the 2D model was used which could account for the biological details such as cell-shape, movement, and alignment-dependent C-signaling. The 2D model simulations have shown how the patterns of spore clusters could be produced by cells moving, aligning and C-signaling to coordinate differentiation. In the simulations, spores begin to form within a disc as small clumps (see figure 10B). The reversals of cells within the disc cause them to move back and forth along specific trajectories or arcs within the fruiting body. Cells that spend time moving along the same trajectories in end-to-end alignment accumulate C-signal at similar rates. This leads to spores forming in clusters throughout the mound. The simulations we performed to test the hypothetical mechanism resulted in pattern formation consistent with the experimental data. (Compare figure 10E with figure 6A).
To summarize, we first formulated a hypothesis based upon the experimental observation of spore patterns in fruiting bodies. We hypothesized that pockets of dense regions of spores form because cell movement, alignment and signaling result in coordination of the cell differentiation. The 1D simulations demonstrated that cell-signaling was capable of regulating the level of clustering inside a fruiting body. The 2D model simulations determined what patterns of spore clustering would emerge from cells aligned movement along slime trails and C-signaling by the end-to-end contact. In addition, the movement and interaction of cells in the 2D model included cell-cell and cell-spore collisions as well as cell reversals that reinforced alignment within the aggregate. We found that the coordinated movement of cells — by way of self-propelled motion, slime trail following, cell-cell and cell-spore collisions, and cell reversals — can facilitate the contact-dependent signal accumulation that drives cell differentiation into spores.
The integration of novel experimental observations with computational simulations provided new insight into the mechanisms that could give rise to the structure with a pattern of dense spore pockets seen during fruiting body formation. This can be improved upon through use of newer OCT devices with better resolution and even applied to other biological systems of cell aggregation such as that seen in dictyostelids, social amoeba known to form multicellular aggregates observed as slugs under starvation conditions.
Understanding how cells can undergo differentiation under specific spatial patterning is important to biology in general. It is known that chemical signals and reaction-diffusion processes can lead to coordination of cell patterning and differentiation. In the fruiting body process, we have shown how this patterning and differentiation could arise in the absence of a diffusive signal.
|
10.1371/journal.pgen.1001030 | Tinkering Evolution of Post-Transcriptional RNA Regulons: Puf3p in Fungi as an Example | Genome-wide studies of post-transcriptional mRNA regulation in model organisms indicate a “post-transcriptional RNA regulon” model, in which a set of functionally related genes is regulated by mRNA–binding RNAs or proteins. One well-studied post-transcriptional regulon by Puf3p functions in mitochondrial biogenesis in budding yeast. The evolution of the Puf3p regulon remains unclear because previous studies have shown functional divergence of Puf3p regulon targets among yeast, fruit fly, and humans. By analyzing evolutionary patterns of Puf3p and its targeted genes in forty-two sequenced fungi, we demonstrated that, although the Puf3p regulon is conserved among all of the studied fungi, the dedicated regulation of mitochondrial biogenesis by Puf3p emerged only in the Saccharomycotina clade. Moreover, the evolution of the Puf3p regulon was coupled with evolution of codon usage bias in down-regulating expression of genes that function in mitochondria in yeast species after genome duplication. Our results provide a scenario for how evolution like a tinker exploits pre-existing materials of a conserved post-transcriptional regulon to regulate gene expression for novel functional roles.
| It is well known that the evolution of gene expression can account for significant phenotypic diversity among species. Gene expression is regulated at various levels. Many studies have demonstrated gene expression changes caused by mutations in transcription. Post-transcriptional regulation is also crucial for splicing, translation, localization, and degeneration of mRNAs in eukaryotes and, thus, is important in determining the abundance of gene expression. Changes in post-transcriptional regulons are also important in the evolution of gene expression. In this study we investigated the evolution of a particular post-transcriptional regulon, Puf3p, in fungal species. Our results illustrated an important evolutionary mode for evolution of post-transcriptional regulon, i.e., pre-existing materials of conserved post-transcriptional regulons can be recruited to regulate gene expression for novel functional roles.
| Evolution of gene expression may account for significant phenotypic diversity among species [1]–[6]. Gene expression is regulated at various levels. Many studies have demonstrated gene expression changes caused by mutations in transcription [7], [8]. Post-transcriptional regulation is also crucial for splicing, translation, localization and degeneration of mRNAs in eukaryotes, and thus is important in determining the abundance of gene expression [9], [10]. It is likely that changes in post-transcriptional regulons are also important in the evolution of gene expression [11].
The control of post-transcriptional regulation is mediated by regulons such as mRNA-binding proteins (RBP) or RNAs (e.g., microRNAs) which usually bind to elements in the 3′ untranslated regions (UTR) and determine the fate of their targeted mRNAs. Evolution of microRNA post-transcriptional regulons has been well studied due to recent improvement in understanding their functions. It was shown that novel microRNAs can turn over rapidly during evolution [12], and for those that are highly conserved over long evolutionary distances, their targets can change dramatically even within populations [13]. Studies on microRNAs have revealed interesting information on evolution of this particular type of post-transcriptional regulon, whereas evolution of RBP regulons remains poorly understood. Furthermore, RBP regulons play major post-transcriptional roles in budding yeast because the species lost the microRNA regulatory machine [14].
One of the best-characterized RBP families is PUF (Pumilio and FBF, FBF represents for fem-3 binding factor), which is conserved in a wide variety of eukaryotes from yeast to humans [15]–[19]. The PUF post-transcriptional regulon regulates diverse gene sets in various model organisms. For example, in the budding yeast, Saccharomyces cerevisiae, genes most commonly targeted by Puf3p are in the mitochondria and play essential roles in mitochondrial biogenesis [17], [20]. In the fruit fly, Drosophila melanogaster, Pumilio (a PUF protein), which binds with the same element as Puf3p in the budding yeast, is necessary for early embryogenesis and development of primordial germ cells [21], . Genome-wide identification of the Pumilio targets in fruit flies uncovered genes involved particularly in nucleotide metabolism, transcriptional regulation and synthesis of membrane proteins [18]. In humans, two paralogous PUF proteins (Pum1p and Pum2p), which interact with the microRNA system in post-transcriptional regulation, share the same binding-element with yeast Puf3p and bind to mRNAs from genes that function in transcriptional regulation and cell proliferation [19], [23].
Previous studies have reported that the binding site of Puf3p is highly conserved in sensu stricto yeasts [24]–[26]. Taking advantage of a large number of genomic sequences, in this study we investigated the evolution of the Puf3p post-transcriptional regulon in fungi. Our results show continuous steps of functional innovation in the Puf3p regulon despite its ultra conservation in these fungal species. First, the regulation of mitochondrial biogenesis by the Puf3p regulon originated in the Saccharomycotina subdivision; second, the Puf3p regulon was coupled with codon usage bias to modulate expression of genes that function in mitochondria in yeasts after whole genome duplication (WGD). Our work and reports from other labs show that mitochondria underwent significant functional changes during the origin of an efficient aerobic fermentation system in the yeast species that went through WGD [27]–[30]. This current report provides evidence suggesting that the Puf3p post-transcriptional regulon was involved in the evolution of this novel life history in yeasts.
The RNA-binding domain of Puf3p, called the PUF homology domain (PUF-HD), consists of eight repeated peptide motifs [31]–[33]. In order to study the evolution of the RNA-binding domain, orthologous genes of PUF3 were identified from the forty-two sequenced fungal species (Table S1). Domain alignments in SMART [34], [35] indicate that almost all Puf3p orthologs contain the eight repeated motifs even though some repeats are not highly conserved (Figure 1A). Puf3p in Lodderomyces elongisporus and Rhizopus oryzae lost one repeat which may have resulted from insufficient genome sequencing or assembly because both orthologs are located at the end of the assembled contigs. This result demonstrates that the binding domain of Puf3p is conserved among all studied fungi.
The evolutionary trajectory of the Puf3p-binding element in its target genes was further investigated. Because the 8nt-core motif of P3E is conserved from yeast to human, we used the 8nt-core P3E profiling in the budding yeast as a reference to identify all possible puf3p targeted genes that contain at least one P3E at their 3′ downstream sequences (Figure 1B) [19]. Because GC content is very low in the P3E, the genomic GC content would inevitably affect the frequency of P3E in each species. Indeed, as shown in Figure 1C, at the genome level, the number of genes with P3E is negatively correlated with the genomic GC content among the studied species.
In order to exclude the impact of genomic GC content on our results, we generated 10,000 random sequences for each species based on the average GC content of 1,000 bp downstream sequences of all the annotated genes in this species. Using the occurrence of P3E in these random sequences as background in each species, we estimated the probability of having the observed P3E frequency in the 3′ downstream sequences of all the annotated genes in the same species. The probability was calculated in each sliding window of 50 bp in the 3′ downstream sequences of each species. As shown in Figure 1D, all studied species exhibited significant enrichment of P3E in the first several sliding windows in the 3′ downstream regions. As the 3′ UTR in yeast is usually shorter than 250 bp [36], our results indicate that the enrichment of P3E in the studied fungi results from P3E conservation in the 3′ UTR sequences. When we used the GC contents in the 250 bp regions after the stop codon or in each 50 bp sliding window to calculate the background P3E motif distribution, similar enrichment of P3E motif in the 3′UTR regions are still observed in most fungi species (Figure S1).
Previous reports showed that Puf3p plays an essential role in mitochondrial biogenesis in S. cerevisiae [17], [37], [38]. This observation prompted us to investigate whether the functional profile of the Puf3p regulon is also conserved among fungi species. Accordingly, we identified all of orthologous genes between each studied fungal species and budding yeast. Genes in each species are classified into categories based on the sub-cellular localization of their orthologs in the budding yeast [39]. We then estimated the enrichment of genes with P3E in each localization category. As shown in Figure 2A, we discovered that all of the studied species in the Saccharomycotina subdivision show significant enrichment of P3E in genes that function in mitochondria. Other clades of the studied fungi did not have this pattern. Indeed, close to 50% of genes that function in mitochondria have P3E in the Saccharomycotina subdivision, which is significantly higher than that of species in other clades (Figure 2B, P-value = 6×10−22). A similar pattern was observed when a slightly different P3E motif profile [17] was used to calculate the motif frequency (Figure S2).
Because regulation of the mitochondrial translational machine is essential for mitochondrial biogenesis [40], we further investigated conservation of P3E among the orthologous genes with this particular function in the Saccharomycotina subdivision. Our results showed that ∼80% of genes with highly conserved P3E were involved in mitochondrial translation, whereas this number is only ∼4% for genes with little P3E conservation, indicating that the Puf3p regulation of genes that are involved in the mitochondrial translational machine is highly conserved in these Saccharomycotina species (Figure 2C).
Gene expression is regulated at multiple levels. Biased usage of preferred codons can result in enhanced accuracy and speed of protein synthesis in highly expressed genes [41], [42]. Previous studies reported that codon usage bias in mitochondrial genes is relaxed, possibly due to a relaxed function of the organelle with the origin of an efficient aerobic fermentation system in the fungal lineage with WGD [27], [28]. We predicted that Puf3p-regulated mitochondrial genes, due to their importance in mitochondrial biogenesis and functions, would experience more relaxation of codon usage bias than other mitochondrial genes in the post-WGD yeast species. In order to test this, we calculated the average codon bias adaptation index (CAI) for the mitochondrial genes, with and without P3E, for each species. As shown in Figure 3A, the mitochondrial genes with P3E in the post-WGD species show significantly smaller CAI than those genes in the fungal species that diverged from the common ancestor before the WGD event (student t-test, P = 3×10−8), whereas mitochondrial genes without P3E did not show such a pattern (Figure 3B, student t-test, P = 0.1).
Due to the importance of Puf3p regulation in mitochondrial gene degradation, we further investigated expression of its target mitochondrial genes under the fermentative condition. Using gene-expression profiling measured by microarray data [43], we discovered that significantly more mitochondrial genes with P3E were down-regulated in the fermentative medium (YPD) than those without P3E (Figure 4A, Fisher's exact test, P = 1.2×10−4). Furthermore, we found that mitochondrial genes with P3E tend to be co-regulated because the average correlation coefficients of gene expression among mitochondrial genes with P3E in different conditions is significantly higher than that of genes without P3E (Figure 4B, student t test, P = 0). Therefore Puf3p regulon plays an important role in regulating mitochondrial genes in different conditions.
PUF protein was first characterized in Drosophila as an mRNA-binding factor that recruits other proteins to inhibit the translation of the bound mRNA [21]. Subsequently, many studies in yeast revealed that the PUF family regulates specific mRNA degeneration by their RNA-binding domains [17], [31], [44], [45]. It was shown that the function of targeting mRNA for degeneration by Puf3p is much more efficient in 2% glucose (YPD, fermentative) than in 3% ethanol (YPE, non-fermentative) [17], [46]. Furthermore, it was shown that Puf3p is crucial for mitochondrial biogenesis and motility under non-fermentative conditions in budding yeast [43]. Saint-Georges and his colleagues reported that Puf3p can transfer its target mRNAs to the peripheral mitochondria in the non-fermentative growth medium [37]. The expression of PUF3 gene is significantly higher in yeast growing in YPE than in YPD (Figure 5A, P-value<0.05). We speculate that this is true because the positive regulation of mitochondrial biogenesis might not be as important for Puf3p in fermentative conditions as that in respiratory conditions: First, based on gene deletion data, the mitochondrial genes with P3E are significantly more important (having more severe growth defects after gene deletion) than those genes without P3E (P = 4×10−6) under non-fermentative conditions, but these two gene groups do not show obvious difference in deletion phenotype under fermentative conditions (Figure 5B). Second, severe growth defect after PUF3 gene deletion was observed in YPE, but not in YPD (Figure 5C). Therefore in non-fermentative condition Puf3p regulates both mitochondrial biogenesis and mRNA degradation, but in fermentative condition, it might only regulate mRNA degradation, albeit more efficiently in this condition. The expression difference of PUF3 in two growth conditions can also be explained by the fact that mitochondrial biogenesis in non-fermentative conditions is extremely important for yeast because the organism relies on respiration, and therefore mitochondria, to generate cellular energy in these conditions. In contrast, the function of mRNA degradation might not be as essential to the organism under fermentative conditions.
Although loss of a functional PUF3 gene shows a negligible effect on organism growth in YPD (Figure 5C), our results in this study indicate that Puf3p regulation of mitochondrial gene degradation might be important for yeast fermentative growth during evolution. After whole genome duplication, the post-WGD species (including budding yeast) evolved efficient fermentative ability [47]. Mitochondrial function became relaxed in these species [27], [28]. Most post-WGD yeast species can live even without a functional mitochondrial genome [47]. A large number of mitochondrial genes are down-regulated in yeast fermentative growth [48]. Degradation of mRNA by Puf3p may accelerate this gene-expression regulation process during environmental switches. Interestingly, our results showed that mitochondrial genes having P3E had significantly relaxed codon usage bias in the post-WGD species, which is not true for other mitochondrial genes.
Understanding the origin of genetic novelties is a challenging issue in evolutionary biology. One of the prominent models proposed by Francois Jacob for evolution of genetic novelties in gene regulatory network is tinkering evolution, in which evolution reorganizes pre-existing networks to produce novelties [6], [49]–[52]. Our results provide an interesting paradigm for the evolution of post-transcriptional regulons that is consistent with this model. The Puf3p post-transcriptional regulon might have changed significantly at least twice in fungal evolution. First, although it is conserved in all the studied fungal species, the dedicated function of the Puf3p post-transcriptional regulon in mitochondrial biogenesis independently evolved in the Saccharomycotina subdivision. Second, although the regulation of mitochondrial mRNAs by Puf3p is conserved across the Saccharomycotina subdivision, the Puf3p target genes evolved a reduced codon usage bias in the post-WGD species, which might be consistent with the functional relaxation of mitochondrial genes in the post-WGD species due to the emergence of their fermentative life-style during evolution.
It was shown that although some microRNAs are highly conserved, their target networks can change dramatically during evolution [13]. Our results, together with those from previous studies, indicate that post-transcriptional regulons by RBP and microRNAs might share similar evolutionary patterns: i.e., the interaction mechanisms between the regulators and their target genes are conserved, whereas the target network is plastic during evolution. As post-transcriptional regulation plays an important role in regulating gene expression, this evolutionary scenario involving post-transcriptional regulons could lead to significant gene-expression divergence among species.
Sequences for the forty-two sequenced fungal species were downloaded from the Fungal Comparative Genomics database [53] and the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/).
Using the InParanoid software package [54], orthologs between budding yeast and other fungal species were identified. The PUF3 orthologs in other species of fungi were identified manually based on the best alignment. Eight repeated motifs of PUF protein were detected by the SMART sequence analysis (http://smart.embl-heidelberg.de/).
Because the Puf3p binding motif in its target genes (P3E) is conserved between yeast and humans, based on the profile of P3E in budding yeast, we used a Perl script to detect the target locus of Puf3p by fixing all seven invariable sites and allowing flexibility in the fifth site (Figure 1B). For each species, we scanned 1,000 bp of DNA sequence downstream of all annotated genes to discern the occurrence of the P3E motif. The percentage of motif occurrences in each 50-bp window among all genes in each species was calculated. Multiple occurrences of the motif in the same sliding window were regarded as independent events. To see whether the motif occurrences in a species are different from random expectation, we calculated GC content of the 1,000bp downstream sequences for all genes in each species. Based on the observed GC content, 10,000 random sequences of 1,000bp were generated by a perl script and the occurrences of PUF3 motif were calculated. The significance of PUF3 motif occurrences in each sliding window was calculated by comparing its motif frequency against the frequencies of P3E in these random sequences by Fisher's exact test (more focused tests were also conducted in Figure S1). The Bonferroni correction was used to correct for multiple comparisons. Genes with P3E in their 250bp 3′ downstream regions were defined as the target of Puf3p.
The sub-localization information for genes in budding yeast was downloaded from the Saccharomyces Genome Database (http://www.yeastgenome.org/) [39]. Based on the identified orthologous genes between each studied fungal species and budding yeast, genes in each species are classified into different localization categories based on the sub-cellular localization of their orthologs in the budding yeast. The hypergeometric test was used to test the enrichment of genes with P3E in each localization category. The Bonferroni correction was used to correct for multiple comparisons.
According to the InParanoid results, we identified all the orthologous clusters that contained the known cytoplasmic ribosomal protein genes in S. cerevisiae, regardless of gene copy number in each species. The ribosomal protein genes in each species were used as references to calculate the codon adaptation index (CAI) value for each individual gene in the same species by CodonW (http://codonw.sourceforge.net/) [55]. In order to compare codon usage bias between different species, CAI values in each species were standardized so that the mean and standard deviation were 0 and 1, respectively.
Cells were grown in the YPD and YPE media until optical densities at 600 nm reaches 1. Total RNA was extracted using the Trizol protocol [56] and cDNA was synthesized using an Invitrogen kit (Cat. No.18080-051). Using the ACT1 gene as reference, the expressional levels of PUF3 in fermentative and non-fermentative conditions were measured by quantitative real-time PCR.
The rate of growth for deletion mutants were downloaded from [57]. Two-tails student t test was used to compare the average fitness contribution of mitochondrial genes with and without P3E in fermentative and non-fermentative growth conditions.
The deletion of the PUF3 gene was conducted in the BY4741 strain (Mata his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) by homologous recombination and ura- was used as the selection marker (the primer sequences for gene deletion are available upon request). The mutant and the wild type strains were grown overnight in YPD (2% glucose) and YPE (3% ethanol) media. Cells were then transferred into fresh media and grown until the optical density (600 nm) reached 0.2. Then 4ul of growth media were dotted onto the YPD and YPE plates with ten-fold dilutions. YPD and YPE plates were incubated at 30°C for 48 and 72h, respectively.
|
10.1371/journal.pcbi.1006228 | Reversing allosteric communication: From detecting allosteric sites to inducing and tuning targeted allosteric response | The omnipresence of allosteric regulation together with the fundamental role of structural dynamics in this phenomenon have initiated a great interest to the detection of regulatory exosites and design of corresponding effectors. However, despite a general consensus on the key role of dynamics most of the earlier efforts on the prediction of allosteric sites are heavily crippled by the static nature of the underlying methods, which are either structure-based approaches seeking for deep surface pockets typical for “traditional” orthosteric drugs or sequence-based techniques exploiting the conservation of protein sequences. Because of the critical role of global protein dynamics in allosteric signaling, we investigate the hypothesis of reversibility in allosteric communication, according to which allosteric sites can be detected via the perturbation of the functional sites. The reversibility is tested here using our structure-based perturbation model of allostery, which allows one to analyze the causality and energetics of allosteric communication. We validate the “reverse perturbation” hypothesis and its predictive power on a set of classical allosteric proteins, then, on the independent extended benchmark set. We also show that, in addition to known allosteric sites, the perturbation of the functional sites unravels rather extended protein regions, which can host latent regulatory exosites. These protein parts that are dynamically coupled with functional sites can also be used for inducing and tuning allosteric communication, and an exhaustive exploration of the per-residue contributions to allosteric effects can eventually lead to the optimal modulation of protein activity. The site-effector interactions necessary for a specific mode and level of allosteric communication can be fine-tuned by adjusting the site’s structure to an available effector molecule and by the design or selection of an appropriate ligand.
| Recent advances in the development of allosteric drugs allow one to fully appreciate the sheer power of allosteric effectors in the avoiding toxicity, receptor desensitization and modulatory rather than on/off mode of action, compared to the traditional orthosteric compounds. The detection of allosteric sites is one of the major challenges in the quest for allosteric drugs. This work proposes a “reverse perturbation” approach for identifying allosteric sites as a result of a perturbation applied to the functional ones. We show that according to the traditional Monod-Changeux-Jacob’s definition of allostery, considering non-overlapping regulatory and functional sites is a critical prerequisite for the successful detection of allosteric sites. Using the reverse perturbation method, it is possible to determine wide protein regions with a potential to induce an allosteric response and to adjust its strength. Further studies on inducing and fine-tuning of allosteric signalling seem to be of a great importance for efficient design of non-orthosteric ligands in the development of novel drugs.
| The traditional emphasis on complementarity between the drug and the catalytic site has inarguably formed a foundation in the current drug discovery approaches. However, many important drug targets share a conserved substrate binding site [1–5], rendering drug toxicity as a result of the off-target binding. The allosteric regulation of protein activity via effector binding has been increasingly favoured in the drug discovery [1, 2, 6]. It is well recognized that potentially druggable allosteric sites are ubiquitous in most if not all dynamic proteins [7], turning the key advantage of targeting allosteric sites [5]—non-competitive fine-tuning of protein activity at a distance—into a new paradigm in the drug design. For example, it was shown that allosteric drugs provide a way to modulating the activity of kinases that underlie a multitude of human diseases, bypassing the problem of low specificity with the conserved ATP binding pocket [8, 9]. The allosteric drugs for GPCRs provide greater subtype selectivity among GPCR receptor families, while avoiding receptor desensitization typical for the orthosteric ones [10–12]. Currently, among the notable marketed drugs targeting GPCRs are cinacalcet [13] and maraviroc [14], an allosteric agonist for calcium-sensing receptor and an antiretroviral allosteric antagonist for the CCR5 receptor, respectively.
One of the major hurdles in the development of allosteric drugs lies in the finding of allosteric sites [5, 15–17], for which a repertoire of experimental and computational methods is being developed. High-throughput fragment-based screening using a large chemical library formed the main thrust in the identification of potential allosteric sites and lead compounds in pharmaceutical research [18–20]. A number of site-directed approaches have been employed for detection and probing of allosteric sites and their modulatory effects, including disulfide trapping [21], alanine scanning [22], hydrogen-deuterium exchange mass spectrometry [23, 24] and photoaffinity [25, 26]. However, above experimental approaches while powerful, are relatively costly and time-consuming compared to any extensive analysis performed in silico. Computational approaches for finding the allosteric sites can be broadly classified as sequence-based and structure-based methods [3, 4, 27]. Sequence-based techniques utilize sequence homology inferred from the multiple sequence alignment to identify the co-evolving amino acids that constitute catalytic and allosteric sites [28]. However, complexity of the site-ligand interactions and energetics result in strong limitations on the predictive power of the sequence-based approaches [5, 28, 29]. Structure-based methods analyse binding pockets based on their topological and physicochemical features [5, 30]. These approaches are strongly biased towards binding pockets that exhibit detectable curvature in the static 3D structure, in which case latent allosteric sites that can exist in a subset of a protein conformational ensemble may be left undetected.
We have recently introduced a structure-based perturbation model of allostery [31], which quantitatively describes the causality and energetics of allosteric communication, by simulating ligand binding as a local alteration in the protein inter-residue network of interactions. Because of the observation that perturbation at allosteric sites can affect distant functional site via modifying the energetics of the whole protein and assuming the reversibility of allosteric signalling, we hypothesized here that allosteric sites could be detected by perturbing the functional ones [5]. In order to test this hypothesis and to explore its predictive power, the reverse perturbation approach was developed here. Using a heterogeneous set of 13 classical allosteric proteins from previous studies [31–33], dubbed here the “classical set”, we found that perturbation at the functional sites allows one to identify known allosteric ones. In order to estimate the predictive power of reverse perturbation method, it was necessary to introduce an operational definition of the allosteric site in the framework of elastic network model of protein. Specifically, assuming that allosteric signalling occurs between non-overlapping distant sites, a distance condition was set to ensure communication and not direct physical interaction between residues of the functional and allosteric sites. Using the classical set and a new collection of 41 allosteric proteins from the benchmarking set [34], predictive power of the reverse perturbation method is shown. Furthermore, we argue that in addition to the widely addressed case of predicting latent allosteric sites [35], the task of inducing allosteric signalling with a desired and tunable level of agonistic/antagonistic activity can be naturally formulated. We show that the reverse perturbation approach opens the way for achieving above goal, allowing one to find targets for allosteric effectors and to optimize structures/compositions and interactions of corresponding regulatory site-effector pairs that will provide a desired allosteric response at the functional site.
In the structure-based statistical mechanical model of allostery [31], ligand binding is modelled as a perturbation of the harmonic network associated with the protein. The perturbation is defined as a stiffening harmonic restraint applied to the residue pairs that compose the binding site of interest. As a result of the perturbation, residues in the binding site experience an increase of rigidity in comparison with the residues in the unperturbed binding site. We have shown that in the event of allosteric communication, the perturbation of the allosteric sites induces a response at the functional ones by altering their energetics and fluctuation dynamics [31]. Specifically, as a result of the perturbation in a binding site, a per-residue free energy change Δgi is obtained for each residue of the protein, which is the signature of the change in the amount of work exerted in the environment of residue i. Here we investigate the hypothesis of reversibility of allosteric communication, according to which one can detect a change in the free energy on the residues of allosteric sites when a perturbation–simulated binding–is applied in the functional sites (Fig 1).
To begin with, we analysed a set of classical allosteric proteins, dubbed here “classical set” [32, 33], to directly test the hypothesis of the reverse perturbation approach in identifying allosteric sites in proteins of different sizes, oligomerization states and functions. Table 1 contains the average free energy changes (averaged over all residues in corresponding sites in case of oligomeric proteins) observed as a result of the allosteric signalling in known allosteric sites (ΔgA) and in the restrained functional sites (ΔgF) in 13 proteins of the classical set.
First column: Protein name, oligomerization state, and total number of residues. Second column: PDB ID of the protein. The third and fourth columns represent the perturbation applied to functional sites and the resulted ΔgF values, respectively (see Eq 7 in Materials and Methods). The fifth and sixth columns show the known allosteric sites (A—allosteric activator, I—allosteric inhibitor) and the ΔgA values in the allosteric sites in response to the reverse perturbation, respectively. The seventh column provides the average free energy differences ΔgU over all the residues in the protein, giving an estimate on the protein stability changes caused by the applied perturbation. The last column gives the proximity between the corresponding functional and allosteric sites in a protein subunit (see Eq 8 in Materials and Methods).
The Anthranilate synthase (AnthS) from Serratia marcescens is a heterotetramer consisting of a dimer of TrpE and TrpG subunits. Upon perturbation of the glutamine substrate binding sites, the tryptophan (inhibitor) binding sites in the larger TrpE subunits show a positive free energy difference response (ΔgTRP(2×GLU) = 0.42 kcal/mol), compared to the overall decrease in the free energy of the structure and stabilization of the entire TrpG subunits upon perturbation (ΔgAnthS(2×GLU) = -0.29 kcal/mol, Fig 2A and Table 1).
The aspartate carbamoyltransferase (ATCase, Fig 2B) from Escherichia coli is a heterododecameric enzyme composed of two trimers of catalytic subunits in the centre of the oligomer and three dimers of the peripheral regulatory subunits. Simulated binding in the catalytic sites of ATCase (PAL sites) increases the configurational work exerted at the allosteric domain in the regulatory subunits that contain allosteric activator ATP and inhibitor CTP (ΔgATP−CTP(6×PAL) = 1.75 kcal/mol, Table 1), but not the zinc-binding domain which plays a structural role in the complex assembly.
Allosterically activated by cyclic AMP, the catabolite activator protein (CAP) from E. coli is a classical model system of transcriptional activation. The cAMP binding causes large rotation in the DNA-binding domain of CAP, which is a prerequisite for interactions with DNA [36]. Negative cooperativity observed for the binding of second cAMP molecule was discussed earlier [31], and it was shown that mutations in the cAMP-binding pocket decreasing the affinity between CAP and cAMP enhance negative cooperativity [37]. Using the DNA-bound conformation of CAP, we simulated binding in the DNA interaction sites and observed that allosteric site in the N-terminal cAMP-binding domains exhibits a large positive free energy change (ΔgcAMP(2×DNA) = 1.73 kcal/mol), higher than the average free energy change (ΔgCAP(2×DNA) = 0.71 kcal/mol) of the homodimer (Fig 2C and Table 1).
For the homotetrameric 3-deoxy-D-arabino-heptulosonate-7-phosphate synthase (DAHPS) from E.coli, binding of the phosphoenolpyruvate (PEP) substrate to the TIM barrel domain was simulated. We observed a positive free energy change (ΔgPHE(4×PEP) = 0.80 kcal/mol) in the center of the quaternary complex, where the inter-subunit pockets for allosteric inhibitor phenylalanine (PHE) are located, while the whole structure yields rather a modest positive free energy change (ΔgDAHPS(4×PHE) = 0.15 kcal/mol, Fig 2D and Table 1).
The dimeric arginine kinase (DAK) from Apostichopus japonicus is known to exhibit negative cooperativity, in which binding of substrates (ARG and ATP) to one subunit causes a conformational change in the free subunit. Specifically, “large outward reorganization of the other subunit (open state) which result in the release of products”, that precludes substrate binding was experimentally observed [38]. We found that perturbing the catalytic sites in one subunit destabilizes the unperturbed one (ΔgARG(1×ARG/ATP) = 1.01 kcal/mol and ΔgATP(1×ARG/ATP) = 0.48 kcal/mol, Fig 3A and Table 1), presumably reducing its affinity to substrates.
The NAD binding sites in the Rossmann fold were perturbed in the human NAD-dependent malic enzyme (NADME). Fig 3B shows that configurational work propagates from the outer domains of the homotetramer to the dimerization interface where the allosteric activator fumarate (FUM) binds, increasing the amount of work exerted in these sites (ΔgFUM(4×NAD) = 2.90 kcal/mol) to a much larger extent than the overall free energy change (ΔgNADME(4×NAD) = 0.67 kcal/mol, Fig 3B and Table 1). Noteworthy, latent allosteric sites could also exist in the extended area near the core of this quaternary complex where large positive configurational work is observed as a result of perturbation at the functional sites.
A classical allosteric enzyme with a long history of studies, the phosphofructokinase (PFK) from Bacillus stearothermophilus, is a homotetramer. When the substrate and cofactor binding sites for fructose-6-phosphate (F6P) and ATP were perturbed, the allosteric response at the regulatory exosites (that bind activator ADP and inhibitor phosphoenolpyruvate (PEP)) located in the dimerization interface is manifested in the increase of the free energy (ΔgADP(4×F6P/ATP) = 1.35 kcal/mol and ΔgPEP(4×F6P/ATP) = 1.47 kcal/mol) detected in these sites, respectively (Fig 3C and Table 1).
The D-3-phosphoglycerate dehydrogenase (PGDH) of E. coli is a homotetramer with a ring-shaped quaternary structure. Simulated binding at the substrate AKG and cofactor NAD sites, which are located in the interface between corresponding substrate and cofactor domains, allows one to identify the binding sites for allosteric inhibitor serine (SER) in the distant peripheral regulatory domains. The serine binding sites show a large positive free energy change (ΔgSER(4×AKG/NAD) = 2.62 kcal/mol) in comparison with the negligible background free energy increase (Fig 3D and Table 1).
The overall change in the free energy of the small monomeric human protein tyrosine phosphatase 1B (PTP1B) is negative upon restraining the catalytic site BPM (ΔgPTP1B(1×BPM) = -2.12 kcal/mol), pointing to the strong stabilizing role of this perturbation. We observed a slight increase of the free energy at the known allosteric site 892 (Δg892(1×BPM) = 0.42 kcal/mol, Fig 4A and Table 1). Additionally, the beta sheet distant from the catalytic site exhibited large increase in the free energy change, suggesting the presence of a potential latent allosteric site (Fig 4A).
The uracil phosphoribosyltransferase from Sulfolobus solfataricus (SSUPRT) is a homotetramer that catalyzes the production of uridine 5′-monophosphate (UMP). As a result of perturbing the catalytic sites UMP in the outer regions (Fig 4B), residues in the binding site for the allosteric inhibitor CTP at the central interface show the positive free energy change (ΔgCTP(4×UMP) = 2.05 kcal/mol, Table 1) compared to the overall small average free energy change (ΔgSSUPRT(4×UMP) = -0.05 kcal/mol).
The homodimer of threonine synthase (ThrS) from Arabidopsis thaliana features an extensive interface between the subunits, in which the allosteric activator S-adenosylmethionine (SAM) binds. Similar to SSUPRT, when the protein’s catalytic sites pyridoxal-L-phosphate PLP are stabilized, the configurational work exerted on the binding sites for SAM increases (ΔgSAM(2×PLP) = 2.27 kcal/mol, Table 1) in contrast to the small average free energy change (ΔgThrS(2×PLP) = 0.05 kcal/mol, Fig 4C).
The bovine glutamate dehydrogenase (BGDH) and the glucosamine-6-phosphate deaminase (G6PD) from E.coli, are large homohexamers consisting of a dimer of trimers, and a trimer of dimers, respectively. The well-studied allosteric regulation of BGDH features a repertoire of metabolites as well as several characterized allosteric sites [39, 40]. It is known that binding of glutamate (GLU) substrate and NADP+ (NDP) cofactor to the catalytic cleft elicits large-scale conformational changes in this complex molecular machine. For example, the antenna region (Fig 5A), which consists of intertwined alpha helices and is evolutionarily conserved in the animal kingdom, is essential to the allostery of BGDH. The antenna serves as the inter-subunit relay of allostery, and its motion is regulated by the allosteric activator ADP and inhibitor GTP. Remarkably, we observed the large increase in the free energy of allosteric response (5 kcal/mol) in the antenna region upon perturbation of the catalytic sites, in comparison to the small background free energy change (ΔgBGDH(6×GLU/NDP) = 0.31 kcal/mol, Fig 5A and Table 1). On the other hand, a weak response is detected in the ADP and GTP binding sites (ΔgADP(6×GLU/NDP) = 0.18 kcal/mol and ΔgGTP(6×GLU/NDP) = -0.18 kcal/mol, Table 1). The configurational work exerted in the antenna region is consistent with experimental data, which shows rotation of the helices as the bound catalytic cleft closes [39, 40]. We found that, in agreement with experimental data in which the presence of antenna region was shown to be required for allosteric signalling, this region plays a role of the allosteric modulator, suggesting, in turn, the presence of latent allosteric sites in this region (Fig 5A).
In the case of homohexameric G6PD (Fig 5B), the allosteric response results in the positive free energy change globally distributed in the complex upon perturbation in all catalytic sites (alpha-D-glucosamine 6-phosphate (AGP), ΔgG6PD(6×AGP) = 0.66 kcal/mol), including the inter-subunit interface where the allosteric activator N-acetylglucosamine 6-phosphate (16G) binds (Δg16G(6×AGP) = 0.78 kcal/mol, Table 1). In particular, residues at the core appeared to be highly affected by the perturbation, especially Cys219, which forms disulphide bridges between subunits. Perturbing the catalytic site of one subunit essentially freezes the entire subunit including the proximal allosteric site for the activator 16G, increasing the free energy in the remaining unperturbed subunits (S1 Fig). It was shown in X-ray crystallography [41] and fluorescence spectroscopy [42] experiments that the substrate binding to the catalytic site of G6PD induces structural and dynamic changes in all subunits of the protein, consistent with our observations.
The classical set of proteins analysed here contains an equal proportion of protein structures in the apo and bound forms. Using crystal structures of PFK and PGDH with or without bound ligands, we show that the free energy profiles are largely similar (S2 and S3 Figs), allowing to work with only one available structure. In all proteins of the classical set, a positive free energy difference was detected in distant allosteric sites upon perturbation of the substrate and/or cofactor binding, as well as an anticipated negative one in the regions where the perturbation is applied. Despite the different modes of allosteric regulation, binding sites for both allosteric activator and inhibitor exhibit an increase of configurational work exerted upon perturbation of the functional ones. For example, in enzymes with overlapping binding sites for activator and inhibitor, such as the ATCase (Fig 2B) and the PFK (Fig 3C), both sites show positive Δgi values. Therefore, the positive free energy difference as a result of the functional site perturbation serves in our approach as the standard quantitative indicator for the allosteric sites.
The profiles of the free energy changes due to the reverse allosteric signalling upon perturbation at the functional sites typically show rather extensive areas of the positive free energy difference, which encompass the binding sites of known allosteric effectors. Therefore, the locations that yield a large free energy change can, actually, be allosterically coupled to corresponding catalytic site, and, therefore, can potentially contain unknown/latent allosteric sites. For example, the reverse perturbation analysis suggests the presence of latent allosteric sites in the large TrpE subunit of AnthS (Fig 2A), in the cAMP-binding domain of CAP (Fig 2C), in PTP1B (Fig 4A), as well as in the dimerization interface in NADME (Fig 3B), PFK (Fig 3C) and ThrS (Fig 4C). Furthermore, the antenna region in BGDH (Fig 5A) and multiple locations in G6PD (Fig 5B) can also contain latent allosteric sites that provide complex allosteric regulation of these large protein complexes.
Fig 5 shows two cases, BGDH and G6PD, in which it was not possible to identify known regulatory sites proximal to the perturbed catalytic ones. In these proteins, the perturbation of the catalytic site stabilizes the region proximal to the binding sites due to the direct interaction of residues in these adjacent sites. These cases show that the separation of allosteric and the regulated functional sites is instrumental in allosteric communication, as postulated in the seminal Monod-Changeux-Jacob paper [43] that allosteric proteins should “…possess two, or at least two, stereospecifically different, non-overlapping receptor sites.” Therefore, for any high-throughput analysis and prediction of regulatory exosites, it is crucial to introduce a quantitative measure for spatial separation between catalytic and allosteric sites.
Here, we have devised an operational definition of the allosteric site in the framework of the elastic network model of protein. To obtain a quantitative criterion for defining the allosteric site we introduce a notion of proximity, which is the fraction of interacting residues among all possible pairs that can be formed between residues of two sites–functional and the candidate allosteric ones. The distance cutoff for defining the interaction between residues is chosen according to the distance cutoff (11 Å) used in the microscopic allosteric potential of the original model [31] (see also Materials and Methods). Taking the most conservative approach, we have chosen the upper limit cutoff 11Å, which allows one to predict most of the allosteric sites. A site is considered allosteric if the corresponding proximity with the functional site does not exceed the threshold value (selection of the threshold value is explained below). We illustrate this definition by two homologous homotetramers, fructose 1,6-bisphosphatase 1 (FBPase 1) from E. coli (PDB ID: 2q8m, Fig 6A) and from Sus scrofa (PDB ID: 1kz8, Fig 6B), which control the gluconeogenesis pathway [44].
Adenosine monophosphate (AMP) and glucose-6-phosphate (BG6) bind to two distinct allosteric sites of FBPase 1 from E. coli, inhibiting hydrolysis of fructose 1,6-bisphosphate (FBP) [45]. The inhibitor binding at BG6 site would perturb the adjacent active FBP site via direct interaction between residues in these overlapping sites, whereas the distant AMP site regulates the FBP site by long-range allosteric signalling. Therefore, using the reverse perturbation approach, the distant AMP site can be readily identified based on the increase in free energy of allosteric response (ΔgAMP(4×FBP) = 1.67 kcal/mol). However, simulated binding in the FBP site (ΔgFBP(4×FBP) = -0.77 kcal/mol) decreases the free energy of the adjacent BG6 site (ΔgBG6(4×FBP) = -1.62 kcal/mol), which is not a true allosteric site because of the 10% proximity to the FBP site (Fig 6A). Non-allosteric nature of interactions between the BG6 and FBP sites can be further confirmed by the weak decrease of the free energy observed upon direct perturbation via simulated binding to BG6 (ΔgFBP(2×BG6) = -0.1 kcal/mol). On the contrary, the direct perturbation at the distant AMP (inhibitor) binding site increases the configurational work exerted at the FBP site (ΔgFBP(4×AMP) = 0.44 kcal/mol), yielding the allosteric nature of communication between these sites.
The drug screening against FBPase 1 from Sus scrofa revealed that one of the anilinoquinazoline compounds, PFE, inhibits gluconeogenesis by binding to an allosteric site at the subunit interface [46]. Upon perturbing the catalytic FBP site, both AMP and PFE sites (with 0 and 2% proximity to the FBP site, respectively), displayed a similar increase in the free energy (ΔgAMP(4×FBP) = 1.36 kcal/mol and ΔgPFE(4×FBP) = 1.45 kcal/mol, Fig 6B). The configurational work exerted at the PFE site shows that the low proximity between functional and allosteric sites is a necessary condition for the correct definition of the latter. Additional analysis of several proteins that result in both successes and failures in detection of allosteric sites have led to the following operational definition: a site is considered allosterically coupled to the regulated catalytic one if the proximity between them is no more than 2%. The sites’ proximities obtained for each protein in the classical set (Table 1) show that all proteins have non-overlapping functional and allosteric sites except the BGDH and G6PD, where the proximities are 7 and 10%, respectively, hence failed to be detected as allosteric ones. This operational definition is important in order to correctly estimate the predictive power of reverse perturbation approach, allowing us to operate with proteins with different architectures and mutual locations of the functional and allosteric sites.
With the set of allosteric sites obtained on the basis of above operational definition, one can analyse the predictive power of the reverse perturbation approach. A successful prediction of allosteric sites is possible if their residues will exhibit a large free energy change upon perturbation of the functional site. The receiver operating characteristic (ROC) curve is used here to quantify the proportion of true positive (those that belong to a known allosteric site) and false positive (those not belonging to a known allosteric site) among residues with a large free energy change, upon perturbation at the corresponding functional sites (see Materials and Methods for details). Using the classical set of allosteric proteins, we show that the true positive rate increases more rapidly than the false positive rate for most allosteric sites, indicating that the majority of residues of known allosteric sites (true positives) are indeed located near the positive tail of the Δgi distribution (Fig 7A).
The known allosteric sites in the NADME (Fig 3B), PGDH (Fig 3D) and SSUPRT (Fig 4B) can be precisely determined based on the large increase in the free energy. It is important to emphasize, however, that presence of latent allosteric sites challenges the task of validating the prediction of the known ones, as some of the residues belonging to these sites can be erroneously scored as false positives. For example, the residues that show a large increase in Δgi values in potential allosteric sites, such as the antenna region in the BGDH, turn out to be false positives. Good predictive power can be achieved for known allosteric sites in the ATCase (Fig 2B), DAHPS (Fig 2D), DAK (Fig 3A), PFK (Fig 3C) and ThrS (Fig 4C). However, for the AnthS (Fig 2A), CAP (Fig 2C) and PTP1B (Fig 4A), the ROC curves are close to the diagonal line, indicating low predictive power for the known allosteric sites in these proteins (Fig 7A). This is likely due to the pronounced increase in free energy over the protein domains caused by catalytic site perturbation, suggesting the presence of latent allosteric sites that contribute to the false positives. For DAK (Fig 3A), the ROC curves for ARG (Fig 7A) and ATP binding sites in the free subunit (Fig 7B) vary greatly as the former site exhibits a higher increase in free energy, allowing its detection as the allosteric site. In the BGDH (Fig 5A) and G6PD (Fig 5B), the regulatory exosites are too close to the restrained catalytic site, hence, not satisfying the aforementioned operational definition (Fig 6B).
To complement the classical set of 13 allosteric proteins, we have used the operational definition of allosteric sites to obtain 41 proteins with 48 unique experimentally-determined allosteric sites from the benchmarking set of allosteric proteins [34] (S1 Table). We calculated the free energy profile for each of these proteins upon simulated ligand binding at the functional site (S4 Fig). Similar to the classical set, the predictive power of the reverse perturbation approach was estimated for each protein in this heterogeneous additional set. The area under the ROC curves (AUC) indicates good predictive power of the method (Fig 7C), showing that reverse perturbation approach allows one to successfully detect the majority of known allosteric sites in the additional set.
In addition to the detection of known allosteric sites, the reverse perturbation approach delineates extended protein regions which are also characterized by the positive free energy change. This observation along with multiple indications of the presence of latent allosteric sites [5] in different proteins show that allosteric response can also be artificially induced by the interactions with rationally selected sets of residues belonging to latent or de novo designated allosteric sites. A classical allosteric enzyme, PFK (Fig 3C), which is regulated by the activator ADP and inhibitor PEP binding to the overlapping binding sites, serves as an excellent illustration that large difference in allosteric response can be caused by minor changes in the composition of an allosteric site (Fig 8). Specifically, ligand binding to the PEP site, which is a part of the larger ADP binding site results in a mild increase of free energy in the functional F6P site (ΔgF6P(4×PEP) = 0.30 kcal/mol) compared to the global free energy change of the homotetramer (ΔgPFK(4×PEP) = 0.10 kcal/mol). At the same time, a large increase in free energy is observed at F6P site upon simulated binding to ADP site (ΔgF6P(4×ADP) = 0.70 kcal/mol, Fig 8). The modes of regulation in PFK, along with many other allosteric enzymes with overlapped activator and inhibitor binding sites in the literature [5] (see also, for example, ATCase, Fig 2B), highlights the importance of tuning allosteric response by varying the binding site composition.
We consider here as case studies, the subunit interfaces of the NADME (Fig 3B), ThrS (Fig 4C) and FBPase 1 (Fig 6A), which can apparently be used to induce a required allosteric response. In our model, an allosteric site consists of a set of residues whose perturbation results in the large free energy change in the corresponding functional site. In a given protein, to obtain a magnitude of regulation comparable with that of the native effectors, the allosteric response at the functional sites originating from the newly designated regulatory sites should be comparable with that obtained from known allosteric sites.
The homotetrameric NADME is allosterically activated by the binding of fumarate (FUM) at the dimerization interface, causing a slight decrease in free energy at the catalytic site (ΔgNAD(4×FUM) = -0.18 kcal/mol). Based on the reverse perturbation analysis in NADME (Fig 3B) and in order to illustrate the possibility to induce an allosteric response, four sites were putatively defined in the protein region corresponding to the increase of the free energy: site 1 (red) and site 2 (green) are located at the dimerization interface where fumarate binds, site 3 (yellow) is situated at the tetramerization interface where ATP binds to stabilize the functional tetrameric form of the enzyme [47], site 4 (cyan) is close to the tetramer’s core, and site 5 (magenta), which is located in the protein region with negligible free energy change is used as a negative control (Fig 9).
We show that a range of allosteric responses can be induced upon simulated binding to sites 1–4 for different modulation of the catalytic activity. Perturbation of site 1 induces a stronger decrease of the free energy in the catalytic site (ΔgNAD(4×site1) = -0.80 kcal/mol) than that of the native FUM allosteric site (-0.18 kcal/mol). Similar to the effect of the FUM–site binding, perturbation of site 2 (ΔgNAD(4×site2) = -0.17 kcal/mol) and the site 3 (ΔgNAD(4×site3) = -0.12 kcal/mol) causes a slight decrease in the free energy in the catalytic site. On the other hand, simulated binding to site 4 strongly increases the free energy in the functional site (ΔgNAD(4×site4) = 0.86 kcal/mol). The reverse perturbation analysis has also detected areas that are not allosterically coupled to the catalytic site, indeed simulated binding in site 5 (chosen as a negative control) induces only weak response in the catalytic site (ΔgNAD(4×site5) = 0.05 kcal/mol). The generic nature of inducing allosteric response on the basis of the reverse perturbation approach is further illustrated with examples of allosteric signalling of different modes and magnitudes obtained in the cases of FBPase 1 (S5 Fig) and ThrS (S6 Fig).
Above analysis shows that in addition to the quest for latent allosteric sites, a more general question about the inducing of desired allosteric response can be formulated. Using site 1 of NAD-dependent malic enzyme as a test case, we show that the response can be further fine-tuned to achieve the necessary modulation in the catalytic site by varying the site’s composition (Fig 10).
Restricting perturbation of site 1 (blue) to a subset of residues (Leu118, Ala119, Gln122) substantially weakens the allosteric response (from -0.80 kcal/mol to -0.14 kcal/mol, Fig 10). Replacing Leu118 with the closest neighbours Gly117 and Ser121 recovers stronger allosteric signal (-0.67 kcal/mol and -1.06 kcal/mol, respectively, S7 Fig). Alternatively, considering the subset of residues Gln122, His125 and Ile126 (red) which produces an allosteric effect, which is comparable (-0.82 kcal/mol) to that induced by perturbing the whole site 1 (Fig 10). Allosteric signalling can also be fine-tuned to be weaker by replacing Gln122 with Ala119, Cys120 or Ser121, resulting in the allosteric responses -0.21 kcal/mol, -0.21 kcal/mol and -0.29 kcal/mol, respectively, at the regulated functional site (S7 Fig). Sites 1–5 are well separated from the distant NAD site with less than 2% proximity, which ensures the allosteric nature of signalling from these putatively designated sites.
Inducing and fine-tuning the allosteric response should be further complemented by the rational design or selection of allosteric effectors. For example, minimal sets of residues as the basis for targeted allosteric response (e.g. blue and red sets, Fig 10 (left, bottom) and S7 Fig) can be incrementally changed residue-by-residue (Fig 10) in order to achieve required strength of the allosteric signal. Alternatively, one can explore the repertoire of all possible binding sites, adjusting them for a given lead ligand (combined red-blue sets, Fig 10, right).
Allostery is a universal phenomenon where ligand-binding at a regulatory site causes a change in the ligand-affinity and/or activity at the coupled functional site [43]. Serious attempts to utilize the advantages of targeting allosteric sites instead of the orthosteric ones have only started in recent years, and the concept of allosteric drugs has since formed an important part in drug discovery [1, 2, 5, 6]. Prediction of allosteric sites that can remotely regulate the dynamics at the functional site of interest has been shown to be a challenging task [5, 15, 16]. In this paper we test the hypothesis of the reversibility of allosteric communication, according to which the perturbation at the functional site results in a signal that propagates towards allosterically active protein regions. We show that in most of the cases reverse perturbation at functional sites causes an increase of the free energy in protein regions that are dynamically coupled to them, which were subsequently used for the detection of allosteric sites. In general, the topology of the protein plays a non-trivial role in the propagation of the allosteric signal, especially when protein activity is regulated by more than one effectors. Using the protein set consisting of 13 classical allosteric proteins and an independent benchmark set of 41 proteins, we show that known allosteric sites can indeed be identified by the reverse perturbation method. Good predictive power of the method was obtained in all proteins of both classical and benchmark sets when the allosteric sites are spatially separated from the functional ones according to the operational definition of allosteric site, which, in turn, is based on the original Monod-Changeux-Jacob’s formulation [43] of allostery and relationship between the functional and the regulatory exosites.
After showing that known allosteric sites can be detected via the reverse perturbation method, we addressed the question of the allosteric sites identification. Using the NAD-dependent malic enzyme as an example, we show that simulated binding to two putatively defined sites in the protein region obtained from the reverse perturbation method (sites 2 and 3) results in the allosteric response at the functional site similar to that caused by the native effector fumarate. Moreover, the prediction of allosteric sites from the protein regions obtained via the reverse perturbation method may result in multiple solutions. A repertoire of overlapping and non-overlapping sites can induce comparable allosteric signals upon binding to these sites. This redundancy, suggests that the practical task on the detection of known and prediction of latent allosteric sites can be turned into the general problem of how to induce and fine-tune an allosteric response. The case studies of NAD-dependent malic enzyme, threonine synthase, and fructose-1,6-bisphosphatase were used here to show that a range of allosteric responses can be induced upon simulated binding in sites putatively defined in the regions with increased free energy obtained via the reverse perturbation approach. For example, while simulated binding in the fumarate and putatively designated sites 1–3 of NAD-dependent malic enzyme result in a free energy decrease at the catalytic site, perturbation in site 4 strongly increases the free energy in the catalytic site. Two additional examples, FBPase 1 and ThrS, further illustrate the possibility to adopt the reverse perturbation method for inducing the allosteric response. Further, a presence of the overlapping sites of activators and inhibitors, such as ADP and PEP in PFK, ATP and CTP in ATCase and many others, calls for the fine tuning of the induced allosteric signalling via the rational design of the interactions between ligand and binding site. We exemplify the fine-tuning of the allosteric effect by varying the composition of a designated binding site 1 of NAD-dependent malic enzyme, which results in different levels of the catalytic activity modulation. In general, knowing the allosteric response in the functional sites upon binding of the native allosteric ligand, it is possible to select new allosteric sites and/or ligands that cause the same effect as the natural ones and, therefore, can be considered as new regulatory exosites. Further, an exhaustive calculation of the allosteric effect caused by every residue would be instrumental for rationally defining the candidate/potential allosteric sites in the absence of preliminary experimental data. The latter can be obtained in mutation experiments that measure the allosteric effects of residue-by-residue substitutions on the protein activity [48], providing thus a foundation for the experimental verification and further improvement of the computational model.
Despite two major limitations of the current structure-based perturbation model, the lack of sequence information, and the coarse-grained description of proteins on the basis of structure-based Cα model, the reverse perturbation method should be regarded as a general strategy in finding and exploring allosteric sites. For example, a reverse perturbation-like strategy was used in a recent experimental work where hydrogen exchange mass spectroscopy method was used for characterizing the allosterically active regions of protein Hsp90 induced by the orthosteric binding [49]. It is important to note, that modularity of the structure-based perturbation model is instrumental for improving the accuracy of calculations, as it was done, for example, in Gehrig, S. et al. [50], where instead of using normal modes the slowest principal components (PCA) calculated from the MD trajectories were used for the calculations of the allosteric potential and the corresponding free energy derived in our model. However, increase of the accuracy by implementing the MD simulations will come at a price of the calculation speed. An alternative way of the model improvement would be via including the sequence dependence in the energy function, which preserves, at the same time, the model’s efficiency in terms of the high speed of calculations.
To conclude, the task of inducing and fine-tuning of allosteric response can be generalized and formulated in the following sequence of steps: (i) finding the potential regulatory exosites via reverse perturbation approach; (ii) optimizing the compositions/structures of the binding exosites that can induce a required allosteric signalling upon binding to them; (iii) selection of the appropriate ligands that interact with the chosen allosteric site with sufficient binding affinity; (iv) allocation of the regulatory exosites that provide required allosteric effect at the corresponding functional site in the case of pre-existing library of ligands. By exploring the possibility to detect known, finding latent, and designing new regulatory exosites and corresponding allosteric effectors, the reverse perturbation method introduced in this work provides a conceptual framework aimed at the optimization of the allosteric regulation of protein activity.
We used here the set of 13 allosteric enzymes included in previous studies [31–33], which we refer to as the “classical set”. An additional set of 41 allosteric proteins with 48 experimentally-determined allosteric sites is obtained from the benchmarking collection of allosteric proteins ASBench [34]. In collecting above additional set, we applied the following requirements: (i) structures lacking information on functional sites, on parts of the structures, proteins that change the oligomerization state and structures in which regulation involves protein-protein interactions were omitted together with other cases of missing annotation; (ii) based on the operational definition of the allosteric site and applying the “proximity threshold” (see below) of no more than 2%, we obtained the final list of 48 sites in 41 proteins (S1 Table). Interacting residues were extracted from the structures, based on the distance cutoff of 4.5 Å between the heavy atoms of protein residues and those of the ligand. The quaternary structure assemblies were obtained from the PDBePISA [51, 52] with all water molecules, ions and ligands removed. The apo form of protein complexes was used whenever available, except for CAP due to the large structural difference between its apo and DNA-bound forms.
The structure-based statistical mechanical model of allostery used in this work consists of three parts, which are described in detail in a previous work [31]. First, Cα harmonic models are built for the ligand-free (unperturbed) and ligand-bound (perturbed) systems from a single crystal structure. The presence of a bound ligand, i.e. perturbation, is modeled via the harmonic restrain of all residue pairs in the binding site. For the unperturbed system, the harmonic potential for all pairs of Cα atoms i and j is given by
E(0)(r−r0)=∑pairsi,jki,j(di,j−di,j0)2
(1)
where r is the 3N-dimensional vector of coordinates of Cα atoms and r0 is a vector of Cα atoms from the reference structure. The di,j is the distance between any pair of Cα atoms i and j, the corresponding distance in the reference structure is di,j0. The distance-dependent force constant ki,j decays as (1/di,j0)6 with a global cutoff of 25 Å [53]. The potential associated with the ligand-bound state (B) with n bound sites S = {s1,s2,…,sn} is given by
E(B)(r−r0)=E(0)(r−r0)+α∑n∑pairsi,j∈snki,j(di,j−di,j0)2
(2)
where α = 100 is a stiffening factor of the perturbed site. Protein’s configurational ensembles are characterized by the first 10 slowest normal modes eμ(0) and eμ(B) for the free and ligand-bound systems, respectively. Previous studies have shown that the conformational transitions in allosteric communication are well described by the first 10 low frequency normal modes [32, 33].
Second, we define a microscopic allosteric potential associated with a residue i
Ui(σ)=12∑μεμ,iσμ2,
(3)
which measures the total elastic work acting on a residue i as a result of the change in displacements of all its neighbours caused by the normal mode μ. The generic configuration of residue i is obtained from the reference configuration ri0 by superimposing the vectors eμ,i such as ri(σ)=ri0+∑μσμeμ,i, where σ = (σ1,…,σμ,…) is a vector of Gaussian amplitudes. Thus, the residue configuration ri(σ) is uniquely identified by the state vector (σ1,…,σμ,…). The parameters εμ,i measures the elastic stress on the residue i and its neighbours j as result of the motion associated with the mode eμ
εμ,i=∑j:di,j0<dcc|eμ,i−eμ,j|2,
(4)
where c = 1 kcal/mol/Å2 and a distance cutoff dc = 11 Å. The allosteric potential in Eq 3 is evaluated for both protein states, ligand free (0) and ligand bound (B) ones, respectively. Finally, in order to obtain a per-residue free energy, the allosteric potential is integrated over all possible configurations σ, resulting in the partition function zi=Πμ(2πkBTεμ,i)1/2, and, consecutively in the free energy gi = −kBT ln zi. The per-residue free energy difference between the unperturbed (0) and perturbed (B) protein states is
Δgi=12kBT∑μlnεμ,i(B)εμ,i(0).
(5)
To compare the relative strength of the free energy change Δgi for one residue to the effects on the corresponding subunit, the following average is considered
ΔgU=1nU∑i∈UΔgi,
(6)
where nU is the total number of residues in the subunit U.
In the reverse perturbation approach, the functional sites are perturbed, and, as a result, the change in the free energy at the levels of residues and sites is evaluated throughout the protein. To analyse the Δgi values of every residue in the oligomeric enzyme and to obtain the corresponding Δgi profile, the Δgi of corresponding residues from different subunits are averaged. Both functional and allosteric sites are indicated on the Δgi profile (Figs 2–6 and S1–S4 Figs). The change of the free energy in the site s is also estimated as the average of the free energy changes among the residues in this site
ΔgS=1ns∑i∈sΔgi,
(7)
where ns is the total number of residues in the site s. Eq 7 is used to obtain the free energy changes for every functional and allosteric site.
The operational definition of the allosteric site is based on the restriction on a spatial proximity of communicating functional and regulatory sites. For every pair of residues i and j, the number of physically interacting pairs on the basis of the distance cutoff dc = 11 Å is obtained. The proximity is defined as the fraction of interacting pairs over the total number of pairs between the residues of functional and candidate allosteric sites under consideration:
PX=ndi,j<dcni,j×100%
(8)
A ligand-binding site is defined as allosteric for the corresponding functional site within the same subunit if the proximity PX is no more than 2%.
A distribution of the free energy changes Δgi is obtained upon perturbation of the catalytic sites for each protein in the classical and benchmark sets. For most of the allosteric proteins, all residues in the known allosteric sites exhibit an increase in the free energy, hence only the positive range of the Δgi distribution is used. In the PTP1B and BGDH, the entire Δgi distributions are used as the known allosteric sites contain equal proportion of residues with gain or loss of the free energy. For plotting the ROC, the first bin corresponding to the residues within the top 5% of the Δgi distribution is first used. A sequence of bins with decreasing thresholds with a 5% step is defined to obtain a series of true and false positive rates of the ROC curve. For residues with Δgi above the threshold, a true positive is scored if the residue in the crystal structure is located within 4.5 Å from the allosteric effector, whereas a false positive indicates that the residue does not belong to a known allosteric site.
The harmonic models of proteins and the normal modes analysis are performed using the MMTK package [54]. Fourier approximation is used in the calculation of normal modes [55]. UCSF Chimera [56] is used to generate the illustrations.
|
10.1371/journal.pntd.0005467 | H+ channels in embryonic Biomphalaria glabrata cell membranes: Putative roles in snail host-schistosome interactions | The human blood fluke Schistosoma mansoni causes intestinal schistosomiasis, a widespread neglected tropical disease. Infection of freshwater snails Biomphalaria spp. is an essential step in the transmission of S. mansoni to humans, although the physiological interactions between the parasite and its obligate snail host that determine success or failure are still poorly understood. In the present study, the B. glabrata embryonic (Bge) cell line, a widely used in vitro model for hemocyte-like activity, was used to investigate membrane properties, and assess the impact of larval transformation proteins (LTP) on identified ion channels. Whole-cell patch clamp recordings from Bge cells demonstrated that a Zn2+-sensitive H+ channel serves as the dominant plasma membrane conductance. Moreover, treatment of Bge cells with Zn2+ significantly inhibited an otherwise robust production of reactive oxygen species (ROS), thus implicating H+ channels in the regulation of this immune function. A heat-sensitive component of LTP appears to target H+ channels, enhancing Bge cell H+ current over 2-fold. Both Bge cells and B. glabrata hemocytes express mRNA encoding a hydrogen voltage-gated channel 1 (HVCN1)-like protein, although its function in hemocytes remains to be determined. This study is the first to identify and characterize an H+ channel in non-neuronal cells of freshwater molluscs. Importantly, the involvement of these channels in ROS production and their modulation by LTP suggest that these channels may function in immune defense responses against larval S. mansoni.
| Schistosoma mansoni is one of four major species of human blood flukes that, together, infect over 250 million people worldwide. Transmission of S. mansoni to humans requires infection of freshwater intermediate host snails, Biomphalaria spp., in order to complete its life cycle. The B. glabrata embryonic (Bge) cell line, derived from a Puerto Rican strain of snail host shares characteristics with circulating hemocytes, the molluscan immune cells, and serves as an in vitro model for snail immune function. Electrical recordings from Bge cells demonstrated the presence of H+ channels that allow hydrogen ions (H+) to cross the membrane. Furthermore, blocking these channels inhibited the production of reactive oxygen species (ROS), an immune defense mechanism shared by Bge cells and hemocytes. Interestingly, Bge cell exposure to proteins produced by S. mansoni larvae exerted the opposite effect, enhancing H+ movement across the cell membrane. An H+ channel-encoding gene was expressed in both Bge cells and hemocytes suggesting that hemocytes may share similar functions with Bge cells.
| Schistosomiasis, a neglected tropical disease afflicting over 250 million people worldwide [1], is caused by parasitic flatworms of the genus Schistosoma. Schistosoma spp. have a two-host life cycle involving sexual reproduction within a mammalian host and asexual reproduction within a snail intermediate host. The pathology associated with the intestinal form of human schistosomiasis arises in chronic infections when eggs released by female worms occupying mesenteric veins become trapped in the liver (and other organs) and elicit an intense inflammatory response leading to the formation of granulomas that damage tissues and block circulation [2, 3]. Eggs from ruptured intestinal capillaries exit the host by fecal excretion, and upon exposure to freshwater, hatch to release the free-swimming snail-infective miracidia. Upon infection of snails, miracidia transform through two sporocyst stages, ultimately completing their life cycle by the production and release of free-swimming cercariae, the human-infective stage [4]. Because of the absolute dependency of human schistosome transmission on the snail host, one of the keys to sustained control of schistosomiasis is to block or eliminate the snail’s participation in the life cycle.
The freshwater snail Biomphalaria glabrata serves as the most common invertebrate host of S. mansoni, the most widely distributed species of Schistosoma [5]. Hemocytes (phagocytic immune cells) of B. glabrata, genetically-selected for susceptibility or resistance to infection by larval S. mansoni, have been shown to react differentially to invading miracidia. Circulating hemocytes of susceptible strains do not recognize and kill invading larvae, whereas in resistant snails developing larvae are rapidly encapsulated by hemocytes and killed within 24–48 hours of infection [6–8]. Hemocyte larvicidal activity has been linked to the production and release of reactive oxygen species (ROS), mainly hydrogen peroxide (H2O2), and the reactive nitrogen species, nitric oxide [9, 10]. Although hemocytes of both resistant and susceptible B. glabrata strains produce H2O2, resistant hemocytes generate and release higher levels than susceptible cells [11], and this production appears to depend on the extracellular signal regulated protein kinase (Erk) [12]. However, a critical question arising from these observations is what are the signaling mechanisms that regulate ROS responses?
A critical period of larval development in the snail host is 24–48 hours post-infection, when the newly invading miracidium completes its transformation to the primary sporocyst stage. Larval killing depends on the ability of circulating hemocytes to recognize and encapsulate the newly formed sporocyst [4, 13–15]. Among various sporocyst factors that may be contributing to hemocyte reactivity are glycoproteins that are released during the miracidium-to-sporocyst transition. In vitro studies have shown that these larval transformation proteins (LTPs) [16] modulate phagocytic activity, motility, and ROS production in B. glabrata hemocytes [17–21], and disrupt hemocyte immune signaling [22–24]. However, questions regarding specific mechanisms by which LTPs modulate hemocyte immune responses remain unanswered.
For over four decades a cell line derived from embryos of a schistosome-susceptible strain of B. glabrata, the B. glabrata embryonic (Bge) cell line [25], has served as an in vitro model for the study of larval schistosome-snail host interactions in schistosomiasis. Bge cells share many characteristics with B. glabrata hemocytes including their morphology, adhesive properties, phagocytic activity, and larval encapsulation response [26]. In fact, co-culture of Bge cells with S. mansoni larvae results in the development of the parasite from the miracidium to the final cercarial stage, similar to the development that occurs with susceptible B. glabrata strains [27–30]. We have therefore adopted Bge cells as an in vitro model system to study the molecular interactions between snail cells and S. mansoni LTP. Because ion channels in the plasma membrane of human immune cells, including eosinophils, macrophages, neutrophils and lymphocytes, play important roles in immune responses, often by regulating the production and release of ROS [31], we explored the role ion channels may play in signaling and ROS production in Bge cells. Using the whole cell patch clamp technique, we discovered an LTP-sensitive H+ channel that serves as the dominant ion conductance of Bge cell membranes. In addition, using a fluorescent probe to measure intracellular ROS, we also found that this channel mediates the production of ROS, thus suggesting a possible function for H+ channels in snail immune responses.
All animal care and procedures were approved by the Institutional Animal Care and Use Committee of the University of Wisconsin-Madison under protocol V00640.
The Bge cell line was originally obtained from American Type Culture Collection (ATCC CRL 1494) and is currently available through the BEI Resources (https://www.beiresources.org). Cells were maintained at 26°C under normoxic conditions in complete Bge (c-Bge) medium consisting of 22% Schneider’s Drosophila Medium, 0.45% lactalbumin enzymatic hydrolysate, and 7.2 mM galactose supplemented with 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin [25, 28]. Bge cells were passaged at 80% confluency.
S. mansoni eggs were isolated, hatched, and miracidia cultured in vitro as previously described [28]. Approximately ~ 5000 miracidia/mL in Chernin’s balanced salt solution (CBSS; 47.9 mM NaCl, 2.0 mM KCl, 0.5 mM Na2HPO4, 0.6 mM NaHCO3 1.8 mM MgSO4, 3.6 mM CaCl2 and pH 7.2) [32] supplemented with glucose (1 mg/mL), trehalose (1 mg/mL), penicillin G (100 units/mL) and streptomycin sulfate (0.05 mg/mL) adjusted to pH 7.2 (CBSS+) were then plated in a 24-well tissue culture plate and incubated at 26°C under normal atmospheric conditions to allow in vitro transformation of miracidia to primary sporocysts. The LTP-containing culture medium was collected after 48 hr, and the newly transformed primary sporocysts were washed once with CBSS+. The LTP and CBSS+ wash were combined, filtered with a 0.45 μm Nalgene syringe filter (Thermo Scientific, Waltham, MA), and concentrated using 3 kDa molecular weight cut-off ultrafiltration tubes (Amicon Ultra Centricon, Billerica, MA). A NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE) was used to determine the protein concentration, after which a protease inhibitor cocktail (Calbiochem, Billerica, MA) was added. Multiple collections of LTP were pooled and stored in aliquots at -20°C. To denature LTP, pools were boiled at 100°C for 5 min.
Bge cells (~4 x 106) were plated in 60x15 mm petri dishes in c-Bge medium, and allowed to attach overnight. In order to make recordings under defined ionic conditions, cells were washed 3X with CBSS before recording and kept in this buffer during subsequent manipulations. In experiments involving the treatment of Bge cells with ZnCl2, 10 mM HEPES replaced NaH2PO4 in CBSS due to the insolubility of Zn3(PO4)2. Adherent cells were viewed with an Axioskop microscope equipped with a 63X water-immersion objective (Carl Zeiss, Thornwood, NY). Bge cells were imaged with a CCD camera and viewed on a monitor. Patch electrodes fabricated from borosilicate glass capillaries had resistances of 3–7 MΩ when filled with a solution containing (in mM) 60 K-gluconate, 1 CaCl2, 1 MgCl2, 1 Mg-ATP, 10 HEPES, and 5 EGTA. The bathing solution for recordings was a slightly modified version of CBSS consisting of (in mM): 47 NaCl, 2 KCl, 0.5 NaH2PO4, 0.6 NaHCO3, 1.8 MgSO4, 3.6 CaCl2. The pH of the pipette solution and external CBSS was adjusted to 5 or 7 with KOH or HCl. Modified versions of the internal and external solutions are stated in the Results section where they are used. Pressure-ejection pipettes were modified patch electrodes with tip diameters of ~2 μm. A Picospritzer II (General Valve Corp.) was used to apply 5–10 PSI of pressure to ejection pipettes.
Patch clamp recordings were made with an Axopatch 200B amplifier (Molecular Devices, Sunnyvale, CA), with data read into a PC through a Digidata 1440 A interface. The computer program pClamp 10 (Molecular Devices) controlled data acquisition, voltage steps, and pressure application by the Picospritzer. Data were filtered with a low-pass Bessel filter at 2 kHz before digitization at 10 kHz.
The fluorescent probe 2’7’-dichlorofluorescein-diacetate (DCFH-DA; Sigma-Aldrich, St. Louis, MO) was used to measure ROS production in Bge cells following a method described previously with hemocytes [33]. Bge cells (~1.5 x 105) in suspension were washed 3X with CBSS before incubation in CBSS (control), CBSS containing either 30 μg/mL LTP, 1 mM ZnCl2 or 30 μg/mL LTP + 1 mM ZnCl2 for 1 hr at 26°C. After treatment, cells were washed 3X with CBSS and centrifuged at 1000 rpm for 10 min. The final cell pellets were then re-suspended in 150 μL of CBSS containing 10 μM DCFH-DA, and distributed in three wells of a 96-well black-walled plate (BD Falcon). The oxidation of DCFH-DA to fluorescent 2’7’-dichlorofluorescein (DCF) was measured in triplicate at 10 min intervals for up to 60 min using a Bio-Tek Synergy fluorescence plate reader (Winooski, VT) with excitation and emission wavelengths of 485 ± 20 and 528 ± 20, respectively. Data analysis was conducted with Origin software (Microcal, Northhampton, MA, USA). Five independent replicates of each experiment were conducted, with the raw data presented as mean ± SEM, and ratios of means of treated groups to controls presented separately.
For molecular analysis of H+ channels, the hydrogen voltage gated channel 1 (HVCN1) gene was identified in the nonredundant NCBI database, and sequence comparisons were conducted with PCR products from Bge cells and B. glabrata hemocytes. Schistosome-susceptible (NMRI) and resistant (BS-90) B. glabrata strains were maintained in laboratory colonies in 10-gallon aquaria at 26°C under 12:12 hr light/dark cycling. Hemolymph, containing hemocytes, was collected by headfoot retraction [34] and immediately transferred to Eppendorf tubes containing an equal volume of CBSS on ice. Hemocytes were then pelleted by centrifugation at 1000 RPM for 10 min and washed 3 times in CBSS. Bge cells, grown in a flask to ~80% confluency, were detached mechanically using a cell scrapper, transferred to a 15 mL conical tube and pelleted by centrifugation as described for hemocytes.
Total RNA was extracted from Bge cells and hemocytes of both B. glabrata strains using TRIzol reagent. Normalized concentrations of isolated total RNA samples were subjected to cDNA synthesis reactions using the GoScriptTM Reverse Transcription System (Promega Corp., Madison, WI). The cDNA was then used as the template for PCR using primers for the B. glabrata voltage-gated H+ channel 1-like gene (BgHVCN1-like; Forward 5’-TGCTATGGGCTTAGCTTACTTC-3’; Reverse 5’-ATGTAGGGTCTTCAAACCATTCT-3’) that were designed using the predicted mRNA sequence for the gene with the National Center for Biotechnology Information (NCBI) database (Accession number XM_013231505). The expected amplicon size is ~362 bp, ~65% of the coding DNA sequence. As a positive control, primers for B. glabrata α–tubulin (Forward 5’ -GTGAGACTGGCTGTGGTAAA-3’; Reverse 5’ -GGGAAGTGAATCCTGGGATATG-3’) with Accession number XP_013094834.1 were used to amplify an expected product of ~643 bp. Gel electrophoresis of the PCR products was performed followed by Big Dye sequencing at the University of Wisconsin Biotechnology Center DNA Sequencing Facility (Madison, WI). The resulting nucleotide sequences were used in a search using BLASTn search against the non-redundant nucleotide NCBI database to confirm that the PCR amplified product encoded an HVCN1-like protein.
Patch clamp data were analyzed with Clampfit (Molecular Devices, Sunnyvale, CA) and Origin Pro (Microcal, Northhampton, MA). One-way RM-ANOVA and post-hoc statistical analyses were conducted in Origin Pro to assess significance. Results are presented as means ± SEM. The asterisks (*) represent p < 0.05 in all figures.
Whole cell patch clamp recordings were made from Bge cells to explore their membrane properties. Voltage steps from -75 to 25 mV for 500 msec induced an outward current that activated rapidly and then weakly inactivated in ~10–20 msec before stabilizing (Fig 1A, control trace, top). To identify the ions responsible for this current, we manipulated the composition of the recording solutions. When Cl- was replaced by gluconate in the internal and bathing solutions, voltage steps induced currents similar to those seen with control solutions (Fig 1A, second trace from top). Further substitution of Cs+ for K+ in the internal solution reduced the current to roughly 68% of control currents (Fig 1A, third trace from top). The mean peak and plateau current amplitudes for these solutions are shown in Fig 1B. For gluconate and Cs+ substitution, current amplitudes were not significantly different from the control. Thus, Cl- and K+ replacement experiments indicated that these are not major permeating ions. In addition, comparisons of the Nernst potentials (equilibrium potential for each ion based on internal and external concentrations) with reversal potentials in current-voltage relationships did not support channels selective for Na+ or Ca2+ (Supplemental S1 Fig). These results suggested that the major ions in our recording solutions do not permeate the membranes of Bge cells.
H+ channels play important roles in many types of immune cells [35], so we explored the possibility that H+ channels reside in the membranes of Bge cells. Subjecting Bge cells to pH gradients (by adjusting the pH of the pipette and bathing solutions–see Methods) [36] altered the current elicited by voltage steps and shifted the relationship between current and voltage (Fig 2). A gradient of two pH units (pH 5in/pH 7out) reduced the current amplitude at all voltages and shifted the reversal potential in the plot of peak current versus voltage in the negative direction by 17.5 mV (Fig 2B, dashed line). Reversing the pH gradient (pH 5out/pH 7in) shifted the peak current-voltage plot in the opposite direction with a positive shift in the reversal potential of 27.5 mV (Fig 2B, dotted line). Plots of plateau current versus voltage showed similar shifts (Supplemental S2 Fig). Table 1 presents the reversal potentials along with the Nernst potentials for H+. The shifts are in the direction of the H+ Nernst potential but smaller in magnitude because the H+ concentration is much lower relative to the concentrations of other ions in the solutions. Channels permeable to other ions generally result in H+ current reversal potential shifts that are less than the change in the H+ Nernst potential [37]. The effects of pH gradients on membrane currents are consistent with the presence of an H+ channel in Bge cell membranes.
As an additional test for the presence of H+ channels we applied the H+ channel blocker Zn2+ [36, 38]. Pressure application of 1 mM ZnCl2 from a glass pipette onto a Bge cell significantly reduced both peak and plateau currents elicited by voltage steps from -50 to 20 mV (Fig 3A). This blockade was reversible, as demonstrated by current recovery after ZnCl2 removal (Fig 3A, wash trace). Time course plots in which ZnCl2 was perfused onto cells through the bathing medium showed a 3.5-fold reduction in current amplitude (Fig 3B and 3C), from 621 ± 4 pA to 177 ± 1 pA (N = 4), supporting the presence of H+ channels in Bge cell membranes. Although other actions of Zn2+ cannot be ruled out, the block of membrane current is consistent with the presence of H+ channels in Bge cells.
As larval schistosome proteins have been shown to modulate a variety of snail hemocyte immune functions [14, 15], we tested the effects of S. mansoni LTP on Bge cell membrane current. Pressure application of LTP onto Bge cells dramatically increased the peak and plateau currents evoked by steps from -50 mV to 20 mV (Fig 4A). LTP increased the current significantly by over 2-fold (478 ± 6 pA) compared to control (212 ± 4 pA), and this increase only partially reversed with a 17% decrease (397 ± 7 pA) following removal of LTP. Recovery was slow, and 5 min after LTP removal the current had decreased only slightly (Fig 4B and 4C). Plotting current versus time also illustrated the opposite effects of LTP and ZnCl2 on Bge cells (Fig 4C). This plot showed a >2-fold increase in current amplitude in the presence of LTP (Fig 4C blue circles) and a >2-fold reduction in the presence of ZnCl2 (Fig 4C, red triangles) compared to control (Fig 4C black squares). The reversal of block by ZnCl2 was rapid and essentially complete, while the reversal of enhancement by LTP was slow. Moreover, when heat-denatured LTP was pressure-applied onto Bge cells, we observed no significant change compared to control current amplitudes (Fig 5C and 5D), indicating that the action of LTP on H+ channels depends on heat-labile factors.
To determine whether LTP increased Bge cell membrane current by opening H+ channels, we applied LTP and ZnCl2 simultaneously, and observed no statistically significant change (Fig 5), indicating that ZnCl2 counters the effect of LTP. Finally, we noted that current-voltage curves shifted in the presence of LTP and ZnCl2; LTP caused a 9 mV right-shift from control, toward the H+ Nernst potential, while ZnCl2 caused a 23 mV left-shift, away from the H+ Nernst potential (Fig 6). These results are consistent with the blockade of H+ channels by ZnCl2 and enhancement of H+ channels by LTP.
Because H+ channels contribute to ROS production in mammalian immune cells [35, 39], we measured the generation of ROS in Bge cells with the fluorescent probe 2’7’-dichlorofluorescein-diacetate (DCFH-DA). We observed a rapid and robust fluorescence increase that reflects constitutive ROS production. ZnCl2 and LTP + ZnCl2 inhibited this activity by ~50% compared to the untreated control (F3, 16 = 24.26, p < 0.05). These results demonstrate a linkage between H+ channels and the production of ROS in Bge cells. LTP alone produced a small apparent increase in ROS production, but this increase was not statistically significant. This suggests that at control level of H+ current in Bge cells, other factors limit ROS production (Fig 7A and 7B).
To identify putative H+ channel proteins expressed by Bge cells and B. glabrata hemocytes we searched the B. glabrata genome (https://www.vectorbase.org/organisms/biomphalaria-glabrata) using Blastp for homologues of human HVCN1 protein. The closest match was an HVCN1-like protein (BgHVCN1-like, Accession number. XM_013231505) with 31% identity to human HVCN1. This sequence contained the motif RLWRVTR, which is consistent with the H+ channel consensus sequence RxWRxxR [36]. A segment of the predicted sequence was then used to design primers for polymerase chain reactions (PCR). Using cDNA from Bge cells and B. glabrata hemocytes (NMRI and BS-90 strains) as templates, PCR using the primers stated in the Methods section yielded amplicons of similar size with 99% sequence identity (E = 0.0) to the predicted B. glabrata HVCN1-like sequence (Supplemental S3 Fig). The amplified products encode 120 amino acid stretch of the 186 residues predicted for molluscan HVCN1-like protein. These results indicate that mRNA with the predicted sequence for a BgHVCN1-like gene is present in both Bge cells and hemocytes.
This investigation revealed the presence of functional ion channels in Bge cell membranes. pH manipulations altered the voltage dependence of membrane currents in a manner consistent with a dominant H+ permeability. Since the H+ concentration was several orders of magnitude lower than the other ions in our solutions, even low permeabilities to other ions can make large contributions to the observed reversal potentials and move them away from the H+ Nernst potential. Thus, although currents did not reverse at the H+ Nernst potential, the shifts were in the appropriate direction and supported the hypothesis that H+ channels are the predominant ion permeability in the plasma membrane of Bge cells. We also found that the H+ channel blocker Zn2+ significantly reduced the current through Bge cell membranes, providing additional support for the presence of H+ channels. Finally, we identified and sequenced an HVCN1-like transcript expressed in both this snail cell line and B. glabrata hemocytes, suggesting a functional linkage between these cell types. Thus, three independent lines of evidence support the conclusion that Bge cells express functional H+ channels. With few exceptions [40], previous studies focusing on ion channels in molluscs almost exclusively have involved neuronal cell systems and/or emphasized Na+, K+, Ca2+ or Cl- channel activities [41–43]. To our knowledge, this is the first report of a functional H+ channel in non-neuronal cells of freshwater gastropods.
Similar to the well documented association between H+ channels and ROS production in mammalian immunocytes [39], we also found that blockade of the H+ channel with Zn2+ significantly abrogated Bge cell ROS production, indicating a functional association between channel-mediated H+ flux across the membrane and the oxidative response. This finding is significant since the formation and release of several ROS, especially H2O2, and RNS are known to be involved in the killing of larval S. mansoni by B. glabrata hemocytes [9, 10]. It is possible that, as in mammalian immune cells [39, 44], changes in membrane potential associated with ROS production also require a compensatory activation of H+ channels to maintain pH balance in immunocyte-like molluscan cells. It is important to note that hemocytes from both resistant and susceptible strains of B. glabrata snails are capable of generating ROS [11, 32], but differ both qualitatively and quantitatively in their responses [11]. Since Bge cells were originally derived from a S. mansoni-susceptible Puerto Rican strain of B. glabrata [25], it is likely that hemocytes from a related susceptible strain (NMRI) also share both molecular and functional similarities to Bge cells. These shared characteristic have been well-documented in previous studies [26, 45, 46], supporting the use of this cell line as a hemocyte-like model, as well as a general model for Biomphalaria-schistosome interactions [29, 47]. Based on the presence and expression of the HVCN1-like gene in B. glabrata hemocytes, it is quite possible that voltage-gated H+ channels are also involved in regulating cellular ROS production as demonstrated in Bge cells.
Proteins released during the S. mansoni miracidium-to-sporocyst transformation (LTP) have been shown to modulate a variety of functions in both hemocytes and Bge cells [14, 24, 45]. Such a role is supported by our finding of an LTP-induced potentiation of H+ channel activity. Exposure to LTP elicited a rapid and sustained enhancement of Bge cell membrane current. Because the reversal potential moved toward the H+ Nernst potential, it is likely that LTP increased the current through H+ channels. This activity was heat-labile, suggesting that the channel-active LTP component(s) may be a protein(s) with irreversible or slowly reversing action. However, it remains unclear whether the regulation of Bge cell H+ channels by schistosome LTPs results from factors thought to play a role in host-parasite compatibility [48–50] or other, yet unidentified, larval factors. The H+ channel may play a role in co-evolutionary mechanisms, known to affect oxidant-antioxidant levels during parasite-host interaction [51]. Identifying the active components of LTP and determining whether this response reflects the action of a single or multiple species will require further investigation.
Despite the channel stimulating action of LTP, LTP treatment of Bge cells resulted in no statistically significant increase in ROS production. These results are consistent with previous findings that exposure of B. glabrata hemocytes to excretory-secretory products of larval S. mansoni exerted little effect on the production of ROS [52]. However, the question remains as to why LTP-stimulated H+ channel activation failed to enhance ROS production. Based on the H+ current data, it might be speculated that LTP binding to Bge cells is linked to the opening of H+ channels through receptor-mediated activation of a channel-associated signaling pathway, possibly through interactions with pathogen recognition receptors such as fibrinogen-related proteins, Toll-like receptors, or bacterial binding proteins that have been implicated in B. glabrata immunity [50, 53–55]. Mitogen-activated and extracellular-signal regulated protein kinases shown to function in molluscan immunity [12, 22] could also play a role in signaling to the H+ channel. A final possibility is that LTP may be acting directly on the channel protein itself to induce opening. The consequence of H+ channel modulation would be an alteration or disruption of H+ ion balance and intracellular pH, but without stimulating ROS production. This may, in turn, serve as a potent anti-immune mechanism used by sporocysts for countering host ROS-mediated effector responses. Thus, H+ channels, while serving an important role in maintaining pH balance within Bge cells and hemocytes, may also be manipulated by schistosome larvae to reduce their immune efficacy. Since Bge cells were originally derived from a S. mansoni-susceptible PR albino strain of B. glabrata [25], it is likely that hemocytes from a related susceptible strain (NMRI) also share sensitivity to H+ channel–reactive anti-immune proteins, thereby supporting a compatible snail-schistosome interaction.
In conclusion, Bge cells possess a functional H+ channel that is responsible for a dominant conductance of their plasma membrane. ROS production is dependent on H+ channels. Exposure of cells to heat-labile LTP stimulates channel opening and H+ flux, but has little if any effect on the generation of ROS. Although H+ channels have not been tested directly in B. glabrata hemocytes, PCR amplification and amplicon sequencing demonstrated the presence of HVCN1-like transcripts in both susceptible and resistant B. glabrata strains. Thus, the association of the Bge cell H+ channel activity with cellular ROS production and the channel’s response to schistosome LTP suggest a role in regulating larval schistosome-snail interactions. Future identification of the specific mechanism(s) tying together these activities should provide important insights into host-parasite compatibility in this system.
|
10.1371/journal.pbio.0050077 | The Sorcerer II Global Ocean Sampling Expedition: Northwest
Atlantic through Eastern Tropical Pacific | The world's oceans contain a complex mixture of micro-organisms that are for the most
part, uncharacterized both genetically and biochemically. We report here a metagenomic
study of the marine planktonic microbiota in which surface (mostly marine) water samples
were analyzed as part of the Sorcerer II Global Ocean Sampling
expedition. These samples, collected across a several-thousand km transect from the North
Atlantic through the Panama Canal and ending in the South Pacific yielded an extensive
dataset consisting of 7.7 million sequencing reads (6.3 billion bp). Though a few major
microbial clades dominate the planktonic marine niche, the dataset contains great
diversity with 85% of the assembled sequence and 57% of the unassembled data being unique
at a 98% sequence identity cutoff. Using the metadata associated with each sample and
sequencing library, we developed new comparative genomic and assembly methods. One
comparative genomic method, termed “fragment recruitment,” addressed questions of genome
structure, evolution, and taxonomic or phylogenetic diversity, as well as the biochemical
diversity of genes and gene families. A second method, termed “extreme assembly,” made
possible the assembly and reconstruction of large segments of abundant but clearly
nonclonal organisms. Within all abundant populations analyzed, we found extensive
intra-ribotype diversity in several forms: (1) extensive sequence variation within
orthologous regions throughout a given genome; despite coverage of individual ribotypes
approaching 500-fold, most individual sequencing reads are unique; (2) numerous changes in
gene content some with direct adaptive implications; and (3) hypervariable genomic islands
that are too variable to assemble. The intra-ribotype diversity is organized into
genetically isolated populations that have overlapping but independent distributions,
implying distinct environmental preference. We present novel methods for measuring the
genomic similarity between metagenomic samples and show how they may be grouped into
several community types. Specific functional adaptations can be identified both within
individual ribotypes and across the entire community, including proteorhodopsin spectral
tuning and the presence or absence of the phosphate-binding gene
PstS.
| Marine microbes remain elusive and mysterious, even though they are the most abundant
life form in the ocean, form the base of the marine food web, and drive energy and
nutrient cycling. We know so little about the vast majority of microbes because only a
small percentage can be cultivated and studied in the lab. Here we report on the Global
Ocean Sampling expedition, an environmental metagenomics project that aims to shed light
on the role of marine microbes by sequencing their DNA without first needing to isolate
individual organisms. A total of 41 different samples were taken from a wide variety of
aquatic habitats collected over 8,000 km. The resulting 7.7 million sequencing reads
provide an unprecedented look at the incredible diversity and heterogeneity in naturally
occurring microbial populations. We have developed new bioinformatic methods to
reconstitute large portions of both cultured and uncultured microbial genomes. Organism
diversity is analyzed in relation to sampling locations and environmental pressures.
Taken together, these data and analyses serve as a foundation for greatly expanding our
understanding of individual microbial lineages and their evolution, the nature of marine
microbial communities, and how they are impacted by and impact our world.
| The concept of microbial diversity is not well defined. It can either refer to the genetic
(taxonomic or phylogenetic) diversity as commonly measured by molecular genetics methods, or
to the biochemical (physiological) diversity measured in the laboratory with pure or mixed
cultures. However, we know surprisingly little about either the genetic or biochemical
diversity of the microbial world [1],
in part because so few microbes have been grown under laboratory conditions [2,3], and also because it is likely that there are immense
numbers of low abundance ribotypes that have not been detected using molecular methods
[4]. Our understanding of microbial
physiological and biochemical diversity has come from studying the less than 1% of organisms
that can be maintained in enrichments or cultivated, while our understanding of phylogenetic
diversity has come from the application of molecular techniques that are limited in terms of
identifying low-abundance members of the communities.
Historically, there was little distinction between genetic and biochemical diversity
because our understanding of genetic diversity was based on the study of cultivated
microbes. Biochemical diversity, along with a few morphological features, was used to
establish genetic diversity via an approach called numerical taxonomy [5,6].
In recent years the situation has dramatically changed. The determination of genetic
diversity has relied almost entirely on the use of gene amplification via PCR to conduct
taxonomic environmental gene surveys. This approach requires the presence of slowly
evolving, highly conserved genes that are found in otherwise very diverse organisms. For
example, the gene encoding the small ribosomal subunit RNA, known as 16S, based on
sedimentation coefficient, is most often used for distinguishing bacterial and archaeal
species [7–10]. The 16S rRNA sequences are highly conserved and can be
used as a phylogenetic marker to classify organisms and place them in evolutionary context.
Organisms whose 16S sequences are at least 97% identical are commonly considered to be the
same ribotype [11], otherwise referred
to as species, operational taxonomic units, or phylotypes.
Although rRNA-based analysis has revolutionized our view of genetic diversity, and has
allowed the analysis of a large part of the uncultivated majority, it has been less useful
in predicting biochemical diversity. Furthermore, the relationship between genetic and
biochemical diversity, even for cultivated microbes, is not always predictable or clear. For
instance, organisms that have very similar ribotypes (97% or greater homology) may have vast
differences in physiology, biochemistry, and genome content. For example, the gene
complement of Escherichia
coli O157:H7 was found to be substantially different from the K12 strain
of the same species [12].
In this paper, we report the results of the first phase of the Sorcerer II
Global Ocean Sampling (GOS) expedition, a metagenomic study designed to address questions
related to genetic and biochemical microbial diversity. This survey was inspired by the
British Challenger expedition that took place from 1872–1876, in which the diversity of
macroscopic marine life was documented from dredged bottom samples approximately every 200
miles on a circumnavigation [13–15]. Through the substantial dataset
described here, we identified 60 highly abundant ribotypes associated with the open ocean
and aquatic samples. Despite this relative lack of diversity in ribotype content, we confirm
and expand upon previous observations that there is tremendous within-ribotype diversity in
marine microbial populations [4,7,8,16,17]. New techniques and
tools were developed to make use of the sampling and sequencing metadata. These tools
include: (1) the fragment recruitment tool for performing and visualizing comparative
genomic analyses when a reference sequence is available; (2) new assembly techniques that
use metadata to produce assemblies for uncultivated abundant microbial taxa; and (3) a whole
metagenome comparison tool to compare entire samples at arbitrary degrees of genetic
divergence. Although there is tremendous diversity within cultivated and uncultivated
microbes alike, this diversity is organized into phylogenetically distinct groups we refer
to as subtypes.
Subtypes can occupy similar environments yet remain genetically isolated from each other,
suggesting that they are adapted for different environmental conditions or roles within the
community. The variation between and within subtypes consists primarily of nucleotide
polymorphisms but includes numerous small insertions, deletions, and hypervariable segments.
Examination of the GOS data in these terms sheds light on patterns of evolution and also
suggests approaches towards improving the assembly of complex metagenomic datasets. At least
some of this variation can be associated with functional characters that are a direct
response to the environment. More than 6.1 million proteins, including thousands of new
protein families, have been annotated from this dataset (described in the accompanying paper
[18]). In combination, these papers
bring us closer to reconciling the genetic and biochemical disconnect and to understanding
the marine microbial community.
We describe a metagenomic dataset generated from the Sorcerer II
expedition. The GOS dataset, which includes and extends our previously published Sargasso
Sea dataset [19], now encompasses a
total of 41 aquatic, largely marine locations, constituting the largest metagenomic dataset
yet produced with a total of ~7.7 million sequencing reads. In the pilot Sargasso Sea study,
200 l surface seawater was filtered to isolate microorganisms for metagenomic analysis. DNA
was isolated from the collected organisms, and genome shotgun sequencing methods were used
to identify more than 1.2 million new genes, providing evidence for substantial microbial
taxonomic diversity [19]. Several
hundred new and diverse examples of the proteorhodopsin family of light-harvesting genes
were identified, documenting their extensive abundance and pointing to a possible important
role in energy metabolism under low-nutrient conditions. However, substantial sequence
diversity resulted in only limited genome assembly. These results generated many additional
questions: would the same organisms exist everywhere in the ocean, leading to improved
assembly as sequence coverage increased; what was the global extent of gene and gene family
diversity, and can we begin to exhaust it with a large but achievable amount of sequencing;
how do regions of the ocean differ from one another; and how are different environmental
pressures reflected in organisms and communities? In this paper we attempt to address these
issues.
Microbial samples were collected as part of the Sorcerer II expedition
between August 8, 2003, and May 22, 2004, by the S/V Sorcerer II, a 32-m
sailing sloop modified for marine research. Most specimens were collected from surface
water marine environments at approximately 320-km (200-mile) intervals. In all, 44 samples
were obtained from 41 sites (Figure
1), covering a wide range of distinct surface marine environments as well as a few
nonmarine aquatic samples for contrast (Table 1).
Several size fractions were isolated for every site (see Materials and Methods). Total DNA was extracted from one or more fractions,
mostly from the 0.1–0.8-μm size range. This fraction is dominated by bacteria, whose
compact genomes are particularly suitable for shotgun sequencing. Random-insert clone
libraries were constructed. Depending on the uniqueness of each sampling site and initial
estimates of the genetic diversity, between 44,000 and 420,000 clones per sample were
end-sequenced to generate mated sequencing reads. In all, the combined dataset includes
6.25 Gbp of sequence data from 41 different locations. Many of the clone libraries were
constructed with a small insert size (<2 kbp) to maximize cloning efficiency. As
this often resulted in mated sequencing reads that overlapped one another, overlapping
mated reads were combined, yielding a total of ~6.4 M contiguous sequences, totaling ~5.9
Gbp of nonredundant sequence. Taken together, this is the largest collection of
metagenomic sequences to date, providing more than a 5-fold increase over the dataset
produced from the Sargasso Sea pilot study [19] and more than a 90-fold increase over the other large marine metagenomic
dataset [20].
Assembling genomic data into larger contigs and scaffolds, especially metagenomic data,
can be extremely valuable, as it places individual sequencing reads into a greater genomic
context. A largely contiguous sequence links genes into operons, but also permits the
investigation of larger biochemical and/or physiological pathways, and also connects
otherwise-anonymous sequences with highly studied “taxonomic markers” such as 16S or
recA, thus clearly identifying the taxonomic group with which they are
associated. The primary assembly of the combined GOS dataset was performed using the
Celera Assembler [21] with
modifications as previously described [19] and as given in Materials and Methods. The assembly was performed with quite
stringent criteria, beginning with an overlap cutoff of 98% identity to reduce the
potential for artifacts (e.g., chimeric assemblies or consensus sequences diverging
substantially from the genome of any given cell). This assembly was the substrate for
annotation (see the accompanying paper by Yooseph et al. [18]).
The degree of assembly of a metagenomic sample provides an indication of the diversity of
the sample. A few substantial assemblies notwithstanding, the primary assembly was
strikingly fragmented (Table 2).
Only 9% of sequencing reads went into scaffolds longer than 10 kbp. A majority (53%) of
the sequencing reads remained unassembled singletons. Scaffolds containing more than 50 kb
of consensus sequence totaled 20.7 Mbp; of these, >75% were produced from a single
Sargasso Sea sample and correspond to the Burkholderia or
Shewanella assemblies described previously [19]. These results highlight the unusual abundance of
these two organisms in a single sample, which significantly affected our expectations
regarding the current dataset. Given the large size of the combined dataset and the
substantial amount of sequencing performed on individual filters, the overall lack of
assembly provides evidence of a high degree of diversity in surface planktonic
communities. To put this in context, suppose there were a clonal organism that made up 1%
of our data, or ~60 Mbp. Even a genome of 10 Mbp—enormous by bacterial standards—would be
covered ~6-fold. Such data might theoretically assemble with an average contig approaching
50 kb [22]. While real assemblies
generally fall short of theory for various reasons, Shewanella data make
up <1% of the total GOS dataset, and yet most of the relevant reads assemble into
scaffolds >50 kb. Thus, with few scaffolds of significant length, we could conclude
that there are very few clonal organisms present at even 1% in the GOS dataset.
To investigate the nature of the implied diversity and to see whether greater assembly
could be achieved, we explored several alternative approaches. Breaks in the primary
assembly resulted from two factors: incomplete sequence coverage and conflicts in the
data. Conflicts can break assemblies when there is no consistent way to chain together all
overlapping sequencing reads. As it was possible that there would be fewer conflicts
within a single sample (i.e., that diversity within a single sample would be lower),
assemblies were attempted with individual samples. However, the results did not show any
systematic improvements even in those samples with greater coverage (unpublished data).
Upon manual inspection, most assembly-breaking conflicts were found to be local in nature.
These observations suggested that reducing the degree of sequence identity required for
assembly could ameliorate both factors limiting assembly: effective coverage would
increase and many minor conflicts would be resolved.
Accordingly, we produced a series of assemblies based on 98%, 94%, 90%, 85%, and 80%
identity overlaps for two subsets of the GOS dataset, again using the Celera Assembler.
Assembly lengths increased as the overlap cutoff decreased from 98% to 94% to 90%, and
then leveled off or even dropped as stringency was reduced below 90% (Table 3). Although larger assemblies
could be generated using lower identity overlaps, significant numbers of overlaps
satisfying the chosen percent identity cutoff still went unused in each assembly. This is
consistent with a high rate of conflicting overlaps and in turn diagnostic of significant
polymorphism.
In mammalian sequencing projects the use of larger insert libraries is critical to
producing larger assemblies because of their ability to span repeats or local polymorphic
regions [23]. The shotgun sequencing
libraries from the GOS filters were typically constructed from inserts shorter than 2 kb.
Longer plasmid libraries were attempted but were much less stable. We obtained paired-end
sequences from 21,419 fosmid clones (average insert size, 36 kb; [24,25]) from the 0.1-micron fraction of GS-33. The effect of these long mate pairs
on the GS-33 assembly was quite dramatic, particularly at high stringency (e.g., improving
the largest scaffold from 70 kb to 1,247 kb and the largest contig from 70 kb to 427 kb).
At least for GS-33 this suggests that many of the polymorphisms affect small, localized
regions of the genome that can be spanned using larger inserts. This degree of improvement
may be greater than what could be expected in general, as the diversity of GS-33 is by far
the lowest of any of the currently sequenced GOS samples, yet it clearly indicates the
utility of including larger insert libraries for assembly.
In the absence of substantial assembly, direct comparison of the GOS sequencing data to
the genomes of sequenced microbes is an alternative way of providing context, and also
allows for exploration of genetic variation and diversity. A large and growing set of
microbial genomes are available from the National Center for Biotechnology Information
(NCBI; http://www.ncbi.nlm.nih.gov). At the time of this study,
we used 334 finished and 250 draft microbial genomes as references for comparison with the
GOS sequencing reads. Comparisons were carried out in nucleotide-space using the sequence
alignment tool BLAST [26]. BLAST
parameters were designed to be extremely lenient so as to detect even distant similarities
(as low as 55% identity). A large proportion of the GOS reads, 70% in all, aligned to one
or more genomes under these conditions. However, many of the alignments were of low
identity and used only a portion of the entire read. Such low-quality hits may reflect
distant evolutionary relationships, and therefore less information is gained based on the
context of the alignment. More stringent criteria could be imposed requiring that the
reads be aligned over nearly their entire length without any large gaps. Using this
stringent criterion only about 30% of the reads aligned to any of the 584 reference
genomes. We refer to these fully aligned reads as “recruited reads.” Recruited reads are
far more likely to be from microbes closely related to the reference sequence (same
species) than are partial alignments. Despite the large number of microbial genomes
currently available, including a large number of marine microbes, these results indicate
that a substantial majority of GOS reads cannot be specifically related to available
microbial genomes.
The amount and distribution of reads recruited to any given genome provides an indication
of the abundance of closely related organisms. Only genomes from the five bacterial genera
Prochlorococcus, Synechococcus, Pelagibacter, Shewanella, and
Burkholderia yielded substantial and uniform recruitment of GOS
fragments over most of a reference genome (Table 4). These genera include multiple reference genomes, and we observed
significant differences in recruitment patterns even between organisms belonging to the
same species (Figure 2A–2I). Three genera,
Pelagibacter (Figure
2A), Prochlorococcus (Figure 2B–2F), and Synechococcus (Figure 2G–2I), were found abundantly in a wide range of samples and together accounted for
roughly 50% of all the recruited reads (though only ~15% of all GOS sequencing reads). By
contrast, although every genome tested recruited some GOS reads, most recruited only a
small number, and these reads clustered at lower identity to locations corresponding to
large highly conserved genes (for typical examples see Figure 2E–2F). We refer to this pattern as nonspecific recruitment as it reflects
taxonomically nonspecific signals, with the reads in question often recruiting to
distantly related sets of genomes. Most microbial genomes, including many of the marine
microbes (e.g., the ubiquitous genus Vibrio), demonstrated this
nonspecific pattern of recruitment.
The relationship between the similarity of an individual sequencing read to a given
genome and the sample from which the read was isolated can provide insight into the
structure, evolution, and geographic distribution of microbial populations. These
relationships were assessed by constructing a “percent identity plot” [27] in which the alignment of a read to a
reference sequence is shown as a bar whose horizontal position indicates location on the
reference and whose vertical position indicates the percent identity of the alignment. We
colored the plotted reads according to the samples to which they belonged, thus indirectly
representing various forms of metadata (geographic, environmental, and laboratory
variables). We refer to these plots that incorporate metadata as fragment recruitment
plots. Fragment recruitment plots of GOS sequences recruited to the entire genomes of
Pelagibacter
ubique HTCC1062, Prochlorococcus marinus MIT9312, and
Synechococcus WH8102 are presented in Poster S1.
Characteristic patterns of recruitment emerged from each of these abundant marine
microbes consisting of horizontal bands made up of large numbers of GOS reads. These bands
seem constrained to a relatively narrow range of identities that tile continuously (or at
least uniformly, in the case when abundance/coverage is lower) along ~90% of the reference
sequence. The uninterrupted tiling indicates that environmental genomes are largely
syntenic with the reference genomes. Multiple bands, distinguished by degree of similarity
to the reference and by sample makeup, may arise on a single reference (Poster S1D and S1F). Each of these
bands appears to represent a distinct, closely related population we refer to as a
subtype. In some cases, an abundant subtype is highly similar to the reference genome, as
is the case for P.
marinus MIT9312 (Poster S1) and Synechococcus RS9917
(unpublished data). P.
ubique HTCC1062 and other Synechococcus strains like
WH8102 show more complicated banding patterns (Poster S1D and S1F) because of the presence of multiple subtypes that
produce complex often overlapping bands in the plots. Though the recruitment patterns can
be quite complex they are also remarkably consistent over much of the reference genome. In
these more complicated recruitment plots, such as the one for P. ubique HTCC1062,
individual bands can show sudden shifts in identity or disappear altogether, producing a
gap in recruitment that appears to be specific to that band (see P. ubique recruitment
plots on Poster
S1B and S1E, and specifically between 130–140 kb). Finally, phylogenetic analysis
indicates that separate bands are indeed evolutionarily distinct at randomly selected
locations along the genome.
The amount of sequence variation within a given band cannot be reliably determined from
the fragment recruitment plots themselves. To examine this variation, we produced multiple
sequence alignments and phylogenies of reads that recruited to several randomly chosen
intervals along given reference genomes to show that there can be considerable
within-subtype variation (Figure
3A–3B). For example, within
the primary band found in recruitment plots to P. marinus MIT9312, individual pairs of overlapping
reads typically differ on average between 3%–5% at the nucleotide level (depending on
exact location in the genome). Very few reads that recruited to MIT9312 have perfect
(mismatch-free) overlaps with any other read or to MIT9312, despite ~100-fold coverage.
While many of these differences are silent (i.e., do not change amino acid sequences),
there is still considerable variation at the protein level (unpublished data). The amount
of variation within subtypes is so great that it is likely that no two sequenced cells
contained identical genomes.
Variation in genome structure in the form of rearrangements, duplications, insertions, or
deletions of stretches of DNA can also be explored via fragment recruitment. The use of
mated sequencing reads (pairs of reads from opposite ends of a clone insert) provides a
powerful tool for assessing structural differences between the reference and the
environmental sequences. The cloning and sequencing process determines the orientation and
approximate distance between two mated sequencing reads. Genomic structural variation can
be inferred when these are at odds with the way in which the reads are recruited to a
reference sequence. Relative location and orientation of mated sequences provide a form of
metadata that can be used to color-code a fragment recruitment plot (Figure 4). This makes it possible to visually identify
and classify structural differences and similarities between the reference and the
environmental sequences (Figure 5).
For the abundant marine microbes, a high proportion of mated reads in the “good” category
(i.e., in the proper orientation and at the correct distance) show that synteny is
conserved for a large portion of the microbial population. The strongest signals of
structural differences typically reflect a variant specific to the reference genome and
not found in the environmental data. In conjunction with the requirement that reads be
recruited over their entire length without interruption, recruitment plots result in
pronounced recruitment gaps at locations where there is a break in synteny. Other
rearrangements can be partially present or penetrant in the environmental data and thus
may not generate obvious recruitment gaps. However, given sufficient coverage, breaks in
synteny should be clearly identifiable using the recruitment metadata based on the
presence of “missing” mates (i.e., the mated sequencing read that was recruited but whose
mate failed to recruit; Figure 4). The
ratio of missing mates to “good” mates determines how penetrant the rearrangement is in
the environmental population.
In theory, all genome structure variations that are large enough to prevent recruitment
can be detected, and all such rearrangements will be associated with missing mates.
Depending on the type of rearrangement present other recruitment metadata categories will
be present near the rearrangements' endpoints. This makes it possible to distinguish among
insertions, deletions, translocations, inversions, and inverted translocations directly
from the recruitment plots. Examples of the patterns associated with different
rearrangements are presented in Figure
5. This provides a rapid and easy visual method for exploring structural
variation between natural populations and sequenced representatives (Poster S1A and S1B).
Variation in genome structure potentially results in functional differences. Of
particular interest are those differences between sequenced (reference) microbes and
environmental populations. These differences can indicate how representative a cultivated
microbe might be and shed light on the evolutionary forces driving change in microbial
populations. Fragment recruitment in conjunction with the mate metadata helped us to
identify both the consistent and the rare structural differences between the genomes of
microbial populations in the GOS data and their closest sequenced relatives. Our analysis
has thus far been confined to the three microbial genera that were widespread in the GOS
dataset as represented by the finished genomes of P. marinus MIT9312,
P.
ubique HTCC1062, and to a lesser extent Synechococcus
WH8102. Each of these genomes is characterized by large and small segments where little or
no fragment recruitment took place. We refer to these segments as “gaps.” These gaps
represent reference-specific differences that are not found in the environmental
populations rather than a cloning bias that identifies genes or gene segments that are
toxic or unclonable in E. coli. The presence of missing mates flanking
these gaps indicates that the associated clones do exist, and therefore that cloning
issues are not a viable explanation for the absence of recruited reads. Although the
reference-specific differences are quite apparent due to the recruitment gaps they
generate, there are also sporadic rearrangements associated with single clones, mostly
resulting from small insertions or deletions.
Careful examination of the unrecruited mates of the reads flanking the gaps allowed us to
identify, characterize, and quantify specific differences between the reference genome and
their environmental relatives. The results of this analysis for P. ubique and
P.
marinus have been summarized in Table 5. With few exceptions, small gaps resulted from
the insertion or deletion of only a few genes. Many of the genes associated with these
small insertions and deletions have no annotated function. In some cases the insertions
display a degree of variability such that different sets of genes are found at these
locations within a portion of the population. In contrast, many of the larger gaps are
extremely variable to the extent that every clone contains a completely unrelated or
highly divergent sequence when compared to the reference or to other clones associated
with that gap. These segments are hypervariable and change much more rapidly than would be
expected given the variation in the rest of the genome. Sites containing a hypervariable
segment nearly always contained some insert. We identified two exceptions both associated
with P. ubique. The first is approximately located at the 166-kb position
in the P.
ubique HTCC1062 genome. Though no large gap is present, the mated reads
indicate that under many circumstances a highly variable insert is often present. The
second is a gap on HTCC1062 that appears between 50 and 90 kb. This gap appears to be less
variable than other hypervariable segments and is occasionally absent based on the large
numbers of flanking long mated reads (Poster S1A). Interestingly, the long mated reads around
this gap seem to be disproportionately from the Sargasso Sea samples, suggesting that this
segment may be linked to geographic and/or environmental factors. Thus, hypervariable
segments are highly variable even within the same sample, can on occasion be unoccupied,
and the variation, or lack thereof, can be sample dependent.
Hypervariable segments have been seen previously in a wide range of microbes, including
P.
marinus [28], but
their precise source and functional role, especially in an environmental context, remains
a matter of ongoing research. For clues to these issues we examined the genes associated
with the missing mates flanking these segments and the nucleotide composition of the
gapped sequences in the reference genomes. In some rare cases the genes identified on
reads that should have recruited within a hypervariable gap were highly similar to known
viral genes. For example, a viral integrase was associated with the P. ubique HTCC1062
hypervariable gap between 516 and 561 kb. However, in the majority of cases the genes
associated with these gaps were uncharacterized, either bearing no similarity to known
genes or resembling genes of unknown function. If these genes were indeed acquired through
horizontal transfer then we might expect that they would have obvious compositional
biases. Oligonucleotide frequencies along the P. ubique HTCC1062 and
Synechococcus WH8102 genomes are quite different in the large
recruitment gaps in comparison to the well-represented portions of the genome (Poster S1).
Surprisingly, this was less true for P. marinus MIT9312, where the gaps have been linked
to phage activity [28]. These results
suggest that these hypervariable segments of the genome are widespread among marine
microbial populations, and that they are the product of horizontal transfer events perhaps
mediated by phage or transposable elements. These results are consistent with and expand
upon the hypothesis put forward by Coleman et al. [28] suggesting that these segments are phage mediated,
and conflicts with initial claims that the HTCC1062 genome was devoid of genes acquired by
horizontal transfer [29].
Though insertions and deletions accounted for many of the obvious regions of structural
variation, we also looked for rearrangements. The high levels of local synteny associated
with P.
ubique and P. marinus suggested that large-scale rearrangements
were rare in these populations. To investigate this hypothesis we used the recruitment
data to examine how frequently rearrangements besides insertions and deletions could be
identified. We looked for rearrangements consisting of large (greater than 50 kb)
inversions and translocations associated with P. marinus; however, we did
not identify any such rearrangements that consistently distinguished environmental
populations from sequenced cultivars. Rare inversions and translocations were identified
in the dominant subtype associated with MIT9312 (Table 6). Based on the amount of sequence that
contributed to the analysis, we estimate that one inversion or translocation will be
observed for every 2.6 Mbp of sequence examined (less than once per P. marinus genome).
A further observation concerns the uniformity along a genome of the evolutionary history
among and within subtypes. For instance, the similarity between GOS reads and
P.
marinus MIT9312 is typically 85%–95%, while the similarity between
MIT9312 and P.
marinus MED4 is generally ~10% lower. However, there are several
instances where the divergence of MIT9312 and MED4 abruptly decreases to no more than that
between the GOS sequences and MIT9312 (Poster S1G). These results are consistent either with
horizontal transfer (recombination) or with inhomogeneous selectional pressures. Similar
patterns are present in the two high-identity subtypes seen on the P. ubique HTCC1062
genome (Poster
S1D). Other regions show local increases in similarity between MIT9312 and the
dominant subtype that are not reflected in the MIT9312/MED4 divergence (e.g., near
positions 50 kb, 288 kb, 730 kb, 850 kb, and 954 kb on MIT9312; also see Poster S1G). These
latter regions might reflect either regions of homogenizing recombination or regions of
higher levels of purifying selection. However, the lengths of the intervals (several are
10 kb or more) are longer than any single gene and correspond to genes that are not
extremely conserved over greater taxonomic distances (in contrast to the ribosomal RNA
operon). Equally, if widespread horizontal transfer of an advantageous segment explains
these intervals, the transfers occurred long enough ago for appreciable variation to
accumulate (unpublished data).
The analyses described above have been confined to those organisms with representatives
in culture and for which genomes were readily available. Producing assemblies for other
abundant but uncultivated microbial genera would provide valuable physiological and
biochemical information that could eventually lead to the cultivation of these organisms,
help elucidate their role in the marine community, and allow similar analyses of their
evolution and variation such as those performed on sequenced organisms. Previous assembly
efforts and the fragment recruitments plots showed that there is considerable and in many
cases conflicting variation among related organisms. Such variation is known to disrupt
whole-genome assemblers. This led us to try an assembly approach that aggressively
resolves conflicts. We call this approach “extreme assembly” (see Materials and Methods). This approach currently does not make use of
mate-pairing data and, therefore produces only contigs, not scaffolded sequences. Using
this approach, contigs as large as 900 kb could be aligned almost in their entirety to the
P.
marinus MIT9312 and P. ubique HTCC1062 genomes (Figure 2J–2L). Consistent patterns of fragment recruitment (see below) generally provided
evidence of the correctness of contigs belonging to otherwise-unsequenced organisms.
Accordingly, large contigs from these alternate assemblies were used to investigate
genetic and geographic population structure, as described below. However, the more
aggressive assemblies demonstrably suffered from higher rates of assembly artifacts,
including chimerism and false consensus sequences (Figure 6). Thus, the more stringent primary assembly was
employed for most assembly-based analyses, as manual curation was not practical.
As just noted, many of the large contigs produced by the more aggressive assembly methods
described above did not align to any great degree with known genomes. Some could be
tentatively classified based on contained 16S sequences, but the potential for
computationally generated chimerism within the rRNA operon is sufficiently high that
inspection of the assembly or other means of confirming such classifications is essential.
An alternative to an unguided assembly that facilitates the association of assemblies with
known organisms is to start from seed fragments that can be identified as belonging to a
particular taxonomic group. We employed fragments outside the ribosomal RNA operon that
were mated to a 16S-containing read, limiting extension to the direction away from the 16S
operon. This produced contigs of 100 kb or more for several of the ribotypes that were
abundant in the GOS dataset. When evaluated via fragment recruitment (Figure 2M–2O), these assemblies revealed patterns analogous to those seen for the
sequenced genomes described above: multiple subtypes could be distinguished along the
assembly, differing in similarity to the reference sequence and sample distribution, with
occasional gaps. Hypervariable segments by definition were not represented in these
assemblies, but they may help explain the termination of the extreme assemblies for
P.
marinus and SAR11 and provide a plausible explanation for termination of
assemblies of the other deeply sampled populations as well.
This directed approach to assembly can also be used to investigate variation within a
group of related organisms (e.g., a 16S ribotype). We explored the potential to assemble
distinct subtypes of SAR11 by repeatedly seeding extreme assembly with fragments mated to
a SAR11-like 16S sequence. Figure 7
compares the first 20 kb from each of 24 independent assemblies. Eighteen of these
segments could be aligned full-length to a portion of the HTCC1062 genome just upstream of
16S, while six appeared to reflect rearrangements relative to HTCC1062. The rearranged
segments were associated with more divergent 16S sequences (8%–14% diverged from the 16S
of HTCC1062), while those without rearrangements corresponded to less divergent 16S
(averaging less than 3% different from HTCC1062). In each segment, many reads were
recruited above 90% identity, but different samples dominated different assemblies.
Phylogenetic trees support the inference of evolutionarily distinct subtypes with
distinctive sample distributions (Figure
8).
Environmental surveys provide a cultivation-independent means to examine the diversity
and complexity of an environmental sample and serve as a basis to compare the populations
between different samples. Typically, these surveys use PCR to amplify ubiquitous but
slowly evolving genes such as the 16S rRNA or recA genes. These in turn
can be used to distinguish microbial populations. Since PCR can introduce various biases,
we identified 16S genes directly from the primary GOS assembly. In total, 4,125 distinct
full-length or partial 16S were identified. Clustering of these sequences at 97% identity
gave a total of 811 distinct ribotypes. Nearly half (48%) of the GOS ribotypes and 88% of
the GOS 16S sequences were assigned to ribotypes previously deposited in public databases.
That is, more than half the ribotypes in the GOS dataset were found to be novel at what is
typically considered the species level [30]. The overall taxonomic distribution of the GOS ribotypes sampled by shotgun
sequencing is consistent with previously published PCR based studies of marine
environments (Table 7) [31]. A smaller amount (16%) of GOS
ribotypes and 3.4% of the GOS 16S sequences diverged by more than 10% from any publicly
available 16S sequence, thus being novel to at least the family level.
A census of microbial ribotypes allows us to identify the abundant microbial lineages and
estimate their contribution to the GOS dataset. Of the 811 ribotypes, 60 contain more than
8-fold coverage of the 16S gene (Table
8); jointly, these 60 ribotypes accounted for 73% of all the 16S sequence data.
All but one of the 60 have been detected previously, yet only a few are represented by
close relatives with complete or nearly complete genome sequencing projects (see Fragment
Recruitment for further details). Several other abundant 16S sequences belong to
well-known environmental ribotypes that do not have cultivated representatives (e.g.,
SAR86, Roseobacter NAC-1–2, and branches of SAR11 other than those
containing P.
ubique). Interestingly, archaea are nearly absent from the list of
dominant organisms in these near-surface samples.
The distribution of these ribotypes reveals distinct microbial communities (Figure 9 and Table 8). Only a handful of the ribotypes appear to be
ubiquitously abundant; these are dominated by relatives of SAR11 and SAR86. Many of the
ribotypes that are dominant in one or more samples appear to reside in one of three
separable marine surface habitats. For example, several SAR11, SAR86, and alpha
Proteobacteria, as well as an Acidimicrobidae group,
are widespread in the surface waters, while a second niche delineated by tropical samples
contains several different SAR86, Synechococcus and
Prochlorococcus (both cyanobacterial groups), and a
Rhodospirillaceae group. Other ribotypes related to
Roseobacter RCA, SAR11, and gamma Proteobacteria are
abundant in the temperate samples but were not observed in the tropical or Sargasso
samples. Not surprisingly, samples taken from nonmarine environments (GS33, GS20, GS32),
estuaries (GS11, GS12), and larger-sized fraction filters (GS01a, GS01b, GS25) have
distinguishing ribotypes. Furthermore, as the complete genomes of these dominant members
are obtained, the capabilities responsible for their abundances may well lend insight into
the community metabolism in various oceanic niches.
The most common approach for comparing the microbial community composition across samples
has been to examine the ribotypes present as indicated by 16S rRNA genes or by analyzing
the less-conserved ITS located between the 16S and 23S gene sequences [7,8,16,17]. However, a clear
observation emerging from the fragment recruitment views was that the reference ribotypes
recruit multiple subtypes, and that these subtypes were distributed unequally among
samples (Figures 2, 7, 8; Poster S1D, S1F, and S1I).
We developed a method to assess the genetic similarity between two samples that
potentially makes use of all portions of a genome, not just the 16S rRNA region. This
similarity measure is assembly independent; under certain circumstances, it is equivalent
to an estimate of the fraction of sequence from one sample that could be considered to be
in the other sample. Whole-metagenomic similarities were computed for all pairs of
samples. Results are presented for comparisons at ≥98% and 90% identity. No universal
cutoff consistently divides sequences into natural subsets, but the 98% identity cutoff
provides a relatively high degree of resolution, while the 90% cutoff appears to be a
reasonable heuristic for defining subtypes. For instance, a 90% cutoff treats most of the
reads specifically recruited to P. marinus MIT9312 as similar (those more similar to
MED4 notably excepted), while reasonably separating clades of SAR11 (Figures 7 and 8). Reads with no qualifying overlap alignment to any
other read in a pair of samples are uninformative for this analysis, as they correspond to
lineages that were so lightly sequenced that their presence in one sample and absence in
another may be a matter of chance. For the 90% cutoff, 38% of the sequence reads
contributed to the analysis. The resulting similarities reveal clear and consistent
groupings of samples, as well as the outlier status of certain samples (Figures 10 and 11).
The broadest contrast was between samples that could be loosely labeled “tropical”
(including samples from the Sargasso Sea [GS00b, GS00c, GS00d] and samples that are
temperate by the formal definition but under the influence of the Gulf Stream [GS14,
GS15]) and “temperate.” Further subgroups can be identified within each of these
categories, as indicated in Figures 10
and 11. In some cases, these groupings
were composed of samples taken from different ocean basins during different legs of the
expedition. A few pairs of samples with strikingly high similarity were observed,
including GS17 and GS18, GS23 and GS26, GS27 and GS28, and GS00b and GS00d. In each case,
these pairs of samples were collected from consecutive or nearly consecutive samples.
However, the same could be said of many other pairs of samples that do not show this same
degree of similarity. Indeed, geographically and temporally separated samples taken in the
Atlantic (GS17, GS18) and Pacific (GS23, GS26) during separate legs of the expedition are
more similar to one another than were most pairs of consecutive samples. The samples with
least similarity to any other sample were from unique habitats. Thus, similarity cannot be
attributed to geographic separation alone.
The groupings described above can be reconstructed from taxonomically distinct subsets of
the data. Specifically, the major groups of samples visible in Figure 10 were reproduced when sample similarities were
determined based only on fragments recruiting to P. ubique HTCC1062
(unpublished data). Likewise, the same groupings were observed when the fragments
recruiting to either HTCC1062 or P. marinus MIT9312, or both, were excluded from the
calculations (unpublished data). Thus, the factors influencing sample similarities do not
appear to rely solely on the most abundant organisms but rather are reflected in multiple
microbial lineages.
It is tempting to view the groups of similar samples as constituting community types.
Sample similarities based on genomic sequences correlated significantly with differences
in the environmental parameters (Table
1), particularly water temperature and salinity (unpublished data). Samples that
are very similar to each other had relatively small differences in temperature and
salinity. However, not all samples that had similar temperature and salinity had high
community similarities. Water depth, primary productivity, fresh water input, proximity to
land, and filter size appeared consistent with the observed groupings. Other factors such
as nutrients and light for phototrophs and fixed carbon/energy for chemotrophs may
ultimately prove better predictors, but these results demonstrate the potential of using
metagenomic data to tease out such relationships.
Examining the groupings in Figure 11
in light of habitat and physical characteristics, the following may be observed. The first
two samples, a hypersaline pond in the Galapagos Islands (GS33) and the freshwater Lake
Gatun in the Panama Canal (GS20) are quite distinct from the rest. Salinity—both higher
and lower than the remaining coastal and ocean samples—is the simplest explanation.
Twelve samples form a strong temperate cluster as seen in the similarity matrix of Figure 11 as a darker square bounded by
GS06 and GS12. Embedded within the temperate cluster are three subclusters. The first
subcluster includes five samples from Nova Scotia through the Gulf of Maine. This is
followed by a subcluster of four samples between Rhode Island and North Carolina. The
northern subcluster was sampled in August, the southern subcluster in November and
December. Though all samples were collected in the top few meters, the southern samples
were in shallower waters, 10 to 30 m deep, whereas most of the northern samples were in
waters greater than 100 m deep. Monthly average estimates of chlorophyll
a concentrations were typically higher in the southern samples as well
(Table 1). All of these
factors—temperature, system primary production, and depth of the sampled water body—likely
contribute to the differences in microbial community composition that result in the two
well-defined clusters. The final temperate subgroup includes two estuaries, Chesapeake Bay
(GS12) and Delaware Bay (GS11), distinguished by their lower salinity and higher
productivity. However, GS11 is markedly similar not only to GS12 but also to coastal
samples, whereas the latter appears much more unique. Interestingly, the Bay of Fundy
estuary sample (GS06) clearly did not group with the two other estuaries, but rather with
the northern subgroup, perhaps reflecting differences in the rate or degree of mixing at
the sampling site.
Continuing to the right and downward in Figure 11, one can see a large cluster of 25 samples from the tropics and
Sargasso Sea, bounded by GS47 and GS00b. This can be further subdivided into several
subclusters. The first subcluster (a square bounded by GS47 and GS14) includes 14 samples,
about half of which were from the Galapagos. The second distinct subcluster (a square
bounded by GS16 and GS26) includes seven samples from Key West, Florida, in the Atlantic
Ocean to a sample close to the Galapagos Islands in the Pacific Ocean. Loosely associated
with this subcluster is a sample from a larger filter size taken en route to the Galapagos
(GS25). The remaining samples group weakly with the tropical cluster. GS32 was taken in a
coastal mangrove in the Galapagos. The thick organic sediment at a depth of less than a
meter is the likely cause for it being unlike the other samples. Sample 00a was from the
Sargasso Sea and contained a large fraction of sequence reads from apparently clonal
Burkholderia and Shewanella species that are atypical.
When this sample is reanalyzed to exclude reads identified as belonging to these two
groups, sample GS00a groups loosely with GS00b, GS00c, and GS00d (unpublished data).
Finally, three subsamples from a single Sargasso sample (GS01a, GS01b, GS01c) group
together, despite representing three distinct size fractions (3.0–20, 0.8–3.0, and 0.1–0.8
μm, respectively; Table 1).
The complete set of sample similarities is more complex than described above, and indeed
is more complex than can be captured by a hierarchical clustering. For instance, the
southern temperate samples are appreciably more similar to the tropical cluster than are
the northern temperate samples. GS22 appears to constitute a mix of tropical types,
showing strong similarity not only to the GS47–GS14 subcluster to which it was assigned,
but also to the other tropical samples.
These results may be compared to the more traditional view of community structure
afforded by 16S sequences (Figure 9).
Some of the same groupings of samples are visible using both analyses. Several ribotypes
recapitulated the temperate/tropical clustering described above. Others were restricted to
the single instances of nonmarine habitats. Several of the most abundant organisms from
the coastal mangrove, hypersaline lagoon, and freshwater lake were found exclusively in
these respective samples. However, while several ribotypes recapitulated the
temperate/tropical distinction revealed by the genomic sequence, others crosscut it. A few
dominant 16S ribotypes, related to SAR11, SAR86, and SAR116, were found in every marine
sample. The brackish waters from two mid-Atlantic estuaries (GS11 and GS12) contained a
mixture of otherwise exclusively marine and freshwater ribotypes; similarity of these
sites to the freshwater sample (GS20) was minimal at the metagenomic level, while the
greater similarity of GS11 to coastal samples visible at the metagenomic level was not
readily visible here. A fuller comparison of metagenome-based measurements of diversity
based on a large dataset of PCR-derived 16S sequences will be presented in another paper
(in preparation).
Differences in gene content between samples can identify functions that reflect the
lifestyles of the community in the context of its local environment [20,32]. We examined the relative abundance of genes belonging to specific
functional categories in the distinct GOS samples. Genes were binned into functional
categories using TIGRFAM hidden Markov models [18], which are well annotated and manually curated [33].
The results can be filtered in various ways to highlight genes associated with specific
environments. One catalog of possible interest is genes that were predominantly found in a
single sample. We identified 95 TIGRFAMs that annotated large sets of genes (100 or more)
that were significantly more frequent (greater than 2-fold) in one sample than in any
other sample (Table 9). Not
surprisingly, this approach disproportionately singles out genes from the samples
collected on larger filters (GS01a, GS01b, and GS25) and from the nonmarine environments,
particularly the hypersaline pond (sample GS33). Another contrast might be between the
temperate and tropical clusters (Figures
10 and 11). We identified 32
proteins that were more than 2-fold more frequent in one or the other group (Table 10). The presence of various
Prochlorococcus-associated genes in this list highlights some of the
potential challenges with this sort of approach. Overrepresentation may reflect: a direct
response to particular environmental pressures (as the excess of salt transporters
plausibly do in the hypersaline pond); a lineage-restricted difference in functional
repertoire (as exemplified by the excess of photosynthesis genes in samples containing
Prochlorococcus); or a more incidental “hitchhiking” of a protein found
in a single organism that happens to be present.
We explored whether clearer and more informative differences could be discovered between
communities by focusing on groups of samples that are highly similar in overall
taxonomic/genetic content. Two pairs of samples provide a particularly nice illustration
of this approach. Samples GS17 and GS18 from the western Caribbean Sea and samples GS23
and GS26 from the eastern Pacific Ocean were all very similar based on the presence of
abundant ribotypes and overall similarity in genetic content (Figures 9–11). Despite these similarities, several genes are found to be up to seven times
more common in the pair of Caribbean samples than the Pacific pair (Table 11). No genes are more than 2-fold higher in the
Pacific than the Caribbean pair of samples. Several of the most differentially abundant
genes are related to phosphate transport and utilization. It is very plausible that this
is a reflection of a functional adaptation: these differences correlate well with measured
differences in phosphate abundance between the Atlantic and eastern Pacific samples [34,35], and phosphate abundance plays a critical role in
microbial growth [36,37]. Indeed, the ability to acquire
phosphate, especially under conditions where it is limited, is thought to determine the
relative fitness of Prochlorococcus strains [38].
The single greatest difference between GS17 and GS18 on the one hand and GS23 and GS26 on
the other was attributed to a set of genes annotated by the hidden Markov model TIGR02136
as a phosphate-binding protein (PstS). This TIGRFAM identified a single
gene in both P.
marinus MIT9312 and P. ubique HTCC1062. In P. marinus MIT9312,
this gene is located at 672 kb lying roughly in the middle of a 15-kb segment of the
genome that recruits almost no GOS sequences from the Pacific sampling sites (Poster S1H). In
P.
ubique HTCC1062, the PstS gene is found at 1,133 kb in
a 5-kb segment that also recruited far fewer GOS sequences from all the Pacific samples
except for GS51 (Poster
S1E). These genomic segments differ structurally among isolates but they are no
more variable than the flanking regions, and thus are not hypervariable in the sense used
previously (unpublished data). Nor are they particularly conserved when present,
indicating that they are not the result of a recent lateral transfer. Phylogenetic
analyses outside these segments did not produce any evidence of a Pacific versus Caribbean
clade of either Prochlorococcus or SAR11 (Figure 3A–3B). The presence or absence of phosphate transporters is not limited to these
two types of organisms. The number of phosphate transporters that were found in the
Caribbean far exceeds the number that can be attributed to HTCC1062- and MIT9312-like
organisms. However, these results indicate that within individual strains or subtypes the
ability to acquire phosphate (in one or more of its forms) can vary without detectable
differences in the surrounding genomic sequences.
Variation in gene content is only one aspect of the tremendous diversity in the GOS data.
The functional significance of all the polymorphic differences between homologous proteins
remains largely unknown. To look for functional differences, we analyzed members of
proteorhodopsin gene family. Proteorhodopsins are fast, light-driven proton pumps for
which considerable functional information is available though their biological role
remains unknown. Proteorhodopsins were highly abundant in the Sargasso Sea samples [19] and continue to be highly abundant
and evenly distributed (relative to recA abundance) in all the GOS
samples. A total of 2,674 putative proteorhodopsin genes were identified in the GOS
dataset. Although many of the sequences are fragmentary, 1,874 of these genes contain the
residue that is primarily responsible for tuning the light-absorbing properties of the
protein [39–41], and these properties have been shown to be selected
for under different environmental conditions [42]. Variation at this residue is strongly correlated
with sample of origin (Figure 12). The
leucine (L) or green-tuned variant was highly abundant in the North Atlantic samples and
in the nonmarine environments like the fresh water sample from Lake Gatun (GS20). The
glutamine (Q) or blue-tuned variant dominated in the remaining mostly open ocean
samples.
Given our limited understanding of the biological role for proteorhopsin, the reason for
this differential distribution is not immediately clear. In coastal waters where nutrients
are more abundant, phytoplankton is dominant. Phytoplankton absorbs primarily in the blue
and red spectra; consequently, the water appears green [43]. Conversely, in the open ocean nutrients are rare and
phytoplanktonic biomass is low, so waters appear blue because in the absence of impurities
the red wavelengths are absorbed preferentially [44]. It may be that proteorhodopsin-carrying microbes
have simply adapted to take advantage of the most abundant wavelengths of light in these
systems.
Proteorhodopsins encoded on reads that were recruited to P. ubique HTCC1062
account for a fraction (~25%) of all the proteorhodopsin-associated reads, suggesting that
the remainder must be associated with a variety of marine microbial taxa (see also [45–47]). Phylogenetic analysis of the SAR11-associated
proteins revealed that each variant has arisen independently at least two times in the
SAR11 lineage (Figure 3C). Consistent
with other findings that proteorhodopsins are widely distributed throughout the microbial
world [48], we conclude that multiple
microbial lineages are responsible for proteorhodopsin spectral variation and that the
abundance of a given variant reflects selective pressures rather than taxonomic effects.
Similar mechanisms seem to be involved in the evolution and diversification of opsins that
mediate color vision in vertebrates [49].
Our results highlight the astounding diversity contained within microbial communities, as
revealed through whole-genome shotgun sequencing carried out on a global scale. Much of this
microbial diversity is organized around phylogenetically related, geographically dispersed
populations we refer to as subtypes. In addition, there is tremendous variation within
subtypes, both in the form of sequence variation and in hypervariable genomic islands. Our
ability to make these observations derived from not only the large volumes of data but also
from the development of new tools and techniques to filter and organize the information in
manageable ways.
Our data demonstrate to an unprecedented degree the nature and evolution of genetic
variation below the species level. Variation can be analyzed in several ways, including
observed differences in sequence, genomic structure, and gene complement. The observed
patterns of variation shed light on the mechanisms by which marine prokaryotes evolve.
Gene synteny seems to be more highly conserved than the nucleotide and protein sequences.
This variation is seen over essentially the entire genome in every abundant group of
organisms sufficiently related for us to recognize a population by fragment recruitment.
(These include, but are not limited to, the organisms shown in Figure 2 and Poster S1.) Notably, we found no evidence of widespread
low-diversity organisms such as B. anthracis [50].
Phylogenetic trees and fragment recruitment plots (Figures 7 and 8) indicate that the variation within a species is not an
unstructured swarm or cloud of variants all equally diverged from one another. Instead,
there are clearly distinct subtypes, in terms of sequence similarity, gene content, and
sample distribution. Similar findings have been shown for specific organisms, based on
evaluation of one or a few loci [2,51–53]. These results rule out certain trivial models of
population history and evolution for what is commonly considered a bacterial “species.”
For instance, it argues against a recent explosive population growth from a single
successful individual (selective sweep) [54]. Equally, it argues against a perfectly mixed population, suggesting instead
some barriers to competition and exchange of genetic material.
In principle, this variation could reflect some combination of physical barriers (true
biogeography), short-term stochastic effects, and/or functional differentiation. Given the
confounding variables of geography, time, and environmental conditions in the current
collection of samples, it is difficult to definitively separate these effects, but various
observations argue for functional differentiation between subtypes (i.e., they constitute
distinct ecotypes). First, individual subtypes may be found in a wide range of locations;
P.
ubique HTCC1062 was isolated in the Pacific Ocean off the coast of
Oregon [55], but closely related
sequences are relatively abundant in our samples taken in the Atlantic Ocean. Second,
geography per se cannot fully explain differences in subtype distributions, as multiple
subtypes are found simultaneously in a single sample. Third, the collection of samples in
which a given subtype was found generally exhibits similar environmental conditions. A
strong independent illustration of this comes from the correlation of temperature with the
distribution of Prochlorococcus subtypes [56]. Fourth, the extensive variation within each subtype
(i.e., the fact that subtypes are not clonal populations) indicates that it cannot be
chance alone that makes genetically similar organisms have similar observed
distributions.
Taken together, these results argue that subtype classification is more informative for
categorizing microbial populations than classification using 16S-based ribotypes, or
fingerprinting techniques based on length polymorphism, such as T-RFLPs [57] or ARISA [58]. For example, the grouping of such disparate
microbial populations under the umbrella P. marinus dilutes the significance of the term
“species.” Indeed, numerous papers have been devoted to comparing and contrasting the
differences and variability in P. marinus isolates to better understand how this
particularly abundant group of organisms has evolved and adapted within the dynamic marine
environment [28,52,56,59–66]. Prior to the widespread use of marker-based
phylogenetic approaches, microbial systematics relied on a wide range of variables to
distinguish microbial populations [67]. Subtypes bring us back to these more comprehensive approaches since they
reflect the influences of a wide range of factors in the context of an entire genome.
Although subtypes are a salient feature of our data, variation within a ribotype does not
stop at the level of subtypes. Variation within subtypes is so extensive that few GOS
reads can be aligned at 100% identity to any other GOS read, despite the deep coverage of
several taxonomic groups. Related findings have been shown for the ITS region in various
organisms [2,51,52], and in a limited number of organisms for individual protein coding and
intergenic regions [2,53,68]. High levels of diversity within the ribotype can be
convincingly demonstrated in the 16S gene itself [69]. The applicability of these results over the entire
genome were recently shown for P. marinus [28] using data from the Sargasso Sea samples taken as a
pilot project for the expedition reported here [19]. We have definitively demonstrated the generality of
these findings, greatly increased our understanding of the minimum number of variants of a
given organism, and shown that these observations apply to the entire genome for a wide
range of abundant taxonomic groups and across a wide range of geographic locations.
Average pairwise differences of several percent between overlapping P. marinus or SAR11
reads imply that this variation did not arise recently. If one uses substitution rates
estimated for E.
coli [70], one
could conclude that on average any two P. marinus cells must have diverged millions of
years ago. Mutational rates are notoriously variable and hard to estimate, and assumptions
of molecular clocks are equally chancy, but clearly within-subtype variants have persisted
side by side for quite some time. This raises a question related to the classic “Paradox
of the Plankton”: how can so many similar organisms have coexisted for so long [71,72]? One explanation, which we favor, is that not only
subtypes but also individual variants are sufficiently different phenotypically to prevent
any one strain from completely replacing all others (discussed further below; see [71] for a recent theoretical treatment).
An alternative is that recombination might prevent selective sweeps within ecotypes, as
proposed by Cohan (reviewed in [73]).
Given the apparent generality of subtypes and intra-subtype variation, it is important to
understand if and how these subpopulations are functionally distinct. At the level of DNA
sequence, a substantial fraction of substitutions are silent in terms of amino acid
sequence, and others may be nonsynonymous but functionally neutral. However, two organisms
that differ by 5% in their genetic sequence (e.g., 100,000 substitutions in 2 Mbp of
shared sequence) will inevitably have at least minor functional differences such as in the
optimal temperature or pH for the activity of some enzyme. At the level of gene content,
the observation of hypervariable segments ([28] and here) implies that there is an additional dimension to functional
variability. Hypervariable genomic islands with preferential insertion sites could
potentially be associated with a wide range of functions, though to date they have been
most closely examined for their role in pathogenicity (for a review, see [74]). However, given their apparent
variability within even a single sampling site, it seems unlikely that these elements
reflect a specific adaptive advantage to the local population. Identifying the source(s),
diversity, and range of functionality associated with these islands by fully sequencing a
large number of these segments and understanding how their individual abundances fluctuate
should be quite informative.
Some might still argue that these differences must be moot for the purpose of
understanding the role these organisms play in an ecosystem. Yet even small differences in
optimal conditions may have profound effects. They may prevent any single genotype from
being universally fittest, allowing and/or necessitating the coexistence of multiple
variants [2,51,69]. Moreover, variation within subtype might afford a form of functional
“buffering,” such that the population as a whole may be more stable in its ecosystem role
than any one clone could be (see also [51]). That is, while any one strain of Prochlorococcus might
thrive and provide energy input to the rest of the community at a limited range of
temperatures, light conditions, etc., the ensemble might provide such inputs over a wider
range of environmental conditions. In this way, microdiversity might provide system
stability or robustness through functional redundancy and the “insurance effect” (reviewed
in [75]). Thus, while the extent of
microdiversity suggests that knowing the behavior of any one isolate in exquisite detail
might not be as useful to reductionist modeling as one might hope, this buffering could
afford a more stable ensemble behavior, facilitating the development and maintenance of an
ecosystem and allowing for system-level modeling.
A direct equation of subtypes with ecotypes is tempting, but not entirely clear-cut. The
correlation of PstS distribution with phosphate abundance suggests a
functional adaptation, but within Prochlorococcus and SAR11 the presence
or absence of PstS subdivides subtypes without apparent respect for
phylogenetic structure. This contrasts markedly with the distribution of
proteorhodopsin-tuning variants within SAR11, which, despite a few convergent
substitutions, are strongly congruent with phylogeny. It is interesting to ask what
distinguishes pressures or adaptations that respect (or that lead to) lineage splits from
those that show little or no phylogenetic structuring. These two specific examples
plausibly reflect two different mechanisms (i.e., convergent but independent mutation in
proteorhodopsin genes and the acquisition by horizontal transfer of genes involved in
phosphate uptake). Yet, we must wonder: given the evidence that proteorhodopsin has been
transferred laterally [48], and that
only a small number of mutations, in some circumstances even a single base-pair change,
are required to switch between the blue-absorbing and green-absorbing forms [39,40], why should proteorhodopsin variants show any lineage
restriction? Perhaps this relates to the modularity of the system in question:
proteorhodopsin tuning may be part of a larger collection of synergistic adaptations that
are collectively not easily evolved, acquired, or lost, while the PstS
and surrounding genes may represent a functional unit that can be readily added and
removed over relatively short evolutionary time scales. If so, perhaps subtypes are indeed
ecotypes, but rapidly evolving characters can lead to phenotypes that crosscut or
subdivide ecotypes.
Phage provide one possible mechanism for rapid evolution of microbial populations or
strains, and have been found in abundance with this and other marine metagenomic datasets
[18,20]. It has been proposed that hypervariable islands are
phage mediated [28]. However, there
are reasons to be cautious about invoking phage as an explanation for rapidly evolving
characteristics. While we see variability of PstS and neighboring genes
in both SAR11 and P.
marinus populations, this variation does not seem to be linked to recent
phage activity. Initially, the distribution of PstS seems similar to the
variation associated with the hypervariable islands, which may be phage mediated [28]. Indeed, phosphate-regulating genes
including PstS have been identified in phage genomes [64], presumably because enhanced
phosphate acquisition is required during the replication portion of their life cycle.
However, the regions containing the PstS genes in both SAR11 and
Prochlorococcus do not behave in the same fashion as clearly
hypervariable regions, being effectively bimorphic (modulo the level of sequence variation
observed elsewhere in the genome), whereas clearly hypervariable regions are so diverse
that nearly every sampled clone falling in such a region appears completely unrelated to
every other. Nor do the other genes in PstS-containing regions appear to
be phage associated. These observations suggest that differences in PstS
presence or absence arose in the distant past, or that different mechanisms are at work.
It seems likely that phage may mediate lateral transfer of PstS and other
phosphate acquisition genes, but it is unclear whether these genes then can become fixed
within the population. Phage require enhanced phosphate acquisition as part of their life
cycle [64], so regulatory or
functional differences in these genes may limit their suitability for being acquired by
the host cell for its own purposes. The rate of phage-mediated horizontal transfer of
genes may reflect a combination of the gene's value to the host and to the agent mediating
the transfer (e.g., phage), suggesting that PstS may have much greater
immediate value than do proteorhodopsin genes and their variants.
In practical terms, these results highlight the limitations associated with marker-based
analysis and the use of these approaches to infer the physiology of a particular microbial
population. At the resolution used here, marker-based approaches are not always
informative regarding differences in gene content (e.g., the PstS gene as
well as neighboring genes), especially those associated with hypervariable segments.
Though phosphate acquisition is known to vary within different strains of P. marinus [64,76], our results clearly show that this variability can
happen within a single subtype (as represented by MIT9312), effectively identifying
distinct ecotypes. Given the correct samples from the appropriate environments, other core
genes might also show similar variation and allow us to more fully assess the reliability
of reference genomes as indicators of physiological potential.
Analysis of the GOS dataset has benefited from the development of new tools and
techniques. Many of these approaches rely on fairly well-known techniques but have been
modified to take greater advantage of the metadata.
The technique of fragment recruit and the corresponding fragment recruitment plots have
proven highly useful for examining the biogeography and genomic variation of abundant
marine microbes when a close reference genome exists. Ultimately, this approach derives
from the percent identity plots of PipMaker [27]. Similar approaches have been used to examine variation with respect to
metagenomic datasets. For example, hypervariable segments and sequence variation have been
visualized in P.
marinus MIT9312 using the Sargasso Sea data [28] and in human gut microbes Bifidobacterium longum
and Methanobrevibacter
smithii [77].
Our primary advance associated with fragment recruitment plots is the incorporation of
metadata associated with the isolation or production of the sequencing data. While simple
in nature, the resulting plots can be extremely informative due to the volume of data
being presented. Being able to present the sequence similarity and metadata visually
allows a researcher to quickly identify interesting portions of the data for further
examination. This is one of the first tools to make extensive use of the metadata
collected during a metagenomic sequencing project. The use of sample and recruitment
metadata is just the beginning. It is not difficult to imagine displaying other variations
such as water temperature, salinity, phosphate abundance, and time of year with this
approach. Even sample independent metadata such as phylogenetic information may produce
informative views of the data. The usefulness of this and related approaches will only
grow as the robust collection of metadata becomes routine and the variables that are most
relevant to microbial communities are further elucidated.
The greatest limitation of fragment recruitment is the lack of appropriate reference
sequences, particularly finished genomes. Using a series of modifications to the Celera
Assembler referred to as “extreme assembly,” we have produced large assemblies for
cultivated and uncultivated marine microbes. On its own, the extreme assembly approach
would be excessively prone to producing chimeric sequences. However, when extreme
assemblies are used as references for fragment recruitment, the metadata provides
additional criteria to validate the sampling consistency along the length of the scaffold.
Chimeric joins can be rapidly detected and avoided. This argues that future metagenomic
assemblers could be specifically designed to make use of the metadata to produce more
accurate assemblies, and that metagenomic assemblies will be improved by using data from
multiple sources. Finding ways to represent the full diversity in these assemblies remains
a pressing issue.
Extreme assembly can produce much larger assemblies but it is still limited by overall
coverage. While many ribotypes are presumably present in sufficient quantities that
reasonable assemblies of these genomes might be expected, this did not occur even for the
most abundant organisms, including SAR11 and P. marinus. Many of the
problems can be attributed to the diversity associated with the hypervariable segments
where the effective coverage drops precipitously. If these are indeed commonplace in the
microbial world, it is unlikely that complete genomes will be produced using the small
insert libraries presented here. However, the ability to bin the larger sequences based on
their coverage profiles across multiple samples, oligonucleotide frequency profiles, and
phylogenetic markers suggests that large portions of a microbial genome can be
reconstructed from the environmental data. This in turn should provide critical insights
into the physiology and biochemistry of these microbial lineages that will inform culture
techniques to allow cultivation of these recalcitrant organisms under laboratory
conditions.
Not every technique described herein relies on metadata. The marker-less, overlap-based
metagenomic comparison provides a quantitative approach to comparing the overall genetic
similarity of two samples (Figures 10
and 11). In essence, genomic
similarity acts as a proxy for community similarity. Marker-based approaches such as ARISA
including the use of 16S sequences described herein can also be used to infer community
similarity, though these approaches more aptly generate a census of the community members
[51,69,78,79]. This census is
biased to the extent that 16S genes can vary in copy number and relies on linkage of the
marker gene to infer genome composition. While our metagenome comparison does not directly
provide a census, the sensitivity can be tuned by restricting the identity of matches.
This means that even subtype-level differences can be detected across samples. It would
also identify the substantial gene content differences between the K12 and O157:H7
E. coli
strains [12]. Such large-scale gene
content differences have yet to be seen between closely related marine microbes, but may
be a factor in other environments. Although the requisite amount of data will vary with
the complexity of the environment or the degree of resolution required, we have found that
10,000 sequencing reads is sufficient to reliably measure the similarity of two surface
water samples (unpublished data). This analysis may become a general tool for allocating
sequencing resources by allowing a shallow survey of many samples followed by deep
sequencing of a select number of “interesting” ones.
The application of this technique for comparing samples along with detailed analysis of
fragments recruiting to a given reference sequence can also help explicate differences
among communities in gene content or sequence variation. For example, recent metagenomic
studies have reported differences in abundance of various gene families or differing
functional roles between samples. Some of these differences correspond to plausible
differences in physiology and biochemistry, such as the relative overabundance of
photosynthetic or light-responsive genes in surface water samples [20,32]. Other differences however are less obvious, such as the abundance of
ribosomal proteins at 130 m or the abundance of tranposase at 4,000 m [20]; some of these may reflect “taxonomic
hitchhiking,” such that a sample rich in Archaea or Firmicutes or
Cyanobacteria, etc., has an overrepresentation of genes more reflective
of their recent evolutionary history than of a response to environmental conditions. Being
able to control or account for these taxonomic effects is crucial to understanding how
microbial populations have adapted to environmental conditions and how they may behave
under changing conditions. The metagenomic comparison method described here provides a new
tool to more accurately measure the impact of taxonomic effects.
In conclusion, this study reveals the wealth of biological information that is contained
within large multi-sample environmental datasets. We have begun to quantify the amount and
structure of the variation in natural microbial populations, while providing some
information about how these factors are structured along phylogenetic and environmental
factors. At the same time, many questions remain unanswered. For example, although
microbial populations are structured and therefore genetically isolated, we do not
understand the mechanisms that lead to this isolation. Their isolation seems contradictory
given overwhelming evidence that horizontal gene transfer associated with hypervariable
islands is a common phenomenon in marine microbial populations. Whatever the mechanism,
the role and rate at which gene exchange occurs between populations will be crucial to
understanding population structure within microbial communities and whether these
communities are chance associations or necessary collections. The hypervariable islands
could be a source for tremendous genetic innovation and novelty as evidenced by the rate
of discovery of novel protein families in the GOS dataset [18]. However, it is not clear whether these entities are
the main source of this novelty or whether this novelty resides in the vast numbers of
rare microbes [4] that cannot be
practically accessed using current metagenomic approaches. Altogether, this research
reaffirms our growing wealth and complexity of data and paucity of understanding regarding
the biological systems of the oceans.
A more detailed description of the sampling sites provides additional context in which to
understand the individual samples. The northernmost site (GS05) was at Compass Buoy in the
highly eutrophic Bedford Basin, a marine embayment encircled by Halifax, Nova Scotia, that
has a 15-y weekly record of biological, physical, and chemical monitoring (http://www.mar.dfo-mpo.gc.ca/science/ocean/BedfordBasin/index.htm). Other
temperate sites included a coastal station sample near Nova Scotia (GS4), a station in the
Bay of Fundy estuary at outgoing tide (GS06), and three Gulf of Maine stations (GS02,
GS03, and GS07). These were followed by sampling coastal stations from the New England
shelf region of the Middle Atlantic Bight (Newport Harbor through Delaware Bay;
GS08–GS11). The Delaware Bay (GS11) was one of several estuary samples along the Global
Expedition path. Estuaries are complex hydrodynamic environments that exhibit strong
gradients in oxygen, nutrients, organic matter, and salinity and are heavily impacted by
anthropogenic nutrients. The Chesapeake Bay (GS12) is the largest estuary in the United
States and has microbial assemblages that are diverse mixtures of freshwater and
marine-specific organisms [80]. GS13
was collected near Cape Hatteras, North Carolina, inside and north of the Gulf Stream, and
GS14 was taken along the western boundary frontal waters of the Gulf Stream off the coast
of Charleston, South Carolina. The vessel stopped at five additional stations as it
transited through the Caribbean Sea (GS15–GS19) to the Panama Canal. In Panama, we sampled
the freshwater Lake Gatun, which drains into the Panama Canal (GS20). The first of the
eastern Pacific coastal stations GS21, GS22, and GS23 were sampled on the way to Cocos
Island (~500 km southwest of Costa Rica), followed by a coastal Cocos Island sample
(GS25). Near the island, ocean currents diverge and nutrient rich upwellings mix with warm
surface waters to support a highly productive ecosystem. Cocos Island is distinctive in
the eastern Pacific because it belongs to one of the first shallow undersea ridges in the
region encountered by the easterly flowing North Equatorial Counter/Cross Current in the
Far Eastern Pacific [81,82]. After departing Cocos Island, the
vessel continued southwest to the Galapagos Islands, stopping for an open ocean station
(GS26). An intensive sampling program was then conducted in the Galapagos. The Galapagos
Archipelago straddles the equator 960 km west of mainland Ecuador in the eastern Pacific.
These islands are in a hydrographically complex region due to their proximity to the
Equatorial Front and other major oceanic currents and regional front systems [83]. The coastal and marine parts of the
Galapagos Islands ecosystem harbor an array of distinctive habitats, processes, and
endemic species. Several distinct zones were targeted including a shallow-water, warm seep
(GS30), below the thermocline in an upwelling zone (GS31), a coastal mangrove (GS32), and
a hypersaline lagoon (GS33). The last stations were collected from open ocean sites (GS37
and GS47) and a coral reef atoll lagoon (GS51) in the immense South Pacific Gyre. The open
ocean samples come from a region of lower nutrient concentrations where picoplankton are
thought to represent the single most abundant and important factor for biogeochemical
structuring and nutrient cycling [84–87]. In the atoll
systems, ambient nutrients are higher, and bacteria are thought to constitute a large
biomass that is one to three times as large as that of the phytoplankton [88–90].
A YSI (model 6600) multiparameter instrument (http://www.ysi.com) was
deployed to determine physical characteristics of the water column, including salinity,
temperature, pH, dissolved oxygen, and depth. Using sterilized equipment [91], 40–200 l of seawater, depending on
the turbidity of the water, was pumped through a 20-μm nytex prefilter into a 250-l
carboy. From this sample, two 20-ml subsamples were collected in acid-washed polyethylene
bottles and frozen (−20 °C) for nutrient and particle analysis. At each station the
biological material was size fractionated into individual “samples” by serial filtration
through 20-μm, 3-μm, 0.8-μm, and 0.1-μm filters that were then sealed and stored at −20 °C
until transport back to the laboratory. Between 44,160 and 418,176 clones per station were
picked and end sequenced from short-insert (1.0–2.2 kb) sequencing libraries made from DNA
extracted from filters [19]. Data
from these six Sorcerer II expedition legs (37 stations) were combined
with the results from samples in the Sargasso Sea pilot study (four stations; GS00a–GS00d
and GS01a–GS01c; [19]. The majority
of the sequence data presented came from the 0.8- to 0.1-μm size fraction sample that
concentrated mostly bacterial and archaeal microbial populations. Two samples (GS01a,
GS01b) from the Sargasso Sea pilot study dataset and one GOS sample (GS25) came from other
filter size fractions (Table 1).
Microbes were size fractionated by serial filtration through 3.0-μm, 0.8-μm, and 0.1-μm
membrane filters (Supor membrane disc filter; Pall Life Sciences, http://www.pall.com), and finally through a Pellicon tangential flow
filtration (Millipore, http://www.millipore.com) fitted with a Biomax-50
(polyethersulfone) cassette filter (50 kDa pore size) to concentrate a viral fraction to
100 ml. Filters were vacuum sealed with 5 ml sucrose lysis buffer (20 mM EDTA, 400 mM
NaCl, 0.75 M sucrose, 50 mM Tris-HCl [pH 8.0]) and frozen to −20 °C on the vessel until
shipment back to the Venter Institute, where they were transferred to a −80 °C freezer
until DNA extraction. Glycerol was added (10% final concentration) as a cryoprotectant for
the viral/phage sample.
In the laboratory, the impact filters were aseptically cut into quarters for DNA
extraction. Unused quarters of the filter were refrozen at −80 °C for storage. Quarters
used for extraction were aseptically cut into small pieces and placed in individual 50-ml
conical tubes. TE buffer (pH 8) containing 50 mM EGTA and 50 mM EDTA was added until
filter pieces were barely covered. Lysozyme was added to a final concentration of 2.5
mg/ml−1, and the tubes were incubated at 37 °C for 1 h in a shaking water
bath. Proteinase K was added to a final concentration of 200 μg/ml−1, and the
samples were frozen in dry ice/ethanol followed by thawing at 55 °C. This freeze–thaw
cycle was repeated once. SDS (final concentration of 1%) and an additional 200
μg/ml−1 of proteinase K were added to the sample, and samples were incubated
at 55 °C for 2 h with gentle agitation followed by three aqueous phenol extractions and
one phenol/chloroform extraction. The supernatant was then precipitated with two volumes
of 100% ethanol, and the DNA pellet was washed with 70% ethanol. Finally, the DNA was
treated with CTAB to remove enzyme inhibitors. Size fraction samples not utilized in this
study were archived for future analysis.
DNA was randomly sheared via nebulization, end-polished with consecutive BAL31 nuclease
and T4 DNA polymerase treatments, and size-selected using gel electrophoresis on 1%
low-melting-point agarose. After ligation to BstXI adapters, DNA was purified by three
rounds of gel electrophoresis to remove excess adapters, and the fragments were inserted
into BstXI-linearized medium-copy pBR322 plasmid vectors. The resulting library was
electroporated into E. coli. To ensure construction of high-quality
random plasmid libraries with few to no clones with no inserts, and no clones with
chimeric inserts, we used a series of vectors (pHOS) containing BstXI cloning sites that
include several features: (1) the sequencing primer sites immediately flank the BstXI
cloning site to avoid excessive resequencing of vector DNA; (2) elimination of strong
promoters oriented toward the cloning site; and (3) the use of BstXI sites for cloning
facilitates the preparation of libraries with a low incidence of no-insert clones and a
high frequency of single inserts. Clones were sequenced from both ends to produce pairs of
linked sequences representing ~820 bp at the end of each insert.
Libraries were transformed, and cells were plated onto large format (16 × 16cm) diffusion
plates prepared by layering 150 ml of fresh molten, antibiotic-free agar onto a previously
set 50-ml layer of agar containing antibiotic. Colonies were picked for template
preparation using the Qbot or QPix colony-picking robots (Genetix, http://www.genetix.com), inoculated into 384-well blocks containing liquid
media, and incubated overnight with shaking. High-purity plasmid DNA was prepared using
the DNA purification robotic workstation custom-built by Thermo CRS (http://www.thermo.com) and based on the alkaline lysis miniprep [92]. Bacterial cells were lysed, cell
debris was removed by centrifugation, and plasmid DNA was recovered from the cleared
lysate by isopropanol precipitation. DNA precipitate was washed with 70% ethanol, dried,
and resuspended in 10 mM Tris HCl buffer containing a trace of blue dextran. The typical
yield of plasmid DNA from this method is approximately 600–800 ng per clone, providing
sufficient DNA for at least four sequencing reactions per template.
Sequencing protocols were based on the di-deoxy sequencing method [93]. Two 384-well cycle-sequencing reaction plates were
prepared from each plate of plasmid template DNA for opposite-end, paired-sequence reads.
Sequencing reactions were completed using the Big Dye Terminator chemistry and standard
M13 forward and reverse primers. Reaction mixtures, thermal cycling profiles, and
electrophoresis conditions were optimized to reduce the volume of the Big Dye Terminator
mix (Applied Biosystems, http://www.appliedbiosystems.com) and to extend read lengths on the AB3730xl
sequencers (Applied Biosystems). Sequencing reactions were set up by the Biomek FX
(Beckman Coulter, http://www.beckmancoulter.com) pipetting workstations.
Robots were used to aliquot and combine templates with reaction mixes consisting of deoxy-
and fluorescently labeled dideoxynucleotides, DNA polymerase, sequencing primers, and
reaction buffer in a 5 μl volume. Bar-coding and tracking promoted error-free template and
reaction mix transfer. After 30–40 consecutive cycles of amplification, reaction products
were precipitated by isopropanol, dried at room temperature, and resuspended in water and
transferred to one of the AB3730xl DNA analyzers. Set-up times were less than 1 h, and 12
runs per day were completed with average trimmed sequence read length of 822 bp.
Fosmid libraries [24] were
constructed using approximately 1 μg DNA that was sheared using bead beating to generate
cuts in the DNA. The staggered ends or nicks were repaired by filling with dNTPs. A size
selection process followed on a pulse field electrophoresis system with lambda ladder to
select for 39–40 Kb fragments. The DNA was then recovered from a gel, ligated to the
blunt-ended pCC1FOS vector, packaged into lambda packaging extracts, incubated with the
host cells, and plated to select for the clones containing an insert. Sequencing was
performed as described for plasmid ends.
Assembly was conducted with the Celera Assembler [21], with modifications as follows. The “genome length”
was artificially set at the length of the dataset divided by 50 to allow unitigs of
abundant organisms to be treated as unique, as previously described [19]. Several distinct assemblies were computed. In the
primary assembly, all pairs of mated reads were tested to see whether the paired reads
overlapped one another; if so, they were merged into a single pseudo-read that replaced
the two original reads; further, only overlaps of 98% identity or higher were used to
construct unitigs. A second assembly was conducted in the same fashion with the exception
of using a 94% identity cutoff to construct unitigs. Finally, series of assemblies at
various stringencies were computed for subsets of the GOS data; in these assemblies,
overlapping mates were not preassembled and the Celera Assembler code was modified
slightly to allow for overlapping and multiple sequence alignment at lower stringency.
An all-against-all comparison of unassembled (but merged and duplicate-stripped)
sequences from the combined dataset was performed using a modified version of the
overlapper component of the Celera Assembler [21]. The code was modified to find overlap alignments
(global alignments allowing free end gaps) starting from pairs of reads that share an
identical substring of at least 14 bp. An alignment extension was then performed with
match/mismatch scores set to yield a positive outcome if an overlap alignment was found
with ≥65% identity. Overlaps involving alignments of ≥40 bp were retained for various
analyses. For the GOS dataset described here, this process resulted in a dataset of 1.2
billion overlaps. Due to the 14-bp requirement and certain heuristics for early
termination of apparently hopeless extensions, not all alignments at ≥65% were found. In
addition, some of the lowest-identity overlaps are bound to be chance matches; however,
this was a relatively uncommon event. Approximately one in 5 × 106 pairs of
800-bp random sequences (all sites independent, A = C = G = T = 25%) can be aligned to
overlap ≥40 bp at ≥65% identity using the same procedure. At a 70% cutoff, the value is
reduced to one in 4 × 107, and one in 5 × 108 at a 75% cut off.
Like many assembly algorithms, the extreme assembler proceeds in three phases: overlap,
layout, and consensus. The overlap phase is provided by the all-against-all comparison
described above. The consensus phase is performed by a version of the Celera Assembler,
modified to accept higher rates of mismatch. The layout phase begins with a single
sequencing read (“seed”) that is chosen at random or specified by the user and is
considered the “current” read. The following steps are performed off one or both ends of
the seed. (1) Starting from the current fragment end, add the fragment with the best
overlap off that end and mark the current fragment as “used,” thus making the added
fragment the new current fragment. (2) Mark as used any alternative overlap that would
have resulted in a shorter extension. The simplest notion of “best overlap” is simply the
one having the highest identity alignment, but more complicated criteria have certain
advantages. A simple but useful refinement is to favor fragments whose other ends have
overlaps over those which are dead ends. For an unsupervised extreme assembly, when the
sequence extension terminates because there are no more overlaps, a new unused fragment is
chosen as the next seed and the process is repeated until all fragments have been marked
used.
Sequencing reads mated to SAR11-like 16S sequences but themselves outside of the
ribosomal operon (n = 348) were used as seeds in independent extreme
assemblies. Since the assemblies were independent, the results were highly redundant, with
a given chain of overlapping fragments typically being used in multiple assemblies. A
subset of 24 assemblies that shared no fragments over their first 20 kb was identified as
follows. (1) Connected components were determined in a graph defined by nodes
corresponding to extreme assemblies. If the assemblies shared at least one fragment in the
first 20 kb of each assembly, the two nodes were connected by an edge. (2) A single
assembly was chosen at random from each of the connected components. The consensus
sequence over the 20-kb segment of each such representative was used as the reference for
fragment recruitment.
Phylogenies of sequences homologous to a given portion of a reference sequence (typically
500 bp) were determined in the following manner. A set of homologous fragments was
identified based on fragment recruitment to the reference as described above. Fragments
that fully spanned the segment of interest and had almost full-length alignments to the
reference sequence of a user-defined percent identity (typically, 70%) were used for
further analysis. A preliminary master–slave multiple sequence alignment of the recruited
reads (slaves) to the reference segment (master) was performed with a modified version of
the consensus module of the Celera Assembler. Based on this alignment, reads were trimmed
to the portion aligning to the reference segment of interest. A refined multiple sequence
alignment was then computed with MUSCLE [94]. Distance based phylogenies were computed using the programs DNADIST and
NEIGHBOR from the PHYLIP package [95]
using default settings. Trees were visualized using HYPERTREE [96].
Based on the low-identity overlap database described above, the similarity of a library
i to another library j at a given percent identity
cutoff was computed as follows. For each sequence s of
i, let ns,i = the number of overlaps to other
fragments of i satisfying the cutoff; ns,j =
the number of overlaps to fragments of j satisfying the cutoff; and
fs,i =
ns,i/(ns,i +
ns,j) = fraction of reads overlapping
s from i or j that are from
i.
A read that can be overlapped to another at sufficiently high-sequence identity was taken
to indicate that they were from similar organisms, and, relatedly, that similar genes were
present in the samples. Only reads with such overlaps contributed to the calculation.
Other reads reflect genes or segments of genomes that were so lightly sampled (i.e., at
such low abundance) that they were not informative regarding the similarity of two
samples. Consequently, the analysis automatically corrects for differences in the amount
of sequencing, and can be computed over sets of samples that vary considerably in
diversity. The resulting measure of similarity Si,j takes on a
value between 0 and 1, where 0 implies no overlaps between i and
j, and 1 implies that a fragment from i and a fragment
from j are as likely to overlap one another as are two fragments from
i or two fragments from j. As with the Bray-Curtis
coefficient [97], abundance of
categories affects the computation. In an idealized situation where two libraries can each
be divided into some number k of “species” at equal abundance, and the
libraries have l of the species in common, the similarity statistic will
approach l/k for large samplings; in
this sense, Si,j = x indicates that the two samples share
approximately a fraction x of their genetic material. It is frequently
useful to define Di,j = 1 − Si,j,
the “dissimilarity” or distance between two samples.
An all-against-all comparison of predicted 16S sequences was performed to determine the
alignment between pairs of overlapping sequences using a version of an extremely fast
bit-vector algorithm [98]. A
hierarchical clustering was determined using percent-mismatch in the resulting alignments
as the distance between pairs of sequences. Order of clustering and cluster identity
scores were based on the average-linkage criterion, with distances between nonoverlapping
partial sequences treated as missing data. Ribotypes were the maximal clusters with an
identity score above the cutoff (typically 97%). Representative sequences were chosen for
each cluster based on both length and highest average identity to other sequences in the
cluster.
Taxonomic classification of 16S sequences was conducted using phylogenetic techniques
based on clade membership of similar sequences with 16S sequences with defined taxonomic
membership. Representative sequences from clustered sequences were analyzed as described
previously [19,99] and by addition into an ARB database of small subunit
rDNAs [100,101]. Results were spot-checked against the Ribosomal
Database Project II Classifier server [102] and the taxonomic labels of the best BLASTN hits against the nonredundant
database at NCBI.
Global ocean sequences were aligned to genomic sequences of different bacteria and phage
using NCBI BLASTN [26]. The following
blast parameters were designed to identify alignments as low as 55% identity that could
contain large gaps: -F “m L” -U T -p blastn -e 1e-4 -r 8 -q -9 -z 3000000000 -X 150. Reads
were filtered in several steps to identify the reads that were aligned over more or less
their entire length. Reads had to be aligned for more than 300 bp at >30% identity
with less than 25 bp of unaligned bases on either end, or reads had to be aligned over
more than 100 bp at >30% identity with less than 20 bp of overhang off either end.
Identity was calculated ignoring gaps. In some instances a read might be placed, but the
mate would not be placed under these criteria. In such cases, if 80% or more of the mate
were successfully aligned, then the mate would be rescued and considered successfully
aligned.
Random pieces of DNA from the genome in question with a length between 1,800 to 2,500 bp
were selected. For each piece a read length N1 was selected from the distribution of
lengths using the GOS dataset. If that GOS sequence had a mate pair, then a second length
N2 was again randomly selected. The length N1 was used to generate a read from the 5′ end
of the DNA. The piece of DNA was then reverse complemented and if appropriate, a second
length N2 was used to generate a second read. The relationship between these two reads was
then recorded and used to produce a fasta file. This approach successfully mimics the
types of reads found in the GOS data with similar rates of missing mates.
A total of 2,644 proteorhodopsin genes were identified from the clustered open reading
frames derived from the GOS assembly [18]. These genes could be linked back to 3,608 GOS clones. Open reading frames
were predicted from these clones as described in [18]. The peptide sequences were aligned with NCBI
blastpgp with the following parameters: -j 5 -U T -e 10 -W 2 -v 5 -b 5000 -F “m L” -m 3.
The search was performed with a previously described blue-absorbing proteorhodopsin
protein BPR (gi|32699602) as the query. The amino acid associated with light absorption is
found within a short conserved motif RYVDWLLTVPL*IVEF, where the asterisk indicates tuning
amino acid [39–41]. In total, 1,938 clones were found to contain this
motif. Clones and the sample metadata were then associated with the tuning amino acid to
determine the relative abundance of the different amino acids at these positions. Clones
could be associated with SAR11 if both mated sequencing reads (when available) were
recruited to P.
ubique HTCC1062.
Given a set of genes identified on the GOS sequences, we can identify the scaffolds on
which these genes were annotated. A vector indicating the number of sequences contributed
by every sample is determined for every gene. This vector reflects the number of sequences
from every sample that assembled into the scaffold on which the gene was identified after
normalizing for the proportion of scaffold covered by the gene. For example, if a 10-kb
scaffold contains a 1-kb gene, then each sample will contribute one GOS sequence for every
ten GOS sequences it contributed to the entire scaffold. The vectors are then summed and
normalized to account for either the total number of GOS sequences obtained from each
sample or based on the number of typically single copy recA genes
(identified as in [18]). Unless
stated otherwise, recA was used to normalize abundance across samples.
When comparisons using groups of samples were performed, the average value for the samples
was compared.
A 1-D profile representing oligonucleotide frequencies was computed as follows. A
sequence was converted into a series of overlapping 10,000-bp segments, each segment
offset by 1,000 bp from the previous one, using perl and shell scripting. Dinucleotide
frequencies are computed on each segment using a C program written for this purpose.
Higher-order oligonucleotides were examined and gave similar results for the genomes of
interest. Remaining calculations were performed using the R package [103]. Principle component analysis (function
princomp with default settings) was applied to the matrix of
frequencies per window position. The value of the first component for each position was
normalized by the standard deviation of these values, and truncated to the range [−5, 5].
For visualizations, the resulting values were plotted at the center of each window.
The unrecruited mated sequencing reads of reads recruited to P. marinus MIT9312 at
or above 80% identity were examined. An unrecruited mate indicated a potential
translocation or inversion if it aligned to the MIT9312 genome in two and only two
distinct alignments separated by at least 50 kb, if each aligned portion was at least 250
bp long, if there was less than 100 bp of unaligned sequence and no more than 100 bp of
overlapping sequence between the two aligned portions in read coordinates, and if each
aligned portion was anchored to one end of the sequencing read with less than 25 bp of
unaligned sequence from each end. In total, 18 rearrangements were identified, six of
which appear to be unique events.
The rate of discovery was estimated by determining the number of rearrangements in a
given volume of sequence. We estimated the volume of sequence that was potentially
examined by identifying recruited mated sequencing reads that fit the “good” category
(i.e., which were recruited in the correct orientation at the expected distance from each
other). For a given read, if the mate was recruited at greater than or equal to 80%
identity, then the expected amount of sequence examined should be the current (as opposed
to mate) read length minus 500 bp. This produces an estimate of the search space to be ~47
Mbp. Given 18 rearrangements, this leads to an estimate of one rearrangement per 2.6
Mbp.
GOS reads assigned to the “missing mate” category that were recruited at greater than 80%
identity outside the gap in question were identified. The mates of these reads were then
identified and clustering was attempted with Phrap (http://www.phrap.org).
Reads that were incorporated end to end into the Phrap assemblies were identified. For
most small gaps a single assembly included all the missing mate reads and identified the
precise difference between the reference and the environmental sequences. For the
hypervariable segments, most of the reads failed to assemble at all, and those that did
show greater sequence divergence than typically seen. In the case of SAR11-recruited
reads, to increase the number of reads associated with the hypervariable gaps we
identified reads that did not recruit to the P. ubique HTCC1062 but aligned in a single HSP
(high-scoring pair) over at least 500 bp with one end unaligned because it extended into
the hypervariable gap.
To facilitate continued analysis of this and other metagenomic datasets, the tools
presented here along with their source code will be available via the Cyberinfrastructure
for Advanced Marine Microbial Ecology Research and Analysis (CAMERA) website (http://camera.calit2.net). The dataset and associated metadata will be
accessible via CAMERA (using the dataset tag CAM_PUB_Rusch07a). Given the exceptional
abundance of Burkholderia and Shewanella sequences in
the first Sargasso Sea sample and the feeling that these may be contaminants, we are also
providing a list of the scaffold IDs and sequencing read IDs associated with these
organisms to facilitate analyses with or without the sequences. In addition to CAMERA, the
GOS scaffolds and annotations will be available via the public sequence repositories such
as NCBI (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=genomeprj&cmd=Retrieve&dopt=Overview&list_uids=13694),
and the reads will be available via the Trace Archive (http://www.ncbi.nlm.nih.gov/Traces/trace.cgi?).
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for proteorhodopsin
protein BPR is gi|32699602.
|
10.1371/journal.ppat.1001197 | TGF-b2 Induction Regulates Invasiveness of Theileria-Transformed Leukocytes and Disease Susceptibility | Theileria parasites invade and transform bovine leukocytes causing either East Coast fever (T. parva), or tropical theileriosis (T. annulata). Susceptible animals usually die within weeks of infection, but indigenous infected cattle show markedly reduced pathology, suggesting that host genetic factors may cause disease susceptibility. Attenuated live vaccines are widely used to control tropical theileriosis and attenuation is associated with reduced invasiveness of infected macrophages in vitro. Disease pathogenesis is therefore linked to aggressive invasiveness, rather than uncontrolled proliferation of Theileria-infected leukocytes. We show that the invasive potential of Theileria-transformed leukocytes involves TGF-b signalling. Attenuated live vaccine lines express reduced TGF-b2 and their invasiveness can be rescued with exogenous TGF-b. Importantly, infected macrophages from disease susceptible Holstein-Friesian (HF) cows express more TGF-b2 and traverse Matrigel with great efficiency compared to those from disease-resistant Sahiwal cattle. Thus, TGF-b2 levels correlate with disease susceptibility. Using fluorescence and time-lapse video microscopy we show that Theileria-infected, disease-susceptible HF macrophages exhibit increased actin dynamics in their lamellipodia and podosomal adhesion structures and develop more membrane blebs. TGF-b2-associated invasiveness in HF macrophages has a transcription-independent element that relies on cytoskeleton remodelling via activation of Rho kinase (ROCK). We propose that a TGF-b autocrine loop confers an amoeboid-like motility on Theileria-infected leukocytes, which combines with MMP-dependent motility to drive invasiveness and virulence.
| Theileria annulata causes tropical theileriosis that is endemic in cattle in North Africa, the Middle East, India and China. T. parva causes East Coast fever that is prevalent in East and Southern Africa. In endemic countries indigenous cattle are more resistant to pathology, but produce little meat and milk and attempts to improve output by importing European and American breeds have failed due to a high susceptibility to these diseases that are often rapidly fatal. We examined T. annulata-transformed macrophages isolated from disease resistant Sahiwal compared to disease-susceptible Holstein-Friesian (HF) cattle, for their capacity to traverse synthetic extra-cellular matrix in vitro. The invasive capacity of all transformed macrophages was TGF-b dependent, but those of disease-susceptible HF animals invaded better i.e. they were more aggressive. The greater invasive capacity of HF transformed macrophages matched their increased production of TGF-b2, since levels of TGF-b1, and all three TGF-b receptors, were the same as in transformed macrophages isolated from disease-resistant Sahiwal animals. TGF-b2 production therefore likely renders Theileria-transformed leukocytes more pathogenic and consistently, in a live attenuated line used to vaccinate against tropical theileriosis transcripts of TGF-b2 and those of a significant number of TGF-target genes drop and consequently, TGF-b-mediated invasiveness decreases.
| Cellular transformation is a complex, multi-step process and leukocyte transformation by Theileria is no exception, as parasite infection activates several different leukocyte-signalling pathways, the combination of which leads to full host cell transformation [1]. However, Theileria-induced leukocyte transformation is unusual in that it is rapid and appears to be entirely reversible with the host cell losing its transformed phenotype upon drug-induced parasite death [2]. Just like most cancer cells however, Theileria-induced pathogenesis (virulence) is associated with the invasive capacity of transformed leukocytes, which is lost upon attenuation of vaccine lines [3]. Attenuation of virulence has been ascribed to decreased matrix-metallo-proteinase-9 (MMP9) production and loss of AP-1 transcriptional activity [4]. Consistently, functional inactivation of AP-1 resulted in reduced tumour formation, when infected and transformed B cells were injected into Rag2gC mice [5].
Host leukocyte tropism differs with T. parva infecting all subpopulations of lymphocytes whereas T. annulata infects monocytes/macrophages, dendritic cells and B lymphocytes [1]. Despite this, the diseases they cause (called tropical theileriosis with T. annulata infection and East Coast fever with T. parva infection) are both severe, as susceptible animals usually die within three weeks of infection. The geographical distribution of their respective tick vector species determines areas where disease is widespread. Tropical theileriosis affects over 250 million animals and extends over the Mediterranean basin, the Middle East, India and the Far East, whereas East Coast fever is prevalent in eastern, central and southern Africa. It is noteworthy that in endemic areas indigenous breeds of cattle are more resistant to disease. For example, when Bos indicus Sahiwals are experimentally infected with T. annulata they exhibit fewer clinical symptoms and recover from a parasite dose that is fatal in the European Holstein-Friesian (HF) B. taurus breed [6]–[7]. Theileria-infected leukocytes are capable of producing IL-1 and IL-6 [8], as well as GM-CSF [9] and TNF [10]. Nonetheless, no differences in the level of expression of the pro-inflammatory cytokines TNF, IL-1b, or IL-6 were detected between disease-resistant Sahiwal- versus HF-infected macrophages [11]. Some additional inherent genetic trait of Sahiwal animals must therefore underlie their disease-resistance. Although transcriptome analysis of 3–5 times passaged Sahiwal and HF macrophages following infection with T. annulata revealed significant breed differences in both the resting and infected gene expression profiles, no clear candidate genetic trait was revealed [12].
Transforming growth factor beta (TGF-b) is a family of cytokines and both TGF-b1 and TGF-b2 can bind with high affinity to the TGF-b type II receptor (TGF-RII) leading to the recruitment of TGF-RI. The constitutive kinase activity of TGF-RII phosphorylates and activates TGF-RI, which in turns recruits and activates Smad2 and Smad3, which bind Smad4, and the whole complex translocates to the nucleus and induces the transcription of target genes [13]. The TGF-b signalling pathway can be negatively regulated [14] and an increasing number of non-Smad-mediated TGF-b signalling pathways have been described [15]. TGF-b can also regulate cytoskeleton dynamics via transcription-dependent and transcription-independent processes [16]. It is likely that all these different pathways contribute in different ways to the pleiotropic effects of TGF-b (see http://www.cell.com/enhanced/taylor). TGF-b can exert opposite effects on cell growth: in most non-transformed cells TGF-b is usually growth inhibitory, but it can increase motility of certain mesenchymal cells and monocytes, but however, at some point in the transformation process TGF-b becomes pro-metastatic [17]–[18], for example in ovarian cancer [19]. We show here that TGF-b plays a role in infected host leukocyte invasiveness. Importantly, the high level of TGF-b2 production by Theileria-infected HF-transformed macrophages renders them more invasive than those of disease-resistant Sahiwal animals. In addition, vaccination against tropical theileriosis uses live attenuated T. annulata-infected macrophages and attenuation leads to the loss of both TGF-b2 transcription and alteration in the expression of a set of TGF-b-target genes, and a drop in TGF-b-mediated invasion. Thus, Theileria-dependent TGF-b2 induction is a virulence trait that underscores susceptibility to tropical theileriosis.
As Theileria-transformed leukocytes are known to secrete a number of different cytokines we examined whether infection by T. annulata sporozites of the same parasite strain (Hissar) could induce TGF-b in macrophages 72h post-invasion, as described [12]. Prior to infection both Sahiwal and HF macrophages produced low levels of TGF-b transcript with slightly higher amounts of TGF-b1 (Fig. 1A). Interestingly, Theileria infection induces preferentially TGF-b2 in both Sahiwal and HF macrophages, and importantly, the induction after 72h is much greater in disease-susceptible HF cells.
We next examined the levels of TGF-b transcripts in a series of T. annulata-transformed cell lines derived from Sahiwal and HF animals. Again, TGF-b1 and TGF-b2 mRNA could be detected in all 10 transformed cell lines (Fig. 1B). Similar to freshly invaded cells the relative mRNA levels of TGF-b1 did not differ significantly across the T. annulata infected cell lines and there was no evidence of a breed-specific difference in TGF-b1 expression (Fig. 1B grey bars, p = 0.710). In contrast, the relative TGF-b2 mRNA levels exhibit statistically significant differences (Fig. 1B black bars, p<0.001), with TGF-b2 mRNA levels being higher in HF cell-lines. Additional qRT-PCR experiments revealed that the levels of expression of TGF-RI, TGF-RII and TGF-RIII were equivalent in HF versus Sahiwal T. annulata-infected cell lines (data not shown). Thus, disease susceptibility correlates to the level of TGF-b2 transcripts that are expressed 7-fold (p<0.001) more by T. annulata-transformed macrophages of HF origin.
In T. parva-transformed B cells a TGF-b-mediated signalling pathway is active and invasion is partially TGF-b-dependent (Fig.S1). We therefore compared the invasive capacity of T. annulata-transformed HF versus Sahiwal macrophages (Figure 2). We found that disease-susceptible HF macrophages displayed 30% greater capacity (p<0.005) to traverse Matrigel than infected Sahiwal macrophages and that traversal is again TGF-b-dependent (Fig. 2A). The invasive capacity of the H7 cell line was reduced to levels equivalent to S3 cells upon treatment with the TGF-R inhibitor (Fig. 2A) and conversely, when S3 cells were stimulated with conditioned medium from H7 cultures, S3 cells displayed increased invasiveness (Fig. 2B). Moreover, the reduced invasive capacity of disease-resistant S3 macrophages could be restored to above virulent levels by addition of either TGF-b1, or TGF-b2 (Fig. 2C), demonstrating that the TGF-b signalling pathways are intact in these cells. Theileria infection therefore preferentially induces up-regulation of TGF-b2 and increased invasiveness of transformed leukocytes.
T. annulata-infected cell lines can be attenuated for virulence by multiple in vitro passages to generate live vaccines that are used to protect against tropical theileriosis [3]. The molecular basis of attenuation is not known, but our above observations on preferential TGF-b2 induction and augmented host cell invasiveness suggest that attenuation might be lead to reduced TGF-b2 transcription and TGF-b2-mediated invasion. To directly test this prediction we examined the Ode vaccine line derived from an infected HF cow in India [20] and estimated the level of TGF-b2 transcripts and TGF-b-target gene transcription in virulent (early passage) and attenuated (late passage) infected macrophages (Figure 3). As predicted, attenuation leads to a significant decrease in the amount of TGF-b2 message and surprisingly, a slight increase in TGF-b1 transcripts (Fig. 3A). This strongly suggests that upon attenuation the parasite's ability to induce host cell TGF-b2 has been impaired. Reduced levels of TGF-b2 should lead to an alteration in the transcription profiles of known TGF-b-target genes and to see if this is indeed the case, we performed microarray analyses and hierarchical clustering of transcript levels. The microarray representing 26,751 bovine genes included 1,158 targets of TGF-b (http://www.netpath.org/) and 76 of these genes were identified as differentially expressed upon attenuation. The heat-map, where low gene expression level is depicted as blue, intermediate as yellow and high expression as red, is present in Fig. 3B. Among the down-regulated TGF-target genes, five were chosen at random and their expression verified by qRT-PCR using mRNA from early and late passage Ode (Fig. 3C). In each case their transcription was reduced upon attenuation and could be restored by adding exogenous recombinant TGF-b2. Consistently, their expression was high in disease-susceptible infected HF macrophages that produce more TGF-b2 (see Fig. 1) and low in Sahiwal macrophages, but could be augmented by exogenous TGF-b2 stimulation (Fig.S2). Attenuation of virulence therefore, leads to ablation of TGF-b2 signalling and an alteration in the profile of expression of a set of TGF-target genes.
The observation that early passage Ode cells express higher levels of TGF-b2 message and have altered expression of 76 TGF-b-target genes led us to compare early with late passage Ode and examine the contribution of TGF-b to their invasive capacity (Figure 4). As previously described [4], attenuation of Ode leads to a significant drop in invasive capacity (***p<0.005) and receptor blockade by SB431542 gives an estimate (***p<0.005) of the contribution of TGF-b to early passage Ode virulence (Fig. 4A). The potential contribution of TGF-b to virulence has been ablated by attenuation, since the invasive capacity late passage Ode is insensitive to receptor blockade (Fig. 4A, right). When conditioned medium from early passage Ode is given to late passage Ode there is a marked (**p<0.05) regain in invasion (Fig. 4B). Virulent Ode therefore, secretes factors into the medium that contribute to invasiveness, one of which is clearly TGF-b2, and this capacity is lost upon attenuation. The partial inhibition of invasion by early passage Ode by SB431542 might also suggest that although virulent following 65/70 in vitro passages some attenuation of TGF-b2 induction might be occurring.
We next studied whether the TGF-b-mediated invasion programme might have a consequence on cellular adhesion and invasion structures such as lamellipodia, podosomes and membrane blebs. We first investigated by time-lapse video microscopy lamellipodia morphology of T. annulata-infected macrophages cultured without (control), or with SB431542 (Fig.S3 and Movies S1 and S2) and found that the size of lamellipodia (shown boxed) decreased upon SB431542 treatment (Fig.S3A). Visualisation of the actin cytoskeleton with Texas red-labelled phalloidin showed that decreased lamellipodia size correlated with reduced actin dynamics (Fig.S3B and C) and suggested that TGF-controls cortical actin dynamics in infected macrophages. We next compared the basal and central cortical actin cytoskeleton of S3 and H7 cells cultured on a gelatin/fibronectin matrix, which facilitates adhesion of these cells. In S3 cells, we observed only small podosomal adhesion structures that were rarely clustered and no actin-rich membrane blebs (Fig. 5A). In contrast, in H7 cells podosomal adhesion structures were markedly enlarged and clustered and the majority of cells displayed actin-rich membrane blebs. Membrane blebbing was confirmed by live-cell imaging (Fig.S4A), which highlights the dynamics of bleb formation. Individual blebs expand within one to three seconds and persist for approximately 30–120 seconds, which is a time frame typically observed in bleb formation and retraction [21]. Reducing the serum concentration from 10% to 0.5% (starvation) resulted in a significant decrease in the number of membrane blebs (Fig. 5B and C). Membrane blebbing in starved cells was rescued by the exogenous addition of TGF-b2. The TGF-b-induced membrane blebs were blocked by the Rho-kinase (ROCK) inhibitor H-1152 and in the presence of serum the formation of actin-rich membrane blebs was significantly reduced after treatment with the TGF-R inhibitor, but completely blocked in the presence of H-1152 (Fig. 5C). Thus, one consequence of increased TGF-b2 production is increased cortical actin dynamics, which likely gives rise to enhanced podosomal adhesion structures (invadosomes) and membrane blebs in macrophages derived from disease susceptible HF cattle.
We next investigated the functional significance of membrane blebs for cell motility in 3-D matrices. In the low rigidity fibrillar collagen or high-density Matrigel matrices, H7 macrophages acquired an amoeboid pattern of motility with characteristic polarized formation of membrane blebs (Fig. 5D and Fig.S4B and C). Membrane bleb formation required ROCK activity and inhibition of ROCK prevents local contractibility, polarized bleb formation and forward movement of the cell. Taken together, these data show that exposure of H7 macrophages to TGF-b2, results in ROCK-dependent membrane blebbing that drives motility in 3-D matrices.
We have shown here that Theileria-induced leukocyte transformation results in the constitutive induction of a TGF-b autocrine loop that augments the invasive potential of infected leukocytes. We could find no evidence for a contribution of TGF-b signalling to host cell survival, or proliferation (data not shown), implying that leukocyte infection by Theileria essentially confers on TGF-b a pro-metastatic role. Smads and p53 are known to associate and collaborate in the induction of a subset of TGF-b target genes [22]. Recently, p53 has been described as being sequestered in the cytosol of Theileria-transformed leukocytes, as part of a parasite-induced survival mechanism [23]. Although not addressed by Haller et al it is possible that cytosol located p53 might ablate nuclear translocation of Smads, thus counteracting the anti-proliferative affect of TGF-b in Theileria-transformed leukocytes?
Importantly, comparison of disease-susceptible HF transformed macrophages to disease-resistant Sahiwal ones, showed that the degree to which Theileria (the same parasite strain) induces TGF-b2 influences the invasive potential of the infected and transformed host cell. The likelihood of developing a life-threatening cancer-like disease therefore appears to be due in part to the inherent genetic propensity of HF macrophages to produce high levels of TGF-b2 upon infection that might render the transformed macrophages more invasive. T. annulata-transformation of HF macrophages leads to the induction of higher amounts of TGF-b2 the levels of induced TGF-b1 mRNA being the same as in Sahiwal macrophages. As TGF-RI and -RII and -RIII [24] are expressed to the same extent in the two types of macrophages (data not shown) it suggests that only the amount of TGF-b2 is crucial. The predisposition of Theileria transformation to induce TGF-b2 over TGF-b1 in HF versus Sahiwal macrophages implies that there could be disease-associated sensitivity to infection linked to TGF-b2 over-production. Species-specific promoter differences, or some other unknown breed difference may explain the greater propensity of Theileria to induce TGF-b2 over TGF-b1 transcripts in HF compared to Sahiwal cattle. However, promoter sequence differences seem unlikely to underlie Theileria's ability to induce TGF-b2 over TGF-b1 transcripts, or explain the drop in TGF-b2 levels upon attenuation of the Ode vaccine line as here, both virulent and attenuated infected macrophages are of HF origin [20]. Loss of virulence of the Ode vaccine line upon long-term in vitro passage is clearly associated with a decrease in TGF-b2 transcripts and it would appear that the parasite's ability to specifically activate host cell transcription of TGF-b2 is impaired and one possibility is that attenuation is associated with altered epigenetic regulation of TGF-b2 promoter activity. Microarray analyses indicate that upon attenuation of virulence, not only do TGF-b2 levels drop, but also 76 TGF-b-target genes display altered transcription. It would appear then that preferential TGF-b2 induction following Theileria infection initiates a host cell genetic programme that contributes to more aggressive invasiveness of transformed HF macrophages. We believe the same TGF-b2-initiated genetic programme also contributes to the invasiveness, albeit reduced, of disease-resistant Sahiwal macrophages, as Theileria infection also preferentially induces TGF-b2, just to a lesser extent.
We have used fluorescence and time-lapse video microscopy to examine the morphology of Theileria-infected Sahiwal and HF infected macrophages and the effect of TGF-b and Rho kinase (ROCK) on actin dynamics and lamellipodia formation. Theileria infected, disease-susceptible HF macrophages show increased actin dynamics in their lamellipodia and podosomal adhesion structures and a remarkable propensity to develop membrane blebs. Figure 5D shows the dynamic behaviour of infected cells embedded in 3-D matrices (see also Fig.S3 and movies S3 and S4), where either fibrillar collagen, or matrigel was used giving two 3-D matrices of low (collagen) and high (matrigel) rigidity. Membrane blebbing of motile H7 cells occurs in both matrices in a polarized fashion at the leading edge, clearly suggesting that membrane blebbing is required for infected cell motility in 3-D. Movie S3 shows bleb-driven membrane protrusions, which results in forward movement of the cell (see also kymographs of movie S3, Fig. 5D). Moreover, ROCK activity is required, because inhibition of ROCK with H-1152 impairs polarized bleb formation and forward movement in fibrillar collagen and matrigel. Spatio-temporal control of Rho-ROCK activity is also required for cell polarization and lamellipodia formation in 2-D [25]. Since TGF-b acts upstream of Rho-ROCK in infected cells, we would therefore predict that spatio-temporal activation of Rho-ROCK – controlled by TGF-b signalling – is required for lamellipodia formation as well. Combined, increased bleb and lamellipodia formation could give rise to more invadosomes on infected virulent macrophages in a similar manner to TGF-b-mediated increased adhesion of immortalised hepatocyte cell lines [26].
We showed that augmented invasiveness by TGF-b2 in disease-susceptible HF macrophages has a transcription-independent element that relies on cytoskeleton remodelling via activation of Rho and its downstream target ROCK [27]–[28]. Given the important role played by ROCK in increased blebbing of infected host cells and the recent description that Rho/ROCK signals amoeboid-like motility [29], it is tempting to speculate that the TGF-b autocrine loop confers on Theileria-infected leukocytes an amoeboid-like motility that contributes to invasiveness and makes them more virulent. However, since prolonged TGF-b stimulation can result in decreased Rho activity due to the action of p190RhoGap [30], Theileria infected cells must have developed a mechanism to balance TGF-b-induced RhoGAP activities. It is possible that the parasite might also regulate the expression, activity or localization of Rho-family GEFs, as the Rho activator GEF-H1/Lfc has been shown to be a TGF-b1 target gene [31] and an analogous mechanism involving TGF-b2 might be exploited by Theileria parasites? Alternatively, the parasite could function by excluding negative regulators of Rho from specific subcellular compartments, analogous to the exclusion of the RhoGAP Myo9b from lamellipodia of macrophages [25].
Whatever the underlying mechanism of inducing TGF-b2 levels in Theileria-infected leukocytes, its induction and the genetic programme it initiates is clearly correlated with the invasive phenotype of transformed macrophages of disease-sensitive hosts. This implies that overall invasiveness of Theileria-transformed leukocytes is made up of amoeboid (TGF-b- & ROCK-dependent) and mixed amoeboid/proteolytic (MMP-dependent [4]) motility; reviewed in [32]. These Theileria-based observations also suggest that in some cases the propensity of human leukaemia patients to develop life-threatening cancer could be due to the inherent genetic predisposition of their tumours to produce high levels of TGF-b2, rather than TGF-b1, and the genetic programme this initiates on promoting an additional amoeboid-like invasive phenotype of their tumours.
TpMD409.B2 is a T. parva Muguga-infected B-cell clone (B2) and its B-cell characteristics have been previously confirmed [33]. The cell lines S1–S5 and H7–H10 have been described previously [6]. In vitro infection of uninfected S and H cells by Theileria sporozoites was done as described [12]. The characterisation of the Ode vaccine line has been reported [20] and in this study virulent/early Ode corresponds to passage 65–70 and attenuated to passage 318–322. It is possible that passages 65–70 have already become slightly attenuated. All cultures were maintained in RPMI-1640 medium supplemented with 10% foetal calf serum (FCS) and 50 uM b-mercaptoethanol. Cell cultures were passaged 24h before harvesting to maintain the cells in the exponential growth phase.
The TGF-bRI/ALK5 inhibitor SB431542 (Sigma #S4317) was added at 10uM for 18h. The Rho kinase (ROCK) inhibitor H-1152 (Alexis Biochemicals, #ALX-270-423) was added at 10uM. Recombinant bovine TGF-b1 and TGF-b2 (rboTGF-b1 and rboTGF-b2; both NIBSC, Potters Bar. UK) were added in the culture media at 10ng/ml and incubated for different times (15 min or 30 min).
All transfections were carried out by electroporation as previously described [34]. The CAGA-luc (the Smad 3/4 binding element-luciferase construct) was transiently transfected into B2 cells with the inhibitory Smad7 plasmid (Flag-Smad7-pcdef3), or an empty vector (pcdef3). The major late minimal promoter (MLP - a minimal promoter consisting of the TATA box and the initiator sequence of the adenovirus major late promoter) was cloned into pGL3 (Promega) to generate the MLP-luc plasmid that was used as minimal promoter negative control [35]–[36]. Efficiencies of transfections were normalized to renilla activity by co-transfection of a pRL-TK renilla reporter plasmid (Promega #E6241). Luciferase assay was performed 24h after transfection using the Dual-Luciferase Reporter Assay System (Promega, #E1980) in a microplate luminometer (Centro LB 960, Berthold). Relative luminescence was represented as the ratio firefly/renilla luminescence.
Nuclear extracts of from T. parva-infected (B2) cells were prepared as described [37]. 20ug of proteins were separated in a denaturing 8% SDS-PAGE gel and electro-transferred onto a nitrocellulose membrane (Scheicher and Schuell). Antibodies used in immuno-blotting were as follows: anti-phospho-Smad2 (Cell Signaling #3101), anti-Smad2 (BioVision, #3462-100), anti-PARP (Clone C2-10, Pharmingen #556362).
Total RNA was isolated from each of the T. annulata infected cell lines using the RNeasy mini kit (Qiagen) according to the manufacturer's instructions. The quality and quantity of the resulting RNA was determined using a Nanodrop spectrophotometer and gel electrophoresis and for microarray screens on an Agilent 2100 Bioanalyser (Agilent Technologies). mRNA was reverse transcribed to first-strand cDNA and the relative levels of each transcript were quantified by real-time PCR using SYBR Green detection. The detection of a single product was verified by dissociation curve analysis and relative quantities of mRNA calculated using the method described by [38]. GAPDH, HPRT1 and C13orf8 relative quantities were used to normalise mRNA levels. For list of primers used, see Table 1.
Host gene expression was investigated using a custom-designed microarray (Roche NimbleGen Inc., Madison, WI), which represented every bovine RNA reference sequence currently deposited in the NCBI database (n = 26,751). Each gene was represented by two identical sets of five 60-mer-oligonucleotide probes that were isothermal with respect to melting temperature. cDNA was generated from 10 ug total RNA using oligo(dT) primer and tagged with 3′-Cy3 dye and labelled cDNA was hybridised to the array. Gene expression values were calculated from an RMA-normalised dataset [39] and differentially expressed genes were identified using Rank Product Analysis [40]. Genes were defined as differentially expressed using the criteria of a false discovery rate of less than 5% and a fold change of greater than two. Selected gene sets were subjected to hierarchical clustering based on Euclidean distance between expression values and the results were illustrated using a heat-map (ArrayStar3, DNASTAR Inc., Madison, WI).
The invasive capacity of the bovine leukocytes was assessed using in vitro Matrigel migration chambers, as described [5]. After 26h of incubation at 37°C, filters were washed twice in PBS and the Matrigel was eliminated. In some cases, during this 12h period cells were also incubated with the TGF-b inhibitor SB431542 (10uM), or with recombinant TGF-b protein (10ng/ml). When added TGF-b was maintained in the top chamber, meaning that overall cells were incubated with TGF-b for 36h. Cells were then counted under the microscope (40× objective) to obtain a statistically significant mean number of cells per field (at least 10 fields per filter). The experiment was performed at least in triplicate.
Time-lapse imaging using video microscopy was performed with cells growing on glass bottom culture dishes (Willco Wells, the Netherlands) using a Nikon Eclipse TE2000-U inverted microscope equipped with a climate-controlled chamber. Data acquisition and image processing was performed using NIS software of Nikon Instruments. DIC images were acquired in intervals ranging from 0.5 msec to 1 min for 30 min and assembled in movies; acceleration of movies is approximately 600-fold. Kymographs were acquired along a one pixel wide line using NIS software. A 40× Plan Achromat Objective of Nikon was used for longworking distance image acquisition in Matrigel.
Theileria-infected macrophages were seeded onto glass coverslips, or glass bottom culture dishes (Willco Wells, the Netherlands) and maintained in growth medium for 48h without or with 10uM SB431542 (SIGMA). SB431542 was replenished after 24h. Cells were either processed for live cell imaging or fixed in 3.5% formaldehyde, 15 min. Actin cytoskeletons of fixed and Triton X-100 permeabilized cells were visualized with Texas red-labelled phalloidin. Lamellipodia area and integrated fluorescence intensities in lamellipodia were determined using photoshop CS3 software.
Glass coverslips were coated with 0.1% poly-L-lysine for 15 min at room temperature and then fixed with 0.5% glutharaldehyde for 15min. After 3 washes with PBS, coverslips were inverted onto droplet containing 2mg/ml (0.2%) gelatin (MERCK) in H2O for 15min. After 3 washes with PBS, coverslips were inverted onto droplet containing 25ug/ml bovine plasma fibronectin (SIGMA) in PBS and then incubated 1h at room temperature. Coverslips were then transferred into 24 well plate washed once with PBS and kept in PBS until seeding of 50,000 cells per well. After 18h cells were fixed in 3.5% formaldehyde, 15min.
Embedding in collagen: 3×105 cells in 50ul medium were added to a mixture of 68µl sodium bicarbonate (7.5%, SIGMA), 240ul 10× PBS and 2ml PureCol (3mg/ml, Inamed). The resulting gelatin solution with the concentration of 2.4mg gelatin/ml was distributed into live-cell imaging wells or dishes and transferred to 37°C for collagen polymerization.
Embedding in Matrigel: 1×105 cells in medium were mixed on ice with 250ul growth factor reduced Matrigel (BD biosciences) and was distributed into live-cell imaging wells or dishes and transferred to 37°C for polymerization.
|
10.1371/journal.pntd.0001764 | Amoebae as Potential Environmental Hosts for Mycobacterium ulcerans and Other Mycobacteria, but Doubtful Actors in Buruli Ulcer Epidemiology | The reservoir and mode of transmission of Mycobacterium ulcerans, the causative agent of Buruli ulcer, remain unknown. Ecological, genetic and epidemiological information nonetheless suggests that M. ulcerans may reside in aquatic protozoa.
We experimentally infected Acanthamoeba polyphaga with M. ulcerans and found that the bacilli were phagocytised, not digested and remained viable for the duration of the experiment. Furthermore, we collected 13 water, 90 biofilm and 45 detritus samples in both Buruli ulcer endemic and non-endemic communities in Ghana, from which we cultivated amoeboid protozoa and mycobacteria. M. ulcerans was not isolated, but other mycobacteria were as frequently isolated from intracellular as from extracellular sources, suggesting that they commonly infect amoebae in nature. We screened the samples as well as the amoeba cultures for the M. ulcerans markers IS2404, IS2606 and KR-B. IS2404 was detected in 2% of the environmental samples and in 4% of the amoeba cultures. The IS2404 positive amoeba cultures included up to 5 different protozoan species, and originated both from Buruli ulcer endemic and non-endemic communities.
This is the first report of experimental infection of amoebae with M. ulcerans and of the detection of the marker IS2404 in amoeba cultures isolated from the environment. We conclude that amoeba are potential natural hosts for M. ulcerans, yet remain sceptical about their implication in the transmission of M. ulcerans to humans and their importance in the epidemiology of Buruli ulcer.
| Buruli ulcer (BU) is a devastating skin disease caused by Mycobacterium ulcerans, an environmental bacterium that is probably linked to slow-running water. It is unlikely to occur free-living, but even though M. ulcerans DNA has been detected in quite a few different organisms (with most studies focusing on insects), it is still not clear what its real reservoir is. Amoeboid protozoa, inhabitants of biofilms in slow flowing water, are good candidates since all previously tested mycobacteria are resistant to the digestion by these macrophage-like organisms. In this paper we demonstrate that M. ulcerans can indeed infect Acanthamoeba polyphaga in the lab, and remain viable intracellularly. We also collected water, biofilm and detritus samples in BU endemic and non-endemic regions in Ghana. We found that several mycobacteria species commonly occur intracellularly in protozoa in these environments. Amoebae were isolated from almost all samples, and an M. ulcerans marker (IS2404) was detected in 4% of the amoeba cultures. We conclude that amoebae are potential hosts for M. ulcerans. However, because these IS2404 positive amoebae originated from both BU endemic and non-endemic areas, we remain sceptical about their implication in the transmission of M. ulcerans to humans.
| Most mycobacteria are environmental opportunistic species that only occasionally infect humans [1]. Only few mycobacterial species are known to be obligate parasites, such as Mycobacterium tuberculosis and M. leprae, the causative agents of tuberculosis and leprosy respectively. The third most common mycobacterial disease, Buruli ulcer (BU), is caused by M. ulcerans, an environmental opportunistic mycobacterium. BU occurs mainly in rural areas of West and Central Africa with over 40,000 cases reported between 2002 and 2010 [2]. To the present day, the main reservoir of M. ulcerans and its transmission from the environment to humans remain unknown.
Epidemiological data from Africa suggest that proximity to slow-flowing or stagnant water best explains the distribution pattern of BU [3]. Nevertheless, it is unlikely that M. ulcerans occurs free-living in these waters, because (i) M. ulcerans evolved recently from the generalist, more rapid-growing environmental M. marinum via lateral gene transfer and reductive evolution to become adapted to a more protected niche [4], and (ii) since M. ulcerans is sensitive to several antibiotics, such as streptomycin and rifampicin, it is unlikely that the bacteria occur free living in an environment where Streptomyces griseus and Amycolatopsis rifamycinica, producers of respectively streptomycin and rifampicin, thrive [5], [6]. In order to survive without protection of a host a or biofilm, M. ulcerans would have developed natural resistance against these antibiotics as is the case for most opportunistic non tuberculous mycobacteria [6].
Recently in Australia, M. ulcerans has been found at high prevalence in two species of possum, thus suggesting a role for these mammals as reservoirs of M. ulcerans [7]. Although in Africa M. ulcerans has never been found in such high numbers in any particular element of the environment, low levels of M. ulcerans DNA were detected in many biotic components of aquatic ecosystems, such as plants, snail, fish and insects [3], indicating that M. ulcerans is ubiquitous in these ecosystems. The most explored hypothesis of M. ulcerans transmission in Africa argues that microphagous arthropods, e.g the heteropteran waterbugs, feed on M. ulcerans in water or biofilms, which are in turn consumed by predatory insects that may occasionally bite humans [8]. M. ulcerans bacilli are thought to be concentrated along this food chain, resulting in a sufficient infectious dose for humans [8]–[12]. Although the exact role of insects in the transmission of M. ulcerans remains to be proven [13], a series of experiments do support this hypothesis [9]–[12] and recent extensive fieldwork found a relatively high prevalence of M. ulcerans DNA in several waterbug species in a BU endemic area [14]. In addition, the only successful cultivation of M. ulcerans from an environmental source was from an aquatic hemipteran [15].
Nonetheless, these transmission hypotheses do not exclude an important role for other host species as M. ulcerans reservoirs. Recently, it has been postulated that amoebae might represent hosts for M. ulcerans and that they could be involved in the transmission from the environment to humans [3], [16], [17]. A study in Benin found that the detection frequency of free-living amoebae in water bodies in BU endemic villages was higher than in non-BU endemic villages [16]. M. ulcerans has been shown to form a biofilm on aquatic plants [18], and amoebae are often the main predators in biofilm communities. As a response to amoebal predation, many bacteria have acquired resistance to digestion in amoebal food vacuoles [19]. Also many mycobacterium species have been shown to survive and even thrive intracellularly in protozoa [19], [20]. Because of their hydrophobic cell wall, mycobacteria tend to attach to surfaces and are easily phagocytised by protozoa [21] and macrophages [22] and can even actively promote their entry into phagocytes [23]. Mycobacteria can use nutrients of protozoa as a food source and the intracellular life offers protection against harmful and fluctuating environmental influences, as protozoan cysts are remarkably resistant to extremes of temperature, drought and all kinds of biocides [24], [25]. However, few studies have investigated whether mycobacteria infect amoebae in their natural environment. Natural resistance to amoebae can have important consequences, as bacteria that infect and evade digestion in amoebae might use the same tools to enter and resist destruction within macrophages [19], [26], which have similar properties, as has been shown for Legionella pneumophila [27], an environmental opportunistic bacterium with amoebae as main reservoir. An intracellular stay in amoebae even enhances the virulence of L. pneumophila against mammalian cells [28]. This feature is not restricted to Legionella; M. avium passaged through Acanthamoeba castellanii is also more virulent towards a mouse model [29]. Even though M. ulcerans was long considered an extracellular pathogen, there is increasing evidence that the bacterium exhibits an important intracellular phase in neutrophils and macrophages during human infection (reviewed in [30]). So far only one study, published in 1978, has shown that M. ulcerans can be phagocytized and retained in an Acanthamoeba, yet this study provided only limited results [31].
In the present study we experimentally determined the capacity of M. ulcerans to infect A. polyphaga. We then further investigated the role of free-living amoebae as hosts for mycobacteria in their natural environment, including the potential of these protozoa as reservoirs for M. ulcerans.
The M. ulcerans strains used were ITM 030216 (Benin), ITM 980912 (China), ITM 5114 (Mexico) and ITM 842 (Surinam) from the collection of the Institute of Tropical Medicine (ITM), Antwerp, Belgium.
A one week old, and therefore starving, axenic culture of A. polyphaga CCAP 1501/15 in PYG712 broth was adjusted to 105 cells/mL and 1 mL transferred into each of the wells of a 24-well tissue culture plate. The amoebae monolayers were seeded with suspensions of M. ulcerans in triplicate at an approximate multiplicity of infection of 1 (M. ulcerans to A. polyphaga) and plates were incubated at 30°C. Three hours after infection, 20 µg/ml kanamycin was added to prevent extracellular growth of M. ulcerans. At times 3 h, 3 d, 7 d and 14 d after infection, the supernatant was aspirated and discarded. At each time point the medium of the unused wells was replaced by fresh PYG712 broth containing 20 µg/mL kanamycin.
To estimate the number of intracellular colony forming units (CFU) present inside amoebae at each time point the monolayer was suspended in 0.1% SDS in order to lyse the amoebae. The lysed amoebae suspension was transferred into a tube containing glass beads and vortexed. This suspension, as well as two 10-fold dilutions were inoculated on Löwenstein-Jensen (LJ) medium. The tubes were then incubated at 30°C and read after 6 weeks.
To localize intracellular bacteria, the amoeba monolayer was suspended in PBS and then processed for electron microscopy and for acid-fast staining using the Ziehl-Neelsen (ZN) method.
For transmission electron microscopy the processing of samples was carried out taking into account that bacteria, including mycobacteria, have specific requirements for adequate preservation as discussed elsewhere [32]. Briefly, a volume of 0.3 mL of the suspended monolayer was pelleted and the pellet was pre-fixed with 4% formaldehyde-1.25% glutaraldehyde-10 mM CaCl2 for 24 h, then fixed in 1% OsO4–10 mM CaCl2 for 16 to 24 h, and then post-fixed in aqueous 1% uranyl acetate for 1 h. Further processing for electron microscopy was carried out with ethanol dehydration and Epon embedding. Ultrathin sections were double-stained with uranyl acetate (saturated aqueous solution) for 5 min followed by lead citrate for 3 min.
Using light microscopy, AFB were observed co-localizing with amoebae after co-incubation of each M. ulcerans strain with A. polyphaga for 3 hours (Figure 2). Electron microscopy was used to confirm the intracellular localization (Figure 3). The bacilli were seen in phagocytic vacuoles with the phagosomal membrane tightly opposed to the bacillary surface (tight phagosomes) (Figure 3D) or, less frequently, inside “spacious vacuoles” (Figure 3C). The phagosomes contained single (Figure 3C and D) or groups of bacilli (not shown). About 45% of the bacilli in the electron micrographs looked normal according to the parameters previously defined [47], including the presence of an asymmetric profile of the cytoplasmic membrane with the outer layer thicker and denser than the inner layer (Figure 3B). As is typical of the ultrastructure of normal mycobacterial cell envelopes [48], an electron-transparent layer of the cell wall of M. ulcerans was observed (Figure 3B). No electron-transparent zone [49] was seen around the intracellular bacilli (Figure 3). Figure 4 shows that viable M. ulcerans persisted within A. polyphaga for the duration of the experiment (two weeks) although their numbers decreased with 1 to 2 log.
181 amoeba cultures were obtained from 134 out of 148 collected samples (90.5%). The isolation frequency of amoebae did not differ significantly between BU endemic and non-endemic sites (p = 0.954, χ21 = 0.004). Different habitats yielded different frequencies of amoeba isolation (p = 0.044, χ23 = 8.084), with the highest detection frequency in detritus (97.8% vs. 76.9% in water, and 88.9% in biofilms). There was no significant difference between the sampled water bodies (estimated effect: 0.03; 95% CI: −0.19 to 0.24). Because 15 of the 181 amoeba cultures did not survive transportation and/or storage, mycobacteria were only searched for in the remaining 166 amoeba cultures (isolated from 124 different samples).
IS2404 was detected by real-time PCR in 3 out of 148 samples, after extracting the DNA directly from these samples. In only one of them IS2606 and KR-B were also detected, strongly suggesting the presence of M. ulcerans in that sample (Table 1). The Δ CT (IS2606 - IS2404) value of 1.96 approaches the known fold difference in copy numbers between IS2404 and IS2606 for M. ulcerans, i.e. 2.3 [44]. However, the high CT values in all 3 of the positive assays (Table 1) imply the presence of less than a genome in the 1 µL DNA extract added to the PCR-mixture so that the Δ CT's cannot be considered as true representations of the relative copy numbers of the repeated sequences. Therefore in the two IS2404 positive yet IS2606 and KR-B negative samples the presence of M. ulcerans cannot be denied nor confirmed.
Out of the 166 amoeba cultures tested (originating from 124 different samples), seven were positive for IS2404 (Table 1). Again, given the high CT values (Table 1), also here less than a genome was present in the 1 µL of DNA extract added to the PCR mixture. The IS2404 positive amoeba cultures were isolated from BU endemic as well as BU non-endemic communities and from different microbial habitats. None of the IS2404-containing amoeba cultures tested positive for IS2606 or KR-B. However, because of the low mycobacterial DNA content neither the absence or presence of M. ulcerans can be confirmed. None of IS2404 positive amoeba cultures were isolated from samples that had already been found positive for IS2404 in DNA extracted directly from the samples.
The following amoebae were identified among the IS2404 positive cultures: Vahlkampfia avara (99% identical with the V. avara sequence in Genbank), a close relative of V. inornata (92% identical with the V. inornata sequence in Genbank), A. lenticulata (T5 genotype), Acanthamoeba sp. T11 genotype and Acanthamoeba spp. T4 genotype. One of the IS2404 positive agar plates that supposedly supported an amoeba culture did not contain amoebae at the time of IS2404 detection.
Identification of the IS2404 negative amoebae will be detailed in a subsequent study by Amissah et al. (in preparation).
The geographical origins of the IS2404 positive samples and amoeba cultures did not show any distribution pattern. We detected IS2404 in at least 1 sample and/or amoeba culture from all sampled localities, except Bebuso.
As described in the methods section, subsamples were made to cultivate extracellular and intracellular mycobacteria. Twenty-six of the 148 samples were excluded from further analysis due to contamination of one or both of the subsamples. From 15 samples (12.2%) only intracellular mycobacteria were isolated, from 17 samples (13.9%) only extracellular mycobacteria, and from 32 samples (26.2%) both intra- and extracellular mycobacteria were isolated. Details are given in Table 2. In general the difference between the isolation frequency of extracellular and intracellular mycobacteria was not significant (χ21 = 0.17, p = 0.89). To assess whether the intracellular life style was more frequent in certain sites or certain habitats, we determined the relative isolation frequency of intracellular mycobacteria (i.e. the number of samples from which intracellular mycobacteria were cultivated divided by the total number of samples from which we cultivated mycobacteria –intracellular and/or extracellular), and related this to BU endemicity, sampling sites and habitat type. The relative isolation frequency of intracellular mycobacteria did not differ between BU endemic and non-BU endemic areas (0.77 vs. 0.68; p = 0.86, χ21 = 0.03). The type of habitat, however, did have a significant effect on the relative occurrence of intracellular mycobacteria (p = 0.002; χ23 = 15.1): intracellular mycobacteria were more frequently isolated from detritus samples (relative isolation frequency of 0.95) than from biofilm samples (relative isolation frequency of 0.63; p = 0.01).
Based on a 821 to 837 bp portion of their 16S-rRNA gene sequence, 76 isolated mycobacteria (of intra- and extracellular origin) could be identified to the species level, with their sequence >99% identical to reference strains of which the sequence is present in GenBank. For 27 isolates, 16S rRNA-DNA sequence based identification was not possible due to the presence of a mixture of different species in the culture. An overview of the identified mycobacterial isolates is shown in Table 3 and Table S1. Species diversity did not show a marked difference between any type of isolation source (Table 3, S1).
Mycobacterial 16S-rRNA-DNA was detected in 29 amoeba cultures (17.5%), isolated from 25 out of 124 samples (20.2%). Mycobacterial presence was confirmed by microscopy in 13 of these positive cultures; 1 to 100 AFB were detected per 100 fields, which approximates to orders of 103 to 105 bacilli per culture of amoebae. No AFB were observed co-localising with the amoebae, however.
Amoebae are good candidates to be a reservoir of the elusive M. ulcerans, but this relationship has not yet been thoroughly investigated. Here, we study the potential for amoebae to host M. ulcerans both experimentally as by sampling an aquatic environment. Our results show that M. ulcerans can indeed be phagocytosed in vitro by A. polyphaga and that viable bacilli persist for at least 2 weeks. We observed both tight and spacious phagosomal vacuoles containing M. ulcerans in infected A. polyphaga with transmission electron microscopy, as has been described for M. avium-infected A. castellanii [29], [47] and for macrophages infected with mycobacteria [50], including M. ulcerans [51]. The observed reduction in the number of viable bacilli is probably due to bacilli that are expelled by the amoebae after a phase of intracellular multiplication, as has been reported for in vitro mycobacteria-infected macrophages [52], [53]. The kanamycin in the medium therefore probably killed released bacteria and resulted in an underestimation of the capacity of M. ulcerans to grow inside the protozoan cells. Compared to a noninfected A. polyphaga monolayer, infection with the M. ulcerans strains did not result in a higher loss of cells (data not shown) indicating that the infection did not affect A. polyphaga viability.
By analysing samples from an aquatic environment in BU endemic and nearby non-endemic communities in southern Ghana, we found several mycobacterium species intracellularly in eukaryotic micro-organisms. Most of the mycobacterium species we identified are potentially pathogenic to humans [54]–[57]. We did not isolate M. ulcerans, even not by successively passaging IS2404 positive specimens and amoeba cultures in mouse footpads, the method that has led to the only successful isolation of M. ulcerans from the environment [15] (data not shown). We isolated mycobacteria as frequently from an intracellular source as free-living, suggesting that it is quite common for several species of mycobacteria to infect micro-organisms in natural circumstances. The intracellular lifestyle was found significantly more frequent in detritus samples compared to water and biofilm samples. This could be due to the low oxygen levels in this organic debris. For several environmental bacteria (including M. avium) it has been shown that oxygen depletion (and other conditions that typically dominate in animal intestines) triggers the invasion of and enhances the survival within host cells [58], [59].
We detected the marker IS2404 in 1 water and 2 biofilm samples collected in a BU endemic and a nearby non-endemic community in southern Ghana. In addition, we detected the same marker in 6 amoeba cultures obtained from other samples. This is the first report of the detection of the marker IS2404, suggestive of M. ulcerans presence, in amoeba cultures isolated from the environment. It is noticeable that we tripled our detection frequency of IS2404 by searching in the amoeba cultures in addition to the original samples. We could not observe AFB in the smears of these amoeba cultures, but one must take into account that M. ulcerans and other IS2404 containing mycobacteria grow very slowly and thus were probably present in very low quantities on the amoeba cultures. On LJ-medium, M. ulcerans colonies only appear after an average of 10 weeks in primary culture from clinical specimens [60]. From environmental sources, M. ulcerans was only isolated once despite numerous attempts [15].
Other mycobacteria were also quite frequently detected in amoeba cultures (in 17.5%), by a PCR assay targeting their 16S-rRNA gene. For these, mycobacterial presence could be confirmed by microscopy in almost half of the positive cultures. However, the AFB were not observed inside or attached to the amoebae. The fact that in our study mycobacteria could still be detected after multiple subcultures of the amoeba cultures, suggests that the mycobacteria were multiplying extracellularly on the agar plates. IS2404 was in fact also detected on one agar plate on which the amoebae did not survive subculturing. Similarly, in a co-culture study of M. avium and A. polyphaga, M. avium was shown to persist and multiply both intracellularly and extracellularly as a saprophyte on the excrement of A. polyphaga, and mycobacterial growth was most extensive extracellularly [47].
The successful uptake and persistence of M. ulcerans inside A. polyphaga in vitro and the higher detection frequency of IS2404 in amoeba cultures as opposed to the crude samples from the environment suggest that amoebae may act as a host for M. ulcerans in natural circumstances. However, our data do not reveal a significant role for protozoa in the distribution patterns of BU disease in humans, so we remain sceptical about their involvement in the direct transmission of M. ulcerans to humans. If a protozoan were to be principally responsible for the observed distribution pattern of BU in humans, one would expect either a particular species with a limited distribution to harbour M. ulcerans, or otherwise several species that only do so in areas where BU actually occurs in humans. In this study, however, we detected IS2404 as frequently in amoeba cultures isolated from BU endemic as from non-BU endemic communities. Moreover, 5 different protozoan species from two divergent families were identified in the IS2404 positive amoeba cultures, some of which are known to be cosmopolitan. On the other hand, we cannot completely rule out that some or all of the IS2404 we detected originated from different mycobacterial species than M. ulcerans.
More environmental research is needed in Africa if we want to understand the distribution of BU, and to prevent its transmission from the environment to humans. Environmental research of M. ulcerans has been severely hampered by the difficulties of detecting the pathogen in the environment. Our results indicate that perhaps amoeba cultures can serve for improved detection of M. ulcerans in environmental samples. Co-cultivation with an existing amoeba culture is a technique to selectively isolate amoebae-resistant bacteria that are difficult to grow from the environment [61] and has already been proven successful in the identification of new pathogens and their distribution patterns [62].
|
10.1371/journal.pgen.1005082 | Glycosyl Phosphatidylinositol Anchor Biosynthesis Is Essential for Maintaining Epithelial Integrity during Caenorhabditis elegans Embryogenesis | Glycosylphosphatidylinositol (GPI) is a post-translational modification resulting in the attachment of modified proteins to the outer leaflet of the plasma membrane. Tissue culture experiments have shown GPI-anchored proteins (GPI-APs) to be targeted to the apical membrane of epithelial cells. However, the in vivo importance of this targeting has not been investigated since null mutations in GPI biosynthesis enzymes in mice result in very early embryonic lethality. Missense mutations in the human GPI biosynthesis enzyme pigv are associated with a multiple congenital malformation syndrome with a high frequency of Hirschsprung disease and renal anomalies. However, it is currently unknown how these phenotypes are linked to PIGV function. Here, we identify a temperature-sensitive hypomorphic allele of PIGV in Caenorhabditis elegans, pigv-1(qm34), enabling us to study the role of GPI-APs in development. At the restrictive temperature we found a 75% reduction in GPI-APs at the surface of embryonic cells. Consequently, ~80% of pigv-1(qm34) embryos arrested development during the elongation phase of morphogenesis, exhibiting internal cysts and/or surface ruptures. Closer examination of the defects revealed them all to be the result of breaches in epithelial tissues: cysts formed in the intestine and excretory canal, and ruptures occurred through epidermal cells, suggesting weakening of the epithelial membrane or membrane-cortex connection. Knockdown of piga-1, another GPI biosynthesis enzymes resulted in similar phenotypes. Importantly, fortifying the link between the apical membrane and actin cortex by overexpression of the ezrin/radixin/moesin ortholog ERM-1, significantly rescued cyst formation and ruptures in the pigv-1(qm34) mutant. In conclusion, we discovered GPI-APs play a critical role in maintaining the integrity of the epithelial tissues, allowing them to withstand the pressure and stresses of morphogenesis. Our findings may help to explain some of the phenotypes observed in human syndromes associated with pigv mutations.
| Cell surface proteins, such as receptors, either integrate into the plasma membrane through a transmembrane domain or are tethered to it by an accessory glycosylated phospholipid (GPI) anchor that is attached to them after they are made. The GPI-anchor biosynthesis pathway is highly conserved from yeast to humans and null mutations in any of the key enzymes are lethal at early developmental stages. Point mutations in several genes encoding for GPI-anchor biosynthesis enzymes have been linked to human disease. Specifically, mutations in PIGV are associated with multiple congenital malformations, including renal and anorectal malformation and mental retardation. It is currently not known how the mutations in PIGV lead to these diseases. Here we describe a point mutation in the PIGV ortholog of the nematode Caenorhabditis elegans, pigv-1, which is found to cause a high degree of embryonic lethality. We documented a substantial reduction in the level of GPI-anchors in the mutant. Importantly, following its development using 4D microscopy and employing tissue-specific rescue, we identified loss of epithelial integrity as the primary cause of developmental arrest. Our results highlight the importance of GPI-anchored proteins for epithelial integrity in vivo and suggest a possible etiology for human diseases associated with PIGV mutations.
| Proteins can attach to the plasma membrane by intrinsic transmembrane domains or by post-translational modifications with lipid moieties. One such lipid modification, tethering proteins to the outer leaflet of the plasma membrane (PM), is a glycosylphospatidylinositol (GPI) anchor, whose synthesis and attachment to proteins in the endoplasmic reticulum (ER) is a multi-step process involving >30 enzymes [1]. Proteins subjected to GPI anchor modification harbor two signal peptides, an N terminal signal peptide that targets them to the ER and a C terminal signal peptide that serves as a marker for GPI attachment [2]. The basic structure of a GPI anchor consists of a phosphoethanolamine linker, a glycan core and a phospholipid tail. The glycan core of GPI can be modified by other phosphoethanolamine or other sugar groups, giving rise to diverse GPI anchor structures [3].
From the ER, GPI-anchor proteins (GPI-APs) are transported to the Golgi, where the phospholipid tail of GPI anchor undergoes lipid remodeling to increase the efficiency of membrane binding. GPI-AP are then sorted and subsequently delivered to the outer leaflet of the PM through the trans-Golgi network [4]. GPI-APs are mostly localized to the apical membrane of polarized cells and are enriched in domains known as lipid rafts [5,6]. Extraction of lipid rafts using weak non-ionic detergent pulls down GPI-APs along with the rafts [7]. Apical polarization of GPI-APs has also been observed in vivo in epithelial cells of pancreas, intestine and urinary bladder in GFP-GPI-expressing mice. GPI-APs are also present in non-polarized tissues with equal distribution across the membrane [8].
GPI-APs have very diverse functions in various cells across species. They are required for viability and cell wall biosynthesis in yeast, act as defense against host immune system in trypanosome, mediate cell-cell interactions, signal transduction, and perform enzymatic activity in mammalian cells [1,3,9]. At the tissue level, GPI-APs were shown to be important for germline and oocyte development in the nematode Caenorhabditis elegans (C. elegans). Mutation in piga-1 (ortholog of mammalian PIGA), the catalytic subunit of phosphatidylinositol N-acetylglucosaminyltransferase complex, the first enzyme playing a role in GPI biosynthesis, decreases the number of germline mitotic cells and compromises oocyte formation and maturation [10]. At the organism level, GPI-APs were shown to be essential for mouse and human embryogenesis. A complete PIGA knockout mouse could never be obtained and this may be explained by the fact that mouse embryonic stem cells depleted of PIGA form embryoid bodies that are arrested at an early stage of differentiation [11,12].
While each individual GPI-AP has a unique function that depends on the protein itself, there is evidence to suggest that GPI anchors themselves, independent of the proteins they anchor, play a role in organizing the PM. Moreover, despite the fact that GPI-anchors are positioned in the outer leaflet of the PM, they have been shown to be affected by the organization of the actin cortex underlying the PM [13,14].
Mounting evidence supporting essential roles for GPI-APs during human embryogenesis comes from human genetic studies conducted in the past decade. Missense mutations in genes encoding enzymes catalyzing various steps of GPI anchor biosynthesis, such as PIGW that catalyzes attachment of acyl group to phosphatidylinositiol [15,16], PIGV that catalyzes transfer of second mannose to GPI intermediate [17,18], PIGT that attaches GPI to proteins [19] and PGAP2 that modifies the phospholipid tail of PI [20] result in congenital diseases known as hyperphosphatasia mental retardation syndrome (HPMRS), Hirschprung disease, morphological malformation and renal anomalies [21,22,23,24].
Studies on the roles of GPI-APs during embryogenesis have been hindered by the difficulties in obtaining viable mutants for GPI biosynthesis enzymes. In this study, we exploit a hypomorphic temperature-sensitive allele of pigv-1 (human PIGV ortholog) in C. elegans to investigate the role of GPI-APs during embryogenesis. We found that GPI-APs are vital for the integrity of epithelial tissues during morphogenesis, suggesting an essential role for GPI-APs in stabilizing the apical membrane of epithelial tissues under stress.
In the course of our whole genome sequencing of maternal-effect morphologically abnormal (mal) mutants isolated by Hekimi et al. [25] in an ethyl methanesulfonate (EMS) mutagenesis screen, we discovered that mal-3(qm34) (Fig. 1A) has a missense mutation at amino acid 361 of previously unassigned gene T09B4.1, converting glycine to glutamate (Fig. 1B). BLAST analysis of the T09B4.1 protein sequence suggested that it is an ortholog of human GPI mannosyltransferase 2, which is known as PIGV (S1A–B Fig.). The mutation was verified by conventional sequencing, and from this point onwards we refer to mal-3(qm34) as pigv-1(qm34).
Sometimes, EMS mutagenesis results in hypomorphic alleles that are temperature-sensitive. We investigated this possibility by growing pigv-1(qm34) worms at 15, 20 and 25°C and measuring their viability at each temperature. We found that pigv-1(qm34) is a heat-sensitive allele, with more than 80% embryonic lethality at 25°C (Fig. 1C, S1 Table). We followed embryogenesis by time-lapse differential interference contrast (DIC) microscopy and observed phenotypes resulting from pigv-1 inactivation at 25°C. At this temperature C. elegans embryogenesis takes 10.5 hours from the first division till hatching (S2 Fig.). The first 3 hours of embryogenesis are characterized by formation of the founder cells, rapid cell division and gastrulation. At around 3 hours the epidermis is born on the dorsal side of the embryo and the next 3.5 hours are dominated by epidermal morphogenesis, a three step process made up of intercalation, enclosure, and elongation [26]. Loss of pigv-1 resulted in defects appearing during elongation with cysts forming inside the embryo and/or cells leaking out from the embryo body, resulting in elongation arrest and embryonic lethality (Fig. 1D). Quantification of 176 pigv-1(qm34) embryos showed that over 80% displayed ruptures and/or cyst formation and arrested in elongation (Fig. 1E). Few escapers hatched and became L1 larva with body shape defects (Fig. 1A).
Utilizing the heat sensitivity of pigv-1(qm34), we determined the developmental period when pigv-1 activity is required through reciprocal temperature shift experiments. We found that pigv-1 activity is essential from the one cell embryo stage until elongation. Once elongation was underway inactivation of PIGV-1 had less effect on embryogenesis (Fig. 1F, S3 Table).
We confirmed that these phenotypes are caused by the mutation in pigv-1 by rescue experiments. Transformation of pigv-1(qm34) worms with a fosmid that contains a wild-type allele of the pigv-1 gene significantly rescued embryonic lethality (P<0.01) (Fig. 1G, S4 Table). Expression of a gfp-tagged pigv-1 under the control of 2.4 kb upstream of the pigv-1 start codon failed to rescue embryonic lethality in pigv-1(qm34) mutant worms, most likely due to low expression. On the other hand, expression of pigv-1 under the control of the erm-1 promoter, which resulted in 3 fold stronger expression, successfully rescued embryonic lethality of pigv-1(qm34) (Fig. 1G, S4 Table), confirming that the mutated gene causing lethality in the pigv-1(qm34) strain is pigv-1.
To visualize GPI-AP distribution during embryogenesis, we used Alexa-488 labeled proaerolysin (FLAER), a bacterial toxin that binds specifically to GPI-AP [27], to label embryos at different stages of development (Fig. 2). In the one-cell embryo, GPI-APs accumulated at perinuclear areas and were enriched in the anterior cytoplasm (Fig. 2, first row). As soon as new membrane was delivered to the cell surface, during cell division, GPI-APs accumulated on the plasma membrane (Fig. 2, second row). During gastrulation we observed GPI-APs accumulated at the membrane of all cells (Fig. 2, fourth row). While being uniformly localized on membrane of all cells in the early embryo, non-uniform GPI-AP distribution was observed upon tissue differentiation. For example, during dorsal intercalation, GPI-APs were highly enriched on pharyngeal cell membranes in a non-polarized manner, whereas later on, during elongation, they became apically enriched (Fig. 2, fifth to seventh rows).
To gain insight into the spatial and temporal activity of the PIGV-1 enzyme during embryogenesis, we visualized an N-terminally GFP-tagged PIGV-1 driven by its endogenous promoter in an extrachromosomal array. We could not detect GFP::PIGV-1 in the early embryo, possibly due to silencing of the transgenes in the germline. Later in development the expression level was low. Nevertheless, we observed GFP::PIGV-1 to be prominent in the epidermis, pharynx, intestine, rectum and excretory cell (Fig. 3A), all tissues with epithelial character. At the subcellular level, PIGV-1 localized to intracellular structures that appear to be ER [28,29].
Using FLAER staining as readout for GPI-anchor biosynthesis, we compared the intensity of FLAER staining in pigv-1(qm34) at the permissive (15°C) and restrictive (25°C) temperatures. At 15°C, pigv-1(qm34) embryos exhibited FLAER levels similar to wild type, whereas at 25°C, abrogation of PIGV-1 activity led to a 4-fold reduction in the FLAER signal (Fig. 3B, second and third rows). FLAER staining was restored to wild-type levels in pigv-1(qm34) embryos at 25°C when pigv-1 was expressed in all epithelial tissues by the erm-1 promoter (Fig. 3B, fourth row). Taken together these data show that GPI-APs are enriched in epithelial tissues and the abundance of GPI-APs at the cell membrane is dependent on the activity of PIGV-1.
In mammalian cells, more than 30 enzymes are known to regulate the GPI anchor biosynthesis pathway. Many of these enzymes have orthologs in C. elegans (S1B Fig.). A previous study has shown that RNAi-mediated knockdown of most C. elegans GPI anchor biosynthesis enzymes does not lead to any phenotype and two of them, namely pigk and pigo resulted in sterility [10]. We scanned a range of feeding RNAi conditions for pigk and pigo, with the rationale that partial loss of function might bypass their requirement for germline development, and expose a possible role in embryogenesis. However, we observed either sterility or no phenotype when each of the two enzymes was depleted. Thus, we turned our attention towards piga-1(tm2939) mutant worms characterized in the previous study [10]. Progeny of homozygous piga-1(tm2939) worms are embryonic lethal, and they display a deformed eggshell due to increased osmotic sensitivity during germline development. To uncouple the functions of PIGA-1 during germline development and embryogenesis, we used piga-1(tm2939) worms rescued by piga-1 expression under the control of lag-2, a distal tip cell promoter. First, we examined whether lag-2 drives piga-1 expression during embryogenesis and found piga-1 expressed ubiquitously during embryogenesis (Fig. 4A). Since the plag-2::piga-1::gfp construct is expressed as an extrachromosomal array, some embryos lose piga-1 expression during embryogenesis. We identified which embryos lost the extrachromosomal array and followed their embryonic phenotypes. While all the embryos retaining piga-1 expression during embryogenesis hatched, ~50% of the embryos devoid of piga-1 expression were arrested during elongation. In one-third of arrested embryos, internal cells leaked out from the embryo body (Fig. 4B), a phenotype reminiscent of pigv-1(qm34) embryos, suggesting that weakening of epithelial tissue integrity is not a specific phenotype of pigv-1 loss of function, but rather a general consequence of disruption of the GPI biosynthesis pathway.
The elongation phase of C. elegans embryogenesis is characterized by the formation of circumferential actin bundles (CFB) in the dorsal and ventral epidermal cells and actomyosin contraction in the lateral epidermal cells [30]. Contractility of the muscle tissues is known to be required for elongation beyond the 2-fold length [31]. We tested whether the elongation arrest occurring upon pigv-1 inactivation is caused by defective CFB and/or muscle organization. We examined CFB in pigv-1(qm34) embryos using an F-actin reporter (VAB-10 actin binding domain tagged with GFP) and found that CFB structure is indistinguishable from that of wild type embryo (S3A Fig.). Myotactin antibodies were utilized to examine muscle organization and we observed no difference between muscle organization in wild type and in pigv-1(qm34) embryos (S3B Fig.). Moreover, some pigv-1(qm34) embryos elongated beyond two-fold stage. These results suggest that elongation arrest in pigv-1(qm34) embryos is not caused by defects in CFB or muscle structure.
We then set out to characterize the embryonic phenotypes resulting from pigv-1 inactivation in more detail. We employed several cell junction and membrane markers expressed in specific tissues to pinpoint the location of the defects. Using AJM-1::GFP and HMP-1::GFP as a marker for epidermal apical junctions, we observed gaps between epidermal cells through which internal cells leaked out, most often from the embryo anterior (Fig. 5A and S1–S4 Movies). In some embryos, the gap is created by misalignment of leading ventral epidermal cells coming from opposite ends to enclose the embryo at the ventral midline (Fig. 5A). Using a plasma membrane marker specifically expressed in the pharynx and intestine, we identified cysts to be located at the basal side of the intestine, and using CED-10::GFP, which highlights plasma membrane of all cells, we observed cysts to be located between the intestine and its surrounding basal lamina (Fig. 5B-C, S4 Fig.). Using AJM-1::GFP to highlight the apical junctions of intestinal cells we observed widening of the lumen in pigv-1(qm34) embryos (Fig. 5D). We measured intestinal lumen width of wild type and pigv-1(qm34) embryos in early (2–2.5 fold) and later (3–3.5 fold) stages of elongation and found that the lumen width of pigv-1(qm34) embryos was significantly wider (P<0.05) than that of wild type embryos at late elongation (Fig. 5D). Furthermore, we noticed that the intestine in pigv-1(qm34) embryos was often twisted (S6C Fig.).
Using mCherry-tagged AQP-8, a water channel specifically localized to the excretory canal, expressed at a low level which maintains a normal translumenal flux, we found the excretory canal to be another location where cysts formed in pigv-1(qm34) embryos (Fig. 5E). In contrast with the intestinal cysts that formed in extracellular space the excretory canal cysts formed within the cell. Few embryos that survived embryogenesis hatched with excretory canal cysts (S5 Fig.). The excretory canals in larvae with excretory canal cysts were usually very short. Not only the length, but the branching of the excretory canal is also affected in pigv-1(qm34) embryo. In wild-type worms the excretory canal extends four tubules shaped like an H: a pair towards anterior and another pair towards posterior from the cell body. However, the excretory canal in pigv-1(qm34) embryo often has one or two more tubules extending from the cell body or branching from the original tubules (S7 Fig.).
GPI-APs are known to be targeted to apical membranes in polarized epithelial cells [5,6]. We therefore examined whether the apicobasal polarity of epithelial cells is affected in pigv-1(qm34) embryos. We observed that the apical markers PAR-6 and PKC-3 were correctly localized on the apical intestinal and excretory cell membranes in pigv-1(qm34) embryos (S6A Fig.), and AJM-1 was localized at the apical side of epidermal and intestinal cell-cell junctions (S6B–C Fig.). Conversely, the basolateral marker LET-413 was localized to the basolateral membranes in epidermal and intestinal cells in pigv-1(qm34) embryos, indistinguishable from wild type (S6B Fig.). Similarly, the intermediate filament IFB-2 was correctly localized beneath the apical membrane in intestinal tissue in pigv-1(qm34) embryos (S6C Fig). Altogether, these results rule out a polarity defect as the underlying cause for the pigv-1(qm34) mutant phenotypes.
We observed pigv-1 loss of function to affect the integrity of three epithelial tissues: epidermis, intestine and excretory canal. However, it was not immediately evident which defective tissue was responsible for the embryonic lethality. To address this question we restored pigv-1 expression specifically in each epithelial tissue or in all epithelial tissues of pigv-1(qm34) worms and determined their embryonic viability. We employed the lin-26 promoter to drive expression in the epidermis, the pha-4 promoter to drive expression in the pharynx and intestine, the aqp-8 promoter to drive expression in the excretory canal, and the erm-1 promoter to drive expression in all epithelial tissues (Fig. 6A). All promoters drove pigv-1 expression at comparable levels.
While restoring pigv-1 expression in the pharynx-intestine or in the excretory canal partially reduced pigv-1(qm34) embryonic lethality, restoring pigv-1 expression in the epidermis did not significantly reduce pigv-1(qm34) embryonic lethality (Fig. 6B, S4 Table). The most significant rescue of embryonic lethality achieved by expression of pigv-1 in a single tissue was a 17% reduction in lethality. In contrast, expressing pigv-1 in all epithelial tissues using the erm-1 promoter sharply decreased pigv-1(qm34) embryonic lethality down by 62%, comparable to the sum of the embryonic rescue of each epithelial tissue (Fig. 6B, S4 Table). Thus, it appears that pigv-1 function is required in all epithelial tissues for embryonic viability.
The cytoskeletal cortex underlying the plasma membrane provides it with structural support and protects the membrane from mechanical stress. Depletion of spectrin in erythrocytes changes membrane rigidity and subsequently leads to cell fragmentation [32]. Thus, we hypothesized that strengthening the actin cortex in pigv-1(qm34) embryos might positively affect membrane integrity. First, we examined whether providing more actin has any impact on epithelial membrane integrity. We examined pigv-1(qm34) embryos overexpressing YFP::ACT-5 in the epidermis and intestine and found that embryonic lethality in these worms is indistinguishable from that of pigv-1(qm34) (Fig. 7A, S5 Table). We then explored whether strengthening the link between the actin cortex and the cell membrane might influence membrane integrity. We chose worms overexpressing ERM-1::GFP at a level which does not cause any phenotypic defect since a previous study showed that at high levels of expression ERM-1 leads to formation of excretory canal cysts [33]. We crossed the ERM-1::GFP-overexpressing worms with pigv-1(qm34) and found that embryonic lethality was significantly reduced in the pigv-1(qm34);ERM-1::GFP strain. Depletion of overexpressed ERM-1 by gfp(RNAi) in this stain reverted embryonic lethality back to pigv-1(qm34) level, confirming ERM-1 overexpression is responsible for rescuing pigv-1-associated embryonic lethality (Fig. 7A, S5 Table). Careful examination of embryogenesis in pigv-1(qm34) embryos overexpressing ERM-1::GFP revealed strong suppression of pigv-1 phenotypes, and a significant portion of embryos (36%) hatched without any visible defects (Fig. 7B, S2 Table). Considering ERM-1 localization at intestine and excretory canal apical membranes, we reasoned that ERM-1 overexpression could rescue apical-associated phenotypes in these tissues. Measuring the width of intestinal lumen we found that it was reduced to the wild type dimension (Fig. 7C).
To gain insight into the mechanism of pigv-1-phenotype rescue by ERM-1 overexpression, we examined whether endogenous ERM-1 distribution and level were altered in pigv-1(qm34) embryos. Immunolabeling with ERM-1 antibodies showed no difference in ERM-1 distribution or level between wild type and pigv-1(qm34) embryos (Fig. 8A). We then asked whether ERM-1 might rescue pigv-1 mutant by enhancing the residual pigv-1 activity and restoring the level of GPI-APs. Using FLAER as the probe for GPI-APs, we found that GPI-AP level in pigv-1(qm34) embryos overexpressing ERM-1 is similar to pigv-1(qm34) embryos alone, suggesting that ERM-1 overexpression does not rescue pigv-1 embryonic lethality by restoring GPI-APs (Fig. 8B).
GPI anchor is an important post-translational protein modification whose functions and mechanisms have been widely studied using unicellular organisms and mammalian cells in culture [1,6,9]. However, the role of GPI biosynthesis in animals remains poorly understood. In humans, somatic mutations in PIGA gene loci lead to paroxysmal nocturnal hemaglobinuria, a disease characterized by increased susceptibility of erythrocytes to lysis by the complement immune system [34]. No heritable mutation in piga gene in human has been identified, suggesting that PIGA function is required during embryogenesis. Indeed, deletion of PIGA gene in mice, which completely abrogates GPI biosynthesis, resulted in early embryogenesis defects [11,35]. However, this condition precludes the study of GPI function throughout embryogenesis.
Also in C. elegans, a null mutation in piga-1 results in germline defects and early embryonic lethality. In this study, we circumvented the early requirements for GPI-APs by using a temperature sensitive hypomorphic allele of pigv-1. The amount of GPI-APs remaining upon pigv-1 inactivation was sufficient for normal germline development, thus enabling us to uncover their requirement during embryogenesis. We showed that GPI-APs are present and function throughout embryogenesis. Interestingly, the phenotypes of pigv-1 inactivation, i.e., weakened epithelial tissues, are manifested only late in embryogenesis during the elongation stage of epidermal morphogenesis. This may be due to increased mechanical forces generated by actomyosin in muscle and epidermis tissues at that stage.
Although present at the membrane of all cells in the C. elegans embryo, pigv-1 loss of function exhibits no defect in early development events, such as gastrulation or tissue differentiation. This is in contrast to mammalian embryogenesis, in which complete PIGA depletion results in defects in tissue differentiation [35]. One possible explanation for this difference is that the residual GPI-APs in pigv-1 animals are sufficient for normal differentiation. Another reason could be differences in the proteins regulating differentiation. Tissue differentiation in mammals is regulated by BMP/ TGF-β signaling whose activation requires GPI-anchored co-receptors, Dragon and Cripto-1 [35,36]. Although present in C. elegans, BMP/TGF-β signaling is not required during embryogenesis, but operates during postembryonic development, regulating body size [37,38].
Inactivation of pigv-1 in C. elegans embryos resulted mainly in epithelial defects. Failures in epidermal enclosure and intestinal cyst formation are consistent with weaker cell-cell adhesion. In the epidermis, improper cell-cell adhesion creates gaps between ventral epidermal cells from which internal tissues leak out during elongation. Compromised cell-cell junctions in the intestine, which has higher osmotic pressure than the surrounding tissues, would allow passage of low molecular weight substances, such as water molecules, from the intestinal lumen to the intestine basal side. The presence of a basal lamina separating the intestine from the pseudocoelom results in accumulation of these substances in the form of cysts.
One possible explanation of these results is that the reduction in the amount of one or more specific cell adhesion proteins that are GPI-anchored causes the observed defects. However, amongst the GPI-APs that have been experimentally identified in C. elegans none are known to mediate cell-cell adhesion [7,10]. While we do not rule out the involvement of yet unknown GPI-anchored adhesion proteins, we propose another mechanism to explain the observed epithelial phenotypes that does not depend on a specific protein, but rather on the GPI anchors themselves. Goswami et. al. have shown that cortical actin affects the organization of GPI-AP in the membrane [13]. We propose that GPI-APs are enriched in apical membranes of polarized epithelial cells where they play a role in organizing the membrane into domains that interact with the actin cortex within the cell and through these interactions stabilize the apical membrane. According to this model, a decrease in GPI-APs will lead to fewer membrane-cortex connections and thus to a weaker apical membrane. In support of this idea, we observed a widening of the intestinal lumen in pigv-1 mutant embryos, as would be expected if the apical membrane of the intestine is weakened and thereby cannot resist as well the osmotic pressure from inside the lumen. Further support for this model comes from the finding that overexpression of ERM-1 rescues lumen width and overall embryonic lethality of pigv-1 mutant embryos. ERM-1, the sole C. elegans ortholog of ezrin, radixin and moesin, is a linker protein that has an actin-binding domain and attaches to the PM through its FERM domain, serving to connect the PM with the actin cortex [39,40]. ERM-1 overexpression did not increase the levels of GPI-APs in pigv-1 embryos and hence the reduction in lethality associated with it is most likely due to its membrane-cortex cross-linking function. From this we deduce that loss of GPI-APs leads to a weakening of apical membranes in epithelial cells, irrespective of the proteins attached to the GPI-anchor.
Another epithelial tissue affected by pigv-1 loss of function is the excretory canal. Down regulation of GPI-APs in the excretory canal leads to cysts formation and this cystic excretory canal is usually short, consistent with apical membrane weakening upon the loss of GPI-APs. Unlike the intestine, which is a multicellular tubule, the excretory canal is a unicellular tubule that extends actively during embryo elongation. It has been demonstrated that a balance between membrane-actin cortex recruitment and translumenal flux is essential for the excretory canal extension [33]. Weakened apical membrane upon down regulation of GPI-APs in pigv-1 mutant embryo may prevent further recruitment of membrane components and actin undercoat to extend the canal, creating an imbalance between the two forces. Consequently, the dominant force, the translumenal flux is transmitted to enlarge the canal diameter, resulting in cysts formation. Another phenotype we observed in the excretory canal of pigv-1 mutants is ectopic branching. To our knowledge such a phenotype has not been associated with loss of function of any gene so far, opening a new avenue to study the regulation of tubular branching.
Besides epithelial tissues, loss of GPI-APs in C. elegans may also affects neuronal and/or muscle tissues, suggested by the lethargic phenotype of pigv-1(qm34) worms, although the GPI-APs responsible for this phenotype remains to be identified. The short excretory canal that has been observed occasionally in pigv-1(qm34) could also result from the loss of GPI-APs from neuronal membrane. The neuronal cell adhesion molecules (NCAM) that are essential for axon outgrowth and pathfinding have been demonstrated to regulate excretory canal extension [41,42,43]. In the absence of NCAM, the excretory canal does not grow to full extent. Supporting this view, the in vitro and the in silico experiments found several NCAMs (rig-3, rig-6, rig-7 and wrk-1) to be potentially GPI-modified [7,44].
Not all epithelial tissues displayed abnormal phenotypes in pigv-1 mutants. The pharynx and rectum are two epithelial tissues that do not seem to be affected by down regulation of GPI-APs. High enrichment of GPI-APs at pharyngeal membranes compared to other tissues could provide an explanation for the absence of weakened membrane phenotypes. However, this reason does not hold for the rectum, as GPI-APs at rectal membranes are not more abundant than other tissues that display weakened membrane. Since both pharynx and rectum are covered by a cuticle, the most likely explanation for the absence of visible phenotypes is that the cuticle protects both tissues from potential damage resulting from weakened membranes.
Restoring PIGV-1 expression individually in the epidermis, intestine and excretory canal in pigv-1(qm34) embryos revealed that the defects in these epithelial tissues do not contribute equally to the embryonic lethality. The intestine and the excretory canal defects have higher contribution to embryonic lethality as compared to the epidermal defects. This is somewhat unexpected because 56% of pigv-1(qm34) embryos die due to tissues leakage from the embryo interior, indicating that a gap between epidermal cells is present from where the tissues pass through. However, uncontained high pressure built in the intestinal and excretory canal due to cell adhesion defects and membrane weakening may be sufficient to open epidermal junction and push the internal tissues out of embryo interior.
Mutations in the human ortholog of C. elegans pigv-1, PIGV, have been associated in genetic studies with hyperphosphatasia-mental retardation syndrome (a.k.a Mabry syndrome). This autosomal recessive syndrome has a wide spectrum of phenotypes including intellectual disabilities, facial anomalies, hyperphosphatasia, vesicoureteral and renal anomalies, and anorectal anomalies [21]. With the exception of hyperphosphatasia, which is known to be the result of loss of GPI-anchored complement inhibitors in blood cells [34], the proteins and cellular functions that are affected in humans with PIGV mutations are unknown. Although our findings in C. elegans cannot possibly fully explain the cellular physiology of the human disease, it does point to a basic mechanism, i.e., weakening of apical membranes in epithelial cells, that may be playing a role in some of the manifestations of the disease. Furthermore, if it will be discovered that epithelial membrane integrity is affected in human patients then our work also suggests a promising avenue for therapy, i.e., strengthening of the membrane-cortex connection, based on our ERM-1 overexpression results.
Strains were grown and maintained at 20°C under standard conditions [45]. Wild type strain N2 was used as a control. The pigv-1(qm34) was retrieved from an EMS screening conducted by Hekimi et al. [25]. For analysis using GFP fusions, F2 progeny exhibiting pigv-1 phenotypes and carrying the markers were selected from crosses between pigv-1(qm34) and the following strains: SU93 jcIs1[ajm-1::gfp, unc-29(+), rol-6p::rol-6(su1006)] [46], SU265 jcIs17[hmp-1p::hmp-1::gfp, dlg-1p::dlg-1::dsRed, rol-6p::rol-6(su1006)] [47], SU467 pIs7[pha-4p::pm::gfp, rol-6p::rol-6(su1006)] [48], FT17 xnIs3[par-6p::par-6::gfp, unc-119(+)]; unc-119(ed3) III, MOT63 temIs59[pIC26::pkc-3]; unc-119(ed3) III, WS4918 opIS310[ced-1p::yfp::act-5::let-858 3'UTR, unc-119(+)] [49], VJ402 fgEx1[erm-1p::erm-1::gfp, rol-6p::rol-6(su1006)] [33], ML1735 mcIs50[lin-26p::vab-10(actin-binding domain)::gfp, myo-2p::GFP] [50], plag-2p::piga-1::egfp-expressing strain was generated by Murata et al [10].
All plasmids generated in this study were constructed in a modified pPD95.75 backbone. For tissue-specific rescue of pigv-1 loss of function, GFP position was changed to be at the N terminal instead of at the C terminal of the multiple cloning sites (MCS), whereas for AQP-8-expressing plasmid, GFP at C terminal was replaced with mCherry. To construct pigv-1p::gfp::pigv-1 plasmid, pigv-1 promoter (2.4 kb sequence upstream of pigv-1 start codon) and coding sequence were amplified and inserted into SbfI and AgeI sites upstream of gfp in original pPD95.75 vector. Circular PCR was performed to amplify the whole plasmid, but the gfp region using a pair of primers harboring XhoI sites at their 5’ ends. PCR product was then ligated to produce a circular plasmid containing pigv-1 promoter and coding sequence, but without gfp. Second circular PCR was conducted to insert two new restriction sites, i.e.: NotI and BglII between pigv-1 promoter and coding sequence. PCR product was then subjected to digestion using NotI and BglII. gfp coding sequence was amplified from original pPD95.75 and subcloned into pJET (Thermo Scientific). The recombinant plasmid was digested using NotI and BglII and gfp sequence-containing product was ligated to pigv-1-containing pPD95.75, resulting in a plasmid expressing gfp::pigv-1 driven by pigv-1 promoter. Four different promoters were used to rescue pigv-1(qm34) in different tissues: 4.1 kb sequence of lin-26 promoter to drive pigv-1 expression in epidermis, 7.1 kb of pha-4 promoter for expression in pharynx and intestine, 2.2 kb of aqp-8 promoter for expression in excretory canal and 3 kb of erm-1 promoter for expression in all epithelial tissues. They are inserted into modified pPD95.75 at SbfI/NotI sites replacing pigv-1 promoter. Transgenic animals generated by injecting the constructs into the gonad of hermaphrodite animals resulted in the following strains: RZB40 (pigv-1(qm34); msnEx40[lin-26p::gfp::pigv-1; rol-6(su1006)]), RZB41 (pigv-1(qm34); msnEx41[pha-4p::gfp::pigv-1; rol-6(su1006)]), RZB129 (pigv-1(qm34); msnEx129[aqp-8p::gfp::pigv-1; rol-6(su1006)]) and RZB128 (pigv-1(qm34); msnEx128[erm-1p::gfp::pigv-1; rol-6(su1006)]).
To construct aqp-8::mCherry-expressing plasmid, mCherry coding sequence was amplified from pAA64 and ligated to circularly amplified pPD95.75 devoid of gfp sequence using Gibson assembly (NEB). Subsequently, 2.2 kb aqp-8 promoter together with aqp-8 genomic sequence were inserted at SbfI/BamHI sites in modified pPD95.75. Injection of this construct resulted in strain RZB221 (pigv-1(qm34); msnEx221[aqp-8p::aqp-8::mCherry; rol-6(su1006)]).
Microinjection was performed as described by Mello and Fire [51]. Injection mix includes 100 μg/μl salmon sperm DNA digested with PvuII, 20 μg/μl rol-6(su1006) digested with SbfI and 5–10 μg/μl each construct digested with SbfI.
Genomic DNA was extracted from pigv-1(qm34) mutant worms using standard method and subjected to whole genome sequencing using Illumina platform and annotated using MAQGene [52]. The whole genome sequencing and its annotation were performed by Hobert lab (Columbia University). Candidate genes altered in pigv-1(qm34) were narrowed down using genetic mapping results done by Hekimi et al. [25]. Point mutation in pigv-1 gene was confirmed by amplification of pigv-1 gene in pigv-1(qm34) mutant worms, subcloning into pJET vector (Thermo Scientific) and followed by conventional sequencing (First Base). Further validation of pigv-1 missense mutation as the phenotype-causing gene in pigv-1(qm34) worms was done by injection of 100 μg/μl fosmid WRM063BcC08, which contains pigv-1 gene, together with the co-transformation marker rol-6(su1006) into the gonad of pigv-1(qm34) hermaphrodites. F2 rollers were upshifted to 25°C and examined for embryonic lethality.
Ten to fifteen gravid hermaphrodites were placed on the plate and incubated for several hours to lay more than 100 eggs. Hermaphrodites were then removed and the number of eggs laid was counted. Twenty-four hours later, the number of larvae hatched was determined. Each experiment was done in duplicate and repeated five times. Beside experiments determining temperature sensitivity that are conducted at three different temperatures (15°C, 20°C and 25°C), the remaining experiments were conducted solely at 25°C to get the highest extent of pigv-1 inactivation. In this case, L4 larvae were upshifted from 20°C to 25°C for 20 to 24 hours prior to the test.
For upshift experiment, embryos were dissected from gravid pigv-1(qm34) worms grown at 15°C and incubated at 25°C for the duration of embryogenesis. Each embryo was staged and scored for hatching. For downshift experiment, similar procedure was performed, except that pigv-1(qm34) worms were kept at 25°C for 24 hours before downshifted to 15°C. Embryos that do not hatch at the end of embryogenesis were considered as lethal.
Larvae or embryos collected from gravid hermaphrodite, mounted onto 3% agarose padded-glass slide, closed with coverslip and sealed with wax. Normaski images shown in Fig. 1A, B and S2B were captured using a Nikon Ti Eclipse widefield microscope equipped with DIC 1.40NA oil condenser and a charged-coupled device camera Cool Snap HQ2 (Photometrics). All other imaging were done using spinning disk confocal system composed of a Nikon Ti Eclipse microscope with a CSU-X1 spinning disk confocal head (Yokogawa), DPSS-Laser (Roper Scientific) at 491 and 568 nm excitation wavelength and an Evolve Rapid-Cal electron multiplying charged-coupled device camera (Photometrics). For both microscopes, Metamorph software (Molecular Devices) was used to control acquisition. Projected images were created using Fiji.
After 24 hour DIC recording, wild type, pigv-1(qm34) and pigv-1(qm34) embryos expressing ERM-1::GFP were scored as viable or lethal and each category is further classified into four subcategories; i.e.: without visible defects, with cysts and rupture, with cyst only and with rupture only.
IPTG plate used for gfp(RNAi) feeding was prepared as described [53]. Wild type and pigv-1(qm34) L1 larvae expressing ERM-1::GFP were fed using bacterial-feeding strain of gfp for 3 days at 15°C till they become L4 and then upshifted to 25°C for overnight. The absence of GFP signal was verified by using fluorescent stereomicroscope and only those devoid of the signal were subjected for embryonic lethality test.
Fixation and indirect immunofluorescence were performed essentially as described [54]. The following primary mouse antibodies were used: ERM-1 (DSHB; 1/20), AJM-1 (MH27, DSHB; 1/10), myotactin (MH46, DSHB; 1/5) and LET-413 (DSHB; 1/2) and IFB-2 (MH33, DSHB; 1/5). Donkey anti-mouse coupled to Alexa 647 (1/500) (Life technologies) was used as secondary antibodies and proaerolysin coupled to Alexa 488 (FLAER, Protox Biotech) was used to detect GPI-APs. Images were taken on a Nikon Ti Eclipse spinning disk microscope with 100x objective and processed further using Fiji. To measure lumen width in wild type and pigv-1(qm34) mutant embryos, N2 and pigv-1(qm34) embryos expressing AJM-1::GFP were fixed, maximum intensity projection of embryonic intestine in GFP channel was constructed and the widest section of intestinal lumen was determined. The same procedure was done to measure lumen width in pigv-1(qm34) embryos expressing ERM-1::GFP, except that AJM-1 antibodies were used instead of AJM-1::GFP expression.
Statistical analyses were done using Microsoft Excel. Two-tailed Student’s t-test was applied to compare the values.
|
10.1371/journal.pgen.1003940 | Crosstalk between NSL Histone Acetyltransferase and MLL/SET Complexes: NSL Complex Functions in Promoting Histone H3K4 Di-Methylation Activity by MLL/SET Complexes | hMOF (MYST1), a histone acetyltransferase (HAT), forms at least two distinct multiprotein complexes in human cells. The male specific lethal (MSL) HAT complex plays a key role in dosage compensation in Drosophila and is responsible for histone H4K16ac in vivo. We and others previously described a second hMOF-containing HAT complex, the non-specific lethal (NSL) HAT complex. The NSL complex has a broader substrate specificity, can acetylate H4 on K16, K5, and K8. The WD (tryptophan-aspartate) repeat domain 5 (WDR5) and host cell factor 1 (HCF1) are shared among members of the MLL/SET (mixed-lineage leukemia/set-domain containing) family of histone H3K4 methyltransferase complexes. The presence of these shared subunits raises the possibility that there are functional links between these complexes and the histone modifications they catalyze; however, the degree to which NSL and MLL/SET influence one another's activities remains unclear. Here, we present evidence from biochemical assays and knockdown/overexpression approaches arguing that the NSL HAT promotes histone H3K4me2 by MLL/SET complexes by an acetylation-dependent mechanism. In genomic experiments, we identified a set of genes including ANKRD2, that are affected by knockdown of both NSL and MLL/SET subunits, suggested they are co-regulated by NSL and MLL/SET complexes. In ChIP assays, we observe that depletion of the NSL subunits hMOF or NSL1 resulted in a significant reduction of both H4K16ac and H3K4me2 in the vicinity of the ANKRD2 transcriptional start site proximal region. However, depletion of RbBP5 (a core component of MLL/SET complexes) only reduced H3K4me2 marks, but not H4K16ac in the same region of ANKRD2, consistent with the idea that NSL acts upstream of MLL/SET to regulate H3K4me2 at certain promoters, suggesting coordination between NSL and MLL/SET complexes is involved in transcriptional regulation of certain genes. Taken together, our results suggest a crosstalk between the NSL and MLL/SET complexes in cells.
| Covalent modification of N-terminal tails of histone proteins is accomplished by a variety of chromatin modifying complexes. These complexes catalyze at least eight distinct types of histone modifications including acetylation, methylation, phosphorylation, and ubiquitination. Histone modifications may act alone or in a coordinated manner to activate or repress chromosomal processes. For example, a particular histone modification may recruit or activate chromatin modifying complexes that generate a different histone modification. Coordination between hMOF-mediated histone H4K16 acetylation and other histone modifications has been reported by several research groups. The presence of subunits shared between the hMOF-containing NSL and MLL/SET family complexes suggests there may be functional links between two complexes. Consistent with this idea, we identified a set of genes that are co-regulated by the NSL and MLL/SET complexes. Both in vitro and in vivo experimental approaches provide evidence that the NSL HAT functions in promoting histone H3K4 di-methylation activity by MLL/SET complexes. Interestingly crosstalk between hMOF/NSL HAT and MLL/SET HMT activity seems to be unidirectional since there we detected no effect of MLL/SET activity on NSL HAT, either in vitro or in cells.
| The precise organization of chromatin is critical for many cellular processes including gene transcription, recombination, DNA replication and damage repair [1]. Changes of chromatin structures are mainly regulated by epigenetic regulations such as ATP-dependent remodeling of nucleosomes, the incorporation of variants histones into nucleosomes and post-translational modifications of histones [2]. Post-translational modifications of the N-terminal tails of histones including acetylation, methylation, phosphorylation, ubiquitination and ADP-ribosylation may act alone or in a coordinated manner to facilitate or repress chromatin-mediated processes [3]–[5]. Crosstalk between different modifications may be accomplished by a number of mechanisms. For example, an initial histone modification may trigger increased activity of a histone-modifying enzyme. Alternatively, one histone and its modifications affect the modification of a different histone [6]. Thus, acetylation of histone H3 on lysine 18 and lysine 23 promotes the methylation of argine 17 by the CARM1 (coactivator-associated arginine methyltransferase 1) methyltransferase, resulting in activation of estrogen-responsive genes [7]. Also, methylation of H3K4 by COMPASS (complex of proteins associated with Set1) and of H3K79 by Dot1 is totally dependent upon the ubiquitylation of H2BK123 by Rad6/Bre1 in Saccharomyces cerevisiae [8].
hMOF (MYST1), a member of the MYST family of histone acetyltransferases (HATs), is the human ortholog of Drosophila male absent on the first (dMOF) protein [9]. Depletion of hMOF in cells leads to genomic instability, spontaneous chromosomal aberrations, cell cycle defects, reduced transcription of certain genes, and defective DNA damage repair and early embryonic lethality [10]–[13]. Moreover, the role of MOF in the DNA damage response is conserved in mammalian cells and Drosophila [14]. Genome-wide analysis demonstrates that MOF is not only involved in the onset of dosage compensation, but also acts as a regulator of gene expression throughout the Drosophila genome, suggesting the functional diversity of MOF [15]. Recent biochemical purifications have revealed that MOF forms at least two distinct multi-protein complexes, MSL and NSL, in Drosophila and mammalian cells [16]–[18]. Although the functions of MSL and NSL complexes in human cells are not entirely clear, both complexes can acetylate histone H4 at lysine 16 (H4K16), suggesting the importance of acetylation of H4K16 in cells [19]–[20]. Besides H4K16, NSL complex is also able to acetylate other histone H4 lysines, such as H4K5 and H4K8 [17]. Intriguingly, NSL complex appears to be involved in more global transcription regulation as it has been found to bind to a subset of active promoters and contribute to housekeeping gene expression in Drosophila [21]–[23]. It is noteworthy that the NSL complex shares subunits with other chromatin regulating complexes. The MCRS1 (Microspherule Protein 1) protein is a shared subunit between the NSL complex and the Ino80 chromatin remodeling complex [24]. The WDR5 protein is a subunit not only of the NSL complex, but also of the MLL/SET-containing histone H3K4 methyltransferase complexes [25] and of the ATAC histone acetyltransferase complex [26]. The presence of these cross-shared subunits suggests functional links between these complexes and the histone modifications they catalyze.
Coordination between hMOF-mediated histone H4K16 acetylation and other histone modifications has been reported by several research groups. In the response of 293 cells to serum stimulation, the phosphorylation of H3S10 is the trigger for H3K9 and H4K16 acetylation, resulting in transcription activation and elongation [27]. In addition, hMOF-mediated H4K16ac and SUV420-H2-mediated H4K20me3 antagonistically control gene expression by regulating Pol II promoter-proximal pausing [28]. Given that WDR5 is part of the MLL/SET-containing methyltransferases and hMOF/NSL acetyltransferase, the interaction between H3K4me and H4K16/K5/K8ac is speculated. FOXP3, an X-linked suppressor of autoimmune disease and cancers, increases both H4K16 acetylation and H3K4 trimethylation at the FOXP3-associated chromatins of multiple FOXP3-activated genes by recruiting MOF and displacing histone H3K4 demethylase PLU-1 [29]. All the above investigations strongly suggest that the coordination between hMOF-mediated H4K16 acetylation and other histone modifications is involved in certain gene transcriptional activation. However, the precise cooperative mechanism between different histone modifications remains unclear.
In an effort to resolve the coordination role in the activities between hMOF-mediated H4K16 acetylation and MLL/SET-mediated H3K4 methylation, we have carried out systematic biochemical and molecular biological analysis of human MOF/NSL HAT and MLL/SET complexes. As we describe below, our findings demonstrate a new regulatory pathway, that the hMOF/NSL-mediated histone H4K16 acetylation facilitates histone H3K4 di-methylation by MLL/SET complexes. This functional interaction leads to a coordinative regulation of certain downstream target genes, such as ANKRD2.
WDR5 has been shown to be a shared subunit of multi-complexes such as histone acetyltransferases and histone methyltransferases [17], [25]–[26]. To elucidate interactions between the WDR5-containing HAT and HMT, we previously generated a human cell line stably expressing Flag-WDR5, purified Flag-WDR5-associating proteins by anti-Flag immunoaffinity chromatography, and analyzed them by MudPIT mass spectrometry and Western blotting. MudPIT analyses of Flag-WDR5-associating proteins identified previously defined subunits of human MLL/SET histone methyltransferase complexes and hMOF-containing NSL histone acetyltransferase complex [17]. These findings were confirmed by immunoblot using available antibodies. As shown in Figure 1A, anti-Flag eluates from Flag-MSL3L1 (a specific subunit for MSL complex), Flag-PHF20 (a specific subunit for NSL complex) and Flag-WDR5 expressing cells were fractionated by SDS-PAGE and analyzed by western blotting. The results revealed that the subunits of both MLL/SET and hMOF-containing NSL HAT complexes could be detected in anti-Flag agarose eluates from Flag-WDR5 expressing cells, consistent with previous evidence that WDR5 complex is associated with at least two multi-protein complexes: MLL/SET HMT and NSL HAT [17].
To characterize the enzymatic activity(s) that copurified with Flag-WDR5-containing complexes, in vitro HAT and HMT assays were performed. As predicted, Flag-WDR5 is associated with both histone acetyltransferase that can acetylate HeLa cell-derived nucleosomes on histone H4 and histone methytransferase that can support mono-, di-, and tri-methylation of histone H3 on lysine 4 (H3K4me1, H3K4me2, and H3K4me3) in recombinant histone octamers. In contrast, complexes containing Flag-Ash2 (a subunit shared between MLL and SET1-containing complexes) copurified only with HMT activity (Figure 1B). To further investigate the potential interplay between H3K4 methylation and H4 acetylation by Flag-WDR5-containing complexes, we performed combined assays for HAT and HMT, in which reactions contained both the acetyl group donor acetyl CoA (AcCoA) and the methyl group donor S-adenosyl methionine (SAM). Although the HAT activity associated with Flag-WDR5 complex was not affected in the presence of SAM (S-adenosyl methionine, a methyl donor) (Figure 1C), HMT activity was dramatically increased in the presence of AcCoA-dependent manner (Figure 1D).
To determine whether the AcCoA-dependent increase in H3K4 methylation activity associated with Flag-WDR5-containing complexes is mediated by hMOF complex(es), we generated HA-tagged hMOF containing a point mutation, G327E, in a highly conserved residue in the hMOF HAT domain (Figure 2A, top) [30]. Complexes containing wild type or mutant HA-tagged MOF purified using anti-HA agarose immunoaffinity chromatography from 293FRT cell lines stably expressing HA-hMOF wild type (hMOFwt) or HA-hMOF G327E (hMOFmt) and fractionated them by SDS-PAGE. As shown in Figure 2A (left), a similar set of polypeptides could be detected in both hMOFwt and hMOFmt complexes by silver staining. To determine how the G327E mutation affects hMOF activity, HA-hMOFwt and HA-hMOFmt complexes containing equivalent amounts of hMOF (Figure 2A, right) were subjected to HAT assays. The results of these experiments indicated that the HAT activity associated with HA-hMOFmt was dramatically reduced (Figure 2B, lane 5–7) compared to HA-hMOFwt (Figure 2B, lane 2–4). To test the role of hMOF-dependent HAT in enhancement of H3K4 methylation, HMT assays were performed according to experimental procedures shown in Figure 2C. In line with the HAT activity, only complexes containing hMOFwt were able to stimulate MLL/SET-mediated H3K4 methylation (Figure 2D, lane 6–8).
In vitro experimental results clearly show that hMOF-mediated acetylation accounts for the positive regulation of H3K4 methylation by MLL/SET complexes. To clarify whether the hMOF-mediated HATs also affect global H3K4 methylation in human cells, we knocked down or overexpressed hMOF, which, as noted above, is the catalytic subunit of both MSL and NSL complexes [16]–[18]. In RNA interference (RNAi) experiments, HeLa cells were transfected with hMOF siRNA three times at 48-hour intervals. After 48 hours, the level of hMOF mRNA was significantly decreased compared to non-targeting (NT) siRNA control (Figure 2F), while the level of hMOF protein was not significantly decreased after the first round of siRNA transfection and gradually decreased thereafter (Figure 2E, top). Global acetylation of H4K16 in cells completely disappeared after the first round of hMOF siRNA transfection, consistent with previous reports that hMOF, but not Tip60, is the major HAT responsible for H4K16 acetylation [16], [31]. We also observed that global acetylation of histone H4K5 and H4K8 were substantially reduced after the second and third rounds of hMOF knockdown (Figure 2E, bottom), consistent with our evidence that the hMOF-containing NSL complex supports H4K5 and H4K8 acetylation in vitro and suggesting that a hMOF-containing complex is also responsible for acetylation of histone H4K5 and H4K8 in human cells [17]. Notably, we also observed that knockdown of hMOF led to a reduction of global histone H3K4 di-methylation in cells. In contrast to the results of our knockdown experiments, hMOF overexpression promoted the global H4K16, K5, and K8 acetylation as well as H3K4 mono-, di- and tri-methylation (Figure 2G).
It has been reported that hMOF, as a catalytic subunit, forms MSL and NSL, which are two distinct human cellular complexes [16]–[18]. According to the description above, acetylation-methylation interaction occurs in both WDR5- and hMOF-containing complexes. Thus, we speculate that the NSL complex is important in this interaction. To verify this, NSL, MSL and MLL/SET complexes were subjected to a series of in vitro HMT assays. We first measured the HAT activities of two complexes purified from stably expressing Flag-tagged indicated proteins. Consistent with our previous reports [17], the MSL complex acetylated histones specifically at the H4K16 site whereas the NSL complex acetylated histone H4 at K5/K8/K16 sites, suggesting a relaxed specificity (Figure 3A). In a combined enzyme activity assay with core histones or recombinant nucleosomes as substrates, the NSL complex facilitated MLL/SET-meditated H3K4me2 in a dose-dependent manner (Figure 3B, 3C). However, the MSL complex did not obviously enhance H3K4me2 in the same assay (Figure 3B). The experimental procedures for Figure 3B and 3C is depicted in Figure 2C. Considering that NSL complex proteins may affect the subsequent HMT assay, reconstituted nucleosomal arrays were prepared and used in combined enzyme activity assay. A scheme of the assay is shown in Figure 3D (top). After removing AcCoA and HATs, the role of promoting H3K4me2 activity by Flag-Ash2 can still be observed in the lane for which the initial HAT assay was performed with the NSL, but not the MSL complex (Figure 3D, bottom). This suggests that the NSL complex—with respect to acetylation of histone H4 specific lysine—might be important for promoting subsequent di-methylation of histone H3K4 via the Flag-Ash2 complex.
Undoubtedly, the HAT NSL complex contributes to H3K4me2 by MLL/SET complexes in vitro. To investigate whether potential target genes are specifically regulated via NSL-MLL/SET co-operativity, gene expression was measured after knocking down the core component of the NSL or MLL/SET complexes. Specific siRNA used to knockdown corresponding component in the complex led to significant reductions of specific mRNA (Figure 4A, top). Global histone modifications were then detected after specific siRNA knockdown (Figure 4A bottom & 4B). In agreement with data from the in vitro assay, knocking down RbBP5, a protein required for the H3K4 methylation by MLL/SET complexes, significant decreased H3K4me2 but had less effect on cellular H4K16ac, in contrast, disruption of hMOF-mediated H4 acetylation, including H4K16, K5 or K8, by knockdown of hMOF or NSL1 led to various degrees of H3K4 methylation depression. Knocking down hMSL3, a protein required for H4K16ac of the MSL complex, only significantly decreased H4K16ac, but not H4K5ac or H4K8ac (Figure 4A, bottom). In addition, as shown in Figure 4C, knocking down the hMOF or NSL1 gene did not affect the protein levels of subunits of MLL/SET complex. Quantified protein is shown in Figure 4D.
To understand variations in gene transcription among hMOF-mediated and MLL/SET complexes, HeLa cells with specific siRNA knocked down were sent to Phalanx Biotech Group, Inc. for gene expression profile analyses. From DNA microarray analyses using Phalanx Human OneArray (HOA 5.2), a total (>2-fold changes of downregulated and upregulated) of 364, 646, 723 and 843 genes were shown to be differentially expressed among hMOF, NSL1, RbBP5 or MSL3L1 and NT siRNA knockdown HeLa cells, respectively (Figure 4E, top). The shared and distinct gene expression in each population is depicted in a Venn diagram (Figure 4E, bottom). Interestingly, 54 genes were co-regulated by hMOF/RbBP5; 40 genes were co-regulated by hMOF/NSL1/RbBP5; and 30 genes were co-regulated by hMOF/MSL3L1/RbBP5. mRNA from different siRNA knockdown cells was measured with qRT-PCR. As shown in Figure 4F and 4G, compared to internal control actin, selected gene mRNA including ANKRD2, HCP5, UNC13D, ACSL5, FHL1, RHEBL1 and STK3 was downregulated with siRNA knockdown of hMOF, hNSL1, and hRbBP5 in HeLa cells. In contrast, NTS, STRADB and CMBL mRNAs was upregulated. However, no obvious changes were observed for those genes in hMSL3L1 siRNA knockdown cells. Of note, our gene expression profile data of hMOF depleted HeLa cells was similar to that from hMOF-depleted HEK293 cells as reported by Sharma and colleagues [12]. This indicates conserved transcriptional regulation of hMOF among different cell lines.
The previously mentioned, our results suggest that the expression of some genes may require coordinated regulation of the NSL-MLL/SET complexes. Therefore, the ANKRD2 gene was chosen for investigating the role of coordinated regulation of the NSL-MLL/SET complexes in gene transcription. Five primer sets were designed to yield ChIP'd DNA for the ANKRD2 promoter proximal region as well as the region far away from the transcriptional start site (Figure 5A). Analysis of ChIP assays for HeLa cells revealed that the distribution of hMOF was restricted to the ANKRD2 transcriptional start site proximal region (from −0.5 kb to +0.5 kb) (Figure 5B), which was in agreement with H4K16ac distribution reported in downregulated genes in hMOF depleted HEK293 cells [20]. To illustrate the correlation between hMOF/H4K16ac and H3K4me2 on the ANKRD2 transcriptional start site region, ChIP assay was performed using H4K16ac, H3K4me1, H3K4me2 and H3K4me3 antibodies. As shown in Figure 5C, marks of H4K16ac and H3K4 methylation, especially H3K4me2, were enriched −0.25kb upstream of the transcription start site, representing the colocalization of H4K16ac and H3K4me with hMOF on the ANKRD2 promoter region.
The correlation between hMOF/H4K16ac and H3K4me2 on the ANKRD2 promoter prompted us to examine transcriptional regulation of ANKRD2 via cooperation of the NSL-MLL/SET complexes. For this, specific genes including hMOF, NSL1, or RbBP5 were depleted by siRNAs in HeLa cells followed by ChIP assays for H4K16ac and H3K4me2 at the ANKRD2 transcriptional start site-proximal region (−0.25 kb). As shown in Figure 5D, depletion of hMOF significantly reduced both H4K16ac and H3K4me2 around the ANKRD2 transcriptional start site proximal region. Around the FHL1 promoter region (−0.4 kb), depletion of hMOF only decreased H4K16ac marks, but not H3K4me2. Similarly, depletion of NSL1 significantly reduced both H4K16ac and H3K4me2 around the ANKRD2 transcriptional start site proximal region. However, depletion of RbBP5 only reduced H3K4me2 marks, not H4K16ac, in the same region of ANKRD2, suggesting that ANKRD2 gene transcription is co-regulated by NSL-MLL/SET complexes (Figure 5E). In addition, distribution of H3K4me2 at the ANKRD2 promoter proximal region was affected by hMOF or NSL1 siRNA-induced H4K16ac reduction.
In this report we confirmed the existence of functional cooperation between hMOF-containing NSL and MLL/SET complexes using both in vitro and in vivo approaches. HAT-HMT combined in vitro assays present evidence that hMOF-containing NSL complex functions to promote histone H3K4 methylation via MLL/SET complexes. Analysis of gene expression indicates that the expression of some genes is coordinated by the NSL and MLL/SET complexes. In addition, coordination between NSL and MLL/SET complexes is involved in transcriptional regulation of certain genes, such as ANKRD2.
Histone H4K16ac and H3K4me are critical in mammalian cells and function as specific transcription regulators that directly linked to either gene transcription activation or repression [18], [27]–[28]. Although the coordinated activity of H4K16ac and H3K4me has been observed in transcription regulation of certain genes, such as HOX and FOXP3-activated genes [29], the precise crosstalk mechanism remains unclear. Based on in vitro HAT and HMT assays with distinct enzymatic complexes purified from stably expressed Flag-tagged proteins, we defined a positive regulation of MLL/SET-mediated H3K4 methylation by Flag-WDR5 complex in AcCoA-dependent manner. Further investigation revealed that enhanced activity of H3K4me by MLL/SET complexes was indeed due to the hMOF-containing complex; no similar enhancement was observed with induction of a G327 point mutation in hMOF. To support this observation, reduction of global H3K4me in cells was confirmed by knocking down hMOF or NSL1 with siRNAs, suggesting that the hMOF/NSL complex may be involved in cellular H3K4 methylation regulation.
Although both the NSL and the MSL HAT complexes contain hMOF as a catalytic subunit, assembly of hMOF HAT into the MSL or NSL complex controls its substrate specificity and transcription regulation. NSL-associated hMOF has less specificity for nucleosomal histone H4 [17] and appears to be involved in more global regulation of transcription [21]–[23]. In our RNAi experiments, knocking down hMOF or NSL1 not only decreased H4K16ac, but also reduced cellular H4K5ac and H4K8ac (Figure 2E, 4A & 4B). In contrast, in hMSL3 knockdown HeLa cells, no decline in H4K5ac or H4K8ac was observed, except for the reduction of H4K16ac. Our data show that the NSL complex, but not the MSL complex, plays a role in promoting histone H3K4me2 activity by MLL/SET complexes (Figure 3B–D), indicating the functional differences between the two complexes. An attractive hypothesis is that acetylation of histone H4 on K5, K8, and/or K16 might be responsible for this effect; however, we cannot rule out the possibility that there were small/trace amounts of residue NSL complex remaining to acetylate some component(s) of the MLL/SET complex. However, knocking down either NSL1 (a core component of the NSL complex) or MSL3 (a core component of the MSL complex) to disrupt the NSL or MSL complex, respectively [32]–[33], suppressed global activity of both cellular H4K16ac and H3K4me. These data indicate that more a complicated regulatory pathway may be involved in the MSL complex in the cellular environment. Recent studies identified that the MSL2 protein in the MOF complex is an E3 ubiquitin ligase for H2BK34 involved in crosstalk with H3K4 and K79 methylation [34]. In addition, when using recombinant nucleosomes, a chromatin structural form more close to native status than to histone octamers, only MLL/SET-mediated H3K4me2 can be detected in vitro (Figure 2D, 3C & 3D). In contrast, knockdown or overexpression of cellular hMOF reveals alterations in all methylation scales (Figure 2E, 2G & 4A). Although only one H3K4 methyltransferase exists in yeast to catalyze H3K4 mono-, di-, or tri- methylation, mammalian cells contain at least 10 H3K4 methyltransferases with various specificities in vivo [35]–[36]. Thus H3K4 methylation is more complicated in vivo than in vitro. So, MLL/SET- mediated H3K4me2 detected in vitro may be more reliable and probably acts as a node of the NSL-MLL/SET cooperative network. However, we cannot exclude the possibility of differences in antibody titers.
Gene expression studies combined with siRNA knockdown data provides rich information about gene transcription regulated by the NSL, MLL/SET and MSL complexes. First, after specific siRNA knockdown, in a core component of the above mentioned three complexes, transcriptions of 2576 genes were found to be altered more than two-fold. Among these genes, 513 genes overlapped in 2 or more knockdown samples, suggesting a strong correlation between histone modifications and gene expression regulation. Next, 160 genes were co-regulated by hMOF and RbBP5 knockdown, suggesting collaborative roles between hMOF-mediated HAT activity and MLL/SET-mediated HMT activity. Finally, among the 160 genes co-regulated by the hMOF and MLL/SET complexes, 40 genes are affected by NSL1 only, whereas 30 genes are influenced by MSL3L1 only. This suggests distinct functions of the NSL and MSL complexes in hMOF-MLL/SET coordination.
It was previously reported that hMOF is essential for embryonic development [11], [37]. It is worth noting that our gene expression profile analyses identified genes co-regulated by hMOF/NSL1/RbBP5 that were related to tissue and organ formation and occurrence of various diseases. ANKRD2, a muscle ankyrin repeat protein, is important in muscle gene transcription regulation, myofibrillar assembly, cardiogenesis, and myogenesis. Abnormal expression of the ANKRD2 gene leads to neuromuscular disorders, cardiovascular diseases, and even cancer [38]. ChIP assay with available antibodies revealed detailed collaborative mechanism of hMOF/NSL1/RbBP5 on ANKRD2 gene expression regulation. Distribution of hMOF, H4K16ac and H3K4me2 were restricted to the ANKRD2 transcriptional start site proximal region. H4K16ac and H3K4me2 are mediated by hMOF/NSL HAT and MLL/SET HMT: knockdown of the core subunit of these complexes significantly blocks corresponding histone modifications (Figure 5D & 5E). Notably, hMOF/NSL-mediated H4K16ac promotes MLL/SET-mediated H3K4me2, a finding that is in agreement with data obtained from in vitro enzyme assays (Figure 5E). In summary, coordination between hMOF/NSL-mediated H4K16ac and MLL/SET-mediated H3K4me2 is involved in ANKRD2 gene activation. Consistent with our data, we noted that NSL-bound genes exhibit elevated H3K4me2/3, H3K9ac and H4K16ac according to genome-wild analyses in Drosophila [23]. This may be another hMOF/NSL-MLL/SET correlation instance that is similar to our findings in mammalian cells.
In our study, the regulatory effect of hMOF/NSL HAT to MLL/SET HMT activity seemed to be unidirectional: no obvious reverse effects from methylation to acetylation were detected in either in vitro assays or endocellular circumstances. These data indicate that hMOF/NSL-dependent acetylation event(s) regulate MLL/SET activity; thus, the pathway is unidirectional. However, some questions persist about this cooperation: 1) Does hMOF/NSL-mediated acetylation of histone H4 induce alteration of chromatin structure and facilitate the subsequent methylation process? 2) Does the NSL complex help with recruiting and assembling of MLL/SET HMT complexes at upstream regions of target genes? In addition, it is important to determine which components of the NSL complex play synergistic roles in the regulation of H3K4 methylation via MLL/SET complexes. Additionally, MLL is a proto-oncogene that is rearranged in a wide variety of human leukemias, so future studies are warranted to investigate whether the NSL complex is implicated in MLL-rearranged leukemia.
Anti-H4K16ac (H9164) antibody was obtained from Sigma (U.S.A.). Anti-H3K4me1 (07-436), anti-H3K4me2 (07-030) and anti-H3K4me3 (07-473), anti-H3 (06-755), anti-H4K5ac (07-327), anti-H4K8ac (07-328) and anti-WDR5 (07-706) rabbit polyclonal antibodies were from Millipore (U.S.A.). Anti-OGT1 (sc-32921) and anti-MLL1 (sc-18214) antibodies and rabbit total IgG (sc-2027) were purchased from Santa Cruz Biotechnology (U.S.A.). Anti-MCRS1, anti-hMOF (residue 210–468 aa), anti-RbBP5, anti-Ash2, anti-MSL3L1 and anti-GAPDH rabbit polyclonal antibodies were raised against bacterially expressed proteins (Jilin University). Anti-MSL2L1 (H00055167A01) was from Novus Biologicals. Anti-MSL1 antiserum was kindly provided by Dr. Edwin R. Smith (Stowers Institute for Medical Research).
Full-legth cDNAs encoding the human WDR5 (NM_017588), NSL2 (FLJ20436; BC009746), NSL3 (FLJ10081; BC063792), PHF20 (NM_016436), hMSL3 (male-specific lethal 3-like 1; BC031210) and hAsh2 ((absent, small, or homeotic)-like; NM_004674) proteins were obtained from the American Type Culture Collection (ATCC), subcloned with FLAG tags into pcDNA5/FRT, and introduced into HEK293/FRT cells using the Invitrogen Flp-in system [17]. hMOF cDNAs (wild type or mutant G327E) were subcloned with HA-tag into pQC and introduced into 293FRT cells. Stably transformed HEK293/FRT cells were maintained in Dulbecco's modified Eagle's medium (Sigma) with 5% glucose and 10% fetal bovine serum.
Nuclear extracts were prepared from HEK293/FRT cells according to the method of Dignam et al. [39]. Flag-tagged or HA-tagged proteins and their associated proteins were purified on anti-Flag (M2) or anti-HA agarose beads as previously described [40]. Identification of proteins was accomplished using a modification of the MudPIT (multidimensional protein identification technology) procedure described by Washburn et al. [41].
Recombinant histone octamers and polynucleosomes (used ∼1500 bp DNA fragment) were prepared as previously described [40], [17].
Histone acetyltransferase assays were performed essentially as described [17].
Reaction mixtures (32 µl) containing 50 mM Tris-HCl (pH 8.5), 20 mM KCl, 10 mM MgCl2, 250 mM sucrose, 10 mM β-mercaptoethanol, 1 mM protease inhibitor cocktail (Sigma), 125 µM S-adnosyl methionine (SAM, Sigma), 0.5 µg E.coli expressed and purified core histones, or 2 µg of long oligo-nucleosomes or reconstituted polynucleosomes, and anti-Flag or HA-agarose eluates prepared from HEK293/FRT cells stably expressing Flag-tagged or HA-tagged proteins were incubated at 30°C for 16–18 hours. Reactions were stopped by addition of 4×SDS sample buffer and was then fractionated on 18% SDS-PAGE and subjected to Western blotting using methylation-specific antibodies to detect modified histone H3 residues.
HeLa cells were cultured in 6-well tissue culture plates (∼2×105 cells/well) in DMEM medium (Sigma) containing 10% fetal bovine serum. The cells were transfected with 20 nM non-targeting siRNA (D-001206), MYST1 siRNA (D-014800), KIAA1267 siRNA (D-031748), MSL3L1 siRNA (D-012319) and RbBP5 siRNA (D-012008) SMART pool (Dharmacon, U.S.A.). Then, 24 hours after transfection, cells were divided into new 6-well plates for immunoblotting, RT-PCR and DNA microarray analysis. Finally, 24 hours later, cells were harvested and lysed. Whole-cell extracts were prepared from cells from 3/4 of the wells in a 6-well plate by adding 4× SDS sample buffer, and total RNA was isolated from 1/4 of the wells in a 6-well plate using TRIzol LS Reagent (Invitrogen). In addition, cells from 1 well of a 6-well plate were rinsed twice with warm PBS and harvested. Cells were then stored in an RNA hold solution (ER501-01, Beijing Transgen Biotech Co., Ltd.) and sent to OneArray by Phalanx Biotech Group for DNA microarray analysis.
Human embryonic kidney (HEK) 293T cells were cultured in 6-well tissue culture plates (∼2×105 cells/well) in DMEM containing 10% fetal bovine serum and antibiotics. The cells were transiently transfected with 2 µg of hMOF cDNAs using Lipofetamine 2000 (Invitrogen). At 48 hrs post-transfection, cells were harvested and lysed by adding 4×SDS sample buffer. Whole-cell extracts were analyzed by Western blotting with the indicated antibodies.
Cells from 1 well of a 6-well plate were lysed and total RNA was isolated using Trizol. Total RNA (1 µg) from each sample was used as a template to produce cDNA with PrimeScript 1st Strand cDNA Synthesis Kit (TAKARA). MYST1, NSL1, RbBP5, MSL3L1 and GAPDH mRNA was measured by quantitative real time PCR with Real Time PCR Detector Chromo 4 (Bio-Rad). All PCR reactions were finished under the following program: initial denaturation step was 95°C for 3 min, followed by 35 cycles of denaturation at 95°C for 30 seconds, annealing at 60°C for 30 seconds and extension at 72°C for 30 seconds. The following qRT-PCR primer sets were used to verify the siRNA knockdown efficiency: hMOF, 5′-GGCTGGACGAGTGGGTAGACAA-3′ (forward) and 5′-TGGTGATCGCCTCATGCTCCTT-3′ (reverse), yielding a 227 bp product; hNSL1, 5′-CTTATTGCTGCCAACGGAACCA-3′ (forward) and 5′-AGGACTGTCTGCTTGCTGAAGA-3′ (reverse), yielding a 196 bp product; hMSL3L1, 5′-CAGGACACATCCGCCAGCAT-3′ (forword) and 5′-AAAGCCAGCAAACACAGCACTC-3′ (reverse), yielding a 128 bp product; hRbBP5, 5′-ATGAACCTCGAGTTGCTGGA-3′ (forword) and 5′-CACTGGATGGATGTGTGCAC-3′ (reverse), yielding a 207 bp product. The primer sets used for qPCR to verify hMOF or NSL1 or RbBP5-regulated genes were as follows: ANKRD2, 5′-TGGCACAGGAGGAGGAGAATGA-3′ (forword) and 5′-CTTCCGCAGCTCGATGAGGTTC-3′ (reverse), yielding a 215 bp product; HCP5, 5′-GACTCTCCTACTGGTGCTTGGT-3 (Forward) and 5′-CACTGCCTGGTGAGCCTGTT-3′ (reverse), yielding a 241 bp product; UNC13D, 5′-GCTGCCACCGTCCTCTTTCT-3′(forword) and 5′-CTCCTCCTGCTGTTCTGCCTTG-3′ (reverse), yielding a 192 bp product; ACSL5, 5′-GCGTCATCTGCTTCACCAGTG-3′ (forword) and 5′-CGTCAGCCAGCAACCGAATATC-3′ (reverse), yielding a 249 bp product; FHL1, 5′-ACTGCGTGACTTGCCATGAGAC-3′ (forword) and 5′-TGGTCCTCCACAGCGGTGAA-3′ (reverse), yielding a 172 bp product; RHEBL1, 5′-CGGGTGCCAGTGGTTCTAGT-3′ (forword) and 5′-CGACGCTCTTGCCCATAGGAA-3′ (reverse), yielding a 206 bp product; STK3, 5′-ATGCGGGCCACAAGCACGAT-3′ (forword) and 5′-TCACCATGGTCCCCAAGTCGGA-3′ (reverse), yielding a 91 bp product; NTS, 5′-GCTCCTGGAGTCTGTGCTCA-3′ (forword) and 5′-CCTTCTTGCAACAAGCTCCTCT-3′ (reverse), yielding a 209 bp product; STRADB, 5′-TGGAGCCGTGAGAGGGTTGA-3′ (forword) and 5′-ACTGATGTGCTGAACTGTGGGA-3′ (reverse), yielding a 189 bp product; CMBL, 5′-GCTAGGCCGTGAAGTTCAAGTC-3′ (forword) and 5′-AAGATAGACCAGTCGCCAGAGG-3′ (reverse), yielding a 222 bp product. The primer sets used for internal control β-Actin were as follows: 5′-ATGGGTCAGAAGGATTCCTATGT-3′ (forword) and 5′-AGCCACACGCAGCTCATT-3′ (reverse), yielding a 153 bp product.
One or two 10 cm dishes (∼1×107) of HeLa cells grown to ∼80% confluence were used for each ChIP assay. Cells were cross-linked with 5 ml 1% formaldehyde in PBS for 15 min at room temperature followed by incubation with 125 mM glycine for 5 min. To shear DNA to lengths ranging between 200–1000 base pairs, cell lysates were sonicated with a SCIENTZ-IID (Ningbo Xinzhi Biotechnology Co., Ltd., China) for 5×60 seconds with every second interval, at a setting of 45% duty, level 2. Equal amounts of sonicated chromatin from each sample were incubated at 4°C overnight with 5–10 µg of antibodies against H3K4me1, H3K4me2, H3K4me3, H4K16ac, or hMOF-1 rabbit polyclonal antibodies. Total rabbit IgG and pre-immune serum were used as control. The next day, 50 µl of protein A agarose containing salmon sperm DNA (10 µg) and BSA (25 µg) (50% slurry) were added, and the mixture was further incubated for 2.5 hours at 4°C to collect the agarose/antibody/protein complexes. The protein A agarose/antibody/protein complexes were washed for 5 min on a rotating platform with 1 mL buffer in the following order: Low Salt (buffer containing 150 mM NaCl, 0.1% SDS, 1% TritonX-100, 2 mM EDTA and 20 mM Tris, pH8)–High Salt (buffer containing 500 mM NaCl, 0.1% SDS, 1% TritonX-100, 2 mM EDTA and 20 mM Tris, pH8)–250 mM LiCl (buffer containing 0.25 M LiCl, 1% NP-40, 1% NaDOC, 1 mM EDTA and 10 mM Tris, pH8)–TE (buffer containing 10 mM Tris, pH8 and 1 mM EDTA)–TE. Finally the washed beads were eluted with 480 µl elution buffer containing 0.1 M NaHCO3 and 1% SDS. DNA was extracted with phenol/chloroform and precipitated by ethanol. Next, 1–2 µl immunoprecipitated or 100-fold diluted input DNA was amplified with a Real Time PCR Detector Chromo 4 (Bio-Rad). Each experiment was performed 2–3 times independently. All ChIP signals were normalized to total input. The primer sets for qPCR on the promoter region of ANKRD2 were as follows: ANKRD2 −0.5 kb (−262∼−88), 5′- GCAGTTCCCTAGCAGATTAACCT-3′ (forward) and 5′-GCCCAGACAGTGCCAGACTT-3′ (reverse); −0.25 kb, 5′-CTTAACGGGGAAGCATGTGG-3′ (forward) and 5′-GACAGTTCTGTACTCCCAGGCTG-3′ (reverse); +0.5 kb, 5′-GGAGGAGGAGAATGAGGTGC-3′ (forward) and 5′-ACCCCCTGCCAGTAATACCC-3′ (reverse); +1.9 kb, 5′-GTAAGCCGAGATCGCACCAC-3′ (forward) and 5′-AACTTCAGCTCCTGCATTTCC-3′ (reverse); +5.2 kb, 5′-CTGGTGGCCTTTAATGTTGTT-3′ (forward) and 5′-GGTGGTCTCAGAGCCCTTCT-3′ (reverse); ; FHL1 promoter (−0.4 kb), 5′-CGGCTTGCTACTAAGGGGAGG-3′ (forward) and 5′-GCAACAAAGACAGCCAAGTGAGG-3′ (reverse).
|
10.1371/journal.ppat.1003312 | Development of a Highly Protective Combination Monoclonal Antibody Therapy against Chikungunya Virus | Chikungunya virus (CHIKV) is a mosquito-transmitted alphavirus that causes global epidemics of a debilitating polyarthritis in humans. As there is a pressing need for the development of therapeutic agents, we screened 230 new mouse anti-CHIKV monoclonal antibodies (MAbs) for their ability to inhibit infection of all three CHIKV genotypes. Four of 36 neutralizing MAbs (CHK-102, CHK-152, CHK-166, and CHK-263) provided complete protection against lethality as prophylaxis in highly susceptible immunocompromised mice lacking the type I IFN receptor (Ifnar−/−) and mapped to distinct epitopes on the E1 and E2 structural proteins. CHK-152, the most protective MAb, was humanized, shown to block viral fusion, and require Fc effector function for optimal activity in vivo. In post-exposure therapeutic trials, administration of a single dose of a combination of two neutralizing MAbs (CHK-102+CHK-152 or CHK-166+CHK-152) limited the development of resistance and protected immunocompromised mice against disease when given 24 to 36 hours before CHIKV-induced death. Selected pairs of highly neutralizing MAbs may be a promising treatment option for CHIKV in humans.
| Chikungunya virus (CHIKV) is a mosquito-transmitted alphavirus that causes outbreaks of polyarthritis in humans, and is currently a threat to spread to the United States due to the presence of its mosquito vector, Aedes albopictus. At present, there is no licensed human vaccine or therapeutic available to protect against CHIKV infection. The primary goal of this study was to develop an antibody-based therapeutic agent against CHIKV. To do this, we developed a panel of 230 new mouse anti-CHIKV MAbs and tested them for their ability to neutralize infection of different CHIKV strains in cell culture. We identified 36 MAbs with broad neutralizing activity, and then tested several of these for their ability to protect immunocompromised Ifnar−/− mice against lethal CHIKV infection. In post-exposure therapeutic trials, administration of a single dose of a combination of two neutralizing MAbs limited the development of resistance and protected Ifnar−/− mice against disease even when given just 24 to 36 hours before CHIKV-induced death. Analogous protection against CHIKV-induced arthritis was seen in a disease model in wild type mice. Our data suggest that pairs of highly neutralizing MAbs may be a therapeutic option against CHIKV infection.
| Chikungunya virus (CHIKV) infection causes a severe febrile illness in humans that is characterized by a debilitating polyarthritis, which can persist for months and cause significant morbidity [1], [2]. There are three genotypes of CHIKV: Asian, East/Central/South African (ECSA), and West African [3]–[5], with 95.2 to 99.8% amino acid identity [4]. The CHIKV strains from the recent epidemics belong to the ECSA genotype and have affected millions in Africa and the Indian subcontinent [3], [6]. Imported cases in the United States and outbreaks in Europe highlight the threat of CHIKV to developed countries [7]. Currently, there are no approved vaccines or therapeutics for CHIKV [8].
CHIKV is an enveloped alphavirus of the Togaviridae family that enters cells via receptor-mediated internalization and a low pH-triggered type II membrane fusion event in early endosomes. The mature virion is comprised of three structural proteins: a nucleocapsid protein and two glycoproteins, E1 and E2, where E2 functions in attachment to cells and E1 participates in virus fusion. Each 700 Å CHIKV virion contains 240 copies of the envelope and capsid proteins, which are arranged in T = 4 quasi-icosahedral symmetry. E1-E2 heterodimers assemble into 80 trimeric spikes on the virus surface [9]. X-ray crystallographic structures of the precursor pE3-E2-E1, mature E2-E1, and E1 proteins [10]–[13] have elucidated the architecture of the glycoprotein shell. The E1 ectodomain consists of three domains. Domain I (DI) is located between DII and DIII, the latter of which adopts an immunoglobulin-like fold. The fusion peptide is located at the distal end of DII. E1 monomers lie at the base of the surface spikes and form a trimer around each of the icosahedral axes. E2 localizes to a long, thin leaf-like structure on the top of the spike. The mature E2 protein contains three domains with immunoglobulin-like folds: the N-terminal domain A, located at the center; domain B at the tip; and the C-terminal domain C, located proximal to the viral membrane.
Mouse models have been developed for CHIKV infection. Newborn outbred and inbred mice are vulnerable to severe CHIKV infection with viral replication observed in muscle, joint, and skin [14], [15]. Adult mice with defects in type I interferon signaling (Ifnar−/− mice) develop lethal disease, with muscle, joint, and skin appearing as the primary sites of infection [15]. CHIKV infection of juvenile C57BL/6 mice by a subcutaneous route results in metatarsal foot swelling with histological evidence of arthritis, tenosynovitis and myositis [16], [17].
Passive transfer of MAbs or immune sera can protect animals against infection of alphaviruses including Sindbis (SINV), Semliki Forest (SFV), and Venezuelan equine encephalitis (VEEV) viruses [18]–[25]. Immune γ-globulin from human donors in the convalescent phase of CHIKV infection exhibited neutralizing activity in vitro and had partial therapeutic efficacy in Ifnar−/− and neonatal wild type mice when administered up to 24 hours after infection [26]. Although mouse and human MAbs that neutralize CHIKV infection have been reported [27], [28], their post-exposure efficacy against lethal infection in vivo has not been clearly established [29].
Here, we investigated the molecular basis of antibody-mediated neutralization of CHIKV using a panel of 230 newly generated, cloned MAbs. CHK-152 protected mice against CHIKV-induced mortality and disease. The inclusion of a second MAb (CHK-166 or CHK-102) prevented the emergence of viral resistance and extended the treatment window in Ifnar−/− mice up to 24 to 36 hours prior to death of the animals. Our results suggest that combination therapy with selected neutralizing MAbs has potential for treatment of CHIKV infection in humans.
We generated a panel of neutralizing MAbs against CHIKV as a first step towards a possible therapy in humans. We infected adult C57BL/6 mice deficient for interferon regulatory factor 7 (Irf7−/−) with 104 PFU of the La Reunion 2006 OPY-1 strain of CHIKV (CHIKV-LR); these mice were boosted with CHIK virus-like particles [30], soluble recombinant CHIKV E2 protein, or live CHIKV-LR. We immunized Irf7−/− rather than wild type (WT) mice, as CHIKV replicated to higher titers, induced stronger neutralizing antibody responses, yet did not cause lethal infection in these innate immune-deficient animals ([31], and data not shown). We screened four independent myeloma cell-splenocyte fusions for binding of hybridoma supernatants to CHIKV-LR infected cells (Fig. S1) and cloned 230 CHIKV-specific MAbs for further analysis (Table S1 in Text S1). Using a single endpoint neutralization assay, we identified 36 MAbs with inhibitory activity against infection of CHIKV-LR in BHK21-15 cells (data not shown).
To assess the inhibitory potential of our anti-CHIKV MAbs against the homologous CHIKV-LR and representative strains from the Asian and West African genotypes (RSU1 and IbH35 respectively), we performed focus reduction neutralization tests (FRNTs) on Vero cells. We determined the concentration of MAb that reduced the number of foci of infection by 50 or 90% (EC50 and EC90 values, Fig. 1A and B, and Table 1). CHK-152 was the most strongly neutralizing MAb we identified; 3 and 15 ng/ml of this MAb prevented 50 and 90% of CHIKV infection against all three CHIKV genotypes (Fig. 1C). Ten other MAbs inhibited CHIKV infection with EC50 values of <10 ng/ml against all three genotypes, and many others inhibited all three strains similarly, with a few exceptions. For example, CHK-9 failed to neutralize the Asian strain to the same extent as the West African or La Reunion (ECSA genotype) strains (Fig. 1D), whereas CHK-151 inhibited infection of the Asian strain better than the others (Table 1). Also, for reasons that are unclear, some neutralizing MAbs (e.g., CHK-143, CHK-264, and CHK-269) were incapable of inhibiting all viruses (EC90>10,000 ng/ml) in this assay, even at high MAb concentrations.
We speculated that some MAbs might show cell type-dependent neutralization if they blocked attachment to cell type-specific factors. To test this hypothesis, we assessed MAb neutralization of CHIKV-LR infection in cells of another species, NIH 3T3 mouse fibroblasts (Table 1). For most MAbs, the EC50 values were comparable to those achieved with Vero cells. However, two MAbs (CHK-96 and CHK-176) showed a 12 to 250-fold reduction (P<0.05) in neutralizing activity on NIH 3T3 compared to Vero cells; although further study is warranted, these MAbs may block a step in the entry pathway that varies among different cell types.
To evaluate whether neutralizing MAbs protect against CHIKV infection in vivo, we initially used a stringent test model: prevention of lethal infection in immunodeficient Ifnar−/− C57BL/6 mice. One hundred micrograms of 14 different MAbs with strong, modest, or poor neutralizing activity were administered to Ifnar−/− mice one day prior to CHIKV-LR infection. As seen previously [15], all Ifnar−/− mice died by day 4 after infection when treated with saline or a negative control MAb (Fig. 2A, and data not shown). Strongly neutralizing (e.g., CHK-102, CHK-152, and CHK-263) and one moderately inhibitory (CHK-166) MAb protected 100% of mice from lethal infection (P<0.0001). In comparison, and somewhat surprisingly, CHK-95, a potently neutralizing MAb of the same IgG2c isotype, protected only 12% of mice from death. The other MAbs tested conferred intermediate levels of protection (Fig. 2A). Thus, although several strongly neutralizing MAbs prevented against lethal CHIKV infection in Ifnar−/− mice, in vitro neutralization activity per se did not directly correlate with protection. To define the relative potency of the four MAbs that completely prevented lethal disease, we administered a lower (10 µg) dose. Whereas CHK-152 and CHK-263 still protected most mice from lethal infection, CHK-102 and CHK-166 protected to a lesser degree or only prolonged survival (Fig. 2B). Consistent with their ability to protect against lethal infection, passive transfer of CHK-102, CHK-152, CHK-166, and CHK-263 MAbs all markedly reduced viral loads in serum, spleen, liver, muscle, and brain at 48 hours after infection relative to a non-binding isotype control (DENV1-E98) MAb (Fig. 2C–G). The level of protection afforded by CHK-102, CHK-152, CHK-166, and CHK-263 MAbs, however, did not correlate directly with their binding strength to CHIKV surface glycoproteins (Fig. S2).
Although a stringent test of MAb protection, CHIKV-infected Ifnar−/− mice do not develop the arthritis observed in humans. To evaluate this, we utilized a WT C57BL/6 mouse model in which inoculation of CHIKV into the footpad results in localized swelling and induction of arthritis and fasciitis within the foot and ankle [16], [17], although infection does not cause lethality. Pretreatment of mice with either 100 µg of CHK-102 or CHK-152 completely protected against CHIKV-induced swelling, compared to control animals, which developed clinically apparent swelling (data not shown). While CHIKV infected control animals developed inflammatory arthritis in the ankle and foot, CHK-102 or CHK-152 MAb treated animals had normal appearing joint tissues (Fig. 2H).
Antibody neutralization of enveloped viruses can occur by inhibiting attachment, internalization, and/or fusion [32], [33]. To determine how many of our most protective MAbs inhibited infection in cell culture, we performed pre- and post-attachment neutralization assays [34], [35]. Anti-CHK MAbs were incubated with CHIKV before or after virus binding to cells, and infection was measured. As expected, all MAbs efficiently neutralized infection when pre-mixed with virus (Fig. 3A). While CHK-102, CHK-152, CHK-166, and CHK-263 also inhibited CHIKV infection when added after virus adsorption to the cell surface, suggesting that at least part of their blocking activity was at a post-attachment step, differences in the extent of neutralization were noted in this context for several MAbs. CHK-152 completely neutralized all CHIKV virions without a resistant fraction when added post-attachment. When studies were repeated with eight other neutralizing MAbs that showed pre-exposure protection in vivo, no other MAb inhibited infection completely when added after virus adsorption to the cell. As expected, an isotype control MAb (DENV1-E98) and a non-neutralizing anti-CHK MAb (CHK-84) had no inhibitory effects in this assay (Fig. S3).
Since CHK-152 neutralized infection efficiently at a post-attachment step, we investigated whether it blocked fusion using a viral fusion from without (FFWO) assay [36]. CHIKV was adsorbed to Vero cell monolayers on ice and then treated with MAbs. Fusion at the plasma membrane was triggered after a brief exposure to low pH buffered medium at 37°C. Subsequently, cells were incubated in the presence of 20 mM NH4Cl to prevent CHIKV fusion via canonical endosomal pathways. As expected, at 14 hours after initial treatment, CHIKV infection was not observed when adsorbed virus was incubated at neutral pH (Fig. 3B). In comparison, in the absence of MAb or in the presence of a control MAb, a short exposure of cell surface-adsorbed virus to acidic pH resulted in infection and CHIKV-antigen positive cells. Notably, CHK-152 completely inhibited (P<0.0001) plasma membrane fusion and infection, whereas other anti-CHIKV neutralizing MAbs showed significant yet incomplete inhibition in this assay (Fig. 3B and C). These studies suggest that CHK-152 efficiently neutralizes infection by preventing the structural changes on the virion necessary for viral fusion with host cell membranes.
We utilized a model liposome fusion assay with pyrene-labeled virus [37], [38] to confirm these results. Pyrene-labeled CHIKV was pre-incubated with different concentrations of MAb, mixed with liposomes at 37°C, and fusion was triggered by addition of a low-pH buffer [37]. In the absence of MAb or in the presence of 10 nM (1.5 µg/ml) of a non-binding control MAb, fusion was complete within seconds of acidification. In contrast, pre-incubation of virus with increasing doses of CHK-152 inhibited fusion (Fig. 3D and E). Thus, CHK-152 can block low-pH-induced fusion of virus with liposomes.
To define additional mechanisms by which our most strongly protective MAb (CHK-152) conferred protection in vivo, we generated a chimeric mouse-human CHK-152 (ch-CHK-152) as well as an aglycosyl variant (ch-CHK-152 N297Q) that lacks the ability to engage C1q or Fc-γ receptors; this mutation does not affect the ability to bind the neonatal Fc receptor (FcRn) or half-life of antibody in mouse serum [39]. The affinity of ch-CHK-152 and ch-CHK-152 N297Q binding to purified pE2-E1 was measured by surface plasmon resonance (SPR) and compared to the parent murine MAb. Notably, ch-CHK-152, ch-CHK-152 N297Q, and the murine CHK-152 all had similar affinity (KD of 3 to 4 nM) (Fig. 4A and data not shown) and neutralizing activity in cell culture (Fig. 4B). As expected, ch-CHK-152 N297Q failed to bind efficiently to soluble Fc-γ receptors or C1q (Fig. 4C).
We transferred ch-CHK-152 and ch-CHK152 N297Q to Ifnar−/− mice prior to infection. Although high doses (100 µg) of ch-CHK-152 and ch-CHK-152 N297Q provided similar protection against CHIKV infection (data not shown), lower doses (10 µg) of the aglycosyl variant were less protective; whereas 62% of the mice receiving ch-CHK152 N297Q survived, all Ifnar−/− mice given ch-CHK-152 MAb remained alive (Fig. 4D, P<0.05). When parallel studies were performed with WT C57BL/6 mice and MAb was administered 18 hours after infection, ch-CHK-152 N297Q also provided less protection against arthritis compared to ch-CHK-152 (Fig. 4E). These data suggest that the Fc effector interactions contribute to the potency of CHK-152 in mice.
We humanized CHK-152 as a first step towards a MAb therapeutic (see Text S1). The affinity for pE2-E1 and neutralizing activity of the hu-CHK-152 were similar to mouse CHK-152 (Fig. S4A and B). Hu-CHK-152 also protected Ifnar−/− mice (P>0.0001) when a single dose (10 or 100 µg) was administered one day before infection (Fig. S4C).
To define the therapeutic potential of our most protective MAbs, a single dose (100 µg) was administered to Ifnar−/− mice 24 hours after CHIKV infection (Fig. 5A). Whereas CHK-152 and 166 protected 58% and 63% of mice from death, respectively (P<0.0001), CHK-263 and CHK-102 had less activity although both MAbs increased the median survival time (7 days versus 4 days with the control DENV1-E98 MAb, P<0.0006). Administration of CHK-152 at 12 or 18 hours post infection also protected WT mice from CHIKV-induced swelling and arthritis (Fig. 5B and Fig. 4E).
We next tested the activity of combinations of the most protective neutralizing MAbs in Ifnar−/− mice. Remarkably, administration of 50 µg each (100 µg total dose) of CHK-102+CHK-152, CHK-263+CHK-152, or CHK-166+CHK-152 at 24 hours post infection completely prevented mortality in all animals (Fig. 5A, P<0.0001 for MAb combinations). This observation was not true for all MAb combinations, as administration of 50 µg each of CHK-102+CHK-263 provided substantially less protection with a 14% survival rate. We then performed a more stringent test in which 100 µg each (200 µg total) of our most protective combinations was delivered as a single dose at 48 hours post-infection (Fig. 5C). Treatment with CHK-102+CHK-152 or CHK-166+CHK-152 protected 62% of the Ifnar−/− mice (P<0.003) and the combination of CHK-263+CHK-152 functioned almost as well, with 50% of animals surviving (P<0.03). To define the limits of protection in Ifnar−/− mice, which all succumb to CHIKV between days 3 and 4, therapy was initiated at 60 and 72 hours after infection. At 60 hours after infection, Ifnar−/− mice receiving 250 µg each of CHK-102+CHK-152 or CHK-166+CHK-152 had survival rates of 28 and 71%, respectively (Fig. 5D, P = 0.03 and P = 0.004). Nonetheless, when combination therapy was given at 72 hours after infection, a time when overt disease was present, no survival benefit was conferred. Thus, combination MAb therapy is superior to monotherapy in protecting against lethal CHIKV infection in highly immunocompromised mice.
To begin to understand the basis for enhanced in vivo activity, we assessed whether CHK-152 and selected MAbs could bind simultaneously to the CHIKV virion. We developed a competition ELISA in which virions were captured by a mouse MAb (CHK-65), and then incubated with increasing concentrations of CHK-102, CHK-152, CHK-166, or CHK-263 mouse MAbs. After washing, hu-CHK-152 MAb was added, and binding was assessed. While pre-bound mouse CHK-152 competed against hu-CHK-152 binding as expected, CHK-102, CHK-166, and CHK-263 minimally competed hu-CHK-152 binding (Fig. S5A), suggesting their epitopes largely were distinct. However, addition of CHK-102, CHK-166, or CHK-263 failed to augment the inhibitory activity of CHK-152 when neutralization was measured in cell culture (Fig. S5B), as no synergy was observed.
To identify epitopes targeted by the therapeutic MAbs, we generated escape mutants in cell culture. After sequential virus passage under CHK-102, CHK-152, CHK-166, or CHK-263 selection, CHIKV became resistant to neutralization by these MAbs (Fig. 6A–D). We assessed whether the escape variants generated in the presence of one MAb remained sensitive to neutralization by the other MAbs. The CHK-152 escape variant was neutralized efficiently by CHK-102, CHK-166, and CHK-263 (Fig. 6B, Table S2 in Text S1, and data not shown), and analogously the CHK-166 escape variant was inhibited by CHK-102, CHK-152, and CHK-263 (Fig. 6C, and data not shown). In contrast, CHK-102 and CHK-263 escape variants reciprocally were resistant, suggesting their epitopes were the same or overlapping (Fig. 6A and D); however, CHK-102 and CHK-263 escape variants remained sensitive to neutralization by CHK-152 and CHK-166. Notably, selection with combinations of MAbs (e.g., CHK-102+CHK-152) failed to produce escape variants despite several independent attempts (data not shown).
To identify the mutations that conferred resistance, we sequenced plaque-purified escape variants (Table 2, top). Six of eight sequences from CHK-102 escape variants contained an L210P mutation in the E2 protein; the remaining two sequences had a G209E mutation in E2. For CHK-152 resistant variants, all sequences (9 of 9) contained a D59N mutation in E2 and two contained a second A89E substitution in E2. For CHK-263, 3 of 4 escape variants had a K215E change in E2, whereas 1 of 4 had mutations in E2 at G209E. All escape variants (14 of 14) of CHK-166 had a single K61T mutation in the E1 protein.
To verify the amino acid changes that conferred MAb resistance in vitro, we introduced several of these substitutions into a chimeric SFV-GFP-CHIKV cDNA comprised of SFV non-structural genes, a GFP reporter gene, and the CHIKV structural genes (T. Lin, K. Dowd, and T. Pierson, unpublished results). Parental and SFV-GFP-CHIKV with single amino acid mutations were analyzed for neutralization by CHK-102, CHK-152, CHK-166, and CHK-263 (Fig. 6E–H). Consistent with our sequencing results, viruses encoding mutations in E2-G209 and E2-L210 were resistant to CHK-102, changes in E2-D59 conferred resistance to CHK-152, substitutions in E1-K61 resulted in resistance to CHK-166, and mutation of E2-G209 and E2-K215 caused resistance to CHK-263. However, introduction of E2-A89E (which was present in 2 of 9 clones) failed to affect the neutralizing activity of CHK-152.
In addition to selecting escape variants in cell culture, we harvested organs from the few mice that became ill after infection despite single MAb treatment (Table 3, bottom). In these moribund Ifnar−/− mice, CHIKV was present in the brain and muscle but absent from the spleen or liver (data not shown). This in vivo-derived virus was tested for MAb resistance and sequenced. For mice receiving a 10 µg dose of CHK-102 as prophylaxis, resistant variants with a L210P mutation in E2 were obtained. For mice receiving CHK-263 or CHK-102 at 24 hours post infection, resistant viruses with a G209E mutation in E2 were identified. None of the animals that were pre-treated with 10 µg of CHK-166 developed escape mutants, as the virus harvested from all 3 mice tested retained sensitivity to CHK-166 (data not shown). However, in one animal receiving CHK-166 at 24 hours post infection, a single resistant virus with a G64S substitution in the E1 gene was recovered (Fig. S6). For mice receiving a 10 µg dose of hu-CHK-152 as prophylaxis, partially resistant viruses with N231D and K233E mutations in E2 were isolated and confirmed by reverse genetics using the chimeric SFV-GFP-CHIKV infectious clone (Fig. S7). In comparison, when CHK-152 was given as a therapeutic, a single mutation at D59N in E2 was obtained in 4 of the 5 mice tested, with a K233T mutation in virus from the remaining animal. For animals treated at 48 hours with combination MAb therapy, all recovered viruses remained sensitive to CHK-152 yet showed partial resistance to CHK-102 or CHK-166 (Fig. S8). Mutations in E2 (N332I, CHK-166+CHK-152) were identified. Comparison of 140 available E1 and E2 sequences from historical and circulating CHIKV strains in a public database (http://www.viprbrc.org/) revealed nearly complete conservation of the residues in which escape mutants were selected: E1-K61, 100%; E1-G64, 100%; E2-D59, 100%; E2-G209, 100%; E2-L210, 99.3%; E2-K215, 100%; E2-N231, 100%; and E2-K233, 99.3%.
To define spatially the location of the amino acids that conferred resistance to our highly protective MAbs, these residues were mapped onto the existing CHIKV protein crystal structures [10] (Fig. 6I, left). Amino acids that conferred neutralization escape to CHK-102 and CHK-263 were located in the B domain of E2. The residues that modulated CHK-152 neutralization mapped to the A domain of E2. In contrast, CHK-166 recognized amino acids on DII of E1, adjacent to the fusion loop. All amino acids that conferred neutralization escape appear solvent accessible and highly exposed when docked onto the E2-E1 spike (Fig. 6I, right).
We set out to identify MAbs with the greatest therapeutic activity against CHIKV in mice as a first step toward generating an immunotherapy for humans. Thirty-six MAbs with neutralizing activity against CHIKV-LR were identified, the majority of which also inhibited infection of strains corresponding to the two heterologous CHIKV genotypes. Although all fourteen of the selected anti-CHIKV MAbs improved outcome in vulnerable Ifnar−/− mice, only four of these (CHK-102, CHK-152, CHK-166, and CHK-263) completely prevented lethality when administered as prophylaxis. CHK-152 provided the greatest benefit as post-exposure therapy, although by itself, the window of treatment activity was limited in the Ifnar−/− mouse model. While addition of a second MAb (CHK-102, CHK-166, or CHK-263) failed to enhance CHK-152 neutralization in vitro, it limited the development of viral resistance in vitro and in vivo. Remarkably, combinations of CHK-102+CHK-152 or CHK-166+CHK-152 protected Ifnar−/− mice against mortality even when a single dose was administered 24 to 36 hours prior to the death of untreated or isotype control MAb-treated animals.
In comparison to the highly therapeutic activity of 0.5 mg of CHK-152+CHK-166, a single 25 mg dose of immune IgG purified from a convalescent human subject protected only 50% of Ifnar−/− mice when administered 24 hours after CHIKV infection [26]. The administered dose of neutralizing antibody likely is critical to post-exposure treatment of CHIKV infection because of the high viral burden [14], [16], [17], [40]. A high viral load impacts therapeutic activity of antibodies as it (a) increases the chance for pre-existing or selected resistant variants to emerge through quasispecies [28], [41]; and (b) results in a low relative fractional occupancy of binding to any individual virion, which allows antibodies recognizing key epitopes to fall below their stoichiometric threshold of neutralization [42]. Although there is extensive literature on the protective efficacy of MAbs or immune sera against alphavirus infection [18]–[25], no prior study has demonstrated reduced CHIKV-induced mortality with MAbs. Although a recent study showed that combination post-exposure therapy with two human anti-CHIKV MAbs (5F10 and 8B10, 250 µg each at +8 h) prolonged survival of AG129 (Ifnar−/−×Ifngr−/−) mice by ten days, they failed to prevent lethal infection [29]; the basis of this treatment failure remains unclear but could reflect the lower neutralizing potency of the MAbs (compared to CHK-152), rapid emergence of resistant mutants, or the relative susceptibility of the immunocompromised mouse host. In comparison, a neutralizing MAb (UM 5.1) administered two days after SFV infection completely protected immunocompetent BALB/c mice [43].
Why were some combinations of two MAbs effective in vivo? (a) Pairs of MAbs may show neutral, additive, or synergistic effects on neutralization. Positive antiviral effects could occur through cooperative binding or by trapping CHIKV in conformations that makes it less competent to bind a receptor or fuse with host membranes. Nonetheless, when we added increasing concentrations of CHK-102, CHK-166, or CHK-263 to CHK-152, we failed to observe synergy. (b) Certain MAb combinations could prevent the emergence of resistance due to the low frequency of two escape mutations occurring simultaneously in a single replication cycle. Although we could readily select for neutralization escape against a single MAb in vitro and in vivo, we failed to isolate resistant mutants against CHK-152 when two MAbs (e.g., CHK-102+CHK-152) were combined. However, some viruses from moribund animals treated with combination MAb therapy showed reduced sensitivity (up to 200-fold) to the other MAb (e.g., CHK-102) in the pair. In comparison, when mice were treated with a combination of 50 µg each of CHK-102+CHK-263, we failed to observe the same survival benefit that was conferred by the combinations of CHK-102, CHK-166, or CHK-263 with CHK-152. Since CHK-102 and CHK-263 appear to share overlapping footprints on domain B of E2, this particular MAb combination may fail to prevent the rapid emergence of escape mutants relative to others targeting distinct epitopes on E1 and E2 proteins. (c) Combinations of MAbs could select for resistant viruses that have reduced fitness [44], and thus are less pathogenic in vivo. Virulence studies with CHIKV encoding selected single and double mutations are planned to evaluate this possibility.
We localized the epitopes of our four highly protective MAbs using neutralization escape selection, sequencing, and reverse genetics. CHK-152, which blocked viral fusion, mapped to the wings of the A domain on E2, a result that we recently confirmed by cryo-electron microscopic analysis of CHK-152 Fab-virus particle complexes [45]. This epitope also was identified as a recognition site for neutralizing MAbs against VEEV [46] and SINV [47]. CHK-166, which was the least neutralizing (EC50 of ∼100 ng/ml) of our highly protective MAbs mapped to an epitope in domain II of the E1 protein, adjacent to the highly conserved fusion loop. While anti-E1 MAbs against SINV and VEEV that protect or neutralize infection have been described [46], [48], [49], none have been characterized against CHIKV. A neutralizing human MAb (8B10) against CHIKV was reported with possible reactivity against E1, although further analysis revealed that it bound to the E1/E2 heterodimer [27], [28]. CHK-102 and CHK-263 mapped to residues within the B domain on E2. A related epitope also was identified in mapping studies of strongly neutralizing antibodies against Ross River virus [50], SINV [51], [52], VEEV [46], [53], [54], and CHIKV [10], [28]. The B domain on E2 comprises an important antigenic domain that is under selective pressure for antibody neutralization [41]. It serves as a cap to the fusion loop on E1 and because of its location at the tip of the heterodimeric spike [10], [11] may contribute to attachment of cellular receptors.
In summary, we identified combinations of MAb pairs that were highly effective as post-exposure therapeutic agents. These findings are consistent with recent studies showing enhanced post-exposure efficacy of MAb combinations against Ebola [55], influenza A [56] and rabies [57] viruses. Our most promising pair of MAbs mapped to distinct epitopes, limited the generation of resistance, blocked multiple stages of the viral entry pathway, and protected Ifnar−/− mice against mortality even when administered 60 hours after infection. CHK-152 was humanized as a first step towards a possible therapeutic for humans and demonstrated similar efficacy compared to the parent murine MAb. Tailored combinations of potently neutralizing MAbs show promise to prevent or treat infection by CHIKV, and likely other pathogenic alphaviruses in humans. Ultimately, a more detailed kinetic analysis of CHIKV infection in humans and determination of a treatment window relative to symptom onset is warranted to establish whether combination MAb therapy can prevent or mitigate acute or chronic and persistent infection and joint disease.
Vero, Vero76 (ATCC), BHK21-15, and NIH 3T3 mouse fibroblast cells were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 5% or 15% (for 3T3 cells) fetal bovine serum (FBS) (Omega Scientific). C6/36 Aedes albopictus cells were grown in Leibovitz-15 medium supplemented with 10% FBS at 27°C. The infectious clones of CHIKV La Reunion 2006 OPY-1 (strain 142, CHIKV-LR) and CHIKV-GFP (strain 145) were gifts from S. Higgs (Manhattan, KS) [58]. CHIKV-RSU1 and CHIKV-IbH35 were gifts of R. Tesh, (Galveston, TX). Infection studies of WT mice used the SL15649 strain of CHIKV, which was generated from an infectious clone [17]. The S27 African prototype CHIKV strain was a gift from Dr. S. Günther (Bernhard-Nocht-Institute for Tropical Medicine, Germany) and isolated from a patient in Tanzania in 1953.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Institutional Animal Care and Use Committee at the Washington University School of Medicine (Assurance Number: A3381-01) and the University of North Carolina (A3410-04). Dissections and footpad injections were performed under anesthesia that was induced and maintained with ketamine hydrochloride and xylazine, and all efforts were made to minimize suffering.
Chimeric SFV-CHIKV virus was generated by complementation of a double sub-genomic DNA-launched SFV replicon “backbone” plasmid (pSFV-GFP-BB) with the structural genes of CHIKV as described recently for WNV [59]. The vectors and methods will be described in detail elsewhere (TY Lin, K. Dowd, and T. Pierson, in preparation). To generate SFV-CHIKV, a DNA fragment encoding WT or mutant CHIKV structural genes was ligated into the pSFV-GFP-BB plasmid and transfected directly into HEK-293T cells using Lipofectamine LTX. The source of CHIKV structural genes was a sub-cloning vector pCHIKV-struct: mutations were introduced into this vector using site-directed mutagenesis and fully sequenced. Virus was harvested at 48, 72, or 96 hours after transfection, filtered, and stored at −80°C.
The CHIKV E2 ectodomain (residues S1-E361) and pE2-E1 (E3-E2-E1: residues S1-R64 of E3, S1-E161 of E2, and Y1-Q411 of E1 including a (GGGS)4 polylinker between E2 and E1) of the CHIKV-LR strain were amplified from the infectious cDNA clone using high-fidelity Phusion PCR (Thermo Scientific). The E2 ectodomain was cloned into pET21a, expressed in E. coli, and purified using an oxidative refolding protocol followed by size-exclusion column purification using fast protein liquid chromatography [60]. pE2-E1 was cloned into the mammalian expression vector pHLsec (Invitrogen) with a C-terminal octa-histidine tag and modified to express the Epstein–Barr virus EBNA-1 protein for enhanced protein expression. pE2-E1 was expressed in serum-free HEK-293F suspension cells and purified by Ni-NTA agarose affinity (Qiagen) and Superdex 200 gel filtration chromatography.
Irf7−/− mice were infected and boosted with 104 PFU of CHIKV-LR and, depending on the experiment, given a final intravenous (i.v.) boost with CHIKV virus-like particles [30], 25 µg of E2 protein, or 2×105 PFU of CHIKV-LR three days prior to fusion with myeloma cells. Hybridomas secreting antibodies that reacted with CHIKV-GFP-infected BHK21-15 cells were identified by flow cytometry and cloned by limiting dilution. MAbs were isotyped by ELISA (Pierce), adapted for growth under serum-free conditions, and purified by protein G affinity and size exclusion chromatography. All MAbs were screened initially with a single endpoint neutralization assay using neat hybridoma supernatant (∼10 µg/ml), which was incubated with 100 FFU of CHIKV-LR for one hour at 37°C. MAb-virus complexes were added to BHK21-15 cell monolayers in 6-well plates. After 90 minutes, cells were overlaid with 1% (w/v) agarose in Modified Eagle Media (MEM) supplemented with 4% FBS. Plates were fixed with 10% formaldehyde in PBS 48 hours later, stained with crystal violet, and plaques were counted. The VH and VL sequence of neutralizing MAbs CHK-102, CHK-152, CHK-166, and CHK-263 were amplified from hybridoma cell RNA by a 5′ RACE procedure Table S3 in Text S1).
The generation of a chimeric mouse-human CHK-9 and CHK-152 with mouse VH and VL and human IgG1 constant regions was performed as described previously [60]. A point mutation that abolishes FcγR and C1q binding (N297Q) was introduced by QuikChange mutagenesis (Stratagene). Recombinant antibodies were produced after transfection of HEK-293T cells, harvesting of supernatant, and purification by protein A affinity chromatography.
Serial dilutions of MAb were incubated with 100 FFU of CHIKV for one hour at 37°C. MAb-virus complexes were added to cells in 96-well plates. After 90 minutes, cells were overlaid with 1% (w/v) methylcellulose in Modified Eagle Media (MEM) supplemented with 4% FBS. Plates were harvested 18 to 24 hours later, and fixed with 1% PFA in PBS. The plates were incubated sequentially with 500 ng/ml of ch-CHK-9 and horseradish peroxidase (HRP)-conjugated goat anti-human IgG in PBS supplemented with 0.1% saponin and 0.1% BSA. CHIKV-infected foci were visualized using TrueBlue peroxidase substrate (KPL) and quantitated on an ImmunoSpot 5.0.37 macroanalyzer (Cellular Technologies Ltd). Non-linear regression analysis was performed, and EC50 values were calculated after comparison to wells infected with CHIKV in the absence of antibody.
96-well tissue culture plates were coated with 100 µl of poly-L lysine and seeded with 3×104 Vero cells/well overnight. For pre-attachment assays, dilutions of MAb were prepared at 4°C in DMEM with 2% FBS and pre-incubated with 100 FFU of CHIKV-LR for one hour at 4°C. MAb-virus complexes were added to pre-chilled Vero cells for one hour at 4°C. Non-adsorbed virus was removed with three washes of DMEM and adsorbed virus was allowed to internalize during a 37°C incubation for 15 minutes. Cells were overlaid with 1% (w/v) methylcellulose in MEM supplemented with 4% FBS. The post-attachment assay was performed similarly, except that an equivalent amount of CHIKV was adsorbed first onto Vero cells for one hour at 4°C. After removing free virus, dilutions of MAb were added to the virus-adsorbed cells for one hour at 4°C. Virus was allowed to internalize and cells were overlaid with methylcellulose as described above. Nineteen hours later, the plates were harvested and analyzed for antigen-specific foci as described above.
The binding of human FcγR and C1q to ch-CHK-152 and ch-CHK-152 (N297Q) was analyzed by SPR using a BIAcore 3000 biosensor (GE Healthcare Life Sciences). MAbs were captured (∼900 RU) after flowing over immobilized F(ab)′2 fragments of goat anti-human F(ab)′2 specific IgG on a CM-5 sensor chip. Binding experiments were performed in HBS-EP buffer (10 mM Hepes, pH 7.4, 150 mM NaCl, 3 mM EDTA, and 0.005% P20 surfactant). Binding of CD16A and CD64 (as monomeric soluble FcγR), CD32A (as dimeric soluble FcγR-aglycosylated Fc fusion), and C1q (Sigma-Aldrich) was analyzed at a single concentration. The FcγR and C1q were injected for 60 sec at a flow rate of 30 µl/min then allowed to dissociate over 2 minutes. Affinity measurements of CHK-152 MAbs for pE2-E1 were performed by SPR in HBS-EP buffer. Ch-CHK-152, ch-CHK-152 N297Q, hu-CHK-152 and mouse CHK-152 were captured (∼300 RU) after flowing over immobilized F(ab)′2 fragments of goat anti-human or anti-mouse Fc specific IgG. Purified pE2-E1 was injected at concentrations of 0, 6.25, 12.5, 25, 50, and 100 nM, at a flow rate of 30 µl/min for 120 sec, and then allowed to dissociate over 2 minutes. Regeneration of capture surfaces was performed by pulse injection of 10 mM glycine pH 1.5. Binding curve at the zero concentration of pE2-E1 was subtracted from each experimental curve as a blank. Data were analyzed using BIAevaluation 4.1 software. Kinetic constants, ka and kd, were estimated by global fitting analysis of the association/dissociation curves to the 1∶1 Langmuir interaction model.
CHIKV-LR (1.2×105 FFU) was incubated with 25 µg/ml of MAbs for one hour at 37°C. Virus-MAb complexes were added to Vero cells and infection proceeded for 24 hours. At each passage, half of the supernatant was mixed (1∶1) with 50 µg/ml of the selection MAb for one hour at 37°C. These complexes were added to a new monolayer of Vero cells for 2 hours, and the procedure was repeated from 3 to 6 times depending on the selection MAb. Individual MAb-resistant viral plaques were picked and virus was grown in Vero cells in the presence of 10 µg/ml of MAb for 24 hours. RNA was isolated from cells using an RNeasy kit (Qiagen) and cDNA was made with random hexamers using the Superscript III Reverse Transcriptase kit (Invitrogen) and amplified by PCR with primers flanking the structural genes (Table S4). The PCR product was sequenced using ten overlapping primer sets (Table S4).
Figures were prepared using the atomic coordinates of CHIKV pE2-E1 (RCSB accession number 3N44) using the program CCP4MG [61].
For survival analysis, Kaplan-Meier survival curves were analyzed by the log-rank test. For growth kinetics and neutralization an unpaired T-test or analysis of variance was used to determine significance. These analyses were assessed using Prism software (GraphPad software). The protective effects of ch-CHK-152 versus ch-CHK-152 N297Q in wild type C57BL/6 mice were analyzed by the Kruskal-Wallace test with Bonferroni correction using the agricolae package of R (R Development Core Team, 2010. Foundation for Statistical Computing, Vienna, Austria).
|
10.1371/journal.ppat.1000453 | A Therapeutic Antibody against West Nile Virus Neutralizes Infection by Blocking Fusion within Endosomes | Defining the precise cellular mechanisms of neutralization by potently inhibitory antibodies is important for understanding how the immune system successfully limits viral infections. We recently described a potently inhibitory monoclonal antibody (MAb E16) against the envelope (E) protein of West Nile virus (WNV) that neutralizes infection even after virus has spread to the central nervous system. Herein, we define its mechanism of inhibition. E16 blocks infection primarily at a post-attachment step as antibody-opsonized WNV enters permissive cells but cannot escape from endocytic compartments. These cellular experiments suggest that E16 blocks the acid-catalyzed fusion step that is required for nucleocapsid entry into the cytoplasm. Indeed, E16 directly inhibits fusion of WNV with liposomes. Additionally, low-pH exposure of E16–WNV complexes in the absence of target membranes did not fully inactivate infectious virus, further suggesting that E16 prevents a structural transition required for fusion. Thus, a strongly neutralizing anti–WNV MAb with therapeutic potential is potently inhibitory because it blocks viral fusion and thereby promotes clearance by delivering virus to the lysosome for destruction.
| Antibodies are essential components of the immune response against many pathogens, including viruses. A greater understanding of the mechanisms by which the most strongly inhibitory antibodies act may influence the design and production of novel vaccines or antibody-based therapies. Our group recently generated a highly inhibitory monoclonal antibody (E16) against the envelope protein of West Nile virus, which can abort infection in animals even after the virus has spread to the brain. In this paper, we define its mechanism of action. We show that E16 blocks infection by preventing West Nile virus from transiting from endosomes, an obligate step in the entry pathway of the viral lifecycle. Thus, a strongly inhibitory anti–West Nile virus antibody is highly neutralizing because it blocks fusion and delivers virus to the lysosome for destruction.
| Neutralizing antibodies can inhibit virus infection by impeding one of several critical steps of the virus lifecycle. These include blocking attachment to the cell surface, interaction with host factors required for internalization, and structural transitions on the virion that drive membrane fusion (reviewed in [1],[2]). Antibodies can independently neutralize virus infection by promoting virus aggregation, destabilizing virion structure, and blocking budding or release from the cell surface (reviewed in [3]). Historically, many of the most potently neutralizing antibodies inhibit infection by interfering with required interactions between viruses and obligate cellular receptors (e.g., rhinovirus and ICAM-1, HIV and CD4 or CCR5, and poliovirus and CD155).
West Nile virus (WNV) is a mosquito-borne positive polarity RNA virus of the Flavivirus genus within the Flaviviridae family. Similar to other Flaviviruses, such as Dengue (DENV), yellow fever, and Japanese encephalitis viruses, WNV has an ∼11 kb RNA genome that encodes three structural (C, prM/M and E) and seven non-structural (NS1, NS2a, NS2b, NS3, NS4a, NS4b, and NS5) proteins that are generated by cleavage from a single polyprotein [4],[5]. WNV has spread globally and epidemic outbreaks of encephalitis now occur annually in the United States. Infection with WNV causes syndromes ranging from a mild febrile illness to severe neuroinvasive disease and death [6],[7]. There is currently no approved vaccine or therapy for WNV infection.
Structural analysis of the WNV and DENV virions by cryo-electron microscopy [8],[9] reveals a ∼500 Å mature virion with a smooth outer surface. The 180 copies of the E glycoproteins lay relatively flat along the virus surface as anti-parallel dimers in three distinct symmetry environments. Following exposure to low pH in the endosomal compartment, the E proteins rearrange from homodimers to homotrimers, exposing a fusion peptide, which interacts with the endosomal membrane and allows uncoating and nucleocapsid escape into the cytoplasm [10].
The atomic structure of the surface E glycoprotein has been defined by X-ray crystallography for DENV, WNV, and tick-borne encephalitis virus (TBEV) [11]–[15], revealing three conserved domains. Domain I (DI) is a 10-stranded β-barrel and forms the central structural architecture of the protein. Domain II (DII) consists of two extended loops projecting from DI and contains the putative fusion loop (residues 98–110), which participates in a type II fusion event [10],[16],[17]. In the mature virus, the fusion loop packs between two anti-parallel dimers and is solvent inaccessible, protecting the virus from premature fusion and inactivation. Domain III (DIII) is located on the opposite end of DI, forms a seven-stranded immunoglobulin-like fold, and has been suggested as a receptor binding site [18]–[20].
The humoral immune response controls WNV pathogenesis as mice lacking B cells are highly vulnerable to lethal infection [21]. During infection with flaviviruses, most neutralizing antibodies are directed against the E protein, although a subset binds the prM protein [22],[23]. To better understand the structural basis of antibody protection against WNV, we recently generated a large panel of monoclonal antibodies (MAbs) against WNV E protein [24]. One antibody, E16, was observed to block WNV infection in vitro and in vivo and was effective as a post-exposure therapy even 5 days after infection [24],[25]. Potent E16 neutralization occurs with strikingly low stoichiometric requirements, as a virion occupancy of ∼25% is sufficient to inhibit infection [26]. Herein, we determine the mechanism by which this therapeutic MAb neutralizes WNV infection. E16 traffics with WNV particles into permissive target cells, and is strongly inhibitory because it blocks pH-dependent fusion, a critical step in the entry pathway of this virus.
A common mechanism of antibody-mediated neutralization of viral infection is to prevent attachment and entry into target cells. Previously published studies suggested that E16 did not dramatically reduce WNV binding to Vero cells but instead inhibited at a post-attachment step [27]. To gain further insight as to how E16 inhibits infection, WNV was pre-incubated with Alexa-488 conjugated E16 or E53, a second inhibitory MAb that binds to the fusion loop in DII, prior to a cell binding assay at 4°C. Subsequently, cells were washed at 4°C, fixed and visualized by confocal microscopy. At 4°C, enveloped viruses, including flaviviruses, remain on the cell surface and are not internalized [28]–[30]. As expected, in the absence of WNV, labeled E16 and E53 were not visualized on the surface or interior of cells (data not shown). When Alexa-488-E53-WNV complexes were added, no fluorescence signal was observed on the surface of Vero cells (Figure 1A, panels F and H), suggesting that E53, as hypothesized previously [27], primarily inhibits WNV attachment to Vero cells. Similar results were obtained with Alexa-488 conjugated E60, a MAb that binds to a similar epitope as E53 in DII (data not shown). In contrast, staining was apparent on the surface of cells incubated with labeled Alexa-488-E16-WNV complexes. Thus, despite saturating and neutralizing concentrations (100 μg/ml) of E16 MAb, WNV binding to Vero cells still occurred (Figure 1A, panels B and D). Analogous results were obtained with the strongly neutralizing DIII-specific E24 MAb (data not shown).
To determine if the E16 MAb restricted virus entry, Vero cells were warmed to 37°C after MAb-WNV complex pre-binding at 4°C, and again visualized by confocal microscopy. As anticipated, Alexa-488-E53-WNV complexes were not detected inside cells (Figure 1A, panels N and P). In contrast, Alexa-488-E16-WNV complexes readily entered cells and accumulated in acidic vesicles that were identified with a pH sensitive dye (Figure 1A, panels J and L). Even after several hours of incubation, E16-WNV complexes remained localized in these acidic cellular compartments (Figure 1B, panels B–D), whereas E53-WNV complexes were not detected within the cells (Figure 1B, panels F–H). In contrast, in the absence of neutralizing antibodies, WNV infection progresses rapidly as demonstrated by the accumulation of E protein in the cell over time (Figure S1).
Because E16-WNV complexes co-localized with an acidified intracellular compartment for several hours, we hypothesized that this MAb prevented virus fusion with endosomal membranes. Because WNV infection requires a pH-dependent structural rearrangement of E proteins for fusion, we evaluated whether concanamycin A1, a vacuolar-ATPase inhibitor [31], blocked WNV infection at a similar cellular stage as did E16. Vero cells were infected at a high multiplicity of infection (MOI) in the presence of 10 nM concanamycin A1 or humanized E16 (hu-E16, 100 μg/ml) or a media control for 3 h or 24 h at 37°C. Cells were washed, fixed, and stained for WNV using an oligoclonal pool of mouse MAbs against the E protein. Samples treated with hu-E16 were also stained with an anti-human IgG secondary antibody to confirm that hu-E16 co-localized with the virus. In the absence of concanamycin A1 or hu-E16, infected Vero cells showed strong staining of E protein at 3 h that was markedly increased at 24 h (Figure S1). Treatment with 10 nM concanamycin A1 resulted in a punctate pattern of E protein staining at 3 and 24 h, suggesting that WNV localized to and likely remained sequestered in endocytic compartments (Figure 2, panels A and D). Analogous to treatment with concanamycin A1, hu-E16-opsonized WNV showed a similar staining pattern up to 24 hours after infection (Figure 2, panels B and E). As co-staining of oligoclonal mouse anti-E protein and hu-E16 was observed over time, it is likely that E16 was still bound to WNV, and these virus-MAb complexes accumulated in endosomal/lysosomal compartments (Figure 2, panels C and F). Of note, in Figure 2C, only a subset of the blue spots (which indicates the presence of the virion) co-stain with hu-E16. This is likely a sensitivity of detection issue as E16 neutralizes infection at both low (∼25% or 30 copies per virion) and high occupancy [26]. Because of the high MOI used, some viruses will be more completely decorated (and thus fluorescent), whereas others will bind fewer antibodies yet still be neutralized. Virions that bind fewer E16 antibodies yet still are neutralized may co-stain less brightly in this microscopic assay.
The ability of E16 to block WNV egress from endosomes suggested that this MAb directly inhibited the pH-dependent fusion step. Initially, to test this, we used a surrogate plasma membrane fusion infection assay that has been validated for alphaviruses and flaviviruses [32],[33]. Normally, flaviviruses enter cells via receptor-mediated endocytosis, with fusion occurring from within acidic endosomes [29],[34],[35]. However, flaviviruses also can be induced to fuse directly with the plasma membrane, at low efficiency, when cell-bound virus is exposed to an acidic solution [32]. To assess the effects of E16 on virus-plasma membrane fusion, WNV was pre-bound to Vero cells at 4°C, and subsequently incubated on ice with saturating concentrations of E16 IgG, E16 single chain Fv (scFv), E60 IgG, or no MAb. Cells were warmed to 37°C in pH 5.5 media (or pH 7.5 media as a negative control) to induce virus-plasma membrane fusion and analyzed at 24 hours for level of infection by flow cytometry. In all experiments, 10 nM concanamycin A1 was added to inhibit infection via the canonical receptor-mediated endocytic pathway. As expected, in the absence of antibody, addition of media at neutral pH (7.5) did not promote productive infection (∼0.7% WNV antigen+ cells, Figure 3A and 3B). Exposure of cell bound WNV to media at pH 5.5 resulted in a ∼7 fold increase in infection (∼5.1% WNV antigen+ cells, P<0.0005, Figure 3A and 3B). The addition of E60 following viral attachment did not appreciably affect virus-plasma membrane fusion (P = 0.4), confirming earlier results that this MAb does not inhibit Vero cell infection at a post-attachment step [27]. In contrast, both E16 IgG and scFv efficiently blocked WNV-plasma membrane fusion (0.15% and 0.08% WNV antigen+ cells, respectively; Figure 3A and 3B, P<0.0001).
To confirm that E16 blocks membrane fusion of WNV, we evaluated the fusogenic properties of WNV in a model liposome system. To this end, WNV particles were metabolically labeled with pyrene hexadecanoic acid and purified by density gradient centrifugation. Subsequently, pyrene-labeled virions were pre-incubated with various concentrations of E16, E60 or E111 (a DIII-specific non-neutralizing control MAb [24]) and mixed with liposomes. The mixture was acidified to pH 5.4 and fusion was measured on-line in a fluorimeter at 37°C as a function of the decrease in pyrene excimer fluorescence. WNV fuses rapidly and efficiently with liposomes. In contrast, no membrane fusion activity was measured with saturating concentrations of E16 (Figure 4A). Inhibition of membrane fusion by E16 was dose-dependent as decreasing concentrations of E16 blocked fusion to a lesser degree (Figure 4A and 4B). E111 did not influence the membrane fusion properties of WNV as efficient fusion was measured at all antibody concentrations tested. MAb E60 was observed to induce a dose-dependent inhibition of membrane fusion activity, although a complete inhibition of fusion was not achieved (Figure 4B).
Previous studies have shown that exposure of WNV or other flaviviruses to acidic (pH<6) media in the absence of target membranes results in E protein rearrangement, premature exposure of the fusion loop, virus aggregation, and rapid irreversible inactivation of fusion competence [36]–[38]. We reasoned that if E16 neutralized WNV infection by directly blocking the pH-dependent fusion event it should prevent adventitious inactivation in solution after exposure to acidic pH. To test this, WNV (3×103 PFU) was pre-incubated with saturating (100 μg/ml) concentrations of E16, E60, or E9 (a DIII non-neutralizing MAb [24]) Fab fragments. Although the E60 MAb did not appear to enter cells or potently neutralize WNV infection [39], we included this fusion loop-specific Fab as a control because it partially inhibited pH-catalyzed virus fusion in the liposome assay. Excess buffered media at pH 7.5 or pH 5.5 was added to the virus-Fab complexes and incubated at 37°C for 15 min. The solution was normalized after dilution with a 25-fold excess of pH 7.5 media and added to Vero cells for 1 h at 37°C to allow infection as the monovalent Fab fragments detached. As expected, exposure to a pH 7.5 solution did not change WNV infectivity, as the monolayer contained ∼3.9×103 PFU (Figure 5). In contrast, treatment with a pH 5.5 solution inactivated WNV and reduced infectivity (P<0.0001) below the limit of detection (∼20 plaques). The E9 Fab failed to protect the virus from low pH inactivation, whereas neutralizing concentrations of E16 and E60 Fabs at pH 5.5 partially protected WNV from pH-induced inactivation as 2.2 and 8.2×102 PFU were detected, respectively (Figure 5; P<0.05 and P<0.0001).
Because less infectious virus was detected with E16 compared to E60 treatment following pH normalization and dilution, we hypothesized that even a small number of bound E16 Fab could still inhibit infectivity since this MAb requires a low fractional occupancy for efficient neutralization [26]. Conversely, even detachment of a few E60 Fabs could significantly increase infectivity because virtually complete occupancy is required for neutralization by this MAb [40]. Experiments were repeated and excess recombinant E protein DIII (0.4 mg/ml) was added at the time of pH normalization and dilution to compete off additional bound E16 Fab. The addition of excess recombinant DIII further increased WNV infectivity by ∼4 fold (data not shown), presumably by lowering the number of bound E16 Fab on some virions below the threshold for neutralization. Overall, these experiments show that saturating concentrations of both E16 and E60 Fabs at least, partially prevent irreversible pH-dependent inactivation of WNV in the absence of target membranes.
Antibody neutralization is essential for protection against infection by many viruses. A greater understanding of the mechanism(s) by which the most strongly neutralizing antibodies act could facilitate strategies for generating targeted vaccines and immunotherapies. To establish the mechanism of action of E16, a strongly neutralizing anti-WNV MAb with therapeutic potential, we performed a series of cellular and biochemical experiments. Cell biology studies demonstrate that E16 blocks WNV infection at a post-entry stage by sequestering the virus in acidic compartments and preventing its egress into the cytoplasm. Biochemical experiments demonstrate that E16 neutralizes WNV by directly blocking the pH-dependent fusion process. Thus, the inhibitory activity of E16 against WNV in vivo is likely defined by its ability to block viral fusion and nucleocapsid penetration into the cytoplasm where replication occurs.
Analysis of the crystal structure of E16 Fab bound to WNV E protein led to a hypothesis that E16 blocked the structural rearrangement required for fusion at low pH [27]. Indeed, E16 engages a large solvent-exposed surface of DIII, a domain that is positioned distinctly in the pre- and post-fusion E protein conformations [10]. The biochemical data presented here demonstrating that E16 Fab blocks the pH-dependent inactivation of WNV in solution is consistent with a direct inhibition of the structural transition of E protein that occurs during fusion. Nonetheless, definitive evidence of this structural mechanism awaits solution of the E16-WNV structure by cryo-electron microscopy in media at acidic pH.
In surface plasmon resonance (SPR) binding studies, E16 bound DIII of the WNV E protein with similar affinity across a range of pH values from pH 5 to pH 8 (B.S. Thompson, M.S. Diamond and D.H. Fremont, unpublished data). This explains why the binding and neutralizing activity of E16 is not altered as the virus-MAb complex transits through the endosomal compartments. Indeed, the confocal microscopy experiments showed co-localization of E16 and virus through acidic compartments into the lysosome. Our investigations with MAbs are consistent with an earlier study showing a strongly neutralizing polyclonal serum against WNV inhibited at a post-attachment step [41]; the authors of that study speculated but did not show that the most potently inhibitory antibodies block viral fusion. One reason why antibody blockade of fusion may be particularly potent in vivo for flaviviruses is because it acts downstream of an increasing number of cellular attachment factors (e.g., DC-SIGN, DC-SIGNR, heparin sulfate, Fc-γ receptors, and αvβ3 integrin [42]–[45]).
The confocal microscopy experiments also suggest that E16-opsonized WNV is retained in acidic compartments that are ultimately targeted for degradation. Antibodies like E16 that block fusion may be particularly potent at clearing viral infection in vivo because in addition to directly limiting transit to and replication in the cytoplasm they effectively convert permissive cells into ones that target virus for destruction. This feature of E16, along with its ability to disrupt transneuronal spread [46], high affinity, and capacity to neutralize at low virion occupancy [26], begins to explain its single-dose potent post-exposure therapeutic activity in animals [24],[47].
The mechanistic analysis of E16 and WNV is supported by recent studies with MAbs against DIII of TBEV, some of which also blocked fusion of pyrene-labeled virus with liposomes [48]. Nonetheless, it remains unclear if the DIII MAbs against TBEV have equivalent neutralizing capacity and bind the same structural epitope as E16. The TBEV study also showed that DII-fusion loop MAbs were effective at blocking liposomal fusion. Although we also observed efficient dose-dependent inhibition of membrane fusion with E60, approximately one-third of the virus particles remained fusion competent even under conditions of antibody excess. This data is consistent with our observation that E53 and E60 are less strongly inhibitory MAbs against WNV [39] and that heterogeneity of WNV particles with respect to their state of maturation (mostly immature, partially mature, or fully mature) affects the ability of fusion loop MAbs to bind and neutralize infection [40]. As the fusion loop epitope is poorly accessible on the mature WNV virion [13],[40],[49], E53 and E60 MAbs require a relatively high fractional occupancy to inhibit infection [40]. Indeed, they may not achieve sufficient MAb concentration in the endosomes to neutralize by this mechanism. Instead, at least for Vero cells, our data with E53 and E60 suggests that antibodies of this class block at a proximal attachment step [27]. Based on these observations, we have developed a model for how the DII-fusion loop and DIII-lateral ridge MAbs neutralize WNV infection (Figure 6).
Blockade of viral fusion by antibodies or pharmacologic agents is usually considered as a therapeutic strategy for viruses that fuse with the plasma membrane. For example, enfuvirtide (Fuzeon™ or T-20 peptide) effectively inhibits entry of HIV at the plasma membrane of CD4+ T cells by interfering with the requisite structural transition that brings viral and cell surfaces into proximity for fusion (reviewed in [50]). In contrast, there have been relatively few descriptions of antibodies that neutralize flaviviruses by blocking endosomal fusion. Butrapet et al described an anti-Japanese encephalitis virus antibody (MAb 503) that inhibited fusion-induced syncytia of infected insect cells and virus internalization into Vero cells. Although they concluded that this MAb functioned at a step post-attachment, they did not clearly demonstrate that it directly blocked fusion [51]. Similarly, the mechanism of action of the potently neutralizing anti-DENV2 MAb, 3H5-1 [52], has been speculated. Whereas He et al, showed that 3H5-1 blocked attachment of DENV2 to Vero cells [53], Se-Thoe et al, using LLC-MK2 cells, concluded that 3H5-1 primarily blocked the DENV2 fusion at the plasma membrane [54]. We recently localized the epitope of 3H5-1 of DENV2 to residues in the N-terminal region and FG loops of the lateral ridge of DIII, in an analogous position to that for E16 and WNV DIII [55]. Although further studies are necessary, based on structural localization and functional potency, we speculate that 3H5-1 and other strongly neutralizing DIII lateral ridge MAbs inhibit flavivirus infections, at least in part through similar fusion blocking mechanisms.
In summary, our experiments define the mechanism of action of a potently inhibitory therapeutic antibody against WNV. E16 prevents egress of WNV from endosomes, leading to retention in progressively acidic compartments and likely destruction in the lysosome. Vaccines that skew the immune response towards production of antiviral antibodies that inhibit fusion may improve protection against challenge. For highly promiscuous viruses like flaviviruses, targeting of the endosomal fusion step may be particularly relevant given the discovery of increasing numbers of distinct entry pathways on mammalian cells [42],[43].
Vero cells were used for confocal microscopy experiments, the plasma membrane fusion assay, and to titrate infectious virus by plaque assay. Vero cells were grown in Dulbecco's modified eagle's medium (DMEM) supplemented with 10% FBS, 10 mM HEPES and 1% penicillin/streptomycin, as described [56]. WNV (strain 3000.0259, New York, 2000) [57] was propagated in C6/36 Aedes albopictus cells, aliquotted, and frozen at −80°C.
Pyrene-labeled WNV was isolated from the medium of infected BHK21 cells that was cultured in the presence of 15 μg/ml of 16-(1-pyrenyl)-hexadecanoic acid (Invitrogen, Breda, The Netherlands), essentially as described before for alphaviruses [58],[59]. BHK21 cells were infected at a MOI of 4. At 24 h post-infection, the supernatant was harvested and pyrene-labeled WNV particles were pelleted by ultracentrifugation (Beckman type 19 rotor; 15 hr at 48,500×g at 4°C). Subsequently, the virus particles were purified on an Optiprep (Axis-Shield, Oslo, Norway) density (15–55% w/v) gradient by ultracentrifugation (Beckman SW41 rotor; 18 hr at 100,000×g at 4°C). The infectivity of the virus preparation was determined by titration on BHK21-15 cells. Protein concentration was determined by micro-Lowry analysis.
Large unilamellar vesicles were prepared by a freeze/thaw extrusion procedure as described [59]. Liposomes consisted of phosphatidylcholine (PC) from egg yolk, phosphatidylethanolamine (PE) prepared by transphosphatidylation of egg PC, and cholesterol in a molar ratio of 1:1:2. Liposomes were prepared with an average size of 200 nm. All lipids were obtained from Avanti Polar Lipids (Alabaster, AL).
The anti-WNV antibodies E9, E16, E24, E53, E60, and E111 have been previously described [24],[27],[39]. Fab fragments were generated by papain digestion and purified by protein A affinity and size exclusion chromatography as described [27]. The generation and purification of the E16 scFv will be described in detail elsewhere (B. Kauffman, S. Johnson, D. Fremont, M. Diamond, and M. Rossmann, manuscript in preparation). Direct conjugation of MAbs to fluorochromes was performed using an Alexa Fluor® 488 (or 647) MAb labeling kit (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. Both anti-human and anti-mouse secondary antibodies conjugated to fluorochromes were purchased (Invitrogen) and used at a 1:200 dilution for confocal microscopy and flow cytometry. Flow cytometric analysis was performed using a BD FACS Calibur and BD Cellquest Pro™ software (Becton Dickinson, San Jose, CA).
Vero cells were plated at ∼7,500 cells/well in 8-well Lab-Tek chambered slides (Nunc, Rochester, NY) and incubated overnight. The cells were infected with WNV (MOI of 100) in the presence or absence of Alexa-488 conjugated antibodies at the indicated temperature and time, washed with PBS, and fixed with 2% paraformaldehyde in PBS for 30 min at room temperature. Acidified endosome and lysosome compartments were identified with Lysotracker red (Invitrogen) by adding the dye (50 nM) to the cells for the last 30 min of the incubation prior to fixation.
To assess whether blockade of endosomal acidification mimics treatment with E16, Vero cells were infected at an MOI of 100 in the presence of 10 nM concanamycin A1 or 100 μg/ml hE16 for 3 h or 24 h, fixed with 2% paraformaldehyde, and permeabilized with PBS supplemented with 0.1% saponin. Cells were stained with a pool of Alexa-488 conjugated mouse anti-E MAbs and in some experiments, Alexa-647-conjugated goat anti-human IgG. After extensive washing and fixation, cells were analyzed by confocal microscopy using a Zeiss LSM510 META Laser Scanning Confocal Microscope (Carl Zeiss Inc., Thornwood, NY) as described [60]. Images were analyzed using the LSM510 software suite and Volocity™ software package (Improvision Inc., Waltham, MA).
The assay for plasma membrane fusion of flaviviruses has been described previously [32]. We adapted the protocol to test the effects of MAbs on WNV fusion at the plasma membrane. Briefly, Vero cells were plated in 12 well plates at 5×104 cells per well and incubated for 24 h at 37°C. The cells were then pre-incubated with 10 nM concanamycin A1 for 30 min. WNV (MOI of 100) was complexed with 100 μg/ml E16 IgG, E16 scFv, E60 IgG or control medium for 30 min at 4°C and bound to Vero cells for 2 h on ice. Subsequently, cells were washed twice with iced PBS and pre-warmed DMEM (buffered to pH 5.5 or pH 7.5) was added at 37°C for ∼7 min. The cells were then washed with PBS and incubated for 24 h at 37°C in DMEM containing 10 nM concanamycin A1, which blocks virus fusion after receptor mediated entry pathways. The cells were washed twice in PBS and fixed in PBS with 2% paraformaldehyde, permeabilized with 0.1% saponin and stained with an oligoclonal pool of Alexa Fluor-488-labeled anti-WNV MAbs. Samples were processed by flow cytometry and data was analyzed using the Cellquest Pro™ software.
WNV (∼3×103 PFU) was incubated alone or with 100 μg/ml E16 Fab, E60 Fab or E9 Fab in DMEM at neutral pH for 30 min at 4°C. The reactions were then diluted 5-fold in DMEM supplemented with 20 mM succinic acid (pH 5.5) or 20 mM HEPES (pH 7.5) and incubated at 37°C for 15 min. Each reaction was subsequently neutralized by a 25-fold dilution in DMEM at pH 7.5 and added to a monolayer of Vero cells in a 6 well plate for 1 h at 37°C. Following this incubation, the cells were overlaid with 2% low melting agarose and a standard plaque assay was performed. In some experiments, recombinant DIII (0.4 mg/ml) purified from E. coli [27] was added at the time of 25-fold dilution to compete bound Fabs.
Fusion of pyrene-labeled WNV with PE/PC/cholesterol (molar ratio of 1:1:2) liposomes was monitored continuously in a Fluorolog 3–22 fluorometer (BFi Optilas, Alphen aan den Rijn, The Netherlands), at excitation and emission wavelengths of 345 nm and 480 nm. Pyrene-labeled WNV (0.35 μg protein; corresponds to 1.5×1010 particles) and an excess of liposomes (140 nmol phospholipid; corresponds to 3×1010 liposomes) was mixed in a final volume of 0.665 ml in 5 mM HEPES pH 7.4, 150 mM NaCl, and 0.1 mM EDTA. The content was stirred magnetically at 37°C. At t = 0 sec, the pH of the medium was adjusted to 5.4 by addition of 35 μl 0.1 MES, 0.2 M acetic acid, pre-titrated with NaOH to achieve the final desired pH. The fusion scale was calibrated such that 0% fusion corresponded to the initial excimer fluorescence value. The 100% value was obtained through the addition of 35 μl 0.2 M octaethyleneglycol monododecyl ether (Fluka Chemie AG, Buchs, Switzerland) to achieve an infinite dilution of the probe. The extent of fusion was determined 60 seconds after acidification. To analyze the influence of E16, E60, and E111 on WNV fusion, pyrene-labeled WNV was incubated with increasing concentrations of MAbs for 1 hr at 20°C prior to mixing with liposomes.
|
10.1371/journal.ppat.1004663 | CD44 Plays a Functional Role in Helicobacter pylori-induced Epithelial Cell Proliferation | The cytotoxin-associated gene (Cag) pathogenicity island is a strain-specific constituent of Helicobacter pylori (H. pylori) that augments cancer risk. CagA translocates into the cytoplasm where it stimulates cell signaling through the interaction with tyrosine kinase c-Met receptor, leading cellular proliferation. Identified as a potential gastric stem cell marker, cluster-of-differentiation (CD) CD44 also acts as a co-receptor for c-Met, but whether it plays a functional role in H. pylori-induced epithelial proliferation is unknown. We tested the hypothesis that CD44 plays a functional role in H. pylori-induced epithelial cell proliferation. To assay changes in gastric epithelial cell proliferation in relation to the direct interaction with H. pylori, human- and mouse-derived gastric organoids were infected with the G27 H. pylori strain or a mutant G27 strain bearing cagA deletion (∆CagA::cat). Epithelial proliferation was quantified by EdU immunostaining. Phosphorylation of c-Met was analyzed by immunoprecipitation followed by Western blot analysis for expression of CD44 and CagA. H. pylori infection of both mouse- and human-derived gastric organoids induced epithelial proliferation that correlated with c-Met phosphorylation. CagA and CD44 co-immunoprecipitated with phosphorylated c-Met. The formation of this complex did not occur in organoids infected with ∆CagA::cat. Epithelial proliferation in response to H. pylori infection was lost in infected organoids derived from CD44-deficient mouse stomachs. Human-derived fundic gastric organoids exhibited an induction in proliferation when infected with H. pylorithat was not seen in organoids pre-treated with a peptide inhibitor specific to CD44. In the well-established Mongolian gerbil model of gastric cancer, animals treated with CD44 peptide inhibitor Pep1, resulted in the inhibition of H. pylori-induced proliferation and associated atrophic gastritis. The current study reports a unique approach to study H. pylori interaction with the human gastric epithelium. Here, we show that CD44 plays a functional role in H. pylori-induced epithelial cell proliferation.
| Chronic gastric inflammation, typically caused by Helicobacter pylori (H. pylori), is the most consistent lesion leading to cancer. During a well-choreographed interaction between H. pylori and the host, the progression from chronic inflammation to cancer involves gastric epithelial changes with evidence of hyperproliferation. Our knowledge of H. pylori pathogenesis is predominantly based on data generated from gastric cancer cell lines or animal models of inflammation. We report the development and use of a novel model of primary human and mouse cultured gastric epithelial cells that are organized into three-dimensional spheroid units containing a lumen, known as gastric organoids. To assay changes in gastric epithelial cell proliferation in relation to the direct interaction with H. pylori, human- and mouse-derived gastric organoids were infected with the bacteria. Cluster-of-differentiation gene (CD44) is a transmembrane receptor responsible for epithelial cell proliferation. We show that CD44 plays a functional role in H. pylori-induced proliferation. In a Mongolian gerbil animal model of H. pylori-induced gastric cancer, we show that inhibiting CD44 blocks epithelial proliferation and subsequently cancer progression in response to bacterial infection. Thus our study provides new insights into the role of CD44 in H. pylori-induced hyperproliferation and progression of gastric disease.
| The major cause of chronic inflammation in the stomach is Helicobacter pylori (H. pylori) [1], and it is widely accepted that chronic inflammation is a trigger for the development of gastric cancer [2]. The severity and localization of the inflammation that results from H. pylori infection is believed to dictate the pathological consequence of disease. Individuals most at risk of developing gastric cancer are those in whom the bacteria colonize the corpus (or fundus) of the stomach, when acid secretion is impaired. The subsequent development of severe inflammation in the gastric fundus leads to atrophy of the acid-secreting parietal cells and subsequently further hypochlorhydria, metaplasia and carcinoma [3,4,5]. Given that individuals most at risk of developing gastric cancer are those in whom the bacteria colonize the corpus [3,4,5], the current research is focused on the use of human- and mouse-derived fundic gastric epithelium, cultured as 3-dimensional structures called gastrointestinal organoids, for the study of H. pylori pathogenesis.
The cytotoxin-associated gene (cag) pathogenicity island is a strain-specific constituent of H. pylori that augments cancer risk [6]. The cag pathogenicity island encodes a type IV secretion system that is a multimolecular complex that mediates the translocation of bacterial factors into the host cell [6,7]. Upon delivery into the host cells by the type IV cag secretion system, CagA translocates into the host cell cytoplasm where it can stimulate cell signaling through interaction with several host proteins [6,8,9] including the tyrosine kinase c-Met receptor [10,11,12]. CagA exerts effects within host cells that mediate carcinogenesis, including aberrant activation of phosphatidylinositol 3-phosphate kinase (PI3K) and β catenin, disruption of apical-junctional complexes, and loss of cellular polarity [13,14,15].
Another host molecule that may influence carcinogenesis in conjunction with H. pylori and CagA is the cluster-of-differentiation (CD) CD44 cell surface receptor for hyaluronate [16]. CD44 is a cell surface adhesion molecule, expressed on a variety of cells including gastric epithelial cells, that has recently been identified as a gastric cancer stem cell marker, whereby cells expressing CD44 have been shown to possess the properties of gastric cancer stem cells [17]. CD44 variant isoforms, in particular CD44v6, was identified as a marker for invasive intramucosal carcinoma and premalignant lesions [18]. Suzuki et al. [19] demonstrated that CagA CM motifs interact with Met leading to sustained PI3K-AKT signaling in response to H. pylori resulting in cellular proliferation. Notably, the isoform containing exon v6 (CD44v6) acts as the coreceptor for c-Met, most probably, through binding of c-Met ligand hepatocyte growth factor (HGF) [20,21]. The coreceptor function of CD44v6 for c-Met is of particular interest given that studies pinpoint CD44v6 as a marker of early invasive intramucosal gastric carcinoma [18]. Whether CD44v6 acts as a coreceptor for the function of c-Met in response to H. pylori infection is unknown.
Our current knowledge of H. pylori pathogenesis is largely based on data generated from gastric cancer cell lines or in vivo animal models of inflammation. Thus, despite extensive evidence demonstrating that H. pylori induces gastric epithelial changes, the direct impact of the bacterium on the normal epithelium is unclear. Culture of primary human- and mouse-derived gastric stem cells as 3-dimensional structures called gastrointestinal organoids are a rapidly emerging approach to study gastrointestinal development, physiology, stem cell biology and disease [22,23,24,25,26,27,28,29]. Troy-positive cells are expressed at the corpus gland base in a subset of differentiated chief cells [23]. Stange et al. [23] demonstrate that Troy-positive chief cells may be used to generate long-lived gastric organoids, but in vitro these cultures are differentiated toward the mucus-producing cell lineages of the neck and pit regions. The Troy-derived organoids are distinct from the cultures that we derive from whole dissociated glands reported here such that we have devised a method to maintain all the major cell lineages of the fundus [22,28]. In this investigation, we used our method of mouse-derived gastric organoid cultures as an approach to assay changes in gastric epithelial cell proliferation in relation to the direct interaction with H. pylori [22,24,29]. To study the functional role of CD44 in the context of human epithelial tissue, we developed a protocol for culturing human-derived gastric organoids. We developed cultures of human-derived fundic gastric organoids independent of the recent report by the Clevers group demonstrating the establishment of a similar culture model for the study of H. pylori pathogenesis [25]. Despite the extensive use of these culture systems for the study of stem cell biology and gastrointestinal development [22,23,24,25,26,27,29], the degree to which these cultures reflect the physiology of native tissue has been reported by our laboratory alone [28]. Here we extend our current knowledge of H. pylori pathogenesis by identifying the signaling mechanism by which bacterial infection induces proliferation in the gastric epithelium. While it is known that c-Met is an important CD44 partner in proliferation, this is the first report that this association occurs in response to H. pylori infection. We find that CD44 plays a functional role in H. pylori-induced proliferation both in vitro and in vivo.
To determine if CD44 plays a functional role in H. pylori-induced proliferation, C57BL/6 (BL/6) control and CD44 deficient mice were infected with mouse adapted LSH100 H. pylori strain for 4 weeks. The LSH100 mouse-adapted strain [30] is a descendant of the clinical isolate G27 [31]. We chose the LSH100 H. pylori to study the mechanism of bacterial-induced proliferation in vivo because this particular strain efficiently expresses virulence factor CagA [14,31,32,33]. The proliferating cells were measured by BrdU incorporation (Fig. 1). There was a significant increase of BrdU positive cells per gland in H. pylori LSH100 strain infected mice (6.23 + 0.53 BrdU+ cells/gland, Fig. 1B, E) compared to Brucella broth control mice (3.64 + 0.04 BrdU+ cells/gland, Fig. 1A, E). Infection of BL/6 mice with LSH100 strain bearing a CagA deletion (∆CagA) lacked the significant increase in proliferation (3.68 + 0.46 BrdU+ cells/gland, Fig. 1C, E). Importantly, CD44 deficient mice infected with H. pylori did not exhibit an increase in proliferation (2.09 + 0.11 BrdU+ cells/gland, Fig. 1D, E) when compared to the Brucella broth uninfected control CD44KO mouse group (2.10 + 0.29 BrdU+ cells/gland, Fig. 1E). These data show that CD44 mediates CagA dependent H. pylori-induced proliferation.
To identify the direct impact of H. pylori on the host gastric epithelium, we employed the use of a mouse-derived gastric organoid culture system. We have previously described a system for culturing gastric organoids derived from mouse fundic tissue, in which the fundic organoids are embedded in Matrigel, provided gastric organoid growth media, and co-cultured in transwell plates with immortalized stomach mesenchymal cells (ISMCs) [22,28]. Organoids were microinjected with H. pylori strain G27. H. pylori strain G27, originally isolated from an endoscopy patient from Grosseto Hospital (Tuscany, Italy) [31], is readily transformable and therefore amenable to gene disruption [34]. Of relevance to the current study, strain G27 efficiently delivers the translocated virulence factor CagA to cells in culture [14,31,32,33]. Therefore, we chose G27 to study the mechanism of H. pylori-induced proliferation using a strain that efficiently expresses virulence factor CagA. Organoids were microinjected with H. pylori G27 strain (Fig. 2A), and bacterial adhesion was confirmed by Warthin-Starry stain (Fig. 2C, D) and culture (Fig. 2E). Quantitative cultures showed a significant increase in the bacteria cultured from organoids infected for 7 days compared to 24 hours and thus confirming bacterial viability within the cultures (Fig. 2E). Therefore, the fundic organoids provided a method by which bacterial-host cell interactions may be studied in the context of an intact normal gastric epithelium in vitro.
When the mFGOs were infected with H. pylori G27 strain, we observed a significant increase in epithelial cell proliferation in response to bacterial infection compared to the uninfected controls (Fig. 3A, B). The proliferative response to H. pylori was significantly blocked when organoids were pretreated with the c-Met inhibitor (c-MetI PF04217903 mesylate) (Fig. 3A, B). Organoids infected with the G27 H. pylori strain that expressed a deletion of CagA (∆CagA) did not differ from the controls with regards to proliferation (Fig. 3A, B). These data show that H. pylori-induced proliferation is mediated by activation of c-Met signaling as previously reported [19]. Therefore, to advance this current knowledge, we examined whether c-Met was associated with CD44. Lysates were prepared from uninfected organoids and organoids infected with either H. pylori (G27 strain) or ∆CagA and immunoprecipitated using an anti-c-Met antibody. Immunoprecipitates analyzed by Western blot using an anti-phosphotyrosine antibody showed an increase in phosphorylated c-Met in response to H. pylori (Fig. 3C). Consistent with published studies CagA coimmunoprecipitated with c-Met [19]. CD44 also coimmunoprecipitated with c-Met. C-Met is an important partner with CD44 in proliferation, but this is the first time that it has been reported that this association occurs in response to H. pylori infection. There is evidence suggesting CD44 binds to hepatocyte growth factor (HGF) and acts as a co-receptor by presenting HGF to c-Met and subsequently activating Met signaling [20,21] (Fig. 3D). In response to H. pylori infection, we show for the first time, HGF also coimmunoprecipitated with c-Met, and HGF expression was significantly upregulated in response to bacterial infection (Fig. 3C). Collectively, these data suggest that H. pylori induces the proliferation of the gastric epithelium by promoting an association between CagA, Met and CD44.
To test the functional role of CD44 in mouse-derived gastric organoids, we cultured organoids derived from mice that lacked the gene for CD44 (CD44KO). Organoids derived from the stomachs of CD44 deficient (CD44KO) mice did not proliferate in response to H. pylori infection (Fig. 3E). CD44KO mouse-derived organoids proliferated in response to a Wnt agonist, and thus showing that this lack of response was specific to H. pylori (Fig. 3E). To determine if CD44 was required for c-Met phosphorylation, lysates from uninfected control, H. pylori infected and ∆CagA infected organoids derived from stomachs of CD44-deficient mice were collected. In H. pylori infected CD44-deficient organoids c-Met was present but not phosphorylated (Fig. 3F). Importantly, in the absence of CD44, CagA co-immunoprecipitated with c-Met and is thus likely to form a complex with this receptor (Fig. 3F). In the absence of CD44, HGF was not detected in the c-Met immunoprecipitated protein complex (Fig. 3F). Collectively, these data further show that both CD44 and c-Met play a functional role in CagA dependent H. pylori-induced epithelial cell proliferation in mouse-derived fundic gastric organoids.
In addition to increased epithelial cell proliferation, we observed striking morphological changes in response to H. pylori infection that were consistent with epithelial-to-mesenchymal transition (EMT) (Fig. 4). In uninfected control (CON) mouse-derived fundic gastric organoids, we observed clear membrane-expressed E-cadherin (Fig. 4A). However, in the H. pylori G27 strain infected organoids there was a disruption in membrane-expressed E-cadherin and transition of the spheroid morphology to a cell monolayer was observed (Fig. 4B). In control mouse-derived fundic gastric organoids infected with the ∆cagA strain, we also observed clear membrane-expressed E-cadherin (Fig. 4C). EMT was also documented by increased expression of markers that included alpha smooth muscle actin (αSMA), SNAIL2, TWIST1, N-cadherin and Zeb1 (Fig. 4D). Organoids derived from the stomachs of CD44KO mice were also microinjected with Brucella broth (uninfected control), H. pylori G27 strain or H. pylori ∆cagA strain. In all three groups membrane-expressed E-cadherin was observed (Fig. 4E). In addition changes in gene expression of αSMA, SNAIL2, TWIST1, N-cadherin and Zeb1 was not observed in infected CD44KO mouse-derived organoids (Fig. 4F). These data show that H. pylori induces an EMT phenotype in the mouse-derived fundic gastric organoids and that this response may be mediated by CD44.
To identify the role of CD44 as a mediator of H. pylori-induced proliferation in human epithelium, we developed the human-derived fundic gastric organoids (hFGOs). Gastric glands were enzymatically dissociated from the human fundic tissue embedded in Matrigel, provided gastric organoid growth media and cultured to form hFGOs epithelial spheres over 7 days (Fig. 5A). These 3 dimensional epithelial spheres contained markers specific for the fundic epithelium, including H+,K+ ATPase (HK), Muc5ac, and Muc6, but do not contain gastrin, a marker for the antral region of the stomach (Fig. 5B). Flow cytometric analysis showed that the majority of cells in the human organoids were positive for the marker for parietal cells, H+,K+ ATPase, as would be expected in oxyntic glands in the human stomach (Fig. 5C). Quantification of the flow cytometry histograms revealed that the hFGOs were comprised of approximately 5% UEAI+ cells (surface mucous pit cells), 15% GSII+ cells (mucous neck cells), 12% pepsinogen C (PgC)+ cells (chief cells), 7% chromogranin A+ cells (ChgA endocrine cells) and 60% H+,K+ ATPase+ cells (HK, parietal cells) (Fig. 5D). As detailed for the mouse-derived fundic gastric organoids, hFGOs were microinjected with H. pylori G27 strain and bacterial adhesion was confirmed by Warthin-Starry stain (Fig. 5E, F). Therefore, the hFGOs provided a method by which bacterial-host cell interactions were studied in the context of an intact normal human gastric epithelium in vitro.
To determine if the co-receptor role of CD44 and c-Met was intact in human tissue, hFGOs were infected with H. pylori G27 strain. Compared to the control treated (CON) hFGOs, infection with H. pylori triggered a significant induction in proliferating cells that was not seen in organoids injected with the ∆CagA mutant strain (Fig. 6A, B). Consistent with the response observed in the mouse-derived organoids, the proliferative response to H. pylori was significantly blocked when hFGOs were pretreated with the c-Met inhibitor (c-MetI PF04217903 mesylate) (Fig. 6A, B). As detailed for the mouse-derived fundic gastric organoids, lysates were prepared from uninfected and hFGOs infected with either H. pylori (G27 strain) or ∆CagA and immunoprecipitated using an anti-c-Met antibody. Immunoprecipitates analyzed by western blot using an anti-phosphotyrosine antibody showed an increase in phosphorylated c-Met in response to H. pylori (Fig. 6C). CagA, CD44 and HGF co-immunoprecipitated with c-Met. It is known that c-Met is an important partner for CD44 in proliferation, but this is the first report of the CD44/c-Met association occuring in response to H. pylori infection of the human gastric epithelium.
To test the function of CD44v6 in hFGOs, we used a neutralizing antibody that specifically targets human CD44v6. We found H. pylori-induced proliferation was blocked in hFGOs pre-treated with the CD44v6 neutralizing antibody for one hour prior to H. pylori injection (Fig. 6D, E). Treating hFGOs with the CD44v6 neutralizing antibody alone did not significantly change the baseline of proliferation seen in the control hFGOs (Fig. 6D, E). Consistent with data shown in Fig. 6C, immunoprecipitates analyzed by western blot showed an increase in phosphorylated c-Met in response to H. pylori and CagA, CD44 and HGF co-immunoprecipitated with c-Met (Fig. 6F). Interestingly, inhibition of CD44 binding to hyaluronic acid using the CD44v6 peptide inhibited the co-immunoprecipitation of CD44 and HGF with c-Met (Fig. 6F). These data indicate that CD44v6 plays a functional role in H. pylori-induced proliferation in hFGOs, and that c-Met may collaborate with CD44 in this proliferative response to H. pylori infection in human tissue.
We next investigated the role of CD44 in cancer progression using a well-established Mongolian gerbil model of gastric cancer [35,36]. We administered an inhibitory peptide, Pep1, that prevents binding of CD44 ligand hyaluronic acid (HA), thus blocking CD44 downstream signaling. Gerbils were infected with H. pylori strain 7.13, an in vivo adapted strain originally isolated from a gastric ulcer [37] and reported to reproducibly induce gastric cancer in Mongolian gerbils [13,37]. We found that gerbils infected with H. pylori strain 7.13, developed atrophic gastritis (Fig. 7B), as documented by neutrophil (Fig. 7E) and lymphocyte (Fig. 7F) infiltration, development of lymphoid follicles (Fig. 7G) and atrophy (Fig. 7H), 6 weeks of infection, compared to control (Brucella broth administered) gerbils (Fig. 7A, E-H). Gerbils that received Pep1 injections three times a week for the duration of the 6 week infection with H. pylori, did not exhibit the development of atrophic gastritis (Fig. 7D, E-H), compared to the gerbils who received H. pylori and the scrambled control peptide (cPep1) (Fig. 7C, E-H).
Our studies show that CD44 plays a functional role in H. pylori-induced proliferation of mouse and human epithelial tissues in vitro. Thus, we next examined the proliferative state in response to H. pylori and Pep1 treatment in the Mongolian gerbils in vivo. Within 6 weeks of infection with H. pylori (7.13) there was a significant induction in the number of proliferating cells in the gastric epithelium (Fig. 8B, E) when compared to Brucella broth controls (Fig. 8A, E). To determine if CD44 signaling was required for this H. pylori induced proliferative response, we treated the H. pylori infected gerbils three times a week with Pep1, and found that the induction in proliferation was blocked (Fig. 8C, E). The gerbils infected with H. pylori and treated with the scrambled control peptide (cPep1) exhibited the expected induction in proliferation (Fig. 8D, E). Collectively, these data show that CD44 signaling mediates H. pylori-induced atrophic gastritis and hyperproliferation in the Mongolian gerbil model of gastric cancer.
Prolonged cell proliferation in the gastric mucosa is a precursor to the progression from chronic inflammation to gastric cancer in response to H. pylori infection [2]. However, the mechanism by which H. pylori induces epithelial cell proliferation is not well defined. Here we report the culture of primary human- and mouse-derived gastric epithelial cells as 3-dimensional structures called gastrointestinal organoids for the study of H. pylori pathogenesis. We utilized mouse-derived fundic gastric organoids (mFGOs), and for the first time we demonstrated that CD44 and c-Met form a complex with virulence factor CagA in response to H. pylori infection. To determine if this mechanism was present in human gastric epithelium, we developed a novel protocol for the culture of differentiated human-derived fundic gastric organoids in culture (hFGOs). In the hFGOs, we found that inhibiting CD44 splice variant 6, a specific marker for early invasive carcinoma [38], blocked H. pylori-induced proliferation. As in the mouse organoids, H. pylori infection triggered the formation of a complex containing CD44, c-Met, and CagA in the hFGOs. In addition, in the well-established Mongolian gerbil model of gastric cancer, animals treated with CD44 peptide inhibitor Pep1, resulted in the inhibition of H. pylori-induced proliferation and associated atrophic gastritis.
Our experiments report for the first time that CD44 acts as a mediator of H. pylori-induced proliferation both in vitro and in vivo. In vivo, we observed that CD44 regulates baseline proliferation whereby, as previously reported [39], basal rate of proliferation was approximately half that of control mice. However, H. pylori infection did not induce epithelial cell proliferation within the CD44KO mice, suggesting a functional role of CD44 as a mediator of bacterial-induced proliferation. Consistent with previous findings in gastric cancer cell lines [10], we also report that the proliferative response to H. pylori was CagA- and c-Met-dependent within the gastric epithelium. Western blot analysis using lysates collected from H. pylori infected mFGOs and hFGOs also showed that CagA coimmunoprecipitates with c-Met, as previously shown in human gastric cancer cells [19]. Although we report that H. pylori-induced proliferation was dependent on CD44 expression and signaling, the CagA/c-Met association was also observed in the absence of CD44 expression and signaling. Suzuki et al. [19] demonstrated that CagA CM motifs interact with Met leading to sustained PI3K-AKT signaling in response to H. pylori, leading to β-catenin activation and cellular proliferation. We advance our understanding of H. pylori-induced epithelial cell proliferation by demonstrating that CD44 acts as a coreceptor for c-Met in response to bacterial infection. The isoform containing exon v6 (CD44v6) acts as the coreceptor for c-Met in tumor cell lines [20,21]. The coreceptor function of CD44v6 for c-Met is of particular interest given that studies pinpoint CD44v6 as a marker of early invasive intramucosal gastric carcinoma [18]. Whether CD44v6 acted as a coreceptor for the function of c-Met in response to H. pylori infection was unknown. However, we found that H. pylori-induced proliferation was blocked in hFGOs pre-treated with the CD44v6 neutralizing antibody, and thus indicating that CD44v6 plays a functional role in H. pylori-induced proliferation in the human gastric epithelium.
Our data suggest a coreceptor function of CD44 for c-Met in response to H. pylori infection. The coreceptor function of CD44v6 for c-Met is shown to be dependent on HGF binding [20]. Indeed, we find that in the absence of CD44 expression and signaling HGF does not coimmunoprecipiate with c-Met and H. pylori fails to induce c-Met phosphorylation. In support of our findings, HGF binding to c-MET results in receptor homodimerization and phosphorylation of two tyrosine residues (Y1234 and Y1235) located within the catalytic loop of the tyrosine kinase domain [40], and subsequent, phosporylation of tyrosines 1349 and 1356 in the carboxy-terminal tail [41]. When these tyrosines become phosphorylated, they recruit signaling effectors that includes phosphatidylinositol 3-kinase (PI3K). When Y1313 is phosphorylated, it binds and activates PI3K, which promotes cell viability, motility and proliferation [42]. The extracellular domain of CD44v6 is necessary for c-Met activation, and this is dependent on HGF binding [20]. Indeed in the mFGOs, in response to H. pylori infection, HGF also coimmunoprecipitated with c-Met. The collaboration between c-Met and CD44v6 contributes to another bacterial infection, for example the invasion of Listeria monocytogenes into target cells [43]. Based on evidence demonstrating a functional role of CD44 in bacterial infection [43] including H. pylori infection, we conclude that the activation of c-Met is not only dependent on binding of H. pylori but in addition requires adhesion molecule CD44v6 as a co-receptor. In addition, although the coreceptor function of CD44v6 for c-Met is shown to be dependent on HGF binding, we show here for the first time that this signaling pathway mediates H. pylori-induced epithelial cell proliferation.
In the well-established Mongolian gerbil model of gastric cancer [35,36], animals treated with CD44 peptide inhibitor Pep1, resulted in the inhibition of H. pylori-induced proliferation and associated atrophic gastritis. Consistent with previous findings, infection with H. pylori for 6 weeks induced a significant inflammatory response, accompanied with the development of atrophic gastritis and metaplasia [35,44]. Using Pep1, a peptide that blocks the interaction between hyaluronic acid and CD44, we found that H. pylori-induced atrophic gastritis was inhibited. Importantly, infected gerbils that received Pep1 exhibited a significant reduction in actively proliferating cells. In support of our studies, a previous report by Khurana et al. [39] found that CD44 is a key coordinator of cell proliferation in a model of chemically-induced parietal cell atrophy, through downstream activation of STAT3. These findings strongly suggest that future therapeutic targets could include CD44 inhibition, to prevent H. pylori-induced hyperproliferation and cancer progression.
Our knowledge of H. pylori pathogenesis is predominantly based on data generated from gastric cancer cell lines or in vivo animal models of inflammation. Thus, limitation in acquiring such knowledge has been attributed to the inability to evaluate molecular mechanisms of bacterial and host cell interactions in a setting of a sustained gastric epithelial cell diversity and polarity. Our current work reports the development and use of a model of primary human and mouse cultured gastric epithelial cells. These models recapitulate key features of the gastric environment including the presence of the major gastric cell lineages and a polarized epithelium. In particular, we and others [25] report that the hFGO culture system represents a vital new technique for modeling H. pylori infection within the normal human tissue in vitro. As reported in similar mouse- and human-derived gastric organoid cultures [22,24,29], our study takes advantage of the presence of a defined lumen in these models, allowing us to inject live H. pylori directly into the gastric organoid and assay the epithelial response without the influence of host recruited factors. This contribution is significant because it provides knowledge required to potentially develop techniques to disrupt bacterial colonization and prevent disease progression. Our study helps to uncover a potential mechanism of H. pylori-induced proliferation in rodent and human tissue, using both in vivo established models of H. pylori infection and gastric cancer, as well as a novel epithelial cell culture system.
All mouse studies were approved by the University of Cincinnati Institutional Animal Care and Use Committee (IACUC) that maintains an American Association of Assessment and Accreditation of Laboratory Animal Care (AAALAC) facility. Human fundus was collected during sleeve gastrectomies (IRB protocol number: 2013–2251).
Helicobacter pylori (H. pylori) strain LSH100, a descendant of the clinical isolate G27 [30], G27 wild type [31] (CagA+), a mutant G27 strain bearing a cagA deletion (ΔcagA::cat) [45] and H. pylori 7.13 strain were grown on blood agar plates containing Columbia Agar Base (Fisher Scientific), 5% horse blood (Colorado Serum Company), 5 µg/ml vancomycin and 10 µg/ml trimethoprim as previously described [46]. Plates were incubated for 2–3 days at 37°C in a humidified microaerophilic chamber [47].
Bacteria were harvested and resuspended in filtered Brucella broth (BD biosciences) supplemented with 10% fetal calf serum. After 12 hours of growth at 37°C in a humidified microaerophilic chamber, bacteria were harvested, resuspended in filtered Brucella broth and C57/BL6 mice (The Jackson Laboratory, stock number: 000664), B6.129(Cg)-Cd44tm1Hbg/J (CD44-defiecient) mice (The Jackson Laboratory, stock number stock number: 005085) or Mongolian gerbils (Charles River) were inoculated by oral gavage with 108 bacteria per 200 µl of Brucella broth. Mice were infected with either G27 wild type or a mutant G27 strain bearing a cagA deletion (ΔcagA::cat). Mongolian gerbils were inoculated with H. pylori strain 7.13. Uninfected control mice were administered 200 µl of Brucella broth. Mice were analyzed 4 weeks post-inoculation. Mongolian gerbils were analyzed 6 weeks post-inoculation.
H. pylori colonization was quantified using the culture method previously published [47]. Briefly, the wet weight of gastric tissue collected from uninfected and infected animals was measured. Tissue was homogenized in 1 ml saline and diluted 1/100 and spread on blood agar plates containing Campylobacter Base Agar (Fischer Scientific), 5% horse blood (BD Diagnostic Systems), 5µg/ml vancomycin and 10µg/ml trimethoprim. Plates were incubated for 7–10 days at 37oC in a humidified microaerophilic chamber. Single colonies from these plates tested positive for urease (BD Diagnostic Systems), catalase (using 3% H2O2) and oxidase (DrySlide, BD Diagnostic Systems). Colonies were counted and data normalized using the tissue wet weight and expressed and colony forming units (CFU)/g tissue.
Specific hyaluronan (HA) blocking peptide Pep 1 (NH2-GAHHWQFNALTVRGGGS-CONH2) and scrambled control (NH2-WRHGFALTAVNQGGGS-CONH2) [48] peptides were synthesized by Pierce Biotechnology (Thermo Scientific-3747, Rockford, IL). Pep1 and control peptides were administered by intraperitoneal injection to Mongolian gerbils at a concentration of 10 mg/kg of body weight three times a week for 6 weeks after H. pylori infection.
Mongolian Gerbils were euthanized 6 weeks post-infection, stomachs were divided longitudinally for representative sections from both sides of the tissue and stained for Hematoxylin & Eosin (H&E). H&E stains were analyzed by using the updated Sydney classification system for histological scoring of gastritis [49].
Mice and Mongolian gerbils were injected with BrdU (300 mg/kg) 24 hours prior to analysis. Stomach sections spanning both the fundic and antral regions collected from experimental animals were fixed for 16 hours in Carnoy’s Fixative, paraffin embedded and sectioned at 5 µM. Prepared slides were deparaffinized with antigen retrieval performed by submerging in boiling solution (1:100 dilution Antigen Unmasking Solution in dH2O, Vector Laboratories, H-3300) for 10 minutes followed by 20 minutes at room temperature. Sections were then blocked with 5% BSA/PBS for 20 minutes at room temperature. BrdU color development was performed according to manufacturer’s protocol (Roche, Cat. No. 11 296 736 001). Immunohistochemical slides were dehydrated and mounted using Permount and images viewed and captured under light microscopy (Olympus BX60 with Diagnostic Instruments “Spot” Camera).
Gastric fundic organoids were prepared based on our recently reported protocol [22,50]. Primary epithelial cells from adult stomach tissue were cultured as 3-dimensional structures called mouse-derived fundic gastric organoids (mFGOs). Stomachs were dissected from mice along the greater curvature and washed in ice-cold Ca2+/Mg2+-free Dulbecco’s Phosphate Buffered Saline (DPBS). The stomach was stripped from muscle and visible blood vessels. Gastric fundus was further separated and cut into approximately 5 mm2 pieces. Tissue was then incubated in 5 ml of 5 mM EDTA for 2 hours at 4°C with gentle shaking. The EDTA was replaced with 5 ml chelation buffer (1 g D-sorbital, 1.47 g sucrose in 100 ml DPBS). Next tissue was shaken vigorously for approximately 2 minutes to dissociate glands. Dissociated glands were centrifuged at 150 g for 5 minutes then were embedded in Matrigel (BD Biosciences) supplemented with Advanced DMEM/F12 medium (Invitrogen), Wnt conditioned medium [22], R-spondin conditioned medium [22] supplemented with gastric growth factors including bone morphogenetic protein inhibitor, Noggin (PeproTech), Gastrin (Sigma), Epidermal grow factor (EGF, PeproTech) and Fibroblast growth factor 10 (FGF-10, PeproTech) as previously described [22,50]. Glands grew into organoids by 1–2 days. After 4 days in culture mFGOs were cultured on the inner well of polyester Transwell inserts (0.4 µm pore size, catalogue number 3460, Corning Lifescience), while immortalized stomach mesenchymal cells (ISMCs) were cultured in the base of the chamber of the Transwell according to our published protocol [22]. Organoids were co-cultured for a further 3 days prior to H. pylori infection and treatments.
Human fundic gastric organoids (hFGOs) were generated independently of the recently reported protocol [25]. The fundic mucosa was stripped away from the muscle layer, and then cut into 5 mm2 pieces and washed 3 times in sterile DPBS without Ca2+ and Mg2+. The mucosa was transferred to DMEM/F12 (catalogue number 1263–010, Gibco Life Technologies) supplemented with 10mM HEPES, 1%Penicillin/Streptomycin and 1X Glutamax, and incubated while stirring and oxygenated in a 37oC water bath with Collagenase (from Clostridium histolyticum, Sigma C9891, 1 mg/ml) and bovine serum albumin (2 mg/ml) to release glands from the tissue. After 15–30 minutes of incubation collected glands were washed in sterile phosphate buffered saline with Kanamycin (50 mg/ml) and Amphotericin B (0.25 mg/ml)/Gentamicin (10 mg/ml), centrifuged at 200 xg, resuspended in the appropriate volume of Matrigel (50 µl of Matrigel/well), and subsequently cultured in human gastric organoid media (DMEM/F12 supplemented with 10mM HEPES, 1X Glutamax, 1% Pen/Strep, 1X N2, 1X B27, 1mM N-Acetylcystine, 10mM Nicotidamide, 50ng/mL EGF, 100ng/mL Noggin, 20% R-Spondin Conditioned Media, 50% Wnt Conditioned Media, 200ng/mL FGF10, 1nM Gastrin, 10uM Y-27632, Kanamycin (50 mg/ml) and Amphotericin B (0.25 mg/ml)/Gentamicin (10 mg/ml)) (Table 1). Glands grew into organoids by 7 days at which time hFGOs were H. pylori infected and treated. We did not observe significant variations in organoid growth between donor gastric glands.
H. pylori strains G27 and ΔCagA were grown on blood agar plates as described above, and prior to injection, a group of organoids were pre-treated for one hour with a highly selective, high affinity c-Met inhibitor (100 µg/ml, PF04217903 mesylate, Tocris Bioscience Cat.# 4239). CD44-deficient mouse-derived organoids were treated with Wnt agonist (1µM CALBIOCHEM, catalogue number 681665). The hFGOs were pretreated with CD44v6 antibody 1 hour prior to H. pylori infection at a concentration of 100 ng/ml. Organoids (mFGOs and hFGOs) cultured for 7 days were injected with 200 nl of Brucella broth using a Nanoject II (Drummond) microinjector, such that each organoid received approximately 2x105 bacteria. Twenty-four hours after injection, organoids were harvested by washing in ice-cold DPBS to remove Matrigel, followed by either RNA isolation by TRIzol, protein isolation by M-PER Mammalian Protein Extraction Reagent (Thermo Scientific, IL) or EdU labeling.
The EdU solution was added to the organoid medium of either mFGOs or hFGOs for uptake for 1 hour. EdU staining was performed using the Click-iT Alexa Fluor 594 Imaging Kit, according to the manufacturer’s instructions (Life Technologies). The mFGOs were fixed with 4% formaldehyde for 20 minutes, followed by permeabilization with 0.5% Triton X-100 in DPBS for 20 minutes at room temperature. Blocking was done with 2% normal goat serum for 20 minutes at room temperature. The organoids were then incubated at 4oC overnight with an antibody specific for E-cadherin (Santa Cruz Biotechnology sc-59778, 1:100). Following this, organoids were incubated with Alexa Fluor 488 again overnight at 4oC. After incubation with the secondary antibodies organoids were counterstained using nuclear stain (Hoechst 33342, 10 µg/ml, Invitrogen) for 20 minutes at room temperature. Organoids were visualized using the Zeiss LSM710.
Both mFGOs and hFGOs were harvested from Matrigel using ice-cold DPBS without Ca2+/Mg2+ and lysed in M-PER Mammalian Protein Extraction Reagent (Thermo Scientific, IL) supplemented with protease inhibitors (Roche) according to the manufacturer’s protocol. For the immunoprecipitation, c-Met antibody (AbCam ab59884) was added to the cell lysates (1:50 dilution) overnight at 4oC. Protein A/G Plus Agarose beads (Santa Cruz sc-2003) were washed with PSB and added to the cell lysate mixture overnight at 4oC. Cell lysate mixtures were resuspended in 40 µl Laemmli Loading Buffer containing eta-mercaptoethanol (Bio-Rad Laboratories, CA) before western blot analysis. Samples were loaded onto 4–12% Tris-Glycine Gradient Gels (Invitrogen) and run at 80 V, 3.5 hours before transfer to nitrocellulose membranes (Whatman Protran, 0.45 µM) at 105 V, 1.5 hours at 4oC. Membranes were blocked for 1 hour at room temperature using KPL Detector Block Solution (Kirkegaard & Perry Laboratories, Inc.). Membranes were incubated for 16 hours at 4°C with a 1:2000 dilution of either anti-GAPDH (Millipore, MAB374), 1:100 dilution of anti-phospho Tyrosine (Santa Cruz sc-7020), anti CagA (Abcam ab90490), anti-CD44 (Abcam ab51037), anti c-Met (Abcam ab59884), or anti-HGF (Abcam ab83760) followed by a 1 hour incubation with a 1:1000 dilution anti-mouse or anti-rabbit Alexa Fluor 680 (Invitrogen). Blots were imaged using a scanning densitometer along with analysis software (Odyssey Infrared Imaging Software System).
Total RNA was isolated from mFGOs, hFGOs and gastric glands using TRIzol (Life Technologies) according to manufacture’s protocol. A High Capacity cDNA Reverse Transcription Kit synthesized cDNA from 100 ng of RNA following protocol provided by Applied Biosystems. Real-time PCR assays were utilized for the following genes in the mouse-derived organoids: GAPDH, alpha-Smooth Muscle Actin (Mm00725412_s1), Zeb1 (Mm00495564_m1) and 2 (Mm0049713_m1), TWIST 1 (Mm04208233_g1) and 2 (Mm000495564_m1), SNAIL 1 (Mm01249564_g1) and 2 (Mm00441531_m1). Cell lineage markers were determined by RT-PCR for HK-ATPase (Hs01026288_m1), gastrin (Hs01107047_m1), Muc5ac (Hs00873651_mH), and Muc6 (Hs01674026_g1). PCR amplifications were done with pre-validated 20X TaqMan Expression Assay primers, 2X TaqMan Universal Master Mix (Applied Biosystems), and cDNA template, in a total volume of 20 µL. Amplifications were performed with duplicate wells in a StepOne Real-Time PCR System (Applied Biosytems), and fold change was calculated at (Ct-Ct high) = n target, 2ntarget/2nGAPDH = fold change where Ct = threshold cycle.
The hFGOs were washed with cold DPBS without Ca2+/Mg2+ to remove Matrigel, and suspended in 2 ml Accutase (Innovative Cell Technologies, Cat.# AT-104) for 5 minutes at 37oC. Organoids were broken into single cells using an 18G needle, passed through 20 times. Single cells were fixed and permeablized using the FIX&PERM Kit from Invitrogen (catalogue number GAS004), according to manufacturer’s instructions. A first tube of cells was co-stained with lectin FITC labeled Ulex europaeus (UEAI, Sigma Aldrich), lectin Griffonia simplicifolia Alexa Fluor 647 (GSII, Molecular Probes) and then rabbit anti-pepsinogen C (Abcam, ab104595) followed by an anti-rabbit IgG PE secondary antibody (Abcam, ab7070), all at a 1:100 dilution. A second tube of cells was co-stained for chromogranin A antibody (Abcam, ab15160) followed by an anti-rabbit IgG PE conjugated secondary antibody, and anti-HK-ATPase (MA3–923, Affinity Bioreagents) followed by an anti-mouse IgG FITC conjugated (Abcam, ab6785), all at a 1:100 dilution. The antibody incubations were done for a period of 20 minutes at room temperature. The stained cells were analyzed using the FACSCalibur flow cytometer (BD Biosciences) followed by the FloJo software (Tree Star, Ashland, OR).
The significance of the results was tested by two-way ANOVA using commercially available software (GraphPad Prism, GraphPad Software, San Diego, CA). A P value <0.05 was considered significant.
|
10.1371/journal.pbio.1000594 | Break-Induced Replication Is Highly Inaccurate | DNA must be synthesized for purposes of genome duplication and DNA repair. While the former is a highly accurate process, short-patch synthesis associated with repair of DNA damage is often error-prone. Break-induced replication (BIR) is a unique cellular process that mimics normal DNA replication in its processivity, rate, and capacity to duplicate hundreds of kilobases, but is initiated at double-strand breaks (DSBs) rather than at replication origins. Here we employed a series of frameshift reporters to measure mutagenesis associated with BIR in Saccharomyces cerevisiae. We demonstrate that BIR DNA synthesis is intrinsically inaccurate over the entire path of the replication fork, as the rate of frameshift mutagenesis during BIR is up to 2,800-fold higher than during normal replication. Importantly, this high rate of mutagenesis was observed not only close to the DSB where BIR is less stable, but also far from the DSB where the BIR replication fork is fast and stabilized. We established that polymerase proofreading and mismatch repair correct BIR errors. Also, dNTP levels were elevated during BIR, and this contributed to BIR-related mutagenesis. We propose that a high level of DNA polymerase errors that is not fully compensated by error-correction mechanisms is largely responsible for mutagenesis during BIR, with Pol δ generating many of the mutagenic errors. We further postulate that activation of BIR in eukaryotic cells may significantly contribute to accumulation of mutations that fuel cancer and evolution.
| Accurate transmission of genetic information requires the precise replication of parental DNA. Mutations (which can be beneficial or deleterious) arise from errors that remain uncorrected. DNA replication occurs during S-phase of the cell cycle and is extremely accurate due to highly selective DNA polymerases coupled with effective error-correction mechanisms. In contrast, DNA synthesis associated with short-patch DNA repair is often error-prone. Break-induced replication (BIR) presents an interesting case of large-scale DNA duplication that occurs in the context of DNA repair. In this study we employed a yeast-based system to investigate the level of mutagenesis associated with BIR compared to mutagenesis during normal DNA replication. We report that frameshifts, which are the most deleterious kind of point mutation, are much more frequent during BIR than during normal DNA replication. Surprisingly, we observed that the majority of mutations associated with BIR were created by polymerases responsible for normal DNA replication, which are assumed to be highly precise. Overall, we propose that BIR is a novel source of mutagenesis that may contribute to disease genesis and evolution.
| Genetic information is preserved through generations by chromosome duplication during S-phase DNA replication, which is highly accurate due to the fidelity of replicative polymerases and efficient elimination of replication errors by polymerase-coupled proofreading activity and post-replicative mismatch repair (MMR). Aside from scheduled DNA replication during S-phase, DNA synthesis is also a part of various types of DNA repair, such as nucleotide-excision repair, base-excision repair, and double-strand break (DSB) repair. It has been shown that short-patch synthesis associated with repair of various kinds of DNA damage is highly error-prone [1]–[5], making these events important contributors to a cell's overall mutation rate.
DSBs as a source of hypermutability have been documented for several repair events, including gene conversion (GC) and single-strand annealing in vegetative cells [4]–[9], and DSB repair in meiosis and non-dividing cells [10],[11]. Also, increased mutability has been associated with senescence in telomerase-deficient cells [12], where shortened chromosome ends behave similarly to DSB ends. At least two mechanisms were demonstrated to contribute to DSB-induced mutagenesis. First, unrepaired lesions accumulated in tracts of single-stranded DNA that form after a DSB result in error-prone restoration of the duplex molecule [9]. A similar pathway was shown to be responsible for hypermutagenesis associated with recovery of dysfunctional telomeres [9]. Second, it has been demonstrated that copying of a donor sequence associated with GC is mutagenic [5],[13],[14], which could be explained by inefficient MMR during GC [5],[6], or by an unusual, conservative mode of synthesis that proceeds without formation of a replication fork [15].
This study was designed to determine the mutation rate associated with a unique cellular process, break-induced replication (BIR), which is a processive type of DNA replication that can duplicate large chromosomal regions comparable in size to replicons. In stark contrast to S-phase replication, BIR is initiated at a DSB site rather than at a replication origin. BIR proceeds by invasion of one DSB end into the homologous template, followed by initiation of DNA synthesis that can continue for hundreds of kilobases. A variety of repair processes is believed to proceed via BIR, including repair of collapsed replication forks and stabilization of uncapped telomeres. BIR can also repair DSBs produced such that either only one of the two free DNA ends can find homology for strand invasion or both ends can find homology but only in different areas of the genome (reviewed in [16],[17]). Notably, a significant fraction of DSB gap repair events also proceed through BIR [18]. The occurrence of BIR often leads to loss of heterozygosity (LOH), chromosomal translocations, and alternative telomere lengthening [19]–[21], which are genetic instabilities associated with cancer in humans.
Unlike other forms of DSB repair, BIR is believed to proceed in the context of a replication fork [21], and the establishment of the BIR fork requires almost all of the proteins required for initiation of normal replication [22]. However, several observations indicate that the BIR replication fork may differ from an S-phase replication fork in several important ways. For example, it has been shown that, in Saccharomyces cerevisiae, BIR requires Pol32p, a subunit of polymerase δ (Pol δ; [21],[23],[24]) that is dispensable for yeast S-phase DNA replication. Further, the roles of the main replicative polymerases may differ between BIR and S-phase replication. Thus, for BIR initiation, only α-primase and Pol δ are essential, while polymerase ε (Pol ε) is involved only in later steps of BIR, and up to 25% of BIR events can complete in the absence of Pol ε [21]. Also, BIR initiation is very slow (takes approximately 4 h [18],[19],[23]) and is associated with frequent template switching that subsides after the first 10 kb of synthesis [25], which led to speculation that there may be slow assembly of an unstable replication fork that shifts to a more stable version later in synthesis. Alternatively, initiation of BIR might be slow due to a “recombination execution checkpoint” that regulates the initiation of DNA synthesis during BIR [18]. All of these unique features of BIR led us to test whether it is more mutagenic than S-phase replication.
Here we demonstrate that DNA synthesis associated with BIR is highly error-prone, as the frequency of frameshift mutations associated with BIR is dramatically increased compared to normal DNA replication. Our results indicate that BIR mutagenesis results from several problems, including increased polymerase error rate and reduced efficiency of MMR.
To assay the accuracy of BIR, we used a modified version of our disomic experimental system in Saccharomyces cerevisiae (Figure 1A), wherein a galactose-inducible DSB is initiated at the MATa locus of the truncated, recipient copy of chromosome III, while the donor copy of chromosome III contains an uncleavable MATα-inc allele and serves as the template for DSB repair [23]. Elimination of all but 46 bp of homology on one side of the break on the recipient molecule via replacement with LEU2 and telomeric sequences results in efficient DSB repair through BIR in this strain (Figure 1B,C). Initiation of BIR in this system is preceded by extensive 5′-to-3′ resection of the GAL::HO-induced DSB at MATa, followed by strand invasion of the 3′ single-strand end into the donor chromosome at a position proximal to MATα-inc (Figure 1B; [26]). To study the accuracy of BIR, we chose to assay the level of frameshift mutagenesis using reversion frameshift reporters in our disomic strain. The frameshift reporters used allowed detection of mutations that occurred during BIR even in the presence of the original wild type gene (an essential feature because the wild type template allele remains after BIR repair) and also allowed us to investigate different aspects of BIR replication (similar to [27], see below). Frameshifts comprise a significant fraction (10%–20%) of all spontaneous mutations [28]–[30] and are the most deleterious type of point mutations, as they almost always eliminate gene function. In contrast, >90% of base substitutions are silent [31]. Notably, an increase in the rate of frameshifts typically correlates with an increase in base substitutions (reviewed in [32],[33]).
Three different frameshift reporters were employed: A4, A7, and A14 [27], which are all alleles of the LYS2 gene with an insertion of approximately 60 bp that includes a homonucleotide run of four adenines (A4), seven adenines (A7), or 14 adenines (A14) (Figure 1D). Insertion of any of the three alleles results in a “+1” shift in the reading frame and a Lys− phenotype, while a Lys+ phenotype is restored by a frameshift mutation that occurs in an approximately 71 bp region of the allele (that includes the inserted sequence) and restores the reading frame. A series of isogenic strains was created with insertion of the described reporter alleles into one of three positions on the donor (MATα-inc-containing) chromosome (Figure 1A): (1) at MATα-inc (“MAT”), (2) 16 kb centromere-distal from MATα-inc in the region between RSC6 and THR4 (“16 kb”), and (3) 36 kb centromere-distal to MATα-inc in the region between SED4 and ATG15 (“36 kb”). In all strains, LYS2 was fully deleted from its native location in chromosome II (see Materials and Methods for details).
BIR-associated mutagenesis was measured by plating appropriate dilutions of cell suspensions to obtain single colonies on rich media (YEPD) and lysine drop-out media after a 7 h incubation in liquid galactose-containing media. The majority of cells undergoing DSB repair remained in G2/M arrest for the duration of the experiment (Figure S3A and unpublished data), consistent with repair of most DSBs by BIR, which exhibits delayed initiation associated with a long G2/M checkpoint arrest [19]. Coherently, the majority of colonies grown with or without selection (on lysine omission media or YEPD, respectively) repaired the DSB by BIR and displayed either an Ade+Leu− or Ade+/−Leu− phenotype (see Materials and Methods and Tables S1 and S2 for details), which were previously confirmed to result from BIR repair of both or one of two sister chromatids, respectively [23]. BIR efficiency in wild type and mutant strains is shown in Table S2.
For all three reporters at all three locations, the rate of Lys+ frameshifts was much higher after DSB repair compared to the spontaneous Lys+ rate (Figure 2; Table S1; see Materials and Methods for details regarding rate calculations). Specifically, for all A4 and A7 strains, the rate of frameshift mutagenesis associated with DSB repair (7 h) exceeded the Lys+ reversion rate before the DSB (0 h) by 100- to 550-fold. Because most strains with a DSB site exhibited residual DSB formation even before addition of galactose (unpublished data), isogenic no-DSB controls were used to estimate more accurately the rate of spontaneous mutagenesis (see Materials and Methods for details). Using these no-DSB control strains lacking the HO cut site, we demonstrate a 780- to 2,800-fold increase in frameshift mutagenesis during BIR compared to spontaneous frameshift mutations. In all strains containing A14, in which spontaneous events were approximately 1,100- to 2,500-fold more frequent compared to A4, the rate of frameshift mutagenesis associated with DSB repair remained 25- to 300-fold higher than the rate of spontaneous events. Similar to unselected colonies, the majority of Lys+ DSB repair outcomes resulted from BIR (Table S1); thus, the substantial increase in frameshift mutagenesis observed in strains with DSBs compared to their no-DSB isogenic controls can be attributed to DNA synthesis during BIR. In control strains that contained the A4, A7, or A14 reporters in the native LYS2 position on chromosome II, no increase in the rates of Lys+ was observed after 7 h in galactose (Figure 2; Table S1), which confirmed that the increased frameshift mutagenesis was specific for the chromosome undergoing BIR.
Lys+ BIR outcomes were primarily 1 bp deletions, the majority of which (70%–100%) occurred in ≥2 homonucleotide runs (Table 1; Figure S1). With data for all strains combined, the majority of Lys+ mutations concentrated in two hotspots: (1) the poly-A run, which is known to provoke replication slippage, and (2) the sequence GGGCCAAGG (Figures 1D and S1; Table 1), which could also promote replication slippage within one of its small homonucleotide runs. Alternatively, the second hot spot could result from template switching involving the first seven nucleotides of this hotspot (GGGCCAA) and its −1 bp quasipalindromic copy (TTGCCC) located approximately 70 bp away (Figures 1D and S1; see Discussion for details). As expected, the proportion of 1 bp deletions in the poly-A run increased with the length of the run, with only 3%, 20%, and 0% of frameshifts occurring in the poly-A run of the A4 reporter at the MAT, 16, and 36 kb positions, respectively, and 100% of frameshifts occurring in the A14 run at all three positions (Table 1 and Figure S1). The proportion of frameshifts in the A7 run varied somewhat across reporter positions. The spectra of Lys+ frameshift mutations were generally similar for BIR-induced compared to spontaneous mutations for each given reporter (Table 1 and Figure S1). One exception was the increase in 2 bp insertions observed in A4 and A7 no-DSB control strains at the 16 kb position. Taken together, our data show that frameshift mutagenesis during BIR is increased 25- to 2,800-fold compared to spontaneous mutagenesis.
We hypothesized that involvement of translesion DNA synthesis during BIR, whether due to a defective replisome or DNA template damage (as discussed in [4],[9],[34],[35]), may contribute to the increased rate of BIR frameshift mutations. To address this, BIR-associated mutagenesis in A4 and A7 strains with deletion of RAD30 (encoding DNA polymerase η [Pol η]) or deletion of REV3 (encoding the catalytic subunit of DNA polymerase ζ [Pol ζ]) was measured at all three positions (Figure 3A,B; Table S1). While rad30Δ mutants showed no change in the rate of frameshift mutations compared to wild type at any position, deletion of REV3 did result in a small but statistically significant decrease (2×–3×) in mutations at MAT and for A4 at 16 kb. No change was observed in the other rev3Δ strains. (Importantly, BIR efficiency in rev3Δ mutants was similar to that observed in wild type [Table S2 and unpublished data]).
To differentiate the role of REV3 in BIR from its role in damage-induced mutagenesis, we exposed our rev3Δ no-DSB control strains containing A7 at the 36 kb position (where there was no effect of rev3Δ on BIR mutagenesis) to 20 J/m2 of UV light (Figure S2). This exposure resulted in an approximately 10-fold increase in Lys+ events compared to the frequency of spontaneous events. Consistent with the observation of Abdulovic and Jinks-Robertson [36], the UV-induced increase in mutagenesis was largely REV3-dependent in our system. Thus, we conclude that BIR-induced mutagenesis differs from UV-induced mutagenesis in its dependency on Pol ζ; while the latter strongly depends on Pol ζ, the former is only modestly dependent on Pol ζ and only at some chromosomal positions.
We eliminated MSH2 from all A4 and A7 strains to test whether MMR corrects frameshift errors made during BIR. In all cases, we observed a significant increase in frameshift mutagenesis during BIR in msh2Δ strains compared to their isogenic wild type strains, suggesting that MMR corrects a large number of BIR frameshift errors (Figure 4A,B; Table S1). The mutation rate observed during BIR in MMR-deficient mutants significantly exceeded the level of spontaneous mutagenesis observed in MMR-deficient no-DSB controls, confirming that MMR deficiency further increased the already high rate of BIR mutagenesis. Strains containing A7 reporters were more sensitive to MMR deficiency and showed higher increases in the rate of frameshifts compared to increases for the corresponding A4 strains. This effect is similar to the effect of msh2Δ during normal replication, where MMR is especially important to correct errors in long homonucleotide runs [27],[37],[38]. Also, consistent with Tran et al. [27], who reported a dramatic shift of spontaneous frameshifts to the poly-A run in MMR-deficient A7 strains, we observed a significantly higher percentage of mutations occurring in the poly-A run in MMR-deficient A7 strains at the MAT and 36 kb positions compared to isogenic MMR-proficient strains (Table 1; Figure S1). At the 16 kb position, where most events in the wild type A7 strain were in the poly-A run, we confirmed that MMR deficiency caused mutation events to shift to the poly-A run in the A4 strain. The ability of MMR to correct BIR errors was further supported by data from mlh1Δ mutants, which were tested at the MAT and 36 kb positions with the A4 reporter (Figure 4A; Table S1). Our data thus suggest that BIR occurs in the context of functional MMR machinery and that long homonucleotide runs are especially susceptible to failure of MMR, as is the case during normal DNA replication.
To better characterize the role of MMR in correction of BIR errors, we compared mutation rates in experimental MMR-deficient strains with their no-DSB controls. This comparison showed that, prior to MMR correction, the level of polymerase errors was significantly higher during BIR compared to normal DNA replication for all constructs (Table 2). Based on the percent of these errors that was repaired by MMR (calculated in Table 2), the efficiency of MMR in BIR was 98%, 97%, and 99.9% for A7 strains at the MAT, 16, and 36 kb positions, respectively, and approached 99.9% for all positions during normal DNA replication. MMR also repaired a high percentage of BIR errors in A4 strains (47%, 87%, and 91% at the MAT, 16, and 36 kb positions, respectively), but this was somewhat lower than the efficiency of MMR during normal DNA replication for these strains (94%, 99%, and 99.7% at the MAT, 16, and 36 kb positions, respectively). These data suggest that, although MMR operates during BIR, the percentage of MMR-repaired polymerase errors is often lower for BIR than for normal replication.
Our previous analysis indicated that BIR is associated with a significantly higher level of polymerase errors than normal replication (see above). To determine the role of proofreading activity during BIR, pol3-5DV, an exonuclease-deficient mutation, was introduced into A4 and A7 strains at all three chromosomal locations to eliminate the proofreading activity of Pol δ (Figure 4A,B; Table S1). Pol δ was chosen because, unlike Pol ε, it is required at all stages of BIR synthesis [21]. pol3-5DV strains consistently showed an increase in BIR-related mutagenesis above the already high mutagenesis observed in wild type strains (a 3- to 6-fold increase for A4 and a 2- to 9-fold increase for A7). Also, the frameshift mutation spectrum of Lys+ outcomes in pol3-5DV strains was similar to that in their respective wild type strains (unpublished data). These results indicate that the proofreading activity of Pol δ is capable of correcting polymerase errors made during BIR.
The role of proofreading activity can be accurately estimated only in the absence of MMR due to redundancy between the two activities. However, haploid pol3-5DV mutants are inviable in combination with full MMR deficiency; thus, we deleted MSH3 (which is known to result in a partial MMR defect) in pol3-5DV strains to better understand the effect of Pol δ proofreading activity. (The growth rate of pol3-5DVmsh3Δ double mutants was similar to wild type [Figure S3B,C] and its viability was not reduced following 7 h incubation in galactose [unpublished data]). We observed synergistic increases in BIR frameshift mutagenesis in pol3-5DVmsh3Δ double mutants compared to their respective single-mutant strains at all positions (Figure 4A,B; Table S1). However, the increase in mutagenesis that resulted from synergism between pol3-5DV and msh3Δ was generally higher for spontaneous events compared to BIR events, which may indicate decreased efficiency of Pol δ proofreading during BIR. Finally, because the increase in mutagenesis was observed in a mutant lacking the exonuclease activity of Pol δ many replication errors during BIR must be produced by Pol δ [39], although it cannot be excluded that DNA synthesis errors by other polymerases contribute as well [40],[41].
Previously, we established that BIR proceeds under conditions of G2/M cell-cycle arrest resulting from the DNA damage checkpoint response [19],[23], and we hypothesized that cells with chromosome(s) undergoing BIR repair may induce ribonucleotide reductase (RNR) and dNTP levels in a manner similar to other damage-induced checkpoint responses ([42], and reviewed in [43]). To test this hypothesis, we employed the strain with the A4 reporter at the 16 kb position to measure dNTP pools before galactose induction of the DSB (0 h), and 3 and 6 h after galactose addition (Figure 5). (BIR-associated DNA synthesis is initiated approximately 4–6 h after galactose addition [19],[23].) Compared to pre-induction levels, a 2- to 3-fold increase in dNTP levels was evident at the 3 h time point, and a 3- to 6-fold increase was observed at the 6 h time point (Figure 5). No increase in dNTP levels was observed at either the 3 or 6 h time points in the isogenic no-DSB control strain. Also, we observed increased levels of the DNA-damage inducible RNR subunits Rnr2p and Rnr4p and decreased levels of RNR inhibitor Sml1p in cells undergoing BIR (Figure S4). In the absence of Dun1p, which is required for degradation of Sml1p and induction of RNR genes in other cases of DNA-damage response [44],[45], the increase in dNTP levels was nearly eliminated during BIR repair. A 1.6- to 2.1-fold reduction in dNTP levels was observed in the dun1Δ no-DSB control strain compared to the wild type no-DSB strain (Figure 5). Finally, deletion of SML1, which is known to increase basal dNTP levels [46], did not affect dNTP levels during BIR, but the sml1Δ mutation in no-DSB control strains did produce an expected increase in dNTP levels. These data are consistent with the roles of Dun1p and Sml1p in the regulation of dNTP levels in vegetative cells and demonstrate that BIR occurs in the context of DUN1-dependent up-regulation of RNR activity.
Increased dNTP levels are known to decrease the fidelity of DNA polymerases and are associated with increased mutation rates. To investigate the role of increased nucleotide pools in BIR-related mutagenesis, the level of frameshift mutagenesis was measured in dun1Δ and sml1Δ A4 strains (Figure 6; Table S1). BIR mutagenesis decreased by 4.0-, 2.4-, and 5.4-fold at the MAT, 16, and 36 kb positions, respectively, in dun1Δ compared to wild type. The efficiency of BIR in dun1Δ cells was slightly reduced (a 1.2-fold reduction at all positions; Table S2 and unpublished data) compared to wild type. This decrease in BIR efficiency results most likely from a checkpoint response deficit in dun1Δ, which may lead to premature recovery from G2 arrest of cells undergoing BIR repair and, therefore, to increased loss of the broken chromosome due to mis-segregation. To accommodate this, the data were re-calculated to determine the rate of Lys+ events per BIR event. These results confirm that the dun1Δ mutation reduced the rate of frameshift mutations by 3.3-, 2.0-, and 4.8-fold at the MAT, 16, and 36 kb positions, respectively. Conversely, sml1Δ mutants did not display any change in the rate of mutations associated with BIR at the 36 kb position but did show small, 1.4-, and 1.8-fold increases at MAT and 16 kb.
We propose that increased dNTP pools contribute to the high mutagenesis associated with BIR. However, the fact that mutagenesis during BIR remained approximately 100- to 500-fold higher than during normal DNA synthesis even in dun1Δ mutants suggests that a large portion of BIR-related mutations may be independent of dNTP levels (see Discussion). Importantly, we tested the effect of the pol3-5DV mutation on dNTP levels both during BIR and in no-DSB controls and confirmed that this mutation did not affect dNTP levels in either case (Figure 5), consistent with our prior conclusion that the observed effects of pol3-5DV on BIR-induced mutagenesis resulted directly from the proofreading defect.
The fidelity of DNA synthesis differs among the various processes in which it is involved. While replication accomplishing genomic duplication is highly accurate, short-patch synthesis associated with repair of DNA damage, such as repair of DSBs by GC, is error-prone. In this study, we demonstrate that BIR, which can duplicate replicon-sized regions of chromosomes and is believed to proceed in the context of a replication fork [21], is associated with frameshift mutation rate increases up to 2,800-fold compared to spontaneous events. BIR-related hypermutability persisted at sites 16 and 36 kb distal to the DSB, differentiating this mechanism from the “template-switching” phenomenon discovered by Smith et al. [25] that was detected only within the first 10 kb of the 97 kb template that was copied.
Overall, the frameshift spectrum observed in lys2::Ins reporters during BIR was similar to the spectrum of spontaneous frameshifts for the same reporters/positions, with the majority of events occurring in di-, tri-, or poly-nucleotide runs. Also, BIR did not cause large deletions such as those increased in the pol3-t mutant, where they were explained by template switching between direct repeats stimulated by formation of extensive regions of ssDNA [47]. For both BIR-related and spontaneous events, two locations were especially susceptible to frameshift mutations. First, as previously described [27], the poly-A run induced frameshift mutations in a length-dependent manner. A second hotspot, the sequence GGGCCAAGG, may promote frameshift errors due to its multiple polynucleotide repeats, or as a result of template switching between this sequence and its partial −1 bp quasipalindromic copy (TTGCCC) located approximately 70 bp away. The mechanism of brief template switching of the nascent DNA strand to a nearly identical, inverted sequence was first proposed by Lynn Ripley [48] and later reported by Strathern et al. [4] and Hicks et al. [5], who observed alteration of nearly palindromic sequences into perfect palindromes during DSB repair by GC. However, in their systems, the inverted repeats were in much closer proximity than in our strains. The overall similarity between spontaneous and BIR-related frameshift mutation spectra observed in our assay could reflect that the nature of mutations is similar for both BIR- and S-phase replication. A second interesting possibility is that inter-sister BIR contributes to spontaneous events.
Cells undergoing BIR repair arrest as a part of the G2/M DNA damage checkpoint response [19]. Here, we observed that this checkpoint response also leads to an increase of Rnr2p and Rnr4p, a decrease of Sml1p, and increased dNTP pools, which require the checkpoint kinase Dun1p. Dun1p is a downstream target of the Mec1p and Rad53p checkpoint pathway that activates RNR by multiple mechanisms (reviewed in [49]). Because induction of RNR is considered the last step in the checkpoint cascade, our data suggest that a single DSB undergoing BIR repair triggers a complete checkpoint response (which includes both cell cycle arrest and RNR induction), which differentiates this process from the truncated checkpoint response observed, for example, in yeast undergoing DSB repair in G1 [50].
The significant decrease in BIR mutagenesis observed in dun1Δ suggests that increased dNTPs contribute to BIR-induced mutagenesis. Nevertheless, even in dun1Δ, BIR mutagenesis remained at least 100-fold higher than the spontaneous level of mutations, which indicates that elevated dNTP pools alone cannot explain the decrease in replication fidelity. These findings are consistent with previous work that has shown elevated dNTP levels to be mildly mutagenic in the presence of MMR [42],[51], and we propose that the role of elevated dNTP pools in our system is to further increase the number of errors made by an already error-prone fork, as discussed below. As expected, sml1Δ increased basal levels of dNTPs at 0 h but did not affect post-DSB increases because, in wild-type cells, Sml1p is degraded during BIR (Figure S4). Consistently, sml1Δ did not change the level of BIR mutagenesis at 36 kb; however, sml1Δ resulted in small but significant increases in BIR mutagenesis at the MAT and 16 kb positions. The reason for these increases is unclear, but exposure of the cell to chronically elevated levels of dNTPs may play a role.
Mutations arise from two sources: uncorrected replication errors left by a replication fork copying an undamaged template and error-prone copying of damaged DNA by a translesion polymerase. We investigated the role of translesion-synthesis polymerases Pol η and Pol ζ in BIR mutagenesis and determined that hypermutability during BIR is independent of Pol η, while modestly dependent on Pol ζ at some chromosomal positions. The activity of Pol ζ is known to be highly mutagenic in yeast, with the majority of damage-induced and over half of spontaneous mutations ascribed to Pol ζ, whereas lesion bypass by Pol η can be error-free or error-prone depending on the type of lesion and experimental assay employed (reviewed in [34],[52]). Here we observed no effect of rad30Δ on BIR-related mutagenesis. At the 36 kb position, where BIR is fast and stable, BIR-mutagenesis was also REV3-independent. In contrast, UV-induced mutagenesis at the 36 kb position was largely REV3-dependent (consistent with data in [36]), emphasizing the difference in the role of Pol ζ in BIR versus damage-induced mutagenesis. Interestingly, a small but significant reduction in BIR mutations occurred in rev3Δ mutants at MAT. One possible explanation is that the slow initiation of BIR in this region [18],[19] results in persistent ssDNA in the D-loop, which leads to higher mutagenesis, presumably by accumulating endogenous damage in ssDNA. Previously, increased spontaneous mutagenesis in regions of artificially created transient ssDNA at DSBs and uncapped telomeres was shown to significantly decrease in the absence of REV3 [9]. Pol ζ dependence was also observed at the 16 kb position in the A4 strain. This location may be more difficult for replication machinery to traverse, as evidenced by the overall increased rate of mutations (both spontaneous and BIR-related) at this position compared to others. Lack of Pol ζ dependence for the A7 construct at the 16 kb position could be explained by additional mutations in the poly-A run, which could be Pol ζ-independent. Overall, we conclude that BIR-associated frameshift mutagenesis is independent of Pol η, while modestly dependent on Pol ζ at some chromosomal positions.
We found that MMR operates during BIR but is often less efficient at correcting BIR-related versus spontaneous errors. This could indicate a decreased efficiency of MMR to correct any individual error made during BIR or that the amount of errors during BIR is sufficiently high to overwhelm MMR repair capabilities. Alternatively, it could indicate that BIR mutants result from both MMR-dependent and MMR-independent pathways, as has been proposed for spontaneous mutations [47]. This final possibility is supported by our observation of higher effects of msh2Δ in A7 versus A4 strains, because the increased number of errors in A7 results from replication slippage in the poly-A run, which is efficiently repaired by MMR [27]. The varying ratio of MMR-dependent to MMR-independent mutation events may explain the varying effect of msh2Δ across the three chromosomal locations on both BIR-associated and spontaneous mutagenesis (Table 2), as well as the context-dependence of MMR previously observed by Hawk et al. [53] for spontaneous mutagenesis. In pol3-5DV mutants, in which Pol δ proofreading was inactivated, we observed a further increase in the mutation rate compared to wild type, suggesting that proofreading activity operates during BIR. This result implicates Pol δ as one polymerase responsible for many BIR elongation errors [39], although other polymerases may contribute as well [40],[41]. The synergistic increase in BIR mutations observed in pol3-5DVmsh3Δ double mutants further supports the involvement of Pol δ proofreading during BIR. However, the efficiency of Pol δ proofreading of BIR errors appeared somewhat lower compared to S-phase replication. Furthermore, this synergism suggests that Pol δ introduces mutagenic errors during BIR replication associated with MMR, versus during other repair-related synthesis.
In summary, we propose that the high level of BIR-associated frameshift mutagenesis is due to uncorrected errors left by a mutagenic replication fork. Our data suggest that undamaged template DNA is copied by a BIR fork that contains multiple deficiencies, including decreased Pol δ replication fidelity in the presence of increased nucleotide pools and reduced MMR efficiency, which act synergistically to markedly increase frameshift mutagenesis. This proposed mechanism is generally similar to the mechanism recently suggested to generate mutations during GC repair [5]. What is unexpected is to observe such similar mutation mechanisms between GC, which proceeds through synthesis-dependent strand annealing that does not assemble a replication fork [15], and BIR, which proceeds in the context of a replication fork [21],[22].
BIR has been documented in both prokaryotes and eukaryotes and has been implicated in various processes of DNA metabolism. BIR is believed to restart collapsed replication forks, which occur even in healthy, dividing cells (reviewed in [17]), and it is also required for telomere maintenance in telomerase-deficient cells [21]. Also, a significant fraction of DSB gap repair events has been shown to proceed through BIR [18]. BIR leads to non-reciprocal translocations similar to those leading to cancer and other human diseases [54],[55]. It has been demonstrated that translocations mediated by BIR are often initiated by DSBs introduced near transposons or other DNA repeats that are present at multiple genomic locations [56]. A BIR-like repair pathway, microhomology-mediated BIR, was reported to generate copy number variations in eukaryotes, including those leading to human disease [57],[58]. Based on its widespread involvement in various processes, we propose that BIR may significantly contribute to the mutation rate and spectrum of many cell types, which is relevant to both disease development and selective adaptation. It may also provide an additional mechanism for so-called “mutation showers” reported to contribute to up to 1% of all mutations in the mouse genome [59]. BIR-associated mutagenesis may have an especially important role in tumorigenesis, because human cancer cells may both activate BIR at an elevated rate and be MMR-deficient (reviewed in [60]). Also, several human tumor-suppressor genes contain homonucleotide runs [61]–[64], which we demonstrated confer hypermutability in the context of BIR in MMR-deficient cells.
All yeast strains were isogenic to AM1003 [23], which is a chromosome III disome with the following genotype: hmlΔ::ADE1/hmlΔ::ADE3 MATa-LEU2-tel/MATα-inc hmrΔ::HYG FS2Δ::NAT/FS2 leu2/leu2-3,112 thr4 ura3-52 ade3::GAL::HO ade1 met13. In this strain, the HO endonuclease-induced DSBs introduced at MATa are predominantly repaired by BIR because the portion of the chromosome centromere-distal to MATa is truncated to leave only 46 bp of homology with the donor sequence [19],[23]. All strains used for measuring mutagenesis were constructed using PCR-based gene disruption and direct genome modification by oligonucleotides as described (see Text S1 for details) [65],[66]. All single-gene deletion mutants were constructed by transformation with a PCR-derived KAN-MX module flanked by terminal sequences homologous to the sequences flanking the open reading frame of each gene [67]. All constructs were confirmed by PCR and by phenotype. Proofreading-deficient mutant pol3-5DV was constructed as described [39] and confirmed by PCR followed by restriction analysis with HaeIII. Control no-DSB strains were obtained from each experimental strain by plating on YEP-Gal media, followed by selection of Ade+Leu+ colonies resulting from GC repair of the DSB at MATa.
Rich medium (yeast extract-peptone-dextrose [YEPD]) and synthetic complete medium, with bases and amino acids omitted as specified, were made as described [68]. YEP-lactate (YEP-Lac) and YEP-galactose (YEP-Gal) contained 1% yeast extract and 2% Bacto peptone media supplemented with 3.7% lactic acid (pH 5.5) or 2% (w/v) galactose, respectively. Cultures were grown at 30°C.
To determine mutation frequency, yeast strains were grown from individual colonies with agitation in liquid synthetic media lacking leucine for approximately 20 h, diluted 20-fold with fresh YEP-Lac, and grown to logarithmic phase for approximately 16 h. Next, 20% galactose was added to the culture to a final concentration of 2%, and cells were incubated with agitation for 7 h. No-DSB control strains were subjected to the same incubation and plating processes.
Samples from each culture were plated at appropriate concentrations on YEPD and lysine drop-out media before (0 h) and 7 h after the addition of galactose (7 h) to measure the frequency of Lys+ cells. Because spontaneous mutation frequencies were calculated based on the number of mutations accumulated during many cell generations, mutation rates were calculated for spontaneous and BIR mutagenesis using modifications of the Drake equation [69]. Specifically, the rate of spontaneous mutagenesis in experimental strains was calculated using mutation frequencies at 0 h in experimental and no-DSB control strains using the following formula: µ = 0.4343 f/log(Nµ), where µ = the rate of spontaneous mutagenesis, f = mutation frequency at time 0 h, and N = the number of cells in yeast culture at 0 h. Because most strains with a DSB site exhibited residual DSB formation even at 0 h, the rate of spontaneous mutagenesis was more accurately determined from 0 h Lys+ frequencies in no-DSB controls using the same formula. For the no-DSB controls with reporters at MAT, the median, calculated based on the equation shown above, was divided by 2 to correct for the presence of two LYS2 reporters in these strains. The rate of mutations after galactose treatment (µ7) was determined using a simplified version of the Drake equation: µ7 = (f7 − f0), where f7 and f0 are the mutation frequencies at times 7 h and 0 h, respectively. This modification was necessary because experimental strains did not divide between 0 h and 7 h, while no-DSB controls underwent ≤1 division between 0 h and 7 h.
Rates are reported as the median value (Figures 2–4,6), and the 95% confidence limits for the median are calculated for the strains with a minimum of six individual experiments as described and reported in Table S1 [70]. For strains with 4–5 individual experiments, the range of the median was calculated. Statistical comparisons between median mutation rates were performed using the Mann-Whitney U test [71].
BIR efficiency was estimated in all strains with a DSB site, typically in a subset of three experiments per strain. Colonies plated on YEPD 7 h after addition of galactose were replica plated onto omission media to examine the ADE1, ADE3, and LEU2 markers. Colonies formed by BIR displayed an Ade+Leu−, Ade+/−Leu−, or Ade+Leu+/− phenotype [23]. The efficiency of BIR in individual experiments was estimated as the sum of all Ade+Leu− events plus one half of all BIR sectored (Ade+/−Leu− or Ade+Leu+/−) events, divided by the total number of colonies analyzed. Typically, ≥50 colonies were analyzed for individual experiments. To compare wild type and mutant strains, BIR efficiency was determined by combining data from isogenic 16 and 36 kb A4 strains (strains with the reporter at MAT were omitted due to the effect of mating type on BIR efficiency [19]). Medians were compared using the Mann-Whitney U test [71].
A portion of the LYS2 gene was sequenced from independent Lys+ outcomes using one or both of the primers used to confirm insertion of the LYS2 reporters (see Text S1 for details). For experimental strains undergoing BIR repair, 7 h Lys+ BIR events (confirmed as Ade+Leu− on selective media) were sequenced. Because these strains did not divide between the 0 h and 7 h time points and the Lys+ frequency at 7 h significantly exceeded that at 0 h, all Lys+ events resulting from DSB repair were considered independent. In msh2Δ A7 experimental strains, in which the 0 h rate was extremely elevated, candidates for sequencing were chosen from experiments with a ≥10-fold difference in mutation frequencies between 0 h and 7 h. For no-DSB controls, independent Lys+ events were obtained by growing cultures from singles in YEPD overnight and choosing only one event from each culture.
The methods of measuring dNTPs in yeast are as described in (see Materials and Methods in the Supporting Information section for details) [51]. Results of three time course experiments performed for each strain were used to calculate the average ± standard deviation level of nucleotides.
|
10.1371/journal.ppat.1002279 | Critical Roles for LIGHT and Its Receptors in Generating T Cell-Mediated Immunity during Leishmania donovani Infection | LIGHT (TNFSF14) is a member of the TNF superfamily involved in inflammation and defence against infection. LIGHT signals via two cell-bound receptors; herpes virus entry mediator (HVEM) and lymphotoxin-beta receptor (LTβR). We found that LIGHT is critical for control of hepatic parasite growth in mice with visceral leishmaniasis (VL) caused by infection with the protozoan parasite Leishmania donovani. LIGHT-HVEM signalling is essential for early dendritic cell IL-12/IL-23p40 production, and the generation of IFNγ- and TNF-producing T cells that control hepatic infection. However, we also discovered that LIGHT-LTβR interactions suppress anti-parasitic immunity in the liver in the first 7 days of infection by mechanisms that restrict both CD4+ T cell function and TNF-dependent microbicidal mechanisms. Thus, we have identified distinct roles for LIGHT in infection, and show that manipulation of interactions between LIGHT and its receptors may be used for therapeutic advantage.
| Visceral leishmaniasis (VL) is a potentially fatal human disease caused by the intracellular protozoan parasites Leishmania donovani and L. infantum (chagasi). Parasites infect macrophages throughout the viscera, though the spleen and liver are the major sites of disease. VL is responsible for significant morbidity and mortality in the developing world, particularly in India, Sudan, Nepal, Bangladesh and Brazil. Because of the intrusive techniques required to analyse tissue in VL patients, our current understanding of the host immune response during VL largely derives from studies performed in genetically susceptible mice. We have discovered that mice which are unable to produce a cytokine called LIGHT have poor control of L. donovani infection in the liver, compared with wild-type control animals. In addition, we demonstrated that LIGHT has distinct roles during VL, depending on which of its two major cell-bound receptors it engages. Finally, we identified an antibody that stimulates the lymphotoxin β receptor (one of the LIGHT receptors), that can stimulate anti-parasitic activity during an established infection, thereby identifying this receptor as a therapeutic target during disease.
| Tumour necrosis factor (TNF) superfamily members are involved in many biological functions, including cell growth and differentiation, apoptosis and organogenesis [1]. This broad range of activities is achieved by TNF family members interacting with functional receptors associated with distinct cell signalling pathways [2]. TNF, lymphotoxin (LT)α, LTβ and LIGHT (TNFSF14) comprise a closely related set of ligands in the TNF family [3], [4]. TNF exists as a cell-bound or soluble homotrimer that binds TNF receptor (TNFR)1 and TNFR2 [5], [6]. LTα can form a soluble homotrimer (LTα3) that binds TNFR1, TNFR2 and HVEM [5], [7], but can also form a cell-bound heterotrimer with LTβ (LTα1β2) that binds and signals through LTβR [8]. LIGHT exists in cell-bound and soluble forms that interact with both LTβR and herpes virus entry mediator (HVEM) [7], [9], [10]. HVEM also engages members of the immunoglobulin superfamily; B and T lymphocyte attenuator (BTLA) [11] and CD160 [12], as well as the envelope glycoprotein D of Herpes Simplex virus [13]. HVEM activates BTLA inhibitory signalling via SHP phosphatases suppressing T cell activation [14]. LIGHT, LTα and the Ig superfamily ligands can also activate HVEM-dependent cell survival signalling via NF-κB [15].
LIGHT has emerged as a key mediator of inflammation and immune homeostasis [4], [14]. There is broad expression of LIGHT and HVEM in the hematopoietic compartment [7], [9], [16], [17], [18], while LTβR expression is largely restricted to stromal and myeloid cells [7], [19], [20]. LTβR and HVEM are implicated as key host defence mechanisms against persistent viral [21] and bacterial pathogens [22]. However, little is known about the role of these receptors in infection with parasites that establish persistent infections in their hosts.
The protozoan parasite Leishmania donovani causes persistent infections in humans and experimental animals [23], [24]. We and others have defined important roles for TNF and LTα in host resistance in a mouse model of visceral leishmaniasis (VL) caused by L. donovani [25], [26], [27]. This disease model is characterised by an acute, resolving infection in the liver involving the formation of pro-inflammatory granulomas around infected Kupffer cells, and the establishment of a chronic infection in the spleen (reviewed in [24], [28], [29], [30]). Mice deficient in TNF are highly susceptible to L. donovani infection, and die in the second month of infection with unchecked parasite growth [25], [26], [31]. However, TNF also induces disease pathology in the spleen, including the loss of marginal zone macrophages and down-regulation of chemokine receptor expression by dendritic cells (DCs) [31], [32]. Mice lacking LTα display a less severe phenotype characterised by disrupted cellular trafficking into the liver and reduced control of hepatic parasite growth, although ultimately, infection is resolved in this organ [26].
Here we investigated the impact of L. donovani infection in LIGHT-deficient mice, as well as the roles of LIGHT binding each of its functional, cognate receptors during infection. We report a critical role for LIGHT in the resolution of hepatic infection, and more specifically, identify an important role for LIGHT-HVEM interactions in stimulating IL-12 production by DCs, and hence in the control of parasitic infections. Conversely, we also discovered that blockade of LIGHT-LTβR interactions dramatically enhanced early anti-parasitic immunity. Thus, we have identified distinct and opposing roles for LIGHT engagement of each of its receptors during infection.
Homeostatic levels of LIGHT mRNA in liver (Figure 1A) and spleen (Figure 1B) differed by an order of magnitude in naïve mice. Following L. donovani infection, LIGHT mRNA accumulation increased in the liver over the first 28 days, and remained elevated despite infection largely resolving (Figure 1C). In contrast, the initially high splenic LIGHT mRNA levels decreased over the first 28 days of infection (Figure 1B), and remained diminished as a persistent L. donovani infection became established (Figure S1A). Thus, an organ-specific pattern of LIGHT mRNA expression emerged in response to L. donovani infection.
To establish whether LIGHT was required to control infection, we infected LIGHT-deficient and control C57BL/6 mice with L. donovani and followed the course of infection in the spleen and liver for 90 days. Despite no difference in hepatic parasite burdens in the first 7 days of infection, parasite growth was significantly greater in the livers of LIGHT-deficient mice from day 14 p.i. onwards. Furthermore, these mice failed to fully resolve hepatic infection in the time period studied (Figure 1C). TNF, IFNγ and nitric oxide (measured as the surrogate marker inducible nitric oxide synthase; NOS2) are all critical for control of L. donovani in the liver [26], [27], [31], [33], [34]. Serum TNF and IFNγ levels were reduced, and the accumulation of hepatic NOS2, IFNγ and TNF mRNA were all lower in LIGHT-deficient mice at 14 days, compared with control animals (Figure S1B–E). In the spleen, there were no significant differences in parasite burdens between C57BL/6 and B6.LIGHT−/− mice at any time point studied (data not shown). The accumulation of NOS2 mRNA was much lower in the spleen of C57BL/6 and B6.LIGHT−/− mice at day 14 p.i., compared with the liver, and no difference in IFNγ, TNF and NOS2 mRNA accumulation in the spleen between mouse strains was observed at this time point (Figure S1B–E). We therefore focused our attention on the liver.
The formation of pro-inflammatory granulomas around infected Kupffer cells is a critical step in host control of parasite growth in the liver [24], [28], [29], [30]. Liver immunohistochemistry revealed an increased number of inflammatory foci associated with increased parasite burden and impaired formation of inflammatory granulomas in B6.LIGHT−/− mice at day 14 p.i., relative to control mice, as indicated by a greater frequency of infected Kupffer cells with no surrounding leukocytes (KC), and a lower frequency of immature (IG) and mature granulomas (MG) (Figure 2A). To ensure that the failure to develop anti-parasitic immunity in the liver did not result from an as yet unidentified developmental defect in LIGHT-deficient mice, BM chimeras were made by engrafting LIGHT-deficient or control (C57BL/6) BM cells into lethally irradiated C57BL/6 mice. These mice were infected and parasite burdens measured 14 days later. BM chimeric mice responded to hepatic infection in accordance with their source of BM (Figure 2B). Hepatic parasite burdens were significantly increased in LIGHT-deficient BM chimeras compared to controls, indicating that LIGHT production by leukocytes was required for the efficient generation of anti-parasitic immunity in the liver at this early time point in infection. Additional experiments in T and B cell-deficient B6.RAG1−/− mice receiving LIGHT-deficient or wild-type T cells showed that LIGHT production by T cells was not required for the development of anti-parasitic immunity in the liver during the first 14 days of infection (Figure 2C). However, we cannot exclude a role for T cell-derived LIGHT in the generation of optimal early host immunity following L. donovani infection because we did find a small, but significant difference (p<0.001) in parasite burden at day 14 p.i. between B6.RAG1−/− mice that received LIGHT-deficient T cells and those that received wild-type T cells (Figure 2C). We also found that T cells per se were not a major source of hepatic LIGHT mRNA, although their presence was required for LIGHT expression to increase in the liver during this early period of infection (Figure 2D). Together, these data indicate that the generation of immune responses against L. donovani in the liver were impaired in the first 14 days of infection the absence of LIGHT.
Treatment with anti-HVEM mAb (LH1) that blocks LIGHT binding to HVEM, but not HVEM-BTLA interactions [35], significantly increased hepatic parasite load at day 14 p.i. in mice, similar to the increase in parasite burden observed in LIGHT-deficient mice (Figure 3A). Surprisingly, hepatic parasite burdens were significantly decreased by treatment of mice with anti-LTβR mAb (LLBT2) that blocks LIGHT binding to LTβR, but not LTα1β2-LTβR interactions [35] (Figure 3A). Antibody treatments had no significant effect on the low splenic parasite burden at this time point (Figure S2). The formation of granulomas was significantly impaired by anti-HVEM mAb, as indicated by a greater frequency of KC and a lower frequency of IG and MG (p<0.05, κ2 analysis; Figure 3B). In contrast, granuloma formation was significantly enhanced by anti-LTβR mAb, as indicated by a lower frequency of KC and a higher frequency of IG and MG (p<0.05, κ2 analysis; Figure 3B). Thus, these results indicate that HVEM and LTβR have distinct and opposing roles during the first 14 days of infection.
To further investigate the role of LTβR in VL, we treated mice with the agonist anti-LTβR antibody (3C8) which blocks binding of both LTα1β2 and LIGHT to LTβR, yet functions as an agonist directly activating LTβR signalling pathways [36], [37]. The anti-LTβR 3C8 enhanced parasite clearance in the liver during an established infection (Figure 4A), but had no anti-parasitic effect in the first 14 days of infection (data not shown), unlike the anti-LTβR mAb LLBT2 (Figure 3A). Importantly, 3C8 also prevented parasite growth in the spleen between days 14 to 28 p.i. (Figure 4B). In contrast, treatment with LLTB2 during established infection (days 14–28 p.i.) had no effect on parasite clearance in the liver or spleen (data not shown). Thus, treatment of L. donovani-infected mice with two different anti-LTβR mAbs had distinct effects on the course of infection, reflecting different functional properties of these mAbs.
We next sought to identify anti-parasitic mechanisms dependent upon LIGHT-HVEM signalling. We previously showed that early splenic IL-12/IL-23p40 production by DC is critical for the efficient generation of immunity in the liver [38], [39]. Although no change in IL-12p35 mRNA accumulation was observed in any treatment group, the anti-HVEM mAb (LH1) inhibited splenic DC IL-12/IL-23p40 mRNA accumulation in response to L. donovani infection (Figure 5A). We next evaluated the importance of LIGHT-HVEM co-stimulatory signals for the development of L. donovani-specific CD4+ T cell priming and Th1 differentiation, the latter being a known IL-12-dependent process [39], [40], [41]. Mice were injected with CFSE-labelled OVA-specific CD4+ (OT-II) T cells, then infected with transgenic OVA-expressing L. donovani [42], and antigen-specific CD4+ T cell proliferation was assessed 4 days later by CFSE dilution. No antigen-specific CD4+ T cell proliferation was observed when mice were infected with wild-type parasites (Figure 5B), so no bystander activation had occurred, and OT-II cell proliferation occurred equally in control and anti-HVEM-treated mice (Figure 5B), indicating that LIGHT-HVEM interactions were not required for early priming of CD4+ T cell proliferation. In support of this result, proliferation of polyclonal antigen-specific, CD4+ T cells was similar between cells isolated from the spleens of control-treated and anti-HVEM treated mice (Figure 5C). However, production of IFNγ and TNF by these antigen-specific CD4+ T cells was inhibited by anti-HVEM mAb (Figure 5C). Furthermore, direct ex vivo production of IFNγ by hepatic CD4+ T cells (both total number and frequency) was significantly reduced by anti-HVEM mAb (Figure 5D), indicating that LIGHT-HVEM interactions play an important role in generating Th1 cell responses following L. donovani infection.
The anti-LTβR mAb (LLTB2) inhibited parasite growth in acute experimental VL (Figure 3A), but had no effect on splenic DC IL-12/IL-23p40 or IL-12p35 mRNA accumulation at 5 hours p.i. (Figure 6A), and no effect on the expansion of OVA-specific CD4+ T cells (OT-II cells) in mice infected with OVA-transgenic L. donovani (Figure 6B). We also found no differences in antigen-specific recall responses in splenic CD4+ T cells isolated from infected mice on day 14 p.i., yet the amount of TNF and IFNγ produced upon antigen-specific CD4+ T cell stimulation was greatly enhanced in these cells from mice treated with anti-LTβR mAb (Figure 6C). In addition, the number and frequency of IFNγ-producing hepatic CD4+ T cells measured directly ex vivo on day 14 p.i. was significantly increased in these mice (Figure 6D), suggesting LIGHT-LTβR binding suppresses the development of Th1 cell responses in VL.
We next investigated timing requirements for treatment with the anti-LTβR mAb (LLTB2) during acute infection with L. donovani. A single dose (100 µg) of anti-LTβR mAb at the time of infection was sufficient to reduce hepatic parasite burden as early as day 7 p.i. (Figure 7A). To test whether treatment with anti-LTβR mAb was simply shunting available LIGHT onto HVEM, we also co-treated mice with anti-LTβR (LLTB2) and anti-HVEM (LH1) mAbs (Figure 7B), and found no additional effect of co-administration over anti-LTβR alone by day 7 p.i., indicating that increased, early anti-parasitic immunity observed after anti-LTβR mAb (LLTB2) treatment was not caused by enhanced HVEM-mediated co-stimulation. Of note, there was no effect of anti-HVEM mAb treatment alone at day 7 p.i., indicating that the effect of this treatment on parasite burden only becomes apparent between days 7–14 p.i., similar to what was observed in LIGHT-deficient mice (Figure 1C). To test whether anti-LTβR mAb (LLTB2) agonist activity might account for the above effect, we treated LIGHT-deficient mice with this antibody and measured liver parasite burdens at day 7 p.i. (Figure 7C). Although a significant reduction in parasite burden was found in C57BL/6 mice treated with anti-LTβR mAb (LLTB2), no such effect was observed in LIGHT-deficient mice, indicating that the likely mechanism of action was via the blockade of LIGHT binding LTβR.
We investigated the cellular requirements for the early anti-parasitic effects of anti-LTβR mAb (LLTB2). Treatment with anti-LTβR mAb had no impact on hepatic parasite burdens in B6.RAG1−/− mice at day 7 p.i. (Figure S3), suggesting that B and/or T lymphocytes are required for the enhanced parasite clearance resulting from this treatment. We focused our attention on T cells because we have previously shown that B cells play a negative regulatory role in the liver during infection [43]. Depletion of CD4+ or CD8+ T cells alone during the first 7 days of infection had no effect on hepatic parasite burden (Figure 8A), despite T cells being required for the control of parasite growth at later stages of infection [26], [44], [45]. However, depletion of CD4+ T cells, but not CD8+ T cells, prevented the anti-parasitic effect mediated by anti-LTβR at day 7 p.i. (Figure 8A). Given that NKT cells comprise a significant proportion of hepatic CD4+ T cells, we also investigated whether this cell subset was required for the increased anti-parasitic activity. Treatment of NKT cell-deficient (B6.Jα18−/−) mice with the anti-LTβR mAb (Figure 8A) had no impact on the decreased liver parasite burden, indicating that conventional CD4+ T cells, but not NKT cells, are required for the enhanced parasite clearance following anti-LTβR mAb treatment.
We observed increased CD4+ T cell TNF and IFNγ production was associated with improved control of parasite growth resulting from anti-LTβR mAb treatment (Figure 6). We next assessed whether these cytokines were required for the enhanced parasite clearance in mice receiving anti-LTβR (LLTB2) mAb. Hepatic parasite burdens were decreased similarly in anti-LTβR mAb treated control and IFNγ-deficient mice (Figure 8B). However, anti-LTβR mAb treatment in TNF-deficient mice had no impact on hepatic parasite burden (Figure 8B), indicating that TNF is critical for this enhanced parasite clearance. The failure of anti-LTβR mAb treatment in TNF-deficient animals was not caused by reduced expression of LTβR on the cells of these mice, as LTβR expression levels were no different to those on immune cells from C57BL/6 control mice (data not shown). Furthermore, adoptive transfer of wild type and TNF-deficient CD4+ T cells into B6.RAG1−/− mice and treatment with anti-LTβR mAb demonstrated that CD4+ T cells did not have to produce TNF (Figure 8C). Thus, anti-LTβR mAb treatment increased early hepatic anti-parasitic immunity by mechanisms requiring conventional CD4+ T cells and TNF, the latter potentially coming from a non-T cell source.
We have identified distinct and opposing roles for LIGHT and its receptors during infection. LIGHT has important roles in T cell costimulation [3], [14]. Blockade of LIGHT impairs allogeneic T cell responses and graft versus host disease [46], [47], while over-expression of LIGHT by T cells causes inflammatory disease of the gut and reproductive tissues [48], [49]. Our results indicate that these effects could be mediated via the LIGHT-HVEM axis between T cells and DC. Early DC IL-12 production depends on the presence of T cells, and this IL-12 production is critical for generating anti-parasitic immune mechanisms that control L. donovani growth [24], [39], [40], [50]. Our finding that anti- HVEM mAb blocks IL-12/IL-23p40 mRNA accumulation during infection is consistent with a previous study that reported BM-derived DCs from LIGHT-deficient animals were impaired in their ability to produce IL-12 following activation in vitro [51]. This study also showed that blockade of LIGHT with soluble receptors in mice infected with L. major, a cause of cutaneous leishmaniasis, resulted in reduced IL-12 generation, associated with diminished CD4+ T cell IFNγ production and increased parasite growth and disease. Our finding that T cells did not have to produce LIGHT in order to promote anti-parasitic immunity, together with data from L. major infection in mice [51], support a model whereby DC-derived LIGHT interacts with T cell HVEM to promote DC IL-12 production.
The defect in anti-parasitic immunity observed in the absence of LIGHT was restricted to the liver, and not the spleen. The reason for this is unclear, but could relate to the requirement for cellular recruitment and granuloma development for control of parasite growth in the liver. Although increased tissue weight and cellular expansion are features of L. donovani infection in the spleen, organised inflammatory granulomas are rarely observed in this tissue [24], [28], [29], [30]. Importantly, parasite growth is contained in the spleen after 1–2 months of infection rather than efficiently controlled, as occurs in the liver. Therefore, it is possible that different anti-parasitic immune mechanisms operate in these two tissue sites during experimental VL with different requirements for LIGHT.
The LIGHT-specific blocking mAbs we have employed (LH1 and LLTB2) have previously been shown to selectively block interactions between LIGHT and its receptors (HVEM and LTβR, respectively) [35]. However, we cannot exclude the possibility that they may trigger some receptor activation following engagement, and that this may contribute to biological effects we have observed. In addition, because these mAbs cause the selective blockade of LIGHT binding to their respective receptors, we cannot rule out that they promote alternative receptor-ligand interactions (Figure 9A). For example, blocking LIGHT interacting with HVEM may allow HVEM to more readily engage BTLA on cells to increase inhibitory signals (Figure 9B), as well as increased CD160 signalling. Similarly, blockade of LIGHT binding LTβR may allow greater amounts of LIGHT to bind HVEM, thereby reducing negative signalling between HVEM and BTLA and potentially promoting LIGHT-HVEM-mediated T cell co-stimulation (Figure 9C). However, this latter possibility seems unlikely in the current study given that co-administration of LH1 and LLTB2 resulted in improved control of parasite growth (Figure 7B). Instead, LIGHT may send inhibitory signals via LTβR early during infection, although no such LTβR-mediated negative signalling pathway has yet been defined.
The anti-parasitic effects of anti-LTβR mAb (LLTB2) were observed when it was administered at the time of infection, but not in mice with an established L. donovani infection, suggesting that a mAb with similar functional characteristics would have limited therapeutic potential for treatment of VL. However, our finding that the defined agonist anti-LTβR mAb (3C8) improved the rate of parasite clearance in the liver and reduced parasite load in the spleen, not only demonstrated fundamentally different biological activities for LLTB2 and 3C8 mAbs, but also shows that LTβR activation can promote beneficial immune mechanisms during established infection. This agonist antibody has previously been shown to promote DC development and maturation in vivo [20], [37], and this may explain the anti-parasitic effects observed after administration to L. donovani-infected mice because we have previously shown that DC adoptive transfer can improve control of parasite growth in infected mice [32]. Hence, the activation of anti-parasitic immune mechanisms by stimulation of LTβR represents a potential therapeutic strategy against chronic infectious diseases like VL. However, a better understanding of the functional characteristics of the different anti-LTβR mAbs will be required in order to better harness their therapeutic potential, including identification of the specific epitopes they recognise and signalling pathways they activate.
We previously reported increased monocyte recruitment into the spleen in an experimental model of cerebral malaria following treatment of mice with the anti-LTβR mAb (LLTB2), and that this treatment protected mice from disease [52]. Interestingly, no protection from experimental cerebral malaria was afforded by treatment with the anti-LTβR (3C8) mAb (Randall and Engwerda, unpublished), again emphasising the functional differences between LLTB2 and 3C8 anti-LTβR mAbs. An intriguing finding from our current studies was an increase in hepatic and splenic monocyte recruitment following anti-LTβR mAb LLTB2 treatment (CD11b+ Ly6Chi cells; Figure S4A and B). Flow cytometry analysis revealed that monocytes, along with DCs (both cDC and pDC), and neutrophils expressed the highest levels of LTβR in the liver, as previously reported [19], [20], [37], and furthermore, that expression of LTβR did not appear to change significantly on any of these cells during the first 5 days of infection with L. donovani (Figure S5). However, the increased monocyte recruitment was not necessary for improved early control of parasite growth in treated animals in the current study because mice lacking CCL2 that have an impairment in monocyte mobilisation [53], also had improved control of parasite growth following anti-LTβR (LLTB2) treatment at day 7 p.i. (Figure S4C). Although the early anti-parasitic effect of anti-LTβR (LLTB2) mAb appeared to involve blocking of LIGHT- LTβR interactions, as indicated by the failure of this antibody to improve parasite control in LIGHT-deficient mice (Figure 7C), we cannot exclude the possibility that some effects of this antibody, such as increased monocyte mobilisation, might involve agonist activities. Regardless, given the important role for monocyte infiltration into sites of infection and tumour growth [54], our results indicate that manipulation of the LIGHT-LTβR signalling axis offers a potential way to improve monocyte mobilisation for therapeutic applications. Furthermore, given the recent report that monocytes can migrate into secondary lymphoid tissues in response to interactions with gram negative bacteria and/or their products, and then develop into CD209a+, CD206+, CD14+, CD11chi DC capable of activating CD4+ T cells and cross-priming CD8+ T cells [55], our results suggest that manipulation of LIGHT signalling pathways may be one way to promote this process that may have applications in vaccination.
In summary, our findings further delineate the complex interactions between LIGHT and its receptors and demonstrate the therapeutic potential of modulating these immune regulatory pathways to improve disease outcomes. Our results provide mechanistic insight into the roles of LIGHT-HVEM interactions on DC function and CD4+ T cell priming, as well as anti-parasitic immune responses activated by blockade of LIGHT-LTβR interactions. Finally, we have identified two different mAbs that target LTβR with distinct functional outcomes on anti-parasitic immunity at different stages of infection.
Inbred female C57BL/6 and B6.SJL.Ptprca (B6.CD45.1) mice were purchased from the Australian Resource Centre (Canning Vale, Western Australia), and maintained under conventional conditions. B6.RAG1−/− [56], B6.LIGHT−/− [57], B6.TNF−/− [58], B6.SJL.Ptprca×OT-II [59], B6.SJL.Ptprca×OT-I [60], B6.IFNγ−/− [61] and B6.Jα18−/− [62] were bred and maintained at the Queensland Institute of Medical Research. B6.CCL2−/− mice [53] were bred at Monash University and maintained at the Queensland Institute of Medical Research. All mice used were age- and sex-matched (6–10 weeks), and were housed under specific-pathogen free conditions. Chimeric mice were prepared by irradiating B6.SJL.Ptprca mice with 11Gy and then engrafting with 3×106 fresh bone marrow (BM) cells i.v. via the lateral tail vein. Mice were maintained on antibiotics for 2 weeks after engraftment and infected with L. donovani 8 weeks after receiving BM, as previously described [26]. Adoptive transfer of equal numbers (106) of purified CD4+ and CD8+ T cells (98% purity as determined by flow cytometry) into B6.RAG1−/− mice was performed as previously described [26].
All animal procedures were approved and monitored by the Queensland Institute of Medical Research Animal Ethics Committee. This work was conducted under QIMR animal ethics approval number A02-634M, in accordance with the “Australian code of practice for the care and use of animals for scientific purposes” (Australian National Health & Medical Research Council).
L. donovani (LV9) and OVA-transgenic LV9 (PINK LV9) [42] were maintained by passage in B6.RAG1−/− mice and amastigotes were isolated from the spleens of chronically infected mice. Mice were infected by injecting 2×107 amastigotes i.v. via the lateral tail vein, killed at the times indicated in the text by CO2 asphyxiation and bled via cardiac puncture. In experiments examining DC IL-12/IL-23p40 production, mice were infected with 1×108 amastigotes intravenously, as previously described [63]. Spleens and perfused livers were removed at times indicated and parasite burdens were determined from Diff-Quik-stained impression smears (Lab Aids, Narrabeen, Australia) and expressed as Leishman-Donovan units (LDU) (the number of amastigotes per 1,000 host nuclei multiplied by the organ weight in grams) [64]. Liver and spleen tissue were also preserved in either RNAlater (Sigma-Aldrich, Castle Hill, Australia) or Tissue-Tek O.C.T. compound (Sakura, Torrence, USA). Hepatic mononuclear cells and splenocytes were isolated as previously described [65].
All antibody-producing hybridomas were grown in 5% (v/v) foetal calf serum, RPMI containing 10 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Purified antibody was prepared as previously described [63]. Mice were administered 100 µg of anti-LTβR mAb (LLTB2) or anti-HVEM mAb (LH1) [35] i.v. on the day of infection and every 5 days thereafter for 14 day experiments, or as a single dose on the day of infection for 7 day experiments.. The anti-LTβR mAb 3C8 was administered at 200 µg i.v. [36], [37] starting at the times indicated in the text and every 5 days thereafter. The anti-LTβR mAb (LLTB2) or anti-HVEM mAb (LH1) specifically block the binding of LIGHT to either LTβR or HVEM, respectively, but do not disrupt interactions between these receptors and other functional ligands (i.e., LTα1β2 for LTβR and BTLA for HVEM) [35]. The anti-LTβR mAb (3C8) blocks binding of both LTα1β2 and LIGHT, but is an agonist directly activating LTβR [66]. Mice were depleted of CD4+ or CD8+ T cells with anti-CD4 (YTS191.1) or anti-CD8β (53-5.8) mAbs, respectively, as previously described [64]. Depletion of T cell subsets was confirmed at completion of experiments by assessing T cell numbers in the spleen by flow cytometry. Greater than 95% of CD4+ and CD8+ T cells were depleted by antibody treatment. In all experiments, control mice received the same quantities of the appropriate control hamster IgG (UC8-1B9; ATCC, Manassas, VA) or control rat IgG (Sigma-Aldrich).
To assess antigen-specific T cell proliferation in vivo, mice were infected with OVA-transgenic PINK LV9 [42]. Splenic OVA-specific OT-II T cells were isolated and labelled with CFSE, as previously described [64]. CFSE-labelled OT-II cells (1×106) were adoptively transferred into mice 2 h prior to infection with LV9 or PINK LV9. Expansion of CFSE+ cells in the spleen was monitored by FACS 4 days later. In all of these experiments, control animals were included that received the same number of CFSE-labelled OT-II cells, but were infected with wild-type parasites. No OT-II proliferation was ever observed in these animals. Re-stimulation assays for endogenous splenic CD4+ T cells were performed as previously described [63].
The maturation of granulomas was scored around infected Kupffer cells in acetone-fixed liver sections as previously described [64], [65].
Allophycocyanin (APC)-conjugated anti-TCRβ chain (H57-597), B220 (clone RA3-6B2), CD11c (clone N418), phycoerythrin (PE)-Cy5-conjugated anti-CD4 (GK1.5), PE-conjugated IFNγ (XMG1.2), CD8β (53-5.8), I-Ab (clone AF6-120.1), Ly6G (clone 1A8), CD45.1 (clone A20), CD45.2 (clone 104), rat IgG1 (RTK2071), fluorescein isothiocyanate (FITC)-conjugated CD19 (clone 6D5), BST2 (clone 120G8), Ly6C (clone AL-21) and biotinylated anti-NK1.1 (PK136), CD11b (clone M1/70), LTβR (3C8) were purchased from Biolegend (San Diego, CA) or BD Biosciences (Franklin Lakes, NJ). Biotinylated antibodies were detected with streptavidin conjugated alexa 488, PE or PE/Cy5 (Biolegend). Leukocyte populations were defined as follows; CD4+ T cells (CD4+, TCR+), CD8+ T cells (CD8+, TCR+), NKT cells (NK1.1+, TCR+), NK cells (NK1.1+, TCR−), B cells (B220+, CD19+), cDC (CD11chi, MHCIIhi, TCR−, B220−), pDC (CD11cint, MHCIIint, 120G8+), monocytes (CD11b+, Ly6C+) and neutrophils (CD11b+, Ly6G+). The staining of cell surface antigens and intracellular cytokine staining was carried out as described previously [63]. FACS was performed on a FACSCalibur or a FACS Canto II (BD Biosciences), and data were analysed using FlowJo software (TreeStar, Oregon, USA). Serum and/or tissue culture supernatants were assessed for the presence of soluble cytokines using flexset bead array kits and a FACSArray plate reader (BD Biosciences) according to the manufacturers' instructions.
RNA extraction and real-time RT-PCR was performed as previously described [63]. The number of IFNγ, TNF, NOS-2, LIGHT and hypoxanthine phosphoribosyltransferase (HPRT) cDNA molecules in each liver tissue sample were calculated using Platinum Sybr Green Master Mix (Invitrogen Life Technologies) [63]. Standard curves were generated with known amounts of cDNA for each gene, and the number of cytokine molecules per 1000 HPRT molecules in each sample was calculated. The number of IL-12/IL-23p40 and IL-12p35 cDNA molecules in each DC sample were calculated using Taqman Gene Expression Assays (Applied Biosystems). Relative quantitation of gene expression was performed using the relative standard curve method as described by Applied Biosystems. Briefly, standard curves were prepared for all target and endogenous control genes using an uninfected control sample. HPRT was used as the endogenous control. The amount of target gene or endogenous control in each sample was calculated from the appropriate standard curves. The target amount was then divided by the endogenous control amount to give the normalized target value. The average normalized values for the four naïve samples were used as the calibrator.
Statistical differences between groups was determined using the Mann-Whitney U test using GraphPad Prism version 4.03 for Windows (GraphPad Software, San Diego, CA) and p<0.05 was considered statistically significant. The distribution of hepatic histological responses was compared using X2 analysis with Microsoft Excel software. All data are presented as the mean values plus or minus standard error unless otherwise stated.
|
10.1371/journal.pgen.1003376 | Alternative Splicing and Subfunctionalization Generates Functional Diversity in Fungal Proteomes | Alternative splicing is commonly used by the Metazoa to generate more than one protein from a gene. However, such diversification of the proteome by alternative splicing is much rarer in fungi. We describe here an ancient fungal alternative splicing event in which these two proteins are generated from a single alternatively spliced ancestral SKI7/HBS1 gene retained in many species in both the Ascomycota and Basidiomycota. While the ability to express two proteins from a single SKI7/HBS1 gene is conserved in many fungi, the exact mechanism by which they achieve this varies. The alternative splicing was lost in Saccharomyces cerevisiae following the whole-genome duplication event as these two genes subfunctionalized into the present functionally distinct HBS1 and SKI7 genes. When expressed in yeast, the single gene from Lachancea kluyveri generates two functionally distinct proteins. Expression of one of these proteins complements hbs1, but not ski7 mutations, while the other protein complements ski7, but not hbs1. This is the first known case of subfunctionalization by loss of alternative splicing in yeast. By coincidence, the ancestral alternatively spliced gene was also duplicated in Schizosaccharomyces pombe with subsequent subfunctionalization and loss of splicing. Similar subfunctionalization by loss of alternative splicing in fungi also explains the presence of two PTC7 genes in the budding yeast Tetrapisispora blattae, suggesting that this is a common mechanism to preserve duplicate alternatively spliced genes.
| The role of duplicated genes in originating new functions is an important question in evolution. Almost all species have duplicated genes that carry out similar but not identical functions. Similar proteins that perform different functions can also be generated when one gene generates multiple mRNAs by alternative splicing that are translated into multiple similar proteins. This alternative splicing is prevalent in animal cells, but much rarer in fungi. Here we show that most fungi use alternative splicing to make a Ski7 protein and a Hbs1 protein from the same gene. Two fungi, budding yeast and fission yeast, have been much better characterized than other fungi, and co-incidentally they both have duplicated this alternatively spliced gene, resulting in two similar genes that are no longer alternatively spliced. Finally, we describe another example where two duplicate genes replace one alternatively spliced gene, suggesting that this is a common mechanism to divide functions among duplicate genes.
| Gene duplication is thought to be a major source of evolutionary innovation. Although the fate of duplicated genes is incompletely understood, it is thought to fit one of three patterns: nonfunctionalization, neofunctionalization or subfunctionalization. Of these nonfunctionalization is thought to be the most common. Immediately after duplication, duplicated genes are typically redundant. Thus, purifying selection cannot provide selective pressure to maintain both. The absence of selective pressure generally leads to loss of function mutations (nonfunctionalization) in one of the copies, followed by loss of that copy of the gene. In neofunctionalization, one of the duplicated copies acquires a new advantageous function that is different from the ancestral function maintained by the other copy [1]. Subfunctionalization can occur when an ancestral gene carries out more than one function. If one duplicated copy mutates so that it loses one of the functions, and the other copy mutates so that it loses a separate function, selective pressure can subsequently maintain both copies by selecting for both functions [2], [3]. Multiple functions in this context can mean being expressed in multiple cell types, encoding proteins localized to different compartments, encoding proteins with distinct biochemical activities, etc.
Saccharomyces cerevisiae is an excellent model organism to study the fate of duplicated gene pairs because an ancestor underwent a whole genome duplication (WGD) approximately 100 million years ago resulting in a transient increase in genome size from around 5000 protein coding genes to 10,000 [4], [5], [6]. Following genome duplication most duplicated genes were lost (nonfunctionalized), but 544 duplicated gene pairs that arose from WGD remain [4]. The genomes of many related species have been sequenced, which revealed through synteny which genes were duplicated as part of the WGD [6], [7], [8]. The related genomes also provide a large amount of sequence information on the duplicated genes and their non-duplicated homologs. The pattern of gene retention in these genomes revealed that nonfunctionalization after WGD is random such that different post-WGD species retained different subsets of duplicated genes [9]. In addition, gene function can be easily assayed in S. cerevisiae. Using these advantages, we have previously shown that subfunctionalization is a major mechanism by which duplicated S. cerevisiae genes were retained, which was confirmed by others [10], [11], .
One example of subfunctionalized genes resulting from the WGD event is provided by SKI7 and HBS1 [11]. The S. cerevisiae Ski7 and Hbs1 proteins both recognize ribosomes stalled during translation and initiate degradation of the mRNA. However, they recognize different stalled ribosomes and initiate mRNA degradation differently. When an mRNA lacks an in frame stop codon, the ribosome is thought to translate until it reaches the 3′ end of that mRNA [13]. The stalled ribosome is then recognized by Ski7, which recruits the RNA exosome to degrade that mRNA. In contrast, Hbs1 recognizes ribosomes stalled within the coding region, for example due to a structure or damage in the mRNA [14]. Recognition by Hbs1 causes cleavage of the mRNA in an RNA exosome-independent manner [14], [15]. Although the SKI7 and HBS1 genes of S. cerevisiae perform distinct functions, we have previously shown that the single ancestral gene performed both functions [11]. The related budding yeast Lachancea kluyveri diverged from S. cerevisiae before WGD and thus contains a single ortholog to SKI7 and HBS1, which we will call SKI7/HBS1. Our key finding was that when the L. kluyveri SKI7/HBS1 gene was introduced into S. cerevisiae it could complement the defects caused by both ski7Δ and hbs1Δ, thus indicating that this single gene carried out both functions [11].
Since the function of duplicated genes can diverge from each other through neo- or subfunctionalization, gene duplication may be one way to generate a more diverse proteome. The proteome can also be diversified through alternative splicing, where one gene generates multiple distinct mRNAs that each encode a distinct protein. Although alternative splicing is important to diversify the proteome in metazoans, it is much rarer in the fungal kingdom. Most fungal alternative splicing events that have been described are of the intron retention type, where the spliced mRNA encodes a functional protein, and the unspliced mRNA is nonfunctional. For example, transcriptome sequencing of Aspergillus oryzae identified only 8.6% of the genes as alternatively spliced, which is 10-fold lower than in humans and 92% of the Aspergillus alternative splicing was intron retention [16]. A well-studied and typical example of fungal intron retention is the S. cerevisiae CYH2 gene. The CYH2 mRNA encodes a 17 KDa ribosomal protein. The intron in the CYH2 pre-mRNA is retained approximately 50% of the time, which results in an mRNA that codes for a 2 KDa peptide with no known function. Furthermore, intron-retained mRNAs are typically very rapid degraded by the nonsense-mediated mRNA decay pathways [17]. In these cases instead of diversifying the proteome, intron retention may function to regulate gene expression. Similarly, the S. cerevisiae SRC1 gene is alternatively spliced using alternative 5′ splice sites, but only the longer splice isoform has been shown to be functional [18], [19]. To the best of our knowledge, the only case in which intron retention or alternative splicing leads to two functional mRNAs in S. cerevisiae is in PTC7, which contains one intron and encodes a protein phosphatase subunit. If this intron is spliced out, the mRNA is translated into a protein that is imported into the mitochondria, while after intron retention the mRNA is translated into a protein that is inserted into the nuclear envelope [20]. A few other cases have been described were fungi use alternative splicing to target a protein to multiple locations [21], [22], [23].
As mentioned above, there are multiple ways a gene can be multifunctional in the context of subfunctionalization. A corollary of that is that subfunctionalization can occur through distinct molecular changes. In yeast, subfunctionalization through changes in the coding region seem to be common [10], [11], [12]. In these cases a single amino acid change can be responsible for subfunctionalization [10], [12]. In multicellular organisms, genes that are expressed in multiple cell types, in response to multiple stimuli, or by multiple transcription factors can be subfunctionalized through changes in expression pattern [e.g. ref 3]. Subfunctionalization through changes in splicing patterns have been described in a few cases [e.g. refs 24], [25]. In these cases, an alternatively spliced gene upon duplication results in two genes where one gene follows one ancestral splicing pattern and the other follows another ancestral splicing pattern. However, the function of these alternative splicing isoforms is often not clear. Thus while loss of alternative splicing happens at the same time as some gene duplications, whether they cause subfunctionalization has not been experimentally demonstrated.
Here we show that the pre-WGD ancestor of SKI7 and HBS1 was alternatively spliced. We also show that in most extant fungi, including ascomycetes and basidiomycetes, the SKI7/HBS1 gene is still alternatively spliced, thereby describing the by far most conserved fungal alternative splicing event. The L. kluyveri alternative splicing isoforms are functionally distinct, such that one spliced mRNA encodes a functional Hbs1, while an alternatively spliced mRNA encodes a functional Ski7. Sequence analysis indicates that a very similar subfunctionalization occurred in an ancestor of the Schizosaccharomyces genus. Finally, while the S. cerevisiae PTC7 gene encodes two differently localized proteins through intron retention, in a related species this gene is replaced by a pair of duplicated genes that arose form WGD. Thus, evolution of a fungal ancestral alternatively spliced gene into two subfunctionalized genes occurred at least three times: twice for the SKI7/HBS1 gene, and once for PTC7. This further suggests that alternative splicing and gene duplication are not independent mechanisms to diversify the proteome, but instead are interrelated.
Several genomes of Saccharomycetaceae have been sequenced, but incompletely annotated. Upon careful analysis of these sequences we noticed that the SKI7/HBS1 genes in five pre-WGD Saccharomycetaceae each have a potential intron (Figure 1A). In contrast, S. cerevisiae and five other post-WGD species lack introns in both SKI7 and HBS1. Furthermore, each pre-WGD gene has two potential 3′ splice sites, resulting in the potential to encode two different conserved proteins. In the case of L. kluyveri these proteins are predicted to be 70 and 96 KDa (Figure 1A). Although alternative splicing is rare in fungi, we speculated that the SKI7/HBS1 gene may be alternatively spliced. We used rt-PCR to show that both predicted splice sites are indeed used. Use of the proximal 3′ splice site was confirmed using rt-PCR with a primer upstream of the 5′ splice site and a primer downstream of the proximal 3′ splice site and sequencing the resulting PCR product (Figure 1B left panel). Use of the distal 3′ splice site was similarly confirmed using a primer downstream of the distal 3′ splice site (Figure 1B right panel). Thus, the L. kluyveri gene is indeed alternatively spliced through the use of alternative 3′ splice sites.
To determine whether both spliced mRNAs were used to generate a protein, we generated a plasmid that introduced the HA epitope at C-terminus of the L. kluyveri ORF. A western blot shows that two proteins of the expected size are indeed made in L. kluyveri (Figure 1C; second lane). We further modified the HA-tagged plasmid by deleting the intron. In one construct we deleted sequences between the 5′ and distal 3′ splice sites, such that only the short 70 KDa splice isoform could be expressed. Figure 1C (third lane) shows that the encoded protein comigrates precisely with the smaller of the two species seen when the intron is included. Conversely, in another plasmid (fourth lane) we deleted sequences between the 5′ and proximal 3′ splice sites and as expected only the large 96 KDa isoform was made. Therefore, the rt-PCR and Western blot data show that the single SKI7/HBS1 gene of L. kluyveri is used to generate two distinct proteins through use of alternative 3′ splice sites.
We have previously shown that the L. kluyveri SKI7/HBS1 can carry out both the Ski7 and Hbs1 functions by showing that the L. kluyveri gene can complement both a ski7Δ and an hbs1Δ in S. cerevisiae [11]. To test whether both L. kluyveri proteins were generated in this context, we introduced the same HA-tagged constructs describe above into a wild-type S. cerevisiae strain. Western blot analysis indicates that when the L. kluyveri SKI7/HBS1 gene is introduced into S. cerevisiae, both L. kluyveri proteins are made (data not shown). To determine whether one splice isoform carries out the Ski7 function and the other splice isoform carries out the Hbs1 function, plasmids expressing one or both proteins were introduced into both a dcp1-2 ski7Δ strain and an rps30aΔ hbs1Δ strain. A ski7Δ by itself does not result in a growth phenotype, but in combination with dcp1-2, results in a failure to grow at 30°C [26]. This ski7Δ phenotype can be complemented by the unmodified L. kluyveri gene (Figure 2A third row top panel) and by the long splice isoform (fourth row) but not by the short splice isoform (fifth row). Thus, only the long splice isoform can perform the Ski7 function. hbs1Δ by itself does not result in a growth phenotype but in combination with rps30aΔ results in slow growth at room temperature [27]. This hbs1Δ phenotype can be complemented by the unmodified L. kluyveri gene (Figure 2B third row bottom panel) and by the short splice isoform (5th row), but not by the long splice isoform (4th row). We conclude that alternative splicing generates two functionally distinct polypeptides, with the long splice isoform functioning similar to Ski7 and the short splice isoform similar to Hbs1.
Multiple sequence alignment (Figure S1) identified several sequence elements that correlated with Hbs1 function in short splice isoforms of pre-WGD Saccharomycetaceae SKI7/HBS1 genes and post-WGD HBS1 genes. The structure of S. cerevisiae Hbs1 has been solved, which shows a structured N-terminal domain and a C-terminal GTPase domain connected by a flexible linker [28], [29], [30]. The structured N-terminal domain is conserved in other post-WGD Hbs1 proteins but not in post-WGD Ski7. In the alternatively spliced pre-WGD homologs, this domain is encoded by exon 1, and thus present in both splice isoforms. The unstructured linker of Hbs1 is poorly conserved, with the exception of one sequence motif (Motif H1 in Figure 2C). A similar sequence motif is encoded in pre-WGD SKI7/HBS1 genes, but has diverged very much in post-WGD SKI7 genes. The GTPase domain is also highly conserved in post-WGD Hbs1 proteins and pre-WGD Ski7/Hbs1 proteins (Motifs G1 to G5 in Figure S1), consistent with previous findings that Hbs1 GTPase activity is important for its function [31], [32], [33]. In contrast, Ski7 has not been shown to be an active GTPase and the domain has diverged rapidly post-WGD. Specifically, a catalytically important His residue in motif G3 is changed to Ser, Asn or Asp in post-WGD Ski7s.
The structure of Ski7 has not been experimentally determined, but the N-terminus is known to be important for interaction with the RNA exosome and three other Ski proteins [34]. Multiple sequence alignment indicated that although the Ski7 N-terminus is generally poorly conserved, it contains three conserved sequence motifs (Figure 2C and Figure S1; motif S1, S2, and S3). Alignment of pre-WGD Ski7/Hbs1 sequences shows that these motifs are also conserved in the pre-WGD species. Motif S1 and S2 are encoded between the two alternative 3′ splice sites and thus are only present in the longer splice form. These observations suggest that the short isoform of pre-WGD SKI7/HBS1 genes may fail to carry out Ski7 function because they lack motifs S1 and S2.
Most ascomycetes contain a single SKI7/HBS1 gene. To determine whether alternative splicing of SKI7/HBS1 was restricted to pre-WGD Saccharomycetaceae such as L. kluyveri or is a more ancient feature, we next looked at the more distantly related ascomycetes. The phylum Ascomycota can be divided in three subphyla, the Saccharomycotina (which includes Saccharomyces and Lachancea), the Pezizomycotina and the Taphrinomycotina. We therefore used the same rt-PCR and sequencing approach described above to analyze SKI7/HBS1 splicing in Aspergillus nidulans and Saitoella complicata, which are members of the Pezizomycotina and Taphrinomycotina, respectively. Figure 3A shows that Aspergillus nidulans also uses alternative 3′ splice sites in SKI7/HBS1 to generate two distinct mRNAs. The single A. nidulans SKI7/HBS1 gene contains four introns. The second intron contains two predicted alternative 3′ splice sites, and rt-PCR and sequencing indicated that both are used. Similarly, the single SKI7/HBS1 gene in Saitoella complicata contains seven introns, and the second intron contains two predicted 3′ splice sites. Figure 3B shows that both 3′ splice sites are used. Notably, the alternative spliced introns in Lachancea, Aspergillus and Saitoella are in the same position, just upstream of motif S3 characteristic of Ski7. Therefore, the capacity to use alternative 3′ splice sites in SKI7/HBS1 is conserved throughout the phylum Ascomycota.
Although we were able to predict alternative 3′ splice sites in the single SKI7/HBS1 gene of many other fungi, we noted four notable differences in the Schizosaccharomyces genus, a subset of the CTG clade (Saccharomycetales that use CUG as a serine codon instead of the canonical leucine), and the basidiomycetes Cryptococcus neoformans and Ustilago maydis. Species within the Schizosaccharomyces genus have duplicated SKI7 and HBS1 genes (see next section), while species in the CTG clade and the two basidiomycetes each contain a single SKI7/HBS1 gene, but these genes lack obvious alternative 3′ splice sites.
The CTG clade can be divided into two smaller clades. One clade contains Candida guilliermondii, Debaryomyces hansenii, and Candida lusitaniae. In all three species there are two potential 3′ splice sites in locations similar to L. kluyveri (Figure S2). Thus, these three species appear to use the same mechanism as other ascomycetes to express two splice isoforms from a single SKI7/HBS1 gene. The other clade includes C. albicans, C. dubliniensis, C. tropicalis, and C. parapsilosis. These four Candida species also contain a predicted intron within their single HBS1/SKI7 gene, however we only detected one potential 3′ splice site, which corresponds to the distal 3′ splice site of other ascomycetes. rt-PCR and sequencing confirmed that this 3′ splice site is used to generate an mRNA that is equivalent to the short splice isoform of L. kluyveri (Figure 4A). Although a proximal 3′ splice site is absent in these species, the capacity to encode Ski7-like sequence upstream of the distal 3′ splice site is conserved in these four species. In all four species motif S1 starts with a methionine. Since motif S1 is at the extreme N-terminus of the protein in post-WGD SKI7 genes, we tested the hypothesis that the four Candida species generated a distinct mRNA that uses the AUG codon at the beginning of motif S1 as start codon. Figure 4A shows that 5′RACE indeed identified an mRNA with a 5′ end five nucleotides upstream of the conserved Ski7 motif S1. Thus, C. albicans uses alternative transcription start sites/first exons instead of alternative 3′ splice sites to generate two distinct mRNAs from the single SKI7/HBS1 gene.
The single SKI7/HBS1 gene in basidiomycetes also appears to be alternatively spliced, although the details differ from the ascomycete situation. The Cryptococcus neoformans and Ustilago maydis genes have 9 and 4 annotated exons, respectively. Several EST sequences indicate that exons 4 and 5 in C. neoformans and exon 2 in U. maydis are skipped to generate an Hbs1-like protein (Figure 4B and 4C). Existing annotations suggest that a Ski7-like protein can be encoded by inclusion of these exons. However, EST, RNA sequencing and protein sequence similarity suggests an alternative where the annotated intron 4 of C. neoformans (and intron 2 of U. maydis) is not a true intron (See Text S1). This alternative mechanism encodes a truncated protein that resembles the N-terminus of Ski7, but is missing the GTPase domain (Figure 4B and 4C). This mechanism is strikingly similar to potential alternative splicing of the metazoan homolog (See Text S1 and Figure S5). Overall, while it is clear that the basidiomycetes use alternative splicing of their single SKI7/HBS1 gene, it is not entirely clear which mechanism they use to generate a Ski7-like protein.
The fourth exception to conserved alternative 3′ splice sites in SKI7/HBS1 occurs in the Schizosaccharomyces genus. The Hbs1 protein of Schizosaccharomyces pombe has been previously studied and appears to function similarly to S. cerevisiae Hbs1 [29]. In addition we found an uncharacterized paralog in S. pombe (Systematic name SPAP8A3.05) that encodes amino acid sequence motifs characteristic of Ski7p (labeled S1, S1′ and S3 in Figure S3). We therefore refer to this S. pombe gene as SKI7. Thus, an ancestor to S. pombe must have independently duplicated its SKI7/HBS1 gene. The other three Schizosaccharomyces species with sequenced genomes each contain one clear ortholog of HBS1 and one clear ortholog of SKI7 (Figure S3), which suggests this duplication occurred before the Schizosaccharomyces species diverged from each other. Of the sequenced fungal genomes, the most closely related species with a single SKI7/HBS1 gene is S. complicata. As discussed above, S. complicata has an alternatively spliced SKI7/HBS1 gene, and thus the duplication in the Schizosaccharomyces genus appears to have occurred after it diverged from S. complicata, but before the Schizosaccharomyces species diverged from each other.
Our above observations indicate that S. cerevisiae SKI7 and HBS1 evolved from a single alternatively spliced ancestral gene and that they correspond to the different splice isoforms of the pre-WGD ancestor. This conclusion suggests that a similar mechanism may apply to other alternatively spliced genes. The only S. cerevisiae gene known to use splicing to generate two different functional proteins is PTC7 [20]. The PTC7 gene contains an intron that can either be spliced out or retained. Both the spliced and unspliced mRNAs encode type 2C protein phosphatases (PP2C) [20]. It has previously been noted that an intron of 3n nucleotides without any in frame stop codons is conserved in the PTC7 gene of twelve species within the Saccharomycetaceae, both pre- and post-WGD ([20]; Figure 5 and Figure S4). We searched for PTC7 genes in additional yeast genomes and noticed that the only species within the Saccharomycetaceae that did not follow this pattern is Tetrapisispora blattae. The genome of this species contains two PTC7 genes (which we will call PTC7a and PTC7b). The synteny pattern (http://wolfe.gen.tcd.ie/ygob/) indicates that after WGD the T. blattae lineage maintained both copies of PTC7, while one copy was lost in the S. cerevisiae lineage. We searched for potential introns in PTC7a and PTC7b, but failed to find one in the PTC7a gene, while PTC7b contains a 103 nucleotide intron. The spliced PTC7b mRNA is predicted to encode a functional protein. In contrast to other post-WGD species, translation of the PTC7b unspliced mRNA does not encode a functional PP2C: translation starting from the normal start codon would end after 20 amino acids at a stop codon within the intron, while the only other in frame AUG codon is only 8 amino acid upstream of the normal stop codon. Thus, unlike other Saccharomycetaceae that encode two proteins from one alternatively spliced PTC7 gene, the T. blattae PTC7a and PTC7b genes each can only encode a single protein.
The two splice isoforms of S. cerevisiae PTC7 are targeted to different compartments. The spliced S. cerevisiae PTC7 mRNA encodes a protein that is localized to the mitochondria, while the intron-retained mRNA encodes a protein localized to the nuclear envelope. Targeting to the nuclear envelope has been attributed to a predicted trans-membrane helix (TM) that is encoded by the retained intron [20]. We used the TMHMM 2.0 server (http://www.cbs.dtu.dk/services/TMHMM/) to predict TMs in Ptc7 proteins of various Saccharomycetaceae. Each of the intron-retained mRNAs from post-WGD species and the T. blattae PTC7a gene encodes a single predicted TM near the N-terminus, suggesting that all of these proteins are targeted to the nuclear envelope. In contrast, none of the spliced mRNAs or PTC7b encode a predicted TM. We also used the PSORT II server (http://psort.hgc.jp/form2.html) to predict TMs and protein localization. The TM results agreed with the TMHMM server. In addition, all of the spliced isoforms and PTC7b were predicted to contain a mitochondrial targeting sequence that was absent from the unspliced isoforms. Thus, T. blattae PTC7a encodes a single PP2C that is predicted to be targeted to the nuclear envelope, like the protein encoded by intron-retained PTC7 mRNA in S. cerevisiae, while T. blattae PTC7b gene appears to encode a single PP2C that is predicted to be targeted to the mitochondria, like the protein encoded by spliced PTC7 mRNA in S. cerevisiae. While separate functions of the S. cerevisiae PTC7 splice isoforms have not been defined, these results strongly suggests that the PTC7a and PTC7b genes are subfunctionalized, and thus that subfunctionalization by loss of splicing isoforms is not restricted to SKI7/HBS1.
We describe alternative splicing in fungal SKI7/HBS1 genes that is unusual in two respects. First, unlike in Metazoa, most fungal alternative splicing events do not produce two different proteins, but instead either have no known function or function to quantitatively regulate gene expression. Both of the mRNAs that are produced through SKI7/HBS1 alternative splicing are predicted to encode functional proteins, western blot analysis indicates that both predicted proteins are produced, and complementation of S. cerevisiae mutants shows that the two proteins are functionally distinct. Second, most alternative splicing events that have been described in fungi are not widely conserved but instead have only been described in one species [MDH1, Ref. 22], genus [GND1, Ref. 21] or family [PTC7 Ref. 20]. Besides SKI7/HBS1, the most conserved fungal alternative splicing events were recently reported for PGK1 in the Ascomycota and GAPDH in the Basidiomycota [23]. In contrast, alternative splicing of SKI7/HBS1 most likely arose before the divergence of the Ascomycota from the Basidiomycota, and it might even have arisen before fungi and animals diverged (Text S1 and Figure S5). Thus, this alternative splicing event has been maintained for at least 500 million years. It has been suggested that the Ski7 function is peculiar to S. cerevisiae and close relatives [6], [35]. The finding of ancient alternative splicing indicates that Ski7 function is much older than appreciated and suggests that the ability to produce both Hbs1 and Ski7 is very important to fungi. Interestingly, one of the reasons why conserved alternative splicing in fungi has not been previously reported is that S. cerevisiae and S. pombe have been chosen somewhat arbitrarily as model fungi, and in both of these species the alternatively spliced SKI7/HBS1 gene has been replaced with duplicate genes.
Although alternative splicing of SKI7/HBS1 is conserved in diverse fungi, we have characterized changes in expression strategies for Ski7 and Hbs1, which are summarized in Figure 6A. The common ancestor of the ascomycetes and basidiomycetes appears to have had an alternatively spliced SKI7/HBS1 gene. Although the exact nature of SKI7/HBS1 alternative splicing event in basidiomycetes remains to be determined, it is clear that the mechanism by which a Ski7-like protein is expressed is different. Fully characterizing this event will require additional data from the basidiomycetes and/or additional early branching fungi. Independent duplications in the Saccharomyces and Schizosaccharomyces lineages allowed loss of alternative splicing (Figure 6 events 2). In the Saccharomyces lineage this duplication was part of a WGD, but in the Schizosaccharomyces lineage this duplication appears to be restricted to a single gene. The fourth evolutionary change occurred in the Candida clade in which a single SKI7/HBS1 gene gained an alternative initiation codon for Ski7 (Figure 6 event 3). Interestingly, after duplication, the S. cerevisiae, and S. pombe SKI7 genes also appear to have gained a new initiation codon (see below).
A major mechanism for intron loss in S. cerevisiae involves a transposon-encoded reverse transcriptase that converts spliced mRNA into cDNA. This cDNA then recombines with the gene, resulting in precise deletion of the intron [36]. Multiple sequence alignment shows that the intron in post-WGD Saccharomycetaceae is precisely deleted (Figure S1), consistent with it being deleted by this mechanism. Similarly, the short isoform from Saitoella complicata aligns very well with Hbs1 sequences from four Schizosaccharomyces species, indicating that the alternatively spliced intron was precisely deleted (Figure S3). The Saitoella SKI7/HBS1 gene contains seven introns. Recombination with a cDNA preferentially deletes introns near the 3′ end of the gene while introns near the 5′ end are more likely to be retained [36]. Consistent with intron loss by recombination with cDNA, HBS1 genes from all four sequenced Schizosaccharomyces species contain two introns that correspond to the first two S. complicata introns. Thus, the alternatively spliced intron was precisely deleted in both the Saccharomyces and Schizosaccharomyces lineage, possibly by reverse transcription of the mRNA into cDNA and recombination.
In contrast to HBS1, the intron in post-duplication SKI7 genes was not precisely deleted. Multiple sequence alignment (Figure S1) of post-duplication Saccharomycetaceae showed that although the Ski7 N-terminus is generally poorly conserved, it contains three conserved sequence motifs (motif S1, S2, and S3). In post-WGD SKI7 genes in both the Saccharomycetaceae and in Schizosaccharomyces, motif S1 is located at the extreme N-terminus of Ski7, starting with the Met translated from the start codon. This is most consistent with the model that after duplication S. cerevisiae SKI7 lost exon 1 and gained a new initiation codon. Similarly, Schizosaccharomyces Ski7 appears to have lost exons 1 to 4, and gained a new initiation codon. This deletion of both the SKI7 intron and the first exon(s) in Saccharomyces and Schizosaccharomyces is inconsistent with simple recombination with a cDNA.
It has previously been noted that a significant number of the genes that were duplicated and retained in S. cerevisiae after WGD are also duplicated in S. pombe [37], suggesting parallel subfunctionalization events in the two species. Our observations provide striking similarities of the evolution of the Hbs1 protein in Saccharomycetaceae and Schizosaccharomyces and of the Ski7 protein in these same clades and in Candida albicans. Thus, after independent duplication in these lineages, a similar sequence of changes occurred, intron deletion through recombination (HBS1) or generation of an alternative start site (SKI7).
The only known example of a S. cerevisiae gene encoding functional proteins from both unspliced and spliced mRNA is PTC7. This capacity to encode two proteins is conserved in most Saccharomycetaceae, but not in other Saccharomycetales (including Candida, Pichia and Yarrowia species) [20]. The time of divergence of the Saccharomycetaceae has not been carefully defined, but estimates indicate that it preceded divergence of mice from humans 75 million year ago. Strikingly, only about 30% of alternative splicing events are conserved from mice to human. Therefore, although alternative splicing of PTC7 is not nearly as well conserved as that of SKI7/HBS1, it still has been conserved for a longer period than many human alternative splicing events.
Since PTC7 homologs outside the Saccharomycetaceae lack an intron in the same position, the alternatively spliced PTC7 intron appears to have been gained by an ancestor of the Saccharomycetaceae (Figure 6B event 4). The PSORT server (http://psort.hgc.jp/form2.html) predicts that Ptc7 proteins outside the Saccharomycetaceae (i.e. from Candida, Pichia, and Yarrowia species) localize to the mitochondria. Thus, the gain of an intron and the alternative splicing of this intron provides the Saccharomycetaceae with a PP2C localized to the nuclear envelope. After WGD (Figure 6 event 5), one copy of PTC7 was lost in the Saccharomyces lineage and the remaining copy maintained the capacity to encode two proteins (Figure 6 event 6). In contrast, in T. blattae both duplicated PTC7 genes were maintained, but each lost the ability to encode two PP2C splice isoforms (Figure 6 event 7) and subfunctionalized into one gene for a mitochondrial PP2C and one gene for a PP2C in the nuclear envelope.
Our combined bioinformatic and experimental analysis shows that alternative splicing and gene duplication may be interrelated events in a cycle that diversifies the proteome. In this cycle, gain of alternative splicing, duplication, and loss of alternative splicing and subfunctionalization result in functionally distinct paralogs. Although there have been some previous descriptions of subfunctionalization by loss of alternative splicing [e.g. ref 24], our findings extend these descriptions in three important ways. First, previous descriptions are generally limited to two closely related species and thus cover only part of the evolutionary history of the gene. The ever-increasing number of sequenced fungal genomes allowed us to analyze SKI7/HBS1 and PTC7 gene structure and expression in diverse fungi thereby identifying when alternative splicing arose and was lost. The whole cycle of gain of an alternative splicing event, duplication, and loss of alternative splicing can be observed in the PTC7 gene of Saccharomycetaceae. In contrast, although we have not been able to identify when alternative splicing of SKI7/HBS1 arose, we have described independent subfunctionalization events by loss of alternative splicing in the Schizosaccharomyces and Saccharomyces lineages.
Second, our observations suggest that subfunctionalization by loss of alternative splicing occurred very similarly in the Saccharomyces and Schizosaccharomyces lineages. Thus, unlike previously described isolated examples this phenomenon appears to have occurred multiple times.
Third, in most previously described cases of loss of alternative splicing in duplicated genes it was not clear whether the splicing isoforms have distinct functions, and thus it is not clear in those cases that subfunctionalization and loss of alternative splicing are causally linked. Similarly, lack of one PTC7 splice isoform does not cause an easily identifiable phenotype under lab conditions, making it impossible to test whether the duplicate T. blattae genes can substitute for one but not the other splice isoform. In contrast, SKI7 and HBS1 have well-described functions, allowing us to demonstrate that the splice isoforms of L. kluyveri SKI7/HBS1 are functionally distinct.
The S. cerevisiae, C. albicans, and wild-type L. kluyveri strains have been described [11], [38]. The L. kluyveri ura3 mutant strain FM628 was a kind gift of Mark Johnston. The S. complicata type strain Y-17804 was obtained from the USDA ARS culture collection.
A draft sequence of the S. complicata genome is available [39]. BLAST analysis using the S. pombe Hbs1 identified two non-overlapping contigs that encoded N- and C-terminal parts of a S. complicata homolog. Extensive BLAST analysis with other queries did not reveal additional homologs. We hypothesized that these contigs represented different parts of the same gene. We used PCR to close the gap between the contigs and sequenced the PCR product directly. The assembled sequence of the S. complicata SKI7/HBS1 gene has been submitted to Genbank (Accession number JQ928880). Since exons proved difficult to predict due to their small size, we sequenced rt-PCR products to determine the gene structure depicted in Figure 3B, which was then used to align the encoded protein with Schizosaccharomyces homologs.
L. kluyveri and S. complicata were grown in YPD and RNA was extracted using our standard method for S. cerevisiae. C. albicans growth and RNA extraction was performed as described [40]. Aspergillus RNA isolated from strain R21 was a kind gift from Taylor Schoberle and Greg May (UT MD Anderson Cancer Center). When contamination with genomic DNA proved to be a problem, we treated the RNA with DNase (Promega).
rt-PCR was done using a commercial kit per manufacturer's instructions (Sigma-Aldrich). To close the gap between the two S. complicata contigs, we used the same kit but omitted the reverse transcriptase to amplify genomic DNA. 5′ RLM-RACE was done using a commercial kit per manufacturer's instructions (Invitrogen). All PCR, rt-PCR and RACE products were sequenced directly (Genewiz) and exactly confirmed the predicted splice sites.
pAv231 has been described [11]. It contains the L. kluyveri SKI7/HBS1 gene, including the promoter, intron and 3′UTR sequences. pAv844 and pAv847 have the long form and short form of the intron removed, respectively. They were generated by overlap PCR using the forward overlap oligonucleotides oAv963 (tgctcaaccaaagcaagaag aagagaagaaattatctaaactgg) for pAv844 and oAv965 (tgctcaaccaaagcaagaag ccaaaaaacaagctatctctaatttc) for pAv847 and their reverse complements oAv964 and oAv966. To generate the HA-tagged SKI7/HBS1 gene, a plasmid encoding the C-terminus and triple HA tag was chemically synthesized (by Genewiz). This fragment was used to replace the Bcl I to Bgl II restriction enzyme fragment of pAv231, generating pAv888, which contains the entire L. kluyveri SKI7/HBS1 gene with a C-terminal triple HA tag. A Bam HI Xba I fragment of pAv888 was then used to replace the Bam HI Xba I fragment of pAv844 and pAv846 to generate pAv903 and pAv905. Thus, pAv903 encodes a C-terminally HA-tagged version of the long Ski7-like isoform of L. kluyveri SKI7/HBS1, while pAv905 encodes the tagged short Hbs1-like isoform.
Plasmids carrying the HA-tagged L. kluyveri SKI7/HBS1 gene, or empty vector controls, were transformed into S. cerevisiae or L. kluyveri strains using a standard method [41]. Transformants were selected on SC-URA, and then grown overnight in SC-URA. Total protein was isolated using the glass bead method and analyzed by western blotting using anti-HA antibodies (Roche). As a control for the western blotting we used a S. cerevisiae strain with the HA epitope integrated at the C-terminus of the endogenous SKI7 locus.
Fungal SKI7 and HBS1 homologs were initially identified by BLAST in the sequenced genomes and the predicted proteomes from 14 Saccharomycetaceae, 10 species from the CTG clade, 4 Pezizomycotina, 5 Taphrinomycotina and 2 basidiomycetes. In none of the cases was the gene annotated as alternatively spliced, and in a number of cases introns were not annotated or incorrectly annotated and were corrected based on our rt-PCR analysis. Multiple sequence alignments of various subsets of protein sequences were generated with the help of the ClustalW (http://www.ch.embnet.org/software/ClustalW.html) BOXSHADE (http://www.ch.embnet.org/software/BOX_form.html), and WebLogo (weblogo.berkeley.edu/) servers. The species trees in Figure 6 and Figures S2 and S3 are adapted from [42], [43], and [44].
|
10.1371/journal.pgen.1008083 | The dPix-Git complex is essential to coordinate epithelial morphogenesis and regulate myosin during Drosophila egg chamber development | How biochemical and mechanical information are integrated during tissue development is a central question in morphogenesis. In many biological systems, the PIX-GIT complex localises to focal adhesions and integrates both physical and chemical information. We used Drosophila melanogaster egg chamber formation to study the function of PIX and GIT orthologues (dPix and Git, respectively), and discovered a central role for this complex in controlling myosin activity and epithelial monolayering. We found that Git’s focal adhesion targeting domain mediates basal localisation of this complex to filament structures and the leading edge of migrating cells. In the absence of dpix and git, tissue disruption is driven by contractile forces, as reduction of myosin activators restores egg production and morphology. Further, dpix and git mutant eggs closely phenocopy defects previously reported in pak mutant epithelia. Together, these results indicate that the dPix-Git complex controls egg chamber morphogenesis by controlling myosin contractility and Pak kinase downstream of focal adhesions.
| A major challenge in biology is to identify the genes and processes that build tissues of correct shape and function. Recently, transmission of mechanical forces through cell adhesions, and control of cell tension via contractile force-generating proteins, have emerged as fundamental to tissue development. Currently, we do not understand how these separate processes are integrated. We gained new insight into morphogenesis and control of contraction through adhesion-localised proteins, by studying mutants of the dPix-Git focal adhesion complex in Drosophila melanogaster egg chambers, a 3D model of tissue morphogenesis. We found that the dPix-Git complex is essential to maintain cell monolayers, and is a regulator of the contractile force-generating protein, myosin. In the absence of the dPix-Git complex, irregular myosin activation led to tissue disruption, however modest suppression of myosin activators rescued this defect. Remarkably, the dPix-Git complex is essential for egg chamber development, but appears dispensable for other D. melanogaster epithelia, indicating the mechanisms that couple adhesion signalling and cell contractile forces are tissue specific. Our study reveals a key molecular link between cell adhesions and contraction, and the conservation of the dPix-Git module in metazoans suggests this mechanism is likely to be evolutionarily conserved in other tube-like tissues.
| Organogenesis requires the coordinated integration of biochemical and mechanical information at the level of cells and their neighbours, as well as across entire tissues [1,2]. Drosophila melanogaster (D. melanogaster) egg chamber development has emerged as a powerful system for identifying myriad developmental processes, and understanding how they interact at the cell and tissue scales [3–5]. Egg chambers originate from stem cells in a structure called the germarium (Fig 1A and 1B), and progress through 14 stages of development. New egg chambers continually ‘bud’ from the germarium and remain connected by stalk structures, like ‘beads on a string’ (Fig 1A and 1B). These egg chambers consist of a monolayer of somatic follicular epithelial cells encapsulating a cyst of growing germline cells. Egg chamber follicular epithelial cells secrete basement membrane proteins to form an extracellular matrix (ECM), which surrounds the growing organ. Over a period of days the egg chamber grows by approximately three orders of magnitude. This growth is preferentially channeled along the egg chamber’s anterior-posterior axis to form a 2–3 fold elongated ellipsoid of approximately 850 cells, which establishes the foundations of the embryonic body plan. During this period of growth and elongation the somatic follicle cells maintain apical-basal polarity [6] and tissue monolayering to preserve egg chamber function and integrity [7–9].
A series of studies have revealed strict spatial and temporal regulation of the diverse roles of non-muscle myosin II (myosin), a contractile force-generating protein, during egg chamber development [10–15]. Amongst other roles, apical myosin provides mechanical resistance to pressure generated by the growing germline cyst early in egg development [14], and drives cell contraction and tissue elongation at the early stages of egg chamber maturation [15]. Later in development, basal myosin oscillations are deployed to directionally constrict growth along the anterior-posterior axis [10]. Landmark studies have also revealed that during early stages (stages 1–8), follicle cells undertake collective epithelial sheet migration and this migration is required for egg chamber elongation [16,17]. Collective cell migration relies on the formation of actin-based protrusions at the leading edge of each follicle cell and contributes to elongation by generating tissue level polarity in basal actin and extracellular matrix fibres [16–18].
Previous genetic studies of egg chamber development identified a similar set of defects in cells lacking different focal adhesion proteins (which engage the ECM), and the sterile-20 kinase p21-activated kinase (Pak). These defects include loss of monolayering [7,19], irregularities in the ordering of basal actin filaments [20–22], and altered myosin activity [10,22]. In addition, in a range of cell culture and animal systems, signal transduction from focal adhesions has been linked to the activity of p21-activated kinase family (PAK) proteins via the oligomeric PIX-GIT complex which includes the P21-activated kinase interacting exchange factor (PIX) proteins, and the G-protein coupled receptor kinase interacting proteins (GIT) [23–25]. PIX proteins are RhoGEFs, and GIT proteins are ArfGAPs, and each homodimerise and heterodimerise to form the PIX-GIT signalling scaffold [26–29]. The PIX-GIT complex binds numerous proteins at a range of subcellular locations. Prominent amongst these, PIX and GIT have been identified as part of the ‘core’ integrin based adhesome [30]. In this context, GIT targets the entire complex to focal adhesions [31], while PIX in turn recruits PAK to focal adhesions for regulation of activation [32,33].
In mammals, the PIX and GIT genes are each duplicated, giving rise to α-PIX/β-PIX, and GIT1/GIT2, generating the potential for redundancy and a practical barrier to genetic studies. In contrast, D. melanogaster has a single copy of each gene, making it an ideal system to interrogate their function. In D. melanogaster the PIX and GIT orthologues are referred to as dpix and git respectively. D. melanogaster studies have revealed that dPix and Git have roles that are distinct from each other in the context of neuronal function [34,35], whilst they appear to co-operate during muscle morphogenesis [36] and also work together to control Hippo pathway dependent tissue growth [37]. To further understand the role of the dPix-Git signalling module in organogenesis, we examined dpix and git mutant D. melanogaster and focused on defects that were common to both mutations. dpix and git were each required during egg chamber development for cell intercalation, correct myosin activation, and to maintain a follicular epithelial monolayer. In the absence of dpix and git, aberrant myosin activation generated force anisotropies that led to cell deformation and loss of epithelial integrity. Interventions designed to mildly suppress myosin activity were sufficient to rescue egg production and elongation defects. The dPix-Git complex was predominantly basally localised, with a planar polarised enrichment at basal filament structures toward the leading edge of cells undergoing collective migration. Altogether, this study identifies a tissue specific and essential requirement for the dPix-Git complex in egg chamber development.
To further understand the role of the dPix-Git signalling module in organ development, we examined dpix (CG10043, also known as rtgef) [34] and git (CG16728) mutant [36] D. melanogaster and looked for shared developmental defects. As previously reported, homozygous mutations for dpix and git were each semi-lethal [34,36]. Surviving adults emerged with ‘crumpled’ wings but were otherwise grossly morphologically normal (S1A Fig). Notably, both dpix and git adult females produced very few eggs. This prompted examination of dpix and git ovaries and revealed that while early stage egg chambers were present, these generally did not progress to maturity (Fig 1C). To test if dpix and git function redundantly in egg chamber development we generated dpix, git double mutant animals. These animals were also semi-viable and eclosed with crumpled wings (S1A Fig), but produced no mature eggs (Fig 1C). This indicates that the dPix-Git signalling complex is essential for egg development, and the more severe phenotype in double mutants compared to each single mutant suggests that dpix and git have partially independent roles. Taken together, these results show that dPix and Git are essential regulators of egg production, whereas they appear to be dispensable for the development of most D. melanogaster organs.
To understand the developmental processes that are misregulated in the absence of dpix and git, we studied the anatomy of mutant egg chambers and found that monolayering of the follicular epithelium was disrupted from as early as stage 3 (Fig 1D–1G’ and see S1B–S1B” Fig). Markers of cell polarity such as E-cadherin were both elevated and mislocalised in supernumerary follicular epithelial layers (Fig 1D–1G and S1C and S1D Fig). In particular, while E-cadherin was apically enriched and also present on lateral junctions in wild-type tissue, it was distributed around the entire cell in the extra follicular epithelium layers of dpix and git mutants. This appears similar to the loss of cell polarity seen in extra follicular epithelium layers of α-spectrin mutant animals [9].
We used the FLP/FRT system to generate clones of dpix or git mutant tissue in otherwise heterozygous animals, and found that loss of dpix or git causes multilayering of the follicular epithelium (Fig 1H–1I’), even in small clones (S1G Fig), indicating that the requirement for dpix and git is autonomous to the follicle cells. Interestingly, while PIX and GIT proteins are known to influence focal adhesion turnover and maturation in cell culture systems [38–41], we did not see an obvious effect on the accumulation of the focal adhesion component talin in monolayered dpix or git mutant tissue (Fig 1H–1I’), however talin levels were often increased in multilayered tissue (S1E and S1F Fig). We did not see a general disruption of apical-basal polarity in monolayered clones, as indicated by unchanged apical localisation of β-heavy-spectrin in dpix clones (S1H–S1H’ Fig).
Follicular epithelium cells are mitotic until stage 6, after which they begin Notch signalling mediated endo-replication [42]. Previous studies of egg chambers have shown that multilayering can be caused by excessive proliferation and re-entry into mitosis [9,43–45], therefore we assessed the presence of mitotic cells in dpix and git homozygous mutants using phospho-Histone H3 staining. We observed no increase in proliferation before stage 6 (S2A Fig), but found prolonged cell divisions occurring in the already multilayered tissues of dpix and git egg chambers between stages 7 and 9 (Fig 2A–2D and S2A and S2B Fig). This raised the possibility that Notch signalling was defective. To investigate this we stained for Eyes Absent, which is normally suppressed by Notch signalling after stage 6 [46], and found increased expression in ectopic cell layers (S2C–S2G” Fig). FasIII is another protein suppressed by Notch after stage 6 [47], and was also increased in multilayered cells (S2H–S2J” Fig). The restriction of ectopic proliferation and Eyes Absent expression to multilayered cells suggests that Notch deregulation and ectopic proliferation are a consequence of multilayering which may exacerbate this phenotype, but are not likely to cause epithelial multilayering in dpix and git mutants.
In addition to multilayering of the follicular epithelium we observed multiple cell intercalation defects in dpix and git mutants. First, both dpix and git mutants showed intercalation defects in the stalk structures that joined consecutive egg chambers (S3A–S3C Fig). Second, we found compound egg chambers containing more than 15 germline cells (S3D–S3F Fig), a defect that has previously been associated with disrupted migration of pre-follicular cells between germline cysts [48]. Finally, we observed egg chamber fusions in some ovarioles (Fig 2E–2I) which we confirmed as side-by-side fusion of egg chambers by oocyte staining (S3G–S3H” Fig). The side-by-side fusion phenotype is very reminiscent of a myosin-mediated cell intercalation defect that was previously reported in developing ovaries of pak mutant animals [49]. This specific phenocopy of pak mutants, combined with the well characterised physical interaction between PIX-GIT and PAK proteins, suggest that a dPix-Git-Pak signalling axis regulates egg chamber morphogenesis (Fig 2J).
In order to understand the effect of dpix or git mutation on overall egg chamber morphology, we measured the length and width of egg chambers between stages 3 and 10 (Fig 2K). dpix and git egg chambers with side-by-side fusions or compound germline cysts were excluded from analysis. This allowed us to view the relationship between egg chamber width and length independent of stage, and revealed that dpix and git mutant egg chambers tended to be wider than wild-type egg chambers of similar length (Fig 2K). Next, we developmentally staged wild-type, dpix and git egg chambers between stages 4–8 using DAPI based features [50], and compared their aspect ratio (length/width) (Fig 2L). Even from early stages, dpix or git mutant egg chambers had a reduced aspect ratio compared to wild-type (Fig 2L) and this persisted through to later stages (Fig 2L–2O). The observed effect on aspect ratio from early stages of elongation is consistent with pak mutants [15], but contrasts with other mutants that affect egg aspect ratio, such as lar, which appear to diverge from wild-type at later stages [20,51].
To investigate whether overall changes in egg chamber shape are associated with cell shape changes, we imaged cortical actin at the basal regions of the follicular epithelium at stage 7–8 (Fig 2P–2Q) and measured a range of cell morphological characteristics (see materials and methods). By stage 7–8, cells in dpix and git homozygous mutant tissues had greater directional elongation than wild-type (greater eccentricity) (Fig 2R–2T), indicating that loss of dpix or git at the whole tissue scale produced cell shape distortions within the follicular epithelium.
Non-muscle myosin II (myosin) is a contractile force-generating protein. The C-terminus of myosin molecules can self-associate to form bipolar filaments that simultaneously bind two anti-parallel actin filaments and ‘pull’ them towards one another. When actin filaments are attached to cell membranes via adherens junctions, or to basal membranes via focal adhesions, contraction from myosin generates forces which drive cell movement and shape changes [52]. Two reasons led us to examine the role of dPix and Git in regulating myosin activity. First, the data above suggest that a Git-dPix-Pak signalling axis functions in follicular epithelia, and Pak is known to antagonise myosin activation in this tissue [22,49]. Second, we had seen changes to the elongation and shape of dpix and git mutant epithelial cells near their basal surfaces (Fig 2P–2T). We considered that these deformations could be caused by deregulated myosin activity, resulting in disorganised cell membrane contractility and tension.
To test for a cell autonomous defect in myosin activation we stained dpix or git mosaic egg chambers with antibodies against phosphorylated Ser19 (Ser21 in D. melanogaster) of myosin regulatory light chain (pMRLC), and examined the intensity and distribution of this activated form of myosin. We examined dpix or git clones that had maintained monolayering and found increased myosin activation at the basal membrane compared to neighbouring wild-type cells (Fig 3A–3E’). While myosin was hyperactivated in both dpix and git clones, we saw some evidence of non-cell autonomous effects at clone boundaries which may be caused by the ability of altered mechanical properties in one cell to affect myosin in neighbouring cells [53]. Additionally, we also observed a cell autonomous increase in pMRLC at the basal membranes of cells in clones which had resulted in epithelial multilayering (shown for git in Fig 3F–3F’).
To understand how removal of the entire dPix-Git complex affects myosin activation we examined dpix, git double mutants and found that pMRLC was increased at both the apical and basal membranes of follicular epithelial cells (Fig 3G–3J). To avoid complications from loss of apicobasal cell polarity, as has been reported in pak mutants [54], we only examined regions of mutant epithelia that had maintained monolayering. We quantified these observations and found an approximately 2 fold increase in pMRLC at basal membranes, and a 1.5–2 fold increase in pMRLC signal at the apical membrane (Fig 3K). To determine if the effect on myosin was specific, and as a control for optical sectioning of the tissue, we also imaged E-cadherin and found no significant difference in apical or basal cell regions (Fig 3L). Therefore, loss of both dpix and git led to myosin activation and this could not be explained by a general effect of enrichment of proteins at membranes or cell junctions. The basal deregulation of pMRLC in dpix or git clones, compared with a combined basal and apical deregulation of pMRLC in dpix, git homozygous mutants suggests that the basal deregulation of pMRLC is a primary cell biological defect of disrupting the dPix-Git complex, while the apical misregulation of pMRLC may be a secondary effect of alterations in tissue wide morphogenetic processes.
We also found cells in dpix, git double mutants where activated myosin was localised to lateral cell junctions and where this loading of activated myosin was associated with almost complete contraction of the length of lateral junctions (Fig 3H–3H’ and 3J, and arrow in 3H–3H’). Imaging optical sections across the basal surface of dpix, git double mutants suggested that the de-regulation of myosin contractility generated force anisotropies that resulted in disorganisation of the epithelium and compromised epithelial integrity (Fig 3M–3N”). The basal surface of dpix, git tissues displayed intense accumulations of activated myosin at the center of aberrant multicellular junctions (arrowhead in Fig 3N”). We also saw purse-string like contractions of pMRLC at the center of cell rosettes that appeared to be ‘holes’, representing a breach of the continuity of the follicular epithelium (Fig 3N–3N”, and asterisk in Fig 3N”). Taken together these data indicate that loss of dpix and git results in dysregulated myosin activity, which causes cell deformation (compare Fig 2P and 2Q) and disruption to tissue integrity (Fig 3M and 3N).
To test whether the ectopic myosin activation we detected could be a major cause of impaired egg chamber development, we set out to reduce the aberrant myosin activity in dpix and git mutants. To do this we generated animals that were homozygous mutant for dpix or git, and heterozygous for either of the canonical activators of myosin, rhoGEF2 or rho1 (Fig 4A). We selected these genes in line with the hypothesis that a Git-dPix-Pak signalling axis regulates egg chamber development; and following previous work showing that heterozygosity for rho1 and rhoGEF2 rescued egg production in pak mutants [22,49], and that RhoGEF2 regulates Rho1 which in turn regulates myosin in the basal region of egg chamber follicle cells [10,11]. Heterozygosity for rhoGEF2 was sufficient to increase egg production in both dpix and git mutants (Fig 4B, with representative images for dpix in Fig 4C). Similarly, heterozygosity for rho1 increased egg production in dpix animals but to a lesser degree than rhoGEF2. rho1 heterozygosity also raised average egg production in git mutants but this increase was not statistically significant for the number of animals analysed (Fig 4B). As rhoGEF2 heterozygosity conferred a robust increase of mature egg production in dpix and git mutants we engineered animals that were homozygous mutant for both dpix and git, and heterozygous for rhoGEF2. Strikingly, although dpix, git mutants never produced mature eggs, when we halved the gene dose of rhoGEF2 egg production was often restored (Fig 4B).
While the major defect of dpix and git mutants is reduced egg chamber viability, we also noticed that surviving mature eggs had an elongation defect (Fig 4D and 4E and S3I and S3J Fig). Remarkably, heterozygosity for rhoGEF2 significantly increased egg aspect ratio in git mutants, and generated an average aspect ratio of dpix mutant eggs comparable to wild-type (Fig 4D and 4E). Taken together, the increase in egg production, restoration of egg production in dpix, git mutants, and normalization of egg morphology indicate that aberrant myosin activation in dpix and git mutant animals has severe physiological consequences for egg chambers, and is one of the major defects in the absence of the dPix-Git complex.
Having established an essential role for the dPix-Git complex in egg chamber development, and a central role in regulating epithelial monolayering, cell intercalation and myosin inhibition, we wanted to investigate the expression and localisation of these proteins during development. Therefore, we used CRISPR-Cas9 genome editing to add fluorescent tags to the 3’ end of the endogenous dpix and git open reading frames, creating dpix-venus and git-tag Red Fluorescent Protein (tRFP) transgenic animals, which were viable with no obvious defects. dPix-Venus and Git-tRFP were expressed in the germ line and follicular epithelium throughout ovary development with a basally enriched localisation in follicular epithelium (Fig 5A and 5B). We characterized this in stage 8 follicular epithelia, where dPix-Venus and Git-tRFP had a predominantly basal localisation (Fig 5A and 5B), but also accumulated to a lesser degree at apical membranes (Fig 5B). dPix-Venus and Git-tRFP appeared to co-localise and, consistent with this, the signal from these proteins co-varied (Fig 5B). Strikingly, on the basal side of the follicular epithelium dPix-Venus had a polarised localisation and was localised onto filament like structures, that extended perpendicular to the axis of egg chamber elongation (Fig 5C).
Our endogenously tagged dpix and git strains revealed a predominately basal localisation of this complex. However, endogenous expression levels produced very low fluorescent signals, and in an effort to better visualise localisation to subcellular structures we also generated animals expressing transgenes encoding C-terminal green fluorescent protein tagged dPix (dPix-GFP) and Git (Git-GFP), under the control of the ubiquitin promoter (ubi>) [55]. The dpix locus is annotated as encoding seven transcripts, and here we chose to clone a cDNA encoding RtGEF-A (dPix-A) as it contains all major conserved protein domains. To confirm that these transgenes are functional, and to avoid competition between dPix-GFP, Git-GFP and endogenous dPix and Git we recombined each of these transgenes into a corresponding mutant background.
Confocal imaging between stages 4 to 8 revealed a striking planar polarised localisation of dPix-GFP across the entire basal surface of the follicular epithelium (Fig 5D–5D”). This mirrored our observations with endogenously tagged dPix, but was more readily observable. We found two distinct polarised localisations of dPix-GFP. First, consistent with our observations for endogenous dPix-Venus, we saw that in each cell dPix-GFP localised along filament like structures that resembled basal actin filaments, perpendicular to the axis of elongation in each egg chamber (Fig 5E–5G). Second, during stages of collective cell migration, dPix-GFP was highly enriched towards one cell membrane and the direction of polarised localisation of dPix-GFP was shared between every cell in a single egg chamber (Fig 5E–5G). Further, dPix-GFP punctae were observed at the ends of dPix-GFP filament structures (Fig 5E–5G). To determine whether dPix-GFP localised towards the trailing or leading edge of these cells, we live imaged egg chamber rotation and found that dPix-GFP accumulated most strongly immediately behind the leading edge membrane of migrating follicle cells (S4A Fig). Notably, the planar polarised localisation of dPix-GFP was lost at stage 9 (Fig 5H) following the termination of collective migration. A similar pattern of localisation was seen for Git-GFP, however, enrichment at the leading edge was less pronounced (Fig 5I).
Given that dPix and Git often function in a complex and are able to influence each other’s localisation in cell culture [31,56], we assessed whether Git was responsible for targeting dPix to basal filaments and the leading edge. To test this we made clones of git in the context of dPix-GFP expression and found that loss of git prevented dPix-GFP localisation to both structures at the basal region of cells (Fig 5J–5J’). Having seen that Git controls the basal localisation of the dPix-Git complex we also wanted to know what determined Git localisation in this system. In mammals, GIT1/2 contain a C-terminal paxillin binding domain which targets GIT proteins to the focal adhesion protein paxillin, allowing for recruitment of PIX and PAK [25,31]. This paxillin binding domain appeared conserved in D. melanogaster, and so we removed the C-terminus of Git (ΔPBD-Git-GFP), and found that while Git-GFP accumulated in the basal region of cells, any basal GFP enrichment above average levels found throughout the cell was lost in ΔPBD-Git-GFP (Fig 5K–5M).
The basal and leading edge localisation of dPix and Git prompted us to look for defects in actin alignment which are indicative of defects in follicle cell migration [57]. We found incompletely penetrant defects in the global alignment of basal actin filaments in stage 8 git egg chambers (S4B and S4C Fig). However, live imaging showed conclusively that egg chamber rotation continues in dpix and git mutants (S4D–S4F Fig), and the speed of rotation is in fact enhanced in git mutants (S4F Fig). Collectively, these experiments reveal that dPix and Git are predominantly basally localised, and that Git’s paxillin binding domain targets the entire complex to basal filaments, and the leading edge of each migrating cell. They further suggest that Git plays a role in limiting the speed of follicle cell migration.
Given the central role of dPix-Git in egg chamber development, we wanted to know how this signalling module is activated and how it activates downstream effectors such as Pak. To begin to dissect the signalling mechanisms of this complex we compared the phenotypes of wild-type dPix-GFP and Git-GFP transgenes with a series of structure function alleles (Fig 6A) we had generated and incorporated into identical genomic locations via the phiC31 integrase to ensure even expression. In each case dpix and git rescue transgenes were expressed in the corresponding mutant background.
First, we confirmed that ubiquitously expressing a single copy of wild-type dPix-GFP or Git-GFP in the corresponding mutant background, produced healthy adult D. melanogaster and also increased egg production (Fig 6B and 6C). Next, to investigate whether localisation of dPix and Git to focal adhesions was functionally important for egg chamber development we used mature egg production as an assay (Fig 6B and 6C). We counted the number of mature eggs in females and found that although Git’s focal adhesion targeting domain was not essential for Git function in egg development, removing this domain (ΔPBD-Git-GFP) reduced the average number of mature eggs per female by about 30 percent (Fig 6B). To investigate signalling downstream of dPix and Git, we removed the dPix SH3 domain (ΔSH3-dPix-GFP), as the SH3 domain has been shown to mediate the interaction of PIX and PAK proteins [33,58]. ΔSH3-dPix-GFP expression in dpix mutant animals increased egg production to a similar level as full length dPix-GFP, suggesting that Pak binding via the SH3 domain is not essential for much of dPix function in egg chamber development (Fig 6C). Additionally, removing the dPix SH3 domain did not noticeably affect its localisation (S5A–S5E Fig, and compare to Fig 5D–5G). As dPix is a putative activator of Rac1 and Cdc42 GTPases via its RhoGEF domain, we also mutated a key serine residue that is conserved as either serine/threonine in RhoGEF domains, and is known to be required for activation of Rac1 in vitro [59]. Expression of a transgene encoding dPix where serine 89 was mutated to glutamic acid (rtGEF*dPix-GFP) rescued egg production in dpix mutants similar to full length dPix-GFP (Fig 6C), indicating that this conserved residue is not essential for dPix function in egg chamber development. Thus, although these experiments do not provide clarity on the molecular mechanism by which dPix functions in egg development they do suggest the possibility of redundant signalling mechanisms of dPix in this tissue.
To further investigate the requirement of focal adhesion targeting of Git in egg chambers, we compared the frequency of morphological defects in homozygous Git-GFP and ΔPBD-Git-GFP expressing D. melanogaster. We saw that removal of focal adhesion targeting resulted in a large increase in rates of multilayering (Fig 6D–6D’). We also found that expression of ΔPBD-Git-GFP led to compound egg chamber defects where developing egg chambers contained more than the normal complement of 15 germline nurse cells (Fig 6E–6E’). Thus loss of focal adhesion targeting increases the rate of defects normally seen in git homozygous mutants, implicating paxillin binding in the maintenance of monolayering and germline cyst encapsulation.
Precise subcellular control of the contractile force-generating protein myosin has emerged as one of the most important determinants of tissue morphogenesis and organ development [60–63]. While much is known about apical regulators of myosin [64–67], less is known about the basal control of myosin activity in developing tissues [10–12]. Here we have identified an essential requirement for the basal dPix-Git complex in epithelial morphogenesis during egg chamber development, and revealed a major physiological role for this complex in the spatial control of myosin activation. The basal localisation of the dPix-Git complex relies on Git’s focal adhesion targeting domain, and loss of this targeting reduced egg production, suggesting that focal adhesions are a key site of dPix-Git activity. Additionally, the precise phenocopy of a side-by-side egg chamber fusion defect that to our knowledge has only been reported in pak mutants, suggest that a major part of dPix-Git function is mediated through Pak. Taken together our data indicate that the dPix-Git complex functions with Pak to respond to information sensed at focal adhesions during egg chamber development.
We have shown that loss of either dpix or git leads directly to aberrant myosin activation and that misregulated myosin is a major physiological cause of defective development of dpix and git egg chambers. Loss of dpix and git leads to force anisotropies at the basal surface of epithelia and collapsed lateral membranes. In dpix and git cells, myosin activation is likely to be a consequence of lowered Pak activation [22,49]. Thus, our results support the role of Pak in antagonising myosin during egg chamber development, and also identify a potential mechanism for both the basal localisation of Pak and spatial control of Pak activation. The regulation of myosin is critical to the morphogenesis of egg chambers. This protein of diverse functions requires different sets of regulators in order to be manipulated into an extraordinary range of emergent behaviours, often in different compartments of the very same cell. In early stage egg chambers myosin is required apically to resist the pressure of the growing germline cyst which would otherwise deform the follicular epithelium [14]. There is now also evidence that apical-medial myosin generates pulsatile contractions in the earliest stages of egg chamber elongation [15]. At the basal cell surface during the processes of collective cell migration, retrograde flow of myosin has been reported opposite to the direction of cell movement [13]. At the cessation of migration (stage 9), basal myosin switches from retrograde flow to patterns of asynchronous oscillations controlled by an integrin-ROCK cascade [10,11]. Given our identification of dPix and Git as key regulators of myosin inhibition in egg chambers, it will be important to determine which subsets of these dynamic processes dPix and Git control.
In addition to increasing mature egg production, we also found that halving the gene dose of myosin activators could rescue the aspect ratio of mature dpix and git eggs. Two potential causes for this short mature egg phenotype seen in dpix and git mutants are defects in nurse cell dumping, and defects in vitellogenesis [68]. Each of these phenotypes can also be explained by alterations in myosin contractility. In the case of nurse cell dumping, this defect would be in the germline nurse cells. In the case of vitellogenesis this defect would be in the muscle sheath which encloses developing egg chambers. Taken together this elongation data is consistent with a physiological role for dpix and git in regulating myosin in cells beyond the follicular epithelium.
Given our evidence that a dPix-Git-Pak module potently limits the activation of myosin, it is important to understand the molecular mechanism of myosin regulation downstream of this complex. Interestingly, while mammalian studies have shown that PAK1 can inactivate myosin by inhibiting myosin light-chain kinase (MLCK) [69], D. melanogaster MLCKs that mediate Pak’s role in egg chamber development have not been identified [22]. However loss of pak has been shown to lead to a redistribution of RhoGEF2 within follicle cells [22], highlighting misregulation of Pak and RhoGEF2 as a plausible mechanism for the deregulation of myosin in the absence of dPix and Git. The importance of targeting of dPix-Git to focal adhesions in egg chambers is also consistent with the hypersensitivity of integrin mutants to myosin overexpression [9]. In particular, the dPix-Git orthologues bind to the focal adhesion protein paxillin, and previous D. melanogaster studies have shown that overexpression of paxillin can suppress Rho-myosin induced phenotypes, and enhance Rac signalling phenotypes [70]. Thus, our data reinforces a model that some signalling cascades downstream of integrins and paxillin can suppress Rho activation during epithelial morphogenesis, and extend this model to suggest that dPix-Git binding to paxillin is a critical part of this mechanism.
Cell biological and in vivo studies have identified PIX and GIT as regulators of PAK in a number of biological systems. Prior to this study, the regulators of Pak during egg chamber development were unknown. Here we report that dpix and git phenocopy loss of pak in egg chamber development in terms of multilayering, aberrant myosin activation, and in the generation of cell intercalation defects [49]. The biochemistry of PAK activation is well understood, however the in vivo cell biological basis of PAK activation by PIX-GIT proteins is less clear [71]. Two possibilities in egg chambers include that: dPix-Git activate Pak via the Rho family GTPases Rac1/Cdc42; and/or that dPix-Git promote local accumulation and autophosphorylation of Pak kinase. Using ubiquitin promoter-driven rescue constructs we found that removing the SH3 domain (Pak binding), or mutating the GEF domain of dPix still strongly increased egg development in dpix mutants. This raises the possibility of redundant mechanisms for Pak activation in this system. For instance, in the absence of an SH3 domain which mediates the dPix-Pak interaction, basal levels of activation of Rac or Cdc42 may still provide sufficient Pak activation. Similarly, in the absence of an efficiently functioning GEF domain, the dPix-Git complex may still serve as a basally located scaffold for accumulation and activation of Pak. Further clarity on dPix signalling in egg development could be provided by manipulating the endogenous dpix locus.
Mechanical forces can guide tissue development by passively constraining tissue shape, but also by the processes of mechanotransduction where forces applied to the cell surface are translated into biochemical signals that propagate inside the cell. Focal adhesions contain proteins which stretch and change molecular interactions under tension, and these molecules form the basis of much of our knowledge of mechanotransduction. The focal adhesion localised PIX, GIT and PAK module functions as a bona fide mechanotransducer in a range of systems. In D. melanogaster, Git is recruited to integrin-based adhesions in response to embryonic muscle contractions [72]. In C. elegans, the PIX-GIT-PAK orthologues respond downstream of cell attachments to promote intermediate filament phosphorylation and embryonic elongation [73]. In mammalian cell culture, PIX and GIT proteins are thought to act as mechanosensors due to their tension mediated interaction with paxillin at focal adhesions. For example, mechanical activation of focal adhesion kinase (FAK) via integrins and substrate stiffness [74,75], or by direct stretching of the FAK molecule [76] can lead to FAK’s interaction with SRC and the direct phosphorylation of GIT proteins [25,77]. GIT2 phosphorylation by FAK/SRC unmasks paxillin binding sites and localises the PIX-GIT-PAK complex to focal adhesions where it becomes a platform for signal propagation [77] (reviewed in [78]). Given that D. melanogaster egg chambers exhibit robust gradients of anisotropic extracellular matrix stiffness along the anterior-posterior axis [79], it will be interesting to further test whether dPix and Git act as mechanostransducers downstream of the stiffness gradient in this system.
At present, few proteins have been identified that are specifically enriched at the leading edge of each follicle cell during collective migration, and the dPix-Git complex represents a new component of this domain. Although we found evidence that this targeting depends on Git, it is not yet clear how the dPix-Git complex becomes enriched at the leading edge. One possibility is that the molecular composition and phospho-signalling profile of newly formed focal adhesions at the leading edge of cells favours the interaction between dPix-Git and paxillin, relative to mature adhesions in the medial-basal region of cells. Alternatively, Fat2 cadherin is known to signal from the trailing edge of one cell, in order to organise the leading edge of the cell immediately ‘behind’ [80]. It will be interesting to see if Fat2 is required to organise dPix at the leading edge of cells in this system.
A major finding of this study is that dpix and git are essential for epithelial morphogenesis in some tissues, but dispensable or act redundantly in others. We found that dpix, git mutation in D. melanogaster is semi-lethal, indicating that this complex is required in unidentified developmental contexts beyond egg chamber formation. Given this, it will be interesting to determine the role of dPix-Git in regulating myosin in other Pak-dependent developmental contexts, such as the epithelial sheet movement that drives dorsal closure in the embryo [81,82]. While cell culture studies have identified potential in vivo biological processes that the dPix-Git-Pak module regulates, our results underscore the essential role of animal studies to identify the precise situations where different protein complexes act in vivo. A key property of the integrin-dPix-Git-Pak signalling axis is that it can connect chemical and physical information from the ECM with intracellular signalling. This study and the work of others suggest a pattern where ECM sensing properties have made this module suitable for coordinating relatively specific developmental processes. These include control of cell migration [83,84], response to mechanical tension during epithelial morphogenesis [73], and the development of tube forming vascular systems (which require resistance to pressure) [23,85,86]. In the present case, the D. melanogaster follicular epithelium undergoes collective migration and must adjust to the pressure generated by an exponentially growing germline cyst [14,87]. Considering these animal studies together, our data suggest an underlying similarity between D. melanogaster egg chambers and tissues such as vertebrate vasculature. In the future it will be interesting and useful to transfer the understanding of how integrin-dPix-Git-Pak signalling guides development and maintains tissue integrity in these seemingly distant biological systems.
All D. melanogaster used in experiments were reared at 25°C unless otherwise indicated. For experiments assaying egg chamber development, adult female D. melanogaster were yeast fed in addition to a diet on standard medium, and were maintained in the presence of males. Unless otherwise indicated, homozygous dpix animals were generated by crossing the dpixp1036 allele to the Df(2L)ED1315 deficiency, and homozygous git animals were generated by crossing the gitex21C allele to the Df(2R)BSC595 deficiency. For stock information, including stocks from the Bloomington Drosophila Stock Center (BDSC), and Drosophila Genomics and Genetics Resources (DGGR) stock center, see Table 1. For key information on the genotypes described in figures, see Table 2.
Ovaries were dissected in phosphate buffered saline (PBS) and fixed in 4% paraformaldehyde for between 10 and 15 minutes. Tissues were rinsed three times in PBS solution with 0.1% (v/v) Triton-X (PBS-T), and permeabilised for 20 minutes in PBS-T. Primary and secondary antibody incubations were overnight at 4°C in PBS-T with 10% (v/v) Normal Goat Serum (Sigma). Ovaries were washed 3 times for 10 minutes each in PBS-T following each antibody staining. DAPI staining was incorporated into the penultimate wash. Phalloidin stainings were for 1 hour at room temperature or overnight at 4°C when co-staining with antibodies. Stained tissues were stored and mounted in 90% glycerol with 10% PBS (v/v). Primary antibody and dye stains, including antibodies from the Developmental Studies Hybridoma Bank (DSHB), are as indicated in Table 1.
The dpix-venus and git-tRFP strains were created via CRISPR/Cas-9 targeted transgene integration [91]. The sequence encoding either Venus fluorescent protein or tag Red Fluorescent Protein (tRFP) was inserted immediately 3’ of the stop codon of D. melanogaster dpix and git respectively, creating C-terminal fusion proteins. The donor vector contained approximately 1kb of homology on either side of a “knock in” cassette, which included the coding sequence for either Venus or tRFP, and a 3xP3-RFP [92] flanked by loxP sites. The gRNA expression vector used a 20-bp protospacer sequence, designed to include the dpix and git stop codons. The donor and gRNA vectors were each injected into fertilised eggs laid by nos-Cas9 flies [93]. Transformants were identified by eye specific red fluorescence from the 3xP3-RFP transgene, and this construct was then removed by crossing to a strain of D. melanogaster bearing a hs-Cre construct.
Experimental females were collected 1–3 days after eclosion and aged on yeast for 2 days in the presence of males. Ovaries were dissected in live imaging media (Schneider’s Drosophila medium with 15% FBS and 200 μg/ml insulin) containing either CellMask Deep Red or CellMask Orange Plasma Membrane Stain (Thermo-Fisher; 1:1000). Individual ovarioles were removed from muscle sheathes with forceps, transferred to fresh live imaging media, and then transferred to a glass slide. 51 μm Soda Lime Glass beads (Cospheric LLC) were added to support a 22 x 22 μm coverslip and Vaseline was used to seal the coverslip edges. Migration rates were determined for Stage 6 and Stage 7 egg chambers with the following exceptions: egg chambers with major structural defects including stalks fused along the follicle cell surface or multiple germ cell cysts within one egg chamber were excluded from analysis, as were damaged egg chambers as indicated by CellMask uptake. Egg chambers that exhibited follicle cell multilayering were included. Egg chambers were imaged with a Zeiss LSM 800 with a 40x/1.3 NA EC Plan-NEOFLUAR objective and Zen 2.3 acquisition software. Frames of a single plane near the basal epithelial surface were captured every 30 seconds for 20 minutes. To calculate epithelial migration rates, kymographs were generated from these movies in FIJI (ImageJ) by drawing a line across the egg chamber parallel to the migration path. The migration rate for each epithelium was then determined by measuring the slope of 4 kymograph lines and taking the mean of these values. This technique is illustrated in Barlan et al. 2017 [80].
Live imaging of ovary rotation to determine localisation of dPix-GFP at the leading or trailing edge followed the protocol in [94], and differed from the above as follows. Stained egg chambers were transferred to a gas permeable membrane (Ibidi) and imaged by encircling with petroleum jelly to serve as a spacer, with a glass coverslip placed on top. Images were acquired on a Nikon upright laser scanning confocal at intervals of 30 seconds, for approximately 20 minutes. To determine the localisation of dPix-GFP at the leading or trailing edge of cells, dPix-GFP and cell membranes were imaged at time zero. After finding the relative position of dPix-GFP accumulation and membranes, migration of membranes was imaged. dPix-GFP was not imaged as the signal from this construct photo-bleached rapidly.
dpix and git rescue and structure function constructs were all generated by using Gateway cloning to introduce a sequence of interest into a derivative of the pKC26w-pUbiq rescue plasmid [55] (courtesy of Nic Tapon), containing a C-terminal green fluorescent protein tag (GFP). This plasmid provides ubiquitous expression under the control of the ubiquitin-63E promoter (denoted as “ubi>” in figures, and “ubi:” in key resources Table 1). This plasmid contains a mini-white coding sequence as a selectable marker. Constructs were inserted into Drosophila bearing attP landing sites via PhiC31 integrase-mediated transgenesis.
All dpix transgenes were inserted into the same chromosomal location on 2L, at site name VK37 (cytogenetic position 22A3) (BDSC 9752). All git transgenes were inserted into the same chromosomal location on 2R, at site name VK22 (cytogenetic position 57F5) (BDSC 9740). Injections were performed at BestGene.
As a wild-type dpix construct, we used coding sequence corresponding to dpix isoform A (http://flybase.org/reports/FBtr0081356) [37,95] (originally courtesy of Ed Manser), to create a construct that deletes the SH3 domain of dPix (aa 9–56) we designed primers to amplify a sequence that removes the first 59 amino acids of dPix. To remove RhoGEF activity from dPix we used site directed mutagenesis to convert serine 89 to glutamic acid [37,59]. To generate a git construct that does not bind to paxillin we designed primers to remove nucleotides encoding the last 125 amino acids of Git.
After creation of transgenic animals these dpix and git constructs were recombined onto chromosomes bearing dpix or git mutations, so that the transgenic construct was the only source of dpix or git in the genome. Recombinant offspring positive for the transgene were scored by the mini-white selectable marker and GFP expression, whereas mutation of endogenous dpix or git was scored by PCR of the relevant locus.
For image acquisition of pMRLC and E-cadherin in dpix, git mutants, stained ovaries were imaged on a Nikon laser scanning confocal. Staining was E-cadherin detected by anti-Rat-647, and pMRLC detected by anti-Rabbit-488.
Measurements of pMRLC intensity were performed in FIJI. Plots of intensity were generated using the “ggplot2” package in the R programming language. Frequency polygons were generated in GraphPad Prism 8 and used a bin width set to 0.2. For quantification of pMRLC and E-cadherin in dpix, git mutants, egg chambers from approximately stages 6–7 (before flattening of anterior follicular epithelium) were selected. Average apical and basal pMRLC and E-cadherin signal for an egg chamber was measured by selecting apical and basal regions from at least 19 cells per egg chamber, from regions that had maintained monolayering. pMRLC and E-cadherin signal were measured from the same cells and regions of interest for each egg chamber. Statistical tests were Welch’s two-sample t-test, performed in the R programming language.
To quantify follicle cell proliferation, we imaged z-stacks through egg chambers and the number of PHH3 positive cells were counted. Egg chamber staging used DAPI based features [50]. Boxplots for pMRLC intensity, E-cadherin intensity, and number of PHH3 positive cells, were generated using the “geom_boxplot” function in the “ggplot2” package of the R programming language.
Measurements of transgene expression intensity in follicle cells were performed using FIJI software and GraphPad Prism 8. For Venus and tRFP transgenes, signal intensity was measured along three transects within the same cell. Intensity values for each channel were min-max normalised in GraphPad Prism 8. Pearson’s correlation coefficients were calculated using GraphPad Prism 8, and were obtained by first computing the mean intensity for each channel, and then analysing those means. For basally localised dPix-Venus (Fig 5C), the image was treated with Gaussian blur for noise reduction (σ = 0.5). For comparisons of basal localisation between Git-GFP and ΔPBD-Git-GFP, and dPix-GFP and ΔSH3-dPix-GFP, the apical membrane, cytoplasmic and basal membrane regions of each egg chamber were identified and labelled. Next GFP intensity was measured along a 10 pixel wide transect from the apical to basal region of individual cells. The average basal membrane GFP intensity was calculated for each cell, and normalised to the average total GFP intensity for the same cell. These measurements were made for at least three cells per egg chamber, and averaged to give an overall basal GFP enrichment value. Statistical tests to compare basal GFP enrichment between genotypes were Student’s t-test and were performed in GraphPad Prism 8.
For dpix mosaic egg chambers, dpix FRT40A D. melanogaster were crossed to a ubi-eGFP FRT40A; T155-gal4, UAS-FLP strain, except for (S1H Fig) which used hsFLP, arm-lacZ FRT40A. The hsFLP system was used for git mosaic egg chambers. Clones were induced by heat shocking at 37 degrees Celsius for two hours, on two consecutive days, beginning when D. melanogaster had developed to 3rd instar wandering larvae. Female offspring from these crosses were yeast fed for 3 days before dissection of egg chambers.
To measure the effect of rho1 and rhoGef2 heterozygosity on egg production in dpix and git mutants, females were yeast fed for 3 to 5 days before dissection, at which point the number of stage 14 eggs were counted. For dpix and git transgene rescue experiments, females were allowed to mature on yeast for at least 3 and 5 days respectively, and stage 14 eggs were counted. Stage 14 eggs were scored by the presence of elongated dorsal filaments.
Relative length and width of mature eggs between genotypes were measured by dissecting egg chambers into PBS and imaging with an Infinity camera mounted to a dissecting microscope. Length, width, and aspect ratio for mature eggs were determined in FIJI by measuring the ratio between the length of the longest and widest sections of each egg. Developing egg chambers were staged using DAPI based features [50]. In developing egg chambers length was measured along the anterior-posterior (AP) axis, and width was measured from the widest section perpendicular to the AP axis. Statistical tests were ANOVA, and were performed in GraphPad Prism 8. Boxplots of aspect ratio for developing egg chambers were generated using the “geom_boxplot” function in the “ggplot2” package of the R programming language.
To measure cell morphological features of wild-type, dpix and git mutant egg chambers between stages 7 and 8, specimens were stained with phalloidin and the basal surface of follicle cells were imaged. We imaged main body follicle cells and avoided imaging cells at the egg chamber poles. Optical sections were segmented in Ilastik and exported to FIJI. Segmentation errors were manually corrected in FIJI and the Voronoi function was used to produce a skeletonized map of cell areas for object identification and measurement in CellProfiler. Cells at the edge of egg chambers were excluded from segmentation maps and were not measured. In CellProfiler the “IdentifyPrimaryObjects” module was used to identify cells, and the “MeasureObjectSizeShape” module was used to measure eccentricity. Statistical analysis was ANOVA with post hoc Dunnett’s test, performed in GraphPad Prism 8.
Significance values throughout are indicated as: * = p < 0.05; ** = p < 0.01; *** = p < 0.001; **** = p < 0.0001; ns = not significant. Numerical values underlying charts in the figures are provided (S1 Table).
|
10.1371/journal.pntd.0004365 | Leishmania infantum Asparagine Synthetase A Is Dispensable for Parasites Survival and Infectivity | A growing interest in asparagine (Asn) metabolism has currently been observed in cancer and infection fields. Asparagine synthetase (AS) is responsible for the conversion of aspartate into Asn in an ATP-dependent manner, using ammonia or glutamine as a nitrogen source. There are two structurally distinct AS: the strictly ammonia dependent, type A, and the type B, which preferably uses glutamine. Absent in humans and present in trypanosomatids, AS-A was worthy of exploring as a potential drug target candidate. Appealingly, it was reported that AS-A was essential in Leishmania donovani, making it a promising drug target. In the work herein we demonstrate that Leishmania infantum AS-A, similarly to Trypanosoma spp. and L. donovani, is able to use both ammonia and glutamine as nitrogen donors. Moreover, we have successfully generated LiASA null mutants by targeted gene replacement in L. infantum, and these parasites do not display any significant growth or infectivity defect. Indeed, a severe impairment of in vitro growth was only observed when null mutants were cultured in asparagine limiting conditions. Altogether our results demonstrate that despite being important under asparagine limitation, LiAS-A is not essential for parasite survival, growth or infectivity in normal in vitro and in vivo conditions. Therefore we exclude AS-A as a suitable drug target against L. infantum parasites.
| It was recently described that asparagine synthetase A (AS-A) of trypanosomatids uses not only ammonia but also glutamine for asparagine formation, which was a surprising feature for a type A AS. Interestingly, Leishmania donovani AS-A was reported to be essential for parasite survival, and once a human homologue was absent, this enzyme emerged as a novel drug target candidate. Leishmania infantum encodes for a functional AS-A enzyme, which also uses either ammonia or glutamine as nitrogen donor for asparagine synthesis. In L. infantum, ASA ablation drives parasites auxotrophic to asparagine, however, LiAS-A is not detrimental for parasite survival, growth or infectivity. AS-A is therefore unlikely to be a suitable drug target against Leishmania parasites.
| Leishmaniasis is a vector borne human disease, caused by several species of digenetic protozoan parasites belonging to genus Leishmania. The clinical presentations of this neglected tropical disease vary from selfhealing cutaneous manifestations to potentially fatal, if untreated, visceral ailment [1]. The most severe form of the disease, designated as visceral leishmaniasis (VL) is mainly associated to Leishmania donovani or Leishmania infantum. Due to the absence of human vaccines, VL control relies mainly on chemotherapy and appropriate vector control [2]. The traditional therapeutic options are associated with significant limitations (cost, toxicity, complex administration regimes, resistance) averting disease control in endemic areas [3]. As consequence, according to World Health Organization between 20,000 and 30,000 people (mostly children) die every year, rendering the search for novel chemotherapeutic options a priority [4].
Asparagine (Asn) metabolism has been under the spotlight in the recent years. Asn is the last nonessential amino acid to be synthesised from glucose metabolism [5]. For many years it seemed it was not involved in any other pathway but protein synthesis in mammalian cells, contrasting with the other 19 common amino acids [6]. Nonetheless, several recent studies suggest Asn somehow coordinates cell responses with metabolic reserves and ultimately regulates cell fate [5]. In many pathogenic microorganisms, functional studies on L-asparaginase and Asn transporters have implicated Asn metabolism in survival, invasion and/or virulence [7–14].
Asparagine synthetase (AS) is another key player in Asn metabolism, it catalyses Asn formation from aspartate in an ATP dependent manner using ammonia or glutamine as nitrogen donors. The reaction mechanism comprises two crucial steps: 1) the formation of β-aspartylAMP, in which β-carboxylate group of aspartate is activated by ATP; 2) nucleophilic attack by an ammonium ion. This mechanism mirrors the close evolutionary relation to aminoacyl-tRNA synthetase enzymes [15]. There are two structurally distinct types of AS: A and B [16]. Type B (AS-B, EC. 6.3.5.4) uses preferably glutamine over ammonia and can be found in prokaryotes and eukaryotes (mammalian cells, yeasts, Chlamydomonas reinhardtii, higher plants) [17–23]. Type A (AS-A, EC. 6.3.1.1) are found mainly in prokaryotes (Escherichia coli [24] and Klebsiella aerogenes [25]) or in archaea (Pyrococcus abyssi [26]) and described as strictly ammonia dependent. Surprisingly, kinetoplastids and other protozoans, despite being eukaryotes, possess not only a putative AS-B but also a bacterial type AS-A [27–29]. Moreover AS-A from Trypanosoma brucei, Trypanosoma cruzi [28] and L. donovani [29] parasites were reported to use glutamine as nitrogen donor as well.
Several roles have been associated to bacterial AS. For instance, in Pasteurella multocida, AS-A is significantly upregulated during host infection, in Mycobacterium smegmatis AS-B is involved in natural resistance to antibiotics and in Mycobacterium tuberculosis, AS-B was reported to be required for in vitro growth [30–33].
Recently our group showed that in T. brucei bloodstream forms AS-A knockdown has no impact on parasites growth or infectivity, except upon Asn deprivation. These results suggest Asn main sources are AS-A mediated synthesis and extracellular uptake [28]. Surprisingly, in L. donovani, AS-A was claimed to be essential for parasites survival and emerged as a promising drug target due to the absence of a human homologue [29]. Additionally, these results also suggest that Asn homeostasis could be differently regulated among trypanosomatids. These parasites present different amino acid requirements for either energetic or osmotic functions in different stages of their life cycles and as a reflex of the different environmental stimuli they receive in the vector or mammalian host [34]. Across trypanosomatids’ species, the amino acid transporters (AAT) repertoire has a high interspecific variation, regarding number, affinity, specificity and capacity [34]. For instance, in the case of cysteine, a crucial amino acid for thiol biosynthesis, Leishmania major contrarily to T. brucei, fails to uptake it at a rate that ensures the intracellular pool is enough for optimal growth. Therefore, these parasites rely mainly on pathways that enable cysteine synthesis [35].
In this work, we have biochemically characterized L. infantum AS-A (LiAS-A), and to gain further insights on AS-A essentiality across different Leishmania species, we have performed gene replacement studies in L. infantum.
All experiments were carried out in accordance with the IBMC.INEB Animal Ethics Committee and the Portuguese National Authority for Animal Health (DGAV) guidelines, according to the statements on the directive 2010/63/EU of the European Parliament and of the Council. DGAV approved the animal experimentation presented in this manuscript under the license DGAV number 25268/2013-10-02.
L-asparagine, L-aspartic acid sodium salt monohydrate, L-glutamine, L-glutamatic acid salt hydrate, ATP disodium salt hydrate, AMP disodium salt, sodium pyrophosphate decahydrate, ninhydrin, dNTPs, ammonium chloride, magnesium chloride, tween-20, tris-base, urea, thiourea, DTT, triton X-100 and IPTG (isopropyl-β-D- thiogalactopyranoside) were purchased from Sigma. Oligonucleotide primers were obtained from STAB VIDA. Restriction endonucleases were from New England Biolabs. Polyclonal antibodies against LiAS-A were obtained in rabbits inoculated with purified recombinant His-tagged LiAS-A. E. coli L-asparaginase was purchased from Prospec.
L. infantum (MHOM/MA/67/ITMAP-263) promastigote forms were grown at 26°C in complete RPMI 1640 medium [36]. For in vitro and in vivo characterization, different cell lines were firstly recovered from the spleen of infected BALB/c to restore virulence, and subsequently maintained in culture no longer than 10 passages [36]. Axenic amastigotes were grown in MAA complete medium [36], at 37°C, 5% CO2. Depending on the analysis, protein extracts were prepared as follows: 1) 1 x 107 late-stationary promastigotes were resuspended in T8 lysis buffer (tris-base 0.6%, urea 42%, thiourea 15%, DTT 0.3%, triton X-100 1%); or 2) 1 x 108 promastigotes or axenic amastigotes were resuspended in 100 μL of PBS containing protease inhibitor (Roche) and following 6 freezing/thaw cycles, the parasite suspensions supernatants were recovered and then quantified using Bio-Rad DC Protein Assay (Biorad).
EcAS-A, LiAS-A, LmAS-A, TbAS-A and TcAS-A protein alignments were performed using the ClustalW program [37]. Aline program, Version 011208 [38], was used for editing protein sequence alignments. LiAS-A and LmAS-A homology models were obtained with SWISS-MODEL, using EcAS-A crystal structure (Protein Data Bank (PDB) 12AS [15]) as a template (percentage of sequence identity of ~50–60% in both cases) [39–41]. The 3D models were illustrated using PyMOL program (The PyMOL Molecular Graphics System, Version 1.3, Schrödinger, LLC).
Asparagine synthetase A (ASA) from L. infantum (LinJ.26.0790; chromosome LinJ.26; 234298–235360) was obtained by performing PCR on genomic DNA, extracted using DNAzol (Invitrogen) [42–44], using primers 1 + 2 (S1 Table). PCR conditions were as follows: initial denaturation (2 min at 94°C), 35 cycles of denaturation (30 s at 94°C), annealing (30 s at 50°C) elongation (2 min at 68°C) and a final extension step (10 min at 68°C). Another restriction strategy was required to clone the gene into a Leishmania overexpression vector–pSPαBLASTα, and the sequence was amplified using primers 3 + 4 (S1 Table). PCR conditions were as follows: initial denaturation (2 min at 94°C), 30 cycles of denaturation (15 s at 94°C), annealing (30 s at 55°C) elongation (1 min at 72°C) and a final extension step (10 min at 72°C). All PCR products were cloned into a pGEM-T Easy vector (Promega) and sent for sequencing.
The LiASA gene was excised from the pGEM-T Easy vector (using NdeI/EcoRI), and subcloned into pET28a(+) expression vector (Novagen). The resulting construct presented a poly-His tag (6x Histine residues) at the N-terminal and was transformed into E. coli BL21DE3. The recombinant protein was expressed by induction of log-phase cultures with 0.5 mM of IPTG at 18°C O/N. Bacteria were harvested and resuspended in buffer A (0.5 M NaCl, 20 mM Tris.HCl, pH 7.6). The sample was sonicated, according to the following conditions: output 4, duty cycle 50%, 10 cycles with 15 s each (Branson sonifier 250), followed by centrifugation to obtain the bacterial crude extract. For enzymatic activity experiments and rabbit polyclonal antibody production, the recombinant enzyme was purified in one step using Ni2+ resin (Qiagen) pre-equilibrated in buffer A. The column was washed sequentially with buffer A, bacterial crude extract, and buffer A with increasing concentrations of imidazole. LiAS-A was eluted in the fractions of buffer A containing 100 to 500 mM of imidazole. Dialysis was performed against PBS.
For additional activity tests, oligomeric form and Stokes’ radius assessment, a deeper purification was performed. Firstly, the enzyme was purified by affinity chromatography, using a Histrap HP column (GE Healthcare), charged with nickel sulphate and equilibrated in buffer A, and posteriorly mounted in an AKTAPrimer Plus (GE Healthcare) system, at 4°C. Secondly, it was purified by size exclusion chromatography, in a Hiprep 26/60 Sephacryl S-200 column (GE Healthcare), previously equilibrated with running buffer (150 mM NaCl, 20 mM Tris, pH 7.6). The last purification step was a preparative ion exchange chromatography, using an UNO Q-1 (Bio-Rad, Cat. No 720–0001) column, mounted in a BioLogic DuoFlow (Bio-Rad) device, at 4°C. The fractions were finally analysed by analytic size exclusion chromatography and analytic ion exchange chromatography, using AktaPurifier10 system (GE Healthcare), using Superose 12 10/300GL (GE Healthcare) column and a UNO Q-1 (Bio-Rad, Cat. No 720–0001) column, respectively. The final fractions were concentrated using Millipore centrifugal filter 30K (Amicon Ultra).
Concentration was determined measuring the absorbance at 280 nm using the theoretical molar extinction coefficient of 46910 M-1.cm-1 for LiAS-A, making use of NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies). The purified recombinant protein was resolved in SDS/PAGE and stained with Coomassie Brilliant Blue G-250 (Biorad).
For estimation of the LiAS-A oligomeric state the purified recombinant protein was analysed by analytic size exclusion chromatography, using the above described conditions. Blue dextran (2,000 kDa), catalase (MW 232 kDa, Stokes radius (SR) 5.22 nm), aldolase (MW 158 kDa, SR 4.81 nm), albumin (MW 67 kDa, SR 3.55 nm), ovalbumin (MW 43 kDa, SR 3.05 nm), chymotrypsinogen A (MW 25 kDa, SR 2.09 nm) and ribonuclease (MW 13.7 kDa, SR 1.64 nm) were used as standards. A calibration curve relating Log (MW) or Log(SR) with Kav was performed (Kav is (Ve-V0)/(Vt-V0), in which Ve is elution volume, V0 is the exclusion volume given by blue dextran and Vt is the total volume of the column).
In a 96-well, thin-walled white PCR plate, 5 μl of LiAS-A (2.4 μM) were mixed with 5 μl of 10x SYPRO Orange (λexc 485 nm; λem 625 nm) and 40 μl of water or the ligands and ligands combinations to be tested. Plates were then sealed and placed into a BioRad iCycler5 PCR instrument. Measurements were taken every minute in 0.5°C increments from 25° to 95°C. Subsequent analysis of the fluorescent data using Biorad iCycler iQ Optical System Software Version 3.1 yielded the protein melting temperature (Tm) for LiAS-A.
Western-blot was performed aiming different purposes: (1) His-tag labelling of recombinant proteins, (2) LiAS-A labelling in total soluble parasite extracts to assess protein expression throughout the life cycle, (3) LiAS-A labelling in mutants and (4) to assess protein distribution upon digitonin fractionation. One μg of recombinant LiAS-A, 20 μg of total soluble extracts from both promastigote and amastigote forms, or 1 x 107 parasites were resolved in SDS-PAGE and transferred onto a nitrocellulose membrane (TransBlot Turbo, Bio-Rad), which was blocked, probed, washed and developed as previously described [28]. The following primary antibodies were used: rabbit anti-His-tag (MicroMol-413, 1:1000), mouse anti-α-tubulin (clone DM1A, Neomarkers, 1:1000), rabbit anti-LiAS-A (1:1000), rabbit anti-LiCS (cysteine synthase, 1:2000), rabbit anti-LdHGPRT (hypoxanthine guanine phosphoribosyl transferase, 1:2000), and rabbit anti-TbEnolase (1:5000). Horseradish peroxidase-conjugated goat anti-rabbit or goat anti-mouse IgG (Amersham) (1:5000 for 1 h, at RT) were used as the secondary antibody. ImageJ software (version 1.43u) was used for protein semi-quantification.
A 150 μl enzymatic mixture containing 85 mM Tris.HCl, 8.4 mM magnesium and varying concentrations of aspartate, ammonia and ATP was assayed. The assay was performed as previously described [28], and ultimately absorbance at 340 nm was measured [45]. To determine the optimal conditions for kinetic parameters determination, reaction linearity was checked by varying enzyme concentration and time. The final reaction conditions used 7.5 μg of enzyme per assay and 15 min incubation at 37°C. A pH range of 7.0 to 9.0 was assessed, and pH 7.6 was selected as the optimal one to perform the following enzyme assays. To determine the Km of each substrate a certain range of concentration was used and the remaining substrates were maintained in excess. For aspartate, ammonia, ATP and glutamine, the following concentrations were used: 1.25 to 20, 0.78 to 50, 0.625 to 10 and 1.56 to 25 mM, respectively.
A targeted gene replacement strategy was used for L. infantum ASA gene knock-out. Briefly, ASA flanking regions were amplified from L. infantum genomic DNA and were linked to neomycin phosphotransferase (NEO) or hygromycin phosphotransferase (HYG) genes using a fusion PCR approach. The 5’ and 3’ UTR were amplified using primers 1 + 2 and 3 + 4 (S2 Table), respectively. NEO and HYG were amplified from pSP72αNEOα and pGL345HYG templates, using primers 5 + 6 and 7 + 8 (S2 Table) respectively, which possess around 30 nucleotides of the 5’ UTR in the sense primer and the first 30 nucleotides of the 3’ UTR in the antisense primer. 5’ UTR_NEO_3’ UTR and 5’ UTR_HYG_3’ UTR constructs were obtained using primers 1 + 4 (S2 Table). Mid-log promastigotes were transfected with approximately 10 μg of linear construct, obtained by fusion PCR, using an AMAXA Nucleofector II device with Human T-cell nucleofector kit (Lonza). The day after transfection drug selection was carried out at 20 μg/mL of G418 (Invitrogen) and 50 μg/mL of hygromycin B (InVivoGen). Parasite cloning was performed by diluting the parasite suspension to a concentration of 0.5 cells/well, using SDM culture medium. The drug concentrations for clone maintenance correspond to half of the selection concentrations.
LiASA gene was excised from the pGEM-T Easy vector (using XbaI/NdeI) and subcloned into pSP72αBLASTα vector. Mid-log promastigotes, WT and dKO mutants, were transfected with approximately 10 μg of plasmid DNA as above in order to generate an overexpressing line (OE) or complemented null mutants, respectively. Drug selection was carried out at 30 μg/ml of blasticidin (InVivoGen).
LiASA mutants were analysed by PCR using Taq polymerase (NZYTech) for the following events: LiASA presence; NEO 5’ integration; NEO 3’ integration; HYG 5’ integration and HYG 3’ integration, using primers pairs 9 + 10, 11 + 12, 13 + 14, 15 + 16 and 17 + 18 (S2 Table), respectively. Additionally, a non-related gene from chromosome 28, encoding a putative ribose-5-phosphate isomerase B (RPIB, ~570 bp) was used as control, using primers 19 + 20 (S2 Table). For Southern-blot analysis, total genomic DNA was extracted. Ten μg of genomic DNA were digested O/N with a 5 fold excess of SacI and NdeI, at 37°C and samples were run O/N in an agarose gel. The gel was sequentially incubated with 0.25 M HCl, 1.5M NaCl 0.5M NaOH and 3M NaCl 0.5M Tris.HCl pH 7. DNA was then transferred O/N onto a Nylon membrane (Amersham), using 10x SSC (saline sodium citrate: 300 mM sodium citrate, 1 M NaCl). Nucleic acids fixation was achieved at 65°C for 5 hours. Hybridization and revelation were undertaken using Gene Images AlkPhos Direct Labelling and Detection System kit (GE Healthcare Amersham). Pre-hybridisation, hybridisation and washes took place at 65°C, probe labelling and membrane stripping were performed according to the manufacturer instructions. The blots were probed sequentially with 5’ UTR, LiASA, HYG and NEO, which were PCR amplified, using primers 1 + 2, 9 + 10, 7 + 8 and 5 + 6 (S2 Table), respectively.
Cultures were launched and monitored microscopically every 24h for 8 days or were maintained in log phase by subculturing every 2 days and cumulative growth was assessed for 5 consecutive passages. The growth experiments were performed in complete RPMI (cRPMI) or Asn depleted cRPMI (cRPMI + L-asparaginase) obtained by cRPMI O/N incubation with 1250 U/L of L-asparaginase at 37°C. Growth curves were also undertaken in a serum-free RPMI (sfRPMI [46]) incubated with L-asparaginase, that was removed afterwards by flowing the medium through a 3 kDa Millipore centrifugal filter (Amicon Ultra), generating an Asn free medium (sfRPMI + L-asparaginase). Asn was directly added to the Asn free sfRPMI (cf = 380 μM) to generate complemented sfRPMI (sfRPMI + L-Asparaginase + Asn). Finally, growth curves were performed in complete M199 (cM199) [29] or cM199 supplemented with Asn (cf = 380 μM). Growth curves of LiASA mutants and WT were seeded at 1 x 106 parasites/ml at 26°C, except in the case of sfRPMI, whose initial parasite density was 2 x 106 parasites/ml, grown with agitation. Before launching growth curves, the parasites were maintained in log phase for 2–3 passages in the absence of selection drugs.
Five to six weeks old female BALB/c mice were obtained from Charles River. For each mouse injection, 1 x 108 promastigotes recovered from 4 days old stationary culture were washed, resuspended in PBS, and injected intraperitoneally. Mice of each group (n = 4) were sacrificed at 2 weeks post-infection. The parasite burden in the spleen and liver was determined by limiting dilution as previously described [47].
For each sample condition, 1 x 108 promastigotes were washed once with cold trypanosome homogenisation buffer (THB), composed by 25 mM Tris, 1 mM EDTA and 10% sucrose, pH 7.8. Just before cell lyses, peptidase inhibitor (Roche) and different digitonin (Calbiochem) quantities (final concentrations of 12.5, 25, 50, 100, 200, 500 and 1000 μg/ml) were added to 250 μl of cold THB, for cell pellet resuspension. Untreated cells and those completely permeabilised (total release, the result of incubation in 1% Triton X-100) were used as controls. Each sample was incubated 60 min on ice, and then centrifuged at 13,000 rpm, 4°C, for 10 min. Supernatants were taken off into new pre-cold tubes and 250 μl of cold THB was added to each pellet and then mixed. All fractions were analysed by WB.
L. infantum mid-log promastigotes were fixed, permeabilised and stained as previously described [48]. Cells were incubated with primary antibody O/N at 4°C. The following primary antibodies were used: rabbit anti-LiAS-A (1:1000) and sheep anti-LiTDR1 (thiol-dependent reductase 1, 1:2000). Subsequently, slides were incubated for 1h at RT in a dark humidified atmosphere with a secondary antibody (1:500). The following secondary antibodies were used: goat anti-rabbit Alexa Fluor 488 or 568 and donkey anti-sheep Alexa Fluor 488 (Molecular probes, Life Technologies). In the case of Mitotracker Orange (Invitrogen), we stained the parasites by adding 1 μM to culture medium (without FBS) for 1h at 26°C, prior to the above described procedure. Slides were stained and mounted with Vectashield-DAPI (Vector Laboratories, Inc.). Images were captured using fluorescence microscope AxioImager Z1 (Carl Zeiss), equipped with a Axiocam MR v. 3.0 camera (Carl Zeiss), using either 63x (Plan-Apochromat 63x/1.40 Oil DIC) or 100x (Plan-Apochromat 100x/1.40 Oil DIC) objective. Images analysis and deconvolution was performed using ImageJ software (v. 1.47) and image deconvolution lab plugin (2010 Biomedical Imaging Group, EPFL, Switzerland) with Richardson-Lucy algorithm.
For statistical analysis, one-way ANOVA and two-tailed Student’s test were used. Statistical analysis was performed using GraphPad Prism Software (version 5.0): statistical significance p ˂ 0.05 (*), p ˂ 0.01 (**), p ˂ 0.001 (***), p ˂ 0.0001 (****).
The open reading frames (ORFs) encoding putative AS-A and AS-B enzymes were identified in the genomes of L. infantum JPCM5 (LinJ.26.0790; LinJ.29.1590) and L. major Friedlin (LmjF.26.0830; LmjF.29.1490) [42–44]. The ASA amplified sequence from L. infantum strain matched 100% the annotated sequence from JPCM5 genome. To obtain structural and functional insights on AS-A enzymes, we have performed in silico analysis using the L. infantum (LiAS-A), L. major (LmAS-A), T. brucei (TbAS-A), T. cruzi (TcAS-A) and E. coli (EcAS-A) sequences that generate polypeptides containing 353, 353, 351, 348 and 330 residues, respectively (Fig 1A). Overall, the sequence alignment shows a high conservation of the main structural features, including the active site residues (Fig 1A). Indeed, the amino acids involved in Asn binding are strictly conserved across species, whereas in the case of AMP binding pocket, the majority of residues are conserved with a few exceptions. For instance, in the case of LiAS-A and LmAS-A, there is a sole residue replacement, namely EcAS-A L109, corresponding to I111 in both cases. This residue is not involved in polar interactions with AMP molecule, but instead integrates the outer wall of the nucleotide binding pocket [15].
Analysing the homology models of LiAS-A and LmAS-A (Fig 1B) obtained by superimposition with EcAS-A crystal structure (PDB 12AS [15]), there is a divergent region highlighted with a dashed rectangle (Fig 1B) in L. infantum and L. major enzymes, which is strictly conserved in these two species. This region also exists in trypanosomes, although little conservation is found when comparing to Leishmania sp. (Fig 1A, [28, 29]).
Recombinant LiAS-A, comprising a 6 histidine N-terminal tag, was expressed in E. coli and purified by affinity chromatography in native conditions in order to evaluate and characterize its enzymatic activity. The protein presented the expected MW for the monomer, ~42 kDa, as presented on Fig 2A and 2B, with either Coomassie staining or Western-blot analysis with an anti-HisTag antibody, respectively. Subsequently, LiAS-A was further purified sequentially by size exclusion and ion exchange chromatographies, and the final fractions were analysed by analytic size exclusion chromatography (Fig 2C). Using the latter chromatography and calibration standards, Stokes’ radius (~3.52 nm, Fig 2D) and MW (~78.8 kDa, Fig 2E) were extrapolated in the protein native state. LiAS-A corresponds to a homodimer, as predicted.
For the characterization of the enzymatic activity of recombinant LiAS-A, a specific colorimetric assay that quantifies Asn formation was used [28, 45]. The optimal pH for the enzymatic activity was 7.6. The kinetic characterization of the enzyme was undertaken in steady-state conditions, using a fixed concentration of 8.4 mM of Mg2+ (Table 1). LiAS-A displayed ammonia and glutamine dependent activity
When comparing Km values for ammonia and glutamine, no statistical significant difference is found (p = 0.03), but there is significance in the differences found in kcat (p = 1.80 x 10−4). In order to discard the possibility that utilization of glutamine as a substrate was an artefact resulting from contamination with EcAS-B (EcAS-B ~120 kDa), highly purified fractions of LiAS-A (LiAS-A ~84 kDa) were tested and glutamine utilization were clear in all protein samples tested.
Differential scanning fluorimetry was also used in order to further understand the relevance of the different substrates for thermal stabilization of the enzyme. AS-A forms a crucial β-AspartylAMP-Mg2+ intermediate, which then undergoes a nucleophilic attack of ammonia, forming Asn and releasing AMP e pyrophosphate [15]. According to our data, ammonia can be free or glutamine-derived, although the glutaminase domain of LiAS-A remains to be identified. Looking at the differential scanning fluorimetry data, AMP leads to a 10 degrees shift in LiAS-A Tm, thermally stabilising this protein (Fig 2F) in a concentration dependent fashion (Fig 2G).
A targeted gene replacement strategy was used for inactivation of the ASA gene of L. infantum. Two constructs, obtained by fusion PCR, linking NEO or HYG to the 5’ and 3’ UTRs of the LiASA gene were used to remove the first and second LiASA allele, respectively. Two sKO mutants (clones A and B) were transfected with the HYG construct. We successfully obtained 5 dKO mutants, 3 from clone A (A1, A2 and A3) and 2 from clone B (B1 and B2). The integration of the resistance markers in the expected locus was confirmed by PCR using primers upstream of the 5’ UTR or downstream of the 3’UTR coupled with primers in their ORFs (the strategy is illustrated on Fig 3A). NEO 5’ and 3’ integration was positive in both sKO and dKO mutants, as for HYG, only in dKO parasites, as expected (Fig 3B). Also by PCR analysis, we could not amplify LiASA ORF in null mutants (a non-related gene–LiRPIB- was amplified as control–Fig 3B).
Southern-blot analysis confirmed the genotypes: the expected fragments upon digestion with SacI and NdeI are represented on Fig 3A. A first hybridisation was performed using 5’ UTR as a probe: in WT a single band of ~1696 bp corresponding to LiASA was generated, with twice the intensity observed in the sKO mutants that possess a single copy, and absent in the dKO mutants, confirming the successful gene removal (Fig 3C). In both sKO and dKO clones, a band of ~2973 bp was generated corresponding to NEO, and then only in dKO clones, a band of ~2052 bp corresponding to HYG was observed (Fig 3C). The blot was then stripped and reprobed three additional times to confirm each one of the bands (faint bands of incomplete stripping can be observed), sequentially using LiASA, NEO and HYG. All the mutants were also analysed by Western-blot, showing a protein reduction in sKO mutants and a complete absence in the dKO clones (Fig 3D). LiASA gene was cloned into a pSP72αBLASTα vector in order to obtain an overexpressor mutant (OE) as well (Fig 3D).
Rabbit polyclonal antibodies produced against recombinant LiAS-A recognised a major band in total WT promastigotes extract with the expected molecular weight (~39.8 kDa [web.expasy.org/protparam], S1A Fig), but not in a dKO mutant (dKO A2). Prior to immunolocalisation studies, LiAS-A antibody was also validated (S1B Fig) by performing an IFA and comparing the labelling intensity in WT promastigotes versus LiASA null mutants and OE. A positive correlation between protein level and fluorescence intensity was found on WT versus OE (S1C Fig). As expected, no specific labelling was detected for the LiAS-A null mutants (S1B and S1C Fig).
Using α−tubulin (~50 kDa) as loading control we compared the expression levels of LiAS-A in different developmental stages: promastigotes (logarithmic, early stationary and late stationary phase) and axenic amastigotes (Fig 4A). No significant differences were observed.
Immunofluorescence analysis showed that in promastigotes LiAS-A co-localises with LiTDR1 (thiol-dependent reductase 1), which is a cytosolic protein involved in thiol metabolism [49] (Fig 4B, upper panel). LiAS-A subcellular localisation in promastigotes was also assessed by digitonin fractionation. The fractioning profile was evaluated using antibodies for proteins present in different subcellular compartments, namely, anti-TbEnolase (LiEnolase versus TbEnolase 79% identity, LiEnolase 39.6 kDa) as cytosolic marker [50], and anti-LdHGPRT (hypoxanthine guanine phosphoribosyltransferase, 23.6 kDa) as glycosomal marker [51]. LiEnolase (Fig 4C) can be found in the supernatant for digitonin concentrations as low as 12.5 μg/ml, and retained in the pellets up to 25–50 μg/ml. LiHGPRT (Fig 4C), which localises to the glycosomes, is detected in the supernatant in appreciable amounts for higher digitonin concentrations and is retained longer in the pellet (up to 200–500 μg/ml, and residually at 1000 μg/ml of digitonin). As expected, LiAS-A presents a profile similar to LiEnolase, supporting a cytosolic location (Fig 4C).
Intriguingly, LdAS-A was reported to have dual localisation between the cytoplasm and mitochondria in promastigote form [29]. Due to the high identity (~99%) between both enzymes, we investigated whether LiAS-A also localised to the mitochondria. Immunofluorescence analysis of LiAS-A subcellular distribution on promastigotes labelled with mitotracker showed no evidence of mitochondrial location (Fig 4B, lower panel). Moreover, by using tools for protein localisation prediction (TargetP, CELLO, MITOPROT and Predotar), mitochondria localisation seems unlikely and actually, CELLO predicts cytoplasmic localisation. In conclusion, our data shows LiAS-A localises to the cytosol.
All mutants displayed similar growth patterns comparing to WT promastigotes in cRPMI (Fig 5A). However, in Asn depleted medium, achieved upon L-asparaginase treatment (cRPMI + L-asparaginase), the behaviour was quite different for some of the mutants. Parasites overexpressing LiAS-A displayed a significant higher growth during log phase when comparing to the WT, whereas the dKO mutants (clones A2 and B1) displayed a major growth defect (Fig 5B). The complementation of these null mutants with an episome (pSP72αBLASTα) carrying LiASA gene rescued the growth (Fig 5B). Moreover, an upregulation in LiAS-A levels could be observed in these mutants in Asn limiting conditions (Fig 5C). Western-blot analysis also showed that in the same conditions, an upregulation in LiAS-A could also be observed over time in the sKO parasites (clones A and B), enabling the growth recovery in these mutants (Fig 5B–5D and S3 Table). This recovery was faster in sKO clone B that had higher basal levels of LiAS-A than clone A (Fig 5B and 5C). We also evaluated the cumulative growth under constant multiplicative conditions, in which high amino acids levels are required. For that, parasites were maintained in log phase in Asn replete or Asn depleting conditions, and the same patterns were observed (S2 Fig).
To ensure the defective growth phenotype of sKO and dKO parasites in L-asparaginase treated medium was due to Asn depletion, we supplemented this medium with Asn. Surprisingly the addition of this amino acid to L-asparaginase treated RPMI medium fails to reverse the observed growth delay/arrest phenotype. The fact that L-asparaginase was not inactivated or neutralized, and consequently may have remained active, may explain this result. Consequently, we used another strategy by undertaking growth curves in a serum free medium (sfRPMI [46]) incubated with L-asparaginase that was removed afterwards using a 3 kDa Amicon column. In sfRPMI devoid of Asn (sfRPMI + L-asparaginase), the same growth defect of the sKO and dKO mutants was observed (Fig 5E). And then again, in the sKO clones the upregulation of LiAS-A allowed the growth rescue (Fig 5G and S3 Table). When adding back Asn (sfRPMI + L-Asparaginase + Asn), all mutants grew in a similar fashion (Fig 5F). In the absence of drug pressure and in normal conditions, parasites provided of an episome carrying LiASA (OE) hardly overexpress AS-A, however, under Asn depleting conditions, they upregulate its expression (comparing to the levels in the WT, there is an increase from ~130% to ~300% and from ~110 to ~130%, in panels C and G, respectively, and S3 Table).
Moreover, besides the experiments using L-asparaginase treatment, we have also performed growth curves in a medium formally lacking Asn–complete M199 (cM199)—in order to further confirm Asn auxotrophy upon ASA ablation. In this medium, null mutants presented a growth defect comparable to the one observed in cRPMI + L-asparaginase, which again was reversed when these mutants were complemented with an ectopic copy of ASA gene (Fig 5B versus 5H). The addition of Asn to the final concentration of 380 μM (like in RPMI) rescues the growth defect displayed by the null mutants in cM199 (Fig 5I). Interestingly, the experiments to assess LiAS-A essentiality in L. donovani were performed in cM199 [29], therefore the inability to generate LdASA null mutants may be due to the performance of those attempts in Asn limiting conditions.
In conclusion, ASA deletion renders parasites auxotrophic to Asn, but is dispensable for parasite growth in normal conditions.
Notwithstanding, we intended to evaluate the impact of LiASA ablation on, in vivo infectivity. Five to six old female BALB/c mice were infected and were sacrificed at 2 weeks post-infection. The parasite burden in the spleen (Fig 6A) and liver (Fig 6B) was not statistically different in LiASA mutants when compared to the WT. The same scenario was observed for sKO A and dKO A2 mutants. No differences in LiAS-A expression levels were found when comparing parasites used in mice infection (Culture) to parasites recovered from spleen (S) or liver (L) (Fig 6C). Thus, LiAS-A ablation does not compromise parasite infectivity in the context of an acute in vivo infection.
Despite being eukaryotes, trypanosomatids, present AS-A enzymes of bacterial origin. Moreover, these enzymes are aminoacyl-tRNA synthetase paralogues, displaying an AsnRS catalytic core with conserved class II motifs, yet lacking the tRNA binding domain [27]. In this work, we have demonstrated that LiAS-A is able to synthesize Asn using either ammonia or glutamine as nitrogen donors, as previously described for TbAS-A, TcAS-A and LdAS-A [28, 29]. Km values for aspartate and ATP are close to the ones determined for TbAS-A and TcAS-A [27]. As for ammonia, the Km value found for LiAS-A is 5 fold lower in comparison to TbAS-A, TcAS-A and LdAS-A [27, 28]. In the case of LdAS-A, the Km values for aspartate were around 10 fold lower [29] than the ones obtained for LiAS-A. Regarding the high conservation of the active sites among Leishmania AS-A enzymes, we cannot exclude that the observed kinetic differences may be due to the differences in the amount of protein that is properly folded, especially taking into account they are expressed in a heterologous system. Moreover, it is important to emphasize that the kinetic determinations for LdAS-A were performed using a different experimental set up. Importantly, TbAS-A and TcAS-A use preferably ammonia [28], whereas LiAS-A seems to use both roughly in the same extent (Table 1). AS-A activity in trypanosomatids more resembles AS-B enzymes, concerning both the optimal pH for enzymatic activity (7.6 instead of 8) and also the ability to use both nitrogen donors. AS-B enzymes use preferably glutamine, with exception of the human enzyme that presents approximately the same affinity for both nitrogen sources [18, 19, 25, 52–57]. This biochemical feature, so far only described for trypanosomatids AS-A enzymes [28, 29], becomes particularly interesting in the context of the presence of an ORF encoding a hypothetical, yet non-classical, AS-B, in the genome of these organisms (L. infantum [LinJ.29.1590], L. major [LmjF.29.1490], T. brucei [Tb927.3.4060] and T. cruzi [Tc00.1047053510001.40]) [42–44]. These sequences contain a Pfam AS domain (pfam00733) and glutamine hydrolysing domains in the C and N-terminus, respectively. BLASTp analysis of L. infantum sequence, for instance, revealed several hits that corresponded to hypothetical proteins from a broad range of eukaryotes. However, we have no evidence AS-B is functional at all.
Much remains to be disclosed regarding the AS-A enzymes from trypanosomatids, for instance, we still lack information on their glutamine binding and hydrolysing sites. TbAS-A crystallisation only emphasised the high conservation of Asn and AMP binding pockets, as the only divergent region from EcAS-A (a 19 residues insertion, also present in LiAS-A and LmAS-A, Fig 1) was not visible in the experimental electron density maps and therefore likely disordered [29]. This insertion displays little conservation when comparing Leishmania and trypanosomes, and its role on a structural or functional level is still unclear.
AS-A is a key enzyme in Asn metabolism that was proposed as a potential drug target due to its absence in the human host. Moreover, AS-A was reported to be essential for L. donovani survival, contrasting with T. brucei bloodstream forms, as in the latter it was shown to be dispensable for both in vitro growth and infectivity. These findings pointed to a differently regulated Asn homeostasis across trypanosomatids. In L. infantum, our efforts to generate ASA null mutants were successful, indicating the gene is not essential for survival. Moreover, the null mutants did not present any growth or infectivity defect. Our in vitro growth data demonstrate that upon LiASA deletion, promastigotes become dependent on extracellular Asn for optimal growth (Fig 5). These results suggest that even if AS-B is functional, it does not compensate LiAS-A activity, as LiASA null mutants fail to grow in Asn limiting conditions. Additionally, WT parasites grew normally in Asn depleted medium without AS-A upregulation, suggesting Asn synthesis by basal AS-A suffices the cellular needs, although the mutants overexpressing this enzyme had a metabolic advantage in an Asn deprivation environment during log phase (S2B and S2E Fig). We can actually infer the parasite can both synthesise and take up this amino acid, and the latter fully compensates the former. Furthermore, our results indicate that LiAS-A levels are regulated according to Asn availability, and it was equally surprising to see how fast and efficiently sKO mutants were able to upregulate AS-A when cultured in Asn limiting conditions (Fig 5D). It is also noteworthy that the two sKO mutants displayed a substantial difference in AS-A levels, which has also been observed among other sKO mutants generated in this study. A possible explanation might be that the two allelic copies may differently affect ASA expression.
In trypanosomatids, much remains to be unravelled concerning amino acid transporters (AATs) and mostly the pathways involved in amino acid sensing and regulation of their synthesis and uptake [34]. Very few data is available in the literature concerning Asn transport in these parasites. In T. brucei, a protein presenting putative orthologues in Leishmania [42–44] was characterized as a transporter of several neutral amino acids, including asparagine (TbAATP1) [58]. In mammalian cells, AS-B is a transcriptional target of the well characterized GCN2/elF2α/ATF4 axis, in response to amino acid starvation [59, 60]. The phosphorylation of elF2 leads to a repression of general protein synthesis, as well as an activation of gene-specific translation. In Saccharomyces cerevisiae, GCN2, which is activated by amino acid, glucose or purine deprivation, is the only elF2 kinase, contrasting with mammals that possess some additional three, HRI, PKR and PEK/PERK [61]. T. brucei and L. donovani PERK orthologues [62, 63] have been implicated in the response to ER stress and their activation leads to a decrease in the overall translation [62]. At the moment, it is still not clear whether phosphorylation of elF2 in trypanosomatids would result in a downstream signalling cascade, as bZIP type transcription factors, that could act like GCN4 or ATF4, are absent in these organisms [64].
The close relation between L. infantum and L. donovani species and the 99% identity of AS-A between both makes the discrepant phenotype intriguing. In the literature, several cases in which knocking out a gene can have different impact on virulence depending on the species can be found. [65]. Nevertheless, to our knowledge, there is no documented example among cutaneous or among visceral species of a gene that is detrimental for survival in one species and dispensable in other closely related species. However, we did find a case of differences at a strain level for instance [66]. Nonetheless, firstly we must highlight that LdASA essentiality was claimed solely based on the consecutive failure in the removal of the second gene copy [29]. Secondly, our results suggest that the medium in which the experiments were performed, cM199, may explain this difference. The former lacks Asn and LiASA null mutants could not grow unless upon Asn supplementation (Fig 5H and 5I). These results reinforce the importance of the medium composition when attempting gene knock-out of metabolic enzymes, and supplementation may be detrimental when potentially generating auxotrophs [67, 68].
LiASA dKO mutants displayed no compromised infectivity in mice, suggesting that in intracellular amastigote form, either AS-B is functional or, most likely, parasites are able to uptake Asn in such an extent that compensates the lack of intracellular synthesis. In vivo treatment with L-asparaginase, which induces a decrease in Asn bloodstream levels, has been successfully used for years in the treatment of acute lymphoblastic leukemia [69] and recently it was proposed as a promising strategy to treat bacteremia caused by group A Streptococcus and eventually other extracellular bacteria [13]. However, if for some extracellular pathogens, L-asparaginase treatment seems promising, in the case of an obligate intracellular microorganism, even when simultaneously inhibiting the microbial AS-A, several issues may arise, namely the potential contribution of the host cell for Asn de novo synthesis.
Taken all together, we conclude AS-A is not a suitable drug target candidate in L. infantum, and therefore, with regard to drug development, such a protein target becomes pointless against Leishmania.
|
10.1371/journal.pntd.0004302 | Phenotypic Features of Circulating Leukocytes from Non-human Primates Naturally Infected with Trypanosoma cruzi Resemble the Major Immunological Findings Observed in Human Chagas Disease | Cynomolgus macaques (Macaca fascicularis) represent a feasible model for research on Chagas disease since natural T. cruzi infection in these primates leads to clinical outcomes similar to those observed in humans. However, it is still unknown whether these clinical similarities are accompanied by equivalent immunological characteristics in the two species. We have performed a detailed immunophenotypic analysis of circulating leukocytes together with systems biology approaches from 15 cynomolgus macaques naturally infected with T. cruzi (CH) presenting the chronic phase of Chagas disease to identify biomarkers that might be useful for clinical investigations.
Our data established that CH displayed increased expression of CD32+ and CD56+ in monocytes and enhanced frequency of NK Granzyme A+ cells as compared to non-infected controls (NI). Moreover, higher expression of CD54 and HLA-DR by T-cells, especially within the CD8+ subset, was the hallmark of CH. A high level of expression of Granzyme A and Perforin underscored the enhanced cytotoxicity-linked pattern of CD8+ T-lymphocytes from CH. Increased frequency of B-cells with up-regulated expression of Fc-γRII was also observed in CH. Complex and imbricate biomarker networks demonstrated that CH showed a shift towards cross-talk among cells of the adaptive immune system. Systems biology analysis further established monocytes and NK-cell phenotypes and the T-cell activation status, along with the Granzyme A expression by CD8+ T-cells, as the most reliable biomarkers of potential use for clinical applications.
Altogether, these findings demonstrated that the similarities in phenotypic features of circulating leukocytes observed in cynomolgus macaques and humans infected with T. cruzi further supports the use of these monkeys in preclinical toxicology and pharmacology studies applied to development and testing of new drugs for Chagas disease.
| T. cruzi is the parasite responsible for Chagas disease, a neglected tropical illness, present in endemic and also in non-endemic countries. T. cruzi parasites are spread mainly by a vector’s bite but can also be transmitted by blood transfusion, organ transplant, laboratory accidents, congenitally and by ingestion of contaminated food. Non-human primates that are also predisposed to become infected, live in places where vectors and T. cruzi exist. Similar clinical sequelae are observed in these animals when compared to humans who are infected with T. cruzi. A better understanding of the pathogenesis of T. cruzi-infected non-human primates may bring new advances to understanding human infection. Here, we explored the immunological features of cynomolgus macaques naturally infected by T. cruzi, aiming to contribute to the validation of this species as an appropriate experimental model. Infected animals displayed a similar immunological profile to that observed in humans, with high activity of cytotoxic cells and expansion of macrophages and T-cell subsets. Furthermore, by using bioinformatics tools, we demonstrated that CD14+CD56+ and CD3+HLA-DR+ cells are major determinants to segregate CH from NI group, followed by innate and adaptive cell subpopulations. Altogether, our data suggest that non-human primates are an appropriate model to study Chagas disease.
| The haemoflagellate Trypanosoma cruzi causes Chagas disease, one of the most important neglected tropical diseases of humankind [1]. There are currently an estimated 6 million to 7 million people infected worldwide, predominantly in Latin America, where infection with T. cruzi is endemic, and more than 25 million people are at risk of becoming infected [2]. Nevertheless, non-endemic areas are also at risk of an increasing health curve burden of Chagas disease, mainly due to the high level of emigration from endemic to developed countries [3].
T. cruzi infection usually progresses from an acute infection to a chronic disease characterized by low, but persistent parasitism, accompanied by a complex host–parasite relationship and imbricate activation and modulation of immunological events [4]. Besides the relevance of the immune system to the development and maintenance of different clinical forms of Chagas disease [4], immunological events seem to be associated with the therapeutic efficacy of benznidazole [5,6], which is the drug of choice for treating Chagas disease. Despite the rapid advances in Chagas disease research from basic research, further investigation is required to decipher several parasite-host interaction mechanisms in order to support the rational proposal of novel diagnostic strategies, supportive clinical monitoring tools, the discovery of new drugs, and the establishment of combined multi-drug therapeutic protocols.
In the field of drug development, the validation of experimental models is essential for enabling valid pre-clinical trials. Although murine and canine experimental models have been used for research on Chagas disease, in regard both to clinical disease manifestation and pre-clinical drug testing [7,8,9], particular physiological features of these mammalian hosts suggest that other models more closely related to humans are required for pre-clinical trials to ensure validity of translation of results to the human condition.
Several non-human primates are predisposed to get naturally infected by T. cruzi and develop similar clinical outcomes to those observed in human Chagas disease [10,11]. There have been reports of natural infection of T. cruzi in cynomolgus macaques (Macaca fascicularis), and the development of cardiomyopathy consistent with Chagas disease supporting the notion that these Old World non-human primates can manifest similar clinical disorder as observed in human Chagas disease [11,12]. Cynomolgus macaques are an important non-human primate in biomedical research and are widely used in drug development, drug testing, and toxicology. In addition to their small body size, the similarities to humans in physiological features and susceptibility to infectious diseases make cynomolgus macaques as experimental models for Chagas disease pre-clinical investigations and drug trials. To date, despite several studies that have been conducted with non-human primates infected with T. cruzi [11,12], the detailed immunological events triggered by the T. cruzi infection in any non-human primate remain to be elucidated. The investigation reported here has applied a systems biology approach to bring insights that improve our comprehension of the immunological aspects of T. cruzi infection in the cynomolgus macaque model. Cytomics represents an innovative tool of systems biology that aim to determine the molecular phenotype at the single cell level and further represent its neighborhood connections in cellular systems [13,14]. Conventional and multi-color fluorescence-based flow cytometry at the single-cell level, associated with bioinformatics software, has become an important tool in cytomics systems biology, and we have used it for analyses that link the dynamics of cell phenotype and function at high content and high throughput.
In this study we have performed a detailed single-cell phenotypic analysis of peripheral blood leucocytes and applied conventional and systems biology approaches to evaluate the immunological features of cynomolgus macaques naturally infected with T. cruzi, aiming to identify putative biomarkers that have similarities to those of humans infected with T. cruzi. Our findings provide further data to validate cynomolgus macaques as a model for pre-clinical studies of Chagas disease.
The experiments were carried out with 26 cynomolgus macaques consisting of 21 females and five males. All subjects were submitted to serological screening tests to detect anti-T. cruzi antibodies by enzyme-linked immunoassay (ELISA; Bio-Manguinhos; Oswaldo Cruz Foundation, Rio de Janeiro, Brazil) and immuno-chromatographic assay (Chagas STATPAK; Chembio Diagnostic Systems, Medford, NY). Based on the serological status, the primates were segregated into two groups, referred to as: T. cruzi naturally infected primates (CH), presenting positive serology in both tests, comprising 12 females and three males (median age = 12 years, age ranging from 2–20 years; median weight = 3.5kg, ranging from 1.9–7.9 kg); and non-infected controls (NI), including nine females and two males (median age = 13 years, age ranging from 1–20 years; median weight = 4.9kg, ranging from 1.9–7.6 kg), presenting negative serology in both tests. All T. cruzi-naturally infected cynomolgus enrolled in the present investigation presented the indeterminate chronic phase of Chagas disease, defined by the absence of patent parasitemia characteristic of chronic Chagas disease and by meticulous organ inspections carried out during necropsy to access the macroscopic aspects of esophagus, colon and heart. The gastrointestinal tract did not present any macroscopic sign of megaesophagus or megacolon, suggestive of digestive clinical form of Chagas disease. Moreover, the myocardium of all animals presented a macroscopically normal aspect, without signs of wall aneurysms. Moreover, the volume and the weight of all hearts were within normal limits. Together, these features fulfilled the criterion described by Dias et al. [15].
The cynomolgus macaques (Macaca fascicularis) included in this cross sectional study were housed in metal and concrete indoor/outdoor enclosures at the Southwest National Primate Research Center (SNPRC), San Antonio, TX, USA. The macaques were provided water and food ad libitum, the food consisting of commercial monkey chow, vegetables and fruits. The animals were maintained in accordance with the Guide for the Care and Use of Laboratory Animals under protocols approved by the Institutional Animal Care and Use Committee (#1050MF). This study was conducted in accordance with the U.S Animal Welfare Act, and the Public Health Service Policy on Humane Care and Use of Laboratory Animals.
General anesthesia was achieved by immobilizing the animals with an intramuscular injection of ketamine hydrochloride (10mg/kg) and valium (5mg). Besides that, the animals were kept up on isofluorane (1.5%) inhalation. Following anesthesia, 5mL sample of peripheral blood was collected from the femoral vein of each animal using ethylenediamine tetraacetic acid (EDTA) as the anticoagulant. After blood collection, the immunophenotypic features of peripheral blood leucocytes were analyzed by flow cytometry.
Mouse anti-human monoclonal antibodies (mAbs) specific for cell surface markers, showing cross-reactivity to non-human primates, were used in this study. Multiparametric flow cytometry immunophenotyping approaches were carried out by simultaneous use of fluorescein isothiocyanate-FITC, phycoerythrin-PE, PerCP-Cy5.5, APC or Alexa fluor 700 conjugated mAbs. The panels were assembled with anti-CD4 (L200), anti-CD14 (322A-1), anti-CD16 (3G8), anti-CD32 (FLI.826), anti-CD64 (10.1), anti-Granzyme A (CB9), anti-Granzyme B (GB11) and anti-Perforin (DG9) antibodies labeled with FITC; anti-CD4 (L200), anti-CD14 (MϕP9), anti-CD54 (LB-2), anti-CD56 (B159) and anti-CD69 (FN50) antibodies labeled with PE; anti-CD4 (L200), anti-CD8 (SK1) and anti-HLA-DR (L243) antibodies conjugated with PerCP-Cy5.5; anti-CD8 (3B5), anti-CD16 (3G8) and anti-CD20 (2H7) antibodies conjugated with APC and anti-CD3 (SP34-2) antibodies conjugated with Alexa fluor 700. Fluorescent labeled mouse isotypic reagents were included as internal controls in all flow cytometric batches.
Whole blood cell samples were used for immunophenotypic analysis as recommended by the monoclonal antibody manufacturer, Becton- Dickinson (Mountain View, CA, USA), modified as follows: 100μL of whole peripheral blood was incubated with 5μL of undiluted fluorescent labeled mAb in 12x75mm tubes in the dark, for 30 min at room temperature. Following incubation, lysis of erythrocytes was performed by the addition of 2mL of FACS Lysing Solution (Becton Dickinson Biosciences Pharmingen, San Diego, CA, USA) vortexing, followed by incubation in the dark for 10 min at room temperature. The leukocyte suspension was then washed twice with phosphate-buffered saline (PBS) containing 0.01% sodium azide. Stained cells were fixed with 200μL of FACS-FIX Solution (10g/L paraformaldehyde, 10.2g/L sodium cacodylate, 6.65 g/L sodium-chloride, 0.01% sodium azide) and stored at 4°C, up to 24h, until flow cytometry processing.
Intracellular analyses of Granzyme A, Granzyme B and Perforin in CD16+ and CD8+ cells were performed by staining 100μL of whole blood with 5μL of anti-CD16 or anti-CD8 mAbs, in the dark for 30 min at room temperature. Following incubation, erythrocytes were lysed and leukocytes were fixed with FACS-FIX Solution, and the remaining cell suspension was permeabilized with 2mL of FACS perm-buffer (FACS buffer supplemented with 0.5% saponin, Sigma), in the dark for 10 min at room temperature. Following, cells were washed with 2mL and resuspended into 100μL FACS perm-buffer. The fixed/permeabilized, membrane-stained leukocyte suspension was then incubated with 5μL of anti-Granzyme A, anti-Granzyme B or anti-Perforin in the dark, for 30 min at room temperature. After intracytoplasmic staining, the cells were washed once with FACS perm-buffer, followed by one wash with FACS buffer and then fixed in 200μL of FACS-FIX Solution and stored at 4°C, up to 24h, until flow cytometry processing.
A total of 30,000 events per sample were acquired in a CyAn ADP flow cytometry analyzer (Beckman Coulter, Inc., Brea, CA, USA). Data acquisition and analyses were performed using the Summit software 4.3.01 (Beckman Coulter, Inc., Brea, CA, USA). Distinct gate strategies, as previously described by Vitelli-Avelar et al. [16], were applied for data analysis using the FlowJo software (version 9.4.1, TreeStar Inc. Ashland, OR, USA).
The analyses of monocyte subsets and activation status are shown in Fig 1A. Data analyses revealed that the frequency of CD14+CD16+ macrophage-like cells and CD14+ HLA-DR++ pro-inflammatory monocytes subsets did not differ between groups. However, despite the unaltered expression of CD64, monocytes from T. cruzi-infected monkeys showed increased expression of activation-related surface markers, such as CD32 and CD56 (Fig 1A).
The frequency of NK-cell subsets, along with the intracytoplasmic expression of cytotoxicity-linked molecules and activation-related surface markers are shown in Fig 1B. Despite no difference in the frequency of circulating NK-cell subsets, a higher percentage of NK Granzyme A+ cells was observed in T. cruzi-infected monkeys as compared to non-infected monkeys. No differences in the percentage of Granzyme B+ and Perforin+, or in CD69+ or CD54+ NK cells, were observed between groups (Fig 1B).
The analyses of T-cell subsets, adhesion molecule expression and activation status are shown in Fig 2A. Although no differences were observed in the frequency of circulating T-lymphocytes and CD4+ or CD8+ T-cell subsets, higher percentages of CD3+CD54+ cells and CD8+CD54+ T-cells were found in infected monkeys than in non-infected monkeys (Fig 2A). The analysis of activation-related markers revealed an increased percentage of HLA-DR+ T-cells, selectively within the CD8+ T-cell subset, in infected monkeys, with no difference in the percentage of early activated CD69+ T-cells. No significant differences were found in adhesion molecule expression or activation status of the circulating CD4+ T-cell subset (Fig 2A).
Additional analyses of cytotoxicity-linked intracytoplasmic marker expression by CD8+ T-lymphocytes are shown in Fig 2B. The results revealed an increased frequency of Granzyme A+ as well as Perforin+ CD8+ T-cells in infected monkeys than in non-infected monkeys. No difference was observed in the expression of Granzyme B by CD8+ T-lymphocytes (Fig 2B).
The percentage of peripheral blood B-cells along with their activation/regulatory status are shown in Fig 2C. The statistical analysis demonstrated that despite the increased percentage of circulating B-cells observed in infected primates, no significant difference in their early activation status (CD69) was found in T. cruzi-infected as compared to non-infected monkeys. Interestingly, up-regulated expression of the regulatory cell surface molecule CD32 by B-cells was characteristic of infected monkeys as compared to non-infected controls (Fig 2C).
Exploratory analysis of biomarker networks demonstrated that non-infected controls displayed a balanced cross-talk between innate and adaptive immunity cells, represented by evenly distributed attributes, including positive and negative axis correlations. On the other hand, T. cruzi infected monkeys showed a clear shift toward a bimodal network profile with preferential circuit involving the adaptive immunity compartment, represented by moderate and strong positive correlation axes (Fig 3).
To verify the profile of innate immunity associated with T. cruzi infection in primates, we constructed a matrix in a heat map representation (Fig 4A). Moreover, we carried out a decision tree classification in other to identify the innate immunity biomarkers most able to discriminate infected from control monkeys (Fig 4B). The heat map analysis of innate immunity clearly demonstrated the ability of CD14+CD56+ biomarker to cluster most infected monkeys apart from the uninfected controls (Fig 4A). This finding was further confirmed in the decision tree classification analysis that indicated this biomarker as the most relevant element, followed by NK Granzyme A+ cells and NK CD16+CD56- cells which together represent the most promising set of innate immunity biomarkers with putative clinical application. The performance of these selected biomarkers was further investigated by scatter plot distribution and ROC curve analysis (Fig 4C). Our data demonstrated that CD14+CD56+/NK Granzyme A+/NK CD16+CD56- cells together display a moderate global accuracy, ranging from 0.72 to 0.88 (Fig 4C). The decision tree displayed a mean accuracy of 0.54 by 10-fold cross validation, being more efficient in classifying monkeys as being infected (10 out of 15).
To verify the profile of adaptive immunity associated with T. cruzi infection in primates, we constructed a matrix in a heat map representation (Fig 5A). Additionally, we assembled a decision tree classification in order to identify the most promising adaptive immunity biomarkers able to distinguish infected monkeys from control monkeys (Fig 5B). The heat map analysis suggested that the HLA-DR activation marker, expressed by T-cells, especially by the CD8+ T-cell subset, is a reliable biomarker to identify infected monkeys (Fig 5A). This finding was further confirmed by the decision tree classification analysis that indicated that these biomarkers along with CD8+Granzyme A+ cells represent a good set of adaptive immunity biomarkers to support clinical investigations of T. cruzi infection of non-human primates. The performance of these selected biomarkers was further investigated by scatter plot distribution and ROC curve analysis (Fig 5C). Data analysis demonstrated that, together, CD3+HLA-DR+/CD8+HLA-DR+/ CD8+Granzyme A+ T-cells displayed a moderate global accuracy, ranging from 0.82 to 1.0 (Fig 5C), adjusted to 0.73 by 10-fold cross validation, for identifying infected subjects (12 out of 15) with low false-positive identification in the control group (4 out of 11).
In the present study, we have performed the phenotypic features of circulating leukocytes, focusing on the frequency of subsets and their activation status in cynomolgus macaques naturally infected with T. cruzi. The main findings revealed a similar pattern of immunological status likewise that observed in human indeterminate Chagas disease. The relevance of these results is to support the use of cynomolgus macaques in preclinical toxicology and pharmacology studies applied to development and testing of new drugs for Chagas disease.
It is well known that the immunological response plays a major role in the pathogenesis of Chagas disease [4]. To establish effective treatments, drug trials must be conducted in experimental models, prior to be administered to humans. Thus, it is important to validate an animal model that present similar immunological and clinical manifestations as those observed in humans. Indeed, non-human primates have demonstrated to be of great potential for such proposal, since they show similarities with human Chagas disease [10,11].
Macaques are important models for a remarkable diversity of human infectious diseases. Using these models, many studies have contributed novel insights into physiological and pathogenic mechanisms and also have revealed the involvement of distinct immunological events that can be used as comparative parameters in research on human diseases [10,12,19,20,21]. In fact, the study of experimentally induced or naturally occurring infectious diseases in non-human primates has enabled the establishment of valuable strategies for the development of improved vaccines, diagnostic tools, and therapeutic schemes for human illnesses [20]. In the present investigation, based on the occurrence of natural infections of T. cruzi in cynomolgus macaques, and the fact that the infected monkeys develop similar clinical sequelae [21], as well as equivalent histopathological patterns [12], to those of infected humans we have tested the hypothesis that these similarities are supported by equivalent immunological events in the two species. In addition to describing the immune mechanisms triggered by T. cruzi infection in these non-human primates, this study has presented a comprehensive overview of cells involved in innate and adaptive immunity using novel systems biology approaches to describe the cross-talk between immunological elements and to select candidate biomarkers relevant for clinical studies.
The immunophenotypic analysis of circulating leukocytes from cynomolgus macaques naturally infected with T. cruzi (CH) demonstrated that CH displayed increased expression of CD32+ and CD56+ in monocytes as compared to non-infected controls (NI). There is a general consensus that the innate immune response represents an important mechanism to control parasite replication during early and chronic Chagas disease [22]. The CD14+CD56+ monocyte subset has been related to human inflammatory chronic diseases [23] and is found on peripheral blood cells of healthy monkeys [24], but little is known about its role in humans and non-human primates infected with T. cruzi. These cells are capable of becoming more frequently positive for TNF-α cytokine, express higher levels of reactive oxygen species and FcγR (CD16+ and CD32+), are more efficient antigen presenting cells, and are a good generator of cytotoxic response [23,25]. These data suggest that CD14+CD56+ monocytes could represent an important cell population in determining Chagas disease progression, and further investigations of CD14+CD56+ monocytes are needed to better define their role in the immune response in T.cruzi-infected primates.
Our data also established that CH displayed higher frequency of cytotoxicity markers, represented by NK Granzyme A+ cells, by comparison to NI. In fact, it has been demonstrated by several studies that macrophages are efficiently activated by NK derived IFN-γ, which invokes nitric oxide production and controls parasite replication during T. cruzi infection [4,26]. Furthermore, the cytotoxic activity of NK cells could also contribute to control of parasitemia, through lysis of infected host cells or killing free parasites by contact-dependent exocytosis of lytic granules, independently from perforin [27]. These data are consistent with the hypothesis that higher NK cell cytotoxic activity could be important in helping to suppress the parasitemia to very low levels, resulting in avoidance of developing a strong acquired immune response involved with disease severity [16].
Although it is widely accepted that the adaptive immune response plays a critical role in ability to control Chagas disease progression in humans, in non-human primates its mechanisms remain unclear. Our findings showed that CH developed a similar pattern of T-lymphocytes as observed in human T. cruzi infection. In fact, higher expression of CD54 and HLA-DR by T-cells, especially within the CD8+ subset, along with outstanding expression of Granzyme A and Perforin was observed in CH, underscoring the enhanced cytotoxicity-linked pattern of CD8+ T-lymphocytes. In patients in the chronic phase of Chagas disease, a robust expansion of T-cell response to parasites has been clearly demonstrated [4]. Dutra et al. [28] showed a high frequency of activated T-cells in peripheral blood of indeterminate and cardiac patients, and further studies evaluated inflammatory infiltrate from heart tissue of cardiac patients, verifying a higher level of adhesion molecule expression by endothelial cells, as well as an increased frequency of Granzyme A+ CD8+ T-cells [29]. Previous to the findings reported here, we and other investigators have reported that non-human primates infected with T. cruzi develop chronic cardiomyopathy similar to that of humans. There are also reports of amastigote nests and parasite DNA with similar inflammatory infiltrates in heart tissue from T. cruzi-infected non-human primates [12,30,31,32], further supporting the premise that these animals are excellent experimental models for research on Chagas disease.
The analysis of the B-cell compartment revealed an expansion of these cells concomitant with up-regulated expression of Fc-γRII in CH. Previous studies have demonstrated that B-cells play an important role in systemic protection against T. cruzi through participation in the synthesis of anti-T. cruzi antibodies and in the maintenance of CD8+ memory T-cells, as well as in the determination of the T-cell cytokine functional pattern [33,34]. In regard to the role of CD32+ B-cells in human Chagas disease, it has been demonstrated that patients with asymptomatic clinical forms present lower levels of these modulatory CD32 surface marker with concomitant higher antibody levels, whereas cardiac patients presented baseline expression of CD32 by B-cells with lower antibody titers. A putative up-regulation of CD32 observed in B-cells from cynomolgus macaques could influence the degree of myocardiopathy found in these animals.
Several methods have been developed to draw networks of phenotypic aspects of the immune system in order to illustrate pathways and to describe clustering of cellular cross-talking relevant to understanding the dynamics of the immune system. Data mining of innate and adaptive immunity using the biomarker network approach revealed that whereas a balanced cross-talk between innate and adaptive immunity cells was observed in non-infected controls, CH primates demonstrated a clear shift toward a bimodal network profile with a preferential circuit involving the adaptive immunity compartment, supporting the existence of strong interaction between T. cruzi infection and the adaptive immunity cells, as observed in humans. The movement toward clustered nodes in the adaptive immunity compartment observed in CH is consistent with the chronic stage of T. cruzi infection in these animals. It is possible that the analysis of animals during early stages of infection would reveal a predominant involvement of innate immunity cells as is observed in humans with the early indeterminate form of Chagas disease [35].
Aiming to further identify cell phenotypes of innate and adaptive immunity compartments as promising biomarkers with putative clinical application, we have applied computational bioinformatics tools to explore the immunological findings observed in CH and control animals. Results from this approach suggest that the CD14+CD56+/NK Granzyme A+/NK CD16+CD56- cells represented a good set of attributes from innate immunity to distinguish CH from NI. As for adaptive immunity, the CD3+HLA-DR+/CD8+HLA-DR+/CD8+Granzyme A+ T-cells were identified as the major subsets to discriminate CH from NI. The performance of these selected biomarkers was further validated by scatter plot distribution and ROC curve analysis. ROC curves were calculated to evaluate the capacity of these biomarkers to discriminate CH primates from the NI group. Moreover, they confirmed the superior performance of CD14+CD56+/NK Granzyme A+/NK CD16+CD56- cells and CD3+HLA-DR+/CD8+HLA-DR+/ CD8+Granzyme A+ T-cells to distinguish CH animals from control animals. The data analysis demonstrated that these phenotypic attributes displayed a moderate global accuracy in identifying infected subjects, even after cross validation. In this context, CD14+CD56+ and CD3+HLA-DR+ were elected as the root attributes, as is consistent with the findings in human Chagas disease that highlight the robust role of macrophages and active T-cells as relevant biomarkers in the immune response triggered by T. cruzi infection [26]. Moreover, the secondary attribute branches, composed by NK-cells and CD8+ T-cells, are also in agreement with the recognized function of these cytotoxic cells in distinct processes during Chagas disease.
Importantly, our data demonstrated that the phenotypic features of circulating leukocytes from naturally infected non-human primates resembled the pattern in human Chagas disease. The current study may present some limitations concerning particularities of T. cruzi infection by genotype TcI that may differ from those triggered by infections with other genetic groups. Moreover, further analysis of putative immunological similarities related to cardiac Chagas disease may also be accomplished in future investigations. Nonetheless, the findings presented in the current investigation clearly demonstrated that, likewise humans, non-human primates with indeterminate T. cruzi infection develop an immunological profile involving both, innate and adaptive immune response that support the use of this experimental model for testing of new drugs for Chagas disease. Since cynomolgus macaques have an immunological response to T. cruzi very similar to that of humans, they also represent a useful experimental model for testing vaccines for Chagas disease.
|
10.1371/journal.pgen.1005268 | Phylum-Level Conservation of Regulatory Information in Nematodes despite Extensive Non-coding Sequence Divergence | Gene regulatory information guides development and shapes the course of evolution. To test conservation of gene regulation within the phylum Nematoda, we compared the functions of putative cis-regulatory sequences of four sets of orthologs (unc-47, unc-25, mec-3 and elt-2) from distantly-related nematode species. These species, Caenorhabditis elegans, its congeneric C. briggsae, and three parasitic species Meloidogyne hapla, Brugia malayi, and Trichinella spiralis, represent four of the five major clades in the phylum Nematoda. Despite the great phylogenetic distances sampled and the extensive sequence divergence of nematode genomes, all but one of the regulatory elements we tested are able to drive at least a subset of the expected gene expression patterns. We show that functionally conserved cis-regulatory elements have no more extended sequence similarity to their C. elegans orthologs than would be expected by chance, but they do harbor motifs that are important for proper expression of the C. elegans genes. These motifs are too short to be distinguished from the background level of sequence similarity, and while identical in sequence they are not conserved in orientation or position. Functional tests reveal that some of these motifs contribute to proper expression. Our results suggest that conserved regulatory circuitry can persist despite considerable turnover within cis elements.
| To explore the phylogenetic limits of conservation of cis-regulatory elements, we used transgenesis to test the functions of enhancers of four genes from several species spanning the phylum Nematoda. While we found a striking degree of functional conservation among the examined cis elements, their DNA sequences lacked apparent conservation with the C. elegans orthologs. In fact, sequence similarity between C. elegans and the distantly related nematodes was no greater than would be expected by chance. Short motifs, similar to known regulatory sequences in C. elegans, can be detected in most of the cis elements. When tested, some of these sites appear to mediate regulatory function. However, they seem to have originated through motif turnover, rather than to have been preserved from a common ancestor. Our results suggest that gene regulatory networks are broadly conserved in the phylum Nematoda, but this conservation persists despite substantial reorganization of regulatory elements and could not be detected using naïve comparisons of sequence similarity.
| Similar expression patterns of orthologous genes imply similarity of developmental programs in different species. Numerous such examples have been uncovered, including hox [1], dlx [2], and dpp/BMP [3] genes, as well as genetic programs regulating photoreceptor [4] and muscle [5] development in distantly related bilaterian animals. Largely based on these and similar findings, a current view of evolution of development emerged that emphasizes the conservation of the genetic “toolkit” within animals and the relative importance of regulatory changes in driving morphological change [6].
The mechanisms responsible for expression pattern conservation are less clear, however. One possibility is that ancestral gene regulatory programs are strictly retained. An alternative is that expression similarity is mediated by divergent regulatory processes [7,8], a phenomenon known as “developmental system drift” [9]. Regulatory rewiring of the latter type is known to occur even when individual components of the diverged networks are highly conserved developmental regulators [10–12]. One way to probe the evolution of regulatory linkages is with enhancer swap experiments, in which cis-regulatory DNA from one species is used to drive expression of a reporter gene in another species (reviewed in [13]). The resulting pattern of gene expression can be compared to the pattern driven by the endogenous regulatory element, with the similarities and differences giving evidence of conservation and divergence in the gene regulatory network.
We wanted to assess the conservation of gene regulatory programs among distantly-related members of the phylum Nematoda, a group of morphologically similar worms with mostly small, vermiform bodies. This body plan is largely conserved, with numbers of certain neuronal subtypes nearly identical in even deeply diverged taxa [14,15], and the intestine arising from a clonal cell lineage [16] in most (but not all, see [17]) nematodes studied. However, instances of developmental divergence have been documented in this clade [18–23]. We therefore performed enhancer swap experiments with regulatory elements of genes expressed in two subsets of neurons and in the developing intestine. By examining the function of cis regulatory sequences from four different nematode species in transgenic C. elegans, we sought to determine the extent of cis-regulatory conservation within this phylum.
The phylum Nematoda is comprised of animals with simple vermiform body plans and diverse life-history strategies. To look for evidence of gene regulatory conservation across this phylum, we carried out a series of enhancer-swap experiments between several distantly-related nematodes and a C. elegans host. Regulatory regions from orthologous C. elegans genes driving the mCherry reporter were co-expressed as controls with the exogenous cis elements driving expression of the GFP gene. This approach allows us to isolate and compare cis-regulatory functions of the two orthologous regulatory elements in a common trans-regulatory background. Any observed differences can then be attributed to the divergence of the cis-regulatory DNA.
We sought broad coverage of the phylum, which is hypothesized to have diversified in the Silurian [24]. Representatives from two basally branching nematode groups have sequenced genomes [25]. These are the Chromodorea (comprised of Clades III-V) and the Dorylaimia (Clade I). No Enoplia (Clade II) genomes have been sequenced to date. For this study we used C. elegans [26] as the transgenic host species, and its congeneric C. briggsae [27] to test divergence of regulatory elements among close relatives (both are from Clade V). The next most closely related nematode species is Meloidogyne hapla (Clade IV, [28,29]), followed by Brugia malayi (Clade III, [30,31]). Finally, as a representative of Clade I, we used Trichinella spiralis [32]. Divergence of Clade I was one of the earliest events in nematode evolution. The relationships among these five species are shown in Fig 1. We leveraged both this phylogeny and the amenability of C. elegans to genetic manipulation to create a series of comparisons of expression of cis-regulatory elements from progressively more distantly-related species in transgenic C. elegans. C. elegans have been used as transgenic hosts of regulatory DNA from a number of different species (reviewed, along with similar studies using Drosophila melanogaster, in [13]), however, to our knowledge, this study is the first explicit test of the relationship between evolutionary relatedness and conservation of cis-regulatory function among a set of genes.
While our selection of species gave us unprecedented ability to test the phylogenetic limits of regulatory conservation, it also rendered reciprocal transgenesis infeasible due to the complex modes of reproduction of the parasitic species.
We selected genes that have considerable conservation of their coding sequences and are single-copy orthologs among the species. Two genes are expressed in GABAergic neurons in Caenorhabditis nematodes, unc-47 [33,34] and unc-25 [35]. We have previously investigated the evolution of their regulation within this clade [36–39]. The third gene, mec-3, is expressed in another neuronal cell type, the touch-receptor neurons, in C. elegans [40,41]. The regulatory region of the C. briggsae ortholog of mec-3 has previously been shown to drive gene expression in C. elegans [42]. Finally, we chose the gene elt-2, which is expressed in the endoderm [43,44], and shows evidence of regulatory conservation outside the genus Caenorhabditis [45]. These cis-regulatory elements are expressed in different cell types, and drive expression of terminal differentiation genes (unc-47 and unc-25) as well as transcription factors (mec-3 and elt-2). Where possible (see Materials and Methods), the putative regulatory regions we investigated ranged from the start of recognizable protein-coding sequence conservation with C. elegans on the 3’ end to the next upstream coding element on the 5’ end. This choice of putative regulatory sequences in no way depended on non-coding conservation between species.
All but one of the 12 regulatory sequences from distantly related species that we tested in C. elegans directed expression in at least a subset of the expected cells, so some degree of functional conservation is preserved even at these great phylogenetic distances. Since the putative regulatory regions from the distant relatives were selected without regard for non-coding conservation, we next examined them for sequence similarity with the C. elegans orthologs. We did not know, a priori, what types of sequence similarity to expect, and did not find any extended sequence conservation. For this reason, we conducted three types of sequence comparison to ascertain the extent of sequence similarity between C. elegans and each of the distantly-related nematodes.
First, we created dotplots, which depict the positions of nucleotide strings of a certain length that are shared by the C. elegans unc-47 sequence and a sequence from another nematode (10 bp examples shown in Fig 6A–6D). Only the C. briggsae cis element displayed substantial evidence of sequence conservation, represented by collinear blocks of sequence with conserved spacing upstream of the translation start site (upper right diagonal, Fig 6A). Not only do the distantly-related nematodes lack any such collinear blocks of sequence (evidence of conservation), they lack much in with way of sequence similarity as well, with only a few scattered motifs found in both the C. elegans unc-47 upstream region and those upstream regions from M. hapla, B. malayi, and T. spiralis (Fig 6B–6D).
We looked more closely at the few 10 bp motifs in each of the divergent sequences that are shared with the C. elegans cis element (Fig 6B–6D). Since the functional units of cis-regulatory elements are thought to be short binding sites, we next hypothesized that the divergent cis elements might be enriched for such short, shared motifs. We tested this in two ways. First, we broke the sequences down into their component k-mers, and asked what percentage of the total sequence length was made up of k-mers shared with the C. elegans sequence. For example, by definition, 100% of the M. hapla unc-47 cis element is made up of 1-mers (A, T, G, or C) that are also found in the C. elegans unc-47 cis element. Approximately 40% of the examined cis elements of C. briggsae and the other 3 nematodes are made up of 8-mers that are also found in the C. elegans sequence (Fig 6E), suggesting that window sizes shorter than 9 nucleotides are not likely to be informative for this comparison. For 9-mers, slight differences in the proportion of shared sequence can be detected among species; at window sizes of 10–12 nucleotides, the difference between C. briggsae and the distantly-related nematodes becomes apparent (Fig 6E). Note that the B. malayi unc-47 cis element, while it functions remarkably better than the T. spiralis ortholog (Fig 2), is not substantially more similar in sequence to the C. elegans regulatory element. None of the three distantly-related nematodes had any identical sequence blocks longer than 12 nucleotides, and blocks longer than 10 nucleotides were primarily low-complexity polynucleotide sequences (S6 Fig), while C. briggsae had identical sequences of up to 23 nucleotides in length (Fig 6E). Alignments showing all of the identical sequence matches in the unc-47 upstream regions that are 9 nucleotides or longer can be found in S6 Fig. These identical blocks are not enriched proximal to the start of the coding sequence. Similar levels of conservation were found for unc-25, mec-3, and elt-2 as well (S7–S9 Figs).
The next method that we used to test whether the orthologous cis elements were enriched for short motifs shared with the C. elegans unc-47 upstream region compared the number of shared motifs detected with the number that might be expected by chance. Here, “chance” refers to a random reordering of the C. elegans sequence that preserves nucleotide, dinucleotide, or trinucleotide frequencies. For each of the four genes, we reshuffled the C. elegans sequence 1000 times. The cis elements from C. briggsae, M. hapla, B. malayi, and T. spiralis were compared to each of the 1000 reshuffled C. elegans sequences, and we calculated the numbers of nucleotide blocks (length 8–12) that were identical between each reshuffled C. elegans cis element and each of the orthologs. This provided empirically derived distributions of sequence identity that could be expected solely as a result of basic nucleotide composition properties. The results for tests of 10 nucleotide blocks are shown in Fig 6F–6I. For the 1000 comparisons between the reshuffled C. elegans sequences and the other nematode’s upstream unc-47 sequence, the number of identical motifs was plotted (Fig 6F–6I). The number of motifs form distributions centered between about 10–20 motifs per reshuffled sequence, depending on the length of the ortholog. Comparing the actual number of conserved blocks of various lengths between C. elegans cis elements and their orthologs revealed that only C. briggsae had more sequence identity than our “chance” rearrangements, with 63 identical 10-mers (Fig 6F). The other three distantly-related nematodes’ sequences had no more similarity than expected by chance, with numbers of shared 10-mers that fell close to the means of the distributions (Fig 6G–6I). The same was true for the upstream noncoding sequences of unc-25, mec-3, and elt-2 (S10 Fig).
Comparisons of noncoding sequence identity did not reveal any substantially conserved regions likely to be responsible for the functional conservation of orthologous cis elements. And yet, 11 of the 12 cis-regulatory elements from deeply diverged nematodes drove gene expression in C. elegans that recapitulated at least some of the expected endogenous expression pattern. We therefore searched the orthologous sequences for motifs known to be functionally important in the C. elegans sequences. Expression of unc-47 is regulated by direct binding of UNC-30 [47] to TAATCC sites. Mutations to this motif abolish expression in the D-type neurons [47]. Perhaps functional conservation of the unc-47 cis elements from distantly-related nematodes is due to the presence of this and other short sequences below the level of detection in our naive sequence comparison.
Searching for the TAATCC site revealed a perfect match, including one flanking base pair on either side in C. briggsae, with similar spacing from the translational start site (Fig 7). The noncoding sequence upstream of M. hapla unc-47 has three instances of this motif, all on the reverse strand, with additional identical nucleotides flanking the core site (Fig 7). The upstream sequences from B. malayi and T. spiralis lack perfect matches to this consensus, but do have 5/6 bp core matches with some additional flanking identity (Fig 7). Either these close matches are divergent cis-regulatory sites, hinting at evolved differences in TF-TFBS recognition, or else there is more to the control of expression in D-type neurons than we have recognized in C. elegans thus far.
The C. elegans UNC-30 binding site controls expression in D-type neurons, but not in DVB in the tail or AVL, RIS, or the RMEs in the head [47]. One site that contributes to expression in DVB, RIS, and AVL is the AHR-1-like motif [37]. This motif has the sequence CACGC and is conserved in sequence and position between C. elegans, C. briggsae, C. brenneri, and C. remanei [37]. A match to this motif is found on the reverse strand of the M. hapla unc-47 cis element (Fig 7). A palindromic sequence CACGCGTG, that is, two overlapping AHR-1-like motifs on opposite strands, along with an additional single instance of this motif, are present upstream of the T. spiralis unc-47 gene (Fig 7).
Similar motif-matching analyses were carried out for the other three sets of orthologous cis-regulatory elements. Matches to motifs known to be necessary for function in C. elegans were identified in almost all tested orthologs from distantly-related nematodes (S1 Text; S11–S13 Figs). However, the occurrence of even multiple instances of motifs corresponding to transcription factor binding sites should not be construed as evidence of conservation. First, these motifs are not found any more frequently than in randomly reshuffled C. elegans sequences. We explicitly estimated the probability of finding these motifs in the randomly reshuffled C. elegans sequences. The probability of finding TAATCC (the UNC-30 binding site) in sequences preserving the single-nucleotide composition of the C. elegans cis element was 0.558, conserving dinucleotides it was 0.378, and trinucleotides it was 0.566. The probabilities of finding CACGC (the AHR-1-like motif) in these same sequences were 0.632, 0.606, 0.674, respectively. Second, these motifs were routinely found in the cis elements of the other genes we examined (S3 Table). Third, these motifs are often found on the opposite strand, suggesting that, while individual motifs are born and die, this sequence turnover maintains at least one instance of the motif in each of the orthologous regulatory elements. Therefore, identical motifs are not, strictly speaking, conserved.
It is suggestive that the unc-47 regulatory sequence from a distantly-related nematode that retains the best function in D-type neurons, that of M. hapla, has the best match to the UNC-30 binding site. Similarly, the regulatory sequence with the best function in DVB—T. spiralis unc-47—has the best matches to the AHR-1-like motif. We therefore tested the contribution of these motifs to functional conservation.
We introduced mutations into an UNC-30 binding motif in the M. hapla unc-47 cis element. This motif was selected because it shares the longest similarity in the flanking sequences with the UNC-30 binding site of the C. elegans unc-47 cis element (Fig 8A). The mutant M. hapla unc-47 sequence directed less consistent expression in the D-type neurons than its wild-type counterpart (Fig 8B, 8C and S14A), suggesting that this UNC-30 motif contributes to control of gene expression in the D-type neurons.
Next, we introduced mutations into the palindromic double AHR-1-like motif of the T. spiralis unc-47 element, eliminating the consensus sequence on both strands (Fig 8D). This resulted in a substantial decrease in the fraction of animals expressing the transgene in RIS and DVB (Figs 8E–8H and S14B). This suggests that the palindromic AHR-1-like motif upstream of the T. spiralis unc-47 gene is partially responsible for expression in DVB and RIS, just as the AHR-l-like motif is in C. elegans [37].
Neither mutation eliminated expression in the affected cells entirely, implying that these sites contribute to but are not strictly essential for expression. This could be due to the redundancy of these binding sites in both cis elements (Fig 7). As another possible explanation, consider the case of the B. malayi unc-47 element that lacks good matches for either the UNC-30 or the AHR-1-like motifs, and yet is reasonably well expressed in both the D-type neurons and in DVB. It is possible that some orthologous cis elements retain functional conservation via sequences that can be recognized by C. elegans transcription factors, but that we currently cannot recognize as functional.
We investigated cis-regulatory function in an explicitly evolutionary framework. The extent of divergence between the species involved in this study ranged from that of congenerics (C. elegans and C. briggsae) to the deepest in the phylum Nematoda (C. elegans and T. spiralis). This allowed us to test how regulatory information breaks down over time. Transgenic experiments were conducted in the “common garden” of C. elegans to control for the effects of trans-regulatory divergence and to focus comparisons on the cis elements (see Potential Caveats in Materials and Methods). We selected the putative regulatory regions without regard for non-coding sequence similarity, which then permitted us to comment on the relationship between functional and sequence conservation. We used a standard methodology to look at the cis elements of four genes, allowing us to make four generalizations.
First, despite the vast spans of evolutionary time that we sampled—the most distantly-related species diverged perhaps as long as 400 million years ago [24]—the majority of the cis-regulatory elements exhibited appreciably conserved gene regulatory function in C. elegans (Table 1). Two reasons compelled us to focus explicitly on the conserved, rather than divergent, aspects of expression. First, at such great phylogenetic distances, any conservation might be less expected than divergence. Second, for technical reasons, we can only know the endogenous function of the C. elegans regulatory elements, not the patterns driven by the divergent cis elements in their native species (see Potential Caveats in Materials and Methods). Our findings are consistent with previous reports of functional conservation of cis-regulatory elements between distantly-related members of the same phylum, most extensively tested in arthropods and chordates [48,49]. Although there have been a number of reports of functional conservation of cis elements between different phyla [50–56], this is not true for cis elements of all genes tested [39,57]. It is possible that the evolutionary dynamics of regulatory elements may be sufficiently idiosyncratic to preclude general conclusions about the “outer limits” of cis-regulatory conservation.
Second, in most cases cis-regulatory elements from more distant relatives have retained less function than elements from closer relatives. However, there are notable exceptions and, importantly, the pattern of functional divergence that we observed reflects modular organization of cis-regulatory elements—separable elements control different aspects of expression [58–60]. Due to modularity of cis elements, evolution can “tinker” with some functions while avoiding pleiotropic effects on others [61]. In C. elegans, expression of unc-47 is controlled by different mechanisms in D-type neurons and DVB, RIS, and AVL [37,47]. Accordingly, we see that whereas the T. spiralis unc-47 element is not expressed in D-type neurons, it functions relatively well in DVB and RIS (Figs 2 and S1). In contrast, the M. hapla unc-47 element is expressed well in the D-type neurons, but not in DVB (Fig 2). Similarly, the elt-2 elements from M. hapla and B. malayi are expressed reasonably well during embryogenesis, but not in later stages (Fig 5). We consider this good evidence for separate regulation of pattern, timing, and levels of expression, as well as substantiating evidence that the weak expression of some of these regulatory elements is due to genuine divergence of regulatory information rather than experimental artifacts of weak transgene expression. We conclude that modular organization of cis elements manifests in different rates of divergence for different aspects of expression patterns [62] and may be quite common [13]. Mechanisms controlling spatial, temporal, and levels of expression may be particularly prone to different rates of divergence (e.g. [45,63]).
Third, despite their substantially conserved functions, the regulatory elements of all species but C. briggsae have not retained more sequence similarity than would be expected by chance. This finding is consistent with previous reports that suggested that conservation of cis-regulatory function does not, strictly speaking, require extended sequence conservation [64–70]. Since different types of regulatory elements evolve under different constraints [36], relying on sequence conservation to find cis-regulatory elements might bias discovery to only particular types of elements with highly constrained sequences [71]. Additionally, because sequences of different elements evolve at different rates [38], it is not a priori clear how distant the species to be compared should be to discover cis elements of different types. Even when some short stretches of identical nucleotides are discovered between distantly-related orthologous cis elements, this should not be taken as evidence of conservation. This is because many short matches will always be found by chance, particularly in regions with biased nucleotide composition. For instance, the co-occurrence of the UNC-30 and AHR-1-like motifs upstream of unc-47 orthologs (Fig 7) is more plausibly explained by a birth-and-death process rather than strict conservation, considering that these motifs are found on opposite strands of DNA in different species.
Fourth, despite the lack of extended sequence conservation, for all four genes we could readily identify motifs corresponding to transcription factor binding sites previously identified as functionally important for regulation of C. elegans orthologs. The motifs that we tested contributed to gene regulation of the orthologous cis elements, implying that gene regulatory output can be conserved, even among distantly-related organisms, as long as key gene regulatory connections—“kernels” [10,72] or “input-output devices” [73]—are maintained. This further reinforces the view that when developmental programs evolve, the regulatory “toolkit” controlling major patterning and cell-type specification programs remains relatively static [6]. Of course, the mere presence of these short motifs is not likely to be sufficient to explain regulatory output. For instance, we can find chance matches to GATA motifs important for elt-2 expression in many of the other sequences we tested, which do not drive expression in the intestinal precursor cells. Similarly, we can find matches to the AHR-1-like motif (that regulates unc-47 expression) in the elt-2 cis elements of C. elegans, C. briggsae, and T. spiralis, none of which drive expression in DVB, RIS, or AVL.
In this study we aimed to understand how the patterns of divergence of gene regulatory mechanisms between closely related species scale up over long evolutionary times. Models have predicted [74] that regulatory control can be shifted from one site to another within a cis-regulatory sequence; if these sites arise somewhat stochastically, longer wait times increase the likelihood of new sites originating and being optimized. These new sites could diminish the strength of purifying selection acting on ancestral motifs [75]. On shorter evolutionary time scales, new motifs do not have the time to arise, so function relies on conservation of existing sites [76]. As the same process plays out over different timespans, cis-regulatory conservation remains common among close relatives, but is mostly absent among more distantly-related species.
Naturally, the rates of divergence and motif turnover are different for different genes. An important factor determining the rate of evolution could be the organization of a cis element, whether it is flexible [77,78] or constrained [79], a billboard or an enhanceosome [80,81]. Modeling suggests that some enhancer sequences are inherently more prone to higher rates of turnover than others [74]. Better understanding of the structure of cis-regulatory elements may provide clues to their evolution [70,82,83].
Practically, our results advocate the use of C. elegans as a convenient and reliable experimental system for testing the functions of putative regulatory elements from nematode species, many of them parasites of major economic and medical significance, that are not amenable to transgenic studies [45]. Furthermore, the fact that C. elegans has been a genetic model system for decades means that the wealth of information about gene regulation in this species could be leveraged into hypothesis-driven investigation of non-model organisms.
As discussed above, functionally conserved sequences can retain no more sequence conservation than would be expected by chance. Indeed, motifs that mediate functional conservation, namely transcription factor binding sites, are short enough that they would be likely to be found by chance in sequences of the lengths of these cis elements. By the measures of sequence conservation we applied, including alignment-free methods, M. hapla does not have appreciably greater sequence similarity to C. elegans than does T. spiralis. Nevertheless, M. hapla cis elements of all four tested genes drive more consistent and correct expression in transgenic C. elegans than elements from T. spiralis do. This means that some sequence properties were retained to a greater extent by the more closely related species. Identification of these properties would lead to a better understanding of function and evolution of gene regulatory elements.
Orthologous genes from C. briggsae, M. hapla, B. malayi, and T. spiralis were identified as best tblastn/blastx matches with the C. elegans protein sequence. For C. briggsae, B. malayi, and T. spiralis, the genome browser on Wormbase was used. For M. hapla, the genome browser at www.hapla.org was used. Forward primers were designed proximal to the next upstream gene, or failing that the 5’-most part of the contig on which the orthologous coding sequence was found. Reverse primers were selected to make in-frame translational fusions with GFP in the 5’-most part of the gene with protein coding sequence similarity with C. elegans. The only cases in which this was not possible were B. malayi and T. spiralis elt-2, in which protein-coding conservation started deep in the protein-coding sequence, and the fusions were generated in the first exon. A previous study of elt-2 from a parasitic nematode, the less divergent Haemonchus contortus [84], found that despite protein sequence divergence from C. elegans, the H. contortus protein retained function when expressed transgenically in C. elegans by a C. elegans heat shock promoter, so this increases our confidence that these can be elt-2 orthologs despite coding sequence divergence. In all cases, the start codon of the ortholog was included in the fusion. To generate reporter transgenes, upstream non-coding sequences were PCR amplified from genomic DNA and cloned upstream of GFP into the Fire vector pPD95.75, or upstream of mCherry (for C. elegans genes), which was inserted in place of GFP in a modified vector pPD95.75 [85]. elt-2 transgenes carried a nuclear localization signal upstream of GFP or mCherry. Prior to injection, all transgenes were sequenced to ensure accuracy.
We injected a mixture (5 ng/μL (for C. briggsae; 10 ng/μL for the other species) promoter::GFP plasmid, 5 or 10 ng/μL promoter::mCherry plasmid, 5 ng/μL pha-1 rescue transgene, 100 ng/μL salmon sperm DNA) into temperature-sensitive C. elegans pha-1(e2123) strain [86]. Transformants were selected at 25°C. Multiple strains were examined for each transgenic construct. Statistical analyses of consistency of expression patterns between strains and individuals are presented in S2 Table, since extrachromosomal transgenes are known to have more variable expression than integrated transgenes. Our previous reports [36,37] thoroughly addressed the similarity of expression driven by transgenes of different types—extrachromosomal, multicopy integrated, and single-copy integrated. We found that while the strength of the signal increases with multiple copies, and variability increases with extrachromosomal transgenes, the patterns generated by these different methods are consistent.
The structures of extrachromosomal transgene arrays are generally not known. Although there is a possibility of cross-talk between promoters from different species if they land close enough when the DNA is concatenated, we mitigate against this by including an excess of salmon sperm DNA and vector sequence to create distance between the promoters and reduce the repetitiveness of the arrays. We measure expression in multiple independent strains. We also tested several of the highly divergent promoters alone, without a coexpressed C. elegans-DNA-driven reporter (S15 Fig). Without the coexpressed mCherry marker, cells were more difficult to identify, so counts were not attempted for these strains, but expression was observed in the same subsets of cells that it was observed in coexpressing lines.
The coexpressing strains also allowed us to control for the mosaicism inherent in extrachromosomal transgenes. Since the transgenes are concatenated, mCherry and GFP are inherited together by cells, and if array loss or silencing causes the loss of expression of one marker, the other will also disappear. This is why, for most of our quantification, we describe expression as the ratio of mCherry (control) positive cells that also express GFP (see Figs 2–5 and 7).
We tested the functions of motifs corresponding to consensus sequences of binding sites of UNC-30 [47] (TAATCC) and AHR-1-like [37] (CACGC). Motifs were identified using the ConsensusSequence feature on the GeneGrokker web server (https://genegrokker.biology.uiowa.edu). Of the several UNC-30 motifs in the M. hapla unc-47 element, we selected for mutagenesis the longest extended match to the C. elegans sequence: aTAATCCcc (reverse complement, since the motif is found on the (-) strand). This motif was mutagenized to aTAGGCGac (changes highlighted). Of the several matches to the AHR-1-like motif in the T. spiralis unc-47 element, the motif selected for mutagenesis was a palindromic sequence (CACGCGTG), which matches two overlapping instances of the AHR-1-like motif (one on each strand). This sequence was mutagenized to CACAAGTG, changing the CACGC sequence on the (+) strand to CACAA and on the (-) strand to CACTT.
All mutations were introduced by PCR with overlapping, opposite-facing primers carrying the mutant sequence. Primers were used to amplify plasmid DNA carrying the wild-type sequence. Following PCR, the reaction was digested with the methylation sensitive restriction enzyme DpnI to selectively digest the wild-type plasmid template. A second PCR reaction was performed, amplifying the mutagenized cis element and some flanking vector sequence. This PCR product was purified and digested for directional cloning back into the expression vector. Mutations were verified by sequencing before microinjection.
Mixed-stage populations of C. elegans carrying transgenes were grown with abundant food. Worms of appropriate stages were selected. These were immobilized on agar slides with 10 mM sodium azide in M9 buffer. The slides were examined on a Leica DM5000B compound microscope under 400-fold magnification, except in S4 and S15 Figs, which include micrographs taken at 1000-fold magnification (as labeled). Exposure times varied as necessary for each transgene. Each photograph showing worms in figures is composed of several images of the same individual capturing anterior, middle, and posterior sections, as well as shallow and deep focus. False-colored composite images were generated with QCapturePro. Brightness, contrast, and scaling of images were adjusted where necessary in final display items.
The stronger background visible in the GFP images relative to their mCherry counterparts may have several explanations. First, GFP has higher background relative to mCherry, and the autofluorescence of the gut is detectable with GFP filters. Second, longer exposure times were necessary to capture expression of the more weakly expressing exogenous cis-regulatory elements. Finally, GFP fluorescence in the gut is a known site of off-target expression [38]. Worms were also injected with a subset of the GFP transgenes carrying the other nematode’s cis elements alone (without a C. elegans mCherry control), and results were consistent (S1 and S2 Tables, S15 Fig).
Young adult individuals were examined for gene expression, except for elt-2, in which case pretzel stage embryos and L1 larvae were counted. Worms without any visible fluorescence were assumed to have lost the transgene and were ignored. Presence of mCherry was a precondition for the worm to be counted, but without regard for the strength or completeness of the mCherry expression pattern.
Motifs matching between C. elegans and each orthologous cis element (identified by the Mirror tool on the GeneGrokker web server https://genegrokker.biology.uiowa.edu) were mapped back to the orthologous sequence, and the total amount of the sequence covered by blocks of conservation of different sizes is plotted in Fig 6A.
Empirical p-values for the sequence similarity of the C. elegans elements to their orthologs were calculated by generating 1000 reshuffled replicates of the C. elegans sequence. Replicates were generated using single, di-, and tri-nucleotide sampling from the C. elegans sequence. Each replicate was compared to each ortholog and scored for similarity in windows of different sizes. The distributions of these similarity scores were plotted (Fig 6F–6I). The actual number of observed motif matches between the C. elegans sequence and its relevant orthologs were indicated on those distributions. The reported p-value is equal to the number of shuffled replicates that had more motif matches than the actual number, divided by 1000. Only C. briggsae had more similar motifs than would be expected by chance.
We used multicopy extrachromosomal transgenes, which could have made the detected levels of expression higher and less consistent than what would have been produced by single-copy transgenes. In previous work [36,37] we did determine that, at least in the case of unc-47 from C. elegans and C. briggsae, the nature of the transgene (multi- vs. single-copy, extrachromosomal vs. integrated) did not change the pattern, but rather the amount and consistency of expression. If the same principle holds for the genes examined here, the conserved patterns we detected represent the cell types where the foreign cis elements are truly active in C. elegans, but the expression levels could be overestimated. The fact that in most instances only subsets of the overall pattern were conserved suggests that artificially higher expression levels were not solely responsible for the conserved expression patterns we detected.
Any apparent divergence—i.e. incongruence between the pattern driven by the C. elegans cis element and its orthologs—could be due to cis-regulatory changes (in the function of the donor element), trans-regulatory changes (in the function of transcription factor(s) in C. elegans), or due to the experimental combination of the two. In addition, endogenous expression patterns may have diverged between C. elegans and other species. For technical reasons, it is difficult to determine endogenous patterns of gene expression in divergent parasitic nematodes used in this study. It is even more difficult to generate transgenic animals in these species. These technical limitations make it essentially impossible to assess divergence in endogenous expression patterns or to disentangle their causes (that is, cis vs. trans changes). For these reasons, we focused on enumerating similarities, rather than differences in expression. Our tests actually underestimate the extent of regulatory conservation, because a failure of a cis element from a distant nematode when tested in C. elegans may reflect a genuine divergence in cis-regulation in that species that was compensated in trans, therefore maintaining the same overall expression pattern.
|
10.1371/journal.ppat.1007494 | Prophage induction, but not production of phage particles, is required for lethal disease in a microbiome-replete murine model of enterohemorrhagic E. coli infection | Enterohemorrhagic Escherichia coli (EHEC) colonize intestinal epithelium by generating characteristic attaching and effacing (AE) lesions. They are lysogenized by prophage that encode Shiga toxin 2 (Stx2), which is responsible for severe clinical manifestations. As a lysogen, prophage genes leading to lytic growth and stx2 expression are repressed, whereas induction of the bacterial SOS response in response to DNA damage leads to lytic phage growth and Stx2 production both in vitro and in germ-free or streptomycin-treated mice. Some commensal bacteria diminish prophage induction and concomitant Stx2 production in vitro, whereas it has been proposed that phage-susceptible commensals may amplify Stx2 production by facilitating successive cycles of infection in vivo. We tested the role of phage induction in both Stx production and lethal disease in microbiome-replete mice, using our mouse model encompassing the murine pathogen Citrobacter rodentium lysogenized with the Stx2-encoding phage Φstx2dact. This strain generates EHEC-like AE lesions on the murine intestine and causes lethal Stx-mediated disease. We found that lethal mouse infection did not require that Φstx2dact infect or lysogenize commensal bacteria. In addition, we detected circularized phage genomes, potentially in the early stage of replication, in feces of infected mice, confirming that prophage induction occurs during infection of microbiota-replete mice. Further, C. rodentium (Φstx2dact) mutants that do not respond to DNA damage or express stx produced neither high levels of Stx2 in vitro or lethal infection in vivo, confirming that SOS induction and concomitant expression of phage-encoded stx genes are required for disease. In contrast, C. rodentium (Φstx2dact) mutants incapable of prophage genome excision or of packaging phage genomes retained the ability to produce Stx in vitro, as well as to cause lethal disease in mice. Thus, in a microbiome-replete EHEC infection model, lytic induction of Stx-encoding prophage is essential for lethal disease, but actual phage production is not.
| Enterohemorrhagic Escherichia coli (EHEC), a food-borne pathogen that produces Shiga toxin, is associated with serious disease outbreaks worldwide, including over 390 food poisoning outbreaks in the U.S. in the last two decades. Humans acquire EHEC by ingesting contaminated food or water, or through contact with animals or their environment. Infection and toxin production may result in localized hemorrhagic colitis, but may progress to life-threatening systemic hemolytic uremic syndrome (HUS), the leading cause of kidney failure in children. Treatment for EHEC or HUS remains elusive, as antibiotics have been shown to exacerbate disease. Shiga toxin genes reside on a dormant bacterial virus present in the EHEC genome, but are expressed when the virus is induced to leave its dormant state and begin to replicate. Extensive virus replication has been thought necessary to produce sufficient toxin to cause disease. Using viral and bacterial mutants in our EHEC disease mouse model, we showed that whereas an inducing signal needed to begin viral replication was essential for lethal disease, virus production was not: sufficient Shiga toxin was produced to cause lethal mouse disease, even without viral replication. Future analyses of EHEC-infected human samples will determine whether this same phenomenon applies, potentially directing intervention strategies.
| Shiga toxin-producing Escherichia coli (STEC) is a food-borne zoonotic agent associated with worldwide disease outbreaks that pose a serious public health concern. Enterohemorrhagic Escherichia coli (EHEC), a subset of STEC harboring specific virulence factors that promote a specific mode of colonization of the intestinal epithelium (see below), is acquired by humans by ingestion of contaminated food or water, or through contact with animals or their environment. EHEC serotype O157:H7 is a major source of E. coli food poisoning in the United States, accounting for more than 390 outbreaks in the last two decades [1–5]. EHEC infection usually presents as localized hemorrhagic colitis, and may progress to the life-threatening systemic hemolytic uremic syndrome (HUS), characterized by the triad of hemolytic anemia, thrombocytopenia, and renal failure [5, 6]. HUS is the leading cause of renal failure in children [7].
EHEC, along with enteropathogenic E. coli and Citrobacter rodentium belong to the family of bacteria known as attaching and effacing (AE) pathogens that are capable of forming pedestal-like structures beneath bound bacteria by triggering localized actin assembly [8–10]. While this ability of EHEC leads to colonization of the large intestine, production of prophage-encoded Shiga toxin (Stx) promotes intestinal damage resulting in hemorrhagic colitis [11–17]. Shiga toxin may further translocate across the colonic epithelium into the bloodstream, leading to systemic disease. Distal tissue sites, including the kidney, express high levels of the Shiga toxin-binding globotriosylceramide (Gb3) receptor, potentially leading to HUS [14, 15, 18–21].
Genes encoding EHEC Shiga toxin are typically encoded in the late gene transcription region of integrated lambdoid prophages [22, 23] and their expression is thus predicted to be temporally controlled by phage regulons [24–27]. Early studies showed that high levels of Stx production and release from the bacterium in vitro required prophage induction, i.e., the mechanism by which quiescent prophages of lysogenic bacteria are induced to replicate intracellularly and released as phage particles by host cell lysis [27, 28]. Lambdoid phage inducers are most commonly agents that damage DNA or interfere with DNA synthesis, such as ultraviolet light or mitomycin C. These inducing stimuli trigger activation of the bacterial RecA protein, ultimately leading to the cleavage of the prophage major repressor protein, CI, allowing expression of phage early and middle genes. Late gene transcription, which requires the Q antiterminator, results in the expression of many virion structural genes and of endolytic functions S and R, which lyse the bacterium and release progeny phage [29]. Other signaling pathways involving quorum sensing or stress responses have also been implicated in lysogenic induction [30, 31].
Unfortunately, antibiotics commonly used to treat diarrheal diseases in children and adults are known to induce the SOS response. Trimethoprim-sulfamethoxazole and ciprofloxacin have been shown to enhance Stx production in vitro [32–34], and antibiotic treatment of EHEC-infected individuals is associated with an increased risk of HUS [35]. Hence, antibiotics are contraindicated for EHEC infection and current treatment is limited to supportive measures [36].
A more detailed understanding of the role of prophage induction and Stx production and disease has been pursued in animal models of EHEC infection. Although some strains of conventional mice can be transiently colonized by EHEC, colonization is not robust and typically diminishes over the course of a week [13, 37], necessitating use of streptomycin-treated [16] or germ-free mice [38, 39] to investigate disease manifestations that require efficient, longer-term intestinal colonization. In streptomycin-treated mice colonized with EHEC, administration of ciprofloxacin, a known SOS inducer, induces the Stx prophage lytic cycle, leading to increased Stx production in mouse intestines and to Stx-mediated lethality [40]. Conversely, an EHEC strain encoding a mutant CI repressor incapable of inactivation by the SOS response was also incapable of causing disease in germ-free mice [41].
A potential limitation of the antibiotic-treated or germ-free mouse infection models is the disruption or absence, respectively, of microbiota, with concomitant alterations in immune and physiological function [42]. For example, a laboratory-adapted E. coli strain that lacks the colonization factors of commensal or pathogenic E. coli is capable of stably colonizing streptomycin-treated mice [43], and, when overproducing Stx2, is capable of causing lethal infection in antibiotic-treated mice [17]. Further, as up to 10% of human gut commensal E. coli were found to be susceptible to lysogenic infection by Stx phages in vitro [44], it has been postulated that commensals may play an amplifying role in EHEC disease by fostering successive rounds of lytic phage growth [44–47]. Finally, gut microbiota may also directly influence expression of stx genes. For example, whereas a genetic sensor of phage induction suggests that the luminal environment of the germ-free mouse intestine harbors a prophage-inducing stimulus [41], several commensal bacteria have been shown to inhibit prophage induction and/or Stx production in vitro [48–50]. Alternatively, colicinogenic bacteria produce DNAse colicins that may trigger the SOS response, increasing Stx production [51].
Our laboratory previously developed a murine model for EHEC using the murine AE pathogen C. rodentium [52, 53], which efficiently colonizes conventionally raised mice and allows the study of infection in mice with intact microbiota. The infecting C. rodentium is lysogenized with E. coli Stx2-producing phage Φ1720a-02 [52, 54] encoding Stx variant Stx2dact (Stx2d activatable), which is particularly potent in mice [55, 56]. Infection of C57BL/6 mice with C. rodentium(Φ1720a-02), (herein referred to as C. rodentium(Φstx2dact)), produces many of the features of human EHEC infection, including colitis, renal damage, weight loss, and potential lethality, in an Stx2dact-dependent manner [52].
In the current study, we address phage, bacterial, and host factors that lead to lethal EHEC infection. We found that C. rodentium(Φstx2dact) strains lacking RecA, which is required for induction of an SOS response, or phage Q protein, which is required for efficient transcription of the late phage genes, did not produce high levels of Stx in vitro or cause lethal disease in mice. In contrast, mutants defective in prophage excision, phage assembly, or phage-induced bacterial lysis retained the ability to both produce Stx2dact upon prophage induction in vitro and to cause lethal disease. Excised phage genomes, potentially undergoing DNA replication leading to phage production or representing packaged phage, were detected, albeit at low levels, in fecal samples of mice infected with wild type C. rodentium(Φstx2dact), but not in mice infected with excision-defective C. rodentium(Φstx2dact). Thus, in a microbiome-replete EHEC infection model, lytic induction of Stx-encoding prophage, but not actual production of viable phage particles, is essential for Stx production and lethal disease.
Lambdoid phage Φ1720a-02 was originally isolated from EC1720a-02, a STEC strain found in packaged ground beef [54]. Our novel C. rodentium-mediated mouse model of EHEC infection encompasses C. rodentium DBS100 (also known as C. rodentium strain ICC 168 (GenBank accession number NC_013716.1)), lysogenized with phage Φ1720a-02 marked with a chloramphenicol (cam)-resistance cassette inserted into the phage Rz gene, creating strain DBS770 [52, 53]. A second lysogen, DBS771, was lysogenized with the same phage but with an additional kanamycin (kan)-resistance cassette inserted into and inactivating the prophage stx2A gene. For simplicity, strains DBS770 and DBS771 will herein be referred to as C. rodentium (Φstx2dact), and C. rodentium(ΦΔstx2dact::kanR), respectively (Table 1).
To identify phage genes critical for lethal mouse infection, we sought to inactivate specific prophage genes and then assess their resulting phenotypes in the C. rodentium mouse model. As a first step, we sequenced the parental strain DBS100 and the genomes of C. rodentium (Φstx2dact) and C. rodentium (ΦΔstx2dact::kanR), revealing that the three genomes were identical except for prophage sequences present in C. rodentium (Φstx2dact) and C. rodentium (ΦΔstx2dact::kanR) (see Materials and Methods).
We then annotated the entire Φstx2dact prophage (GenBank accession number KF030445.1; Figs 1 and S1). As is typical of Stx phages, the sequence revealed a lambdoid phage with a mosaic gene organization that does not precisely match that of phage λ, but is nevertheless somewhat syntenic with other lambdoid phages [62], (S2 Fig). Further, although lysogenized independently, C. rodentium(Φstx2dact) and C. rodentium(ΦΔstx2dact::kanR) prophages were integrated at the same location, i.e. 100 bp into the coding sequence of dusA (encoding tRNA-dihydroxyuridine synthase A). A recent study revealed that known integrase genes, at least half of which belong to prophages, were found adjacent to the host dusA gene in over 200 bacterial species [63]. Furthermore, a 21 base pair motif found at the prophage-host DNA junctions in many bacteria was present at the prophage junctions, attL and attR, of C. rodentium(Φstx2dact) and C. rodentium(ΦΔstx2dact::kanR), as well as at the presumed attB phage insertion site in the parental C. rodentium dusA gene (Fig 1). A seven-base segment within this 21-base sequence is completely conserved between attL, attR, and attB and likely represents the ‘core’ recombination site for integration or excision (Fig 1, bolded sequence; [64]). Note that, although the Φstx2dact and ΦΔstx2dact::kanR prophages interrupt the dusA gene, they encode a 184 bp ORF (designated “ΦdusA’” in Fig 1) that is in frame with the 3’ 937 nucleotides (positions 101 to 1038) of dusA
A prior analysis of the host C. rodentium DBS100 genome sequence revealed the presence of 10 additional partial and intact prophages distributed around the genome [65], although it is not known if any of these prophages can give rise to intact phage. Sequence analysis showed only 2 regions of homology between Φstx2dact and these prophages (S3 Fig): one resident prophage encoded 70% homology to a region encoding Φstx2dact Cro, CI repressor, and a hypothetical protein, and a second resident prophage showed 79% homology to another Φstx2dact gene encoding a hypothetical protein.
Although Φstx2dact harbors a cat insertion in the Rz gene, a gene that contributes to phage λ lysis under some conditions [2], prophage induction with mitomycin C resulted in lysis of C. rodentium(Φstx2dact) (S4 Fig), suggestive of lytic phage induction. Nevertheless, pilot experiments revealed that Φstx2dact plaques were not detectable on any of numerous E. coli K12 and other indicator strains (Materials and Methods). This finding is not unusual for Stx-producing phages [66–68]. To more rigorously test whether this phage can infect E. coli K12, we selected for E. coli K12 lysogens by infecting E. coli K12 strain DH5α with supernatants of mitomycin C -induced cultures of C. rodentium (Φstx2dact), then selecting for kanamycin-resistant clones. These clones were verified as lysogens by PCR-detection of phage genes (S5A Fig). DH5α lacks RecA and thus cannot undergo an SOS response to trigger prophage induction. However, when the RecA-producing plasmid pER271 was introduced to the DH5α lysogens, they were more sensitive to UV light than non-lysogens containing the same plasmid (S5B Fig), consistent with lysogenic induction. Hence, Φstx2dact is a functional phage that is capable of infecting bacteria, including E. coli K12.
In the course of EHEC infection of streptomycin-treated mice, Stx phage can be induced by antibiotic treatment to lysogenize other E coli strains in the intraluminal environment [40] [69]. It has been postulated that successive cycles of infection of non-pathogenic commensal E. coli could amplify Stx production and exacerbate disease [38, 44, 45, 47]. We first addressed this question by testing whether lysogeny of commensal bacteria by phage Φstx2dact was detectable following oral C. rodentium(Φstx2dact) infection of mice. Mice orally gavaged with C. rodentium(Φstx2dact) normally exhibit weight loss and lethal disease [52], typically succumbing to disease after day 7 post-infection. DNA was extracted from fecal samples of a group of five mice at days 1 and 6 post-infection. The DNA samples were used as a template to generate a library of sequences encompassing the sequence downstream of attL (specifically, spanning the region from the phage int gene, through Φ dusA and into the adjacent host sequence; see Fig 1). This strategy is a modification of that used for Tn-seq library analysis ([70], Materials and Methods).
Although we were unable to obtain detectable amplified DNA from fecal samples produced on day 1 post-infection, consistent with the low titer of C. rodentium(Φstx2dact) in the stool at this early time point, the day 6 post-infection sample yielded a DNA library, which was subjected to massively parallel sequencing to identify the origin of the host DNA into which the prophage was integrated. Of 17,142,098 readable sequences generated, 99.56% showed homology to C. rodentium (Φstx2dact), i.e. included C. rodentium (Φstx2dact) attL and the adjacent C. rodentium dusA gene sequence, indicating prophage integration into the original C. rodentium strain (Table 2; see Materials and Methods). For the remaining 0.44% of sequences, the C. rodentium dusA sequences adjacent to the attL core sequence were replaced by phage-specific attR sequences, thus regenerating attP. These latter sequences likely reflect excised circular phage genomes generated following induction of the C. rodentium(Φstx2dact) lysogen. Thus, C. rodentium(Φstx2dact) undergoes lytic induction in the murine host, consistent with previous findings of EHEC infection in streptomycin-treated mice. Furthermore, no integration of the Φstx2dact prophage into either a different site in C. rodentium, or into a different bacterial host was observed, leading to the conclusion that lysogeny of intestinal bacteria by Φstx2dact is not a common event in this model.
Prophage induction of lambdoid phages is often initiated by DNA damage, in which SOS pathway activation leads to RecA-promoted autocleavage of CI repressor, followed by transcription of early genes from the from PL and PR promoters. Subsequent temporally programmed transcription of the prophage genome results in the production of delayed early (middle) proteins such as Int (integrase), essential for prophage integration and excision, and antiterminator protein Q. Production of Q in turn mediates the transcription of late genes, including portal protein gene B, responsible for translocation of phage DNA into the virion protein capsid, and lysis genes S and R, encoding endolysins that disrupt the bacterial plasma membrane causing release of intact phage progeny (for a review, see Gottesman and Weisberg [71]). Late genes in EHEC phages also encompass stx.
To uncover the roles of specific phage and bacterial functions in EHEC disease, we used lambda red recombination (Materials and Methods) to construct C. rodentium(Φstx2dact) strains defective for prophage genes SR, int, B, or Q, or the host gene recA, which is well documented to be central to the SOS response and lytic induction. In addition, we inactivated three other genes that have been implicated as having more subtle roles in the lytic induction of Shiga toxin-encoding phage [30, 31]: rpoS, which controls the bacterial stress response, and qseC and qseF, which control quorum sensing pathways (Materials and Methods, Table 1). The production of Shiga toxin phage has been shown to be influenced by growth medium [66], but none of the mutants displayed a growth defect upon in vitro culture in LB or DMEM medium (S6 Fig).
We then tested C. rodentium(Φstx2dact) and several of the mutant derivatives predicted to have dramatic effects on phage production for the ability to generate Φstx2dact following SOS induction. Given that Φstx2dact was found to not form plaques on indicator strains tested, we instead utilized qPCR to quantify phage [72–74]. Specifically, we employed primers flanking the phage attP site to distinguish integrated and excised phage DNA, as only the latter will have reconstituted the attP site [71]. This technique detects both unpackaged phage genomes and those packaged in phage capsids, as in our initial experiments lysates were not treated with DNase prior to qPCR enumeration. Note that protease digestion of the capsid prior to qPCR quantitation was also eliminated, as capsid undergoes melting during the high heating steps of the PCR procedure [75]
Supernatants of mid-logarithmic phase (t = 0h) LB cultures contained 1.3×109–3.8×109 attP copies (phage genomes) per ml (Table 3 legend), compared to approximately 108 viable bacteria per ml, indicating significant spontaneous prophage induction during the period leading to mid-log growth. After four additional hours (t = 4h), supernatant phage concentration increased 3.2-fold relative to t = 0h, consistent with continued spontaneous prophage induction (Table 3, “Relative attP production”, “- Mito C”). Prophage induction of the wild type lysogen with the SOS inducer mitomycin C led to a 234-fold increase in relative attP production (Table 3, “+ Mito C”), a 73-fold increase above baseline levels. As predicted [76, 77], the generation of circular phage genomes required Int recombinase, as at all time points tested, attP copies were below the level of detection of 1× 104/ml in uninduced or mitomycin C-induced cultures of the C. rodentium(Φstx2dactΔint) mutant (Table 3).
Host and phage functions contributed to the amount of phage production. In the absence of inducer (Table 3, “- MitoC”), the concentration of attP copies in culture supernatants of C. rodentiumΔrecA(Φstx2dact), predicted to be defective for SOS induction, did not increase between t = 0h and t = 4h, with an average relative attP production of 0.5. Lysogens deficient in the antiterminator Q, required for late gene transcription, or deficient in the S and R endolysins, which promote the efficient release of phage from infected bacteria, were also deficient in relative attP production in the absence of inducer (Table 3). Finally, C. rodentium(Φstx2dactΔB), predicted to replicate but not package phage genomes, showed no defect in the production of attP copies in the culture supernatant in the absence of inducer, with relative phage production ratio of 4.1. However, as described below, DNAse sensitivity assays suggested that these attP sequences are likely not packaged into phage particles.
The mutants defective in baseline phage production were similarly defective in the titer of attP copies after induction with mitomycin C (Table 3, “+ Mito C”). Induction of C. rodentiumΔrecA (Φstx2dact) resulted in an increase in attP production, consistent with low levels of phage production by RecA-deficient λ lysogens of E. coli following induction [78], but the relative attP value of 29 was eight-fold lower than wild type. C. rodentium(Φstx2dactΔQ) and C. rodentium(Φstx2dactΔSR) each also demonstrated dramatically diminished attP copies in mitomycin C-induced culture supernatants, with relative attP production of approximately 6. The small increase in attP levels for each of these mutants upon induction is consistent with readthrough of early transcription of Q-deficient λ mutants [79] and low level bacterial lysis in the absence of phage-encoded endolysins, respectively.
Finally, C. rodentium (Φstx2dactΔB), generated wild type levels of phage genome copies, with a 209-fold increase in relative attP production. However, DNAse treatment of supernatants diminished this value more than 23-fold, whereas parallel treatment diminished the relative attP production by wild type C. rodentium(Φstx2dact) less than 1.5-fold (Table 3, “+Mito C + DNAse”), consistent with a defect in packaging of Φstx2dact genomes in the absence of the B portal protein.
To determine which host or phage functions are required for production of Stx2dact in vitro, we measured Stx2dact in culture supernatants by ELISA [53]. To quantitate non-induced levels of toxin, and to provide ample time for toxin to accumulate, we grew triplicate cultures of the C. rodentium(Φstx2dact) or the mutant derivatives described above for four hours (t = 4h) beyond mid-log phase (defined as t = 0h). Stx2dact was present in the culture supernatants of wild type C. rodentium(Φstx2dact) at approximately 50 ng/ml/OD600 unit, consistent with previous measurements [53] (Fig 2A, “WT”). Prophage excision and phage production were not required for this basal level of Stx2dact: culture supernatants of C. rodentium(Φstx2dactΔint), which did not harbor detectable phage (Table 3), contained equivalent amounts of toxin (Fig 2A, “Δint”). Uninduced culture supertants of C. rodentium(Φstx2dactΔSR) contained levels of Stx2dact two-fold lower than (and statistically indistinguishable from) wild type, consistent with the moderately (5-fold) lower levels of phage found in cultures of wild type C. rodentium (Φstx2dact) (Table 3, “- MitoC”). Supernatants of C. rodentium(Φstx2dactΔB), which contained attP DNA but relatively few packaged phage (Table 3), also produced levels of Stx2dact statistically indistinguishable from wild type. Finally, in contrast, C. rodentiumΔrecA (Φstx2dact), which is unable to mount an SOS response, and C. rodentium(Φstx2dactΔQ), which cannot transcribe phage late genes, including stx2dactA and stx2dactB, were defective for basal levels of Stx2dact production (Fig 2A, “ΔrecA”, “ΔQ”).
To test whether the defect in Stx2dact production was due to the lesion in the Q gene, we complemented C. rodentium(Φstx2dactΔQ) with plasmid pTOPO-Q, the wild type Q gene (Table 1). The complemented strain indeed increased Stx2dact production 286-fold (S7 Fig, “ΔQ + pTOPO-Q”). Nevertheless, this level of Stx2dact was 12-fold lower than that produced by the wild type C. rodentium (Φstx2dact) strain, a defect that is likely due to unregulated Q production in trans [80] because we found that pTOPO-Q similarly diminished Stx production by the WT strain (S7 Fig, “WT + pTOPO-Q”). The exquisite developmental control of gene expression during the lysogenic and lytic cycle is a hallmark of lambdoid phages [23], making complementation of many of phage mutants technically challenging [80].Hence, to minimize the risk that phenotypes observed were due to off-target lesions, we isolated two independent clones of each mutant and tested both clones for each of the phenotypes observed throughout this study.
We also assessed toxin production by wild type C. rodentium(Φstx2dact) and mutant derivatives after 4h of mitomycin C induction. Given that mitomycin C-induced Φstx2dact functions may be involved in the release of toxin from the bacterial host [27], we assessed toxin in cell pellets and in culture supernatants separately. As previously observed [52], mitomycin C induction resulted in a more than 100-fold increase of Stx2dact in culture supernatants (Fig 2B, “WT"). A nearly equivalent amount of toxin remained associated with the bacterial cell pellet, suggesting that under these conditions, a significant fraction of bacteria remained unlysed. Culture supernatants or cell pellets of the C. rodentiumΔrpoS (Φstx2dact) mutant predicted to be defective in the bacterial stress response, or the C. rodentiumΔqseC(Φstx2dact) and C. rodentiumΔqseF(Φstx2dact) mutants defective for quorum sensing, showed wild type levels of Stx2dact (S8 Fig), indicating that neither the bacterial stress response nor the QseC- or QseF-mediated quorum responses were required for toxin production. Culture supernatants of C. rodentium(Φstx2dactΔint) and C. rodentium(Φstx2dactΔB), which showed no defect in basal levels of toxin production (Fig 2A), also contained amounts of cell-associated toxin and supernatant-associated Stx2dact indistinguishable from wild type (Fig 2B,”Δint” and” ΔB”), despite the lack of prophage excision and/or phage production in these mutant strains. The ΔSR lysogen, defective for phage endolytic functions, produced wild type levels of cell-associated Stx2dact at 4 h post-induction, but supernatant-associated toxin was approximately ten-fold lower than wild type levels (Fig 2B, “ΔSR”). This difference is consistent with a defect in bacterial lysis and Stx2dact release, but did not reach statistical significance. In addition, by 16 h post-induction of the ΔSR lysogen, Stx2dact was detected in supernatants at levels similar to that of the WT strain (Fig 2C), suggesting that any defect in R and S proteins results in a delay rather than an absolute block in toxin release. Finally, however, deficiency in the RecA or Q proteins was associated with a near-complete absence of Stx2dact in cell supernatants (Fig 2B,”ΔrecA” and” ΔQ”), reinforcing the notion that these proteins, which are required for the SOS response and/or transcription of the stx2dact genes ([81] [71]), are essential for large amounts of Stx2dact production.
Stx-encoding prophages undergo lytic induction during EHEC infection of germ-free or antibiotic-treated mice [40, 41, 69], and our comprehensive survey of prophage integration sites in fecal microbiota (Table 2) indicated that C. rodentium(Φstx2dact) undergoes some degree of lytic induction during infection of conventional mice. To assess this induction further, we infected conventionally raised C57BL/6 mice with C. rodentium(Φstx2dact) by oral gavage and measured fecal shedding of both the infecting strain, by plating for CFU, and Φstx2dact, by quantitating attP (non-integrated phage) copies by qPCR. As previously observed, by day 3 post-infection, C. rodentium(Φstx2dact) was detected in the stool at 8 x 107 per gram, and reached 9 x 1010 per gram by day 6 post-infection ([82]; Fig 3, “CFU of WT"). Further, murine infection by this strain was indeed associated with lytic induction, as excised phage genomes were detected in stool at all time points (Fig 3, “Phage from WT”).
Interestingly, given the relatively high phage production by induced C. rodentium(Φstx2dact) in vitro, the amount of phage detected in stool was quite low. At day 3 post-infection, 5 x 106 attP copies were detected per gram of stool, a value 16-fold lower than the concentration of viable C. rodentium(Φstx2dact) in stool at that time point. By day 6 post-infection, attP copies had increased to 5 x 107 per gram of feces, but were approximately 600-fold lower than the fecal bacterial counts. These results indicate that C. rodentium(Φstx2dact) thus undergoes lytic induction and growth in this murine model, although not to the degree seen upon induction in vitro.
To test the importance of SOS induction and phage functions on disease in our microbiota-replete model of infection, we infected C57BL/6 mice with C. rodentium(Φstx2dact) and mutant derivatives by oral gavage. The wild type and the mutant lysogens colonized mice similarly, although the ΔB and Δint mutant lysogens appeared to colonize at somewhat higher levels (S9 Fig). C. rodentium ΔrecA(Φstx2dact) and C. rodentium(Φstx2dactΔQ), the two mutant lysogens that displayed dramatic defects in basal and mitomycin C-induced levels of Stx2dact in vitro, were the only ones incapable of causing sickness or death (Fig 4,”ΔrecA” and”ΔQ”), supporting the hypothesis that induction of an SOS response and the subsequent expression of phage late genes, including stx genes, are required for Shiga toxin production during infection of a microbiota-replete host.
The RpoS-deficient and QseC-deficient C. rodentium(Φstx2dact) mutants that are compromised in bacterial stress and quorum-sensing responses, respectively, retained the ability to cause weight loss and lethality with kinetics that were indistinguishable from that of WT C. rodentium(Φstx2dact) (S10 Fig). Thus, although previous results indicated that some quorum sensing mutants display diminished virulence during infection by non-Stx-producing C. rodentium [83], our results are consistent with the the ability of these strains to produce wild type levels of Stx2dact after SOS induction (S8 Fig). In addition, the lack of endolysins that appeared to somewhat delay release of Stx2dact into supernatants by C. rodentium (Φstx2dactΔSR) was not reflected by any delay in the kinetics of weight loss or lethality in infected mice (Fig 3,”ΔSR”), consistent with the ability of this strain to produce wild type levels of Stx2dact upon extended culture in vitro. Thus, C. rodentium (Φstx2dactΔSR) is capable of triggering Stx2dact–mediated disease in the absence of phage-induced lysis.
Finally, the production of intact phage is not essential to disease in this model. C. rodentium (Φstx2dact-ΔB), which is unable to generate intact phage in vitro, and C. rodentium(Φstx2dactΔint), which can neither generate excised phage genomes in vitro or in vivo, both retained full virulence in this model. We conclude that in this microbiota-replete model of EHEC infection, disease progression correlates exclusively with the ability to produce Stx2dact, regardless of the lysogen’s ability to amplify the stx2 genes by phage excision and genome amplification, or by the production of phage that are capable of secondary infection of commensal bacteria.
Commensal organisms have the potential to suppress or enhance phage induction and Stx production. Although a role for induction of stx-encoding prophages in the production of Stx and serious disease during animal infection has been well documented in antibiotic-treated and germ-free mice [40, 41, 69], we used a murine model of EHEC infection that features an intact microbiome.
To investigate phage functions required for C. rodentium(Φstx2dact) to produce Stx and cause disease in conventional mice, we first characterized prophage genetic structure. Φstx2dact prophage was integrated into the C. rodentium dusA gene, an integration site utilized by prophages in over 200 bacterial species [63]. Although the orientation of the regulatory and late genes within the Φstx2dact prophage is noncanonical with respect to attL and attR (with int adjacent to attL; Fig 1), this orientation has been previously observed in at least one other lambdoid phage. In addition, Φstx2dact genes encoding several key phage proteins were identified by homology, and their inactivation had the predicted effects on phage development and production (Table 3; [77]). For example, antiterminator Q and integrase were required for phage production, as measured by detection of attP, and portal protein B was required for packaging of phage DNA into DNAse-resistant virions.
Stx production in vitro by the prophage mutants, as well as by a host recA mutant, confirmed that prophage induction, i.e., the SOS-dependent process required to initiate a temporal program of phage gene expression that normally leads to phage lytic growth, is essential for high-level Stx2 production in vitro. Mitomycin C treatment of C. rodentium(Φstx2dact) resulted in a greater than 100-fold increase in Stx2dact in culture supernatants, similar to the mitomycin C-mediated increase in Shiga toxin production by EHEC ([41]; Fig 2). Three signaling pathways, mediated by RpoS, QseC, and QseF, previously demonstrated to influence SOS induction of EHEC in vitro, had no effect on Stx2dact production by C. rodentium (Φstx2dact). In contrast, and as expected, RecA, required for mounting an SOS response, was necessary for this enhanced production of Stx2dact (Fig 2). It was previously shown that inactivation of the EHEC prophage repressor CI, a key step in the SOS response, is required for the increase in EHEC Stx production upon mitomycin C induction in vitro [41].
Despite the previous observation that the increase in phage genome copy number plays the most quantitatively important role in mitomycin C-enhanced Stx1 production by Stx phage H-19B [27], we found that integrase-deficient C. rodentium(Φstx2dact), which is deficient in phage excision and replication (Table 3; [76]), produced levels of Stx2dact indistinguishable from wild type (Fig 2). Apparently, enhanced expression of late genes stx2dactA and stx2dactB still occurs in the absence of integrase and is sufficient for wild type levels of Stx2dact production. As expected, antiterminator protein Q, required for the transcription of late genes including stx, was essential for Stx2dact production by C. rodentium (Φstx2dact), consistent with previous findings for the Stx2 phage Φ361 [26]. Finally, the S endolysin of Stx phage H-19B was previously shown to promote the timely release of toxin after mitomycin C induction [27]; we found that deficiency of the RS endolysins encoded by Φstx2dact appeared to diminish the release of Stx2dact into culture supernatants at 4 hours post-induction (Fig 2B). However, the decrease was not statistically significant, and RS-deficiency had no discernible effect on toxin release by 16 hours (Fig 2C). Dead and dying E. coli cells are known to release their contents into the surroundings at the end of stationary phase [84]; additionally, E. coli O157:H7 has been shown to release Shiga toxin via outer membrane vesicles [85].
Whereas previous work in streptomycin-treated or gnotobiotic murine models has demonstrated that induction of the lytic developmental program of Stx phage occurs during infection and is required for disease [40, 41, 69], we document here that prophage induction occurs during infection of mice with intact microbiota. attP sequences (indicative of excised, uningegrated phage genomes) were detected in the feces of infected mice, as revealed by deep sequencing (Table 3), or by qPCR (Fig 3).
Deep sequencing of phage genomes in the stool of mice revealed no evidence of Φstx2dact lysogeny of commensal bacteria during C. rodentium(Φstx2dact) murine infection, suggesting that secondary infection of commensals by this phage is rare. Furthermore, C. rodentium(Φstx2dact) mutants deficient in phage integrase or portal protein B, which retained the ability to produce Stx2dact, but were incapable of generating phage or infecting commensal bacteria, caused weight loss and lethality of mice with kinetics indistinguishable from wild type C. rodentium(Φstx2dact) (Fig 4). Indeed, the only C. rodentium(Φstx2dact) derivatives incapable causing disease in animals were those with a demonstrated defect in the production of Stx2dact in vitro (Table 3 and Fig 4). For example, RecA, essential for the initiation of the SOS response that leads to prophage induction, was required for lethality after oral inoculation of C. rodentium (Φstx2dact), consistent with the previous finding that RecA was required for lethality following intravenous EHEC infection of conventional mice [74]. We conclude that amplification of Stx2dact production by successive rounds of lytic infection of commensal bacteria, as has been postulated [38, 44, 45, 47], is not required for toxin-mediated disease in this microbiota-replete model.
We detected more than 1 x 109 phage/ml in uninduced mid-log cultures, suggesting that there is a high level of spontaneous induction under in vitro culture conditions. In contrast, despite severe Stx2dact-mediated disease manifestations during productive infection by C. rodentium(Φstx2dact), the number of attP sequences detected in feces was extremely low, suggesting that the level of prophage induction during infection may also be low. On day 6 post-infection, only 0.44% of all phage genomes detected were excised, compared to 99.66% that were integrated, reflecting intact prophage (Table 3). Depending on the day post-infection, excised phage detected by qPCR numbered 20- to 1000-fold fewer than viable C. rodentium(Φstx2dact) cells (Fig 3). Notably, previous work using a genetic reporter to indicate activation of lytic promoters of EHEC Stx phage 933W showed that the intestinal environment of a gnotobiotic mouse was strongly inducing [41]. While we cannot rule out the possibility that the low number of Φstx2dact attP sequences detected in feces reflects an instability of phage particles or some other factor in the intestinal milieu, our findings are consistent with the possibility that a low rate of Φstx2dact induction may be sufficient to promote disease in this model. Given that the methods to measure phage particles utilized in this study can be applied to patient samples, future studies will focus on the extent of lytic induction of Stx phage during human infection, and how it may correlate with disease outcome.
Mice were purchased from Jackson Laboratories and maintained in the Tufts University animal facility. All procedures were performed in compliance with Tufts University IACUC protocol B2014-87. If examination revealed signs of suffering, manifested by greatly diminished activity, poor grooming/appearance, biting, greatly increased respiratory rate or diminished appetite, or weight loss greater than 15% of body weight, then the animal was euthanized. Primary euthanasia method: CO2 asphyxiation or CO2 followed by cardiac stick. Secondary euthanasia method. Cervical dislocation, decapitation, thoracotomy or major organ removal is performed following the primary method."
Strains and plasmids used in this study are listed in Table 1.
Genomic DNA was isolated from 5 ml of strain C. rodentium(Φstx2dact::kanR) (Table 1) grown overnight at 37°C in LB broth containing chloramphenicol (12.5 μg/ml) and kanamycin (25 μg/ml). DNA was extracted using a DNeasy kit (Qiagen), according to the manufacturer’s protocol for Gram negative bacteria. A library of this DNA was then constructed for Illumina sequencing using Illumina TruSeq DNA Sample Preparation Kit per the manufacturer’s instructions. Following sequencing, the bacterial genome was assembled de novo into 1500 contigs using assemblers ABySS [86], and Edena [87]. The Bowtie2 program [88] was then used to map the stx2 gene against this assembled genome and the contig containing this gene was identified. When aligned to the C. rodentium genome, a 69594-bp contig revealed a 47,343 bp prophage containing the stx2 gene and other phage lambda-like gene sequences inserted into the host dusA gene. (Although the C. rodentium dusA gene is interrupted by the prophage genome, a potentially functional dusA gene is reconstituted at the attL bacterial/phage DNA junction by fusion with a prophage-derived open reading frame that we term “ΦdusA’” in Fig 1.) The prophage sequence was deposited in GenBank as Φ1720a-02, accession number KF030445.1.
Integration of the prophage in both C. rodentium(Φstx2dact) and C. rodentium (ΦΔstx2dact::kanR) into the host dusA gene was verified by PCR amplification of the attL and attR phage-host junctions using primers DusF/PhageR and DusR/PhageF, respectively (Table 4), then DNA sequencing of the amplified junctions. Subsequent whole genome sequencing of C. rodentium(Φstx2dact) and C. rodentium(ΦΔstx2dact::kanR) showed that, except for the Φstx2dact prophage sequences, they are identical to C. rodentium ICC 168, also known as strain DBS100 (GenBank accession number NC_013716.1), and to each other. The encoded Φstx2dact prophage sequences were identical except for the presence of the kanR gene in stxA of strain C. rodentium (ΦΔstx2dact::kanR) (S1 Fig) flanked by the sequence TCCCCGGGTCATTATTCCCT CCAGGTA upstream of the kanR gene and the sequence CTTATTCCTCCTAGTTAGTCACCCGGGA downstream of the kanR gene.
The Φstx2dact genome sequence was first annotated using the program RAST (http://rast.nmpdr.org/ [89]). The annotation was further refined by analyzing each open reading frame using the NCBI program MEGABLAST against the GenBank nucleotide database. Note that although the insertion of the marker into the Rz, gene affects lysis by phage λ lysogens in the presence of high magnesium, this gene has been altered in other studies of Stx phage [66] and in this study, lysis of C. rodentium (Φstx2dact) occurred upon in vitro induction (S4 Fig).
DNA was extracted from fecal samples of 5 infected sick mice at 6 days post-infection, according to the method of Yang et al. [90]. Twenty mg stool samples were suspended in 5 ml PBS, pH7.2, and centrifuged at 100 × g for 15 min at 4°C. The supernatant was centrifuged at 13,000 × g for 10 min at 4°C, and the resulting pellet was washed 3 times in 1.5 ml acetone, centrifuging at 13,000 × g for 10 min at 4°C after each wash step. Two hundred μl of 5% Chelex-100 (Bio-Rad) and 0.2 mg proteinase K were added to the pellet and the sample was incubated for 30 min at 56°C. After vortexing briefly, the sample was centrifuged at 10,000 × g for 5 min and the supernatant containing the DNA was harvested and stored.
To characterize bacteria that harbor the Φstx2dact prophage, we sequenced the bacterial bacterial-host attL prophage junction and adjacent bacterial DNA by following, with slight modifications, the methodology of Klein et al. [70] for constructing high-throughput sequencing libraries that contain a repetitive element (in this case, the phage int (integrase) gene). Briefly, genomic DNA was sheared by sonication to a size of 100–600 bp, followed by addition of ~20 deoxycytidine nucleotides to the 3’ ends of all molecules using Terminal deoxynucleotidyl Transferase. Two rounds of PCR using a poly-C-specific and phage int gene-specific primer pair (PCR primers 1 and 2, Table 4) were used to amplify attL and to add on sequences necessary for high-throughput sequencing (PCR primers 3 and 4, Table 4).
Amplicons were sequenced using the MiSeq desktop sequencer (Ilumina) and primer Seq-P (Table 4), providing reads of up to 300 bp. As amplicons spanned the region from the phage int gene, through attL, and into the adjacent host genome (see Fig 1), reads of this length were required. 17,868,095 sequences encompassing 5 Gb were downloaded to the Galaxy server (https://usegalaxy.org/) and analyzed (Table 3). We first excluded sequences that clearly reflected attL (i.e., contained the 184 bp of ΦdusA’ followed by C. rodentium dusA), indicating the prophage inserted into the C. rodentium genome. Of the remaining 801,959 sequences, 75,962 (0.44% of the total) encoded the intact attP site, implying that they were circular. These latter sequences presumably reflected excised circular phage genomes, possibly undergoing early theta DNA replication, ultimately leading to phage production. The remaining 725,997 sequences encoded only strings of A’s and/or C’s, and were eliminated from consideration.
Deletion mutants of C. rodentium(Φstx2dact) in the prophage or the host genome were generated using a modified version of a one-step PCR-based gene inactivation protocol [61, 82]. Briefly, a PCR product of the zeocin-resistance gene and its promoter region flanked by 70–500 bp homology of the region upstream and downstream of the targeted gene was generated using the primers listed in Table 4. The chromosomal DNA served as template when the flanking regions were 500 bp in length on either side of the zeocin cassette. The PCR product was electroporated into competent C. rodentium(Φstx2dact) cells containing the lambda red plasmid pKD46 and recombinants were selected on plates containing chloramphenicol and zeocin (75 μg/ml). Replacement of the gene of interest with the zeocin resistance cassette was confirmed using specific primers (Table 4). At least two independent clones, validated using PCR, were obtained and subsequently analyzed.
To complement C. rodentium(Φstx2dactΔQ), the only phage mutant with a defect in Stx2dact production, the region encoding the anti-terminator Q was amplified from WT genomic DNA using primers Cr (ΦΔQ) validation F and Q100 R (Table 4), cloned into the pCR4-TOPO vector and transformed into Top 10 cells using the TOPO TA cloning kit (Life Technologies). Kanamycin-resistant colonies were screened for the presence of vector carrying the Q gene (pTOPO-Q). pTOPO-Q was then transformed into electrocompetent wild type C. rodentium(Φstx2dact) or C. rodentium(Φstx2dact ΔQ), using standard cloning techniques.
Overnight 37°C cultures of C. rodentium(Φstx2dact) or deletion derivatives were diluted 1:25 into 10 ml of fresh medium with appropriate antibiotics. Two independently derived clones for each mutant were tested, with indistinguishable results. The cultures were grown at 37°C with aeration to an OD600 of 0.4, and one ml of each culture was set aside (Table 3, “t = 0h”). The remaining culture was split into 2 cultures. These cultures were grown for a further 4 hours (Table 3, “t = 4h”) in the absence or presence of 0.25 μg/ml mitomycin C. (We first measured phage and Stx2 production at various times post-induction and found the 4-hour time point to be optimal for obtaining maximal phage and Stx2 following mitomycin C induction). Supernatants depleted of intact bacteria were then harvested by centrifugation at 17,800 × g for 5 minutes at room temperature. For C. rodentium(Φstx2dact) and C. rodentium (Φstx2dactΔSR), a portion of each culture was also collected after ~16 h of incubation (“t = 16h”). Supernatants and pellets were quantitated for Stx2dact by ELISA, as described previously [52].
Attempts to quantitate phage using plaque titers were unsuccessful. Attempts included the use of various host strains, including C. rodentium non-lysogens, E. coli K12 strains Epi300, LE392, or DH5α, E. coli OP50, and Shigella. Plate modifications included the addition of subinhibitory concentrations of antibiotics, addition of 10 mM CaCl2 or MgCl2 or both, addition of 5% glycerol to bottom agar, addition of tetrazolium to bottom agar, or Sybr staining and fluorescence microscopy of phage. Instead, excised phage genomes in cell supernatants were quantitated by qPCR. Protease digestion of capsids prior to qPCR quantitation was not required, as capsid undergoes melting during the high heating steps of the PCR procedure [75]. Supernatants were serially diluted 1:10, 1:100 and 1:1000 in distilled water. Separate reactions using two μl of the various dilutions as a template were carried out in duplicate. qPCR master-mix (Bio-Rad) was prepared according to the manufacturer’s instructions, using the attP primer set (Table 4) to detect copies of excised phage DNA. Results were compared to a standard curve, derived from a known concentration of a template fragments generated from amplifying C. rodentium(Φstx2dact) DNA using attP primers. The template was serially diluted, in duplicate, to detect copy numbers ranging from 1010 to 102. qPCR reactions were carried out as follows: 95°C for 3 min, followed by 35 cycles of 95°C for 1 min, 58°C for 30 sec, and 72°C for 1 min. Phage genomes in the supernatant of C. rodentium (Φstx2dactΔB), which lacks the portal protein required for genome packaging, was diminished 18-fold by DNAse treatment, supporting our method of qPCR quantitation of phage (see Table 3).
Mice were purchased from Jackson Laboratories and maintained in the Tufts University animal facility. Seven to eight-week-old female C57BL/6J mice were gavaged with PBS or ∼5×108 CFU of overnight culture of C. rodentium(Φstx2dact) or deletion derivatives in 100 μl PBS. Inoculum concentrations were confirmed by serial dilution plating. Fecal shedding was determined by plating dilutions of fecal slurry on either chloramphenicol, to detect wild type C. rodentium(Φstx2dact), or chloramphenicol-zeocin plates, to detect deletion derivatives marked with a zeomycin resistance gene [52]. Body weights were monitored daily, and mice were euthanized upon losing >15% of their body weight.
DNA from infected mouse fecal pellets was isolated using the QIAGEN DNeasy Blood and Tissue kit with modifications. Fecal pellets were incubated with buffer ATL and proteinase K overnight at 55°C. Buffer AL was added, and after mixing, pellets were further incubated at 56°C for 1 h. Pellet mixtures were then centrifuged at 8000 rpm for 1 min and the pellets were discarded. Ethanol was added to the supernatants, which were processed according to the manufacturer’s protocol. DNA concentrations were determined using a NanoDrop spectrophotometer. qPCR was performed as described above.
Data were analyzed using GraphPad Prism software. Comparison of multiple groups were performed using the Kruskal-Wallis test with Dunn's multiple comparison post-test, or 2-way ANOVA with Bonferroni’s post-tests. In all tests, P values below 0.05 were considered statistically significant. Data represent the mean ± SEM in all graphs.
|
10.1371/journal.pgen.1007548 | Binary addition in a living cell based on riboregulation | Synthetic biology aims at (re-)programming living cells like computers to perform new functions for a variety of applications. Initial work rested on transcription factors, but regulatory RNAs have recently gained much attention due to their high programmability. However, functional circuits mainly implemented with regulatory RNAs are quite limited. Here, we report the engineering of a fundamental arithmetic logic unit based on de novo riboregulation to sum two bits of information encoded in molecular concentrations. Our designer circuit robustly performs the intended computation in a living cell encoding the result as fluorescence amplitudes. The whole system exploits post-transcriptional control to switch on tightly silenced genes with small RNAs, together with allosteric transcription factors to sense the molecular signals. This important result demonstrates that regulatory RNAs can be key players in synthetic biology, and it paves the way for engineering more complex RNA-based biocomputers using this designer circuit as a building block.
| In this work, we have engineered a distinctive genetic system, based on regulatory RNAs that control the process of protein translation, that is able to perform arithmetic logic computations (additions) in a single bacterial cell. The system expresses as output fluorescent proteins according to the molecular concentrations of the inputs (binary code). In the future, this circuitry might be instrumental to develop smart bacterial cells that can make appropriate decisions after certain computation for biomedical applications.
| In 1945, von Neumann established the foundations of the logic architectures behind computers in his famous “First draft of a report on the EDVAC” [1]. There, the arithmetic logic unit (ALU) appeared as a principal element in the central processing unit. An ALU is a digital-like circuit that performs arithmetic and logic operations over bits of information. Certainly, today’s computers mount complex ALUs to deal with large volumes of information [2]. But in an emerging scenario of unconventional modes of computation [3], we could wonder whether ALUs, even if in simple forms, are implementable by other means. In particular, is it possible to engineer genetically such a device in a single living cell? Importantly, this question had a positive answer with the engineering of a genetic half adder in mammalian cells [4], by combining transcription factors (TFs) and RNA-binding proteins. A half adder is a basic implementation of an ALU to perform the binary sum of two bits of information. This requires generating two output channels, one for the sum (multiple of 1) and another for the carry (multiple of 2). Later, a genetic half adder was also engineered in bacterial cells exploiting combinatorial transcriptional regulation [5]. However, these designs are centered on regulatory proteins, which are limited in number, especially those with high propensity for composability and orthogonality with the host machinery, and do not allow an easy computational design of de novo sequences. In this regard, and even though ground-breaking work is being accomplished on circuit design automation [6], directed evolution of TFs [7], and de novo protein design [8], alternatives to protein-based regulation are required.
In recent years, RNA has been exploited as an ideal substrate to engineer gene expression programs that robustly run in vivo, thanks to its functional versatility [9, 10] and model-based designability at the nucleotide level [11, 12]. Examples of this suitability are novel mechanisms of gene expression control through the modulation of transcription with non-coding RNAs [13–15], or chimeric RNA molecules integrating different domains that are able to transduce molecular signals [16–18]. Moreover, efforts in RNA synthetic biology to increase the sophistication of the designer systems have led to combinatorial logic gates [19], serial cascades [20, 21], a feed-forward loop [22], and a pulse counter [23]. In this work, we go one step further with the engineering of a genetic half adder in Escherichia coli centered on regulatory RNAs. In particular, we focused on riboregulators of translation initiation [24, 25] to implement our design. The whole system also relies on TF-mediated regulation, especially to sense the molecular signals and express accordingly those riboregulators.
Interestingly, a genetic half adder would allow mounting a common response against two different molecules acting individually (mediated by the sum), and mounting a new response when they act together (mediated by the carry). This would be useful, for instance, in scenarios in which there is synergy between molecules [26].
A half adder receives two input signals and processes them to generate two output responses. In this work, isopropyl β-D-1-thiogalactopyranoside (IPTG) and anhydrotetracycline (aTc) are the two molecules that work as input signals. Moreover, the expressions of a superfolder green fluorescent protein (sfGFP) [27] and a monomeric red fluorescent protein (mRFP1) [28] constitute the output responses. The computation is accomplished in two different genetic modules, both receiving IPTG and aTc as inputs, but each producing one different output. The first genetic module implements a XOR logic gate and generates the sum in the red fluorescence channel. That is, mRFP1 is expressed in presence of IPTG alone or aTc alone. The second genetic module implements an AND logic gate and generates the carry in the green fluorescence channel. That is, sfGFP is expressed in presence of both IPTG and aTc. To implement these logic circuits, we used a synthetic PL-based promoter repressed by LacI, PLlac, and another PL-based promoter repressed by TetR, PLtet [29]. This way, the genes controlled by these two promoters can be induced by IPTG and aTc, respectively, in a strain constitutively expressing the TFs LacI and TetR (here E. coli MG1655-Z1).
We started by engineering the AND logic gate, as this circuit is much simpler than the XOR logic gate. The AND behavior was conceived as the expression, on the one hand, of a cis-repressed messenger RNA (mRNA) coding for a GFP with the PLlac promoter and, on the other hand, of a small RNA (sRNA) able to trans-activate translation with the PLtet promoter (Fig 1A); a scheme already proposed [30, 31]. Cis-repression can be achieved by trapping the ribosome binding site (RBS) in the stem of a strong hairpin formed in the 5’ untranslated region (5’ UTR) of the mRNA, and trans-activation requires a suitable seed region between the sRNA and that hairpin [12]. According to our previous work with the riboregulatory system RAJ11 [31], there is a substantial increase in green fluorescence when both IPTG and aTc are present in the medium, a result obtained again here in new conditions (section A in S1 Appendix). In this case, a GFPmut3b [32] was used as output, following the original system. The no apparent expression in the other induction conditions (readouts even below the fluorescence of cells that do not express GFP) indicated a tight RBS repression. In addition, we considered the riboregulatory system RAJ12 [31] to implement another AND logic gate. We also observed in a fluorometer a substantial increase in green fluorescence only with both inducers, now with sfGFP, but apparently with less dynamic range (Fig 1B). The tight RBS repression was also noticeable in this case. Indeed, previous single-cell analyses of the systems RAJ11 and RAJ12 (by flow cytometry) revealed fluorescence distributions almost coincident with the distribution coming from cells that do not express GFP, even with plasmids of high copy number [31]. Accordingly, we decided to keep the RAJ12-based AND logic gate (implemented in one single plasmid, pRHA12) as one final module, and exploit the riboregulatory system RAJ11 for the engineering of the XOR logic gate.
Our next goal was to engineer an OR logic gate, proposing two trans-activations of translation in parallel [19]. For that, we placed a cis-repressed mRNA coding for the mRFP1 under the control of a constitutive promoter (J23119 [33]), and the RAJ11 sRNA under the control of the PLtet promoter. Subsequently, we designed a minimal version of such sRNA (RAJ11min), also able to trans-activate the translation of that mRNA. This was done to avoid repeated regulatory genes in the circuit, which presumably enhances genetic stability. The RAJ11 and RAJ11min sRNAs produce the same intermolecular base pairs with the corresponding 5’ UTR. The RAJ11min sRNA was then expressed with the PLlac promoter (Fig 2A). We found a significant expression boost either with IPTG or aTc (Fig 2B). The similar expression levels indicated fully functionality of the RAJ11min sRNA. Moreover, we found that the expression levels are almost the double upon induction with both IPTG or aTc. This is expected if we assume that (synthetic) riboregulation, in contrast to transcriptional regulation, rests on decreased binding affinity in vivo (sRNA-mRNA interaction) and then operates in the linear regime, without reaching saturation [34–36]. Afterwards, we decided to replace the promoter that controls mRFP1 expression. In particular, we chose the PR promoter from λ phage [37]. In absence of the TF cI, this promoter is also constitutive in E. coli. We found a similar expression pattern as before, but with less than half expression levels (Fig 2C). This is in tune with previous work on promoter characterization showing that the J23119 promoter is stronger than the PR promoter [33]. Running the EFM calculator, devised for assessing evolutionary failure modes [38], we obtained a RIP score (lower is more stable) of 270.7 for this last OR logic gate (implemented with the RAJ11 and RAJ11min sRNAs and the PR promoter); while it would be 773.0 if this gate were implemented with two copies of RAJ11 or 373.9 with two copies of RAJ11min.
In addition, we conceived the XOR behavior as the combination of such an OR logic gate and an additional AND logic gate. To this end, we expressed, on the one hand, a cis-repressed mRNA coding for cI with the PLlac promoter and, on the other hand, of a sRNA able to trans-activate translation of that gene with the PLtet promoter (Fig 3A). This way, cI is only expressed in the presence of both IPTG and aTc (AND behavior). In turn, cI represses the PR promoter, which expresses mRFP1. To implement this system, we tried two different riboregulatory systems, RR12 [24] and RAJ21 [31], knowing that the apparent dynamic range is much larger for RR12. However, we only found the intended behavior with the system RAJ21, as mRFP1 was not expressed with the system RR12 (Fig 3B and 3C). We argued that cI was relatively expressed with only IPTG or aTc when the system RR12 implements the logic circuit, and that this cI expression was sufficient to repress the PR promoter. As cI is a potent repressor [39] and the circuit was expressed from a high-copy plasmid, any expression leakage, due to inefficient transcriptional or translational control, can end in repression of mRFP1. In terms of translation, previous single-cell analyses (by flow cytometry) revealed a small expression leakage from the cis-repressed mRNA in the case of RR12 [24], but not in the case of RAJ21 [31]. Hence, the RAJ11/RAJ21-based XOR logic gate (implemented in one single plasmid, pRHA40) resulted in the other module.
Finally, we integrated the two modules in a single cell to generate the RNA-based half adder (Fig 4A). That is, E. coli was co-transformed with pRHA12 and pRHA40. Importantly, the riboregulatory systems RAJ11, RAJ12, and RAJ21 were shown computationally, with the NUPACK web application [11], to not suffer cross-talk, i.e., a given sRNA is not able to release the RBS of a non-cognate 5’ UTR. We measured again red (sum) and green (carry) fluorescence with IPTG and aTc, demonstrating the biological computation (Fig 4B and 4C). Nevertheless, we observed that sfGFP was marginally expressed with aTc, perhaps because the transcriptional repression exerted by LacI (less potent than TetR [29]) was slightly abated due to multiple PLlac promoters in the system [40] (section B in S1 Appendix). Further work might try to reduce this leakage to enhance the digital behavior of the system. We quantified the performance of the system as the minimal fold change (f) between the ON and OFF states. We obtained f = 9.4 for mRFP1 (aTc vs. IPTG + aTc) and f = 5.2 for sfGFP (IPTG + aTc vs. aTc). An overall fold change was obtained by averaging geometrically these two values, resulting in f = 7.0. Moreover, we inspected the possibility of getting a visual outcome of the circuit computation. For that, we monitored different cell cultures induced with IPTG and aTc with a microscope, showing that the two bits of processed information, corresponding to the sum and the carry, can be easily recognized (Fig 4D and 4E).
To study whether each E. coli cell was able to perform the computation (i.e., respond to the inducers in a relatively homogeneous manner), we further characterized the functionality of our genetic half adder at the single cell level by flow cytometry. Certainly, cell-to-cell variability in gene expression within a clonal population (noise) is an inherent feature of biology [41]. This assay revealed that the whole population significantly shifted its fluorescence in both channels according to the induction condition (Mann-Whitney U-tests, P ≈ 0; Fig 5). Again, we quantified f = 16.3 for mRFP1 (now the minimal fold change was in IPTG vs. none) and f = 4.9 for sfGFP (IPTG + aTc vs. aTc) using mean values of fluorescence. The overall fold change was in this case f = 8.9. These values are in tune with those reported at the population level. The single cell data also revealed that the slight increase in GFP with only aTc was associated with an increase in cell-to-cell variability regarding sfGFP expression (3.3 times more deviation with aTc than with IPTG). Definitely, more theoretical work is needed to recognize how noise performs in systems of increasing complexity based on intricate transcriptional and post-transcriptional regulation [42].
We have programmed a bacterial cell so that it can perform the binary sum of two bits of information, encoded into the concentrations of IPTG and aTc (signal molecules). The bacterial cell reports the sum of this computation as red fluorescence and the carry as green fluorescence, a sort of minimal biocomputer. To achieve this dynamic behavior, we engineered a genetic system exploiting riboregulation [24]. The whole system consists of four synthetic riboregulators (RAJ11, RAJ11min, RAJ21, and RAJ12), three TFs (LacI, TetR, and cI), and two fluorescent proteins (mRFP1 and sfGFP), which work together within the cell in an articulate manner. Such a system did not require fine-tuning promoters or RBSs to perform as designed, in contrast to what might happen in other cases [43, 44]. Moreover, one important advantage of our designer circuit over the two previous genetic half adders [4, 5] is that the reporter gene conveying the sum is not duplicated. This makes the architecture to be better organized and more scalable, as already pointed out [1].
In addition, the genetic footprint of our designer circuit was greatly reduced thanks to the use of regulatory RNAs, with respect to circuits fully implemented with TFs [6]. The RAJ11, RAJ12, and RAJ21 sRNAs are of 55–71 nucleotides (excluding the terminators), and the RAJ11min sRNA is even of 30 nucleotides. Certainly, the DNA sequence required to encode a protein of average size is much longer. The cis-regulating regions at the DNA or RNA levels, by contrast, are of similar size. The PLlac and PLtet promoters are of 54 nucleotides and the 5’ UTRs involved in riboregulation of 52 nucleotides. Beyond this, by only mutating the seed region between the sRNA and the 5’ UTR it is possible to create riboregulatory systems that perform orthogonally in vivo [45]. This way, we might easily scale up our designer circuit. Following this strategy, of note, we already created a RAJ11-derived orthogonal system [36].
Definitely, we chose a given molecular implementation, but other implementations might be possible maintaining the same regulatory architecture. As the system does not rely on combinatorial promoters, nothing prevents the use of other input signals (e.g., endogenous substances of the cell) to perform the computation replacing the PLlac and PLtet promoters by suitable responsive promoters [43, 44]. Alternatively, LacI and TetR might be computationally redesigned to sense new compounds [46]. The riboregulatory mode, here characterizing an internal layer of gene expression activation, is also flexible. Cis-repression of translation might occur by trapping the start codon, instead of the RBS, in the 5’ UTR structure [25]. More distinctly, the activation might be transcriptional with sRNAs that act in trans as anti-terminators [14]. In addition to LacI and TetR (working in the sensory layer), our system also involves the TF cI to implement an internal repression in the XOR logic gate. We tried to implement this repression by antisense RNA [47] or CRISPR interference [15], without successful results (section C in S1 Appendix); arguably, because the expression of mRFP1 was from a high-copy plasmid. This reveals the necessity of pursuing the development of novel RNA-based mechanisms and circuits. All in all, our genetic implementation of an ALU promises to be important in the future to develop smart cells (e.g., diagnostic bacteria for clinical use) that can make appropriate decisions after certain processing (computation) of the signals perceived from the medium [48].
Synthetic PL-based promoters regulated by the TFs LacI and TetR [29] were used as elements to sense the input signals (IPTG and aTc). Riboregulatory sequences (sRNAs and 5’ UTRs) of systems RAJ11, RAJ12, and RAJ21 were obtained from previous work [31], as well as the sequences of system RR12 [24]. A minimal version of the sRNA RAJ11 was designed by removing the nucleotides not contributing to the intermolecular interaction. The structural models of these systems are shown in S2 Appendix. The PR promoter and a codon-optimized version of the TF cI from λ phage [37] were also used.
Six plasmids were characterized in this work: pRAJ11, pRHA12, pRHA25, pRHA36, pRHA37, and pRHA40. First, pRAJ11 (ampR, pUC ori) and pRAJ12 (kanR, pSC101m ori) were taken from previous work [31]. pRHA12 was constructed by removing the mRFP1 gene from pRAJ12. pRAJ11 expresses in a controlled way GFPmut3b and pRHA12 sfGFP. Moreover, pRHA25 (ampR, pUC ori) was synthesized by IDT. This expresses in a controlled way mRFP1. pRHA36 was constructed by inserting in pRHA25 an expression cassette of cI regulated by ribosystem RR12 (synthesized by IDT), also changing the J23119 promoter by the PR promoter. pRHA37 was constructed by removing the expression cassette of cI from pRHA36. Finally, pRHA40 was constructed by inserting in pRHA37 an expression cassette of cI regulated by ribosystem RAJ21 (synthesized by IDT). See sequences in S3 Appendix.
For cloning purposes, E. coli Dh5α was used following standard procedures [49]. To express the circuits, E. coli MG1655-Z1 (F-, λ-, rph-1, lacIq, PN25:tetR, SpR) was used (i.e., a strain that is lacI+ and tetR+). LB medium was used for overnight cultures, while M9 minimal medium (1x M9 salts, 2 mM MgSO4, 0.1 mM CaCl2, 0.4% glucose, 0.05% casamino acids, and 0.05% thiamine) for characterization cultures. IPTG was used at the concentration of 1 mM and aTc at 100 ng/mL. Ampicillin and kanamycin were used as antibiotics at the concentration of 50 μg/mL. Compounds provided by Sigma-Aldrich.
Cultures (2 mL) inoculated from single colonies (three replicates) were grown overnight in LB medium at 37°C and 200 rpm. Cultures were then diluted 1:200 (1:100 in the case of cells expressing pRHA40) in M9 minimal medium (2 mL) with appropriate inducers (IPTG, aTc) and were grown for 5–8 h, depending on the genetic system and induction condition, at 37°C and 200 rpm to reach an OD600 around 0.5. Cultures were then used to load the wells (200 μL) of the microplate (96 wells, black, clear bottom; Corning). This was assayed in a fluorometer (Perkin Elmer Victor X5) to measure absorbance (600 nm absorbance filter), green fluorescence (485/14 nm excitation filter, 535/25 nm emission filter), and red fluorescence (570/8 nm excitation filter, 610/10 nm emission filter). Mean background values of absorbance and fluorescence, corresponding to M9 minimal medium, were subtracted to correct the readouts. Normalized fluorescence was calculated as the ratio of fluorescence and absorbance. The mean value of normalized fluorescence corresponding to cells transformed with control plasmids was then subtracted to obtain a final estimate of expression.
A culture (2 mL) inoculated from a single colony was grown overnight in LB medium at 37°C and 200 rpm. The culture was then diluted 1:100 in M9 minimal medium (2 mL) and was grown for 5 h at 37°C and 220 rpm to reach exponential phase. The culture was then diluted 1:40 in M9 minimal medium (2 mL) with appropriate inducers (IPTG, aTc) and was grown for 8 h at 37°C and 220 rpm to reach an OD600 around 0.7. 200 μL of each culture were transferred to small tubes. Culture images were acquired with a light microscope (Leica DFC7000T) with the fluorescence filters for GFP and DsRed. Exposition parameters were manually adjusted to enhance the quality of the image.
A culture (2 mL) inoculated from a single colony was grown overnight in LB medium at 37°C and 200 rpm. The culture was then diluted 1:100 in M9 minimal medium (2 mL) and was grown for 5 h at 37°C and 200 rpm to reach exponential phase. The culture was then diluted 1:50 in M9 minimal medium (200 μL) and placed in a microplate with appropriate inducers (IPTG, aTc) and was grown for 5 h at 37°C and 1,000 rpm in a plate shaker (Biosan PST-60HL). Cultures were spun down at 13,000 rpm for 2 min and resuspended in PBS (2 mL). Fluorescence was measured with a flow cytometer (BD LSRFortessa, lasers of 488 nm and 561 nm) with the emission filters for GFP (530/30 nm) and DsRed (585/15 nm). Events were then gated and compensated (~15,000 after this process). The mean value of the autofluorescence of the cells was subtracted in each channel to obtain a final estimate of expression.
|
10.1371/journal.pgen.1006106 | Dynamics of Chloroplast Translation during Chloroplast Differentiation in Maize | Chloroplast genomes in land plants contain approximately 100 genes, the majority of which reside in polycistronic transcription units derived from cyanobacterial operons. The expression of chloroplast genes is integrated into developmental programs underlying the differentiation of photosynthetic cells from non-photosynthetic progenitors. In C4 plants, the partitioning of photosynthesis between two cell types, bundle sheath and mesophyll, adds an additional layer of complexity. We used ribosome profiling and RNA-seq to generate a comprehensive description of chloroplast gene expression at four stages of chloroplast differentiation, as displayed along the maize seedling leaf blade. The rate of protein output of most genes increases early in development and declines once the photosynthetic apparatus is mature. The developmental dynamics of protein output fall into several patterns. Programmed changes in mRNA abundance make a strong contribution to the developmental shifts in protein output, but output is further adjusted by changes in translational efficiency. RNAs with prioritized translation early in development are largely involved in chloroplast gene expression, whereas those with prioritized translation in photosynthetic tissues are generally involved in photosynthesis. Differential gene expression in bundle sheath and mesophyll chloroplasts results primarily from differences in mRNA abundance, but differences in translational efficiency amplify mRNA-level effects in some instances. In most cases, rates of protein output approximate steady-state protein stoichiometries, implying a limited role for proteolysis in eliminating unassembled or damaged proteins under non-stress conditions. Tuned protein output results from gene-specific trade-offs between translational efficiency and mRNA abundance, both of which span a large dynamic range. Analysis of ribosome footprints at sites of RNA editing showed that the chloroplast translation machinery does not generally discriminate between edited and unedited RNAs. However, editing of ACG to AUG at the rpl2 start codon is essential for translation initiation, demonstrating that ACG does not serve as a start codon in maize chloroplasts.
| Chloroplasts are subcellular organelles in plants and algae that carry out the core reactions of photosynthesis. Chloroplasts originated as cyanobacterial endosymbionts. Subsequent coevolution with their eukaryotic host resulted in a massive transfer of genes to the nuclear genome, the acquisition of new gene expression mechanisms, and the integration of chloroplast functions into host programs. Chloroplasts in multicellular plants develop from non-photosynthetic proplastids, a process that involves a prodigious increase in the expression of chloroplast genes encoding components of the photosynthetic apparatus. We used RNA sequencing and ribosome profiling to generate a comprehensive description of the dynamics of chloroplast gene expression during the transformation of proplastids into the distinct chloroplast types found in bundle sheath and mesophyll cells in maize. Genes encoding proteins that make up the chloroplast gene expression machinery peak in protein output earlier in development than do those encoding proteins that function in photosynthesis. Programmed changes in translational efficiencies superimpose on changes in mRNA abundance to shift the balance of protein output as chloroplast development proceeds. We also mined the data to gain insight into general features of chloroplast gene expression, such as relative translational efficiencies, the impact of RNA editing on translation, and the identification of rate limiting steps in gene expression. The findings clarify the parameters that dictate the abundance of chloroplast gene products and revealed unanticipated phenomena to be addressed in future studies.
| The evolution of chloroplasts from a cyanobacterial endosymbiont was accompanied by a massive transfer of bacterial genes to the nuclear genome, and by the integration of chloroplast processes into the host’s developmental and physiological programs [1]. In multicellular plants, chloroplasts differentiate from non-photosynthetic proplastids in concert with the differentiation of meristematic cells into photosynthetic leaf cells. This transformation is accompanied by a prodigious increase in the abundance of the proteins that make up the photosynthetic apparatus, which contribute more than half of the protein mass in photosynthetic leaf tissue [2]. Both nuclear and chloroplast genes contribute subunits to the multisubunit complexes that participate in photosynthesis. The expression of these two physically separated gene sets is coordinated by nucleus-encoded proteins that control chloroplast gene expression, and by signals emanating from chloroplasts that influence nuclear gene expression [1, 3]. Beyond these general concepts, however, little is known about the mechanisms that coordinate chloroplast and nuclear gene expression in the context of the proplastid to chloroplast transition. Furthermore, a thorough description of the dynamics of chloroplast gene expression during this process is currently lacking.
Despite roughly one billion years of evolution, the bacterial ancestry of the chloroplast genome is readily apparent in its gene organization and gene expression mechanisms. Most chloroplast genes in land plants are grouped into polycistronic transcription units [4] that are transcribed by a bacterial-type RNA polymerase [5] and translated by 70S ribosomes that strongly resemble bacterial ribosomes [6]. As in bacteria, chloroplast ribosomes bind mRNA at ribosome binding sites near start codons, sometimes with the assistance of a Shine-Dalgarno element [6]. Superimposed on this ancient scaffold are numerous features that arose post-endosymbiosis [7]. For example, a phage-type RNA polymerase collaborates with an RNA polymerase of cyanobacterial origin [5], and chloroplast RNAs are modified by RNA editing, RNA splicing, and other events that are either unusual or absent in bacteria [8].
Ribosome profiling data from E. coli revealed that the rate of protein output from genes encoding subunits of multisubunit complexes is proportional to subunit stoichiometry, and that proportional synthesis is typically achieved by differences in the translational efficiency of genes residing in the same operon [9, 10]. As the majority of chloroplast gene products are components of multisubunit complexes, it is of interest to know whether similar themes apply. Furthermore, the gene content of polycistronic transcription units in chloroplasts has diverged from that in the cyanobacterial ancestor. Has “tuned” protein output been maintained in chloroplasts despite this disrupted operon organization? If so, what mechanisms achieve this tuning in light of the new gene arrangements and the new features of mRNA metabolism?
In this work, we used ribosome profiling to address these and other questions of chloroplast gene regulation in the context of the proplastid to chloroplast transition. For this purpose, we took advantage of the natural developmental gradient of the maize seedling leaf blade, where cells and plastids at increasing stages of photosynthetic differentiation form a developmental gradient from base to tip [11]. By using the normalized abundance of ribosome footprints as a proxy for rates of protein synthesis, we show that the rate of protein output from many chloroplast genes is tuned to protein stoichiometry, and that tuned protein output is achieved through gene-specific balancing of mRNA abundance with translational efficiency. This comprehensive analysis revealed developmentally programmed changes in translational efficiencies, which superimpose on programmed changes in mRNA abundance to shift the balance of protein output as chloroplast development proceeds.
We analyzed tissues from the same genetic background and developmental stage as used in previous proteome [2] and nuclear transcriptome [12, 13] studies of photosynthetic differentiation in maize. Four leaf sections were harvested from the third leaf to emerge in 9-day old seedlings (Fig 1A): the leaf base (segment 1), which harbors non-photosynthetic proplastids; 3–4 cm above the base (segment 4), representing the sink-source transition and a region of active chloroplast biogenesis; 8–9 cm above the base (segment 9), representing young chloroplasts; and a section near the tip (segment 14) harboring mature bundle sheath and mesophyll chloroplasts [2, 12]. The developmental transitions represented by these fractions are illustrated in the immunoblot assays shown in Fig 1B. The mitochondrial protein Atp6 is most abundant in the two basal sections, subunits of photosynthetic complexes (AtpB, PetD, PsaD, PsbA, NdhH, RbcL) are most abundant in the two apical sections, and a chloroplast ribosomal protein (Rpl2) exhibits peak abundance in the two middle sections. These developmental profiles are consistent with prior proteome data [2].
To explore the contribution of differential chloroplast gene expression to the distinct proteomes in bundle sheath and mesophyll cells, we also analyzed bundle sheath and mesophyll-enriched fractions from the apical region of seedling leaves. Standard protocols for the separation of bundle sheath and mesophyll cells involve lengthy incubations that are likely to cause changes in ribosome position. We used a rapid mechanical fractionation method that minimizes the time between tissue disruption and the generation of ribosome footprints (see Materials and Methods). Markers for each cell type were enriched 5- to 10-fold in the corresponding fraction (Fig 1C). This degree of enrichment is comparable to that of the fractions used to define mesophyll and bundle sheath-enriched proteomes in maize [14].
We modified our previous method for preparing ribosome footprints from maize leaf tissue [15] to reduce the amount of time and tissue required, and to reduce contamination by non-ribosomal ribonucleoprotein particles (RNPs). In brief, leaf tissue was flash frozen and ground in liquid N2, thawed in a standard polysome extraction buffer, and treated with Ribonuclease I to liberate monosomes. Ribosomes were purified by pelleting through a sucrose cushion under conditions that leave chloroplast group II intron RNPs (~600 kDa) [16] in the supernatant (S1A Fig). RNAs between approximately 20 and 35 nucleotides (nt) were gel purified and converted to a sequencing library with a commercial small RNA library kit that has minimal ligation bias [17]. rRNA contaminants were depleted after first strand cDNA synthesis by hybridization to biotinylated oligonucleotides designed to match abundant contaminants detected in pilot experiments (S1 Table). Approximately 35 million reads were obtained for each “Ribo-seq” replicate, roughly 50% of which aligned to mRNA (S2 Table). RNA-seq data was generated from RNA extracted from aliquots of each lysate taken prior to addition of RNAse I. Replicate RNA-seq and Ribo-seq assays showed high reproducibility (Pearson correlation of >0.98, S2 Fig). Almost all plastid genes were represented by at least 100 reads per replicate in all datasets (S3 Fig). Several clusters of low abundance reads mapped to small unannotated ORFs, but further investigation is required to evaluate which, if any, of these are the footprints of translating ribosomes.
Ribosomes in the cytosol, mitochondria, and chloroplasts have distinct genetic origins. Accordingly, the ribosome footprints from each compartment displayed different size distributions (Fig 2A). The cytosolic ribosome footprints showed a minor peak at 23 nucleotides and a major peak at 31 nucleotides, similar to observations in yeast [18]. The mitochondrial data showed a major peak at 28–29 nucleotides and a minor peak at 36 nucleotides, similar to the 27 and 33-nt peaks reported for human mitochondria [19]. The plastid ribosome footprints had a broad size distribution suggestive of two populations, with peaks at approximately 30 and 35 nucleotides. A similar distribution was observed in pilot experiments involving the gel purification of RNAs up to 40-nt (S1B Fig) indicating that the peak at 35-nt was not an artifact of our gel purification strategy. A broad and bimodal size distribution was also observed for chloroplast ribosome footprints from the single-celled alga Chlamydomonas reinhardtii, albeit with peaks at slightly different positions [20]. The two prior reports of ribosome footprint size distributions in plants [21, 22] did not parse the data from the three compartments, but the 31-nucleotide modal size reported in those studies is consistent with our data. Our data show the 3-nucleotide periodicity expected for ribosome footprints (Fig 2B and 2C). Interestingly, the degree of periodicity varies with footprint size (S4 Fig). The reads are largely restricted to open reading frames in the cytosol (Fig 2C) and chloroplast (Fig 2D). Taken together, these results provide strong evidence that the vast majority of the Ribo-seq reads come from bona-fide ribosome footprints.
The placement of ribosome P and A sites with respect to ribosome footprint termini has not been reported for any organellar ribosomes or for cytosolic ribosomes in maize. A meta analysis of our data showed that the position of the 3’ end of ribosome footprints from initiating and terminating ribosomes in chloroplasts and mitochondria is constant with respect to start and stop codons, respectively, regardless of footprint size; however, the position of the 5’ ends varies with footprint size (Fig 2E, S4C Fig). Therefore, the positions of the A and P sites in organellar ribosomes can be inferred based on the 3’-ends of their footprints, as is also true for bacterial ribosomes [23, 24]. The modal distance between the start of the P site in chloroplast ribosomes and the 3’-ends of chloroplast ribosome footprints is 7 nucleotides. By contrast, cytosolic ribosome footprints are approximately centered on the P site regardless of footprint size (S4B Fig).
The partitioning of ribosome footprints among the three genetic compartments shifts dramatically during the course of leaf development (Fig 2G). The contribution of cytosolic translation drops from 99% at the leaf base to 57% in the apical leaf sections due to the increasing contribution of ribosome footprints from chloroplasts. This shift of cellular resources towards chloroplast translation corresponds with the massive increase in the content of photosynthetic complexes harboring plastid-encoded subunits (Rubisco, PSII, PSI, cytochrome b6f, ATP synthase, NDH) (Fig 1). Ribosome footprints from mitochondria accounted for a very small fraction of the total at all stages. However, our protocol was not optimized for the quantitative recovery of mitochondrial ribosomes so these data may not reflect the total mitochondrial ribosome population.
In the discussion below we define the “translational output” of a gene as the abundance of ribosome footprints per kb per million reads mapped to nuclear coding sequences (RPKM), and we use this value to compare rates of protein synthesis among genes on a molar basis. This is a typical interpretation of Ribo-seq data, and it is based on evidence that the bulk rate of translation elongation on all ORFs is similar under any particular condition, despite the fact that ribosome pausing can lead to the over-representation of ribosomes at specific positions [9, 25]. Although this may be an over simplification in some instances, this interpretation of our data produced results that are generally coherent with current understanding of chloroplast biogenesis (see below). Group II introns interrupt eight protein-coding genes in maize chloroplasts. These present a challenge for data analysis because the unspliced transcripts make up a substantial fraction of the RNA pool [16] and translation can initiate on unspliced RNAs and terminate within introns [15]. We therefore calculated translational output based solely on the last exon (normalized to exon length). Data summaries presented below include RNA-seq data only for that subset of intron-containing genes for which multiple methods of analysis provided consistent values for the abundance of spliced RNA isoforms (see Materials and Methods).
Fig 3 summarizes the abundance of Ribo-seq and RNA-seq reads from protein-coding chloroplast genes in each of the four leaf segments. To display the low values from Segment 1, they are replotted with a smaller Y-axis scale in S5 Fig. The abundance of mRNA from genes in the same transcription unit (Fig 3A and S5A Fig, bracketed arrows) is typically similar, but the protein output of co-transcribed genes varies considerably. Translational efficiency (translational output /mRNA abundance) varies widely among genes (Fig 3A and S5A Fig, bottom). The atpH mRNA is the most efficiently translated of any chloroplast mRNA at all four developmental stages, surpassing even psbA, whose product is the most rapidly synthesized protein in photosynthetic tissues [26]. Prodigious psbA expression results from very high mRNA abundance in combination with a translational efficiency that is comparable to that of other photosystem genes.
When the data are grouped according to gene function, correlations between function and translational output become apparent (Fig 3B). For example, the translational output of genes encoding subunits of ribosomes and the NDH complex are consistently very low, whereas the translational output of genes encoding subunits of PSI, PSII, the ATP synthase, and the cytochrome b6f complex are consistently much higher. These trends mirror the abundance of these complexes as inferred from proteome data [27]. The data for complexes whose subunits are not found in a 1:1 ratio show further that translational output is tuned to subunit stoichiometry. For example, the chloroplast-encoded subunits of the ATP synthase (AtpA, AtpB, AtpE, AtpF, AtpH, AtpI) are found in a 3: 3: 1: 1: 14: 1 molar ratio in the complex [28, 29]. The translational output of their genes mirrors this stoichiometry quite well, whereas mRNA abundance does not (Fig 4A). These genes are distributed between two transcription units (Fig 4A). A single mRNA encodes AtpB and AtpE, whose rates of synthesis are tuned via differences in translational efficiency. The atpI-atpH-atpF-atpA primary transcript is processed to yield various smaller isoforms [30] but the abundance of RNA from each gene is nonetheless quite similar (Fig 4A). The translational output of the atpH gene is boosted relative to that of its neighbors primarily through exceptionally high translational efficiency (Fig 4A bottom). In a second example, the unequal stoichiometry of subunits of the plastid-encoded RNA polymerase (PEP) (2 RpoA:1 RpoB:1 RpoC1:1 RpoC2) [5] is mirrored by the relative translational output of the corresponding genes (Fig 4B). In this case, however, tuning occurs primarily at the level of mRNA accumulation.
The plastid-encoded subunits of PSI, PSII, the cytochrome b6f complex, the NDH complex, and chloroplast ribosomes are found in equal numbers in their respective complex. Genes encoding subunits of each of these complexes are distributed across multiple transcription units, many of which also encode subunits of other complexes. This gene organization sometimes results in considerable disparity in mRNA level among subunits of the same complex (Fig 3B bottom). In general, such differences are buffered by opposing changes in translational efficiency, such that translational outputs more closely reflect protein stoichiometry than does mRNA abundance (see, for example, the NDH complex in S6B Fig). In the case of PSI (Fig 4C), the structural genes (psaA, psaB, psaC, psaJ, psaI) exhibit an approximately three-fold range of translational output, but all of these genes vastly out produce two genes encoding PSI assembly factors (ycf3 and ycf4) [31–33]. The psaI and ycf4 genes are adjacent in the same polycistronic transcription unit (Fig 4C bottom), and their difference in translational output is programmed primarily by a difference in translational efficiency. The translational output of psbN, which encodes a PSII assembly factor [34], is likewise much less than that of structural genes for PSII (Fig 4D). Taken together, this body of data shows that the tuning of translational output to protein stoichiometries is accomplished via trade-offs between mRNA level and translational efficiency, with this balance differing from one gene to the next. Where mRNA abundance closely matches protein stoichiometry, differences in translational efficiency make only a small contribution (as observed for rpoA, rpoB, rpoC1 and rpoC2). Where mRNAs are severely out of balance with protein stoichiometry, differences in translational efficiency compensate.
The translational output of PSII structural genes is well matched, with the notable exception of psbA (Fig 4D), whose output vastly exceeds that of other genes in photosynthetic leaf segments (segments 9 and 14). This behavior is consistent with the known properties of the psbA gene product, whose damage and rapid turnover during active photosynthesis is compensated by a high rate of synthesis to support PSII repair [26]. Setting psbA aside, the relative translational outputs of other genes only approximate the stoichiometries of their products: several-fold differences between relative output and stoichiometry are common among subunits of a particular complex, suggesting that proteolysis of unassembled subunits serves to fine-tune protein stoichiometries. It is also possible that the calculated translational outputs do not perfectly reflect rates of protein synthesis due to differences in translation elongation rates among mRNAs. That said, instances in which translational outputs are particularly discordant among subunits of the same complex are worthy of note, as this may reflect physiologically relevant behaviors. For example, the translational output of ndhK is balanced with other ndh genes in non-photosynthetic leaf segments but ndhK substantially out produces the other ndh genes in mature chloroplasts (S6B Fig). This behavior is reminiscent of psbA, and suggests that NdhK may be damaged and replaced during active photosynthesis.
To explore the dynamics of chloroplast gene expression during the proplastid to chloroplast transition, we calculated standardized values for translational output, mRNA abundance and translational efficiency such that developmental shifts can be compared despite large differences in signal magnitude. This analysis shows that the developmental dynamics of translational output varies widely among genes (Fig 5A top). The standardized values were used as the input for hierarchical clustering, which produced four clusters from the translational output data, four from the mRNA data, and five from the translational efficiency data (Fig 5B, S7 Fig). The genes in each cluster are identified by color in Fig 5A. Although the transitions between clusters are not marked by obvious distinctions, the distinct trends defining each cluster are clear in the plots in Fig 5B. Genes whose translational output and mRNA abundance peak early in development (segment 4) generally encode components of the chloroplast gene expression machinery (rpl, rps, rpo, matK) (Fig 5A and 5C). Most genes encoding components of the photosynthetic apparatus (psb, psa, atp, pet genes) have peak mRNA and translational output in young chloroplasts (segment 9). A handful of photosynthesis genes either maintain or increase translational output and mRNA in mature chloroplasts (segment 14) (Fig 5A and 5C).
There is considerable similarity among the clusters produced from the translational output and mRNA data (Fig 5A and 5B), implying that programmed changes in mRNA abundance underlie the majority of developmental shifts in translational output. However, changes in translational efficiency also influence the developmental shifts in translational output (Fig 5A bottom). In general, ORFs encoding proteins involved in photosynthesis are more efficiently translated later in development and those encoding gene expression factors are more efficiently translated early in development, albeit with numerous exceptions (Fig 5A bottom, 5C right). Transcription units that encode both photosynthesis and gene expression factors provide revealing examples of distinct translational dynamics. In the psaA-psaB-rps14 transcription unit, for example, rps14 is found in a translational output cluster with other genes involved in gene expression, whereas psaA and psaB reside in a translational output cluster with other photosynthesis genes (Fig 5A top). This results from distinct developmental shifts in translational efficiency: the rps14 ORF is translated more efficiently early in development whereas psaA and psaB are more efficiently translated later in development (Fig 5D). The psaI-ycf4-cemA-petA transcription unit provides a second example. The translational output of psaI, cemA, and petA show similar developmental dynamics, but ycf4 clusters with different genes due to more efficient translation earlier in development (Fig 5D). Again, these distinct patterns correlate with function, as psaI and petA encode components of the photosynthetic apparatus, whereas ycf4 encodes an assembly factor for PSI [31, 32].
Many polycistronic RNAs in chloroplasts are processed to smaller isoforms. Although the impact of processing on translational efficiencies remains unclear [35, 36], it is plausible that programmed changes in the accumulation of processed isoforms could uncouple the expression of cotranscribed genes during development. To address this possibility, we used RNA gel blot hybridization to analyze transcripts from two transcription units that include genes whose translational efficiencies exhibit distinct developmental dynamics: psaI-ycf4-cemA-petA and psaA-psaB-rps14 transcription units (S8 Fig). Processed rps14-specific transcripts accumulate preferentially in immature chloroplasts (segment 4), correlating with the stage at which rps14 is most efficiently translated. Analogously, a monocistronic psaI isoform accumulates preferentially in segments 4 and 9 where psaI is most efficiently translated. Various cause and effect relationships may underlie these correlations, as is discussed below.
In maize and other C4 plants, photosynthesis is partitioned between mesophyll (M) and bundle sheath (BS) cells. Three protein complexes that include plastid-encoded subunits accumulate differentially in the two cell types: Rubisco and the NDH complex are enriched in BS cells whereas PSII is enriched in M cells [2, 14]. Differential accumulation of several chloroplast mRNAs in the two cell types has been reported [37–41], but a comprehensive comparison of chloroplast gene expression in BS and M cells has been lacking. To address this issue we performed RNA-seq and Ribo-seq analyses of BS- and M- enriched leaf fractions. The translational output of genes encoding subunits of Rubisco, PSII, and the NDH complex (Fig 6A) correlated well with the relative abundance of subunits of these complexes in the same sample preparations (Fig 1C), and with quantitative proteome data [2]. Cell-type specific differences in mRNA accumulation (Fig 6B) can account for many of the differences in translational output (Fig 6A), indicating that differences in transcription and/or RNA stability make a strong contribution to preferential gene expression in one cell type or the other. However, the data suggest that differences in translational efficiency contribute in certain instances (Fig 6C). Four genes encoding PSII core subunits (psbA, psbB, psbC, psbD) provide the most compelling examples, as their translational output is considerably more biased toward M cells than are their mRNA levels.
Organellar RNAs in land plants are often modified by an editing process that converts specific cytidine residues to uridine [42, 43]. Some sites are inefficiently edited, which raises the question of whether the translation machinery discriminates between edited and unedited RNAs. The protein products of several unedited mitochondrial RNAs have been detected in plants [44, 45]. We used our Ribo-seq and RNA-seq data to examine this issue for chloroplast RNAs. Fig 7 summarizes the data for those sites of editing that are represented by at least 100 reads in both the Ribo-seq and RNA-seq data in at least two replicates (17 of the 28 edited sites in the maize chloroplast transcriptome). In general, the percent editing was similar in the RNA-seq and Ribo-seq data, implying little discrimination between edited and unedited RNAs by the translation machinery. There were, however, two major exceptions: rpl2 (nt 2) and ndhA (nt 563). In these cases a large fraction of the RNA-seq reads came from unedited RNA, whereas virtually all of the Ribo-seq reads came from edited sites. These two sites have unusual features that can account for the preferential translation of the edited RNAs. Editing at the ndhA site is linked to the splicing of the group II intron in the ndhA pre-mRNA: the site is not edited in unspliced transcripts and it is fully edited in spliced transcripts [46–48]. Failure to edit unspliced RNA is presumably due to the position of the intron between the edited site and the cis-element that specifies it. Translation that initiates on unspliced ndhA RNA would terminate at an in-frame stop codon within the intron. Thus, exon 2 is translated only from spliced RNAs, and these are 100% edited. In the case of rpl2, the editing event creates an AUG start codon from an ACG precursor; this is the only editing event in maize chloroplasts that creates a canonical start codon. Although it has been reported that ACG can function as a start codon in chloroplasts [49, 50], our data show that this particular ACG is strongly discriminated against by initiating ribosomes.
The fact that the Ribo-seq data show the expected strong bias toward edited rpl2 and ndhA(563) instills confidence that valid conclusions can be made from our data for other edited sites. Approximately 40% of the petB and ndhA(nt 50) sequences are unedited in both the RNA-seq and Ribo-seq data, indicating that these unedited sequences give rise to a considerable fraction of the translational output of the corresponding genes. Editing of the petB site is essential for the function of its gene product (cytochrome b6) [51]. It seems likely that the product of this unedited RNA is either unstable or selected against during complex assembly, as has also been suggested for the products of two unedited transcripts in mitochondria [52, 53]. The remaining sites show almost complete editing in the RNA-seq data and, as expected, in the Ribo-seq data as well. That said, there is an overall trend toward less representation of unedited sequences in the Ribo-seq data than in the RNA-seq data. This may simply be a kinetic effect as would be expected if ribosome binding is slow in comparison to editing, such that ribosomes generally translate older (and therefore more highly edited) mRNAs.
Ribosome profiling has provided a wealth of new insights into translation and associated processes in a wide variety of organisms [54], but its application to questions in organellar biology is just beginning. The method has been used to analyze the effects of a disease-associated mutation in mitochondria [19], to define targets of nucleus-encoded translational activators in chloroplasts [15] and to characterize the cotranslational targeting of chloroplast-encoded proteins to the thylakoid membrane [36]. The results reported here provide the first comprehensive description of an organellar transcriptome and translatome in a developmental context. The data revealed dynamic changes in RNA abundance and translational efficiency during the differentiation of proplastids into chloroplasts, elucidated mechanisms that dictate the abundance of chloroplast-encoded proteins, clarified the relationship between RNA editing and translation, and provided new insights that suggest hypotheses to be explored in future studies.
Ribosome profiling data from bacteria revealed a striking correspondence between the stoichiometry of subunits of multisubunit complexes and their relative rates of synthesis [9, 10]. Our results show that the relative translational outputs of chloroplast genes likewise approximate the relative abundance of the gene products. This tuning is apparent when comparing sets of genes encoding different complexes (e.g. compare genes encoding the low abundance NDH complex to genes encoding the highly abundant PSI and PSII complexes) (Fig 3B), and when comparing genes encoding subunits of the same complex (e.g. the PEP RNA polymerase and the ATP synthase) (Fig 4A and 4B).
Our calculations of translational output rest on the assumption that the rate of translation elongation on all mRNAs is similar under any particular condition. This same assumption produced remarkable concordance between protein stoichiometry and inferred translational output in bacteria [9, 10]. Although our results show a clear trend toward “proportional synthesis”, they also suggest that the tuning of protein output to stoichiometry is less precise in chloroplasts than it is in bacteria. Subunits of photosynthetic complexes are subject to proteolysis when their assembly is disrupted [55], and a similar (albeit wasteful) mechanism could contribute to balancing stoichiometries when proteins are synthesized in excess under normal conditions. That said, instances in which inferred translational outputs are particularly incongruent with protein stoichiometries may reflect physiologically informative behaviors. The most prominent examples of “over-produced” proteins in our data are PsbA and PsbJ in PSII, PsaC and PsaJ in PSI, NdhK in the NDH complex, Rps14 in ribosomes, and PetD, PetL and PetN in the cytochrome b6f complex (Fig 4 and S6 Fig). Disproportionate synthesis of PsbA is well known, and compensates for its damage and proteolysis during photosynthesis [26]. The other proteins suggested by our data to be produced in excess may likewise be subject to more rapid turnover than their partners in the assembled complex. A proteomic study in barley demonstrated that subunits of each photosynthetic complex generally turn over at similar rates [56], but data for these particular proteins were not reported.
Interestingly, the inferred rates of synthesis of PsbA, PetD, and NdhK are well matched to those of their partner subunits early in development, but outpace those of their partners in mature chloroplasts (Fig 4D,S6 Fig). This feature of psbA expression coincides with the need to replace its gene product, D1, following photo-induced damage and proteolysis [26]. By extension, the developmental dynamics of petD and ndhK expression suggest that their gene products may turn over more rapidly than their partners as a consequence of photosynthetic activity.
In bacteria, proportional synthesis of subunits within a complex is achieved largely through the tuning of translational efficiencies among ORFs on the same mRNA [9, 10]. In chloroplasts, genes encoding subunits of the same complex are generally distributed among multiple transcription units [4] and RNA segments within a transcription unit often accumulate to different levels [8]. It is interesting to consider how this shift in the gene expression landscape is reflected in the mechanisms that balance protein output among genes. In the case of the four genes encoding the PEP RNA polymerase, relative translational outputs closely match the 2:1:1:1 protein stoichiometry, and this is programmed primarily at the level of mRNA abundance (Fig 4B). By contrast, widely varying translational efficiencies are superimposed on small variations in mRNA abundance to tune translational output to protein stoichiometry in the ATP synthase complex (Fig 4A). Genes for ribosomal proteins are distributed among ten transcription units, several of which also encode proteins involved in photosynthesis (see Fig 3A). For example, rps14 is cotranscribed with genes encoding the reaction center proteins of PSI (psaA/psaB), and translational outputs within this transcription unit are balanced by large differences in translational efficiency (Fig 4C). Similarly, the psaI transcription unit encodes subunits of the abundant PSI and cytochrome b6f complexes, a low abundance PSI assembly factor (Ycf4) and a protein of unknown function (CemA); large differences in translational efficiency adjust the translational outputs to meet these different needs (Fig 4C bottom). For complexes harboring plastid-encoded subunits in equal stoichiometries (ribosomes, NDH, PSI, PSII, cytochrome b6f), compensating differences in translational efficiency generally buffer differences in mRNA level. Taken together, these results imply that mRNA abundance and translational efficiencies have coevolved in chloroplasts to produce proteins in close to the optimal amounts. In some instances, mRNA levels are sharply out of balance with protein stoichiometries, in which case differential translational efficiencies compensate. In other instances, mRNA levels approximate protein stoichiometries, and translational efficiencies are similar. These observations further suggest that for most genes in maize chloroplasts, mRNA levels and translational efficiencies are poised such that they limit the rate of protein synthesis to a similar extent. This view is further supported by the developmental dynamics discussed below.
In Chlamydomonas chloroplasts, synthesis of subunits within the same photosynthetic complex is coordinated through assembly-dependent auto-regulatory mechanisms [57]. By contrast, current data for angiosperm chloroplasts suggest that translational efficiencies are generally independent of the assembly status of the gene products [15, 58]. It seems likely that translational efficiencies are dictated by the interplay between the sequence and structure of RNA proximal to start codons and the proteins that bind this region. Translation initiation in chloroplasts sometimes involves a Shine-Dalgarno interaction and is facilitated by an unstructured translation initiation region [6, 59]. Additionally, the translation of some chloroplast ORFs requires the participation of gene-specific translation activators [15, 60–75]. Such proteins provide a means for tuning protein synthesis within and between transcription units. The atpH ORF and its nucleus-encoded translational activator PPR10 exemplify this mechanism. The exceptionally high translational efficiency of atpH (Fig 3A) boosts its translational output to match the high stoichiometry of AtpH in the ATP synthase complex (Fig 4A); this high translational efficiency requires the binding of PPR10 adjacent to the atpH ribosome binding site, an interaction that prevents the formation of inhibitory RNA structures involving the translation initiation region [15, 30, 62].
Our results provide a comprehensive view of the dynamics of chloroplast mRNA abundance and translation during the proplastid to chloroplast transition. The majority of genes involved in chloroplast gene expression exhibit peak mRNA abundance and translational output in developing chloroplasts (segment 4) whereas the majority of genes encoding subunits of the photosynthetic apparatus exhibit peak mRNA abundance and translational output in young chloroplasts (segment 9) (Fig 5C). That said, even in proplastids (segment 1), genes involved in photosynthesis are generally represented by more mRNA and a higher translational output than are those involved in chloroplast gene expression (S5 Fig). Our data show that programmed changes in translational efficiency combine with changes in mRNA abundance to produce developmental shifts in translational output (Fig 5A). In general, translational efficiency is lowest at the leaf base, reflecting the low ribosome content in proplastids. The translational efficiency of most ORFs peaks in young chloroplasts (segment 9). In this context, it is intriguing that one subset of genes exhibit peak translational efficiency in the basal leaf segments (Fig 5A, bottom; Fig 5C, right), whereas another subset increases in translational efficiency right out to the leaf tip (Fig 5A and 5C). The former group is strongly enriched for “biogenesis” genes (RNA polymerase, ribosomes, assembly factors), and the latter for photosynthesis genes. Possible mechanisms underlying these distinct “translational regulons” are discussed below.
A study in Chlamydomonas showed that changes in chloroplast mRNA abundance are not reflected by corresponding changes in rates of protein synthesis, leading to the conclusion that translation is the primary rate-limiting step [76]. The data presented here suggest that this is not the case in maize chloroplasts. The developmental shifts in mRNA abundance were largely mirrored by shifts in translational output (Fig 5A and 5B), implying that mRNA abundance has considerable impact on the output of most chloroplast genes in maize. Likewise, chloroplast DNA copy number limits gene expression in developing maize chloroplasts [77] but does not limit gene expression in Chlamydomonas chloroplasts [76]. It is perhaps unsurprising that mechanisms of gene regulation have diverged in the chloroplasts of vascular plants and single-celled algae, given their very different developmental and ecological contexts.
Our data revealed a strong correlation between gene function and the developmental dynamics of mRNA abundance (Fig 5C middle): mRNAs encoding proteins involved in gene expression generally peak in abundance earlier in development than do those encoding components of the photosynthetic apparatus. This finding was foreshadowed by analyses of several chloroplast mRNAs during leaf development in barley and Arabidopsis [78–80]. Land plant chloroplasts harbor two types of RNA polymerase, a single-subunit nucleus-encoded polymerase (NEP) and a bacterial-type plastid-encoded polymerase (PEP) [5]. The ratio of NEP to PEP drops precipitously during chloroplast development, and this likely makes a large contribution to the changes in chloroplast mRNA pools [5, 78, 80, 81]. There is evidence that NEP plays an especially important role in the transcription of “house keeping” genes, and PEP in the transcription of photosynthesis genes [5]; however, most chloroplast genes can be transcribed by both NEP and PEP [82], and the degree to which each polymerase contributes to the transcription of each gene during the course of chloroplast development remains unknown. Chloroplasts harbor several nucleus-encoded sigma factors that target PEP to distinct promoters [83], and these provide an additional means to tune transcription rates in a developmental context.
Changes in RNA stability combine with changes in transcription to modulate mRNA pools during chloroplast development [78, 80, 81, 84, 85]. Determinants of chloroplast mRNA stability include various ribonucleases, RNA structure, ribosome occupancy, and proteins that protect RNAs from nuclease attack [8]. Most mRNA termini in chloroplasts are protected by helical repeat RNA binding proteins that provide a steric blockade to exoribonucleases [8, 61, 62, 86]. The majority of such proteins belong to the pentatricopeptide repeat (PPR) family, a large family of sequence-specific RNA binding proteins that influence virtually every post-transcriptional step in gene expression in mitochondria and chloroplasts [87]. In addition, chloroplasts harbor abundant hnRNP-like proteins, and these have been shown to impact the stability of several chloroplast mRNAs [88, 89]. Programmed changes in the abundance and/or activities of PPR and hnRNP-like proteins might contribute to the shifting mRNA pools during the proplastid to chloroplast transition.
Changes in translational efficiency superimpose on changes in mRNA abundance to modulate the output of plastid genes during the transformation of proplastids into chloroplasts. ORFs encoding proteins involved in photosynthesis generally exhibit maximal translational efficiency in young or mature chloroplasts (segments 9 and 14), whereas those that function in gene expression generally peak in translational efficiency earlier in development (Fig 5C right). Furthermore, our data suggest that mRNAs encoding PSII reaction center proteins are translated with higher efficiency in mesophyll chloroplasts than in bundle sheath chloroplasts (Fig 6C). It will be interesting to explore the mechanisms that underlie these differential effects on translational efficiency. Some possibilities include shifts in stromal pH, Mg++, or the polymerase generating the mRNA (NEP versus PEP), which might impact the formation of RNA structures at specific ribosome binding sites. Programmed changes in the activities of nucleus-encoded gene-specific translational activators could modulate translational efficiencies in a developmental context. Most such proteins in land plant chloroplasts are PPR (or PPR-like) proteins, and several of these also stabilize processed mRNAs with a 5’ end at the 5’ boundary of their binding site [15, 30, 60–62, 64–67, 90–93]. Indeed, many polycistronic transcripts in chloroplasts are processed to smaller isoforms whose ends are defined and stabilized by PPR-like proteins [7, 8, 86]. The impact of this type of RNA processing on translational efficiencies in vivo remains unclear. The removal of upstream ORFs is not required for the translation of several ORFs that are found on processed RNAs with a proximal 5’-terminus [35, 36]. Some proteins have dual translation activation and RNA processing/stabilization functions, implying that the two activities are coupled [15, 30, 60–62, 65, 66, 91–93]; however, the translation activation and RNA processing/ stabilization effects of such proteins could be independent consequences of their binding upstream of an ORF [62, 86]. We showed here that there is a correlation between the accumulation of processed RNA isoforms and changes in relative translational efficiencies in two polycistronic transcription units (S8 Fig). Deciphering the cause and effect relationships underling these correlations presents a challenge for the future.
The data presented here lead to numerous new questions for future exploration. Is the synthesis of nucleus-encoded subunits of photosynthetic complexes tuned to that of their chloroplast-encoded partners? What is the mechanistic basis for the preferential translation of some mRNAs in developing chloroplasts and others in photosynthetic chloroplasts? To what extent do environmental inputs such as light and temperature modify the developmental dynamics of chloroplast mRNA abundance and translation? The use of ribosome profiling can be anticipated to accelerate progress in addressing these and many other long-standing questions relating to the biology of organelles.
For the developmental analysis, Zea mays (inbred line B73) was grown under diurnal cycles for 9 days and harvested as described [12]. Leaf sections from twelve plants were pooled for each of three replicates; each pool contained between 0.15 g and 0.3 g tissue. Plants used to prepare mesophyll and bundle sheath fractions were grown similarly, except the light was set at 300 μmol·m-2·s-1 and the tissues were harvested 13 days after planting, 2 hours into the light cycle. The apical one-third of leaf two and three were pooled from fifteen seedlings for each replicate, and the bundle sheath and mesophyll-enriched fractions were obtained with a rapid mechanical procedure. The tissue was cut into ~ 1 cm-sections, placed in a pre-chilled mortar and pestle, and lightly ground for 2 min in 5 ml of ice-cold modified polysome extraction buffer lacking detergents (0.2 M sucrose, 0.2 M KCl, 50 mM Tris-acetate, pH 8.0, 15 mM MgCl2, 20 mM 2-mercaptoethanol, 2 μg/ml pepstatin A, 2 μg/ml leupeptin, 2 mM phenylmethanesulfonyl fluoride, 100 μg/ml chloramphenicol, 100 μg/ml cycloheximide). The material that was released into solution constituted the mesophyll cell-enriched fraction. A portion of this was frozen in liquid N2 for RNA isolation, and the remainder was stored on ice while bundle sheath strands were purified from the tissue remaining in the mortar. The tissue was subject to four additional rounds of light grinding (2 min), each time in a fresh aliquot of 5-ml modified polysome extraction buffer. The light green fibers remaining constituted the bundle sheath enriched fraction; these cells were broken by hard grinding in 5 ml of modified polysome extraction buffer. A portion of this material was flash frozen for future RNA isolation and the remainder was used immediately for ribosome footprint isolation. Polyoxyethylene (10) tridecyl ether and Triton X-100 were added to the mesophyll and bundle sheath fractions retained for ribosome profiling (final concentrations of 2% and 1%, respectively), and the material was filtered through glass wool. The isolation of ribosome footprints and total RNA were performed as described below.
Ribosome footprints were prepared using a protocol similar to that described in [15], but with two key modifications: (i) RNAse I rather than micrococcal nuclease was used to generate monosomes, and (ii) the centrifugation time used to pellet ribosomes through the sucrose cushion was shortened to reduce contamination by other RNPs. Tissues were pulverized in liquid N2 with a mortar and pestle, and thawed in 5 ml of polysome extraction buffer (0.2 M sucrose, 0.2 M KCl, 50 mM Tris-acetate, pH 8.0, 15 mM MgCl2, 20 mM 2-mercaptoethanol, 2% polyoxyethylene (10) tridecyl ether, 1% Triton X-100, 100 μg/ml chloramphenicol, 100 μg/ml cycloheximide). A 2.4-ml aliquot was removed and frozen in liquid N2 for total RNA isolation. The remaining suspension was filtered through glass wool and centrifuged at 15,000xg for 10 min. The supernatant was digested with 3,500 units of RNAse I (Ambion) at 23°C for 30 min. 2.5 ml lysate was layered on a 2 ml sucrose cushion (1 M sucrose, 0.1 M KCl, 40 mM Tris-acetate, pH 8.0, 15 mM MgCl2, 10 mM 2-mercaptoethanol, 100 μg/ml chloramphenicol, and 100 μg/ml cycloheximide) in a 16 x 76 mm tube and centrifuged in a Type 80 Ti rotor for 1.5 h at 55,000 rpm. The pellet was dissolved in 0.7 mL of ribosome dissociation buffer (10 mM Tris-Cl, pH 8.0, 10 mM EDTA, 5 mM EGTA, 100 mM NaCl, 1% SDS). RNA was isolated with Tri reagent (Molecular Research Center). RNAs between ~20 and ~35 nt were purified on a denaturing polyacrylamide gel, eluted, extracted with phenol/chloroform, precipitated with ethanol, and suspended in water. We have subsequently modified our protocol to purify RNAs between 20 and 40 nt; this results in a small shift in the size distribution of the reads (S1B Fig).
The ribosome footprint preparation was treated with T4 polynucleotide kinase. Twenty ng of the kinased RNA was converted to a sequencing library using the NEXTflex Small RNA Sequencing Kit v2 (Bioo Scientific), which minimizes ligation bias by introducing four randomized bases at the 3’ ends of the adapters [17]. rRNA fragments were depleted by subtractive hybridization after first-strand cDNA synthesis, using 54 biotinylated DNA oligonucleotides corresponding to the most abundant rRNA fragments detected in pilot experiments (see S1 Table). 10 μl of the oligonucleotide mixture (concentrations as in S1 Table) was added to 40-μl of the first-strand synthesis reaction and heated to 95°C for 2 min. A 50-μl aliquot of pre-warmed 2X hybridization buffer (10 mM Tris-Cl pH 7.5, 1 mM EDTA, 2 M NaCl) was added and incubated at 55°C for 30 min. The solution was transferred to a new tube containing 1 mg of prewashed Dynabeads M-270 Streptavidin (Invitrogen) and incubated at room temperature for 15 min with frequent agitation. The tube was placed on a magnet for 5 min and the supernatant was collected and desalted using Sephadex G-25 Fine (GE Healthcare). The sample was concentrated to 18 μl and used as input for the PCR amplification step in the library construction protocol. After 14 cycles, PCR products were separated by electrophoresis through a 5% polyacrylamide gel and a gel slice corresponding to DNA fragments between markers at 147 and 180 bp (representing insert sizes of 20–53 bp) was excised. The DNA was eluted overnight, phenol/chloroform extracted, precipitated with ethanol, suspended in water, and stored at -20°C.
For RNA-seq, rRNA was depleted from the RNA samples using the Ribo-Zero rRNA Removal Kit (Plant Leaf) (Epicentre). One hundred ng of the rRNA-depleted RNA was used for library construction using the NEXTflex Rapid Directional qRNA-Seq Kit (Bioo Scientific) according the manufacturer’s instructions. The adapters provided with the kit include 8-nt molecular labels that were used during data processing to remove PCR bias. The libraries were combined and sequenced using a HiSeq 2500 or NextSeq 500 instrument (Illumina). The read lengths were 50 or 75 nt for Ribo-seq and 75 nt for mRNA-seq.
Adapter sequences were trimmed using cutadapt [94]. Ribo-seq reads between 18 and 40 nt were used as input for alignments. Alignments were performed using Bowtie 2 with default parameters [95], which permits up to 2 mismatches, thereby allowing edited sequences to align. Reads were aligned to the following gene sets, with unaligned reads from each step used as input for the next round of alignment: (i) chloroplast tRNA and rRNA; (ii) chloroplast genome; (iii) mitochondrial tRNA and rRNA; (iv) mitochondrial genome (B73 AGP v3); (v) nuclear tRNA and rRNA; nuclear genome (B73 AGP v3).
Maize genome annotation 6a (phytozome.jgi.doe.gov) was reduced to the gene set annotated in 5b+ (60,211 transcripts) (gramene.org). For metagene analysis, all coding sequence (CDS) coordinates from all transcript variants were combined to make a union CDS coordinate. Custom Perl scripts extracted mapping information using SAMtools [96] and analyzed mapped reads as follows. The distribution of ribosome footprint lengths and the RPKM for both the Ribo-seq and RNA-seq data were calculated based only on reads mapping to CDS regions. For RPKM calculations, we defined the total number of mapped reads as the number of reads mapping to nuclear CDSs. Translation efficiency was calculated from the division of ribosome footprint RPKM by RNA-seq RPKM.
Because unspliced RNAs constitute a substantial fraction of the RNA pool from intron-containing genes in chloroplasts, these genes require special treatment to infer the abundance of spliced (functional) mRNA. The fraction spliced at each intron was calculated in several ways. (i) RNA-seq reads were aligned to the chloroplast genome with splicing-aware software TopHat 2.0.11 [97]. The number of reads spanning each exon-exon junction (spliced) was divided by the sum of spliced (exon-exon) and unspliced (exon 1-intron or intron-exon 2) reads; (ii) RNA-seq reads were aligned with Bowtie2 to a reference gene set that included both spliced and unspliced forms (100-nt on each side of each junction). The fraction of spliced RNA was calculated as for method (i); (iii) RNA-seq reads were aligned with TopHat to the genome and the spliced fraction was calculated from (exon RPKM—intron RPKM)/exon RPKM. Values calculated by each method are provided in S3 Table. Summary plots report mRNA abundance and translational efficiencies only for genes for which all of these methods gave similar results. We cannot confidently infer the amount of fully spliced RNAs from genes with two introns (ycf3 and rps12), so these are also excluded.
Hierarchical clustering was performed using the Bioinformatic Toolbox of MATLAB software (Mathworks) using standardized values as input: values from the four leaf segments for each gene were standardized to have a mean of 0 and a standard deviation of 1 such that developmental shifts can be compared among genes despite differences in signal magnitude. Hierarchical clustering was performed using Pearson correlation coefficient values and unweighted average distance.
Antibodies to AtpB, D1 and PetD were raised by our group and have been described previously [30]. Antibodies to Atp6, NdhH, PPDK, and Rpl2 were generously provided by Christine Chase (University of Florida), Tsuyoshi Endo (Kyoto University), Kazuko Aoyagi (UC Berkeley), and Alap Subramanian (University of Arizona), respectively. Antibodies to PEPC, malic enzyme, and RbcL were generous gifts of William Taylor (University of California, Berkeley).
Illumina read sequences were deposited at the NCBI Sequence Read Archive with accession number SRP070787. Alignments of reads to the maize chloroplast genome used Genbank accession X86563.
|
10.1371/journal.pbio.3000147 | A polyploid admixed origin of beer yeasts derived from European and Asian wine populations | Strains of Saccharomyces cerevisiae used to make beer, bread, and wine are genetically and phenotypically distinct from wild populations associated with trees. The origins of these domesticated populations are not always clear; human-associated migration and admixture with wild populations have had a strong impact on S. cerevisiae population structure. We examined the population genetic history of beer strains and found that ale strains and the S. cerevisiae portion of allotetraploid lager strains were derived from admixture between populations closely related to European grape wine strains and Asian rice wine strains. Similar to both lager and baking strains, ale strains are polyploid, providing them with a passive means of remaining isolated from other populations and providing us with a living relic of their ancestral hybridization. To reconstruct their polyploid origin, we phased the genomes of two ale strains and found ale haplotypes to both be recombinants between European and Asian alleles and to also contain novel alleles derived from extinct or as yet uncharacterized populations. We conclude that modern beer strains are the product of a historical melting pot of fermentation technology.
| The budding yeast Saccharomyces cerevisiae has long been used to make beer. Yeast strains used to make ales are known to differ genetically and phenotypically from strains used to make wine and from strains isolated from nature, such as oak isolates. Beer strains are also known to be polyploid, having more than two copies of their genome per cell. To determine the ancestry of beer strains, we compared the genomes of beer strains with the genomes of a large collection of strains isolated from diverse sources and geographic locations. We found ale, baking, and the S. cerevisiae portion of lager strains to have ancestry that is a mixture of European grape wine strains and Asian rice wine strains and that they carry novel alleles from an extinct or uncharacterized population. The mixed ancestry of beer strains has been maintained in a polyploid state, which provided a means of strain diversification through gain or loss of genetic variation within a strain but also a means of maintaining brewing characteristics by reducing or eliminating genetic exchange with other strains. Our results show that ale strains emerged from a mixture of previously used fermentation technology.
| The brewer's yeast Saccharomyces cerevisiae is known for its strong fermentative characteristics. The preference for fermentation in the presence of oxygen arose as a multistep evolutionary process around the time of an ancient genome duplication, endowing numerous species with the ability to produce levels of ethanol toxic to many microorganisms [1,2]. One of these species, S. cerevisiae, also gained the ability to competitively dominate many other species in high-sugar, low-nutrient environments, such as grape must [3]. Wine is largely fermented by S. cerevisiae and is thought to be the first fermented beverage, having been made for over 9,000 years [4]. However, S. cerevisiae is not the only Saccharomyces species used to make fermented beverages; others, particularly S. uvarum, S. kudriavzevii, S. eubayanus, and hybrid derivatives, are also used, particularly for fermentations at low temperatures [5–8]. Besides S. cerevisiae, the most widely used species is S. pastorianus, an allopolyploid hybrid of S. cerevisiae and S. eubayanus, used to make lager beer [7]. The use of this hybrid emerged during the 15th century in Europe and was formed from an S. eubayanus strain closely related to wild populations from North America and Tibet [9,10] and a S. cerevisiae strain related to those used to ferment ales [11–13]. The origin of ale and other domesticated strains of S. cerevisiae is beginning to emerge through comparison with wild populations [12–16].
Multiple genetically distinct populations of S. cerevisiae have been found associated with fermented foods and beverage. These include grape wine, Champagne, sake and rice wine, palm wine, coffee, cacao, cheese, and leavened bread [14,17–20]. Ale strains have also been found to be both genetically and phenotypically differentiated from other strains [12,13]. Multiple populations of ale strains have been identified and found to exhibit high rates of heterozygosity and polyploidy [12,13,16]. However, the origin of such domesticated groups is not always clear because it requires comparison to wild populations from which they were derived, and these wild populations have not all been identified. The best characterized wild populations of S. cerevisiae have been isolated from oak and other trees in North America, Japan, China, and Europe [21–24], the latter of which is most closely related to and the presumed wild lineage from which European wine strains were derived.
Despite clear differences among many domesticated groups, human-associated admixture is common [20,22,25,26] and can blur the provenance of domesticated strains. For example, wine strains show a clear signature of admixture with other populations, and clinical strains appear to be primarily derived from admixed wine populations [27–29]. Ale strains, with the exception of a few found related to sake and European wine lineages, have no obvious wild population from which they were derived [12,13].
In this study, we examined the origin of ale and lager strains in relation to a diverse collection of S. cerevisiae strains. Through analysis of publicly available genomes and 107 newly sequenced genomes, we inferred a hybrid, polyploid origin of beer strains derived from admixture between close relatives of European and Asian wine strains. This admixture suggests that early industrial strains spread with brewing technology to give rise to modern beer strains, similar to the spread of domesticated plant species with agriculture.
We sequenced the genomes of 47 brewing and baking strains and 65 other strains of diverse origin for reference. Combining these with 430 publicly available genomes, we found nearly all the brewing strains closely related to previously sequenced ale and lager strains (S1 Fig). Through analysis of population structure, we identified 13 populations, 4 of which contain the majority (64/76) of beer strains. The four beer-associated populations consisted of predominantly lager strains, German ale strains (Ale 1), British ale strains (Ale 2), and a mixture of beer and baking strains (Beer/baking) and are consistent with previously identified groups of beer strains [12,13]. The remaining populations were similar to previously characterized groups [20,21,27] and were classified by the most common source and/or geographic region of isolation as Laboratory, Clinical, Asia/sake, Europe/wine, Mediterranean/oak, Africa/Philippines, China/Malaysia, and two populations from Japan/North America (S2 Table).
To identify the most likely founders of the four beer populations, we used a composite likelihood approach to infer population relationships while accounting for admixture, which can obfuscate population phylogenies [30]. The inferred admixture graph grouped the four beer populations together, with the lager and two ale populations being derived from the lineage leading to the Beer/baking population (Fig 1). The four beer populations are most closely related to the Europe/wine population. However, the admixture graph also showed strong support for two episodes of gene flow into the beer lineages resulting in 40% to 42% admixture with the Asia/sake population. We confirmed these admixture events using f4 tests for discordant population trees, which are caused by admixture [31,32]. All f4(Europe, test; Asia, Africa) statistics were significant for tests of the four beer populations (Z-scores < −21.3), whereas the f4(Europe, Mediterranean; Asia, Africa) statistic was much closer to zero for the Mediterranean/oak population (Z-score = 3.4). Similar results were obtained using the China or Japan/North American populations rather than Africa (S3 Table). Therefore, the beer populations were derived from admixture events between a population closely related to the Europe/wine population and Asia/sake population. Consistent with prior studies [20,29], we also found admixture events between the Europe/wine population and both the lab and clinical population.
To quantify the degree of admixture for each of the beer strains, we calculated f4 admixture proportions using the Europe/wine and Asia/sake populations [31,33]. For the 64 beer strains, we estimated an average of 39.6% (range 36.7%–46.7%) of their genome was derived from the Asia/sake population and 60.4% was derived from the Europe/wine population. The high proportion yet narrow range of admixture implies little to no subsequent back-crossing following admixture.
Polyploidy is enriched in beer and baking strains [12,16,18] and has been shown to contribute to reproductive isolation [34]. The Beer/baking population includes 10 previously studied strains, all of which were found to be triploid or tetraploid and to exhibit high rates of heterozygosity [20]. These strains were previously found to group with other strains isolated from diverse sources around the world (Pan/Mixed 2 in [20]). To identify triploids and tetraploids strains, we used the expected allele frequency at heterozygous sites: 50% for diploids, 33% and 66% for triploids, and 25%, 50%, and 75% for tetraploids (Fig 2). We note that this approach can miss triploid or tetraploid strains due to low heterozygosity or read coverage but should not yield any false positives. Nevertheless, out of 105 strains with an abundance of heterozygous sites, we identified 23 triploid and 28 tetraploid strains (S2 Fig). Of the 51 polyploid strains (N > 2), 45 (88%) were in one of the four beer populations, of which 29 were beer strains and 6 were baking strains. The remaining 10 polyploids assigned to the beer populations include three isolates from green coffee beans and were previously assigned to a Pan/Mixed 2 population [20], a group of predominantly human-associated strains. These results are comparable to the high rates of polyploidy found in prior studies of beer strains [12,16].
Given the admixed origin of beer strains, we wanted to know which populations the heterozygous sites indicative of polyploidy were derived from. We examined heterozygous sites in the beer populations in relation to other strains by clustering SNPs and grouping strains by their inferred population membership (Fig 3). Excluding the lager population, which are not heterozygous, the three beer populations are predominantly heterozygous for alleles abundant within either the Asia/sake or Europe/wine population. However, all four beer populations also carry alleles not present in any of the other populations. The presence of heterozygous, beer-specific alleles suggests that these alleles were derived from admixture between a lineage closely related to the Europe/wine lineage and/or the Asia/sake lineage. The large number of beer-specific alleles is unlikely to have accumulated in the recent past subsequent to the formation of the polyploids. The beer groups have between 6,558 and 13,728 alleles present at 25% frequency or more in the group but not in any other population. Using these beer-specific alleles, we found the divergence at four-fold degenerate synonymous sites was 0.153%, 0.100%, 0.087%, and 0.069% along the Ale 1, Ale 2, Beer/baking, and Lager lineages. These rates are higher than expected to have accumulated since the use of these strains for brewing purposes (see Discussion) and not much less than the rate of divergence between the Europe/wine and Asia/sake population (0.592%). Combined with the observation that these variants are mostly heterozygous, we infer that many were present when the strains became polyploid and originated from an extinct or as yet to be characterized yeast population related to the Europe/wine and/or Asia/sake population.
The presence of heterozygous European, Asian, and beer-specific alleles enabled us to use haplotype phasing to test whether there has been recombination between European and Asian haplotypes and whether beer-specific alleles reside on European or Asian haplotypes. We used long-read sequencing to phase two beer strains—a German ale strain in the Ale 2 population (A.2565) and a Belgian ale strain in the Beer/baking population (T.58). Both strains were inferred to have over 99% ancestry to their assigned population, and both are likely polyploids (S2 Table). As a control, we phased the genome of a hybrid (YJF1460) generated between a Europe/wine strain and a Japan/North America 2 oak strain. Because of uncertainty in the ploidy as well as the possibility of variable ploidy levels (aneuploidy) across the genome, we developed a phasing algorithm that merges reads into consistent haplotypes and makes no assumptions about ploidy.
Phasing of the three strains yielded predominantly two haplotypes in the YJF1460 control and three or four haplotypes in the two ale strains (Fig 4 and S4 Fig). The majority of phased haplotypes in the two ale strains carried a mixture of European and Asian alleles. In contrast, the YJF1460 control showed few haplotypes with both European and Asian alleles, which is indicative of haplotype switching errors or mitotic exchange. Eliminating regions of haplotype switching less than 4 kb in length, which could result from genotype errors or mitotic gene conversion, we counted the number of switches within the phased haplotypes between European and Asian alleles and found 12 in YJF1460, 346 in T.58, and 199 in A.2565 (S4 Table). Consequently, most haplotypes present in the two ale strains represent recombinant haplotypes as opposed to pure European-related or Asian-related haplotypes (Fig 4 and S4 Fig); only 22% of the T.58 genome and 19% of the A.2565 genome carried haplotypes with over 95% European or 95% Asian alleles. In contrast, 88% of the YJF1460 genome was inferred to carry pure European or Asian haplotypes.
The amount of recombination between European and Asian alleles is indicative of time since admixture. We measured the decay in linkage disequilibrium between European and Asian alleles on phased haplotypes as a function of distance and found a 50% drop in linkage disequilibrium corresponds to 6.3 kb in A.2565 and 30 kb in T.58 and no decay in YJF1460 (S5 Fig). Assuming 0.34 kb/cM [35], this translates to an equivalent of 46.9 meiotic events in A.2565 and 9.8 meiotic events in T.58. For comparison, we estimated the number of meiotic equivalents since admixture of the Clinical (9.5), Laboratory (13.7), Lager (72.1), Beer/baking (6.7), Ale 1 (75.0), and Ale 2 (67.5) populations. These results indicate more recent admixture of the Clinical, Laboratory, and Beer/Baking populations compared with the Lager and two ale populations.
Although it is difficult to know how much recombination occurred prior to polyploidy and how much occurred subsequently through mitotic recombination or gene conversion, mitotic events have contributed to diversification of beer strains subsequent to polyploidy. There are four large and a number of smaller regions in A.2565 that exhibit loss of heterozygosity (Fig 4), and loss of heterozygosity is a distinct signature of mitotic recombination. Recombination of European- and Asian-derived haplotypes occurred prior to loss of heterozygosity because the fixed haplotypes in regions where there is loss of heterozygosity are recombinants of European- and Asian-derived haplotypes.
The polyploid beer strains also carry beer-specific alleles not present in other strains. These beer-specific alleles could have been inherited from an ancestral population that split from either the European/wine or Asian/sake population, or from an admixed population. To distinguish between these possibilities, we examined the distribution of beer-specific alleles on the phased haplotypes and found most (approximately 80%) were on mixed haplotypes, having at least 5% European and 5% Asian alleles. Therefore, beer-specific alleles were present during admixture of European and Asian alleles. The remaining beer-specific alleles were equally distributed between predominantly European or predominantly Asian haplotypes (S4 Table). We also counted alleles at heterozygous sites as an indicator of ploidy contribution. At sites with four phased haplotypes, we found Asian alleles had an average of 1.93 and 1.96 copies in T.58 and A.2565, respectively. In contrast, beer-specific alleles had an average 1.31 and 1.56 copies in T.58 and A.2565, respectively. The lower copy number of beer-specific alleles suggests either lower beer-specific allele frequencies compared to European/Asian allele frequencies in the ancestral admixed population prior to polyploidy or that polyploidy involved parents from populations with unequal representation of beer-specific alleles compared to European/Asian alleles. Finally, because many of the beer-specific alleles are not shared between the Ale 1, Ale 2, Beer/baking, and Lager populations, we can infer multiple origins of the four beer populations despite similar episodes of admixture and polyploidy, demonstrating that the S. cerevisiae contribution to lager strains did not simply come from one of the other beer strain lineages.
Inferring the origin of domesticated organisms can be complicated by extinction of wild progenitor populations, human-associated migration, polyploidy, and admixture with wild populations. In this study, we find that extant beer strains are polyploid and have an admixed origin between close relatives of European and Asian wine strains. Ale genomes, like lager genomes, carry relics of their parental genomes captured in a polyploid state as well as novel beer alleles from an extinct or undiscovered population. Loss of heterozygosity through mitotic exchange provided a means of strain diversification but has also potentially eroded precise inference of the timing and order of events giving rise to modern beer strains. Below, we discuss models and implications for an admixed, polyploid origin of beer strains.
Polyploidy is thought to mediate rapid evolution [36], and prior work showed that polyploidy is common in beer and baking strains [12,18,31]. We find that the Ale 1, Ale 2, and Beer/baking population all have a polyploid origin. Although not all strains had sufficient coverage for calling polyploidy, all those that did were either triploid or tetraploid. Chromosome level aneuploidy is also more common in strains within the Ale 1 (52%), Ale 2 (19%), and Beer/baking (52%) populations than in the nonbeer populations (5.1%). A notable consequence of both polyploidy and aneuploidy is that they can limit admixture with haploid or diploid strains due to low spore viability [34,37,38], thereby maintaining their brewing characteristics. Indeed, beer strains exhibit low sporulation efficiency and spore viability [12]. Both grape wine and particularly sake wine strains have also evolved more limited capacities to interbreed through low sporulation efficiencies [39,40].
Human-associated admixture is well documented in wine strains, which have been dispersed around the globe with the spread of viticulture [20,22,25,26]. However, admixture between close relatives of European grape wine and Asian rice wine populations presents a conundrum regarding where and how these populations became admixed. A crucial yet unresolved piece of information is where European wine strains were domesticated. The discovery of a Mediterranean oak population closely related to European wine strains suggests a European origin of wine strains [21]. An alternative model is that the Mediterranean oak population is a feral wine population and both the European wine and Mediterranean oak populations are nonnative. Analysis of a diverse collection of Asian strains suggested an East Asian origin of all domesticated S. cerevisiae strains, including European wine strains [14]. Domestic populations from solid and liquid state fermentations (bread, milk, distilled liquors, rice wines, and barley wines) were found related to wild populations from East Asia. In support of European wine and Mediterranean oak populations also originating in East Asia, these populations carry duplicated genes involved in maltose metabolism and grouped with fermented milk and other strains isolated from China. However, this model also has some uncertainty given the small number of Chinese isolates within the European wine group, the dispersion of European wine strains with viticulture, and the absence of samples from the Caucasus where grapes are thought to have been domesticated [4,41].
Considering the uncertainty of where European wine strains were domesticated, we put forth two hypotheses regarding the admixed origin of beer strains. First, European wine strains were domesticated in East Asia and admixed in situ with a population related to the Asia/sake group, which contains eight sake/rice wine strains, seven distillery strains, and seven bioethanol strains, mostly from Asia. Second, European wine strains were domesticated in Europe from a Mediterranean oak population, or perhaps in the Caucasus, and the admixed beer populations arose through East–West transfer of fermentation technology, including yeast by way of the Silk Route. Resolving these scenarios would be greatly facilitated by finding putative parental populations of diploid but not necessarily wild strains that carry alleles we find to be unique to the Ale 1, Ale 2, Beer/baking, and Lager groups. As yet, such populations have not been sampled or are extinct.
Even with a clear signature of a polyploid and admixed origin of beer strains, there are uncertainties regarding the founding strains and the order of events. The decay in linkage disequilibrium suggests that admixture occurred prior to polyploidy, and the distribution of beer-specific alleles suggests that admixture involved at least one uncharacterized population. However, polyploid genomes are often labile, and it is hard to know the extent to which mitotic recombination and gene conversion have altered genetic variation in the beer strains. In yeast, the rate of mitotic gene conversion and recombination has been estimated to be 1.3 × 10−6 per cell division and 7 × 10−6 per 120 kb, respectively [42,43], and both can lead to loss of heterozygosity. Converting to the size of a tetraploid genome (approximately 48 Mbp), we expect 0.0038 (using a median track length of 16.6 kb) conversion events and 0.0028 recombination events across the genome per cell division. Three lines of evidence support the role of these mitotic events in beer strains. First, many of the switches between the European and Asian alleles involved one or a small number of adjacent SNPs rather than long segments, indicative of gene conversion (S4 Table). Second, one strain (A.2565) shows clear loss of heterozygosity on multiple chromosomes, indicative of mitotic recombination (S4 Fig). Third, there is substantial genotype diversity within each of the beer populations (Fig 3). This would be expected to occur if loss of heterozygosity occurred during strain divergence but subsequent to the founding of each beer population.
Two other factors besides mitotic gene conversion and recombination must be considered in regards to diversity within the beer populations—outcrossing and de novo mutation. Outcrossing with strains outside of the beer population is unlikely because there is no evidence for this type of admixture in our analysis and admixture proportions from the Asian population is fairly constant at 37% to 47% across beer strains. However, it is worth noting that outcrossing of strains within or between different beer populations may not easily be detected. De novo mutations have undoubtedly occurred, but even using a reasonable estimate of 150 generations per year for brewing strains [12] and a per base mutation rate of 5 × 10−10 [44], the beer lineage substitution rates yield divergence times of 2.0 × 104 (Ale 1), 1.3 × 104 (Ale 2), 1.1 × 104 (Beer/baking), and 9.2 × 103 (Lager) years. Therefore, a sizable fraction of beer-specific alleles was likely inherited from populations closely related to European wine and Asian wine populations rather than de novo mutations that accumulated subsequent to polyploidy. Regardless of the relative impact of mitotic recombination, gene conversion, outcrossing, and de novo mutation, beer strains have diversified from one another but have remained relatively distinct from other populations of S. cerevisiae [12,13].
In conclusion, beer strains are the polyploid descendants of strains related to but not identical to European grape wine and Asian rice wine strains. Therefore, similar to the multiple origins of domesticated plants, including barley [45] and rice [46,47], beer yeasts are the products of admixture between different domesticated populations and benefited from historical transfer of fermentation technology.
Genome sequencing was completed for 47 commercial yeast strains, which include 33 ale, 7 lager, 2 whiskey, and 5 baking strains. For reference, sequencing was also completed for 60 strains of diverse origin, including 22 isolates from trees or other nonhuman-associated sources and 38 isolates from human-associated ferments such as togwa, coffee, and cacao (S1 Table). For each strain, DNA was extracted and indexed libraries were sequenced on Illumina machines (NextSeq, HiSeq2000, or HiSeq2500). A median of 10.7 million reads per strain was obtained, ranging from 272,000 to 26 million. The sequencing data is available at NCBI (PRJNA504476).
Genomic data was obtained for 430 strains from publicly available databases. These include 138 additional beer strains from [12,13]. We also obtained reference genomes for S. paradoxus, S. mikatae [48], and S. eubayanus (SEUB3.0) [49]. Two large sets of recently published genomes [14,16] were obtained for comparison with our set of 537 genomes. Genotype calls for SNPs identified in this study were obtained from gvcf files of the 1,011 yeast genomes project [16], and genotype calls were generated for 266 strains from China [14] using the mapping and genotyping pipeline described below. Because these two later sets of data were only available recently, they were only incorporated into the S1 Fig heatmap.
Reads were aligned to the S. cerevisiae S288c reference genome (R64-1-1_20110203) using BWA-v0.7.12-r1039 [50]. Lager strains were mapped to a concatenated S. cerevisiae and S. eubayanus genome and reads mapping to S. eubayanus were discarded. For short reads (<70 bp), we used BWA-sampe, and for the remainder, we used BWA-mem. Duplicate reads were marked prior to genotyping. Assembled genomes were also mapped using BWA-mem, and flags for secondary alignments were removed to facilitate complete mapping of large contigs. For S. paradoxus and S. mikatae, we obtained higher coverage of the S288c genome by mapping synthetic reads fromshredded contigs compared to mapping of full contigs and so used the former.
SNPs were called using short read data and then genotyped in those strains with assembled genomes. For SNP calling, we used GATK-UnifiedGenotyper-v3.3–0 [51] and applied the hard filters: QD < 5, FS > 60, MQ < 40, MQRankSum < −12.5, and ReadPosRankSum < −8. The dataset was filtered to remove strains and sites with more than 10% missing data. Among those strains removed were lager strains of the type 1 Saaz group [11], but we retained S. paradoxus and S. mikatae for which we obtained calls at 78% and 40% of sites, respectively. Biallelic SNPs with a minor allele frequency of at least 1% and with at least four minor allele genotype calls were selected for analysis, resulting in a total of 273,963 SNPs. The 399 strains retained for analysis are listed in S2 Table, and the genotype data is available in variant call format from http://doi.org/10.6084/m9.figshare.7550009.v1. Genotype calls for these SNPs were also obtained for the 1,277 strains in the comparative data set [14,16].
To estimate our genotyping error rate, we compared six pairs of strains that were independently sequenced. Two of the strains, YJF153 and BC217, were haploid derivatives of diploids strains, YPS163 [52] and BC187 [53], respectively, that were also sequenced. The other four pairs were all beer strains independently obtain from Wyeast (Wyeast 1728, 1968, 2565, 2112) and independently sequenced at Washington University in St. Louis and University of Washington in Seattle. Between the pairs of strains, we found genotype discordance rates of 9.62 × 10−4 (YJF153/YPS163), 1.31 × 10−3 (BC217/BC187), 3.57 × 10−3 (L.2112/YMD1874), 3.00 × 10−3 (A.2565/YMD1952), 1.81 × 10−2 (A.1968/YMD1981), and 5.74 × 10−3 (A.1728/YMD1866). We retained the six pairs of strains throughout the analysis as a measure of robustness.
Ploidy and aneuploidy were assessed by read counts at heterozygous sites and read coverage, respectively. For ploidy analysis, genotypes of 317 strains were from assemblies, and so no information on heterozygous sites was available, and 117 strains had few heterozygous sites indicating they were haploid or homozygous diploid. Of the remaining 105 strains, 66 had sufficient coverage at heterozygous sites to make visual designations of ploidy [20,54,55]. Visual designations were based on dominant trends consistent with expected percentage of read counts supporting—diploid (50:50), triploid (33:66), tetraploid (25:50:75) allele configurations. Of the 39 strains without sufficient coverage to distinguish triploids from tetraploids, most (33) showed distributions consistent with polyploidy (ploidy > 2), and of these, 29 were beer strains (S2 Fig). Aneuploidy was assessed by visual inspection of read coverage across the genome. Aneuploidy was only called for clear cases in which one or more chromosomes showed a deviation in read coverage compared to all other chromosomes.
Population structure was inferred by running ADMIXTURE [56] on a set of 20,394 sites with a minimum physical distance of 500 bp. The variants from 138 strains in a recent study of beer strains [12] were removed because the assemblies eliminated heterozygous sites and raw reads for these genomes were not available. Based on 20 independent runs using between 4 and 20 populations for the 399 strains, we chose 13 based on an average change in the log-likelihood greater than 3 standard deviations of the variation in the log-likelihood among independent runs (S3 Fig). The beer populations of interest were not affected by this choice; with 12 populations, the 2 Japanese populations merged and with 14—a new population of admixed European wine strains was formed (S3 Fig).
Population admixture graphs were inferred using Treemix [30]. A subset of 199 strains with less than 1% admixture were used to generate a population admixture graph. The population from China was used to root the tree because two strains in the China population, HN6 and SX6, were most closely related to both S. paradoxus and S. mikatae, and blocks of 500 SNPs were used to obtain jacknife standard errors. Five episodes of migration were inferred (P < 4.9 × 10−12), with weights ranging from 0.18 to 0.49. Migration events were validated using f4 tests of admixture (S3 Table). For tests of tree discordance, we did not use the clinical and lab populations as reference populations because these showed evidence of admixture. f4 admixture proportions were estimated by the ratio of f4(Mediterranean, Africa; test, Europe) to f4(Mediterranean, Africa; Asia, Europe), in which each of the 64 beer strains in the Ale 1, Ale 2, lager, and beer/baking populations were individually tested.
Three strains were selected for PacBio sequencing and variant phasing. Two of the strains were beer strains, A.2565 and A.T58, and the third, YJF1460, was a hybrid we generated by mating a European/wine strain (BC217) and a Japan/North America 2 oak strain (YJF153). PacBio reads were aligned to the S288c reference genome using Blasr [57], and heterozygous variants in each genome were phased using HapCUT2 [58], and our own heuristic phasing method that accounts for variable ploidy levels across the genome. Average coverage at 56k, 59k, and 33k variant sites was 13.1, 18.8, and 13.0 for YJF1460, A.T58, and A.2565, respectively. Our custom phasing method used the variant call format files and fragment files from HapCUT2 as input, and output a variable number of phased haplotypes. HapCUT2 fragment files were generated with minimum base quality of 10. Reads were merged into haplotypes using a minimum overlap of four matching SNPs and a minimum of 80% matching SNPs. Reads were iteratively joined to haplotypes using the best scoring overlap based on score = matches– 5 × mismatches. Haplotypes were formed by three rounds of merging. In the first round, reads were merged into haplotypes without any mismatches. In the second and third rounds, haplotypes were merged using the criteria defined above. Error rates were estimated by counting the minimum number of mismatches of reads to the final set of haplotypes. Error rates of 1.84%, 2.03%, and 1.90% were obtained from comparison of reads to 3337, 2452, and 2607 haplotype alleles for YFJ1460, A.T58, and A.2565, respectively. The average number of haplotypes at phased sites was 2.29, 3.27, and 2.98 for YFJ1460, A.T58, and A.2565, respectively. Sites where three haplotypes were inferred in the YJF1460 control are largely due to overlapping haplotypes that were too short to merge. The long read data, custom phasing script and inferred haplotypes are available from http://doi.org/10.6084/m9.figshare.7550009.v1.
After phasing, two sets of SNPs were selected for analysis. The first set consisted of nearly fixed differences between the Europe/wine and Asia/sake populations. After excluding strains with more than 1% admixture, there were 34,022 sites with an allele frequency of 99% in Europe/wine strains (n = 47) and less than 1% frequency in Asia/sake strains (n = 28) or vice versa. The nearly fixed differences between Europe/wine and Asia/sake strains were used to quantify switching between European and Asian haplotypes. Switching events were measured by counting switches involving one or more sites, five or more sites, or sites spanning 4 kb or longer (S4 Table). The latter two measures were used to avoid counting switches caused by sequencing errors or mitotic gene conversion events, which should not affect multiple adjacent sites or regions longer than 4 kb [59], respectively. The switching rate for the YJF1460 control was similar to that obtained using HAPCUT2 (S4 Table), which minimizes errors when merging reads but assumes a ploidy of two, and SDhaP [60] run assuming a ploidy of two for YJF1460 and four for the two ale strains. The second set consisted of alleles abundant in the four beer populations but absent in all others. After excluding strains with more than 1% admixture, there were 32,829 sites with allele frequencies over 25% in either the Ale 1 (n = 13), Ale 2 (n = 12), or Beer/baking strains (n = 2), but less than 1% in all other populations. To avoid problems with low-coverage strains, we estimated population allele frequencies from counts of homozygous calls and half of heterozygous calls.
Decay in linkage disequilibrium was measured by the covariance in alleles between sites [61]. An exponential decay function was fit to the average covariance of sites binned every 100 bp from 1kb to 50kb. Rather than weight linkage disequilibrium based on the allele frequency differences between the two admixed populations, we used the unweighted covariance across 34,022 sites that show nearly fixed differences between the Europe/wine and Asia/sake population. For the phased strains, we used the covariance across sites on the same haplotypes. For the population decay estimates, we only used strains with 99% or more ancestry assigned to either the Clinical, Laboratory, Ale 1, Ale 2, Beer/baking, and Lager populations. Invariant sites were excluded in each case. We assumed 0.34 kb/cM [35] to translate decay in physical distance to genetic distance and infer the number of meiotic equivalents.
We estimated divergence using four-fold degenerate sites in coding sequences. Excluding splice sites and sites with overlapping gene annotations, there were 1,036,317 four-fold degenerate sites surveyed. At these sites, we found 1586, 1040, 899, and 716 alleles at a frequency of 25% or more in the Ale 1, Ale 2, Beer/baking, or Lager population, respectively, but not in any other population.
|
10.1371/journal.pgen.1002410 | Ancestral Components of Admixed Genomes in a Mexican Cohort | For most of the world, human genome structure at a population level is shaped by interplay between ancient geographic isolation and more recent demographic shifts, factors that are captured by the concepts of biogeographic ancestry and admixture, respectively. The ancestry of non-admixed individuals can often be traced to a specific population in a precise region, but current approaches for studying admixed individuals generally yield coarse information in which genome ancestry proportions are identified according to continent of origin. Here we introduce a new analytic strategy for this problem that allows fine-grained characterization of admixed individuals with respect to both geographic and genomic coordinates. Ancestry segments from different continents, identified with a probabilistic model, are used to construct and study “virtual genomes” of admixed individuals. We apply this approach to a cohort of 492 parent–offspring trios from Mexico City. The relative contributions from the three continental-level ancestral populations—Africa, Europe, and America—vary substantially between individuals, and the distribution of haplotype block length suggests an admixing time of 10–15 generations. The European and Indigenous American virtual genomes of each Mexican individual can be traced to precise regions within each continent, and they reveal a gradient of Amerindian ancestry between indigenous people of southwestern Mexico and Mayans of the Yucatan Peninsula. This contrasts sharply with the African roots of African Americans, which have been characterized by a uniform mixing of multiple West African populations. We also use the virtual European and Indigenous American genomes to search for the signatures of selection in the ancestral populations, and we identify previously known targets of selection in other populations, as well as new candidate loci. The ability to infer precise ancestral components of admixed genomes will facilitate studies of disease-related phenotypes and will allow new insight into the adaptive and demographic history of indigenous people.
| Admixed individuals, such as African Americans and Latinos, arise from mating between individuals from different continents. Detailed knowledge about the ancestral origin of an admixed population not only provides insight regarding the history of the population itself, but also affords opportunities to study the evolutionary biology of the ancestral populations. Applying novel statistical methods, we analyzed the high-density genotype data of nearly 1,500 Mexican individuals from Mexico City, who are admixed among Indigenous Americans, Europeans, and Africans. The relative contributions from the three continental-level ancestral populations vary substantially between individuals. The European ancestors of these Mexican individuals genetically resemble Southern Europeans, such as the Spaniard and the Portuguese. The Indigenous American ancestry of the Mexicans in our study is largely attributed to the indigenous groups residing in the southwestern region of Mexico, although some individuals have inherited varying degrees of ancestry from the Mayans of the Yucatan Peninsula and other indigenous American populations. A search for signatures of selection, focusing on the parts of the genomes derived from an ancestral population (e.g. Indigenous American), identifies regions in which a genetic variant may have been favored by natural selection in that ancestral population.
| During the past decade, data generated by high-throughput genotyping technologies have enabled studies probing into two central questions in human evolutionary biology: the characterization of human population genetic structure, and the search for the molecular signature of natural selection. Insights gleaned from these studies have provided important clues for understanding the phenotypic diversity of our species, and variables representing population structure are routinely incorporated as covariates in genome-wide association studies of complex traits and diseases. At a global level, as well as within a continent or even a sub-continental region, geography has been shown to act as the leading driving force in shaping the pattern of genetic variation that we observe today [1]–[5]. In parallel, analyses based on European, African and East Asian populations have revealed that recent positive selection is a prevalent phenomenon throughout the genome [6]–[8]. Using data from the Human Genome Diversity-CEPH Panel (HGDP), a recent and comprehensive survey suggests that, while adaptation to local environment is a common theme throughout human evolution, the genetic loci involved in adaptation show little overlap among non-contiguous geographic regions [9].
While geography poses a significant reproductive barrier, multiple waves of massive trans-continental migration have occurred during the past centuries, giving rise to admixed populations. The ancestry of non-admixed individuals can often be traced to precise regions based solely on genetic data, but characterizing the sub-continental ancestry origins of an admixed individual has not been demonstrated to date. For example, the two largest minority groups in North America, Latinos and African Americans, both arose as a result of mating among populations that had been in historical reproductive isolation. The “Hispanic” or “Latino” populations include the ethnically diverse groups of Latin America; although significant genetic contributions can be traced to Indigenous American, European and West African populations, it has been challenging to determine whether one's Indigenous American ancestors originate from North, Central, or South America. Solving this problem has implications for both a deeper understanding of human evolution and for human disease, since genetic diversity between Latino populations is characterized both by variation in continent-level ancestry – e.g. Mexicans on average have lower African ancestry than Puerto Ricans – and by the population structure among the ancestral Indigenous American populations [4], [10].
The assessment of the precise ancestral origin and the quantification of genetic structure within an ancestry component are limited, in part, by analytic challenges. Principal Component Analysis (PCA) is a classic technique for multivariate data analysis, which aims to project high-dimensional data to a much lower dimension while capturing the greatest level of variation [11]. This approach has gained popularity in genetic analyses due to both computational efficiency and interpretability: when the underlying population structure is driven mainly by reproductive isolation and subsequent genetic differentiation, the principal components (PCs) mirror the geographic origins of individuals [3]. By itself, however, PCA is not well suited for studying admixed populations: while leading PCs usually represent the relative contributions of continentally-divided ancestral populations, subsequent PCs may be simultaneously influenced by structures within one or more of the ancestral populations, and are consequently difficult to interpret.
We tackled this problem by employing an analytic strategy that works backwards according to the temporal nature of demographic events that underlie human admixture: genomes are first separated into the major and most recent components that reflect inter-continental migration, then each of those components is further investigated separately. As described below, we apply a probabilistic method for inferring locus-specific ancestry along the chromosome, followed by a variant of PCA to further investigate each of the ancestry-specific genomic components, which we term “virtual genomes”. This hierarchical strategy yields a fine-scale view of genetic structure in admixed populations, and provides insight into the population history of nonextant ancestral populations. As an example, we study a cohort of 492 parent-offspring trios recruited from Mexico City. Our results confirm the a priori expectation that the most significant European contributors to the Mexican gene pool are populations from the Iberian Peninsula, but reveal that the Indigenous American component of the Mexican genomes is more complex.
Studying the genetic structure of admixed genomes also offers the unique opportunity to probe the adaptive landscape of the ancestral populations. This is particularly powerful for studying the Indigenous American populations, for which limited genotype data is available. As proof of principle, we report a novel application of the extended haplotype homozygosity test for recent positive selection to the European and Indigenous American “virtual genomes” evident in the Mexican cohort, and identify numerous loci as potential targets of positive selection.
Our analytic strategy for studying population structure in admixed populations is shown in Figure 1; details of the approach are described in what follows, and in the Materials and Methods section. This approach first applies a model-based clustering method, frappe, to the intact genotype matrix, identifying components that correspond to variation in continental-level admixture proportions, and estimating the relative proportion of those components for each individual. Locus-specific continental ancestry along a genome is then inferred using SABER+, an extension of a Markov-Hidden Markov Model method [12] that partitions each genome into ancestral haplotype segments or “virtual genomes”. Finally, within-continent population structure is determined by applying PCA to the virtual genomes, treating the rest of the genome as missing. To account for the large amount of the missing data resulted from the continent-specific genomes, we implement a variation of the subspace PCA (ssPCA) algorithm [13].
Most of the results described here are from a panel of 492 Mexican parent-offspring trios recruited from Mexico City (MEX1) as part of a previous genome-wide association study using genotype data from the Illumina 550K platform [14]. For comparison, we also examined data from 23 HapMap Phase3 Mexican trios recruited from Los Angeles, California (MEX2; http://hapmap.org). Reference populations for inferring continental-level ancestry were taken from HapMap (CEU, YRI), and additional sources as described below and in Table S1.
Among the 984 parents of the Mexico City trios (MEX1), we used frappe to estimate median ancestry proportions of 65% Indigenous American, 31% European, and 3% African; the corresponding statistics in the 46 HapMap Mexican individuals from Los Angeles (MEX2) are 45%, 49%, and 5%, respectively (Figure 2A). The distribution of Indigenous American ancestry in the Mexico City population is shifted upward compared to the Los Angeles population (Figure 2B), which may reflect differences in the extent of European admixture. African ancestry is low in both cohorts, although the distribution is skewed to the right, reaching over 40% for some individuals.
We next used SABER+ to estimate recombination breakpoints between ancestral chromosomes and thus locus-specific ancestral origin— Indigenous American, European, or African—in individuals from the MEX1 and MEX2 cohorts. For the work described here, the primary goal of SABER+ is to partition the Mexican genomes into haplotype segments according to continental ancestry that can be used for subsequent analysis. However, the output of SABER+ can also be used as an independent means of assessing global ancestry, simply by averaging locus-specific ancestries across all markers, and yields estimates that are highly correlated (r>0.99) with frappe (Figure S1).
To facilitate the analyses of sub-continental genetic structure, we constructed virtual genomes by retaining haplotype segments from a single continental-ancestral population, while masking (i.e. setting to missing) segments from all other ancestral populations; for example, MEX1AMR and MEX1EUR denote the sets of Indigenous American and European haplotype segments from the Mexico City individuals, respectively.
In a principal component analysis of this data that includes the YRI and CEU HapMap populations, the Indigenous American, European, and African virtual genomes mark vertices of a triangle (Figure 2C) in which the intact genomes of the MEX1 and MEX2 individuals are distributed broadly along an Indigenous American – European axis represented by PC1. The exact position of the intact MEX1 and MEX2 genomes depends on admixture proportions; individuals with the greatest level of African ancestry, which corresponds to PC2, mostly lie at intermediate positions along the Indigenous American -European axis. Importantly, the MEXEUR and MEXAFR virtual genomes (red and blue points, respectively) form discrete clusters whose locations coincide with those of the HapMap CEU and YRI, respectively, and, while there is no reference population in this analysis for Indigenous American, the MEXAMR virtual genomes also form a discrete cluster at a vertex of the triangle. These observations suggest that the ability of SABER+ to assign local ancestry to a specific continental origin is highly accurate, which is essential for subsequent analyses.
The distribution of the length of ancestry blocks is shaped by population history since admixture. When two individuals from different parental populations mate, the first generation offspring inherits exactly one chromosome from each parental population. In subsequent generations, recombination events in an admixed individual generate mosaic chromosomes of smaller ancestry segments. Intuitively, more recent admixing gives rise to longer ancestry blocks than older admixture. Furthermore, conditioning on the time since admixing within an individual's pedigree, block length distribution also depends on the individual level ancestry proportions: e.g., an individual with 90% European ancestry tends to have long European ancestral blocks because recombination events in the person's genealogy are likely to have joined two European haplotypes, and therefore fewer ancestry changes are expected.
A likelihood-based model has been proposed that can estimate several aspects of admixture history [15]. However, the admixing rates in Mexicans from the European, Indigenous American and African ancestral populations are likely dependent and difficult to model with this likelihood-based method; therefore, we attempted to estimate admixing time using a different approach. We first computed the theoretical number of ancestry blocks for individuals according to their ancestral proportions, and carried out that computation assuming a series of different admixing times (5–25 generations, dotted lines in Figure 2D). The parabolic shape of these curves conforms to the intuitive idea outlined above that the number of block peaks at an intermediate ancestry proportion.
We then superimposed the observed number of ancestry blocks in each MEX1 individual onto the theoretical curves; these results suggest an admixing time of 10–15 generations ago (Figure 2D). The admixing time of the European component appears slightly longer than that for the Indigenous American component (15 generations vs. 12); one potential explanation is that some mixing occurred between the European and the African ancestral individuals prior to admixing with the Indigenous American populations.
With the MEXEUR uncoupled from the MEXAMR genomes, we investigated structure within each of these virtual genomes separately. (We did not investigate the MEXAFR virtual genomes due to their small sample size). Because there is a large amount of missing data, e.g. the virtual genome of one individual may cover very different loci from the virtual genome of other individuals, we used the ssPCA approach as described in Materials and Methods. To help evaluate the robustness of our approach, we carried out simulation experiments, in which the effects of random error in the inference of continental locus-specific ancestry were measured with regard to their impact on accuracy of within-continent substructure estimates. Results summarized in Materials and Methods and Figure S2 indicate that European substructure can be well separated in the presence of up to 5% error, i.e. Indigenous American alleles mistakenly included in the European virtual genomes, which is well above the level of uncertainty (<2%) associated with the SABER+ approach.
In addition to the HapMap CEU, who are mostly of Northern European ancestry, we used individuals recruited from Dublin, (Ireland), Warsaw (Poland), Rome (Italy) and Porto (Portugal) to provide references for different areas within Europe. The first two PCs provide good separation of these reference populations, and correspond roughly to North-South and West-East gradients (Figure 3A). Both the MEX1EUR and MEX2EUR virtual genomes are most closely related to intact genomes from Porto, which we interpret as a surrogate for populations from the Iberian Peninsula, [3], consistent with the historical record that the first European migrants to Mexico were Spaniards.
For analysis of the MEXAMR virtual genomes, we introduced 129 individuals representing 8 different Indigenous American populations as reference genomes (Table S1) [16]. Initially, we also included the HapMap CEU based on previous results in which some Indigenous American individuals from the Human Genome Diversity Panel (HGDP) were observed to have non-negligible levels of European ancestry [2]. Indeed, the first two PCs for this analysis occur along European-Indigenous American and within-America axes (Figure 3B), and reveal varying levels of European ancestry in the Mayan, Quechua and Colombian populations. In this analysis and subsequent ones carried out in which certain reference populations were removed (CEU removed from Figure 3C; CEU, Surui, Karitiana and Pima removed from Figure 3D), the MEXAMR virtual genomes are most closely related to intact genomes of individuals from southwestern Mexican state of Guerrero (Guerr), which includes Nahua, Mixtec and Tlapanec indigenous groups. Although the Guerrero individuals and the Pima individuals cluster together in Figure 3C, they are separable on PC 3 (Figure S3), along which the Guerrero, but not Pima individuals, cluster with MEXAMR. The Indigenous American virtual genomes of Mexicans from Mexico City (MEX1AMR) are similar to those from Los Angeles (MEX2AMR); further, we observe a gradient with varying contribution from Mayans, with some Mexicans deriving their Indigenous American ancestry predominantly from Mayans. One individual from Mexico City has an Indigenous American virtual genome that is localized with the Quechua (arrow, Figure 3C) and therefore is likely to have a source of Indigenous American ancestry that is distinct from that of the other Mexicans.
The ability to accurately construct ancestral virtual genomes from admixed genomes provides a number of opportunities in the areas of human evolution and genetic anthropology. As an example of how such data can be used more generally, we examined the Mexican ancestral components for regions of extended haplotype homozygosity, which mark loci that have undergone recent positive selection. We used the integrated haplotype score (iHS) statistic [8], with a modified normalization procedure so as to fit a standard normal distribution.
For the virtual genome SNPs that show the strongest evidence of positive selection, the degree of overlap between the Europeans and Indigenous Americans is similar to that expected by chance (Figure 4A). Specifically, we considered SNPs with |iHS|>2.5, which represent approximately the top 1% scores in either components; 3874 and 3931 SNPs meet this criterion in MEXAMR and MEXEUR, respectively, with 57 SNPs overlap between the two sets (expected overlap = 40, p = 0.094). Similarly, we found little overlap between the iHS scores in MEXAMR and those computed based on the HapMap populations [8] (Figure 4B). In contrast, the correlation is much higher between the iHS in MEXEUR and those from HapMap CEU (r = 0.79), which reflects shared population and adaptive histories of Southern Europeans (the MEXEUR) and the CEU (mostly from Northern and Central Europe). Specifically, of 3257 and 3460 SNPs with |iHS|>2.5 in MEXEUR and CEU, respectively, 655 are overlapping (expected overlap = 32, p<2.2−16) (Figure 4C). These findings are consistent with previous observations that intact genomes from the HGDP collection exhibit histories of positive selection that differ according to continent [9].
We also asked what genes might underlie the strongest signatures of positive selection. Towards this end, we grouped SNPs into 50kb windows, selected regions with at least 20 SNPs and at least 10% of SNPs with |iHS|>2.5, and ranked windows by the maximum |iHS| score. The top 10 regions within the MEXEUR and MEXAMR components are shown in Table 1 and Table 2. The only genomic location that features in both lists is the HLA region on chr 6p, a region known to have experienced strong selection [17]; however, the precise variants that show high iHS scores differ between the Europeans and Indigenous Americans. Outside the HLA region, the most prominent signal in the European component coincides with APBA2 on chr 15q, which is in close proximity to a known pigmentation gene, OCA2. In the Indigenous Americans component, the strongest signal occurs in chr 6p12.3-2; this region harbors numerous genes, including IL17A which is associated with chronic inflammatory diseases such as rheumatoid arthritis, and PKHD1 which is associated with polycystic kidney disease [18], [19].
Previous approaches to analyzing the ancestry in admixed individuals have largely focused on estimating continental-level admixture proportions. Within continental ancestry analyses have been performed at a population level but not at an individual level. Our approach is distinct in several ways from a recent study, which reports the affinity, at a population level, of admixed populations to various ancestral groups [20]. This latter approach requires a pre-defined notion of subpopulation, such as Mexicans versus Puerto Ricans, and produces a population level summary of genetic relationship. In contrast, our approach does not rely on such pre-defined ethnic groups, and thus has the ability to identify previously unrecognized substructure at an individual level, such as the detection of one individual with South American ancestry. In what follows, we first discuss aspects of the approach that may be generally relevant, and then provide some insights in the evolutionary history of the Mexican population. In the context of genome-wide association studies of complex traits and diseases, variables representing both continental-level and within-continent population structure need to be adjusted to provide a more accurate correction for population stratification [21], [22].
Two methodological innovations contributed to the hierarchical depiction of individual ancestry origin: an improved algorithm for locus-specific ancestry inference, which accommodates multiple ancestral populations, and a subspace PCA algorithm that permits varying degrees of missing data. SABER+ uses a graphical model to account for haplotype structure within an ancestral population, and is more accurate for analyzing high-density genotype data. The accuracy of the locus-specific ancestry is supported by two observations. First, in the continental-level PCA analysis (Figure 1A), all “virtual genomes” that are attributed to a single ancestral population, cluster tightly with the reference individuals. Had there been substantial error in the local ancestry inference, some of these genomes would appear admixed and lie in between the vertices. Second, we included the HapMap CEU individuals in the analysis of the Indigenous American components of the genomes because some of the Mayan individuals have been shown to have European admixture [2]. Indeed, although Figure 3B clearly reveals European admixture in some Mayan and Quechua individuals, little European admixture is detected in the putative Indigenous American genomes of the Mexicans, MexAMR. We note that, although many methods for estimating local ancestry, including the method used in this study, are applicable to unphased data, the parent-offspring trio structure of the Mexican data (both the Mexico City cohort and the HapMap Mexican sample) allows accurate haplotype phasing for each individual, which likely improves the accuracy of inference of local ancestry along each haplotype. As a result, both phasing and ancestry inference are likely more accurate than those estimated based on unphased genotype data.
Typically, application of PCA for genetic structure analyses makes use of the program Eigenstrat, which is based on the eigen-decomposition of the covariance matrix, , where G′ is the transpose of the centered and scaled genotype matrix, G [21]. In computing this covariance matrix, missing genotypes are set to the column means; thus Vij is computed based on genotypes that are non-missing in both individuals i and j. While this approach is adequate for analyses based on high-density genotypes with very low levels of missing genotypes, it is not appropriate for analyzing the virtual genomes, which feature large and varying proportions of missing data. Consider two individuals each with 30% ancestry from the population of interest (e.g. Europe): within each individual, 9% of the genome is expected to be homozygous in European ancestry, and therefore <1% of the markers are expected to be non-missing in both genomes after excluding non-European genotypes. This leads to two problems: first, it has reduced power for detecting population substructure because it uses only a small fraction of informative genotypes for the continent of interest; more importantly, the sampling variability of the covariance estimates depends heavily on the proportion of missing genotypes, biasing the PCs such that individuals with high missing rate become outliers along each PC. The ssPCA we implement does not require the computation of the covariance matrix, and uses all informative markers in each genome; hence it is less sensitive to the missing data. Since the algorithm can compute the first k PCs without computing all PCs, it also has computational advantages over the current covariance-based implementation, especially when the number of individuals is large.
We find extensive variation with respect to continental-level ancestry proportions, both between geographic regions – shown by the much higher Indigenous American ancestry in the Mexico City cohort compared to the HapMap Mexican Americans from Los Angeles – and between individuals within each cohort. This study benefits from the ability to divide the genome of a single Mexican individual into its constituent ancestral components. The ability to trace chromosomal segments to their respective ancestral populations allows us to scrutinize the ancestry origin of each individual within a continent. Within the European component of the Mexican genomes (MexEUR), nearly all individuals, both from Mexico City and from Los Angeles, trace their European ancestries to a Southern European population, as represented in our study by the Portuguese. Within the Indigenous American component of the genomes (MexAMR), a majority of individuals trace their ancestries to groups from the southwest coastal regions of Mexico, consistent with a previous study, which found Zapotec individuals from the State of Oaxaca to best approximate the Indigenous American ancestral population for Mestizos [23]. Importantly, we find evidence of varying levels of Mayan admixture, as well as one individual with Indigenous American ancestry from Bolivia/Peru. Of note, individuals with high levels of Mayan or South American ancestries do not stand out in the continental-level PCA, as their continental-level ancestry proportions are comparable to the rest of the Mexicans.
The finding that most Mexican individuals trace their European and Indigenous American ancestry to well-defined geographic regions contrasts sharply the lack of structure in the African ancestry in the African Americans: not only did we trace each African American individual to multiple West/Central West African groups, but the relative proportions are nearly constant across all individuals [24]. This difference can be reconciled by the distinct migratory histories: the African ancestry in African-American populations is largely derived from the trans-Atlantic slave trade, which forcibly departed African individuals from various geographic regions of Western Africa, ranging from Senegal to Nigeria to Angola [25]. In contrast, no evidence suggests massive relocation of the Indigenous Americans during the colonization in North America, and hence reproductive isolation likely has been maintained between geographically separated Indigenous American populations. One limitation of the current study is the incomplete sampling of the Indigenous American populations in our reference panel, which represents two distinct regions in Mexico: the Southwest coastal State of Guerrero and the Yucatan Peninsula. Thus, while most Mexicans trace their Indigenous American ancestries to the indigenous groups from the State of Guerrero (Guerr), it is possible that the true ancestors of the extant Mexicans are an un-sampled group that is genetically similar to Guerr. With the coming of whole genome sequencing data, it is possible that indigenous populations from neighboring states can be distinguished, and thus it may even be possible to detect admixture from closely related Indigenous American groups.
The EHH analyses of the Southern European and the Indigenous American components of the Mexican genomes separately revealed numerous intriguing putative targets of recent positive selection. We note that many other approaches have been developed to detect specific types of selective events, and are equally applicable [26], [27]. We chose to use the iHS test because it has been applied to both the HapMap dataset and the HGDP dataset, thus facilitating comparison. The goal of this paper is not to conduct a comprehensive survey of the selective landscape in the ancestral populations of the present day Mexicans, but rather to illustrate the potential benefits of such endeavors. Given the difficulties in recruiting large samples of non-admixed indigenous individuals from each well-defined Indigenous American group, we argue that admixed populations will provide valuable insight in future endeavors in understanding the evolutionary histories of the Indigenous American populations, some of which may have been extinct. For example, individuals with full Taíno ancestry are rare, but approximately 15% of the contemporary gene pool of Puerto Ricans may have been derived from Taínos. Hence, admixed Puerto Rican genomes can be used to learn about those of the ancestral Taínos [28]. We note that this approach of assembling an ancestral population from a mixed population has also provided important insights in the Aboriginal Australian population in a recent study [29].
Distinguishing between selective events that occurred within the ancestral populations and those that occurred post-admixing requires careful consideration of the tests and associated assumptions. In the current setting, we reasoned that, since a novel adaptive allele is unlikely to be swept to a substantial frequency within a period of less than 500 years (since the arrival of the Europeans in Mexico), and since the EHH method does not have appreciable power to detect low frequency adaptive alleles [9], most of the signals detected by the EHH had occurred prior to admixing, and hence represent selection within the ancestral populations. On the other hand, the preservation of a long haplotype excludes the possibility of very ancient selective events; this belief is also supported by the observation that there is little overlap between the signatures detected in the Southern European and the Indigenous American components. In previous studies of Puerto Ricans and African Americans, numerous genomic locations were found where locus-specific ancestry deviate from the genome-wide average, and could represent targets of selection in the admixed populations [30], [31]. In the current analyses, the only locus showing deviation from the genome-wide average is the HLA region on chr 6, again supporting a population-specific pattern of selection. Therefore, the adaptive history of the Indigenous American groups may vary considerably, and should be studied separately and not as a whole group. Such analyses can be achieved, for example, by examining the Indigenous American components in Mexicans versus that of Puerto Ricans.
Our results have important implications for the design of genome-wide association studies based on admixed populations. Epidemiologic studies have found varying prevalence of conditions such as asthma, diabetes and alcohol-related problems across Hispanic national groups [28], [32], [33]. Distinct population and adaptive history among Hispanics ethnic groups can give rise to heterogeneity in complex traits. Therefore, the importance of accounting for intra-continental genetic structure in disease mapping studies, in addition to adjusting inter-continental admixture proportions, needs to be carefully evaluated.
The Mexican individuals analyzed in this project come from two sources: a panel of 492 Mexican parent-offspring trios recruited from Mexico City as part of a previous genome-wide association study (MEX1) [14], and 23 HapMap Phase3 Mexican trios recruited from Los Angeles, California (MEX2; http://hapmap.org). For estimating locus-specific ancestry, we used the HapMap CEU (N = 88) and YRI (N = 100) individuals for the ancestral populations. To analyze the European component of the admixed genome, we augmented the Mexican datasets with individuals recruited from Dublin, Ireland (N = 43), Rome, Italy (45), Warsaw, Poland (N = 45) and Porto, Portugal (N = 43). For the Indigenous American component analyses, we combined the data generated in two previous studies [2], [16]. Four Mayan individuals with substantial European admixture are removed. The combined set used for the subsequent analyses includes 14 individuals from Guerrero, Mexico (two Nahua, seven Mixtec and five Tlapanec), 24 Mayan individuals from the Yucatan Peninsula, 24 Quechua collected in Cerro de Pasco, Peru, 25 individuals of largely Aymara ancestry collected in La Paz, Bolivia, 13 Karitiana and eight Surui from Brazil, seven Colombians, and 14 Pima. Because the sample sizes for Nahua, Mixtec and Tlapnec are small, and all individuals were recruited from the same state, we considered these individuals as one group. Table S1 summarizes the individuals used for each analysis.
Genotyping and quality control procedures have been described in the primary publications for each dataset, except for the dataset of 176 European individuals. Briefly, MEX1 and the HGDP individuals were genotyped on Illumina 550K and on 650K Beadchip, respectively. The Indigenous American individuals from Bigham et al. (2009) were genotyped on Affymetrix 1M SNP arrays [16]. The set of 176 European individuals were genotyped using Illumina HumanHap300 arrays; this dataset originally included 180 individuals; four individuals were found with non-negligible non-European ancestry and were excluded. SNPs with a call rate of less than 95% were excluded. The number of individuals and markers used for each analysis is summarized in Table S1. We used BEAGLE to construct haplotypes for Mexican trios [34]. As children provide no additional information regarding population structure or adaptation, they are not used in subsequent analyses.
Continental-level admixture proportions were estimated two ways: (1) a model-based clustering algorithm implemented in frappe [35], and (2) average locus-specific ancestries across all markers. Locus-specific ancestry was estimated with SABER+, an extension of a previously described approach, SABER, that uses a Markov-Hidden Markov Model [12]. SABER+ differs from SABER in implementation of a new algorithm, an Autoregressive Hidden Markov Model (ARHMM), in which haplotype structure within the ancestral populations is adaptively constructed using a binary decision tree based on as many as 15 markers, and which therefore does not require a priori knowledge of genome-wide ancestry proportions (Johnson et al., in preparation). In simulation studies, the ARHMM achieves accuracy comparable to HapMix [36] but is more flexible in modeling the three-way admixture in the Mexican population and does not require information about the recombination rate.
HapMap CEU and YRI individuals were used as the reference ancestral populations. Based on frappe and supported by PCA, 50 individuals in MEX1 set have more than 95% Indigenous American ancestry. These individuals were initially used to approximate the Indigenous American ancestors in the locus-specific ancestry analyses; an iterative procedure is used to identify and correct for the non-Indigenous American segments in these individuals. Accuracy of the locus-specific ancestry is verified by performing a PC analysis, treating each individual as three non-admixed genomes, MexEUR, MexAMR, and MexAFR (see section “subspace PCA” below).
We implemented this algorithm to accommodate the large amount of missing genotype data in partially masked virtual genomes, and used it to derive all the PCA results reported here. The statistical theory of the algorithm in a general data mining context can be found in [13]; however, various modifications are required for the current setting, as described below. Let Gh (h = 1,2) be two N×M matrices, in which denote the unordered pair of alleles at SNP m (m = 1,…,M) in individual n (n = 1,…,N); the columns of Gh are standardized to have mean 0 and variance 1. To compute the subspace spanned by the first k principal components (PC), we begin by finding a matrix decomposition, , which minimizes the reconstruction error, R, defined as:subject to the constraints that the column vectors of A are of unit norm and mutually orthogonal and the row vectors of S are also mutually orthogonal. Here A is a N×d matrix, S is a M×d matrix, and d<N≤M represents the desired number of leading PC's. The algorithm we use is a generalized instance of the coordinate descent approach [37], which iteratively optimizes matrix A for fixed S and then optimizes S fixing A according to the rules:where λ is a learning rate, the superscripts, r, indicate iteration, and the subscripts, j, denote the j-th column of a matrix. It can be shown that the columns of A and S span the subspace of the first d PCs, and that the leading PCs can be computed by orthogonalizing the columns of A and S [13]. To evaluate the accuracy of our modified ssPCA approach, we applied it in parallel with Eigenstrat [21] to the intact Mexican genomes, and found the leading PCs produced by the two algorithms were virtually identical, up to a permutation of signs.
We carried out two simulation experiments to evaluate the impact of statistical uncertainties associated with estimating locus-specific ancestry, and to investigate the performance of the ssPCA approach.
In the first set of simulations, we created 10 datasets in which 400 admixed genomes were modeled to mimic a Latino population: each individual draws chromosomal segments from European and Indigenous American ancestry, and the proportion of Indigenous American ancestry in each individual matches what we observed in MEX1. For 200 individuals, European-derived segments were sampled from the HapMap CEU haplotypes, representing Northern and Western European ancestry, while for the remaining 200 individuals, the European-derived segments were sampled from MexEUR inferred from the actual Mexican genotype data, representing Southern European ancestry. The chromosomal segments from CEU and MexEUR in the admixed individuals were treated as the true European virtual genomes.
To evaluate the potential impact of statistical uncertainty, we introduced random errors in which the true identities of European vs. Indigenous American segments were switched with probability ε. The top PC for each set of simulated virtual genomes (at each of 8 error rates, ε = 0.01–0.20) was computed with ssPCA. We evaluate the effect of these errors by calculating a confusion fraction, ξ, that quantifies the accuracy with which the estimated first PC separates individuals with Northern vs. Southern European ancestry, and is defined as the proportion of individuals that lie on the “wrong” side of a threshold that best separates the two groups. Thus, ξ can range from 0 (perfect separation) to nearly 50% (complete confusion as would be observed for genetically homogenous groups). Finally, we analyze each of the 10 datasets using SABER+, exactly as was done for real data: apply ssPCA to estimate substructure, and calculate a confusion fraction. The results for this simulation experiment are depicted in Figure S2, and show that the confusion fraction increases substantially, from a mean of 2.14% to 17.5%, at error rates between 0.03 and 0.05. Using SABER+ on these same 10 datasets yields a mean confusion fraction of 1.58%, which corresponds to an error rate <0.02 (indicated by the arrow in Figure S2).
In a second set of simulations to investigate the ability of the ssPCA approach to deal with missing data, we created five datasets in which the proportion of genome-wide European ancestry in each of 400 admixed genomes was fixed at either 50% or 30%, respectively. Applying SABER+ and ssPCA to these datasets yields mean confusion fractions of 0 and 0.7%, respectively, indicating that our approach performs well for situations such as the one described here, where mean genome-wide continental ancestry proportions are above 30% for both the European and the Indigenous American components.
We used the number of ancestry blocks in an individual as summary statistics. Tracing through a pedigree of T generations, the expected number of recombination events in a haploid genome is 0.01×TL, where L is the total genome length (taken to be 3435cM [38]). Under a hybrid-isolation model and assuming a genome-wide ancestry proportion of z, a fraction of 2×z(1−z) of the recombination events occurs between two haplotypes of opposite ancestry and thus leads to transitions in ancestry. When we count the number of ancestry blocks in the real data, we do not observe recombination events that occur between two haplotypes of the same ancestry. Hence the expected number of ancestry switches in a diploid genome is B = (2×2×0.01)×TL×z(1−z), and each ancestry switch creates one additional ancestry block. When there is no ancestry switch in a genome, the number of ancestry blocks is defined to be the same as the number of chromosomes. Therefore, for each specific time of admixing, T, we computed the expected number of ancestry blocks as B+2×22, with the genome-wide ancestry proportion, z, varying from 0 to 1 at 100 equally spaced grid points. Each curve in Figure 2D shows the expected number of ancestry blocks as a function of admixture proportions for a specific admixing time. The estimated numbers of European and non-European ancestry blocks from the Mexican individuals were tallied and compared to the expected values. To assess the impact of uncertainty associated with estimating the number of ancestry blocks, we note that errors in estimating locus-specific ancestry often create very short ancestry blocks. Hence, we simulated admixed genomes according to the hybrid-isolation model, but removed extremely short blocks (segments with <10 SNPs) from both simulated genomes and real data. The estimated admixing time remained the same under this alternative analysis, suggesting the estimated admixing time is relatively robust. The hybrid-isolation model was chosen because of the mathematical simplicity; under a more realistic continuous gene-flow model, the estimated times of admixing should be interpreted as an approximation of average admixing time, weighted by the relative level of gene-flow in each generation.
To assess the sub-continental population structure, each Mexican genome was partitioned into three non-admixed genomes, by masking (i.e. setting to missing) alleles from all but one ancestral population. In other words, the European component of a Mexican's genome (MexEUR) was derived by treating as missing all alleles whose origins were inferred as African or Indigenous American. For within European analysis, we applied ssPCA on the dataset consisting of MexEUR (including both Mexico City and HapMap samples), 88 HapMap CEU and 176 European individuals from four cities: Dublin (Ireland), Warsaw (Poland), Rome (Italy) and Porto (Portugal). Because of the limited number of informative haplotype segments, Mexican individuals with less than 25% European ancestry were excluded from this analysis. In an analogous fashion, we analyzed the Mexican component of the genome, MexAMR, along with 129 indigenous Indigenous American individuals representing 8 populations (Table S1). Because the African ancestry is low in both Mexican cohorts (3% and 5%, respectively), we did not analyze the within-Africa population structure.
or all SNPs with frequencies between .05 and .95 in the respective populations, iHS was calculated following Voight et al. [8] with two modifications. First, haplotype homozygosity scores for a core SNP is computed on the subset of haplotypes in which the core SNPs are derived from a specific population. If a haplotype is truncated because of an ancestry change, the haplotype beyond the ancestry switch point is considered different from all other haplotypes in the corresponding interval. We have also considered an alternative strategy in which only haplotypes that do not have an ancestry change within 400 SNPs from the core SNPs are included in the calculation; the results are virtually identical. Second, instead of binning SNPs by the inferred ancestral allele frequencies and calculating the standard deviation of iHS in each bin, we used a quantile regression to estimate the 25th- and 75th-percentile of the empirical null distribution as a function of the minor allele frequency. The raw iHS scores were then normalized by the estimated inter-quartile range within each chromosome; the resulting standardized iHS scores fit a standard normal distribution well. To define regions that may harbor recently adaptive alleles, we seeded a region by a window of 50kb around a SNP with extreme iHS scores; we then successively scanned to the left and to the right, 50kb a time, merging neighboring regions in which at least one SNPs has an |iHS|>2.5 (which represents the 99th-percentile of the scores); finally, the top 10 list in Table 1 requires that at least 10% of the SNPs in the region have |iHS|>2.5. The proportion of SNPs with high |iHS| was the criterion used by Pickrell et al. [9]. It has been suggested that genome-wide and locus-specific ancestries may show particular poor correlation at loci under selection compared to neutrally evolving loci [39]. We did not observe such a trend at loci with the highest iHS scores in either MEXAMR or MEXEUR.
|
10.1371/journal.pbio.1001229 | Alternative Splicing of RNA Triplets Is Often Regulated and Accelerates Proteome Evolution | Thousands of human genes contain introns ending in NAGNAG (N any nucleotide), where both NAGs can function as 3′ splice sites, yielding isoforms that differ by inclusion/exclusion of three bases. However, few models exist for how such splicing might be regulated, and some studies have concluded that NAGNAG splicing is purely stochastic and nonfunctional. Here, we used deep RNA-Seq data from 16 human and eight mouse tissues to analyze the regulation and evolution of NAGNAG splicing. Using both biological and technical replicates to estimate false discovery rates, we estimate that at least 25% of alternatively spliced NAGNAGs undergo tissue-specific regulation in mammals, and alternative splicing of strongly tissue-specific NAGNAGs was 10 times as likely to be conserved between species as was splicing of non-tissue-specific events, implying selective maintenance. Preferential use of the distal NAG was associated with distinct sequence features, including a more distal location of the branch point and presence of a pyrimidine immediately before the first NAG, and alteration of these features in a splicing reporter shifted splicing away from the distal site. Strikingly, alignments of orthologous exons revealed a ∼15-fold increase in the frequency of three base pair gaps at 3′ splice sites relative to nearby exon positions in both mammals and in Drosophila. Alternative splicing of NAGNAGs in human was associated with dramatically increased frequency of exon length changes at orthologous exon boundaries in rodents, and a model involving point mutations that create, destroy, or alter NAGNAGs can explain both the increased frequency and biased codon composition of gained/lost sequence observed at the beginnings of exons. This study shows that NAGNAG alternative splicing generates widespread differences between the proteomes of mammalian tissues, and suggests that the evolutionary trajectories of mammalian proteins are strongly biased by the locations and phases of the introns that interrupt coding sequences.
| In order to translate a gene into protein, all of the non-coding regions (introns) need to be removed from the transcript and the coding regions (exons) stitched back together to make an mRNA. Most human genes are alternatively spliced, allowing the selection of different combinations of exons to produce multiple distinct mRNAs and proteins. Many types of alternative splicing are known to play crucial roles in biological processes including cell fate determination, tumor metabolism, and apoptosis. In this study, we investigated a form of alternative splicing in which competing adjacent 3′ splice sites (or splice acceptor sites) generate mRNAs differing by just an RNA triplet, the size of a single codon. This mode of alternative splicing, known as NAGNAG splicing, affects thousands of human genes and has been known for a decade, but its potential regulation, physiological importance, and conservation across species have been disputed. Using high-throughput sequencing of cDNA (“RNA-Seq”) from human and mouse tissues, we found that single-codon splicing often shows strong tissue specificity. Regulated NAGNAG alternative splice sites are selectively conserved between human and mouse genes, suggesting that they are important for organismal fitness. We identified features of the competing splice sites that influence NAGNAG splicing, and validated their effects in cultured cells. Furthermore, we found that this mode of splicing is associated with accelerated and highly biased protein evolution at exon boundaries. Taken together, our analyses demonstrate that the inclusion or exclusion of RNA triplets at exon boundaries can be effectively regulated by the splicing machinery, and highlight an unexpected connection between RNA processing and protein evolution.
| The split structure of eukaryotic genes impacts gene expression and evolution in diverse ways. Most directly, the presence of introns enables multiple distinct mRNA and protein products to be produced from the same gene locus through alternative splicing, which is often regulated between tissues or developmental stages [1],[2]. Alternative inclusion or exclusion of exons—“exon skipping”—can generate protein isoforms with distinct subcellular localization, enzymatic activity or allosteric regulation, and differing, even opposing, biological function [3]–[5]. Splicing is often regulated by enhancer or silencer motifs in the pre-mRNA that are bound by splicing regulatory proteins that interact with each other or with the core splicing machinery to promote or inhibit splicing at nearby splice sites [6]. Such enhancer and silencer motifs are common throughout constitutive as well as alternative exons and their flanking introns [7]–[9]. In turn, the presence of splicing regulatory motifs in exons, and their higher frequency near splice junctions, impacts protein evolution. For example, the frequencies of single nucleotide polymorphisms (SNPs) and amino acid substitutions are both reduced near exon-exon junctions relative to the centers of exons as a result of selection on exonic splicing enhancer motifs [10],[11]. Thus, a gene's exon-intron structure and its evolution are intimately linked.
Alternative 3′ and 5′ splice site use, in which longer or shorter versions of an exon are included in the mRNA, are among the most common types of alternative splicing in mammals [1] and can generate protein isoforms with subtly or dramatically differing function. For example, production of the pro-apoptotic Bcl-xS or the anti-apoptotic Bcl-xL protein isoforms is controlled through regulated alternative splice site usage [12]. Binding of splicing regulatory factors between the alternative splice sites or immediately adjacent to one site or the other can shift splicing toward the (intron-) proximal or distal splice site [6],[13],[14], providing a means to confer cell type-specific regulation. The distance between the alternative splice sites can vary over a wide range, from hundreds of bases to as few as three bases in the case of NAGNAG alternative 3′ splice sites.
NAGNAG alternative splicing (Figure 1A) has been observed in vertebrates, insects, and plants, and is known to be very common. Bioinformatic analyses of expressed sequence tag (EST) databases have identified thousands of examples [15]–[18]. However, most of the mechanisms known to regulate other alternative 3′ splice site pairs, particularly those that involve binding of regulatory factors between the sites, or much closer to one site than the other, cannot apply to NAGNAGs because of the extreme proximity of the two sites. Thus, regulation of NAGNAGs is more difficult to envisage. Furthermore, analyses of select genes using PCR and capillary electrophoresis approaches reached differing conclusions about NAGNAG tissue specificity [15],[19],[20], and several authors have argued that NAGNAG splicing is purely stochastic, is not evolutionarily conserved, and is not physiologically relevant [21],[22]. However, analyses of NAGNAG splicing at a genome-wide scale have been hampered by the impracticality of distinguishing such similar isoforms by microarray hybridization and the insufficient depth of EST databases for assessment of tissue specificity.
In order to assess the abundance and potential regulation of NAGNAG splicing events genome-wide, we analyzed polyA-selected RNA-Seq data generated using the Illumina HiSeq platform from 16 human tissues at depths of ∼8 Gbp per tissue, similarly deep RNA-Seq data that we generated from eight mouse tissues, and data generated by the modENCODE consortium across a developmental time course in Drosophila. NAGNAG isoforms can be uniquely distinguished by short reads that overlap the splice junction, and the quantity of data available from each tissue in human and mouse typically represented at least 80-fold mean coverage of the transcriptome, a depth sufficient to detect potential tissue-specific differences in many cases. Sequence features were identified which can shift splicing toward the proximal or distal NAG, providing clues to regulation. We also analyzed the impact of NAGNAGs on exon evolution, obtaining evidence that NAGNAGs dramatically accelerate addition and deletion of sequence at the beginnings of exons.
Our initial analyses used the Illumina Body Map 2.0 dataset of polyA-selected RNA-Seq data from 16 human tissues (adipose, adrenal, brain, breast, colon, heart, kidney, liver, lung, lymph node, ovary, prostate, skeletal muscle, testes, thyroid, and white blood cells) sequenced at depths of ∼80 million paired-end 2×50 bp reads per tissue. This sequencing depth generates ∼8 Gbp of data, representing >80-fold coverage of the human protein-coding transcriptome. Enumerating all possible NAGNAG splicing events, we mapped both ends of each read against NAGNAG splice junctions (Figure 1A). Isoform ratios were estimated across all tissues as “percent spliced in” (PSI or ψ) values (Figure 1B), representing the fraction of mRNAs that use the intron-proximal splice site, thereby including the second NAG in the mRNA. The reliability of such RNA-Seq-based estimates of isoform abundance has been established previously [23].
Using a conservative approach that has comparable power to detect each of the major types of alternative splicing events, we estimated that NAGNAGs comprise slightly more than 20% of reading frame-preserving alternative splicing events in coding regions, making NAGNAGs the most common form of protein-producing alternative splicing after exon skipping (Figure 1C). In all, more than 2,000 NAGNAG events were detected in protein-coding regions of human genes where both isoforms were expressed at ≥5% in at least one tissue. Strikingly, 73% of these NAGNAGs showed evidence of tissue-specific regulation (p<0.01 by multinomial test). Furthermore, approximately 42% were “strongly regulated,” with changes in ψ of at least 25% between tissues (Table S1). For example, a NAGNAG in the gene encoding FUBP1, a transcriptional regulator of MYC, undergoes dramatically different splicing between kidney and lymph node (Figure 1B). Here, we report absolute rather than relative differences in splicing levels, e.g., a change from 10% to 35% between tissues is considered an increase of 25%, not 250%, and the largest difference in ψ between tissues is defined as the “switch score” [1]. Other genes containing NAGNAGs with switch scores of 50% or more included HOXD8, CAMK2B, ATRX, CAPRIN2, and MLLT4 (a complete list of human genes containing alternative NAGNAGs, sorted by switch score, is provided in Table S2). Technical replicates—sequencing of the same RNA-Seq libraries with 75 bp single-end reads at depths of ∼50 million reads per tissue—yielded similar estimates of NAGNAG abundance and regulation (Table S3).
Regulation that contributes to fitness is expected to be evolutionarily conserved. A previous study reported the existence of selection against alternatively spliced NAGNAGs in coding sequences [24]. Nevertheless, some NAGNAGs are quite deeply conserved, e.g., a NAGNAG that generates an arginine insertion/deletion in a RNA-binding domain of the splicing factor PTBP2 (also known as nPTB or brPTB). Both isoforms of this NAGNAG event are observed in ESTs from human, mouse, and chicken, and the potential for alternative splicing is conserved at the sequence level to lizard (Figure 1D). Consistent with this example, a previous analysis of EST databases suggested that a subset of alternatively spliced NAGNAGs are under purifying selection in vertebrates [25]. We systematically assessed the global conservation of NAGNAG isoform levels using RNA-Seq data generated from eight mouse tissues (brain, colon, kidney, liver, lung, skeletal muscle, spleen, and testes). Restricting to the set of NAGNAGs which were alternatively spliced in human (both isoforms expressed at ≥5% in at least one tissue), we found that NAGNAGs which were strongly regulated were approximately 10 times more likely than unregulated NAGNAGs to exhibit alternative splicing in their mouse orthologs, and vice versa (Figure 1E). This large and consistent increase in conservation of alternative splicing with increasing switch score suggests that regulated NAGNAGs are much more likely to contribute to organismal fitness, and therefore to be selectively maintained, than are alternatively spliced events which do not exhibit tissue specificity. If NAGNAG alternative splicing were selectively neutral, then we would not expect to see a correlation between the observed degree of tissue specificity in one species and conservation of alternative splicing in the other species.
NAGNAG isoform levels were very well correlated between biological replicates, consisting of individual mice of strains C57BL/6J and DBA/2J, whose genomes differ to an extent similar to that of unrelated humans (r = 0.96, Figure 2A), demonstrating the robustness and reproducibility of these RNA-Seq-based estimates of NAGNAG ψ values. Similar numbers of alternatively spliced NAGNAGs were detected in mouse as in human, with 28% of alternatively spliced NAGNAGs in mouse exhibiting evidence of tissue-specific regulation and 8% being strongly regulated across the eight tissues studied (Table S4). Many orthologous NAGNAGs in human and mouse exhibited tissue-specific regulation in both species, e.g., NAGNAGs in FUBP1, CAMK2B, CAPRIN2, and ATRX (a complete list of alternative NAGNAGs in mouse is provided in Table S5). The higher fraction of regulated NAGNAGs detected in the human data probably results from a combination of factors, including the greater number of tissues sampled (Figure S1), the diverse genetic backgrounds of the human samples, and intrinsically higher read coverage variability in the human RNA-Seq data used. Comparing technical replicates of human tissues, which capture variability in sequencing, we estimated false discovery rates (FDRs) for discovering strongly regulated NAGNAGs ranging from ∼0.8% to ∼13.3%, with a mean FDR of 4.4% (Figure S2). In contrast, comparing biological replicates of mouse tissues, which capture all major sources of variability (tissue collection, library preparation, sequencing, and individual-specific splicing differences), we estimated FDRs ranging from 0.6% to 1.9%, with a mean of 1.1% (Figure S3). Using these estimated FDRs, and extrapolating the mouse data to 16 tissues (Figure S1), we estimated that between 12% and 37% of NAGNAGs are strongly regulated across tissues in mammals, making strong regulation a fairly common occurrence—though somewhat less common than for other types of splicing events. The relatively small differences between samples of the same tissue from mice whose genomes differed to an extent comparable to that of unrelated humans (Figure 2A) suggested that inter-individual variation contributed less than other sources of variation (e.g., tissue-specific differences) to the variations observed between the human libraries.
Orthologous human and mouse NAGNAGs exhibited high quantitative conservation of isoform levels. This was particularly true when the difference between the proximal and distal 3′ splice site scores—using a method that scores the strength of the polypyrimidine tract and AG region—was conserved (Spearman's ρ = 0.67, Figure 2B). The correlation decreased somewhat in cases where the differences in 3′ splice site scores were less conserved (ρ = 0.54, p = 0.013 for test of equality of correlation using the Fisher transformation; Figure S4), suggesting that changes in relative 3′ splice site strength may contribute to species-specific differences in NAGNAG splicing. Notably, many NAGNAGs with diverged splice site scores were alternatively spliced in one species but constitutively spliced in the other, suggesting relatively rapid evolution of 3′ splice site positions.
To better understand how NAGNAG splicing is regulated, and which sequence regions might be involved, we examined sequence conservation of flanking intronic and exonic regions for NAGNAGs grouped by switch score using alignments of the genomes of placental mammals. Tissue-specific NAGNAGs exhibited markedly increased sequence conservation in the upstream intron (Figure 2C–D), with little or no increase in other analyzed regions. The consistent increase in conservation in the upstream intron with increasing switch score provides further evidence that these regulated NAGNAGs contribute to organismal fitness, and is consistent with previous observations that alternatively spliced NAGNAGs have higher upstream sequence conservation than constitutive 3′ splice sites [26]. Enumerating NAGNAGs in introns of the fly Drosophila melanogaster, and comparing isoform usage across 30 developmental time points (embryo to adult) using RNA-Seq data from the modENCODE consortium [2], we identified over 500 NAGNAGs in coding regions of Drosophila genes where both isoforms were expressed at ≥5% in at least one developmental time point. Of these, 14% were developmentally regulated, with 5% being strongly regulated as defined above. As in mammals, more highly regulated fly NAGNAGs were associated with increased sequence conservation within and upstream of the 3′ splice site (Figure 2E). The consistent location of the sequence conservation signal for regulated NAGNAGs in mammalian and insect genomes (Figure 2C–E) suggested that the region ∼50 bp upstream of the NAGNAG motif, encompassing the competing 3′ splice sites themselves, may contain most of the regulatory information that governs NAGNAG alternative splicing. The extensive tissue-specific regulation observed in mammals and developmental regulation seen in flies may indicate that regulated NAGNAG alternative splicing is widespread in metazoans.
The increased divergence in isoform usage observed for NAGNAGs that had undergone divergence in 3′ splice site score difference (Figures 2B, S4) suggested that relative splice site strength is a major determinant of NAGNAG quantitative isoform usage. Supporting this hypothesis, previous EST-based analyses have demonstrated that splice site strength impacts whether or not a NAGNAG will be alternatively spliced [21],[27]. To explore the relationship between splice site strength and quantitative isoform levels, rather than simply the presence or absence of alternative splicing, we created a biophysical model wherein the probabilities of using the proximal and distal splice sites are proportional to and , respectively, where the parameter determines the inherent preference for using the intron-proximal splice site and is a scaling factor for the splice site scores. This simple model, containing just two free parameters, accurately predicted mean isoform usage across human tissues (Figure 3A), suggesting that relative 3′ splice site strength is the primary determinant of basal NAGNAG isoform levels. The fitted value provides a quantitative measurement of preference for the proximal splice site in NAGNAG 3′ splice site recognition, predicting that the distal splice site of a NAGNAG must typically be bit stronger than the proximal splice site in order to be spliced with equal efficiency. Analysis of mouse NAGNAGs yielded similar values of the Q and B parameters (Figure S5), supporting the robustness of these estimates. This preference for the proximal site was obvious even after controlling for the identity of the −3 bases (the Ns of the NAGNAG) (Figure 3B), which are known to be important determinants of NAGNAG isoform choice [18],[26],[27]. Preference for the proximal splice site is consistent with models of 3′ splice site recognition that involve scanning or diffusion from an upstream branch point [28],[29].
While the mean ψ value was accurately predicted by our model, the variability around the mean was substantially higher than expected based on measurement noise (Figure 3A). This observation is consistent with the concept that splice site strength determines the basal levels of the two NAGNAG isoforms, but the presence of regulatory sequence elements not captured by the 3′ splice site score, and variation in the levels of associated trans-acting factors, modulates the isoform ratios that occur in different tissues.
The variability in NAGNAG splicing observed above implies that features outside of splice site strength and the −3 base must also be involved in determining isoform usage. For example, the NAGNAG in the splicing factor PTBP2 (Figure 1D) represents an exception to the pattern observed above: the −3 bases (CAGAAG) predict predominant proximal splice site usage, since C is strongly favored over A and is also proximal, but roughly equal proportions of both isoforms are expressed across all tissues studied (Figure S6). This observation led us to wonder whether other aspects of this 3′ splice site, e.g., the relatively short and distally located polypyrimidine tract and the relatively distal location of the putative branch point (Figure 1D) might favor use of the distal NAG in this and other cases.
While many analyses support the importance of the −3 base combination in NAGNAG alternative splicing [18],[26],[27], there is less consensus in the literature about the relevance of other major elements of the 3′ splice site, including the polypyrimidine tract and branch site. Molecular genetics experiments demonstrated that mutating sequences near the polypyrimidine tract and branch site influenced alternative splicing of specific NAGNAGs [30],[31], but two computational studies that used machine-learning approaches [27],[32] concluded that neither of these elements significantly influenced NAGNAG splicing globally. Notably, the experimental studies [30],[31] measured quantitative isoform ratios, as we do in this study, while the machine-learning studies [27],[32] simply classified NAGNAGs as constitutively or alternatively spliced.
In order to dissect features that impact NAGNAG isoform choice, controlling for the effect of the −3 bases, we considered the large class of NAGNAGs with favored (C or T) nucleotides at both −3 bases (YAGYAGs). We found that exons that predominantly used the proximal splice site (“proximal major” YAGYAGs) had substantially distinct nucleotide preferences from those that predominantly used the distal site (“distal major” YAGYAGs) (Figure 3C), consistent with the experimental results of Tsai et al. [30],[31], who found that modifying the sequence upstream of the 3′ splice site influenced NAGNAG splicing. For example, distal major YAGYAGs tended to have shorter, more distal, polypyrimidine tracts than proximal major YAGYAGs (Figure 3D), implicating polypyrimidine tract length and location in control of NAGNAG splicing. The proportion of CT/TC dinucleotides in the polypyrimidine tract was ∼25% higher for distal major YAGYAGs (Figure 3E), suggesting the possible involvement of CU/UC-binding factors such as those of the PTB family [33]—some of which are tissue-specifically expressed—in promoting use of distal NAGs. The location of the first upstream AG was also shifted several bases downstream in distal major YAGYAGs compared to other 3′ splice sites (Figure 3F), suggesting that the branch site is located further downstream in this class and that use of a distally located branch site favors use of the distal YAG, perhaps because the distance to the 3′ splice site is more optimal.
Strongly regulated YAGYAGs had features that were intermediate between the extremes found for proximal major and distal major YAGYAGs, such as polypyrimidine tracts of intermediate length (Figure 3D), suggesting that the presence of intermediate features facilitates regulation. Increased regulation was also associated with reduced 3′ splice site strength and greater similarity in strength between the competing sites (Figure S7), consistent with previous studies of other types of alternative splicing [34].
The −4 base, four nucleotides upstream of the 3′ splice site, is not generally considered to be important in splicing (with rare exceptions [35]). This position contains little or no information in alignments of constitutive 3′ splice sites [36], although a previous machine-learning analysis of features distinguishing between constitutively and alternatively spliced NAGNAGs included the −4 base in their classifier [27]. Our quantitative analysis strongly supported a special role in NAGNAG regulation for this canonically unimportant position. For distal major YAGYAGs, the −4 position (here referring to the position four nucleotides upstream of the intron-proximal splice site) had the highest information content of any position upstream of the YAGYAG (Figure 3C); furthermore, the −4 base was more conserved in distal major and strongly regulated YAGYAGs than for other classes of 3′ splice sites (Figure S8).
Of the observations in Figure 3, the two that seemed most compelling were the preference for pyrimidines at the −4 position and the more distal positioning of branch points in YAGYAGs that favored the distal splice site. To test the predicted role of the −4 base in regulation of NAGNAG splicing, we used a minigene reporter based on the NAGNAG in PTBP2, whose splicing alters an exon coding for the RRM4 RNA binding domain (Figures 1D, 4A). As predicted based on the data in Figure 3C, mutation of the −4 base (T in the wildtype) to A or G resulted in a substantial shift in splicing toward use of the proximal NAG, while mutation to C had no effect (Figure 4B). These observations confirm that presence of a pyrimidine at the −4 position favors use of the distal NAG, even though no sequence preference was observed at this position in constitutive splice sites (Figure 3C). Presence of a pyrimidine at the −4 position of a NAGNAG might function to shift the location of binding of U2AF65 downstream by a base or more from its normal position, which might then result in preferential binding of U2AF35 to the downstream NAG, though this will require further study.
We also tested the role of the branch point in NAGNAG splicing by manipulating the branch site to 3′ splice site distance in this reporter, either in a context in which the inferred native branch point sequence (BPS) was intact or in a context in which the native BPS had been replaced by the previously mapped BPS of IGF2BP1 intron 11 (Figure 4A). With the native BPS present, an increase of just four bases in the BPS-3′ splice site distance was sufficient to cause a substantial shift in splicing towards the proximal NAG, with little or no additional shift resulting from addition of three more bases (Figure 4C). In the context of the exogenous IGF2BP1 BPS, a somewhat higher basal level of proximal splice site usage was reduced by deletion of six bases, with deletion of three bases producing a modest change (Figure 4D). These data indicate that the BPS plays a significant role in NAGNAG splicing and confirm that shorter BPS-3′ splice site distances can shift splicing toward the distal NAG.
Together, our analyses of proximal/distal major splicing suggested that NAGNAG 3′ splice sites afford broad scope for evolutionary tuning of isoform ratios, even in cases where the sequence of the second NAG is constrained by selection on the encoded amino acid. For example, mutations affecting the upstream −3 and −4 bases, the polypyrimidine tract, or the location of the branch site could all potentially modulate the ratio of the two isoforms across a range from predominantly proximal to predominantly distal isoform usage, which might facilitate evolutionary addition and deletion of single codons at 3′ splice junctions. A previous study observed reduced frequencies of amino acid substitutions near exon-exon junctions relative to the centers of exons, presumably resulting from purifying selection acting on exonic splicing enhancer motifs [10],[11]. By contrast, when we examined exon length changes in alignments of orthologous human and mouse coding exons (Figure 5A), we observed a striking 18.5-fold enrichment for gain/loss of exonic sequence at 3′ splice sites relative to flanking positions (Figure 5B; assignment of gaps is illustrated in example alignments in Figure S9). No particular enrichment for gain/loss of exonic sequence was observed at the 5′ splice site, suggesting that increased addition/deletion of exonic sequence is associated with properties of the 3′ splice site itself, rather than being a generic feature of exon boundaries. This pattern was not changed when restricting to constitutive splice junctions (Figure S10). A majority of the changes plotted in Figure 5B involved gain/loss of precisely three bases, and restricting to changes of exactly this size yielded a similar degree of enrichment at the 3′ splice site (Figure 5C).
While gain/loss of exonic sequence is normally attributed to insertions or deletions (“indels”) in the genome, the increased frequency of changes at the 3′ splice site suggested a prominent role for an alternative mechanism involving genomic substitutions that give rise to three base shifts in exon boundaries without insertion or deletion of genomic DNA. For example, creation of a NAG motif immediately upstream of a 3′ splice site NAG by mutation would be expected to commonly shift splicing upstream by three bases (resulting in exonization of three bases of intron) or generate an alternatively spliced NAGNAG that could subsequently lose splicing at the downstream NAG through mutation. Alternatively, a mutation creating an immediately downstream NAG—or a mutation that weakened the upstream NAG relative to a pre-existing downstream NAG—could result in either alternative splicing or loss of three bases of exonic sequence. As outlined in Table S6, both of these scenarios could arise frequently by single base substitutions, which occur at a rate that is an order of magnitude higher than the rate of genomic indels [37].
Consistent with this substitution/exaptation model and the finding that many NAGNAGs are alternatively spliced in the Drosophila lineage, we observed similar enrichment for gain/loss of three bases of exonic sequence at the 3′ splice site when comparing orthologous D. melanogaster and D. yakuba coding exons (Figure 5D). Notably, the enrichment of three base gaps at the 3′ splice site was 3-fold weaker in comparisons of Caenorhabditis elegans and C. briggsae exons (Figure 5E). NAGNAG alternative splicing is reported to occur rarely in nematodes due to a highly constrained 3′ splice site motif [15]. We confirmed the rarity of NAGNAG alternative splicing in C. elegans using RNA-Seq data from 14 developmental time points and conditions generated by the modENCODE consortium. Enumerating NAGNAGs in introns of C. elegans coding genes, we detected alternative splicing (both isoforms expressed at ≥5% in at least one developmental time point) for only 18% of NAGNAGs with favorable pyrimidine bases at both −3 positions based on RNA-Seq read depths slightly below those used in human. By contrast, 50%–85% of human, mouse, and Drosophila YAGYAGs were detected as alternatively spliced, suggesting that NAGNAG alternative splicing is substantially rarer in worms than in other metazoans. This decrease in abundance mirrors the 3-fold weaker enrichment of three base gaps at 3′ splice sites observed in worms (Figure 5E).
Sequence motif analyses further implicated NAGNAG splicing in the exon length changes observed at exon boundaries. Classifying the borders of orthologous mouse and rat exons as unchanged, expanded, or contracted (comparing to human, cow, chicken, and/or Xenopus laevis as outgroups), we observed evidence of residual NAGNAG motifs in exons with altered boundaries (Figure 5F). Specifically, exons expanded in mouse or rat exhibited a consensus NAG at exonic positions +1 to +3, and contracted exons exhibited a consensus NAG at intronic positions −6 to −4. The presence of this residual sequence motif provides further evidence that a substantial portion of exon length changes observed between orthologous mammalian exons derive from splicing-mediated shifts in exon boundaries rather than genomic indels. Likely because of subsequent selection to optimize the polypyrimidine tract, the residual NAG signal was weaker for contracted than for expanded exons.
Consistent with these findings, we observed a strong association between gain/loss of three bases in the rodent lineage and presence of a NAGNAG in orthologous human exons. Exons that expanded or contracted in rodents were 7.5-fold more likely to have a NAGNAG in the orthologous human exon than were exons with unchanged boundaries (Figure 5G). Further subdividing these exons according to the splicing pattern of the NAGNAG in human, we observed that rodent exons orthologous to alternatively spliced human NAGNAGs were ∼9 times more likely to have gained/lost exonic sequence than those orthologous to constitutively spliced human NAGNAGs (Figure 5H). These analyses implicate NAGNAG alternative splicing as a very common evolutionary intermediate in the gain and loss of single codons from exons.
This model, where frequent alternative splicing at the 3′ splice site leads to gain/loss of exonic sequence, is expected to play out very differently at 5′ splice sites. Competing 5′ splice sites are most frequently four bases apart [22], resulting in a frame-shift which is likely to render one of the protein products non-functional and potentially target the mRNA for nonsense-mediated decay. Although common, competing 5′ splice sites separated by four bases are therefore unlikely to lead to accelerated exon length changes and we observed no significant increase in exon length changes at the 5′ splice site (Figure 5A).
Most three base changes to mRNAs probably minimally affect RNA-level properties such as message stability. However, insertion/deletion of a single amino acid residue can have a profound impact on protein function. For example, deletion of a single codon can alter protein degradation, subcellular localization, DNA binding affinity, or other protein properties [38],[39]; can cause diseases including cystic fibrosis and Tay-Sachs disease [40],[41]; and can even rescue a disease-related phenotype [42]. Insertion or deletion of a codon in a protein structural motif with a periodic hydrogen bonded structure such as a beta sheet or coiled coil domain might have a disproportionate effect on protein structure by altering the hydrogen bonding of a large number of downstream residues. The codon-level effects of NAGNAG splicing are largely determined by intron “phase” (position relative to the reading frame) [15]. Considering the spectrum of codons that occurred opposite three base gaps at the beginnings of exons (corresponding to the peak in Figure 5C), we observed a highly non-random distribution that strongly favored glutamine, alanine, glutamate, and serine and disfavored most other residues including cysteine, phenylalanine, and histidine relative to the background (Table S7). Distinct and far stronger biases were observed when grouping introns by phase. These biases occurred in a pattern consistent with frequent origin via exaptation of NAGNAGs (Figure 5I). For example, glutamine (mostly coded by CAG) was the most commonly added residue at the end of “phase 0” introns, for which the first three bases of the downstream exon form a codon. Serine (mostly AGY) and arginine (mostly AGR) were the most commonly added residues at the boundaries of phase 2 introns, for which the AG of an added NAG would form the first two bases of a codon. These biases contributed to a strong enrichment observed for gain/loss of predicted phosphorylation sites at 3′ splice sites (Figure S11). Together, the analyses in Figure 5 demonstrate that gain and loss of residues along proteins occurs in a strongly biased manner, with a highly accelerated rate and biased codon spectrum at the beginnings of exons that is likely driven by genomic substitutions that alter NAGNAG motifs or their splicing patterns. These observations suggest that the evolutionary trajectories of proteins in metazoans are shaped to a surprising extent by the specific locations and phases of introns that interrupt their coding sequences.
Mapped sequence reads from the human and mouse RNA-Seq experiments are located in NCBI's GEO database (accession number GSE30017). The complete Body Map 2.0 sequence data are in the ENA archive with accession number ERP000546 (available at http://www.ebi.ac.uk/ena/data/view/ERP000546). These data are also accessible from ArrayExpress (ArrayExpress accession: E-MTAB-513). The Body Map 2.0 data were generated by the Expression Applications R&D group at Illumina using the standard (polyA-selected) Illumina RNA-Seq protocol from total RNA obtained commercially (Ambion) using the HiSeq 2000 system. We downloaded D. melanogaster (“Developmental Stage Timecourse Transcriptional Profiling with RNA-Seq”) and C. elegans (“Global Identification of Transcribed Regions of the C. elegans Genome”) RNA-Seq data from the modMINE (http://intermine.modencode.org/) website of the modENCODE consortium. For the C. elegans data, we restricted to 36 bp reads for consistency with other analyses.
We used the set of splicing events from [1] to identify skipped exons, alternative 3′ splice sites (>3 nt apart), alternative 5′ splice sites, and mutually exclusive exons in the human (GRCh37, or hg19) and mouse (NCBIM37, or mm9) genomes (Figure 1C). We enumerated all possible NAGNAGs in the human genome by finding all 3′ splice sites in these alternative splicing events and the Ensembl [43] and UCSC [44] annotation databases and then searching for NAGNAG motifs. We classified splice junctions as constitutive if they did not overlap any alternative splicing event present in the databases described above.
Mouse tissues from a 10-wk-old male were extracted immediately after death and stored in RNAlater per the manufacturer's instructions (Ambion). Tissue was lysed in Trizol and RNA was extracted with Qiagen miRNeasy mini columns. Using 5 µg of total RNA, we performed polyA selection and prepared strand-specific libraries for Illumina sequencing following the strand-specific dUTP protocol [45] and using the SPRIworks Fragment library system (Beckman Coulter). We obtained final insert sizes of approximately 160 bp. We sequenced these libraries using the Illumina HiSeq 2000 and the GAIIx machines.
For each NAGNAG, we extracted the sequence flanking the proximal and distal 3′ splice sites and used Bowtie [46] version 0.12.7 to map reads to these two sequences. We required that short reads have at least 6 nt on either side of the splice junction (an “overhang” of 6 nt), and furthermore that there be no mismatches within the overhang region. In order to eliminate errors in read mapping due to non-unique splice junctions, we restricted the set of NAGNAGs enumerated across the genome to the subset of NAGNAGs for which all 36-mers mapping to either splice site did not map to the genome or any other splice junction (we used 36-mers because they were the shortest reads analyzed in our experiments). We then computed ψ values as (number of reads mapping to the proximal splice junction)/(number of reads mapping to either the proximal or distal splice junction). For all bioinformatics analyses, we only analyzed the subset of tissues for which a particular NAGNAG had a total of at least 10 reads in order to control for variation in junction coverage due to gene expression differences. We experimented with requiring different levels of junction coverage (10–100 reads per NAGNAG) and confirmed that our conclusions were insensitive to the chosen cutoff. We identified alternatively spliced events as those for which both isoforms were expressed at ≥5% in at least one sample (restricting to tissues for which a particular NAGNAG had ≥10 reads), and identified regulated events as those with p≤0.01 by the proportion or z-test (prop.test in R [http://www.R-project.org/]). As described in the text, when computing the fraction of regulated NAGNAGs, we only considered NAGNAGs which were alternative spliced by these criteria (both isoforms expressed at ≥5% in at least one sample).
For Figure 1C and Tables S2, S3, we re-mapped the reads using TopHat [47] version 1.1.4 and restricted to uniquely mapping reads with an overhang of 6 nt and no mismatches in the overhang region. Using only reads mapping to the two 3′ (skipped exons, NAGNAGs, alternative 3′ splice sites, and mutually exclusive exons) or 5′ (alternative 5′ splice sites) splice sites of each event, we computed ψ values and identified alternative spliced and regulated events as described above.
We estimated false-discovery rates as the fraction of events which were differentially expressed between technical (human) or biological (mouse) replicates identified using the procedure described above for regulated events. Briefly, for each tissue and pair of replicates, we restricted to the set of NAGNAGs which were alternatively spliced in at least one of the replicates and computed the fraction of these NAGNAGs which were differentially expressed with p≤0.01 between the replicates. We estimated mean FDRs for human (4.4%) and mouse (1.1%) by taking a weighted average over tissues, where we weighted the FDR computed for each tissue by the number of alternatively spliced NAGNAGs analyzed for that tissue.
The fraction of strongly regulated NAGNAGs increased essentially linearly with the number of tissues considered for both human and mouse (Figure S1). We expect this trend to continue as the number of mouse tissues increases, as it does for the human data. Accordingly extrapolating the mouse data to 16 tissues with a linear fit and subtracting the mean FDR of 1.1%, we estimated that at least 12% of alternatively spliced mouse NAGNAGs are strongly regulated, providing a lower bound on the fraction of strongly regulated NAGNAGs in mammals. We used the human data to compute a corresponding upper bound of 37% by subtracting the mean FDR of 4.4% from the observed fraction of strongly regulated NAGNAGs (Figure S1).
For each NAGNAG event, the probabilities of using the proximal and distal splice sites are proportional to and , where and are the proximal and distal splice site scores. The probability of using the proximal splice site is therefore . We fit the parameters and as follows: For each NAGNAG, we computed the mean ψ (averaging over tissues). We then binned NAGNAGs according to their splice site score differences, using a bin size of 3.25 bits and a bin increment of 0.5 bits, and computed the median ψ for each bin. We fit a straight line to the six bins flanking the point where ψ = 50% and estimated the parameters as and based on a first-order Taylor expansion.
We performed a whole-genome alignment of human and mouse using Mercator (http://www.biostat.wisc.edu/~cdewey/mercator/) and FSA [48], and identified orthologous NAGNAGs as those for which both the 5′ splice site and competing 3′ splice sites were orthologous according to the corresponding sequence alignment. For the Drosophila analysis, we used a previously described D. melanogaster–D. yakuba whole-genome alignment [49].
For all sequence conservation analyses, we downloaded phastCons scores [50] from the UCSC annotation databases [44]. We used phastCons46 (placental mammals) for human, phastCons30way (placental mammals) for mouse, and phastConst15way for D. melanogaster.
Segments of PTBP2 intronic sequence containing the NAGNAG were cloned into a modular splicing reporter [51] upstream of the IGF2BP1 exon using SacI and XhoI restriction enzyme sites. Forward and reverse oligonucleotides (below) were mixed in equimolar ratios, annealed, and double-digested with SacI and XhoI, or in some cases the oligonucleotides were ordered with desired restriction site overhangs, and ligated into the pGM4G9 minigene. For constructs analyzing the effects of distance to the native PTBP2 branch point, the vector (IGF2BP1) branch point sequence was first mutated by site-directed mutagenesis (TCATTGA was deleted, immediately upstream from the SacI restriction site) prior to insertion of the PTBP2 3′ splice site.
All minigene reporters (0.5 µg) were transfected into HEK293T cells using Lipofectamine 2000 (Invitrogen). RNA was isolated 18–24 h post-transfection with RNeasy Mini Kits (Qiagen). RT-PCR was performed with a fluorescent primer (NAGNAG_Forward: 5′ 6FAM- TCTTCAAGTCCGCCATGC and NAGNAG_reverse: 5′ AGTCAGGTGTTTCGGGTGGT). The proximal (63 nucleotides) and distal (60 nucleotides) isoforms were resolved on a 10% TBE gel and detected with a Typhoon 9000 scanner (GE Healthcare). Proximal and distal isoforms were quantified with ImageJ software.
Primers: PTB2_For: cagtgtctaattttataattttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTC; PTB2_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaattataaaattagacactgagct; BPS+4_For: cagtgtctaattttataaataattttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTC; BPS+4_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaattatttataaaattagacactgagct; BPS+7a_For: cagtgtctaattttataaataaatattttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTC; BPS+7a_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaatatttatttataaaattagacact gagct; BPS+7b_For: cagtgtctaatttttttataattttttttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTC; BPS+7b_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaaaaaattataaaaaaattagacactgagct; −4A_For: cagtgtctaattttataattttgttacagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTC; −4_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgtaacaaaattataaaattagacactgagct; −4G_For: cagtgtctaattttataattttgttgcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTc; −4G_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgcaacaaaattataaaattagacactgagct; −4C_For: cagtgtctaattttataattttgttccagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTc; −4C_Rev: TCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctggaacaaaattataaaattagacactgagct; IGF2BP1BPS_For: gcgagctcttataattttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTCTCGAGCGG; IGF2BP1BPS_Rev: CCGCTCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaattataagagctcgc; BPS-3_For: gcgagctctaattttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTCTCGAGCGG; BPS-3_Rev: CCGCTCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaattagagctcgc; BPS-6_For: gcgagctcttttgtttcagAAGATTGCACCACCCGAAACACCTGACTCCAAAGTTCGTATGGTTCTCGAGCGG; BPS-6_Rev: CCGCTCGAGAACCATACGAACTTTGGAGTCAGGTGTTTCGGGTGGTGCAATCTTctgaaacaaaagagctcgc.
We restricted all analyses to “singleton orthologs,” genes without paralogs and with unambiguous orthology assignments in all species considered for each analysis, annotated in Ensembl [43] and queried with PyCogent [52]. For each gene, we required that the longest annotated coding sequence have the same number of exons in all species, and performed all subsequent analyses using this longest coding sequence. For each longest coding sequence, we extracted pairs of consecutive exons, concatenated them, and then aligned them to their corresponding orthologous sequences using FSA [48]. In order to control for alignment error, we required that alignment sequence identity be greater than 70% and that the total inserted sequence be no longer than 20% of the length of the shortest exon. Furthermore, if gaps in an alignment could be moved to lie at exon-exon boundaries rather than within exonic sequence while preserving the alignment quality (number of exact matches), then we modified the alignment accordingly, as FSA is unaware of exon structures. This modification affected only a small fraction of alignments, and our results in Figure 5 are unchanged without this modification.
We classified orthologous mouse and rat exons as unchanged, expanded, or contracted based on comparison with an outgroup (human, cow, chicken, Xenopus laevis, or Danio rerio, in that order, until an informative comparison was found). For each exon in each class, we extracted the corresponding intronic sequence and created a sequence logo using WebLogo (Figure 5F–H) [53].
For analyses of amino acid sequences in Figure 5I, we compared the amino acids gained or lost in alignments with gaps of three bases at the 3′ splice site. If the next gain/loss was a single amino acid (for example, if the human peptide was SR and the mouse peptide was R), then we counted only the single amino acid which was inserted (S); if the gain/loss was two amino acids (for example, if the human peptide was SR and the mouse peptide was K), then we counted both amino acids which were inserted (SR).
For Figure S11, we used a BioPerl module [54] to query Scansite [55] to predict phosphorylation sites (medium stringency) in the translated longest annotated coding sequence, and plotted the location of predicted phosphorylation sites which were gained/lost in human and mouse.
Unless otherwise described, all plots in Figure 5 were created with matplotlib (http://matplotlib.sourceforge.net/).
|
10.1371/journal.pgen.1000676 | Oncogenic Pathway Combinations Predict Clinical Prognosis in Gastric Cancer | Many solid cancers are known to exhibit a high degree of heterogeneity in their deregulation of different oncogenic pathways. We sought to identify major oncogenic pathways in gastric cancer (GC) with significant relationships to patient survival. Using gene expression signatures, we devised an in silico strategy to map patterns of oncogenic pathway activation in 301 primary gastric cancers, the second highest cause of global cancer mortality. We identified three oncogenic pathways (proliferation/stem cell, NF-κB, and Wnt/β-catenin) deregulated in the majority (>70%) of gastric cancers. We functionally validated these pathway predictions in a panel of gastric cancer cell lines. Patient stratification by oncogenic pathway combinations showed reproducible and significant survival differences in multiple cohorts, suggesting that pathway interactions may play an important role in influencing disease behavior. Individual GCs can be successfully taxonomized by oncogenic pathway activity into biologically and clinically relevant subgroups. Predicting pathway activity by expression signatures thus permits the study of multiple cancer-related pathways interacting simultaneously in primary cancers, at a scale not currently achievable by other platforms.
| Gastric cancer is the second leading cause of global cancer mortality. With current treatments, less than a quarter of patients survive longer than five years after surgery. Individual gastric cancers are highly disparate in their cellular characteristics and responses to standard chemotherapeutic drugs, making gastric cancer a complex disease. Pathway based approaches, rather than single gene studies, may help to unravel this complexity. Here, we make use of a computational approach to identify connections between molecular pathways and cancer profiles. In a large scale study of more than 300 patients, we identified subgroups of gastric cancers distinguishable by their patterns of driving molecular pathways. We show that these identified subgroups are clinically relevant in predicting survival duration and may prove useful in guiding the choice of targeted therapies designed to interfere with these molecular pathways. We also identified specific gastric cancer cell lines mirroring these pathway subgroups, which should facilitate the pre-clinical assessment of responses to targeted therapies in each subgroup.
| Gastric cancer (GC) is the second leading cause of global cancer mortality [1]. Particularly prevalent in Asia, most GC patients are diagnosed with advanced stage disease [2]. Deregulation of canonical oncogenic pathways such as E2F, K-RAS, p53, and Wnt/β-catenin signaling are known to occur with varying frequencies in GC [3]–[6], indicating that GC is a molecularly heterogeneous disease. Previous studies describing GC diversity in primary tumors have typically focused on single pathways, measuring only one or a few biomarkers per experiment [4],[6],[7]. In contrast, experimental evidence indicates that most cancer phenotypes (uncontrolled growth, resistance to apoptosis, etc) are largely governed not just by single pathways, but complex interactions between multiple pro- and anti-oncogenic signaling circuits [8]. Narrowing this gap between the clinical and experimental arenas will require strategies capable of measuring and relating activity patterns of multiple oncogenic pathways simultaneously in primary tumors.
Previous studies have proposed using gene expression signatures to predict the activity of oncogenic pathways in cancers [9] – here, we hypothesized that patterns of oncogenic pathway activation could be used to develop a genomic taxonomy of GC. Importantly, this pathway-centric strategy differs substantially from previous microarray studies describing expression changes associated with morphological and tissue type differences in GC [10],[11], as pathway signatures (rather than individual genes) are used as the basis for cancer classification. We developed an in silico method to map activation levels of different pathways in cohorts of complex primary tumor profiles and validated this pathway-directed classification approach using proof-of-concept examples from breast cancer. We then applied this method to GC to evaluate eleven oncogenic pathways previously implicated in gastric carcinogenesis [3]–[7], [12]–[17]. In total, we analyzed over 300 primary GCs derived from three independent patient cohorts, performing to the best of our knowledge the largest genomic analysis of GC to date. We identified three oncogenic pathways (nuclear factor-κB (NF-κB), Wnt/β-catenin, and proliferation/stem cell) that were deregulated in the vast majority (>70%) of GCs, and functionally validated the pathway predictions in vitro using a panel of GC cell lines. Although patient stratification at the level of individual pathways failed to consistently demonstrate significant differences in clinical outcome, patient stratification by oncogenic pathway combinations (e.g. high proliferation/high NF-κB vs. low proliferation/low NF-κB) showed reproducible and significant survival differences in multiple independent patient cohorts, suggesting a critical role for pathway combinations in influencing GC clinical behavior. Our results thus demonstrate that GCs can be successfully taxonomized using oncogenic pathway activity into biologically, functionally, and clinically relevant subtypes.
Our strategy for predicting levels of oncogenic pathway activation in cancers involves four steps (Figure 1A). First, we defined ‘pathway signatures’ - sets of genes exhibiting altered expression after functional perturbation of a specific pathway in a well-defined in vitro or in vivo experimental system. Second, we mapped the pathway signatures onto gene expression profiles from a heterogeneous series of cancers. Third, using a nonparametric, rank-based pattern matching procedure, activation scores were assigned to individual cancers based upon the strength of association to the pathway signature. Finally, the individual cancers were sorted based upon their pathway activation scores.
Before applying this approach to GC, we considered it important to validate this in silico strategy in a series of proof-of-principle experiments. We chose the example of breast cancer, a malignancy for which there is ample evidence of pathway heterogeneity and discrete ‘molecular subtypes’ [18]. To perform this validation, we first asked if previously described pathway signatures associated with impaired estrogen signaling could be used to identify breast cancer cell lines exhibiting high levels of estrogen receptor (ER) activity. We analyzed a gene expression panel of 51 breast cancer cell lines originally described in Neve at al. [18] with an 11-gene ‘tamoxifen sensitivity’ pathway signature derived from a list of genes differentially expressed between MaCa 3366, a tamoxifen-sensitive human mammary carcinoma xenograft, and MaCa 3366/TAM, a tamoxifen-resistant subline of the same xenograft [19]. We found that breast cancer cell lines positively associated with the tamoxifen sensitivity signature exhibited significantly higher expression levels of ESR1, the estrogen receptor and molecular target of tamoxifen, compared to lines showing negative pathway activation scores (p = 2.12×10−7, Accuracy 84.3%, Sensitivity 100%, Specificity 75%) (Figure 1B and Table S1).
Second, we tested if a pathway signature associated with estrogen signaling but derived from non-breast tissue could also be used to stratify the same panel of breast cancer cell lines. We queried the breast cancer cell line panel with a 41-gene ‘estrogen response’ signature derived from a list of genes upregulated by estradiol in U2OS human osteosarcoma cells [20]. Despite the signature originating from a different tissue type (e.g. osteosarcoma), we once again found that, when sorted based upon their predicted estrogen responsiveness, breast cancer cell lines clustered together by their level of ESR1 (estrogen receptor) expression (p = 0.0035, Accuracy 62.7%, Sensitivity 94.7%, Specificity 43.8%) (Figure 1C and Table S1). These results demonstrate that it is indeed feasible to predict patterns of pathway activation in a particular cancer of interest (gastric cancer in our cases) using expression signatures obtained from different experimental conditions and even different tissue types.
After validating this pathway prediction approach, we proceeded to apply the strategy to primary GC. Rather than testing every possible pathway, we selected eleven oncogenic and tumor suppressor pathways previously implicated in gastric carcinogenesis, using in our analysis RAS [4], p53 [5], BRCA1 [12], p21 [13], Wnt/β-catenin [6], E2F [3], SRC [14], MYC [15], NF-κB [21], histone deacetylation (HDAC) [16], and stem-cell related signatures [17]. Whenever possible, we attempted to select multiple signatures for each pathway, preferably from independent published studies. For example, of the two E2F activation signatures used in our approach, one signature was obtained by inducing E2F1 activity in rat fibroblast cells [22] while the other signature was obtained using an osteosarcoma-derived cell line containing an inducible ER-E2F1 fusion protein [23]. Final pathway predictions for further analyses were typically obtained by combining individual signatures belonging to the same pathway (see Materials and Methods).
We computed activation scores for the eleven pathways represented by 20 pathway signatures across three independent cohorts of primary GCs derived from Australia (Cohort 1–70 tumors), Singapore (Cohort 2–200 tumors), and the United Kingdom (Cohort 3–31 tumors). To visualize patterns of pathway activation, we depicted each cohort as a heatmap, where the heatmap color represents the predicted strength of activation for each pathway in the individual GCs. We observed considerable heterogeneity of pathway activation between individual GC patients (Figure 2A–2C). However, signatures derived from independent studies representing similar pathways frequently yielded similar prediction patterns (e.g. NF-κB (skin) and NF-κB (cervix)), and a chi-square test confirmed a significant level of similarity in the overall patterns of pathway activation between the Australia and Singapore cohorts (p = 0.00038), and between the Australia and UK cohorts (p = 0.00051, see Table S2) suggesting that the GC pathway predictions are not tied to a specific patient cohort. We identified two major clusters of co-activated pathways, which were completely preserved in Cohorts 1 and 2 (Figure 2A and 2B) and mostly preserved in Cohort 3 (Figure 2C). These included (i) a ‘proliferation/stem cell’ pathway cluster (brown vertical bar in Figure 2) encompassing pathways associated with various cell cycle regulators (e.g. MYC, E2F, p21) and stem cell signatures; and (ii) an ‘oncogenic signaling’ pathway cluster (grey vertical bar in Figure 2) containing many different oncogenic pathways (BRCA1, NF-κB, p53, Wnt/β-catenin, SRC, RAS, and HDAC pathways).
By analyzing the GC pathway heatmap in Figure 2, we selected three oncogenic pathways (NF-κB, Wnt/β-catenin, and proliferation/stem cell) that were individually activated in a significant proportion of GCs (≥35%), and when combined provided coverage of the majority (>70%) of GCs. Proliferation/stem cell pathways were activated in 40% of GCs in each cohort (range: 38 to 43%), Wnt/β-catenin pathways were activated in 46% of GCs (range: 43 to 48%), and the NF-κB pathway was activated in 39% of GCs (range: 35 to 41%) (color bars below each heatmap in Figure 2). These frequencies and other frequently deregulated pathways (e.g. p53) are listed in Table S3.
To experimentally validate these primary GC pathway predictions, we applied the pathway prediction algorithm to a panel of 25 GC cell lines (GCCLs) (Figure 3). Similar to primary GC, ‘proliferation/stem cell’ and ‘oncogenic signaling’ pathway clusters were also observed in the GCCLs. Furthermore, signatures representing the same pathway, but obtained from different studies, such as the two independent MYC-derived signatures [9],[24] also clustered together in the GC cell lines after unsupervised hierarchical clustering (purple brackets in Figure 3). Guided by the pathway predictions, we identified specific GC cell lines exhibiting patterns of oncogenic pathway activity mirroring primary GCs. Confidence in the selection of specific cell lines as in vitro models was also achieved by repeating the prediction procedure seven times using a variety of reference profiles, ranging from the median GCCL profile to independent profiles such as non-malignant normal stomach profiles (see Materials and Methods and Table S4). Pairwise comparisons confirmed that any two reference profiles were more likely to produce concurring pathway predictions than conflicting predictions (Text S1 and Table S4). Some examples of representative lines include AZ521 and MKN28 cells, which exhibit activation of proliferation/stem cell pathways, YCC3 and AGS cells for Wnt/β-catenin pathways, and MKN1 and SNU5 cells for the NF-κB pathway.
First, we directly measured the proliferative rates of 22 GCCLs and correlated the proliferation rate data with the mean activation score from signatures in the proliferation/stem cell pathway cluster. There was a significant association between the experimentally determined proliferative rates and the pathway activation scores (R = 0.4688, p = 0.0278) (Figure 4A). Supporting the notion that oncogenic pathway signatures are superior predictors of pathway activity compared to the expression of single key pathway genes, no significant associations were observed for either MYC or E2F1 expression (p = 0.48 and 0.38 for MYC and E2F1, respectively) (Figure S1).
Second, in order to validate the Wnt/β-catenin pathway predictions, we analyzed the expression of various Wnt pathway components (β-catenin, TCF4) and relative levels of TCF/LEF transcriptional activity in GC cell lines predicted to be Wnt/β-catenin- activated or Wnt/β-catenin-nonactivated. Of seven cell lines selected for their experimental tractability (e.g. ease of transfection and convenient growth conditions), we found that both β-catenin and the TCF/LEF transcription factor TCF4 (also known as TCF7L2), major components of the Wnt signaling pathway, were expressed in GC cell lines predicted by the pathway activation analyses to have high Wnt/β-catenin activity (AGS, YCC3, Kato III, and NCI-N87), but not expressed in two out of three lines (SNU1 and SNU5) associated with inconsistent or low Wnt/β-catenin activation scores (Figure 4B). Furthermore, in order to directly assay Wnt pathway activity, we determined TCF/LEF transcriptional activity in the GC cell lines using Topflash, a luciferase expressing plasmid containing multimerized TCF binding sites. The Topflash assay confirmed high TCF/LEF transcriptional activity in three out of four GC cell lines predicted to have high Wnt/β-catenin activity (AGS, YCC3, and Kato III), but minimal or no Topflash activity in GC cell lines associated with inconsistent or low Wnt/β-catenin activation scores (SNU1, SNU5, and SNU16). Additionally, the β-catenin pathway activation scores were significantly higher in GCCLs with more than two-fold TCF/LEF transcriptional activity (AGS, YCC3, Kato III, and NCI-N87) than in GCCLs with lower TCF/LEF transcriptional activity (p = 0.007, Figure 4B). When compared against single genes, superior associations to TCF/LEF transcriptional activity were once again observed using the mean activation score from Wnt/β-catenin signatures compared to either β-catenin or TCF4 (aka TCF7L2) expression alone (p = 0.038 for signatures vs. p = 0.31 and 0.58 for β-catenin and TCF4, respectively) (Figure S1).
Third, to validate the NF-κB pathway predictions, we selected 11 GCCLs consistently predicted as either NF-κB-activated (‘NF-κB/on’, six GCCLs) or NF-κB-nonactivated (‘NF-κB/off’, five GCCLs) (Figure S2). Increased gene expression of p50 and p65, the NF-κB heterodimer subunits, were observed in NF-κB/on GC cell lines compared to NF-κB/off GC cell lines (p = 0.0002 for p50, p = 0.046 for p65, Figure 4C) and at the protein level p65 expression was observed largely in the NF-κB/on lines (Figure 4C). Using immunocytochemistry on formalin fixed paraffin embedded GC cell lines, p65 protein expression was more frequently observed in NF-κB/on GC cell lines compared to NF-κB/off GC cell lines in terms of nuclear sublocalization, percentages of cells with staining (either nuclear or cytoplasmic), and staining intensity (Table S5, Figure S3). To determine if NF-κB/on GC cell lines also exhibited differential expression of p65-regulated genes compared to NF-κB/off GC cell lines, we combined the list of genes directly bound by the p65 transcription factor [25] with lists of genes regulated at the mRNA level by TNF-α [26], a known inducer of NF-κB activation. Using Gene Set Enrichment Analysis (GSEA, [27]), we found that p65 target genes upregulated by TNF-α treatment were significantly overexpressed in NF-κB/on GC cell lines compared to NF-κB/off GC cell lines (normalized enrichment score, NES = 1.86; false discovery rate, FDR<0.001, bottom most panel, Figure 4C). Conversely, p65 target genes downregulated by TNF-α were significantly underexpressed in NF-κB/on GC cell lines compared to NF-κB/off GC cell lines (NES = −1.56, FDR = 0.019, bottom most panel, Figure 4C). Finally, to directly confirm the presence of elevated NF-κB activity, we transfected three NF-κB/on GC cell lines and two NF-κB/off GC cell lines with a luciferase reporter containing a NF-κB reporter gene. As shown in Figure 4D, the three NF-κB/on GC cell lines exhibited elevated NF-κB transcriptional activity compared to the two NF-κB/off GC cell lines (p = 0.0084).
Taken collectively, these results support the concept that in silico pathway predictions using gene expression profiles are associated with activation of the relevant pathway in vitro.
To assess the clinical relevance of the identified pathway subgroups, we investigated if patterns of pathway co-activation as illustrated in the heatmaps of the different cohorts might be related to patient survival. We used overall survival data from Cohort 1 and Cohort 2 and stratified patients by their predicted patterns of pathway activation. A primary GC profile was defined as showing high activation level of a pathway when the activation score was above zero – i.e. being positively associated with the pathway signature. Patient groups stratified by either the proliferation/stem cell pathway activation score alone or the NF-κB pathway activation score alone did not differ significantly regarding their overall survival (p>0.05 for proliferation/stem cell and NF-κB in both cohorts, Figure 5A and 5B). However, when the pathway activation scores were combined, patients with high activation levels of both NF-κB and proliferation/stem cell pathways had significantly shorter survival compared to patients with low activation levels of both NF-κB and proliferation/stem cell pathways (p = 0.0399 and p = 0.0109 for Cohorts 1 and 2 respectively, Figure 5D).
Activation of the Wnt/β-catenin pathway was significantly associated with patient survival in Cohort 1, (p = 0.0056, Figure 5C) but not in Cohort 2 (p = 0.0693, Figure 5C). However, patients in Cohorts 1 and 2 with high activation levels of both Wnt/β-catenin and proliferation/stem cell pathways had significantly worse survival compared to patients with low activation levels of both pathways (p = 0.0073 and p = 0.0086, Figure 5E). To benchmark the contributions of the pathway combinations against known histopathologic criteria, we performed a multivariate analysis including combined pathway predictions and pathological tumor stage (TNM classification: stages 1–4), the most important prognostic factor in GC [28]. In both cohorts, combined activation of proliferation/stem cell and NF-κB pathways proved to be a prognostic factor independent from tumor stage (p = 0.003 and 0.048 for Cohorts 1 and 2, respectively) (Table S6). Likewise, combined activation of proliferation/stem cell and Wnt/β-catenin pathways was an independent prognostic factor in Cohort 1 and achieved borderline significance in Cohort 2 (p<0.001 and p = 0.058, Table S7). These results demonstrate that the assessment of the combined pathway activation status is clinically relevant and moreover can provide additional prognostic information over and above the current gold standard of patient prognosis prediction, the TNM based tumor staging.
In this study, we sought to subdivide GCs into molecularly homogenous subgroups as a first step to individualizing patient treatments and improving outcomes. Importantly, unlike previous GC microarray studies relating gene expression patterns to histology or anatomical type [10],[11], we chose to base our GC subdivisions on patterns of oncogenic pathway activity. After developing and validating this novel classification approach, we were able to describe, for the first time, a genomic taxonomy of GC based on patterns of oncogenic pathway activity. Our approach is particularly suited for gene expression microarrays, since these platforms interrogate thousands of mRNA transcripts in each sample, thereby permitting the assessment of multiple pathways simultaneously in a single experiment. In contrast, such an approach is not currently possible at the protein level due to lack of appropriate platforms. Using this strategy, we identified three dominant pathways showing activation in the majority (>70%) of GCs: proliferation/stem cell, Wnt/β-catenin, and NF-κB signaling.
The ability to perform such “high-throughput pathway profiling” opens many interesting avenues. For example, several studies have previously reported inconsistent results regarding the prognostic impact of different oncogenic pathways in GC - the prognostic implications of proliferation-related antigens such as Ki-67 in GC are not firmly established [29], and high NF-κB activation in GC has been associated with both good and bad GC patient outcome in different studies [7],[30]. It is quite possible that some of this inconsistency may have been due to a historical focus on using conventional methods and analyzing either single pathways or individual pathway components (genes/proteins). Our observation that pathway combinations are predictive of patient outcome suggests that pathway combinations, rather than single pathways alone, may play a critical role in influencing tumor behavior.
Another benefit of high-throughput pathway profiling is the ability to define higher order relationships between distinct oncogenic pathways. In the current study, we consistently observed concomitant activation of E2F, MYC, p21(-repression), and stem cell pathways in tumors (the ‘proliferation/stem cell’ pathway cluster). This is most likely due to increased cellular proliferation in tumor cells, as E2F is important in cell proliferation control and MYC is both a p21-repressor and inducer of cyclin D2 and cyclin-dependent kinase binding protein CksHs2 [31]. Furthermore, stem cells, particularly embryonic stem cells (ESCs), are also known to exhibit high cell proliferation rates [32]. More intriguingly, we also observed close associations between apparently functionally different pathways, such as β-catenin and SRC, as well as HDAC inhibition and BRCA1. Such pathway co-activation patterns may suggest functional interactions between these pathways, which deserve to be studied further. For example, it is possible that activated c-SRC may enhance the expression of the Wnt signaling pathway [33]. Exploring the relationships between pathways showing co-activation may thus provide valuable information regarding the ability of the cancer cell to coordinate the activity of multiple pathways.
A third benefit of the pathway profiling approach is that it facilitates identification of major disease-related pathways. Of the pathways analyzed in this study, the finding that NF-κB signaling may be elevated in a significant proportion of GCs deserves some attention as this pathway has been relatively less explored in GC. Interestingly, while we observed a significant difference in both p50 and p65 (the NF-κB subunits) gene expression between NF-κB/on and NF-κB/off GCCLs, we did not observe overt differential p50 protein expression in these lines, in contrast to p65 (Figure 4C). This may be due to a combination of three reasons. First, the absolute range of p65 gene expression across the cell lines is markedly greater than the absolute range of p50 gene expression (>3×, Figure S4). Second, the Western blotting assay used to perform these protein measurements is known to be highly non-quantitative, which may mask subtle differences in expression. Third, beyond gene expression, p50 expression is also subject to a variety of post-transcriptional regulatory mechanisms such as precursor cleavage that might affect the final level of p50 protein, while p65 is not generated from a precursor protein [34]. NF-κB has been shown to be activated by H. pylori [35], a known GC carcinogen, and aberrant NF-κB signaling has also been implicated in multiple inflammation-linked cancers such as GC [36]. NF-κB has been suggested to be constitutively activated in primary gastric cancers in a few studies [7]. Targeted NF-κB-inhibitors are currently being actively developed in many anticancer drug development programs and a subset of GC patients (i.e. those with elevated NF-κB activity) may represent a suitable subclass for evaluating the efficacy of these compounds.
The in silico method used in our study is conceptually similar to the work of Bild et al, which used a binary regression model to classify tumors based on the predicted activity of five oncogenic pathways [9]. Unlike binary regression, our approach, which makes use of a rank-based connectivity metric [37], requires no elaborate training process on each pathway signature and also does not require the availability of raw expression data, facilitating the use of the many publicly available pathway signatures in the literature [27]. However, the gene expression-based approach does have limitations. First, because our pathway predictions are based on gene expression rather than proteins, such predictions are admittedly molecular surrogates of true pathway signaling activity. Second, we are currently limited to analyzing known oncogenic pathways previously identified in the literature. Third, although we were able to use pathway signatures from very different tissue contexts to predict pathway activation status, an examination of the initial proof-of-principle breast cancer examples revealed that the association of ER status to estrogen responsiveness as predicted using the osteosarcoma signature, although significant, was markedly weaker compared to the association of ER status to tamoxifen sensitivity predicted using a signature derived from the same tissue type (i.e. breast). This result implies that there may also exist tissue-specific differences in pathway signatures that may affect prediction accuracy. Fourth, compared to our study which focused on pathways of known biological relevance in GC, it is unclear if this method can be applied to diseases where prior knowledge of involved pathways may not be available. However, it should be noted that a wealth of pathway signatures (>1000) associated with diverse biochemical and signaling pathways already exists in the literature, which can be accessed from public databases such as MSigDB (http://www.broad.mit.edu/gsea/msigdb/genesets.jsp?collection=CGP). Since our approach can be applied to virtually any disease dataset for which gene expression information is available, testing every signature in a high-throughput manner for evidence of pathway deregulation is both conceivable and feasible. In such cases, pathway exhibiting high frequencies of deregulation would then represent candidate pathways involved in the disease in question, which can then be targeted for focused investigation and experimentation. Addressing these issues will form the ground for much future research.
In conclusion, we have shown in this work that pathways signatures can be successfully used to predict the activation status of cellular signaling pathways, even in biological entities as complex as a human GC. One obvious immediate application of such pathway-based taxonomies may relate to the use of targeted therapies. Initial trials assessing the role of targeted therapies in GC have demonstrated only modest results [38]; however, most of these studies have been performed without pre-stratifying patients using molecular or histopathologic criteria. Pathway-based taxonomies may prove very useful in developing personalized treatment regimens for different subgroups of GC, since such oncogenic pathway activation patterns can be readily linked to potential pathway inhibitors and targeted therapies.
Three cohorts of gastric cancer were profiled: Cohort 1–70 tumors (Peter MacCallum Cancer Centre, Australia), Cohort 2–200 tumors (National Cancer Centre, Singapore), and Cohort 3–31 tumors (Leeds Institute of Molecular Medicine, United Kingdom). All GCs were collected with approvals from the respective institutions, Research Ethics Review Committee, and signed patient informed consent. Histopathological data of all GC cohorts are provided in Table S8, S9, S10. The median follow-up period was 16.89 months for Cohort 1 and 13.47 months for Cohort 2. 43 patients from Cohort 1 and 91 from Cohort 2 were dead at the end of the study period.
A total of 25 unique GC cell lines were profiled. GC cell lines AGS, Kato III, SNU1, SNU5, SNU16, NCI-N87, and Hs746T were obtained from the American Type Culture Collection and AZ521, Ist1, TMK1, MKN1, MKN7, MKN28, MKN45, MKN74, Fu97, and IM95 cells were obtained from the Japanese Collection of Research Bioresources/Japan Health Science Research Resource Bank and cultured as recommended. SCH cells were a gift from Yoshiaki Ito (Institute of Molecular and Cell Biology, Singapore) and were grown in RPMI media. YCC1, YCC3, YCC6, YCC7, YCC10, YCC11, and YCC16 cells were a gift from Sun-Young Rha (Yonsei Cancer Center, South Korea) and were grown in MEM supplemented with 10% fetal bovine serum (FBS), 100 units/mL penicillin, 100 units/mL streptomycin, and 2 mmol/L L-glutamine (Invitrogen).
Total RNA was extracted from cell lines and primary tumors using Qiagen RNA extraction reagents (Qiagen) according to the instructions of the manufacturer. Cell line and primary tumor mRNAs from Cohort 1 and Cohort 2 were hybridized to Affymetrix Human Genome U133 plus Genechips (HG-U133 Plus 2.0, Affymetrix), while primary tumor mRNAs from Cohort 3 were profiled using U133A Genechips (HG-U133A, Affymetrix). All protocols were performed according to the instructions of the manufacturer. Raw data obtained after chip-scanning was further processed using the MAS5 algorithm (Affymetrix) available in the Bioconductor package simpleaffy. The microarray data sets are available at http://www.ncbi.nlm.nih.gov/projects/geo/ (Accession: GSE15460).
All signatures used in this study were previously generated [9], [19], [20], [22]–[24], [39]–[49], and obtained from either the MSigDB database [27] (http://www.broad.mit.edu/gsea/msigdb/genesets.jsp?collection=CGP) or original references [9],[39]. Detailed descriptions of the signatures and their sources are available in Table S11 and Table S12. Each signature is represented by a geneset, termed a query signature (QS). Depending on the signature, a QS may consist of only up- (or down-)regulated genes, i.e. genes up- (or down-)regulated during the activation of the pathway. It may also consist of both up- and down-regulated genes. Our approach is capable of handling all of the aforementioned types of QS. QSs were mapped to the probeset domain of the cancer profiles (HG-U133A or HG-U133 Plus 2.0) before computing the pathway activation scores for the cancer profiles. Mapping of QSs were performed using the probe mapping (‘.chip’) files available from ftp://gseaftp.broad.mit.edu/pub/gsea/annotations [27].
For each pathway, we used whenever possible multiple signatures from independent studies, to minimize the possibility of laboratory-specific effects. For further analyses (e.g. survival comparisons), we used the mean of activation scores across independent signatures belonging to the same pathway or group of pathways in order to determine the final activation status of the pathway or group of pathways.
Two pathway activation signatures [19],[20] (Table S11) related to the estrogen signaling pathway were analyzed. The breast cancer cell line dataset [18] was obtained from http://www.ebi.ac.uk/microarray-as/ae/download/E-TABM-157.raw.zip. Activation scores for breast cancer cell lines were computed by comparing each individual line against the median profile of the collection of 51 breast cancer cell lines. P-values for the validation of our predictions against ER status were computed using Pearson's chi-square test, under the null hypothesis that the pathway predictor delivers comparable performance to a random predictor.
20 signatures [9], [19], [20], [22]–[24], [39]–[49] (Table S12) representing the activation of 11 pathways related to gastric carcinogenesis were analyzed. Activation scores for primary GCs were computed by comparing each individual GC against the median profile of the patient cohort being analyzed. For the analysis of GC cell lines, final activation scores were obtained by computing the mean activation scores across the seven reference profiles (Table S4). Unsupervised hierarchical clustering (average linkage with centered Pearson correlation metric) was applied to establish patterns of co-activation between different pathways using BRB-ArrayTools.
Pathway activation scores were computed using two inputs: 1) cancer profiles, comprising lists of probesets sorted by differential gene expression between individual cancer gene expression profiles and a reference profile (see Text S1), where n is defined as the total number of probesets in each cancer profile i, and 2) a query signature QS (pathway activation signature). Probesets representing either up- (or down-) regulated genes in the QS are defined as ‘tags’, and t the number of tags in the up- (or down-) regulated portion of the QS. Raw enrichment scores were computed using a Kolmogorov-Smirnov metric previously described in [37]. Here, ‘direction’ in may be considered as ‘up’ or ‘down’, depending on whether the set of tags in question represents the up-regulated () or the down-regulated () portion of the QS. For a cancer profile i and a set of t QS tags, the position of tag j in the cancer profile i is defined as V(j), forming the vector V.(1)The elements of V are then sorted in ascending order of V(j) such that . In this manner, the tags indexed by j are ordered based on their position in the cancer profile (e.g. tag 1 is the probeset with the highest rank in the cancer profile among all t tags in the up- (or down-) regulated portion of the QS). Using the sorted elements of V, two parameters are computed:(2)(3)If , is set to a. Otherwise, (if ), is set to −b.
To compute the pathway activation score , if and have the same signs then for cancer profile i is set to zero. Otherwise, the raw activation score is obtained.(4)The maximum and minimum of across all cancer profiles in the cohort are defined as p and q, respectively. The activation score is the normalized form of , where(5)and(6)In cases where more than one profile exists for a sample, the final activation score represents the mean activation score across the replicate profiles.
Cell proliferation assays were performed on 22 lines (except SNU1, SNU5, and SNU16) using a CellTiter96 Aqueous Nonradioactive Cell Proliferation Assay kit (Promega) following the manufacturer's instructions. Briefly, cell lines were plated at concentrations of 1×103 to 5×103 cells per well in 96-well plates. Growth rates, representing proliferative activity, were analyzed after 48 hours.
Western blotting was performed as previously described [50] using the following antibodies and dilutions: 1∶500 β-catenin (catalogue number 06-734, Upstate), 1∶500 TCF7L2 (05-511, Upstate), 1∶1,000 β-actin (sc-8432, Santa Cruz), 1∶500 p65 (sc-372, Santa Cruz), 1∶500 p50 (sc-1191, Santa Cruz), and 1∶1,000 GAPDH (ab9483, Abcam). Processing of cell line TMAs (tissue microarrays), blocking, and antigen retrieval was performed as previously described [51]. p65 antibodies were incubated at a dilution of 1∶50 for 2 hours at 37°C. Signal detection was performed using the REAL system (DAKO) at 37°C for 30 minutes, using the DAB chromogen (1∶50 dilution), and Mayer's haematoxylin counterstain. The slides were scored by an experienced histopathologist (H.G.) and the percentage of positive nuclei, percentage of cells with cytoplasmic staining, and staining intensity were assessed.
TOPFLASH assays for validation of Wnt/β-catenin activation were performed as previously described [50]. For validation of NF-κB activation, MKN1, MKN7, Hs746T, AGS, and SCH cells were transfected with a pNFκB-Luc reporter (Clontech, Cat. No. 631904) using FuGENE 6 Transfection Reagent (Roche) in 96-well plates. pNFκB-Luc contains the Photinus pyralis luciferase gene and multiple copies of the NF-κB consensus sequence fused to a TATA-like promoter region from the Herpes simplex virus thymidine kinase promoter. The same cells were also transfected with pGL4.73[hRluc/SV40] vector (Promega) as a normalization control. Cells were collected 48 hours after transfection and luciferase activity was measured using a dual-luciferase reporter assay system (Promega). All experiments were repeated three independent times.
Kaplan-Meier analysis (SPSS, Chicago) was used for survival comparisons of patient cohorts where clinical follow-up and mortality information were available. P-values representing the significance of the differences in survival outcome (metric: overall survival) were calculated using the Log Rank (Mantel-Cox) test, with p-values of <0.05 being considered significant. Cox regression models were used for computing hazard ratios and implementing multivariate analyses including combined status of two pathways and overall tumor stage (TNM classification: 1–4) as variables. Patients from Cohorts 1 and 2 analyzed in survival comparisons exhibit a significant relationship between overall survival and overall tumor stage, suggesting that patient selection is likely non-biased (data not shown). P-values denoting the significance of a correlation coefficient R between two N-element vectors were estimated from the Student t-distribution, against the null hypothesis that the observed value of t = R/√[(1−R2)/(N−2)] comes from a population in which the true correlation coefficient is zero. Unless otherwise specified, all other p-values (used in comparisons of two groups) were computed using Student's t-test. All p-values are two-tailed. Gene Set Enrichment Analysis (GSEA) was performed as described in Subramanian et al. [27].
|
10.1371/journal.pcbi.1003973 | Comprehensive Sieve Analysis of Breakthrough HIV-1 Sequences in the RV144 Vaccine Efficacy Trial | The RV144 clinical trial showed the partial efficacy of a vaccine regimen with an estimated vaccine efficacy (VE) of 31% for protecting low-risk Thai volunteers against acquisition of HIV-1. The impact of vaccine-induced immune responses can be investigated through sieve analysis of HIV-1 breakthrough infections (infected vaccine and placebo recipients). A V1/V2-targeted comparison of the genomes of HIV-1 breakthrough viruses identified two V2 amino acid sites that differed between the vaccine and placebo groups. Here we extended the V1/V2 analysis to the entire HIV-1 genome using an array of methods based on individual sites, k-mers and genes/proteins. We identified 56 amino acid sites or “signatures” and 119 k-mers that differed between the vaccine and placebo groups. Of those, 19 sites and 38 k-mers were located in the regions comprising the RV144 vaccine (Env-gp120, Gag, and Pro). The nine signature sites in Env-gp120 were significantly enriched for known antibody-associated sites (p = 0.0021). In particular, site 317 in the third variable loop (V3) overlapped with a hotspot of antibody recognition, and sites 369 and 424 were linked to CD4 binding site neutralization. The identified signature sites significantly covaried with other sites across the genome (mean = 32.1) more than did non-signature sites (mean = 0.9) (p < 0.0001), suggesting functional and/or structural relevance of the signature sites. Since signature sites were not preferentially restricted to the vaccine immunogens and because most of the associations were insignificant following correction for multiple testing, we predict that few of the genetic differences are strongly linked to the RV144 vaccine-induced immune pressure. In addition to presenting results of the first complete-genome analysis of the breakthrough infections in the RV144 trial, this work describes a set of statistical methods and tools applicable to analysis of breakthrough infection genomes in general vaccine efficacy trials for diverse pathogens.
| We present an analysis of the genomes of the HIV viruses that infected some participants of the RV144 Thai trial, which was the first study to show efficacy of a vaccine to prevent HIV infection. We analyzed the HIV genomes of infected vaccine recipients and infected placebo recipients, and found differences between them. These differences coincide with previously-studied genetic features that are relevant to the biology of HIV infection, including features involved in immune recognition of the virus. The findings presented here generate testable hypotheses about the mechanism of the partial protection seen in the Thai trial, and may ultimately lead to improved vaccines. The article also presents a toolkit of methods for computational analyses that can be applied to other vaccine efficacy trials.
| The HIV pandemic is responsible for more than 34 million deaths worldwide. Analysis of the RV144 vaccine trial yielded an estimated efficacy to prevent HIV infection of 31%, with a 95% confidence interval (CI) of 1% to 51% [1]. In this phase III efficacy trial, 16,402 Thai HIV-1-negative volunteers were randomized to receive a prime-boost vaccine regimen that consisted of four priming injections of a recombinant canarypox vector [ALVAC-HIV vCP1521: subtype B gag, pro (from HIV-1 strain LAI) and CRF01_AE gp120 (92TH023)], and two booster injections of a recombinant gp120 subunit vaccine [AIDSVAX B/E: subtype B (MN) and CRF01_AE (CM244)]. Follow-up studies highlighted possible mechanisms behind the modest RV144 protection. Multiple sources of evidence indicated a role for vaccine-induced antibody responses targeting the V2 region of the envelope glycoprotein (Env): (1) the case-control study of immune correlates of risk showed that the magnitude of IgG antibodies binding to the V1/V2 region of Env was inversely correlated with risk of infection [2–5]; (2) the magnitude of binding of IgG antibodies to linear peptides in the V2 loop was inversely correlated with risk of infection [3,6]; and (3) sieve analysis targeted to the V2 region (described below) demonstrated vaccine pressure at two sites [7]. The case-control correlates study also showed that IgA antibodies to envelope and to the C1 region of Env were directly correlated with risk of infection [3]. In addition, among vaccine recipients with low IgA antibody responses to Env, HIV-1 infection risk was inversely correlated with IgG Env antibody avidity, antibody-dependent cellular cytotoxicity, neutralizing antibodies, and Env-specific CD4+ T cell responses [3], as well as with IgG to V3 linear peptides [6].
“Sieve analysis” is the statistical assessment of whether and how the efficacy of a vaccine depends upon characteristics of the pathogen. Genomic sieve analysis compares breakthrough HIV-1 sequences between the infected vaccine and infected placebo groups. A sieve analysis of the HVTN 502/Step trial, with a vaccine inducing cytotoxic T-lymphocyte (CTL) responses, found evidence of CTL epitope-specific variation [8,9]. Based on HIV-1 breakthrough infections in the RV144 trial sequenced at the time of HIV-1 diagnosis, a sieve analysis that focused on the V1/V2 region of Env identified two sites in the V2 loop (HXB2 amino acids Env 169 and 181) at which the level of efficacy of the vaccine significantly differed depending on whether the genome of the infecting HIV-1 virus matched the vaccine immunogen sequence at the site [7].
Here we present a comprehensive genome-wide exploratory sieve analysis of the breakthrough HIV-1 sequences of 109 of the 110 RV144 participants who were infected with CRF01_AE (excluding one subject whose infection was epidemiologically linked and secondary to another study participant’s infection [7]). Our investigation was based on a pre-specified analysis plan, and included multiple sieve analysis methods, each of which evaluates a different immunological hypothesis (Fig. 1). The analysis focused on amino acid (AA) site-, peptide-, and protein-specific methods, with investigation of (1) differential deviation (vaccine versus placebo) from the immunogen sequences at specific loci or in peptide regions that are relevant to antibody binding; (2) differential vaccine efficacy versus HIV-1 sequences that do not match immunogen sequences at individual sites and in each of several pre-specified antibody-relevant protein regions; (3) differential codon selection, and differences in physico-chemical properties across treatment groups; (4) greater or more rapid viral escape (vaccine versus placebo) at predicted class I and class II HLA-restricted T cell epitopes; and (5) differences in phylogenetic diversity of the breakthrough amino acid sequences or differential evolutionary divergence from the vaccine immunogen sequences. The results of these analyses generate testable hypotheses about the mechanisms underlying the modest protection induced by the RV144 vaccine regimen and about potential paths to more effective HIV-1 vaccines to be investigated in future research.
We applied an array of methods designed to evaluate distinct hypotheses regarding vaccine-induced effects on the genetic sequences of the breakthrough HIV-1 viruses. The methods and their relative merits are outlined in Fig. 1 and in the Materials and Methods section. All analysis methods used here have been described previously except for the “Expected Gilbert, Wu, Jobes” (EGWJ) method, the “Quasi-Earth Mover’s Distance” (QEMD) method, and the “Physico-chemical Properties” method (PCP), which are described here for the first time. The EscapeCount method was also developed for this analysis; it has been reported previously [10], but is described more thoroughly here. This paper is the also first published application of the PRIME method (http://hyphy.org/w/index.php/PRIME).
All results reported here have not been reported previously except for the V2 crown signature sites (Env 169 and Env 181) [7], and the V3 signature site (Env 317) [11]. In addition, one epitope region found by the EscapeCount method (at the crown of the V2 loop) was reported previously [10]. The SmoothMarks method has been applied to evaluate a related genetic distance, but as described below the results shown here are novel.
Our dataset includes genome sequences from 109 of the 110 subjects previously identified with HIV-1 CRF01_AE infections [7]. All sequences were obtained from the earliest available sample for sequencing, and all were prior to the subjects’ initiation of antiretroviral therapy. There were a median of 10 sequences available for analysis per subject in vaccine proteins (S1 Table) and also a median of 10 in non-vaccine proteins (S2 Table). We found 19 signature AA sites contained within the vaccine immunogens (Fig. 2) – 12 in Env-gp120, 3 in Gag and 4 in Pro – that showed a p-value ≤ 0.05 in at least one of the three primary site-scanning methods (Table 1). In addition, proteins that were not included in the vaccine immunogens were scanned for sieve effects versus the consensus CRF01_AE sequence (CON-AE) [12]: 37 sites were significant by at least one of the primary methods, and these were distributed across all non-vaccine proteins (Table 2 and Fig. 2; complete results in S1 Dataset). Four pairs of sites overlap different proteins across reading frames: one pair in the immunogen region, three pairs outside of the immunogen region. These overlapping sites are described in the “context” columns in Tables S3 and S4. 10 of the 19 immunogen signature sites were more likely to match a vaccine immunogen AA in the placebo group (a “vMatch” or “typical” sieve effect), and 10 of 19 in the vaccine group (a “vMismatch” or “atypical” sieve effect), with one site (Env 369) having both a vMatch effect vs the CRF AE immunogen AA and a vMismatch effect versus the subtype B immunogen AA, see Fig. 3; additional information about each site is provided in S3 Table (see Figs. 4 and 5 and S4 Table for non-immunogen signature sites). In contrast to the hypothesis-driven V1/V2 study, in this exploratory analysis we used uncorrected p-values at the 0.05 significance level to identify putative signature sites, a strategy taken to maximize sensitivity. To control for false positives, we used a conventional 0.20 false discovery rate (FDR) significance threshold [13], evaluated separately by gene for each analysis method. Only 1 of the 19 signature sites within the immunogen region, Pol 51, had q-value < 0.20 (Table 1, Fig. 2).
The first of the three primary site-scanning sieve analysis methods, differential vaccine efficacy (DVE), uses Cox survival analysis to test whether vaccine efficacy (VE) for preventing infection by viruses that are AA-matched to the vaccine immunogen sequence at a particular locus is significantly different from the VE versus mismatched infections, [14,15]. Point estimates of VE associated with each mutation at which the DVE is significant show that VE can be eliminated or greatly strengthened with the mutation of a single residue (Table 3 and Table 4). Because this method evaluates all trial time-to-event data (including all randomized subjects HIV negative at baseline) and yielded a p-value for differential VE close to that for testing overall VE, the strength of evidence is comparable to the strength of evidence for overall efficacy, with the important caveat that multiple testing could lead to false discoveries due to chance variation. The other two primary methods compare the AA distribution of breakthrough infections at an individual site. Both methods employ numeric weights determined by the substitution frequency of the immunogen AA to the breakthrough sequence AA [16]. The Gilbert, Wu, Jobes (GWJ) method compares these substitution weights across treatment groups [17], and the Model-Based Sieve (MBS) method employs the weights in a Bayesian model comparison that is more sensitive to detect treatment effects that alter the distribution among non-vaccine-matched amino acid categories [18]. These three primary methods evaluate a single representative sequence (the mindist sequence) per subject. This sequence, chosen as the closest actual sequence to the consensus of a subject’s multiple sequences (S1 Text), is selected to represent the founder of the subject’s infection. To more fully represent the viral population, two secondary site-scanning methods utilize all available sequence data: the Mismatch Bootstrap (MMBootstrap, or simply MMB) method, which compares the frequency of vaccine-mismatched AAs in all of the subjects’ sequences across treatment groups (this is the method employed previously in the V2-focussed analysis [7]), and the new Expected GWJ (EGWJ) method that generalizes the GWJ method by replacing subject weights with weight averages over a subject’s multiple sequences (Table 1 and Table 2). Only 9 of the 19 sites identified in proteins represented in the vaccine were significant with all five scanning methods: five in Env (19, 181, 317, 369, 424), along with site 11 in Gag and sites 12, 44 and 51 in Pro, reflecting the variety of alternative hypotheses tested by the five methods (Fig. 1).
In a related analysis, we used 9-mer scanning (the KmerScan method as previously described [8]) to compare all 9-mers in subjects’ sequences to the corresponding 9-mer in each reference sequence (the vaccine sequences for immunogen proteins and the CRF01_AE consensus sequence, CON-AE, for non-immunogen proteins). This analysis evaluates contiguous amino acids that could be the target of a CTL response, but does so without incorporating subject-specific HLA information. For a given pair of 9-mers the similarity score was the sum of HIVb similarity scores [16] over the nine sites. We compared the distribution of these scores for all of the individual sequences between the infected vaccine group and the infected placebo group. S5 Table lists 9-mers that had significantly different similarity to a vaccine immunogen sequence 9-mer across treatment groups (38 9-mers) and S6 Table lists 9-mers outside of the vaccine immunogen regions that significantly differed versus CON-AE (82 9-mers). Thirty 9-mers in Tat (involving sites 1–72) and one 9-mer in Vpu (sites 30–38) passed the pre-specified 20% q-value multiplicity adjustment threshold. The significant 9-mers in Tat comprised seven distinct contiguous regions ranging in length from 9 to 25 AA. In four of these seven regions there was a signature site that could explain the 9-mer scanning results (with concordance of “vMatch” or “vMismatch” sieve effects) while in three of the regions at least one of the 9-mers did not overlap a signature site.
We sought to test the hypothesis that vaccine efficacy declined as a function of the distance between the HIV-1 breakthrough viruses and the immunogen sequences. The SmoothMarks method [19,20] (S2 Text) evaluates VE as a continuous function of each of several distances between the mindist sequences and each immunogen sequence. In addition to distances corresponding to all gp120 sites, we considered four of the pre-specified immunologically-relevant subsets of gp120 amino acid sites: contactsites, contactsites-augmented, hotspots, EPIMAP (Table 5). For all of these analyses the first 41 sites of Env were excluded, because they were not present in CM244 and MN and the first approximately 30 sites corresponded to the signal peptide, which is cleaved from the mature protein. S1 Fig. shows boxplots of the genetic distances for the vaccine and placebo groups for the five sets of sites against the two CRF01_AE vaccine sequences (92TH023 and CM244), computed using the HIVb substitution matrix. These distances are tightly correlated with Hamming distances (percent amino acid mismatch), with Spearman rank correlations ranging between 0.91 and 0.95 across the 10 distances. The distances are approximately equal when measured to the 92TH023 and CM244 reference sequences, and all of the distances except hotspots are approximately equal across the sets, whereas the hotspots distances tend to be lower. The median (range) number of amino acid mismatches to the reference sequences are 13.4 (4.3–24.6) per 100 sites for all of the distances except the hotspots distances, and are 9.2 (3.6–14.8) per 100 sites for the hotspots distances. The likely reason for the closer hotspots distances is that the linear peptides used to measure antibody binding reactivity included 7 distinct HIV-1 subtypes, indicating that hotspots sites are sites with cross-subtype-reactivity, and such sites are expected to be relatively conserved because the vaccine can more readily induce cross-reactive antibodies to more conserved peptides. While the hypothesis testing analyses presented next are of main interest given they assess vaccine efficacy directly, we note that the distances between HIV-1 breakthrough and immunogen sequences did not significantly differ between infected vaccine recipients and infected placebo recipients.
Vaccine efficacy was estimated as a function of genetic distance v for each of the ten distances (Fig. 6 for contactsites, S2 Fig. for all ten distances). A similar analysis of a subset of these antibody contact sites was previously reported in Gilbert and Sun [20], using a set of monoclonal antibody contact sites that was current through 2011; here we analyzed distances of Ab contact sites based on information that is up-to-date as of August 2014. The tests for distance-variability of vaccine efficacy were all non-significant (p-values > 0.20). These results support no strong sieve effects but cannot rule out moderate sieve effects, as power calculations showed that for the setting of the RV144 trial, the SmoothMarks method has only 50% power to detect VE declining from 67% to 0%. However, these distance-based analyses contribute additional evidence supporting the hypothesis that the vaccine regimen conferred some protection. In particular, overall vaccine efficacy against CRF01_AE HIV-1 ignoring the genetic distances resulted in a p-value for positive VE of 0.026, whereas the tests of Gilbert and Sun [20] for positive vaccine efficacy against at least one HIV-1 genotype (10 tests) gave p-values ranging from 0.006 to 0.024, with median p = 0.013. This shows that accounting for the genetic distances increased the evidence for positive vaccine efficacy against CRF01_AE HIV-1.
To complement the site-scanning sieve analysis that identified individual Env-gp120 signature sites as potential discriminators of vaccine efficacy, we combined the signature sites into a global distance and assessed how the vaccine efficacy varied with this distance. In particular, the global sieve analysis above was repeated for the distances calculated over just the 10 identified Env-gp120 signature sites (listed in Table 1), excluding Env 6 and Env 19 because they are not in the CM244 and MN inserts and they are part of the signal peptide. Because 5 of the 10 signature sites had “vMatch” sieve effects and 5 had “vMismatch” sieve effects, we do not expect the vaccine efficacy curves to exhibit the “classical” sieve effect shape with vaccine efficacy highest for smallest distances and waning to zero for the greatest distances; rather the vaccine efficacy curves could take many shapes depending on the joint distribution/covariation of the amino acids at the 10 sites, and the curves provide new information about the aggregate impact of the non-contiguous decapeptides on vaccine efficacy.
S3 Fig. shows the distributions of the signature-site distances to 92TH023 and CM244 together with the estimated vaccine efficacy curves. For 92TH023, the estimated curve is approximately horizontal, indicating that the 5 “vMatch” and 5 “vMismatch” signature sites have a “balancing” effect, with the net impact being that the combined decapeptide patterns do not associate with vaccine efficacy. However, for CM244 the estimated VE peaks against HIV-1s with an intermediate number of mismatches to the vaccine (zenith at estimated VE = 59% for genetic distance 0.28, an average of 2 mismatched residues) and declines to zero against HIV-1s with increasing distance (estimated VE = 0% for genetic distance 0.53, an average of 5 mismatched residues). To help interpret this relationship, S4 Fig. shows the signature site decapeptide AA patterns for each of the 109 infected subjects, aligned to the vaccine efficacy curve for reference. S4 Fig. indicates that the “vMismatch” signature site Env 413 has the greatest influence to create the increasing VE curve in the distance region 0.066 to 0.166 and the vMismatch signature sites Env413, Env 268, and Env 317 have the greatest influence to create the declining VE curve in the distance region 0.28 to 0.53.
To search for functional sequence differences in the vaccine and placebo groups, we evaluated treatment-group differences in the physicochemical properties of amino acids in the mindist sequences. Unlike the methods with results presented in Table 1 and Table 2, which compare divergences of breakthrough AA from a vaccine AA between treatment groups, the physicochemical properties (PCP) method compares the sequences between groups directly, without regard for the vaccine reference sequences, on a per-property basis. We evaluated each of two different property scales: (1) the vector of ten indicator values determined by Taylor [21], indicating the presence or absence of ten particular physicochemical properties for each amino acid; and (2) the vector of five “z scales” for each amino acid, principal components of observed physicochemical properties used to determine quantitative structure-activity relationships between peptides [22–24]. We scanned the sequences at individual sites (S7 and S8 Table) as well as at contiguous 3-mers (S9 and S10 Table) and 9-mers (S11 and S12 Table) across the HIV-1 proteome, comparing counts of each of the ten Taylor properties and five z-scale components across treatment groups (complete results are included in S1 Dataset). The results of this method can help interpret the physicochemical and structural differences between the vaccine and placebo viral populations.
Of the 19 vaccine immunogen signature sites shown in Table 1, only Pol 51 was also found to have site-specific significant PCP results (S3 Table), and of the 37 out-of-immunogen signature sites shown in Table 2, eight coincided with site-specific PCP results (S4 Table). A total of 16 individual sites were significant at the p ≤ 0.05 level (S7 and S8 Table), two of which with q-values below 0.2: Pol 51 as noted above (property z3, q = 0.19) and Vpu 30 (hydrophobicity, q = 0.10). These two sites were also the only locations with q-values below 20% in the scanning of 3-mer peptides (Peptide starting at Pol 49: property z5, q = 0.024; Peptides starting at Vpu 28, 29 and 30: hydrophobicity, q = 0.050). Additional sites had q-values below 20% when scanning 9-mer peptides, all of which were located in the non-immunogen proteins, concordant with the KmerScan 9-mer results. A negatively charged region in the vicinity of Nef 150 differed between the treatment groups (q = 0.059) as did component z4 in the vicinity of Tat 73 (q = 0.19). In addition, hydrophobic residues in the vicinity of Vpu 30 spanning positions Vpu 25 through Vpu 30 differed between the treatment groups (the q-values in this region ranged from a minimum of q = 0.021 for the 9-mer starting at Vpu 25 to a maximum of q = 0.23 for the 9-mer starting at position Vpu 29).
To further elucidate the role of selection for particular physicochemical properties, we conducted a codon-based phylogenetic analysis that detected Env-gp120 sites at which natural selection has operated to preserve or change one or more of five physicochemical properties: chemical composition, polarity, volume, iso-electric point or hydropathy [25,26]. To do so, we modeled the rate of nonsynonymous substitution from codon x to codon y, β(x, y) at a site as a function of the difference in properties between x and y: β(x, y) = exp(−∑pcpdp[x, y]), where p indexes the five properties. If cp is significantly different from 0 for a particular property p at a site (measured by a likelihood ratio test, with 5-fold multiple testing correction at each site using the Holm-Bonferroni procedure) along the vaccine-group lineages, then we conclude that the property is preserved (cp >0) or driven to change (cp <0) by natural selection. Fig. 7 shows the sites found to have selection acting on one or more properties, along with signature sites on a crystal structure of Julien et al.[27], and S5 Fig. provides an alternate viewing angle. Notably, at several sites almost all physicochemical properties tested were under selection, including Env 85–87, 353, 365, and 425 (S13 Table and S1 Dataset).
Many of the signature sites described above are located in genomic regions of known functional or immunological relevance. Specifically, those in HIV-1 Env-gp120 include sites in the antibody binding regions at the crowns of the V2 (Sites 169, 181) and V3 (Site 317) variable loops, sites in the co-receptor binding site (Sites 317, 353), and in the CD4 binding loop motif (Site 369). We considered six pre-specified subsets of Env sites known to be immunologically relevant (Table 5). We found that the Env-gp120 signature sites were significantly more likely to be found in a subset of Env sites known for their relevance to neutralization potency (nAb-sites set) (Fisher’s exact test p = 0.0035). Signature sites were also more likely to be part of the focus set that includes only those sites that were identified as hotspots of antibody binding reactivity in RV144 and are either in the known antibody contactsites set or are in the nAb-sites set (p = 0.0049). Results for all six site sets are presented in Table 6.
Using tests for codon selection that identify important biochemical properties as discussed above, as well as a method that estimates the ratio of non-synonymous and synonymous substitution rates (dN/dS) separately in internal and terminal branches of the tree connecting these sequences [25], we found 91 sites in gp160 that were under differential selective pressure between the vaccine and the placebo groups (54 among the 511 sites of gp120 and 37 among the 345 sites of gp41, S13 Table and S1 Dataset), including signature sites Env 6 and Env 353. Interestingly, a high proportion of these sites were located in V3 (8/54, significantly more than the proportion of sites located outside of V3: Fisher’s exact test two-sided p-value = 0.049).
Covariation was assessed within proteins between pairs of sites, both pooling over the vaccine and placebo groups and separately, together with a test for whether the degree of covariation differed for vaccine versus placebo, which could imply vaccine-induced functional or structural constraints (S3 Text and S2 Dataset). Covariation was generally weak, and there was no evidence that covarying site pairs were restricted to the vaccine group. Among gp160 residues, there were 630 covarying site pairs with p < 0.05 but only two had a q-value below 0.2: between sites 276 and 343 (q = 0.10), and between sites 65 and 181 (q = 0.12). Both sites 181 and 343 were identified above as signature sites. An interesting pattern was seen when we considered how many associations (defined as treatment-arm-pooled covariation p < 0.05) linked each residue. While most Env sites (n = 584) showed no association with any other site, 22 sites interacted with more than five other sites (19, 169, 181, 276, 307, 308, 317, 332, 343, 347, 353, 360, 365, 369, 379, 412, 413, 424, 465, 564, 658, 822). Ten of these 22 sites were also signature sites, and there is evidence for a hub of covariation in V2 (S3 Text).
Importantly, the signature sites identified showed significantly more associations with other signature sites (mean = 32.13) than with non-signature sites (mean = 0.88) (p < 0.0001). This difference was also significant if we considered only gp120 (p < 0.0001). Interestingly, there were more associations between residues in gp120 (mean = 2.03), which corresponded to the vaccine sequence, than in gp41, which was not included in the vaccine (mean = 0.63) (p = 0.10; p = 0.001 if zero values are excluded).
We have developed multiple methods to evaluate potential T cell-driven sieve effects based on comparisons of computationally-predicted epitopes in sequences from vaccine and placebo recipients. Results are shown in S14 and S15 Tables. For all methods, we begin by predicting T cell epitopes in the vaccine and breakthrough sequences using the HLA haplotypes of the infected trial participants.
First we evaluated each viral protein using the novel EscapeCount method (S4 Text), which counts the number of high-affinity predicted epitopes in the vaccine sequence that bind with much lower affinity to the corresponding k-mer in the subject’s mindist sequence. Since the power of this method is improved when there is more variability in the predicted epitope binding affinities, for class I predictions we used the adaptive double threading (ADT) epitope prediction software [28] rather than the more well-known NetMHCpan software [29] that we used in this and previous applications of the EpitopeDistance method (discussed below). Using the EscapeCount method, we found significant evidence of greater class I binding escape among placebo recipients in Env versus the CM244 reference sequence than among vaccine recipients (p = 0.031).
We also applied the EscapeCount method to evaluate individual k-mers for evidence of greater class I (9-mers) and class II (15-mers) binding escape in vaccine versus placebo groups (S14 and S15 Tables). Twelve 9-mers (ten in Env and two in Gag) showed an unadjusted p-value <0.05 for differential binding escape, though none of these surpassed the q-value threshold of 20% (0.24 < q-value < 0.93). Seven 9-mers with a q-value < 0.5 were found in Env-gp120 (start positions: 5, 128, 299, 328, 335, 363, 445). In S6 Fig. we show as an example the V3 loop 9-mer “PSNNTRTSI” (PI9, HXB2 start position 299) at which there was a greater number of HLA binding escapes in vaccine versus placebo recipients (p = 0.0084). Vaccine to placebo differences are concentrated in the 9th position, site 307, which forms part of the core of the V3 loop. Although site 307 did not qualify as a signature site, its DVE p-value was 0.065 and its EGWJ p-value was significant at 0.029. This site forms (with Env 308 and 317) the core of the V3 loop and is a well-studied target of antibody binding [6,30]. Since also many of the infected RV144 subjects had HLA types capable of binding the CRF01_AE vaccine sequence epitope, and the variation at site 307 abrogated class I HLA binding in vaccine but not placebo recipients, the V3 sieve effect may be partially due to T cell mediated effects.
As the second of three methods, we applied the PercentEpitopeMismatch procedure, which was applied previously to the HVTN 502/Step sieve analysis [8] (S14 and S15 Tables). This method complements the EscapeCount method by considering any class I epitope that is predicted (for a given subjects’ HLA type) in the vaccine sequence, regardless of whether it is also predicted in the breakthrough sequence, and regardless of the number of changes between the breakthrough and vaccine k-mers. The percent of vaccine-predicted epitopes for which there is any change in the corresponding breakthrough sequence was computed for all of a subject’s sequences and these were compared across treatment groups as previously described [8] (S5 Text). Using the PercentEpitopeMismatch method, we found no evidence of T cell escape in insert-relevant genes when using the NetMHCpan- or ADT-predicted epitopes.
For the third method we applied the EpitopeDistance procedure that was also previously applied for the HVTN 502/Step trial [8]. This method compares the predicted epitopes in each subject’s breakthrough sequences to HLA-matched epitopes estimated in the vaccine sequence (S6 Text). In summary, we found no concordant evidence for a T cell-driven sieve effect across Gag, Pro and Env (S14 Table) or the non-vaccine proteins (S15 Table). However, some significant results were found in the V2 region of Env when binding affinities were considered (CM244 p = 0.022; 92TH023 p = 0.047), although there was only a trend suggesting a difference between the vaccine and placebo groups when evolutionary distances were considered (CM244 p = 0.058; 92TH023 p = 0.23).
To analyze sequences at the gene/protein level, we assessed whether sequences from vaccine recipients were (a) more phylogenetically diverse or (b) more divergent from the vaccine insert sequences than sequences found in placebo recipients. For the phylogenetic diversity, we constructed maximum-likelihood phylogenetic trees using all amino acid sequences available for each subject; we then subset the leaves of these trees to retain only the mindist sequences. For each tree, we then computed the differential amino acid phylogenetic diversity (PD) [31], defined as the difference in the total branch length of two subtrees: one corresponding to placebo recipients and the vaccine insert, and the complementary sub-tree corresponding to vaccine recipients and the vaccine insert, and compared this to an estimated null distribution as described in Methods. We found trending evidence of greater phylogenetic diversity in Gag for the vaccine group compared to the placebo group (p = 0.089) (S17 Table).
In addition to the phylogenetic diversity, we also calculated the phylogenetic divergence from the vaccine sequence based on the same mindist trees (S7 Fig. shows the distributions of these distances to the CM244 reference sequence for each tree). We compared these values across treatment groups for each tree, and found trending evidence of greater divergence from the LAI sequence in Pro among placebo-recipient sequences (p = 0.059) (S18 Table), a “vMismatch” result. We repeated this analysis using the median distance over the multiple sequences available from each subject (n = 3–14) rather than the mindist sequence distance (as previously described [8]), and found that the results were consistent. When applying a variant of this analysis using nucleotide sequence trees, we found a trend toward greater divergence of vaccine recipient Gag sequences to the LAI insert sequence (p = 0.072) and no significant or trending result in Pro.
In addition to the phylogenetic analyses, we also evaluated whether the pairwise alignment similarity scores between all vaccine recipient sequences versus all placebo recipient sequences were different than what would be expected under the null hypothesis that the vaccine and placebo sequences came from the same distribution. We computed the Quasi-Earth Mover’s Distance (QEMD) using BLOSUM90 [32] pairwise alignment scores and compared it to the null distribution of the QEMD computed based on repeatedly permuting treatment assignments; this approach does not use the vaccine reference sequences and does not depend on an estimated phylogeny. While the PD analysis evaluates the across-group difference between within-group phylogenetic diversity, the QEMD analysis evaluates the between-group sequence variation. We found significantly less QEMD similarity in Gag sequences than would be expected under the null hypothesis (p = 0.041), consistent with the trend toward a sieve effect found via the phylogenetic analysis. We did not find significant evidence for QEMD dissimilarity for Pol (p = 0.16) or Env (p = 0.54).
We compared variable loop lengths, numbers of cysteines, and frequencies of potential N-linked glycosylation sites (PNG sites) between vaccine and placebo recipient sequences. There were no significant differences in the mindist variable loop lengths in Env-gp120 between vaccine and placebo recipients in any of the five variable loops (Wilcoxon rank sum p-values > 0.20). Next, based on mindist sequences we compared the per-subject median number of cysteines in gp120 between the treatment groups; this analysis was motivated by the finding in Vax004 that 20% of trial participants had atypical cysteine variants [33]. The distributions of per-subject median numbers of cysteines were similar in the two treatment groups (average number of cysteines = 19.45 in vaccine recipients and 19.73 in placebo recipients).
Next, we identified PNG sites by searching each breakthrough sequence for tripeptide motifs of the form N-X-S or N-X-T, where X is any amino acid other than proline [34]. We compared the numbers of PNG sites between the treatment groups using the mindist sequences as well as with all sequences using the multiple outputation (MO) method [35], and found no significant or trending difference. We also tested for a difference across treatment groups in the distribution of PNG sites at each of the sites at which any subject had a PNG site, restricting to sites with sufficient diversity (defined as at least 4 sequences with a PNG site and at least 4 sequences without a PNG site). Of the 75 sites tested, only one had an unadjusted Fisher’s exact test p ≤ 0.05 (site 186s, p = 0.04). S8 Fig. shows the percentage of mindist sequences with a PNG site at each alignment position that had one or more PNG site.
Sieve analysis is a powerful tool for the evaluation of breakthrough infections in vaccine studies and complements related studies of immune correlates of infection risk among vaccine recipients. Sieve analysis leverages the randomized design of the trial by comparing features of infections across treatment groups, and can further suggest testable hypotheses about the targets of vaccine-induced immunity. By comparing HIV-1 breakthrough viruses that were isolated from vaccine and placebo recipients in the RV144 trial, we identified HIV-1 genetic determinants potentially associated with (unmeasured) vaccine-induced immune responses. Scanning across the HIV-1 proteome, we identified 19 signature sites in the vaccine proteins Env-gp120, Gag, and Pro that differed between the vaccine and placebo groups. In addition, we identified 37 signature sites in parts of the proteome that were not included in the vaccine. Four pairs of signature sites overlapped in different proteins across reading frames, resulting in a total of 52 unique sites across the proteome. Because our exploratory study was designed to identify all potential sieve effects, we reported all sites with unadjusted p-values below 0.05. Of the signature sites identified in vaccine immunogen regions, only Pol 51 passed the q-value ≤ 0.20 threshold.
Sieve analysis, by comparing breakthrough HIV-1 viruses across treatment groups, can test some of the specific hypotheses generated by correlates of risk analyses, such as the Haynes et al. [3] study that identified anti-V2 antibodies as a correlate of risk. For example, sieve analysis can test whether the breakthrough infections in the vaccine group are less viable targets for the vaccine-induced anti-V2 antibodies than the infections in the placebo group. The V1/V2 focused sieve analysis that identified V2 signature sites 169 and 181 [7] and follow-up studies [36] lent support to the hypothesis that anti-V1/V2 antibodies were involved in a mechanism of partial vaccine protection and that these sites are important for antibody binding. Confirming our previous study [7], the full proteome site-scanning analysis also identified signature sites 169 and 181, although unlike the previous V1V2-focused analysis results, the exploratory results reported here did not pass multiplicity correction, partly due to the much larger number of analyzed sites (8 compared to 248 in Env alone). We hope that additional follow-up experiments will further elucidate the role, if any, of the other newly-identified signatures in the partial protection conferred by the vaccine regimen.
Among all the signature sites, some are worth singling out because they were found by multiple methods and/or there is biological evidence supporting their potential vaccine-associated immunological relevance. In particular, Env 19, 169, 181, 317, 413 and 424 appear important because they are known antibody contact sites or belong to functionally important regions of the HIV envelope (S3 Table). The finding that Env signatures preferentially map to sites known to have a role in antibody neutralization or binding supports the hypothesis that the results are biologically meaningful. For example, the tridimensional structure of Env-gp120 showed that site 169 was in the vicinity of site 317. Env 169 is located at the crown of the V2 loop and was previously identified in the V1/V2-focused sieve analysis [7]. It is contained in a linear binding antibody epitope hotspot for RV144 vaccine-induced antibodies, and is a known contact site for neutralizing and binding antibodies. It is also part of a predicted HLA binding hotspot in the MN vaccine immunogen for both class I and class II alleles. Furthermore, this position is in the seventh position of a 9-mer that had significant treatment group differential binding escape versus the subtype B immunogen sequence (MN), a surprising discovery that motivated further analysis, leading to the finding that the differential vaccine efficacy at Env 169 was significantly associated with the class I HLA allele A*02 [10]. Env 317, identified by all three of the primary site-specific sieve methods, is in the core of the V3 loop and is part of the conserved co-receptor binding site. It is also known to be a contact site for neutralizing antibodies (nAb-sites), is part of an antibody hotspot defined using antigen microarrays [6], and is predicted to be on antibody interfaces using the EPIMAP method [7]. It exhibited a “vMismatch” sieve effect, in that there was greater divergence from the vaccine immunogen AA among the placebo recipient sequences than among the vaccine recipient sequences. It has been shown previously that mutations in V2 can interact with V3, and thereby have an impact on phenotypic changes such as co-receptor usage [37,38]. In addition, mutations in V3 can modulate the neutralization sensitivity of the conserved V2 epitope that is recognized by PG9/PG16-like antibodies. Interestingly, some antibodies isolated in RV144 vaccine recipients mapped to the same V2 region as PG9/PG16-like antibodies, implying that the mutations that we identified in V2 and V3 may have a synergistic impact on the neutralization sensitivity of breakthrough viruses [39].
Signature site Env 413 had the strongest influence on the variation of vaccine efficacy against HIV-1 as a function of genetic distance to CM244 computed using the Env-gp120 signature sites, and exhibited a vMismatch sieve effect. Env 413 is close to the CCR5 and 17b binding sites, and, together with signature site Env 424, it surrounds the binding motif RIKQ (residues 419–422). Most importantly, Env 413 was linked to the breadth of neutralizing antibody responses in a study that compared subjects with strong or weak neutralizing antibody responses [40]: an increase in breadth was associated with asparagine (N) at position 413. Here the consensus residue in CRF01_AE was T and the second most frequent residue found at that site was asparagine (N), which creates a site for potential N-linked glycosylation.
The signature sites identified with the site-scanning methods were characterized by greater amino acid covariation with other sites; there were more interactions with signature sites than at other sites as well as more interactions in gp120 (in the vaccine) than in gp41 (not in the vaccine). When across-protein interactions were considered, vaccine proteins showed greater connectedness: they covaried with more proteins. We conjecture that a vaccine-induced constraint at a highly connected site would have a greater impact on viral fitness than a change at a weakly connected site.
The differential vaccine efficacy (DVE) analysis allowed us not only to identify sites that differed between the infected vaccine and infected placebo groups but also to estimate the site-specific vaccine efficacy against viruses with a matched or mismatched residue to that present in a vaccine reference sequence at this given site. In Env, the DVE analysis identified six sites where estimated vaccine efficacy was increased to at least 43% (Env 19) and up to 85% (Env 317) against viruses with a specific residue at that site. Conversely, the vaccine efficacy was abolished with a different residue at these sites (point estimates ranging from less than zero percent to 17%). These results suggest that vaccine efficacy can essentially disappear with a single mutation. Better evaluating the VE/mutation relationship is critical for our understanding of vaccine immunity as it pertains to HIV-1. Knowing the genetic diversity of HIV-1, the disappearance of vaccine efficacy with a point mutation raises important questions as to the future efficacy of a successful vaccine. By analogy with drug-resistance mutations, we can envisage that the broad usage of a vaccine may lead to the increased frequency of mutations such as those that we found to be associated with null vaccine efficacy, and that such mutations would rapidly be selected in the population, hence reducing the vaccine efficacy. This also emphasizes that it may be necessary to have vaccines with multiple specificities in order to avoid the focusing of immune responses, which may lead to more rapid escape from vaccine-induced immunity, or that it may be important to include only essential protein sequences that cannot be mutated without impacting viral fitness.
The SmoothMarks sieve analysis did not find significant evidence that vaccine efficacy varied against HIV-1 genotypes with genotype defined by the genetic distance of breakthrough viruses to the CRF01_AE vaccine inserts. However, we found evidence of global T-cell based sieve effects relative to the CM244 and MN Env gp120 vaccine inserts using the EscapeCount T-cell sieve method. These results are surprising, since CD8+ T-cell response rates of RV144 vaccine recipients were low overall; depending on the sample time point and the assay that was used, 12–63% of vaccine recipients had a T-cell response to Env peptides, but these responses were predominantly CD4 responses [1,41,42]. One explanation is that the vaccine primed natural infection and an anamnestic response caused earlier escape in the vaccine group. We also note that these results were not found with the EpitopeDistance method. This may be due to differences in the definition of a T-cell sieve effect by the two methods. The global effects found here are also related to a V2-specific T-cell sieve analysis using the EscapeCount method that reported evidence of a T-cell sieve effect in the V2 region of the MN immunogen sequence [10].
By employing methods that incorporate T cell epitope binding predictions, our analysis indicates that vaccine-primed T cells and participants’ HLA alleles may have played a role either in early T cell escape or in modulating vaccine efficacy, even possibly at sites that are part of known antibody binding epitopes, such as the tips of the V2 and V3 loops and the CD4 binding site. Identification of potential sieve effects and vaccine-induced T cell epitopes motivates further study both experimentally and computationally, including, for example, testing for amino acid covariation within the PI9 peptide among infected participants (S6 Fig.). The putative effect within Env 299 PSNNTRTSI, along with those identified by the EscapeCount method in other 9-mers (S16 Table), generates hypotheses that can be further investigated computationally and tested experimentally with T cell assays. The suggestion of a trend toward greater phylogenetic diversity and divergence in Gag sequences for vaccine than placebo recipients could reflect the genetic effect of some T-cell mediated responses, although the signal is weak and there was no evidence of a sieve effect at predicted T-cell epitopes in Gag. In addition, our finding of 30 9-mers in Tat with a T cell sieve effect (passing the 20% q-value multiplicity adjustment) could possibly be explained by the fact that Tat is a viral regulatory factor for HIV gene expression and CD8 T-cell responses have been shown to select for viral escape variants in Tat during acute HIV and SIV infection [43].
It remains unclear whether the observed vaccine sieve effects are due to an acquisition barrier, reflecting the selection of viruses that managed to break through the protective effects of vaccination by the RV144 vaccine regimen, versus reflecting early post-acquisition immune pressures that affected within-host viral evolution after infection, or some combination of the two. We note that significant results found using methods that focus on T cell epitopes are not necessarily driven by T-cell pressure, and that signatures in Env may be driven by either T cell pressure, antibody pressure, or by a combination of the two.
Given that one could expect that sieve effects would be restricted to the proteins included in the RV144 vaccine, how can the 37 non-vaccine signature sites be interpreted? In addition, how can we explain that there are an approximately equal number of “vMismatch” and “vMatch” effects? Additionally, the physico-chemical property sieve effect sites tended to occur in non-vaccine proteins, as did all thirty-one sieve effect 9-mers that were significant after multiplicity correction. While some of these signature sites and 9-mers are false positives, others may be true effects. Certain non-immunogen sites/9-mers may be implicated because they are in linkage with other mutations in vaccine sequences; for instance such sites could act as compensatory mutations for vaccine-associated mutations that would be destabilizing. In addition, certain residues that are matched to the vaccine may be required for HIV-1 to be infectious/transmittable. Alternatively, some non-vaccine-immunogen signature sites/9-mers and vMismatch effects could reflect true effects of post-acquisition immune pressure that affect vaccine recipients more strongly or more rapidly than placebo recipients.
This comprehensive whole-genome sieve analysis generates additional testable hypotheses about the nature and mechanism of the vaccine’s partial efficacy, by identifying individual sites, peptide regions and proteins at which the genomic sequences significantly differed between vaccine and placebo recipients. By using a variety of methods, each tailored to detect different types of signals, we both increased the chance of finding differences and provided means for potentially explaining the differences. With additional support from independent analyses, such as viral inhibition experiments based on individual amino acid substitutions, a subset of the site-specific sieve effects identified here may prove to reflect vaccine immune pressure and thus be significant for future vaccine design and analysis. Directions for future research include experimental determination of vaccine-induced antibody binding in identified Env regions, evaluating functional consequences of the observed mutations, and further elucidating the extent to which differences in non-vaccine-immunogen regions of the breakthrough viruses could be directly attributed to vaccination, or indirectly attributed to constraints on the virus or to chance sampling variability.
The list of specific testable hypotheses includes evaluation of all of the signature sites for evidence of vaccine-induced immune pressure targeting each site. While in the absence of an additional efficacy trial it is not possible to directly evaluate the statement that “vaccine efficacy can essentially disappear with a single mutation”, it is possible to test the hypothesis that the vaccine-induced antibodies bind viruses differentially depending on individual mutations. This has been done for V2-targeting antibodies by Liao and colleagues [36] as well as for V3-targeting antibodies as reported by Susan Zolla-Pazner [11]) and could in principle be done for any of the (Env) signature sites. Neutralization assays could also be applied to assess differential neutralization, though the antibodies induced by RV144 appear to be non-neutralizing. More generally, effector function assays (e.g., ADCC) could be applied to assess differential functional responses.
After over 30 years of effort to develop an effective public health vaccine to prevent infection by HIV-1, the only vaccine to show statistically significant efficacy was the regimen used in the RV144 Thai trial. The borderline significant p-value of this result leaves open the possibility that the regimen had no overall efficacy. It would be possible for a vaccine with no overall vaccine efficacy to nevertheless have differential efficacy against different viruses. One example is a balancing effect, in which the vaccine has both protective and harmful effects, depending on the virus. Another possibility is that subjects who experience multiple exposures to HIV are protected against some viruses but ultimately become infected despite this partial protection (but later than they otherwise would have been); if the time delay is short (if the multiple exposures are close together in time), this could lead to negligible overall efficacy despite true acquisition sieve effects. In our view the strength of evidence for overall VE is increased by the findings of this study, for example the SmoothMarks analysis provided smaller p-values for overall VE by accounting for viral distances. However, experimental confirmation is still crucial, both because we are not able to prove that the observed sieve effects are acquisition effects and because of the expectation that many of these are false discoveries. Even if an effect is a true discovery (in that the treatment group differences are not due to chance variation), it may be an effect of vaccine-induced changes to the evolutionary course of the virus after infection (post-acquisition effects) rather than effects to prevent infection (acquisition effects). There were significant effects found in the sieve analysis of the Step 502 trial that are presumed to be post-acquisition effects because the vaccine immunogen had no Envelope component (and no evidence of antibody induction) and because the strongest effect was found at a known T-cell epitope and was strongest in subjects with the necessary HLA types to target that epitope [8,9]. In the absence of confirmatory studies, the signature site analysis would not increase the strength of evidence. However, the strength of evidence for overall efficacy has already been increased, in our view, by the experimental confirmation of RV144-induced V2-targeting mAbs that differentially bind depending on the amino acid at site Env 169, in conjunction with the significant correlation of vaccine efficacy with induction of V2-targeting Abs.
It has become clear that future vaccine studies should be designed for a more rapid iterative process, to maximize the information gleaned from each trial and ultimately to minimize the time to an effective global intervention strategy [44,45]. The correlates and sieve analyses of the RV144 trial demonstrate the importance of designing future trials with sufficient power to conduct such analyses. In particular, both types of analyses are improved by more precise resolution of the timing of HIV infection (e.g., accomplished through more frequent visits for diagnostic testing of HIV-1 infection that capture a sizable fraction of HIV infection events in the pre-seroconversion acute phase), which would allow use of statistical methods that can help tease apart acquisition sieve effects from post-acquisition differential within-host viral evolution [19].
The protocol was approved by the Institutional Review Boards of the Ministry of Public Health, the Royal Thai Army, Mahidol University, and the US Army Medical Research and Materiel Command. Written informed consent was obtained from all participants.
The study conduct and results have been published previously [1].
The vaccine regimen was a combination of HIV-1 subtype B and HIV-1 CRF01_AE: the prime corresponded to HIV-1 Gag and Pro of subtype B LAI and the CRF01_AE HIV-1 gp120 (strain 92TH023) linked to the subtype B transmembrane domain of gp41 (strain LAI); the boost corresponded to the CRF01_AE HIV-1 gp120 strain CM244 with the subtype B HIV-1 gp120 strain MN. (CRF01_AE is subtype E in the HIV-1 Env.) We aligned these three sequences to the breakthrough sequences for analysis.
Of the 16,395 participants who entered the trial HIV-1 negative [the modified intention-to-treat (MITT) cohort], 125 acquired HIV-1 infection during the 3.5-year follow-up period. Of the 125 MITT infected subjects, we analyzed the subset of subjects who were infected by HIV-1 CRF01_AE viruses, for whom we have sequence data, and who were not infected by another trial participant (n = 109 subjects). Subjects infected with subtype B viruses (n = 11) were excluded because of the much larger HIV-1 genetic distances to the vaccine immunogen sequences compared to CRF01_AE, such that their inclusion would likely reduce statistical power of the sieve analysis by contributing genetic variation unrelated to a sieve effect. Four of the 125 infected subjects had no sequence data, three because the Sanger sequencing technology failed to deliver a result due to low HIV-1 viral load, and one because of drop-out. Finally, we excluded subject AA100 because this subject was the second to acquire HIV in the AA118/AA100 transmission pair; excluding AA100 avoids complications arising from the non-independence of these infections, and helps maintain plausibility of the ‘sieve conditions’, which are sufficient assumptions to justify interpretability of observed genotype-specific vaccine efficacies and infected-case sequence differences as prospective, per-contact estimates of genotype-specific vaccine efficacy [46]. The other linked transmission pair was subtype B, so both of those subjects were excluded on the grounds of their infecting subtype. Note that while most of the sieve analyses conditioned on infection (and therefore truly excluded from analysis all subjects other than the 109), the estimates of genotype-specific vaccine efficacy and differential vaccine efficacy, as well as the SmoothMarks multi-site acquisition sieve analyses, were time-to-infection analyses that included the entire MITT cohort (and right-censored, rather than truly excluded, the infected subjects outside of the 109). The vaccine efficacy to prevent acquisition of CRF01_AE HIV-1 (based on the MITT cohort and these 109 infections) was estimated to be 35.2% (95% CI 4.8% to 55.8%, score test p = 0.026).
The RV144 HIV-1 sequencing methodology has been published previously[8], and further information is provided in S1 Table and S2 Table. Sequences are available under GenBank accession numbers JX446645–JX448316.
For each subject, we defined the mindist sequence to be the closest actual sequence to the consensus of that subject’s full-genome nucleotide sequences, as measured by the Tamura-Nei ‘93 (TN93) distance correction model [47]. A full description of our mindist selection process is presented in S1 Text. In short, subjects with a full-length nucleotide sequence that measured closest to their consensus with TN93 had that sequence used and translated for all mindist protein sequences. For subjects with only right-half or left-half sequences that measured closest to their consensus, the closest right- and left-half genomes were selected and thence translated into the appropriate mindist protein sequences. Ties were broken by (a) excluding sequences with the most ambiguous, incomplete or stop codons, (b) for right-half genomes, selecting the sequence with the shortest env distance, and (c) for left-half genomes, selecting the sequence with the shortest gag distance. Five ties remained after this procedure, which were broken randomly.
Only the SmoothMarks and vaccine efficacy (VE) and differential VE (DVE) analyses utilize the entire “Modified Intent-to-Treat” (MITT) cohort of the RV144 trial, including subjects who did not become infected and subjects lost to followup. These methods are particularly well-suited to detect acquisition sieve effects, because under fairly general conditions these have been shown to be robust to post-hoc selection biases engendered by conditioning on infection. The other methods only include in the analysis infected subjects (who by definition are the only subjects with HIV-1 sequences available for analysis; see the Trial Data subsection of Methods for details), and (while generally applicable) are best-suited to evaluate post-acquisition sieve effects. Because of the six-monthly sampling scheme employed in the RV144 trial, the evaluated sequences are likely to have evolved between acquisition and sampling, and, of the methods applied here, only the SmoothMarks method attempts to recapitulate the genetic distance of the founder variant using missing-data methods. To our knowledge there is no existing method that can differentiate between acquisition and post-acquisition effects without incorporating longitudinal sequence data, which are not available for this trial.
The DVE method is designed to detect acquisition sieve effects of differential VE by Match vs. Mismatch of breakthrough sequences to the immunogen sequences, and the SmoothMarks method is designed to detect acquisition sieve effects of differential VE by continuous genetic distance of breakthrough sequences to the immunogen sequences. The other methods are designed to detect post-acquisition effects such as weighted mutation rates at single sites (GWJ uses a T-type test comparing AA substitution costs versus the vaccine immunogen, MBS uses a Bayesian model of post-acquisition sieve effects), incorporate multiple sequences per subject (SMMB and EGWJ), employ a phylogenetic model of sequence relatedness (divergence, diversity, PRIME, and FEL), evaluate codons for selection pressure (dN/dS and PRIME), and/or evaluate immunological hypotheses such as physico-chemical selection (PCP and PRIME), T cell escape (EscapeCount, EpitopeDistance, and PercentEpitopeMismatch), and antibody binding (signature site set enrichment, and SmoothMarks when applied to Ab site sets). The application of these varied methods provides a comprehensive exploratory evaluation of the effects of vaccination on breakthrough HIV-1 sequences.
Maximum likelihood phylogenetic trees were constructed (one tree per protein and per vaccine immunogen sequence) using PhyML (version 3.0) [48], using the HIV-between (HIVb) PAM substitution matrix[16], invariant sites, and four gamma-distributed rate categories. For each tree, the differential amino acid phylogenetic diversity (PD) [31] was defined as the difference in the total branch length of two subtrees (defined by holding the tree fixed and excluding a subset of leaves corresponding to one treatment group): the subtree excluding placebo recipients (retaining only sequences from vaccine recipients and the vaccine immunogen sequence); and the complementary subtree excluding vaccine recipients (but retaining the vaccine immunogen sequence). We estimated a null distribution by randomly permuting vaccine/placebo labels 10,000 times, and computed a (two-sided) p-value by comparing the observed difference in PDs to this null distribution.
The phylogenetic divergence analysis computed shortest-path distances between each subject’s sequence(s) and the vaccine immunogen sequence. For mindist analyses we used the trees computed for the PD analysis. For multiple-sequence-per-subject analyses, we constructed AA trees as above using all available sequences, and we constructed nucleotide trees using the GTR + I + G nucleotide substitution model using PhyML (version 3.0) [48] implemented in DIVEIN [49] (http://indra.mullins.microbiol.washington.edu/DIVEIN/diver.html). Tree-based distances were extracted from these trees using the NewickTermBranch algorithm (http://indra.mullins.microbiol.washington.edu/perlscript/docs/NewickTermBranch.html) and the ape package in the R computing language [50], and per-subject median distances were computed to each reference sequence. These distances were compared between the vaccine and placebo groups using a Wilcoxon rank sum test (one test per gene/reference combination).
We introduce a new application of the Earth-Mover’s Distance statistic to sieve analysis. The QEMD statistic equals the maximum over W of ∑(S * W) where S is the n by m matrix containing the pairwise alignment scores between vaccine and placebo mindist sequences, * denotes entrywise multiplication and W is an n by m weight matrix subject to the following constraints: W > 0, every row sums to 1/n and every column sums to 1/m. Note that in this case the QEMD statistic measures similarity (not distance). The QEMD hypothesis test reports two-sided “mid p-values” [51] based on random permutation of treatment assignments. We applied this approach with n and m, the total number of vaccine and placebo recipient sequences, respectively.
We used the mindist sequence as an approximation of the founder virus, and we computed distances between the immunogen sequences and the mindist sequence measured from blood samples drawn at or before the HIV diagnosis date. The SmoothMarks method [19,46] was used for estimation and testing of VE(v) over the range of distances v from 0 to 1, where the vaccine efficacy against HIV-1 with distance v, VE(v), is one minus the distance v-specific hazard ratio (vaccine/placebo) of HIV-1 infection multiplied by 100%. This method employs a missing-data framework to analyze VE(v) as a function of the “true distance” v between the transmitted founder sequence and the vaccine immunogen sequence. This can in principle improve the analysis over the other analysis methods that analyze the “observed distances” of available sequences that are measured weeks or months after infection; by not accounting for post-acquisition evolution these methods may obscure acquisition sieve effects. Since we do not have longitudinal sequence data, we are limited in our ability to estimate the transmitted founder sequence, so for the present analysis we defined “true” genetic distances as the HIVb-computed distance between the immunogen sequences and the mindist sequence measured from blood samples drawn at or before the HIV diagnosis date, where the 10% of infected subjects (11 of 109) with later sampled sequences were treated as missing data. See S2 Text for additional details about the method and its implementation.
We assessed genotype-specific VE using the Cox proportional hazards model and score test as described by [52], and we assessed differential VE (DVE) by genotype using the same model, via the procedure described by Lunn and McNeil [14]. These were the primary analysis methods used previously [7]. Negative VE values are shown in symmetrized form (as the negative of the VE value calculated with vaccine and placebo groups interchanged).
Two additional primary site-scanning methods were used that assess at each site whether the amino acid distances to a reference immunogen at that site differ for vaccine compared to placebo recipient sequences: a nonparametric weighted distance comparison test (GWJ) [17], and a model-based method (MBS) [18]. Both of these methods were based on the mindist amino acid sequences. Code for these methods was published previously [7].
We introduce the PCP analysis method, which compares counts of each of the ten Taylor properties [21], and five z-scale components [22–24] across treatment groups using parametric two-sample pooled-variance two-sided t-tests. The analysis can apply to individual sites or to arbitrary site sets (we evaluated 3-mers and 9-mers), in the latter case by summing counts over sites. The resulting p-values are then Bonferroni-corrected across the properties for each of the two property scales at each site (for all k-mers overlapping that site, separately for each value of k).
Peptide microarrays designed to cover the entire gp160 consensus sequences for HIV-1 Group M, subtypes A, B, C, D, CRF01_AE and CRF02_AG for a total of 1423 peptides (15-mers overlapping by 12 amino acids) were used to detect reactive regions for RV144 vaccine recipients. Using the analysis method of Imholte et al. [53], four dominant responses were detected in the C1, V2, V3 and C5 regions of gp120 [6]. These sites are listed in S3 Dataset.
This is a set of known and published monoclonal antibody contact sites provided by Ivelin Georgiev, Peter Kwong, Robin Stanfield, and Ian Wilson [54,55]. They are listed in S3 Dataset.
These are the sites identified as relevant to the neutralization activity of known neutralizing antibodies in Wei et al. (2003) [54], Moore et al. (2009) [55], and Tomaras et al. (2011) [56]. They are listed in S3 Dataset.
These are the union of sites in contactsites and nAb-sites.
Potential antibody contact “patches” were calculated by the method described previously [7], but considering all of the Env protein rather than only the V1/V2 region. Sites were sorted by frequency of inclusion in these patches (by the mean of their frequency of inclusion in patches versus the 92TH023 sequence and the maximum of their frequencies of inclusion in patches versus the CM244 and MN sequences), as shown in S9 Fig.. The same threshold used previously [7] was used to select top-scoring sites. The EPIMAP site set contains the 71 sites that passed this threshold, 38 of which overlap vaccine sequence sites. They are listed in S3 Dataset.
We defined two sets of sites where selective pressure by T cells was putatively highest. These analyses considered only the vaccine immunogen sequences (and the HLA types of the subjects), and were conducted blinded to subject treatment assignment. They are listed in S4 Dataset.
|
10.1371/journal.pbio.1002111 | Cell Cycle–Dependent Differentiation Dynamics Balances Growth and Endocrine Differentiation in the Pancreas | Organogenesis relies on the spatiotemporal balancing of differentiation and proliferation driven by an expanding pool of progenitor cells. In the mouse pancreas, lineage tracing at the population level has shown that the expanding pancreas progenitors can initially give rise to all endocrine, ductal, and acinar cells but become bipotent by embryonic day 13.5, giving rise to endocrine cells and ductal cells. However, the dynamics of individual progenitors balancing self-renewal and lineage-specific differentiation has never been described. Using three-dimensional live imaging and in vivo clonal analysis, we reveal the contribution of individual cells to the global behaviour and demonstrate three modes of progenitor divisions: symmetric renewing, symmetric endocrinogenic, and asymmetric generating a progenitor and an endocrine progenitor. Quantitative analysis shows that the endocrine differentiation process is consistent with a simple model of cell cycle–dependent stochastic priming of progenitors to endocrine fate. The findings provide insights to define control parameters to optimize the generation of β-cells in vitro.
| In order to form organs of the right size and cell composition, progenitor cells must balance their proliferation and their differentiation into functional cell types. Here we study how individual progenitor cells in the developing pancreas execute their choices to either expand their pool or differentiate into hormone-producing endocrine cells. Using live microscopy to track the genetically marked progeny of single cells, we reveal that after they divide, individual cells generate either two progenitors, two cells on the endocrine path, or one progenitor and one cell on the endocrine path. Quantitative analysis shows that endocrine differentiation is largely stochastic and that the probability of progenitor cell differentiation by the end of mid-gestation is about 20%. We propose a model in which the production of a progenitor and a differentiated cell in the pancreas results from the stochastic induction of differentiation in one daughter after cell division, rather than the unequal partitioning of molecules between two daughters at the time of division, as observed in the nervous system. Furthermore, when two daughters become endocrine cells, this results from the induction of differentiation followed by cell division—rather than two independent induction events. This model may be applicable to other organs and provides insights to optimize the generation of β-cells in vitro for diabetes therapy.
| The pancreas is an organ performing vital exocrine and endocrine roles in nutrient metabolism and glucose homeostasis. In the mouse, multipotent pancreatic progenitor cells (MPCs) emerge from the endoderm around embryonic day 9.0 (E9.0) [1]. This population, characterized by the expression of transcription factors PDX1 (GenBank NP_032840), SOX9 (GenBank NP_035578), and HNF1B (GenBank AAH25189), eventually gives rise to all three major cell lineages of the pancreas: endocrine, acinar, and ductal [2–4]. Following early progenitor expansion, three-dimensional (3-D) organization of the pancreatic epithelium leads to the generation of an apico-basally polarized [5–7], branched tubular network. By E13.5, it exhibits its final functional compartmentalization: the distal tip domains give rise to the acinar cells of the exocrine lineage [8], whereas the SOX9+/HNF1B+ proximal trunk domain is bipotent at the population level, giving rise to the ductal and endocrine cells [3]. The endocrine lineage arises from transient NEUROG3+ (GenBank AAI04328.1) endocrine progenitors, as demonstrated by lineage tracing studies [2] and the absence of all pancreatic endocrine cells in Neurog3−/− mice [9]. NEUROG3+ endocrine progenitors originate from pancreatic progenitors expressing PDX1/SOX9/HNF1B during the early phases of MPC expansion and during the secondary transition spanning E12.5–15.5, with specific endocrine subtypes being specified during discrete time windows [10]. Whereas the majority of NEUROG3+ endocrine progenitors are post-mitotic [11] and unipotent, giving rise to only one endocrine subtype [12], we do not know whether individual PDX1/SOX9/HNF1B pancreatic progenitors give rise to both ductal and endocrine cells or are heterogeneous, encompassing cells with pre-specified lineage-restricted potentials. In this study, we ask how individual pancreas progenitors contribute to the population dynamics enabling organ expansion and endocrine differentiation.
Over the last few years cell-labelling and tracing methods have brought forth quantitative descriptions of cell differentiation. In homeostatic systems, for instance, the maintenance of a hierarchy of stem and differentiating cells can be accounted for by populations of equipotent progenitors exhibiting probabilistic fate choices [13–15]. An attempt to extrapolate these notions to developing systems has encountered some difficulties because, in these instances, the growth of the tissue needs to be taken into consideration. Notwithstanding these complications, lineage analysis of progenitor cells in the vertebrate retina indicates that, similarly to the abovementioned homeostatic systems, the distribution of clone sizes is compatible with a model in which progenitors stochastically divide in three modes: (1) symmetric self-renewing, (2) asymmetric, and (3) symmetric differentiating divisions [16–20]. Contrary to homeostatic systems, however, the probabilities of each division mode are not assumed to be fixed but to vary over time, following phases of proliferation and differentiation. These models have proven successful in explaining the distributions of clone sizes but do not explain the observed frequencies of each division type. Alternative models have been put forward that invoke deterministic asymmetric inheritance of differentiative cues at the time of division [21–24]. In general, how decisions of single cells contribute to global organ growth and differentiation in developing organs remains an open question.
Here we test some of these notions in the context of the emergence of endocrine progenitor cells from uncommitted pancreatic progenitors in the embryonic pancreas. This developmental model has a simple lineage configuration, with a reduced number of fates over well-characterized time windows, and thus provides an optimal testing framework. We use 3-D live imaging of pancreatic explants ex vivo and in vivo, together with lineage tracing at a clonal density, to address the dynamics of the progenitors of the endocrine lineage. In addition to monitoring their lineage, we determined measurements for cell cycle length, synchrony, and differentiation dynamics of these progenitors. This revealed three types of pancreatic progenitor behaviours: (1) symmetric progenitor self-renewal, (2) symmetric endocrinogenic divisions leading to two NEUROG3+ endocrine progenitors, and (3) asymmetric divisions generating a pancreatic progenitor and an endocrine progenitor. By live tracing individual cell fate specification events, we uncover the relationship between Neurog3 expression timing and mitosis. We identify major differences in the onset of Neurog3 transcription between cells stemming from symmetric and asymmetric divisions, and further show that this onset is highly synchronized between symmetrically generated sibling cells. Our analysis of such findings leads to a novel interpretation of the choice between symmetric and asymmetric cell divisions. We posit that asymmetric cell divisions are the result of the stochastic induction of endocrine fate in one of the progenitor daughters, rather than a decision made during cell division. Alternatively, if this progenitor divides a last time after induction, which is expected if the induction happens late in G1, the division will be seen as symmetric differentiative. These results argue against conventional views of asymmetric inheritance of differentiative cues at the time of division [21–24] and are instead consistent with a model of cell cycle–dependent stochastic specification of organ-specific progenitors.
To study how individual pancreatic progenitors contribute to pancreas expansion and to monitor their differentiation into endocrine progenitors, we conducted live imaging of explants of dorsal pancreatic buds from E12.5 Pdx1tTA/+;tetO-H2B-GFP embryos (Fig. 1A). The buds were dissected and laid on a fibronectin-coated coverslip bottom plate, where they grew (Fig. 1B) [25,26]. After 24 h of settling time, we initiated high-magnification time-lapse live imaging in 3-D with 6-min intervals for up to 24 h. Nuclear H2B-GFP fusion protein enabled us to observe cell divisions and to track individual cell nuclei. At the end of the experiment, the explants were fixed and immunostained for markers of pancreatic progenitors (SOX9) and endocrine progenitors (NEUROG3) (S1F–I Fig.), while preserving the native green fluorescent protein (GFP) signal (S1G Fig.), which enabled us to match to the cells from the last frame of the time-lapse movies. The SOX9+ cells constituted the majority of GFP+ epithelial cells (S1I Fig.), and NEUROG3+ cells were mainly observed in the middle trunk region of explant (S1H Fig.) [8]. In spite of the constant exposure to laser, explants grew, showed active cell movements, apico-basal polarization, branching, acini morphogenesis, and differentiation similarly to explants that were not subjected to imaging (S1A–E,J,N,O Fig. and S1 and S2 Movies). After 18–24 h of time-lapse imaging (42–48 h post-dissection), NEUROG3+ cells were detected by immunostaining, showing that the differentiation process occurred ex vivo, albeit less efficiently than in vivo (S1 Table).
To determine how single progenitor cells contribute to balancing global pancreas expansion with endocrine progenitor generation (Fig. 1C), we systematically back-tracked NEUROG3+ endocrine progenitors in 3-D, as well as a random subset of SOX9+ pancreatic progenitors that were identified from immunostaining images and mapped onto the last frame of time-lapse movies. Once a cell division was observed through back-tracking, the sister cell was then forward-tracked, and its fate was determined by referring to the immunostaining. The tracking revealed that pancreatic progenitors divided in one of three different modes. The first type of division was symmetric, giving rise to two SOX9+ progenitor cells (P/P division; S3 Movie and Fig. 1D,F–I). The second type was asymmetric, giving rise to a SOX9+/NEUROG3− pancreatic progenitor and a NEUROG3+ endocrine progenitor (P/N division; S3 Movie and Fig. 1E,F–I). The last type was symmetric endocrinogenic, producing two NEUROG3+ cells (N/N division; S4 Movie and Fig. 1J–N). In order to quantitatively account for each division mode, we analysed 1,628 divisions comprising all observed division events of Pdx1tTA/+;tetO-H2B-GFP cells from randomly selected positions from four time-lapse movies. Thus, non-NEUROG3-producing divisions include both bi-potent progenitors and acinar cells, since Pdx1+ cells are multipotent at E13.5. This quantification revealed 6.6% ± 1.6% of divisions producing endocrine progenitors, and 93.4% ± 1.6% generating either progenitors or exocrine cells (Fig. 1O and S2 Table). Of all the divisions producing NEUROG3 cells that could be tracked, 56.3% ± 13.8% produced a SOX9+ cell and a NEUROG3+ cell (P/N division), and 43.7% ± 13.8% produced two NEUROG3+ cells (N/N division; Fig. 1O). We could determine the origin of approximately half of NEUROG3+ cells through P/N or N/N divisions in the past 24 h, while some NEUROG3+ cells either did not exhibit prior division or were either lost or dead during tracking (Fig. 1P). Cell death might be a consequence of the explant culture since apoptosis is very rare in the pancreas epithelium in vivo [27]. Taken together, these results show that at E13.5–14.5 most progenitors undergo symmetric renewing divisions, accounting for pancreas size increase, while the remaining progenitors are approximately evenly split into those undergoing symmetric endocrinogenic division and asymmetric division.
While ex vivo imaging enables constant monitoring of cell behaviours, it is performed in a somewhat artificial context. In order to determine whether pancreas progenitors undergo the same pattern of symmetric and asymmetric divisions in an in vivo context, we devised a clonal lineage tracing strategy (Fig. 2A) using Hnf1bCreER mice. Previously, this line was used to demonstrate that the E13.5 HNF1B+ progenitor cells give rise to ductal and endocrine cells [3]. This could be accounted for either by individual cells giving rise to endocrine and ductal cells or by heterogeneity among HNF1B+ cells, some giving rise to endocrine cells and others to ductal cells. To investigate this question, we subjected pregnant mice carrying E13.5 Hnf1bCreER;mT/mG embryos to a single low-dose intraperitoneal injection of 4-hydroxytamoxifen (Fig. 2B) to label pancreatic progenitors at a clonal density. We optimized conditions for clonal tracking leading to two-cell clones at E14.5 (Fig. 2B–G, S3 Table, and S5 Movie). Since we know from the time-lapse experiments that the majority of daughters from the same cell do not move more than 30 μm apart, we called labelled cells within 30 μm of each other a clone (S8 Fig.). Reiterations with a 60 μm radius led to similar outcomes. Whole-mount immunostaining of 22 pancreata and detection of 244 two-cell clones revealed that the majority of progenitors in which the fate could be determined divided symmetrically (P/P; Fig. 2E) into two SOX9+ progenitors (59.8%; Fig. 2H). This proportion is lower than the 93.4% found in the explants, in part because the cells traced by HNF1B are a subgroup of PDX1+ and SOX9+ cells traced in the explants and some of the latter will give rise to acinar cells [2,4]. In vitro lineage tracing with Hnf1bCreER;mT/mG explants showed that 6.3% of clones became endocrine (S2 Fig. and S4 Table). This shows that the in vivo differentiation process is more efficient than in vitro differentiation. After the 24-h tracing period, we could not yet observe any INSULIN+ clones in vivo, suggesting NEUROG3− or SOX9− clones might be in transition to endocrine differentiation. Of the NEUROG3-producing two-cell clones in which the fate of both daughters could be determined, 61.8% were asymmetric, generating one NEUROG3+ daughter and a SOX9+ progenitor (P/N; Fig. 2F,H), and the remaining were symmetric with two NEUROG3+ daughters (N/N; Fig. 2G,H). As a consequence, more NEUROG3+ cells originated from symmetric divisions (Fig. 2I). These results thus provide in vivo evidence of asymmetric and symmetric endocrinogenic progenitor divisions, as well as of symmetric renewal of progenitors, confirming the modes of divisions detected by the explant tracking data.
From the above data with regard to fate-determinable two-cell clones, we estimated expected ratios of P/P, P/N, and N/N divisions to be 69.9%, 18.6%, and 11.5%, respectively, after excluding indeterminable clones. Progenitors undergoing symmetric differentiating divisions will contribute all of their progeny to the differentiated pool, whereas asymmetrically dividing progenitors will contribute only one half of their progeny to this pool. We can therefore directly estimate the probability of differentiation of new-born cells to be 20.8% ([0.5 × 18.6] + 11.5)%, which is consistent with a net expansion of developing pancreas (S1.5 Text).
If sibling cells adopted their fate independently of each other, the expected fractions for each division type would be 62.7% for symmetric proliferative, 4.3% for symmetric differentiative, and 33% for asymmetric. Notably, these last two fractions deviate from the experimentally reported ones (Fig. 2H), contradicting the hypothesis of independent sibling fate allocation. This is further supported by statistical tests indicating a significant divergence from the independence ratios (S1.5 Text). Similar calculations can be made on the in vitro data leading to the same conclusion that a single conversion event leads to symmetric endocrine cell production (Fig. 1O and S1.5 Text).
To investigate the dynamics of differentiation, we generated transgenic Neurog3-RFP reporter lines that can be used for live imaging together with H2B-GFP (Figs. 3A,B and S3A,B). Immunostaining for NEUROG3 and comparison with red fluorescent protein (RFP) from E14.5 Neurog3-RFP pancreas revealed that 40.1% ± 4.5% of NEUROG3+ cells were co-expressing RFP, while the remaining NEUROG3+ cells were RFP− (S3C,D Fig.). Some discrepancies may be expected because of the transient nature of Neurog3 expression and the different onset and decay kinetics of the RFP protein compared to the NEUROG3 protein (S1.3 Text). Moreover, 77.1% ± 2.8% of RFP+ cells were NEUROG3− due to probable delay and perdurance of RFP as compared to that of NEUROG3 [28], as also seen for other reporters [29–31]. This maintenance was attested by the detection of hormones in 18.5% ± 2.3% of RFP+ cells. To further address the reliability of the reporter and assess its incidence in our analysis, we compared this line to the enhanced yellow fluorescent protein (EYFP) knock-add-on allele, which has been reported to show a greater overlap with NEUROG3 protein [29] and which is, in principle, less susceptible to exogenous chromatin environments, being in the endogenous locus. Our imaging of explants expressing one allele of EYFP and one of RFP (S4A,B Fig.) showed that all RFP+ cells were also EYFP+ (S4C Fig. and S5 Table), indicating no false positives due to genomic environment. Single cell tracing showed that RFP was turned on 4.7 ± 1.1 h after EYFP was detected (S4D Fig.); 11.6% ± 3.7% of EYFP+ cells never became RFP+, indicating the system was largely faithful.
From time-lapse imaging extended to 48 h, we could observe a dynamic change of RFP signal in single cells from the onset of fluorescence: gradual increase and a subsequent decrease, which reflects the transient expression of NEUROG3 [32]. Our analysis predicts a short half-life of 5–6 h for RFP in a cell, most probably due to continuous laser exposure. We estimate a “perdurance” of detectable fluorescence of more than 20 h (see S5 Fig. and S1.3 Text) and a minimum delay between cell priming and RFP onset of approximately 5 h. Monitoring all events of RFP onset (n = 323; Fig. 4A) initially suggested waves of cellular differentiation at the tissue level. However, statistical analysis of the timing of onset events showed that these are also compatible with a stochastic process of cell differentiation with homogeneous differentiation rate (i.e., a Poisson process) throughout the imaging period (S6 Fig. and S1.4 Text). While we cannot rule out a periodic process underpinning Neurog3 expression, confirmation of this would require more data points.
Similar to earlier tracking, RFP+ cells were back-tracked from the last time point in time-lapse movies, their prior division was monitored, and sister cells were forward-tracked. Quantifications (S6 Table) revealed that among the RFP-producing divisions where the fate of the two sisters was determinable, as follows: 60.2% ± 11.9% were asymmetric divisions producing a progenitor and a RFP+ daughter (P/R; S6 Movie, and Fig. 3C–G,N), and 39.8% ± 11.9% were symmetric divisions producing two RFP+ daughters (R/R; S7 Movie, and Fig. 3H–N). In these long time-lapse movies, many RFP+ cells moved out of frame or were lost due to weak GFP signal before acquiring RFP expression (Fig. 3O and S6 Table). Excluding those indeterminable RFP cells, 18.8% ± 6.6% were generated through P/R division, 25.0% ± 10.0% through R/R division, and 3.2% ± 2.8% through RFP division, while no division was seen during the movie duration for 53.0% ± 10.3% of RFP cells (Fig. 3O). These RFP tracking results thus confirmed the calculated proportions of asymmetric versus symmetric divisions established from the previous live imaging and in vivo clonal analysis (Fig. 3P).
The dynamic reporter revealed highly synchronized differentiation after divisions producing two NEUROG3 cells, the RFP signal being detected in both daughters within 0.8 ± 0.4 h of each other (Fig. 4B, correlation coefficient between lag times 0.98 ± 0.002). This outstanding synchrony confirms that it is unlikely that the two daughters are induced by independent events and suggests that mother cells have been primed to differentiate into NEUROG3 cells prior to their division. This observation further suggests a defined time between priming and NEUROG3 expression (or its RFP proxy). In addition, asymmetrically generated NEUROG3 cells exhibited a significantly longer lag time between the division and RFP onset, as shown in Fig. 4C, further supporting an interplay between cell cycle, the differentiation priming event, and the division mode.
These results on the contrasting dynamics of differentiation between cells stemming from symmetric versus asymmetric division events are obtained with the RFP reporter, for which we have established a false negative rate of 11.6% (S4C Fig.). This implies that, far from amplifying the differences between the dynamics of differentiation between the two groups, we might be underestimating them. Specifically, our reporter may miss a subset of Neurog3-expressing cells, thus leading to mis-allocation of around 11.6% of symmetric events to asymmetric and therefore homogenizing the two categories and reducing the differences between them (See below).
To try to understand the mechanism underlying the emergence of endocrine progenitors, we devised a simple mathematical model [33] of cell proliferation dynamics based on the lineage and differentiation dynamics data. The model is based on the premise that proliferating progenitor cells primed for Neurog3-dependent differentiation might either exit the cell cycle and become terminally differentiated or commit to complete the cell cycle and produce two terminally differentiated cells via mitosis (Fig. 5A–C). Thus in this model there are three, rather than two, cell types: (i) NEUROG3-primed (N) cells, which are post-mitotic; (ii) cells primed for differentiation but committed to cell cycle completion (L cells); and (iii) progenitor (P) cells, which will not differentiate (Fig. 5D). We assigned a probability q (the differentiation probability) for the differentiation of progenitors and a probability, θ, for primed cells to become N (1- θ to become L). Thus, the model describes the differentiation process in terms of two probabilities, which can be directly inferred from the lineage data (S1.6 Text), i.e. it has no free parameters. From the in vivo clonal analysis data, we estimate θ = 56.5% and, as we have already seen, q = 20.8%. This means that approximately one-half of the progenitors primed for differentiation become post-mitotic (P→N; 56.5%), while the other half (P→L[→N/N]; 43.5%) will undergo one last division before differentiating. Because this latter group of cells (L) is transient and contributes terminally differentiated cells (N), its expected abundance in the tissue is residuary. The model predicts that, at any given time, only 9.8% of cells in the developing tissue are primed progenitors committed to division-cycle completion (L), yet this small fraction accounts for 93.8% of the symmetric differentiative divisions (L→N/N) and might therefore explain our observation that the fates of sibling cells are linked (S1.6 Text). The remainder of symmetric differentiative divisions is interpreted to result from random, independent priming in two sister P cells.
The model also allows multiple interpretations for the probability of becoming L versus N (e.g., exposure to differentiation signals, gene expression noise, etc.). One such interpretation is the timing of the priming event after division (i.e., θ can be construed as accounting for a cell cycle restriction point). For instance, if a cell is primed early after division it might differentiate and halt the cell cycle, whereas if the priming event occurs late in G1, the cell might have already committed to cell cycle completion. Such specific reading of the model, which we adopt hereafter, leads to a few qualitative predictions on the dynamics of differentiation. First and foremost, the vast majority of sibling cells (93.8%) from symmetric divisions will have a perfectly synchronized differentiation program. According to the model, differentiated cells stemming from symmetric divisions shall turn on the differentiation program, on average, much earlier than those from asymmetric divisions (S1.6 Text). To quantitatively account for these predictions and compare them to the experimental data, we performed computational simulations of the model (n = 10,000 clones, S11 Fig.) including the observed variability in the cell cycle length as well as the dynamics of the fluorescent reporter (Figs. 5E–G, S9, S11 and S1.3, S1.4, and S1.6 Text; data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.4b58d [34]). The simulations reproduced the differences in the onset of the reporter in cells stemming from symmetric versus asymmetric divisions (Figs. 5E,F, S6E,F, and S9) and also for the high degree of synchronization between sibling cells (Figs. 5G, S6G, and S9). Furthermore, when we included the 11.6% false negative rate in the reporter (S4C Fig.) of the model, the results were not significantly affected (S10 Fig. and S1.6 Text).
The results of the model led us to experimentally characterize the cell cycle. We used FACS-sorting of Pdx1tTA/+;tetO-H2B-GFP+;Ngn3-RFP− cells marking pancreatic progenitors at E14.5 to establish their cell cycle partition and observed that 69% were in G0/G1, 27% in S and 5% in G2/M, whereas 98% of Neurog3-RFP+ cells were in G0/G1 (S7 Fig.). The progenitors thus spend the majority of their time in G1. Our hypothesis is that priming in early G1 would lead to differentiation and exit of the cell cycle, and its mother would thus have an apparent asymmetric cell division. In contrast, priming in late G1 after the cell has committed to complete the cell cycle through DNA replication and mitosis would lead to an apparent symmetric differentiative cell division.
Finally, simulations also predicted the existence of a residual fraction of primed cells that would turn on the reporter immediately before dividing. Noticeably, although most RFP+ cells did not divide, we observed 3 cases of RFP+ cell division producing six cells (Fig. 6, S8 Movie, and S6 Table) and thus accounting for 3.2% ± 2.8% of all tractable RFP cells. The average time between RFP signal acquisition and division was 1.7 ± 0.8 h. This result is in agreement with the previous estimations of the progression of NEUROG3 cells through S-phase (BrdU incorporation) [11] and further shows that the NEUROG3 cells can exceptionally progress through mitosis at an early stage of their life [29].
The longer movies enabled the observation of multiple rounds of division and the quantification of cell cycle parameters (Fig. 3B). They first confirmed that the daughter cells qualified as progenitors after asymmetric cell division based on SOX9 expression (Fig. 1H) were functionally behaving as progenitors. Indeed, among P/R divisions (n = 27), we observed 14 events where the RFP− daughter divided, producing second-generation progeny (S6 Movie and Fig. 3C). Of those 14 cases, we observed two events where one or both granddaughters divided again, producing third-generation progeny. For all of those, immunostaining at the end point revealed that the RFP− progeny were SOX9+ progenitors (Fig. 3E). In such cases, we could calculate the doubling time of daughter progenitor divisions, and we compared it between P/P and P/N or P/R divisions (Fig. 4D). The doubling time of self-renewing progenitors from P/P divisions was shorter than that from P/N or P/R divisions (p = 0.04, 21.0 ± 2.4 h and 26.5 ± 7.3 h, respectively). Moreover, the distribution of data points was greater in the asymmetric cell divisions. Finally, the time-lapse movies revealed that pancreatic epithelial cells were highly dynamic and that two daughters migrated to distances up to 64 μm apart from each other in the 24 h following division regardless of division mode (S8 Fig.).
In this study, we elucidate the contribution of single cell decisions to the balance between expansion and differentiation in the pancreas. Our lineage analysis, combining in vivo genetic clonal tracing with dynamic imaging in explants, reveals the existence of three kinds of divisions: symmetric progenitor self-renewal, symmetric endocrinogenic divisions leading to two NEUROG3+ endocrine progenitors, and asymmetric divisions generating a pancreatic progenitor and an endocrine progenitor. Furthermore, we show that progenitors are stochastically primed for endocrine differentiation, and that timing of induction in NEUROG3+ cells within the cell cycle establishes the division mode. Whereas late-induced cells complete the cell cycle, resulting in a differentiative symmetric division, early induced cells exit the cell cycle, in which scenario their mother would have produced asymmetric progeny. The results can alternatively be interpreted as HNF1B+ cells being a mix of three pre-determined populations, amongst which only one is truly bipotent. However, the clonal analysis performed in vitro shows different proportions of P/P, P/N, and N/N divisions as compared to in vivo, which would not be expected if the three HNF1B+ subpopulations were predetermined (unless some would preferentially die, which was not observed). Our data is most consistent with a model in which all progenitors are similar, except for their cell cycle stage, and can be primed for endocrine specification with a differentiation probability of around 20% in vivo. Future studies should reveal how this probability changes with time. For example, how it evolves to the cessation of differentiation at the end of gestation, leading to homeostatic conditions that rely primarily on slow self-duplication of differentiated populations [35]. On the other hand, a first phase of symmetric progenitor expansion followed by an increase in the probability of differentiation minimizes the time to form mature organs [36] and may also be expected to occur in the pancreas. Analogous studies are also needed in the human pancreas, as the size of the organ and the length of the differentiation stage are much greater, and several parameters such as cell cycle length of progenitors, probability of differentiation, and ratio of symmetric and asymmetric differentiative divisions may differ.
The high correlation between our in vivo and in vitro results (Fig. 3P) rules out erroneous interpretations due to in vitro artefacts and biases caused by subpopulations of progenitors marked by HNF1B at low tamoxifen doses. Spatially, the endocrinogenic divisions were observed in the centre of the pancreas where the HNF1B+ progenitors reside, but no areas of preferential symmetric or asymmetric division were observed.
Our dynamic data, including the synchrony in differentiation of symmetrically produced endocrine progenitor cells and their shorter lag from division to differentiation, argue that the specification event can occur at different phases in the cell cycle conditioning the ability to execute a final division or not (Fig. 5A,B). This is further supported by our analysis of the cell cycle–dependent priming model, which displays a good fit to the experimental results and provides a causal understanding of the dynamics of the process. The model proposes parameters q and θ that can be measured in other systems to test its prevalence, and our analysis of previous data in other organs suggests that it may be more general rather than specific to the pancreas [33].
Although the molecular mechanisms of Neurog3 priming remain to be elucidated, especially whether it is under cell-intrinsic or extrinsic control, our data provide information on the general principles. Intrinsic control may be based on asymmetric inheritance of molecular components during division [21–24] or incremental or oscillatory expression of transcriptional determinants [37]. Our results strongly argue against the iterative asymmetric inheritance of differentiation cues at the time of division, as seen in Drosophila neurogenesis and also reported in the mouse brain [24]. Indeed, if the specification was determined at the time of division, the differentiation should occur after the same lag time in symmetric and asymmetric cell divisions. Moreover, the lag time between division and Neurog3-RFP onset is very heterogeneous ranging from 0 to 20 h (Fig. 4C), which is difficult to reconcile with a specification occurring at the time of division. If either cumulative increase or oscillations of an intrinsic determinant promoting endocrine fate lead to differentiation, the progeny of progenitor daughters arising from asymmetric division may exhibit an endocrinogenic bias. On the contrary, these progeny were all SOX9+ progenitors, which would rather suggest a negative bias. However, the movie duration might have been too short to observe differentiation after the second division. Moreover, we observe a slower doubling time of progenitor daughters from an asymmetric division, which may result from the inheritance of a factor that slows down the cell cycle [38–40]. Incremental specification could explain why the cell cycle time is also more heterogeneous in these progenitors. Our analyses are also compatible with extrinsic specification, for example, in the context of Notch-Delta-mediated lateral inhibition [41]. The apparent discrepancy with differentiation in the nervous system where uneven splitting of molecular cues at mitosis leads to asymmetric cell division requires further investigations in both systems. When quantified, the ratios of asymmetric and symmetric differentiation events are very similar in the pancreas and the nervous system [33], and our model would be compatible with the observation that lengthening of G1 impacts the cell division modes in the cortex [30]. Thus, an assessment of the differentiation dynamics in the nervous system similar to ours would be useful, and the possible existence of asymmetrically inherited of cues in mitotic cells in the pancreas can also be considered.
Our results reveal that the balance between expansion of progenitors and endocrine differentiation can potentially be regulated by either controlling the probability of endocrine cell induction or its timing in the cell cycle to boost the generation of endocrine cells in vitro for a cell therapy of diabetes. Our approach paves the way to establish how the frequency of division and the ratio of the different types of divisions vary over time and how their balance is controlled by signalling pathways such as Notch and FGF.
Genetically engineered mice used for this study were as follows: Pdx1tTA/+ [42], tetO-HIST1H2BJ/GFP (tetO-H2B-GFP) [43], Hnf1bCreER [3], Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J (mT/mG) [44], Neurog3-EYFP [29], and Neurog3-RFP (S3 Fig.). For embryonic stage, noon of the day when vaginal plug appeared was referred as E0.5.
The Neurog3-RFP transgenic construct (S3A Fig.) was generated by fusing 7.6 kb of the Neurog3-promoter [2] with a reporter construct composed of a chimeric intron; turbo RFP (Evrogen); a nuclear localization signal (NLS); a Myc-tagC; a bovine growth hormone polyadenylation signal (bGH-PolyA). Transgenic mouse lines were obtained by pronuclear injection of the construct (Transgenic Core Facility, EPFL, Switzerland). Two different lines were obtained initially, exhibiting similar levels of RFP signal detectable by a wide-field fluorescent microscope, and one of the lines was used for this study. All animals were handled humanely according to the authorized protocols of Switzerland and Denmark.
Dorsal pancreata from E12.5 Pdx1tTA/+;tetO-H2B-GFP or Pdx1tTA/+;tetO-H2B-GFP;Neurog3-RFP were cultured on a fibronectin (Sigma)-coated coverslip, adapted from the previously published protocol [25]. GFP and RFP were readily detectable under wide-field fluorescent microscopes. We used a culture medium composed of Medium 199, 10% fetal calf serum, 1% Fungizone, and 1% penicillin/streptomycin (all from Gibco). After 24 h of culture that enabled stabilization of explant flattening to approximately 80 μm thick, pancreatic explants were imaged at a single-cell resolution using Leica SP5 or SP8 confocal microscopes with a 63X glycerol immersion objective in a humidified, heated, CO2-controlled chamber. Tiled positions (9 [3x3] or 12 [3x4] tiles) were scanned in 256x256–280x280 format with around 40 μm Z-stack (voxel size, 0.506x0.506x1.3 μm3–0.880x0.880x1.25 μm3) every 6 min for 18–48 h. The GaAsP hybrid detection system (Leica HyD™) enabled a substantial reduction of laser power by 62.5% and increase in signal-to-noise ratio resulting in reduced scanning time, compared to conventional photomultiplier detectors. At each time point, it usually took approximately 5 min and 30 s to scan 9–12 tiled positions in 3-D. At the end point of image acquisition, the explants were fixed and prepared for whole-mount immunostaining.
Tiled images were stitched using either Leica AF6000 software or a custom-built Massive Stitcher plugin (Bioimaging and Optics Platform, EPFL, Switzerland) in Fiji. Imaris (Bitplane, Switzerland) software was used to track cells and their divisions in 3-D maximum intensity projection. Once immunostaining was done, NEUROG3+ endocrine progenitor cells from staining images were manually identified on the last frame of time-lapse movies with Pdx1tTA/+;tetO-H2B-GFP explants by GFP superimposition. The identified endocrine progenitors were first back-tracked to monitor their prior divisions. Once a division was observed, the other sister was forward-tracked to the final frame, and its fate was determined from the immunostaining images. For time-lapse movies from explants with Ngn3-RFP in addition to Pdx1tTA/+;tetO-H2B-GFP, RFP+ cells were back-tracked, and the fate of each sister was determined by immunostaining. For the quantification of total cell divisions, due to the technical difficulties in tracking all Sox9+ cells from the immunostaining, we did not trace all the individual cells from those 1,628 divisions, but rather subtracted the tracked divisions that produced NEUROG3 cells from the total number of divisions.
Pregnant mice carrying Hnf1bCreER;mT/mG embryos were injected intraperitoneally with 0.175 mg 4-hydroxy (4-OH) tamoxifen (H6278, Sigma Aldrich) at E13.5. 4-OH tamoxifen was prepared as a 10 mg/mL stock in 90% corn oil and 10% ethanol and diluted to obtain the desired dose. Embryos were harvested at E14.5, and the dorsal pancreas was isolated and subjected to whole-mount immunostaining for GFP, SOX9, and NEUROG3, as indicated below. The fixation procedure eliminates native GFP and Tomato signals. After whole-mount immunostaining, dorsal pancreata were dehydrated through an ascending methanol series and subjected to clearing in a 1:2 solution of benzyl alcohol to benzyl benzoate (BABB). Cleared samples submerged in BABB were mounted on glass depression slides and imaged whole-mount using a Leica SP8 confocal microscope with a 20X oil objective at a 1024x1024 format. 3-D reconstruction of whole-mount imaged pancreata was performed using Imaris (Bitplane), enabling detection of recombined clones while preserving the spatial organization of the pancreas, thereby ensuring detection of clonal progeny by allowing interclone distance measurements. Two-cell clones were identified in 3-D space, and categorized according to SOX9 and NEUROG3 status. Recombined cells were only considered to be of clonal origin if the distance between recombined cells was less than 30 μm after the tracing period, based on live imaging data (S8 Fig.). The results were not sensitive to this parameter as using 60 μm as a maximal distance to be considered as a clone led to the same proportion of the three types of division (S2 Data).
Hnf1bCreER;mT/mG embryos were also used for in vitro clonal analysis by explanting pancreata at E13.5 and growing these at the air–liquid interface on 0.4 μm filters (Millipore). Explants were subjected to a 6 h pulse of 25 nM 4-OH tamoxifen in 100% ethanol to induce labelling at clonal density. Following tracing for 48 h, explants were fixed and subjected to whole-mount staining and imaging as indicated below.
Whole-mount immunostaining was performed after live imaging or for pancreata harvested from the lineage tracing. After fixation with 4% paraformaldehyde (PFA) for 5 min on ice, samples were washed in phosphate buffered saline (PBS) for 5 min three times. Then, they were dehydrated through 50% and 100% methanol, and could be stored at −20°C until later use. When ready, samples were rehydrated through 50% methanol and washing solution, PBS+0.5% Triton X-100 (Tx100). Throughout the procedure, all the solutions contained 0.5% Tx100, and all the incubation was undergone in 4°C. After blocking overnight in blocking solution (1% Bovine serum albumin+0.5% Tx100), samples were incubated with primary antibodies (S7 Table) in blocking solution for 24–48 h. After washing, secondary antibodies were applied overnight, followed by washing. Alexa fluorophore conjugated secondary antibodies (Invitrogen) were used. Stained explants were kept in PBS and imaged using a confocal microscope. For quantification from explants, NEUROG3+ cells were counted manually, and H2B-GFP+ cells were counted using a custom-made macro in Fiji.
Immunostaining of frozen sections from E14.5 Neurog3-RFP pancreata was performed as previously described [6], and images were taken with a Leica DM5500 microscope. Quantification was obtained by manually counting immunopositive cells on every sixth section.
Statistical analyses were done by two-tailed Mann-Whitney U-test using GraphPad Prism software. Values were presented as the mean ± standard deviation.
|
10.1371/journal.pgen.1007177 | WRKY23 is a component of the transcriptional network mediating auxin feedback on PIN polarity | Auxin is unique among plant hormones due to its directional transport that is mediated by the polarly distributed PIN auxin transporters at the plasma membrane. The canalization hypothesis proposes that the auxin feedback on its polar flow is a crucial, plant-specific mechanism mediating multiple self-organizing developmental processes. Here, we used the auxin effect on the PIN polar localization in Arabidopsis thaliana roots as a proxy for the auxin feedback on the PIN polarity during canalization. We performed microarray experiments to find regulators of this process that act downstream of auxin. We identified genes that were transcriptionally regulated by auxin in an AXR3/IAA17- and ARF7/ARF19-dependent manner. Besides the known components of the PIN polarity, such as PID and PIP5K kinases, a number of potential new regulators were detected, among which the WRKY23 transcription factor, which was characterized in more detail. Gain- and loss-of-function mutants confirmed a role for WRKY23 in mediating the auxin effect on the PIN polarity. Accordingly, processes requiring auxin-mediated PIN polarity rearrangements, such as vascular tissue development during leaf venation, showed a higher WRKY23 expression and required the WRKY23 activity. Our results provide initial insights into the auxin transcriptional network acting upstream of PIN polarization and, potentially, canalization-mediated plant development.
| The plant hormone auxin belongs to the major plant-specific developmental regulators. It mediates or modifies almost all aspects of plant life. One of the fascinating features of the auxin action is its directional movement between cells, whose direction can be regulated by auxin signaling itself. This plant-specific feedback regulation has been proposed decades ago and allows for the self-organizing formation of distinct auxin channels shown to be crucial for processes, such as the regular pattern formation of leaf venation, organ formation, and regeneration of plant tissues. Despite the prominent importance of this so called auxin canalization process, the insight into the underlying molecular mechanism is very limited. Here, we identified a number of genes that are transcriptionally regulated and act downstream of the auxin signaling to mediate the auxin feedback on the polarized auxin transport. One of them is the WRKY23 transcription factor that has previously been unsuspected to play a role in this process. Our work provides the first insights into the transcriptional regulation of the auxin canalization and opens multiple avenues to further study this crucial process.
| The phytohormone auxin plays a key role in many aspects of a plant’s life cycle. A unique attribute of auxin is its polarized, intercellular movement that depends, among other components, on the polarly localized PIN-FORMED (PIN) auxin exporters [1–3]. The so-called canalization hypothesis proposes that auxin acts also as a cue in the establishment of new polarity axes during the polarization of tissues by the formation of self-organizing patterns due to the formation of narrow auxin transport channels driven by the polarized auxin carriers from an initially broad domain of auxin-transporting cells [4–6]. Canalization has been implied to mediate multiple key plant developmental processes, including formation of new vasculature [7], regeneration after wounding [8, 9], and competitive control of apical dominance [10–12]. Whereas the molecular details of canalization are largely unknown, the key constituents are (i) the feedback regulation of the auxin transport directionality by auxin and (ii) the gradual concentrating and narrowing of auxin channels [4]. The auxin feedback on the transport directionality can be realized by the auxin impact on the PIN polarity [8] and might be related to an auxin effect on clathrin-mediated internalization of PIN proteins [13, 14], but the connection is still unclear [15]. Presumably, this feedback regulation of the PIN repolarization also plays a role in the establishment of the embryonic apical-basal axis [16, 17], during organogenesis [18], and termination of shoot bending responses [19].
Auxin feedback on the PIN polarity can be experimentally approximated by PIN polarity rearrangements after auxin treatment of Arabidopsis thaliana roots. Under standard conditions, PIN1 is localized at the basal (root-ward) sides of endodermal and pericycle cells and cells of the vascular tissue [20], whereas PIN2 exhibits a basal polarity in the young cortex cells, but an apical (shoot-ward) polarity in epidermal cells [21, 22]. After treatment with auxin, PIN1 changes from predominantly basal to also inner-lateral in endodermal and pericycle cells, whereas PIN2 undergoes a localization shift from the basal to also outer-lateral side of cortex cells [8]. The exact molecular mechanism and biological significance of this effect is unclear, but it has so far successfully served as easy, experimentally tractable proxy for auxin feed-back on PIN polarity [8]. It depends on the transcriptional SCFTIR1-Aux/IAA-ARF auxin signalling pathway [23]. In brief, upon auxin binding to the TIR1/AFB receptor family, transcriptional repressors and co-receptors of the Aux/IAA class are degraded, in turn releasing auxin response transcription activators of the ARF family [24, 25].
In a heat-shock (HS)-inducible HS::axr3-1 line expressing a mutated, nondegradable version of the IAA17 transcriptional repressor [25, 26], as well as in the arf7 arf19 double mutant defective for these two functionally redundant transcriptional activators expressed in primary roots [27], auxin is no longer effective in mediating PIN polarity rearrangements in the root meristem [8]. These results suggest that transcriptional auxin signalling regulates the cellular abundance of so far unknown regulators, which, in turn, modify subcellular sorting or trafficking pathways and other polarity determinants, ultimately leading to changes in the polar PIN distribution.
In this work, we carried out an expression profiling experiment in Arabidopsis roots to identify potential regulators of the PIN polarity that are transcriptionally regulated by auxin signalling. We identified several novel regulators and characterized in more detail the transcription factor WRKY23 and its role in auxin-mediated PIN polarization, thus providing initial insights into a molecular mechanism of the auxin feedback on the directional auxin flow–one of the key prerequisites of canalization.
The rationale behind the microarray approach was to search for genes that were (i) regulated by auxin in roots under conditions when auxin changes PIN polarity and (ii) their auxin regulation is mediated by the IAA17 (AXR3) transcriptional repressor. First, to look for auxin-induced genes, we matched data from NAA-treated and untreated heat-shocked wild type (WT) Columbia-0 (Col-0) control seedlings and found 523 auxin-induced genes, with a minimum of two-fold difference. As in the HS::axr3-1 line under the same conditions auxin fails to induce PIN polarity changes (Fig 1A and 1B) [8], we compared heat-shocked and auxin-treated Col-0 seedlings to similarly handled HS::axr3-1 seedlings, expressing the auxin-resistant version of IAA17 (AXR3) and we identified 667 genes (Fig 1C). The overlap of this set with the 523 auxin-induced genes yielded 245 genes induced by auxin and regulated downstream of IAA17 (S1 Table), including PATELLIN2 and PATELLIN6 that encode phosphatidylinositol transfer proteins, concomitantly characterized to be crucial for the regulation of embryo and seedling patterning in Arabidopsis [28]. Further comparison with published microarray data on arf7 arf19 mutant seedlings [29], which are also ineffective in rearranging the PIN polarity [8], yielded a final list of 125 genes (S2 Table), of which some had previously been found to be involved in PIN polarity regulation, including the AGC3 kinase PINOID (PID). and its homologs WAG1 and WAG2 are known to phosphorylate PIN proteins [30], contributing to the control of their polar distribution [31–33]. Nevertheless, overexpression of PID was shown to be dominant over the auxin-induced PIN lateralization [8]. Another identified candidate with a known role in the PIN polar distribution was the phosphatidylinositol-4-phosphate 5 kinase PIP5K1. This protein, together with its close homolog PIP5K2, is enriched on basal and apical membrane domains and they are required for PIN trafficking [34, 35] and localization [36, 37]. Other candidates for polarity determinants include several previously known players in auxin-mediated plant development, such as RUL1, a leucine-rich repeat receptor-like kinase regulating cambium formation, a process linked to PIN polarity control [38].
Auxin-dependent PIN lateralization in the root meristem requires a rather prolonged auxin treatment [8], hinting at the involvement of a whole cascade of transcriptional processes. Therefore, we looked for additional auxin-induced transcription factor (TF) genes, which, based on their analogous behaviour in similar experiments and on their known functions, would be potential candidates for having a role in auxin-mediated development. The list of candidates contains e.g. MINI ZINC FINGER1 (MIF1), affecting auxin responses during ectopic meristem formation [39], but also WRKY23. WRKY genes belong to a plant-specific family of 72 TFs in Arabidopsis, typically associated with plant defense processes and plant-pathogen interactions [40]. These genes were named by a shared sequence motif of 60 amino acids containing a conserved domain of seven invariant amino acids (WRKYGQK) [41]. The WRKYGQK motif provides a high binding preference and contacts a 6-bp DNA sequence element–the W-box (/TTGACT/C) contained in target gene promoters [40, 42]. Distinct WRKY TFs have distinct selective binding preferences to certain W-box variants [43]. The role of WRKY23 has been established in plant defence processes during plant-nematode interactions, but also in auxin transport regulation by flavonol biosynthesis that affects root and embryo development. In Arabidopsis embryos, the WRKY23 expression attenuates both auxin-dependent and auxin-independent signalling pathways toward stem cell specification [44–46]. In addition, WRKY23 is unique within its gene family, because none of the other WRKY genes in these experimental conditions was responsive to auxin and, thus, present in the gene selection (S2 Table). In this work, we focused on one of the transcription factors fulfilling our selection criteria, and investigated the role of WRKY23-dependent transcriptional regulation in auxin-dependent PIN repolarization.
First, we confirmed and analysed the auxin regulation of WRKY23 expression. Promoters of auxin-inducible genes typically contain tandem-localized auxin response elements (AuxREs) that are recognised by auxin response factors (ARFs) [47, 48]. ARFs dimerize to act as molecular callipers and provide specificity to the auxin-dependent gene regulation by measuring the distance of AuxREs in the element pair at the promoter [48]. The length of the intergenic region between the 3’-UTR of the previous gene UPBEAT (UPB; At2g47270) and the 5’-UTR of WRKY23 (At2g47260) is 4.5 kbp. The predicted 2.4-kbp WRKY23 promoter by the AGRIS tool [49] contains 10 AuxRE and AuxRE-like sites and the extended promoter of 3.2 kbp used for native promoter fusion construct [44] contains two additional AuxRE sites (Fig 2A). Such a density of auxin-regulatory sequences in the promoter makes direct regulation by ARF-dependent auxin signalling a plausible scenario.
In accordance with these results, we found that WRKY23 is auxin inducible in a dose- and time-dependent manner. When we treated Arabidopsis seedlings with 100 nM NAA for 4 h, the WRKY23 transcription increased 2-fold, and 1 μM NAA led to a 6-fold increase (Fig 2B). Time response experiments at the consensus concentration of 10 μM NAA used in PIN lateralization experiments [8] revealed that the WRKY23 transcription starts to increase approximately after 1.5 h of auxin treatment with a stronger increase after between 2 and 4 h (Fig 2C). This relatively slow auxin-mediated transcriptional regulation of WRKY23 is well within the time frame for the auxin-mediated PIN lateralization that also occurs strongly only after 4 h [8]. The dependence on the auxin signalling was further supported by the compromised WRKY23 auxin inducibility in the HS::axr3-1 and arf7 arf19 mutants (Fig 2D and 2E). These results show that the WRKY23 transcription depends on the SCFTIR1-Aux/IAA-ARF auxin signalling pathway and confirm WRKY23 as a candidate regulator of auxin-mediated PIN polarization.
A transgenic line harbouring the uidA reporter gene (or GUS-coding gene) under the control of a 3.2-kb upstream sequence from WRKY23 (WRKY23::GUS), whose expression pattern has previously been confirmed by in situ hybridization [44, 45], revealed that auxin induces the ectopic expression of WRKY23 in root tissues, partly overlapping with root regions, in which the PIN lateralization can be observed (S1G and S1H Fig). Without auxin treatment, the expression pattern of WRKY23 partially overlaps with the DR5 auxin response reporter (S1G and S1I Fig) and auxin distribution as revealed by anti-IAA immunolocalization [44, 45, 50]. Previously, WRKY23 has been shown to be expressed in all apical cells of an octant stage embryo and at heart stage to be detected in both the root and the shoot stem cell niches (S1D and S1E Fig) [46], possibly indicating that WRKY23 has—besides its role in root development—also a function in shoot development. We found WRKY23::GUS expression in pollen grains (S1C Fig), the shoot apical meristem (SAM) (S1A Fig and Fig 2F), as well as at the hydathodes of cotyledons (S1F Fig), coinciding with known auxin response maxima [51]. Sectioning the SAM revealed specific WRKY23 expression in the L1, L2, and L3 layers (S1A Fig). WRKY23 promoter activity was prominently associated with the vascular tissues of flowers, cotyledons, and leaves (S1B and S1F Fig and Fig 2G). Notably, the WRKY23 expression mirrored the pattern of developing leave vasculature with the highest expression in cells adjacent to the differentiated xylem (Fig 2G) and were detected in a venation-like pattern even before any morphological changes typical for the differentiated vasculature were visible (Fig 2F and 2G). In the previous, external auxin source-mediated canalization experiments in pea stems, the PIN channels were preceding the formation of vasculature and later the differentiated xylem formed adjacent to the PIN channels [11]. Thus, the WRKY23 expression pattern in Arabidopsis largely overlaps with presumptive PIN channels being consistent with a role of WRKY23 in venation patterning of leaves–a process regulated by the polarized auxin transport [51, 52].
In summary, the presence of auxin-responsive elements in the promoter, the auxin-inducibility of the WRKY23 expression together with its dependence on AXR3, ARF7 and ARF19 activities indicate that the WRKY23 transcription is regulated by Aux/IAA- and ARF-dependent auxin signalling. In addition, the association of the WRKY23 expression with developing vasculature is consistent with a possible involvement of WRKY23 in the auxin-mediated PIN polarization process.
Next, we tested whether an altered WRKY23 expression or activity affected the auxin regulation of the PIN1 and PIN2 protein localization. A strong constitutive overexpression of WRKY23 was obtained by means of a GAL4-VP16-UAS transactivation system (RPS5A>>WRKY23) [45, 46, 53]. The 35S promoter-driven WRKY23 line (35S::WRKY23) as well as also 35S promoter-driven dexamethasone-glucocorticoid (DEX/GR) receptor system (35S::WRKY23-GR) were used for constitutive overexpression, eventually, with inducible nuclear localization [45, 46]. Constitutive overexpression of WRKY23 had an impact on the PIN2 but not PIN1 polarity. It caused the PIN2 lateralization in root cortex cells, to some extent mimicking the application of auxin (Fig 3A and 3B). Subsequent treatment with NAA further increased lateralization of PIN2 in cortex cells and caused increased lateralization of PIN1 as compared to wild type (Fig 3A and 3B and S2C and S2D Fig). An inducible WRKY23 gain-of-function line had a similar effect: seedlings of a 35S::WRKY23-GR line treated with DEX to induce WRKY23-GR translocation to the nucleus, resulted in PIN2 but not PIN1 lateralization in the cortex cells. Again, additional NAA treatment had an additive effect on PIN2 lateralization and caused a stronger PIN1 lateralization than as seen in the wild type (S3C and S3D Fig and S2C and S2D Fig).
Thus, both constitutive and inducible WRKY23 gain-of-function consistently led to PIN2 lateralization and increased the auxin-mediated PIN1 and PIN2 lateralization.
In complementary experiments, we tested the downregulation effect of the WRKY23 function. The large WRKY family of homologous proteins has an extensive functional redundancy among individual members [54]. As the functional compensation of wrky23 loss-of-function by other members was likely, given the large size of the WRKY gene family, we used a dominant-negative approach with the chimeric repressor silencing technology [55]. This technology is based on a translational fusion of an activating TF with the repressor domain SRDX, thus inhibiting the expression of target genes. The transactivation activity of WRKY23 had previously been verified in a tobacco transient expression assay, in which the activating or repressing potential of the TF fused to GAL4 had been checked in the presence of a UAS::Luciferase construct [45].
Plants expressing WRKY23-SRDX under both the native and constitutive promoters showed a clear auxin insensitivity in PIN2 lateralization, namely the auxin treatment did not lead to lateralization when compared to the controls (S3A and S3B Fig). Notably, PIN1 lateralization did not change visibly after NAA treatment (S2C and S2D Fig).
To investigate intrafamily redundancy and to assess specifically the role of WRKY23 on the auxin effect on the PIN polarity, we isolated two T-DNA insertional mutants in the WRKY23 locus, designated wrky23-1 and wrky23-2 (Fig 4A). The quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis revealed that both alleles are knock-downs, wrky23-1 having more downregulated expression (Fig 4B).
Similarly to the WRKY23-SRDX lines, both wrky23 mutant alleles showed a reduced PIN2 lateralization response to auxin treatment and, additionally, also reduced PIN1 lateralization. Specifically, following the NAA treatment, the PIN1 and PIN2 lateralization in root endodermal cells was diminished in the wrky23-2 weaker knock-down and, even more so, in the stronger wrky23-1 allele (S2A and S2B Fig and Fig 3C and 3D). The observed opposite effects of WRKY23 gain- and loss-of-function on the PIN lateralization suggested that WRKY23 plays an important role in the auxin-mediated PIN polarity rearrangements.
The importance of a tight PIN polarity regulation for directional auxin fluxes and plant growth and development has been demonstrated previously [2, 3]. Therefore, we analysed the phenotypes related to PIN polarity or auxin transport in transgenic lines with an altered expression or activity of WRKY23. 35S::WRKY23 overexpressing plants show growth retardation and root meristem patterning defects [45]. Also, dominant negative lines showed severe defects in lateral root organogenesis [45]. Both WRKY23-SRDX and 35S::WRKY23 lines had shorter roots than those of Col-0 (S4A Fig) and WRKY23-SRDX showed defects in gravitropism, similar to those observed in the auxin transport mutant pin2/eir1 [56, 57]. Notably, native promoter-driven WRKY23-SRDX displayed a significant increase in lateral root density (S4B Fig). Notably, none of these phenotypical defects, including root meristem disorganization, root growth inhibition, and lateral root development alteration, were observed in the wrky23 mutant alleles (S4A and S4B Fig), suggesting that these more pleiotropic defects are not related to the WRKY23 action specifically, but they could reflect a broader role of the WRKY gene family in plant development.
The canalization hypothesis proposed that the leaf venation pattern depends on the auxin feedback on the PIN polarity [58]. We analysed several features of vascular defects in cotyledons.–bottom disconnectivity of l2 vein loops (BD), upper disconnectivity of l1 vein loops (UD), extra loops (EL), less loops (LL) and appearance of higher order structures (HS) (Fig 4C–4E). In plants expressing WRKY23::WRKY23-SRDX and 35S::WRKY23-SRDX, we observed vasculature patterning defects manifested by increased incidence in BD, HS and EL On the other hand, both wrky23-1 and wrky23-2 mutant alleles showed more defects in UD and LL (Fig 4C).
Next, we tested the PIN1 polarity during vascular tissue development by means of anti-PIN1 antibody staining on young first leaves. In the WT leaves, the staining revealed a pronounced PIN1 polarization along the basipetal (rootward) direction (S4C Fig). In the 35S::WRKY23 and WRKY23-SRDX lines, the typical PIN1 polarity was partly or completely abolished in some veins or their parts (S4C Fig). Similar PIN1 polarity defects were also found in wrky23-1 and wrky23-2 lines (Fig 4F and S4C Fig). The venation defects might be interpreted in terms of defective canalization (as suggested by the PIN1 polarity defects), although the venation defects differ somewhat from defects induced by auxin transport inhibition [51, 52]. This observation indicates that interference with the PIN polarization does not have the same consequence as inhibition of PIN auxin transport activity.
In summary, our genetic analysis revealed that from the numerous functions of the WRKY family in the regulation of plant development [45, 46], WRKY23 is more specifically involved in auxin-mediated PIN polarity rearrangements and leave venation patterning.
Classical experiments have led to the formulation of the so-called canalization hypothesis that proposes an auxin feedback on the auxin transport and consequent formation of auxin channels as a central element of multiple self-organizing developmental processes; in particular formation and regeneration of vasculature [7]. In canalization, the auxin transport through an initially homogeneous tissue follows a self-organizing pattern, leading from initially broad fields of auxin-transporting cells to eventually a narrow transport channel, consequently establishing the position of future vascular veins [6]. This hypothesis [4, 5] is further supported by successful modelling efforts based on the concerted cellular polarization via a feedback mechanism, by which auxin influences the directionality of its own flow by polarity rearrangement of auxin carriers [6, 15, 59–62]. Most of these models rely on hypothetical propositions, such as auxin flux sensors or direct cell-to-cell communication, giving testimony of our lack of understanding how canalization works mechanistically. However, the auxin impact on the PIN polarization has been experimentally demonstrated in different contexts and this effect has been shown to rely on the transcriptional gene expression activation through auxin signalling [8, 9, 11, 19].
Our transcriptional profiling experiments on auxin-dependent PIN rearrangements in Arabidopsis roots provide insight into the transcriptional reprogramming during auxin-mediated PIN polarity rearrangements and identify potential downstream molecular components in this process, including established PIN polarity regulators, such as PID, PIP5K, and PATELLINS [28, 30, 37, 63], validating the soundness of the experimental concept. Among a number of novel components awaiting further characterization, we also found the transcriptional activator WRKY23.
WRKY23 is an auxin-responsive gene. The local upregulation of the WRKY23 expression following the auxin application is consistent with a possible involvement in the PIN repolarization process. The WRKY23 transcription is induced by auxin in a dose- and time-dependent manner and it is reminiscent of the expression pattern of the DR5rev auxin signalling reporter. Notably, WRKY genes are traditionally known to be involved in defensive processes in plants. More and more, this limited functional spectrum has been broadened by studies uncovering the involvement of these TFs in developmental and physiological processes other than plant defense [45, 46, 64, 65]. In the case of WRKY23, besides a role in plant-nematode interaction with subsequent activation of auxin responses, participation in auxin transport through flavonol synthesis in the root as well as a function in a mp/bdl-dependent pathway in embryo development have been demonstrated [44–46].
We show that WRKY23 is a crucial factor required for auxin-mediated PIN polarity rearrangements, because gain-of-function and dominant-negative WRKY23 lines as well as wrky23 mutants were strongly affected in this process. These defects at the cellular level revealed by the exogenous auxin application appears to be developmentally relevant, because wrky23 mutants are defective also in the PIN1 polarization process during vascular tissue formation of leaf venation and consequently in vascular tissue formation. Notably, increased PIN2 but not PIN1 lateralization in the WRKY23 overexpression lines and PIN2 but not PIN1 insensitivity to auxin treatment in WRKY23-SRDX lines indicate a partly diverging mechanism controlling PIN1 and PIN2 relocation. This is consistent with reported differences in PIN1 and PIN2 trafficking mechanisms [66].
Our results also suggest that WRKY23 is a critical player in auxin feedback on PIN polar localization. As a TF, WRKY23 is probably not directly involved in regulating localization of transmembrane proteins, such as PIN proteins. Instead, this work opens avenues for future studies revealing the WRKY23-dependent transcriptional network. The identification of WRKY23 and its role in the auxin feedback on the PIN polarity along with other established PIN polarity regulators proves that our transcriptomics dataset can be mined in the future to identify additional regulators. Ultimately, it will provide insights into the molecular mechanism of this key aspect of the canalization-dependent regulation of plant development.
All Arabidopsis thaliana (L.) Heynh. lines were in Columbia-0 (Col-0) background. The insertional mutants wrky23-1 (SALK_003943) and wrky23-2 (SALK_38289) were obtained from NASC and genotyped with the primers listed in S3 Table. The arf7 arf19 double mutant and the HS::axr3-1 transgenic line have been described previously [26, 29] as well as the DR5::GUS [18] and PIN1-GFP [67]. For RPS5A>>WRKY23 analyses, the F1 generation of a RPS5A::GAL4VP16 [53] × UAS::WRKY23 [45] cross was analysed and compared with the F1 generations from the UAS::WRKY23 × WT Col-0 and RPS5A::GAL4VP16 × WT Col-0 crosses. WRKY23::GUS, 35S::WRKY23-GR,35S::WRKY23, WRKY23::WRKY23-SRDX, and 35S::WRKY23-SRDX have been described previously [44, 45]. Seeds were surface-sterilized overnight by chlorine gas, sown on solid Arabidopsis medium (AM+; half-strength MS basal salts, 1% [w/v] sucrose, and 0.8% [w/v] phytoagar, pH 5.7), and stratified at 4°C for at least 2 days prior to transfer to a growth room with a 16-h light/8-h dark regime at 21°C. The seedlings were grown vertically for 4 or 6 days, depending on the assay.
Arabidopsis seedlings were treated with auxin or chemicals in liquid AM+ at 21°C in a growth room with the following concentrations and times: for α-naphthaleneacetic acid (NAA; Sigma-Aldrich) at 10 μM for 4 h; dexamethasone (DEX; Sigma-Aldrich) 10 μM for 24 h. Mock treatments were done with equivalent amounts of DMSO.
Wild type Col-0 and HS::axr3-1 seeds were grown vertically on AM+ plates for 5 days. We applied a 40 min heat shock at 37°C to the seedlings, followed by a 1.5-h recovery at normal growth temperature. Subsequently, the seedlings were transferred to liquid AM+ and treated with 10 μM NAA or DMSO for 4 h. Afterward, the lower third of 100–130 roots from each treatment was cut off, frozen in liquid N2. RNA was extracted with the RNAeasy mini kit (Qiagen). Probes were prepared and hybridized to the Arabidopsis ATH1–121501 gene expression array (Affymetrix) as described [68]. Expression data for Col-0, HS::axr3-1, both NAA and mock treated, had been deposited under the ArrayExpress number E-MEXP-3283. Expression profiling data for arf7 arf19 (ArrayExpress: E-GEOD-627) have been published previously [29]. Raw data were pairwise analyzed with the logit-t algorithm [69] with a cutoff of p = 0.05.
RNA extraction, cDNA synthesis, and quantitative (q)RT-PCR were done as described [37]. Selected candidate gene transcript levels were quantified with qRT-PCR with specific primer pairs, designed with Primer-BLAST (http://www.ncbi.nlm.nih.gov/tools/primer-blast/). Transcript levels were normalized to GAMMA-TUBULIN 2 (TUB2; AT5G05620), which was constitutively expressed and auxin independent across samples. All PCRs were run in three biological replicates per three technical repeats. The data were processed with a qRT-PCR analysis software (Frederik Coppens, Ghent University-VIB, Ghent, Belgium). Primers used in this study are listed in the S3 Table.
PIN immunolocalizations of primary roots and young leaves were carried out as described [70]. The antibodies were used as follows: anti-PIN1, 1:1000 [13] and anti-PIN2, 1:1000 [71]. For primary roots, the secondary goat anti-rabbit antibody coupled to Cy3 (Sigma-Aldrich) was diluted 1:600. For young leaves, the secondary goat anti-rabbit antibody coupled to Alexa Fluor 488 (Sigma-Aldrich) was diluted 1:600. For confocal microscopy, a Zeiss LSM 700 confocal microscope was used. The PIN relocalization was quantitative analysed as described [8], at least 3 experiments were performed for each observation. Note that the absolute levels of the PIN lateralization index may vary between individual experiments (depending on the anti-PIN signal strength), but the relative differences are always consistent.
All measurements were done with ImageJ (http://rsb.info.nih.gov/ij). For the root length analysis 6-day-old seedlings were scanned and root lengths were measured. For the lateral roots analysis 10-day-old seedlings were scanned and lateral root density was calculated from ratio number of LR/root length.
To detect β-glucuronidase (GUS) activity, seedlings were incubated in reaction buffer containing 0.1 M sodium phosphate buffer (pH 7), 1 mM ferricyanide, 1 mM ferrocyanide, 0.1% Triton X-100, and 1 mg/ml X-Gluc for 2 h in the dark at 37°C. Afterward, chlorophyll was removed by destaining in 70% ethanol and seedlings were cleared.
Tissues (seedlings and cotyledons) were cleared in a solution containing 4% HCl and 20% methanol for 15 min at 65°C, followed by a 15-min incubation in 7% NaOH and 70% ethanol at room temperature. Next, seedlings were rehydrated by successive incubations in 70%, 50%, 25%, and 10% ethanol for 5 min, followed by incubation in a solution containing 25% glycerol and 5% ethanol. Finally, seedlings were mounted in 50% glycerol and monitored by differential interference contrast microscopy DIC (Olympus BX53) or a stereomicroscope (Olympus SZX16).
|
10.1371/journal.pcbi.1003053 | Effect of Regulatory Architecture on Broad versus Narrow Sense Heritability | Additive genetic variance (VA) and total genetic variance (VG) are core concepts in biomedical, evolutionary and production-biology genetics. What determines the large variation in reported VA/VG ratios from line-cross experiments is not well understood. Here we report how the VA/VG ratio, and thus the ratio between narrow and broad sense heritability (h2/H2), varies as a function of the regulatory architecture underlying genotype-to-phenotype (GP) maps. We studied five dynamic models (of the cAMP pathway, the glycolysis, the circadian rhythms, the cell cycle, and heart cell dynamics). We assumed genetic variation to be reflected in model parameters and extracted phenotypes summarizing the system dynamics. Even when imposing purely linear genotype to parameter maps and no environmental variation, we observed quite low VA/VG ratios. In particular, systems with positive feedback and cyclic dynamics gave more non-monotone genotype-phenotype maps and much lower VA/VG ratios than those without. The results show that some regulatory architectures consistently maintain a transparent genotype-to-phenotype relationship, whereas other architectures generate more subtle patterns. Our approach can be used to elucidate these relationships across a whole range of biological systems in a systematic fashion.
| The broad-sense heritability of a trait is the proportion of phenotypic variance attributable to genetic causes, while the narrow-sense heritability is the proportion attributable to additive gene effects. A better understanding of what underlies variation in the ratio of the two heritability measures, or the equivalent ratio of additive variance VA to total genetic variance VG, is important for production biology, biomedicine and evolution. We find that reported VA/VG values from line crosses vary greatly and ask if biological mechanisms underlying such differences can be elucidated by linking computational biology models with genetics. To this end, we made use of models of the cAMP pathway, the glycolysis, circadian rhythms, the cell cycle and cardiocyte dynamics. We assumed additive gene action from genotypes to model parameters and studied the resulting GP maps and VA/VG ratios of system-level phenotypes. Our results show that some types of regulatory architectures consistently preserve a transparent genotype-to-phenotype relationship, whereas others generate more subtle patterns. Particularly, systems with positive feedback and cyclic dynamics resulted in more non-monotonicity in the GP map leading to lower VA/VG ratios. Our approach can be used to elucidate the VA/VG relationship across a whole range of biological systems in a systematic fashion.
| The broad-sense heritability of a trait, , is the proportion of phenotypic variance attributable to genetic causes, while the narrow-sense heritability , is the proportion attributable to additive gene action. The total genetic variance includes the variance explained by intra-locus dominance () and inter-locus interactions (). The reasons for and importance of this non-additive genetic variance that distinguishes the two heritability measures has been subject to substantial controversy for more than 80 years (e.g., [1]–[6]). It was recently shown through statistical arguments that for traits with many loci at extreme allele frequencies, much of the genetic variance becomes additive with h2/H2 (or equivalently VA/VG) typically >0.5 [3]. In populations with intermediate allele frequencies, such as controlled line crosses, lower VA/VG ratios are often reported [7], [8]. This is illustrated in Table 1, which summarizes estimated VA/VG ratios from a collection of studies on such populations. This wide range of h2/H2 ratios reported for line crosses cannot be explained by an allele-frequency argument, and putative explanations must be based on how the regulatory architecture of the underlying biological systems shape the genotype-phenotype (GP) map.
It is important to understand the causal underpinnings of the observed variation in h2/H2 ratios within and between biological systems for several reasons. In human quantitative genetics, where twin studies are commonly used, most heritability estimates refer to H2 [9]. In cases where h2/H2 is low, this can lead to unrealistic expectations about how much of the underlying causative variation may be located by linear QTL detection methods [6]. On the other hand, low narrow sense heritability for a given complex trait does not necessarily imply that the environment determines much of the variation. In evolutionary biology, additive variance is the foremost currency for evolutionary adaptation and evolvability. Important questions in this context are for example (i) to which degree is there selection on the regulatory anatomies themselves to maintain high additive variance, (ii) are there organizational constraints in building adaptive systems such that in some cases a low h2/H2 ratio must of necessity emerge while the proximal solution is still selected for? Moreover, in production biology with genetically modified, sexually reproducing organisms, one would like to ensure that the modifications would be passed over to future generations in a fully predictable way. Thus, one would like to ensure that the modification becomes highly heritable in the narrow sense.
As a step towards a physiologically grounded understanding of the variation of the h2/H2 relationship across biological systems or processes, we posed the question: Are there regulatory structures, or certain classes of phenotypes, more likely to generate low VA/VG ratios than others? Addressing this question requires the linking of genetic variation to computational biology in a population context (e.g., [10]–[19]), so-called causally-cohesive genotype-phenotype (cGP) modeling [15], [17], [18]. We applied this approach to five well-validated computational biology models describing, respectively, the glycolysis metabolic pathway in budding yeast [20], the cyclic adenosine monophosphate (cAMP) signaling pathway in budding yeast [21], the cell cycle regulation of budding yeast [22], the gene network underlying mammalian circadian rhythms [23], and the ion channels determining the action potential in mouse heart myocytes [24] These models differ in their regulatory architecture; below, we show that they also differ in the range of VA/VG ratios that they can exhibit. In particular, positive feedback regulation and oscillatory behavior seem to dispose for low VA/VG ratios. The results suggest that our approach can be used in a generic manner to probe how the h2/H2 ratio varies as a function of regulatory anatomy.
The five cGP models were built and analyzed with the cgptoolbox (http://github.com/jonovik/cgptoolbox) an open-source Python package developed by the authors; further source code specific to the simulations in this paper is available on request. In the following we describe the three main parts of the workflow: (i) the mapping from genotypes to parameters, (ii) the mapping from parameters to phenotypes, i.e. solving the dynamic models and (iii) the setup of Monte-Carlo simulations combining the two mappings (Figure S1). For each model, we briefly describe its origins, the software used to solve it, which parameters were subject to genetic variation, what phenotypes were recorded, and criteria for omitting outlying datasets. Figures S2, S3, S4, S5, S6 shows graphical representations of the five model systems and Text S1 contains more detailed descriptions of all five models.
The five cGP models studied in this work differ in several ways, both in their function and the underlying network structure. The glycolysis and cAMP models are both pathway models with an acyclic series of reactions transforming inputs to outputs. The glycolysis model [20] is more advanced than the metabolic models in earlier genetically oriented studies (e.g., [3], [31], [32]) as it contains many different types of enzyme kinetics as well as negative feedback regulation of some enzyme activities through product inhibition. The cAMP model [21] contains a number of negative feedback loops, but overall it also has a clear pathway structure where the glucose signal is relayed from G-proteins to cAMP to the target kinase PKA. Both these two models have in common relatively simple dynamics with solutions converging to a stable steady state (Figure 1A and B).
In contrast, the three other models show cyclic behavior resulting from an interplay between positive and negative feedback loops (Figure 1 C–E). However, there are clear differences between these models too. The heart cell model [24] is an excitable system with feedback mechanisms including calcium-induced calcium release and several voltage-dependent ion channels. In contrast to pacemaker cells, it relies on external pacing to initiate the action potential. The circadian rhythm model [23], [33] is a gene expression network with intertwined positive and negative transcriptional feedback loops, giving a limit cycle oscillator with sustained oscillations even in continuous darkness. The cell cycle model [22] centers around a positive feedback loop between B-type cyclins in association with cyclin dependent kinase and inhibitors of the cyclin dependent kinase, which establishes a hysteresis loop causing the cell cycle to emerge from transitions between the two alternative stable steady states.
This crude classification of the five cGP models into pathway models and more complex regulatory systems is clearly reflected in the effective dimensionality of the phenotypes arising in our Monte Carlo simulations. We studied the phenotypic dimensionality for all five cGP models by Principal Component Analysis (PCA) for each Monte Carlo simulation (Figure 2). Across all simulated datasets, 95% of phenotypic variation of the glycolysis and cAMP models can be explained by the first 3 principal components, the cell cycle and heart cell models require the first 5 principal components, and 7 components are required for the circadian model. Since the genotype-to-parameter maps are additive for all five models, these differences in the effective dimensionality of high-level phenotypes indicate that the mappings from parameters to phenotypes are simpler for the pathway models than the other three models. This, together with results on the effect of positive feedback on statistical epistasis in gene regulatory networks [11], suggested that the glycolysis and cAMP models might result in higher VA/VG ratios than the other three models.
The results confirmed our expectations regarding high VA/VG ratios for the glycolysis and cAMP models. Furthermore, a number of distinct patterns emerged. The cAMP model shows the overall highest VA/VG ratios values (Figure 3A and Table S6), with mean and median values above 0.99 across all recorded phenotypes. The 0.05-quantile (i.e. only 5 percent of the Monte Carlo simulations show lower values than this) VA/VG values were above 0.97 for all phenotypes and no values lower than 0.6 were observed. In other words, an intra- and inter-locus additive model of gene action very well approximates the genotype-phenotype maps arising from this cGP model.
The glycolysis model also has mean and median VA/VG values close to 1 for all phenotypes (Figure 3B and Table S7). But compared to the cAMP model, the numbers are clearly lower; the lowest recorded mean value (phenotype BPG) is 0.9 and 0.05-quantile values are below 0.7 for some phenotypes. A few VA/VG values below 0.5 are observed for all phenotypes. The distribution of VA/VG ratios for the cell cycle model (Figure S7 and Table S8) is quite similar to that of the glycolysis model, with a lowest mean VA/VG value of 0.93 for time to peak for Sic1 and with 0.05-quantiles below 0.8 for some phenotypes. VA/VG values below 0.1 are observed for a few Monte Carlo simulations in some phenotypes.
For each of the cAMP, glycolysis and cell cycle models the distributions of VA/VG ratios were quite similar across all phenotypes, and a large majority of the Monte Carlo simulations showed very high ratios. The circadian clock model differs from these three models both in terms of displaying large variation between phenotypes and in terms of having a much larger proportion of low VA/VG values (Figure 4A and Table S9). Four phenotypes stand out with VA/VG distributions that resemble a uniform distribution U(0,1). These are the time from bottom to peak for the phosphorylated and unphosphorylated proteins of Per and Cry, and they have median VA/VG values ranging from 0.46 to 0.70 and 0.05-quantile values in the range 0.04 to 0.10. The remaining phenotypes have somewhat higher VA/VG values, but over half of the recorded phenotypes have 0.05-quantiles below 0.6. Median VA/VG values are below 0.9 for the majority of phenotypes of the action potential model. And all recorded phenotypes have a large proportion of low VA/VG ratios (Figure 4B and Table S10) with 0.05-quantiles in the range 0.18-0-35. The distributions are quite similar across action potential and calcium transient phenotypes, but the time to 90% repolarization for the action potential shows somewhat higher values than the others.
All five cGP models are capable of creating VA/VG ratios close to 1, and except for two phenotypes for the circadian model all median values of VA/VG are well above 0.5. This supports the hypothesis [30] that biological systems tend to involve regulatory machinery that in general leads to considerable additive genetic variance even at intermediate allele frequencies. That is not to say that selection cannot sometimes produce regulatory solutions that tend towards low VA/VG ratios; in fact, the incidence of low VA/VG ratios varied markedly among the five models that we studied. Because the genotype-parameter maps were purely additive, all non-additive genetic variance is a result of non-linearity in the ODE models. The expected effect of introducing non-additivity in the genotype-parameter maps would be a further decrease in the VA/VG ratios.
Our finding that models with complex regulation involving positive feedback loops tend to give lower VA/VG agrees with a previous study on gene regulatory networks [11]. Considering the relatively high VA/VG ratios of the cell cycle model compared to the circadian and action potential models, the following quote from Tyson and Novak's [34] discussion of why the cell-cycle is better understood as a hysteresis loop than as a limit cycle oscillator (LCO), is highly informative: “Generally speaking, the period of an LCO is a complicated function of all the kinetic parameters in the differential equations. However, the period of the cell division cycle depends on only one kinetic property of the cell: its mass-doubling time.” This seems to explain why the genotype-phenotype maps arising from the cell-cycle models are much more linear than the maps from the circadian model, which is an LCO.
In a given population VA/VG is a function of allele frequencies as well as the GP map, and GP maps with strong interactions can still give high VA/VG values in populations with extreme allele frequencies [3]. In populations with intermediate allele frequencies the VA/VG values are determined mainly by the shape of the genotype-phenotype map, and the observed differences between the five cGP models in the distribution of VA/VG values motivates a search for underlying explanatory principles.
The recently proposed concept of monotonicity (or order-preservation) of GP maps seems to be one such principle. In short, a GP map is said to be monotone if the ordering of genotypes by gene content (the number of alleles of a given type) is preserved in the ordering of the associated phenotypic values (see [30] for details). Figure 5 depicts three extreme types of GP maps seen in our simulations. Nearly additive GP maps as shown in Figure 5A give VA/VG values very close to one. GP maps with strong magnitude epistasis, but still order-preserving, typically result in intermediate VA/VG values (Figure 5B), while highly non-monotone or order-breaking GP maps (Figure 5C) showing strong overdominance and/or sign epistasis result in VA/VG values close to zero.
Based on recent results from studies of gene regulatory networks [30], we anticipated that the three cGP models with complex regulation involving positive feedback would result in considerably more non-monotone or order-breaking GP maps than the two pathway models. To test this, we measured the amount of order-breaking in all simulated GP maps (see Methods) and found a very clear pattern (Figure 6). While the datasets from the glycolysis and cAMP models contained only 1.1% and 1.3% GP maps with order-breaking for any locus, those from the cell cycle, circadian and action potential models contained 20.7%, 44.4% and 69.5%, respectively. Moreover, monotone GP maps gave higher VA/VG values than non-monotone GP maps for all five cGP models (Mann-Whitney test; p-values below 1e-10 for all five models).
However, despite the fact that the glycolysis model rarely shows order-breaking even for a single locus, it possesses much lower VA/VG values than the cAMP model. A plausible explanation is that the steady-state concentrations of metabolites can markedly increase for parameter values close to a saddle-node bifurcation point [26]. Simulation outcomes with unstable steady states were discarded, but in those cases where one of the genotypes (i.e. parameter sets) come close to the bifurcation point without crossing it we get genotype-phenotype maps as in Figure 5B, where one genotype (or a small set) gives much higher phenotypic values than the others. Such GP maps, similar to the duplicate factor model in Hill et al. [3], are known to give low VA/VG ratios despite being monotonic. Similar GP maps giving VA/VG ratios close to zero were also found by Keightley [32] in his study of metabolic models possessing null alleles at all loci.
Our main reason for restricting the sampled genetic variation of parameters to within 30% of the published baseline values was to avoid qualitative (or topological) changes of the dynamics. Such qualitative changes are often biologically realistic descriptions of knockouts or other large genetic changes, for example action potentials of alternating amplitude (alternans) [17]; loss of stable circadian oscillation [23]; and non-viable cell-cycle mutants phenotypes [22]. However, since the heritability and variance component concepts are defined for phenotypes showing continuous rather than discrete variation, we sought to avoid such qualitative changes here.
We ran simulations with five polymorphic loci for the cAMP (Figure S8A), glycolysis (Figure S8B), cell cycle (Figure S9) and action potential (Figure S10) models (the circadian model describes only three genes explicitly). The resulting VA/VG values were slightly lower than with three loci, but the overall shape of the distributions and the clear differences between models did not change. This indicates that our findings are of general relevance for oligogenic traits.
It should be emphasized that the five studied cGP models differ in several other aspects than those highlighted here, such as the system size (number of state variables) and the process time scales. These features could also contribute to the observed variation in the distributions of VA/VG ratios. However, such structural differences are unavoidable when the aim is to compare experimentally validated models designed to describe specific biological systems. A complementary approach is to study generic models where system size and equation structure is fixed, while the connectivity matrix can be changed to describe a family of systems [35]. This facilitates graph-theoretic comparison of systems at the expense of some biological realism. We anticipate that the major conclusions from such studies will be similar to ours, but it may very well be that other important generic insights may also come to the fore.
All the models in our study describe parts of the cellular machinery and the resulting phenotypes are thus cellular rather than organismal. We do not think this is a major shortcoming in terms of the main conclusions that emerge from our results. However, we anticipate that application of our approach on multiscale models including cellular, tissue and whole-organ phenotypes [36] will provide a much improved foundation for explaining how properties of the GP map vary across and within biological systems in terms of regulatory anatomy and associated genetic variation [37], [38].
As our approach can be used together with any computational biology model, it has the potential to contribute substantially to a theoretical foundation capable of predicting when we are to expect low or high VA/VG or h2/H2 ratios from the principles of regulatory biology. Causally cohesive genotype-phenotype modeling thus appears to qualify as a promising approach for integrating causal models of biological networks and physiology with quantitative genetics [39]–[44].
|
10.1371/journal.pcbi.1006099 | Biogeography and environmental conditions shape bacteriophage-bacteria networks across the human microbiome | Viruses and bacteria are critical components of the human microbiome and play important roles in health and disease. Most previous work has relied on studying bacteria and viruses independently, thereby reducing them to two separate communities. Such approaches are unable to capture how these microbial communities interact, such as through processes that maintain community robustness or allow phage-host populations to co-evolve. We implemented a network-based analytical approach to describe phage-bacteria network diversity throughout the human body. We built these community networks using a machine learning algorithm to predict which phages could infect which bacteria in a given microbiome. Our algorithm was applied to paired viral and bacterial metagenomic sequence sets from three previously published human cohorts. We organized the predicted interactions into networks that allowed us to evaluate phage-bacteria connectedness across the human body. We observed evidence that gut and skin network structures were person-specific and not conserved among cohabitating family members. High-fat diets appeared to be associated with less connected networks. Network structure differed between skin sites, with those exposed to the external environment being less connected and likely more susceptible to network degradation by microbial extinction events. This study quantified and contrasted the diversity of virome-microbiome networks across the human body and illustrated how environmental factors may influence phage-bacteria interactive dynamics. This work provides a baseline for future studies to better understand system perturbations, such as disease states, through ecological networks.
| The human microbiome, the collection of microbial communities that colonize the human body, is a crucial component to health and disease. Two major components of the human microbiome are the bacterial and viral communities. These communities have primarily been studied separately using metrics of community composition and diversity. These approaches have failed to capture the complex dynamics of interacting bacteria and phage communities, which frequently share genetic information and work together to maintain ecosystem homestatsis (e.g. kill-the-winner dynamics). Removal of bacteria or phage can disrupt or even collapse those ecosystems. Relationship-based network approaches allow us to capture this interaction information. Using this network-based approach with three independent human cohorts, we were able to present an initial understanding of how phage-bacteria networks differ throughout the human body, so as to provide a baseline for future studies of how and why microbiome networks differ in disease states.
| Viruses and bacteria are critical components of the human microbiome and play important roles in health and disease. Bacterial communities have been associated with disease states, including a range of skin conditions [1], acute and chronic wound healing conditions [2, 3], and gastrointestinal diseases, such as inflammatory bowel disease [4, 5], Clostridium difficile infections [6] and colorectal cancer [7, 8]. Altered human viromes (virus communities consisting primarily of bacteriophages) also have been associated with diseases and perturbations, including inflammatory bowel disease [5, 9], periodontal disease [10], spread of antibiotic resistance [11], and others [12–17]. Viruses act in concert with their microbial hosts as a single ecological community [18]. Viruses influence their living microbial host communities through processes including lysis, host gene expression modulation [19], influence on evolutionary processes such as horizontal gene transfer [20] or antagonistic co-evolution [21], and alteration of ecosystem processes and elemental stoichiometry [22].
Previous human microbiome work has focused on bacterial and viral communities, but have reduced them to two separate communities by studying them independently [5, 9, 10, 12–17]. This approach fails to capture the complex dynamics of interacting bacteria and phage communities, which frequently share genetic information and work together to maintain ecosystem structure (e.g. kill-the-winner dynamics that prevent domination by a single bacterium). Removal of bacteria or phages can disrupt or even collapse those ecosystems [18, 23–32]. Integrating these datasets as relationship-based networks allow us to capture this complex interaction information. Studying such bacteria-phage interactions through community-wide networks built from inferred relationships begins to provide us with insights into the drivers of human microbiome diversity across body sites and enable the study of human microbiome network dynamics overall.
In this study, we characterized human-associated bacterial and phage communities by their inferred relationships using three published paired virus and bacteria-dominated whole community metagenomic datasets [13, 14, 33, 34]. We leveraged machine learning and graph theory techniques to establish and explore the human bacteria-phage network diversity therein. This approach built upon previous large-scale phage-bacteria network analyses by inferring interactions from metagenomic datasets, rather than culture-dependent data [28], which is limited in the scale of possible experiments and analyses. We implemented an adapted metagenomic interaction inference model that made some improvements upon previous phage-host interaction prediction models. Previous approaches have utilized a variety of techniques, such as linear models that were used to predict bacteria-phage co-occurrence using taxonomic assignments [35], and nucleotide similarity models that were applied to both whole virus genomes [36] and clusters of whole and partial virus genomes [37]. Our approach uniquely included protein interaction data and was validated based on experimentally determined positive and negative interactions (i.e. who does and does not infect whom). We built on previous modeling work as a means to our ends, and focused on the biological insights we could gain instead of building a superior model and presenting our work as a toolkit. We therefore did not focus on extensive benchmarking against other existing models [36, 37–40]. Through this approach we were able to provide an initial basic understanding of the network dynamics associated with phage and bacterial communities on and in the human body. By building and utilizing a microbiome network, we found that different people, body sites, and anatomical locations not only support distinct microbiome membership and diversity [13, 14, 33, 34, 41–43], but also support ecological communities with distinct communication structures and robustness to network degradation by extinction events. Through an improved understanding of network structures across the human body, we aim to empower future studies to investigate how these communities dynamics are influenced by disease states and the overall impact they may have on human health.
We studied the differences in virus-bacteria interaction networks across healthy human bodies by leveraging previously published shotgun sequence datasets of purified viral metagenomes (viromes) paired with bacteria-dominated whole community metagenomes. Our study contained three datasets that explored the impact of diet on the healthy human gut virome [14], the impact of anatomical location on the healthy human skin virome [13], and the viromes of monozygotic twins and their mothers [33, 34]. We selected these datasets because their virome samples were subjected to virus-like particle (VLP) purification, which removed contaminating DNA from human cells, bacteria, etc. To this end, the publishing authors employed combinations of filtration, chloroform/DNase treatment, and cesium chloride gradients to eliminate organismal DNA (e.g. bacteria, human, fungi, etc) and thereby allow for direct assessment of both the extracellular and fully-assembled intracellular virome (S1A and S1B Fig) [14, 34]. Each research group reported quality control measures to ensure the purity of the virome sequence datasets, using both computational and molecular techniques (e.g. 16S rRNA gene qPCR) (S1 Table). These reports confirmed that the virome libraries consisted of highly purified virus genomic DNA.
The bacterial and viral sequences from these studies were quality filtered and assembled into contigs (i.e. genomic fragments). We further grouped the related bacterial and phage contigs into operationally defined units based on their k-mer frequencies and co-abundance patterns, similar to previous reports (S2 and S3 Figs) [37]. This was done both for dimensionality reduction and to prevent inflation of node counts by using contigs which are expected to represent multiple fragments from the same genomes. This was also done to create genome analogs that we could use in our classification model which was built using genome sequences. We referred to these operationally defined groups of related contigs as operational genomic units (OGUs). Each OGU represented a genomically similar sub-population of either bacteria or phages. Contig lengths within clusters ranged between 103 and 105.5 bp (S2 and S3 Figs).
The original publications reported that the whole metagenomic shotgun sequence samples, which primarily consisted of bacteria, had an average viral relative abundance of 0.4% (S1 Table) [13, 14, 33, 34]. We confirmed these reports by finding that only 2% (6/280 OGUs) of bacterial OGUs had significantly strong nucleotide similarity to phage reference genomes (e-value < 10−25) [13, 14, 33, 34]. Additionally, no OGUs were confidently identified as lytic or temperate phage OGUs in the bacterial dataset using the Virsorter algorithm [44]. We also supplemented the previous virome fraction quality control measures (S1 Table) to find that, in light of the rigorous purification and quality control during sample collection and preparation, 77% (228/298 operational genomic units) still had some nucleotide similarity to a given bacterial reference genome (e-value < 10−25). It is important to note that interpreting such alignment is complicated by the fact that most reference bacterial genomes also contain prophages (i.e. phages integrated into bacterial genomes), meaning we do not know to what extent the alignments were the result of bacterial contaminants in the virome fraction and what were true integrated prophages. As most phages in these communities have been shown to be temperate (i.e. they integrate into bacterial genomes) using methods that included nucleotide alignments of phages to bacterial reference genomes, we expected that a large fraction of those phages were temperate and therefore shared elements with bacterial reference genomes—a trend previously reported [14]. To ensure the purity of our sample sets, we supplemented the quality control measures by filtering out all OGUs that could be potential bacterial contaminants, as described previously [37]. This resulted in the removal of 143 OGUs that exhibited nucleotide similarity to bacterial genomes but no identifiable known phage elements. We were also able to identify two OGUs as representing complete, high confidence phages using the stringent Virsorter phage identification algorithm (class 1 confidence group) [44].
We predicted which phage OGUs infected which bacterial OGUs using a random forest model trained on experimentally validated infectious relationships from six previous publications [36, 45–49]. Only bacteria and bacteriophages were used in the model. The training set contained 43 diverse bacterial species and 30 diverse phage strains, including both broad and specific ranges of infection (S4A and S4B Fig, S2 Table). While it is true that there are more known phages that infect bacteria, we were limited by the information confirming which phages do not infect certain bacteria and attempted to keep the numbers of positive and negative interactions similar. Phages with linear and circular genomes, as well as ssDNA and dsDNA genomes, were included in the analysis. Because we used DNA sequencing studies, RNA phages were not considered (S4C and S4D Fig). This training set included both positive relationships (i.e. a phage infects a bacterium) and negative relationships (i.e. a phage does not infect a bacterium). This allowed us to validate the false positive and false negative rates associated with our candidate models, thereby building upon previous work that only considered positive relationships [36]. It is also worth noting that while a positive interaction is strong evidence that the interaction exists, we must also be conscious that negative interactions are only weak evidence for a lack of interaction because the finding could be the result of our inability to reproduce conditions in which those interactions occur. Altogether we decided to maintain a balanced dataset at the cost of under-sampling the available positive interaction information because the use of such a severely unbalanced dataset often results in over-fit and uninformative model training. However, as an additional validation measure, we used the extensive additional positive interactions as a secondary dataset to confirm that we could identify infectious interactions from a more diverse bacterial and phage dataset. Using this approach, we confirmed that 382 additional phage reference genomes, representing a diverse range of phages, were matched to at least one reference bacterial host genome of the species that they were expected to infect (S5 Fig). Because the model was built on full genomes and used on OGUs, we also assessed whether our model was resilient to incomplete reference genomes. We found that the use of our model on random contigs representing as little as 50% length of the original reference phage and bacterial genomes resulted in minimal reduction in the ability of the model to identify relationships (S6 Fig).
Four phage and bacterial genomic features were used in our random forest model to predict infectious relationships between bacteria and phages: 1) genome nucleotide similarities, 2) gene amino acid sequence similarities, 3) bacterial Clustered Regularly Interspaced Short Palindromic Repeat (CRISPR) spacer sequences that target phages, and 4) similarity of protein families associated with experimentally identified protein-protein interactions [50]. These features were calculated using the training set described above. While the nucleotide and amino acid similarity metrics were expected to identify prophage signatures, the protein family interaction and CRISPR signatures were expected to aid in identifying lytic phages in addition to temperate phages. We chose to utilize these metrics that directly compare nucleotide sequences between sample phages and bacteria, instead of relying on alignment to reference genomes or known marker genes, because we were extrapolating our model to highly diverse communities which we expect to diverge significantly from the available reference genomes. The resulting random forest model was assessed using nested cross validation, and the median area under its receiver operating characteristic (ROC) curve was 0.788, the median model sensitivity was 0.905, and median specificity was 0.538 (Fig 1A). This balance of confident true positives at the cost of fewer true negatives was ideal for this type of dataset which consisted of primarily positive connections (S7 Fig). Nested cross validation of the model demonstrated that the sensitivity and specificity of the model could vary but the overall model performance (AUC) remained more consistent (S8 Fig). This suggested that our model would perform with a similar overall accuracy despite changes in sensitivity/specificity trade-offs. The most important predictor in the model was amino acid similarity between genes, followed by nucleotide similarity of whole genomes (Fig 1B). Protein family interactions were moderately important to the model, and CRISPRs were largely uninformative, due to the minimal amount of identifiable CRISPRs in the dataset and their redundancy with the nucleotide similarity methods (Fig 1B). Approximately one third of the training set relationships yielded no score and therefore were unable to be assigned an interaction prediction (Fig 1C).
We used our random forest model to classify the relationships between bacteria and phage operational genomic units, which were then used to build the interactive network. The master network, analogous to the universal microbiome network concept previously described [51], contained the three studies as sub-networks, which themselves each contained sub-networks for each sample (S9 Fig). Metadata including study, sample ID, disease, and OGU abundance within the community were stored in the master network for parsing in downstream analyses (S9 Fig). The phage and bacteria of the master network demonstrated both narrow and broad ranges of infectious interactions (S10 Fig). Bacterial and phage relative abundance was recorded in each sample for each OGU and the weight of the edge connecting those OGUs was calculated as a function of those relative abundance values. The separate extraction of the phage and bacterial libraries ensured a more accurate measurement of the microbial communities, as has been outlined previously [52, 53]. The master network was highly connected and contained 38,337 infectious relationships among 435 nodes, representing 155 phages and 280 bacteria. Although the network was highly connected, not all relationships were present in all samples. Relationships were weighted by the relative abundances of their associated bacteria and phages. Like the master network, the skin network exhibited a diameter of 4 (measure of graph size; the greatest number of traversed vertices required between two vertices) and included 433 (154 phages, 279 bacteria, 99.5% total) and 38,099 (99.4%) of the master network nodes and edges, respectively (Fig 1E and 1F). Additionally, the subnetworks demonstrated narrow ranges of eccentricity across their nodes (S11 Fig). Graph node eccentricity, a measurement to supplement diameter, is the shortest distance of each node to the furthest other node within the graph. The phages and bacteria in the diet and twin sample sets were more sparsely related, with the diet study consisting of 80 (32 phages, 48 bacteria) nodes and 1,290 relationships, and the twin study containing 130 (29 phages, 101 bacteria) nodes and 2,457 relationships (Fig 1E and 1F). As a validation measure, we identified five (1.7%) examples of phage OGUs which contained similar genomic elements to the four previously described, broadly infectious phages isolated from Lake Michigan (tblastx; e-value < 10−25) [54].
Diet is a major environmental factor that influences resource availability and gut microbiome composition and diversity, including bacteria and phages [14, 55, 56]. Previous work in isolated culture-based systems has suggested that changes in nutrient availability are associated with altered phage-bacteria network structures [25], although this has yet to be tested in humans. We therefore hypothesized that a change in diet would also be associated with a change in virome-microbiome network structure in the human gut.
We evaluated the diet-associated differences in gut virome-microbiome network structure by quantifying how central each sample’s network was on average. We accomplished this by utilizing two common weighted centrality metrics: degree centrality and closeness centrality. Degree centrality, the simplest centrality metric, was defined as the number of connections each phage made with each bacterium. We supplemented measurements of degree centrality with measurements of closeness centrality. Closeness centrality is a metric of how close each phage or bacterium is to all of the other phages and bacteria in the network. A higher closeness centrality suggests that the effects of genetic information or altered abundance would be more impactful to all other microbes in the system. Because these are weighted metrics, the values are functions of both connectivity as well as community composition. A network with higher average closeness centrality also indicates an overall greater degree of connections, which suggests a greater resilience against network degradation by extinction events [25, 57]. This is because more highly connected networks are less likely to degrade into multiple smaller networks when bacteria or phages are randomly removed [25, 57]. We used this information to calculate the average connectedness per sample, which was corrected for the maximum potential degree of connectedness. Unfortunately our dataset was insufficiently powered to make strong conclusions toward this hypothesis, but this is an interesting observation that warrants further investigation. This observation also serves to illustrate the types of questions we can answer with more comprehensive microbiome sampling and integrative analyses.
Using our small sample set, we observed that the gut microbiome network structures associated with high-fat diets appeared less connected than those of low-fat diets, although a greater sample size will be required to more properly evaluate this trend (Fig 2A and 2B). Five subjects were available for use, all of which had matching bacteria and virome datasets and samples from 8-10 days following the initiation of their diets. High-fat diets appeared to exhibit reduced degree centrality (Fig 2A), suggesting bacteria in high-fat environments were targeted by fewer phages and that phage tropism was more restricted. High-fat diets also appeared to exhibit decreased closeness centrality (Fig 2B), indicating that bacteria and phages were more distant from other bacteria and phages in the community. This would make genetic transfer and altered abundance of a given phage or bacterium less capable of impacting other bacteria and phages within the network.
In addition to diet, we observed a possible trend that obesity influenced network structure. This was done using the three mother samples available from the twin sample set, all of which had matching bacteria and phage samples and confirmed BMI information. The obesity-associated network appeared to have a higher degree centrality (Fig 2C), but less closeness centrality than the healthy-associated networks (Fig 2D). These results suggested that the obesity-associated networks may be less connected. This again comes with the caveat that this is only an opportunistic observation using an existing sample set with too few samples to make more substantial claims. We included this observation as a point of interest, given the data was available.
Skin and gut community membership and diversity are highly personal, with people remaining more similar to themselves than to other people over time [13, 58, 59]. We therefore hypothesized that this personal conservation extended to microbiome network structure. We addressed this hypothesis by calculating the degree of dissimilarity between each subject’s network, based on phage and bacteria abundance and centrality. We quantified phage and bacteria centrality within each sample graph using the weighted eigenvector centrality metric. This metric defines central phages as those that are highly abundant (AO as defined in the Methods) and infect many distinct bacteria which themselves are abundant and infected by many other phages. Similarly, bacterial centrality was defined as those bacteria that were both abundant and connected to numerous phages that were themselves connected to many bacteria. We then calculated the similarity of community networks using the weighted eigenvector centrality of all nodes between all samples. Samples with similar network structures were interpreted as having similar capacities for network robustness and transmitting genetic material.
We used this network dissimilarity metric to test whether microbiome network structures were more similar within people than between people over time. We found that gut microbiome network structures clustered by person (ANOSIM p-value = 0.008, R = 1, Fig 3A). Network dissimilarity within each person over the 8-10 day sampling period was less than the average dissimilarity between that person and others, although this difference was not statistically significant (p-value = 0.125, Fig 3B). Four of the five available subjects were used because one of the subjects was not sampled at the initial time point. The lack of statistical confidence was likely due to the small sample size of this dataset.
Although there was evidence for gut network conservation among individuals, we found no evidence for conservation of gut network structures within families. The gut network structures were not more similar within families (twins and their mothers; intrafamily) compared to other families (other twins and mothers; inter-family) (p-value = 0.547, Fig 3C). In addition to the gut, skin microbiome network structure was conserved within individuals (p-value < 0.001, Fig 3D). This distribution was similar when separated by anatomical sites. Most sites were statistically significantly more conserved within individuals (S12 Fig).
As an additional validation measure, we evaluated the tolerance of these findings to inaccuracies in the underlying network. As described above, our model is not perfect and there is likely to be noise from false positive and false negative predictions. We found that additional random noise, both by creating a fully connected graph or randomly reducing the number of edges to 60% of the original, changed the statistical significance values (p-values) of our findings but not by enough to change whether they were statistically significant (p-value < 0.05). Therefore the comparisons between groups are resilient to potential noise resulting from model false positive and false negative predictions (S13 Fig).
Extensive work has illustrated differences in diversity and composition of the healthy human skin microbiome between anatomical sites, including bacteria, virus, and fungal communities [13, 42, 58]. These communities vary by degree of skin moisture, oil, and environmental exposure; features which were defined in the original publication [13]. As viruses are known to influence microbial diversity and community composition, we hypothesized that these differences would still be evident after integrating the bacterial and viral datasets and evaluating their microbe-virus network structure between anatomical sites. To test this, we evaluated the changes in network structure between anatomical sites within the skin dataset. The anatomical sites and their features (e.g. moisture & occlusion) were defined in the previous publication through consultation with dermatologists and reference to previous literature [13].
The average centrality of each sample was quantified using the weighted eigenvector centrality metric. Intermittently moist skin sites (dynamic sites that fluctuate between being moist and dry) were significantly more connected than the moist and sebaceous environments (p-value < 0.001, Fig 4A). Also, skin sites that were occluded from the environment were less connected than those that were constantly exposed to the environment or only intermittently occluded (p-value < 0.001, Fig 4B). We also confirmed that addition of noise to the underlying network, as described above, altered the values of statistical significance (p-values) but not by enough to change whether they were statistically significant (S14 Fig).
To supplement this analysis, we compared the network signatures using the centrality dissimilarity approach described above. The dissimilarity between samples was a function of shared relationships, degree of centrality, and bacteria/phage abundance. When using this supplementary approach, we found that network structures significantly clustered by moisture, sebaceous, and intermittently moist status (Fig 4C and 4E). Occluded sites were significantly different from exposed and intermittently occluded sites, but there was no difference between exposed and intermittently occluded sites (Fig 4D and 4F). These findings provide further support that skin microbiome network structure differs significantly between skin sites.
Foundational work has provided a baseline understanding of the human microbiome by characterizing bacterial and viral diversity across the human body [13, 14, 41–43, 60]. Here we integrated the bacterial and viral sequence sets to offer an initial understanding of how phage-bacteria networks differ throughout the human body, so as to provide a baseline for future studies of how and why microbiome networks differ in disease states. We implemented a network-based analytical model to evaluate the basic properties of the human microbiome through bacteria and phage relationships, instead of membership or diversity alone. This approach enabled the application of network theory to provide a new perspective while analyzing bacterial and viral communities simultaneously. We utilized metrics of connectivity to model the extent to which communities of bacteria and phages interact through mechanisms such as horizontal gene transfer, modulated bacterial gene expression, and alterations in abundance.
Just as gut microbiome and virome composition and diversity are conserved in individuals [13, 41, 42, 59], gut and skin microbiome network structures were conserved within individuals over time. Gut network structure was not conserved among family members. These findings suggested that the community properties inferred from microbiome interaction network structures, such as robustness (meaning a more highly connected network is more “robust” to network degradation because a randomly removed bacteria or phage node is less likely to divide or disintegrate [25, 57] the overall network), the potential for horizontal gene transfer between members, and co-evolution of populations, were person-specific. These properties may be impacted by personal factors ranging from the body’s immune system to external environmental conditions, such as climate and diet.
We observed evidence supporting the ability of environmental conditions to shape gut and skin microbiome interaction network structure by observing that diet and skin location were associated with altered network structures. We observed evidence that diet was sufficient to alter gut microbiome network connectivity, although this needs to be interpreted cautiously as a case observation, due to the small sample size. Although the available sample size was small, our findings provide some preliminary evidence that high-fat diets are less connected than low-fat diets and that high-fat diets may therefore lead to less robust communities with a decreased ability for microbes to directly influence one another. We supported this finding with the observation that obesity may have been associated with decreased network connectivity. Together these findings suggest the food we eat may not only impact which microbes colonize our guts, but may also impact their interactions with infecting phages. Further work will be required to characterize these relationships with a larger cohort.
In addition to diet, the skin environment also influenced the microbiome interaction network structure. Network structure differed between environmentally exposed and occluded skin sites. The sites under greater environmental fluctuation and exposure (the exposed and intermittently exposed sites) were more connected and therefore were predicted to have a higher resilience against network degradation when random nodes are removed from the network. Likewise, intermittently moist sites demonstrated higher connectedness than the moist and sebaceous sites. These findings agree with previous work that has shown that bacterial community networks differ by skin environment types [51]. Together these data suggested that body sites under greater degrees of fluctuation harbored more highly connected microbiomes that are potentially more robust to network disruption by extinction events. This points to a link between microbiome and environmental robustness toward network homeostasis and warrants further investigation.
While these findings take us an important step closer to understanding the microbiome through interspecies relationships, there are caveats and considerations to our findings. First, as with most classification models, the infection classification model developed and applied is only as good as its training set—in this case, the collection of experimentally-verified positive and negative infection data. Large-scale experimental screens for phage and bacteria infectious interactions that report high-confidence negative interactions (i.e., no infection) are desperately needed, as they would provide more robust model training and improved model performance. Furthermore, just as we have improved on previous modeling efforts, we expect that new and creative scoring metrics will improve future performance. Other creative and high performing models are currently being developed and the applications of these models to community network creation will continue to move this field forward [38–40].
Second, although our analyses utilized the best datasets currently available for our study, this work was done retrospectively and relied on existing data up to seven years old. These archived datasets were limited by the technology and costs of the time. For example, the diet and twin studies, relied on multiple displacement amplification (MDA) in their library preparations–an approach used to overcome the large nucleic acids requirements typical of older sequencing library generation protocols. It is now known that MDA results in biases in microbial community composition [61], as well as toward ssDNA viral genomes [62, 63], thus rendering the resulting microbial and viral metagenomes largely non-quantitative. Future work that employs larger sequence datasets and that avoids the use of bias-inducing amplification steps will build on and validate our findings, as well as inform the design and interpretation of further studies.
Although our models demonstrated satisfactory accuracy and overall performance, it was important to interpret our findings under the realization that our model was not perfect. This caveat is not new to the microbiome field, with a notable example being the use of 16S rRNA sequencing using the V4 variable region [53]. Use of the V4 variable region excluded detection of major skin bacterial members, meaning that the findings were not able to completely describe the underlying biological environment. Despite this caveat, skin microbiome studies provided valuable biological insights by focusing on the community differences between groups (e.g. disease and healthy) which were both analyzed the same way. Similarly, here we focused our conclusions on the differences between the groups which were all treated the same, so that we can minimize our dependence on a perfect predictive model. We also provided explicit evidence that the introduction of noise equally to the compared groups did not significantly impact our findings.
Third, the networks in this study were built using operational genomic units (OGUs), which represented groups of highly similar bacteria or phage genomes or clustered genome fragments. Similar clustering definition and validation methods, both computational and experimental, have been implemented in other metagenomic sequencing studies, as well [37, 64–66]. These approaches could offer yet another level of sophistication to our network-based analyses. While this operationally defined clustering approach allows us to study whole community networks, our ability to make conclusions about interactions among specific phage or bacterial species or populations is inherently limited, compared to more focused, culture-based studies such as the work by Malki et al [54]. Future work must address this limitation, e.g., through improved binning methods and deeper metagenomic shotgun sequencing, but most importantly through an improved conceptual framing of what defines ecologically and evolutionarily cohesive units for both phage and bacteria [67]. Defining operational genomic units and their taxonomic underpinnings (e.g., whether OGU clusters represent genera or species) is an active area of work critical to the utility of this approach. As a first step, phylogenomic analyses have been performed to cluster cyanophage isolate genomes into informative groups using shared gene content, average nucleotide identity of shared genes, and pairwise differences between genomes [68]. Such population-genetic assessment of phage evolution, coupled with the ecological implications of genome heterogeneity, will inform how to define nodes in future iterations of the ecological network developed here. Even though we are hesitant to speculate on phage host ranges at low taxonomic levels in our dataset, the data does agree with previous reports of instances of broad phage host range [54, 69]. Additionally, visualization of our dataset interactions using the heat map approach previously used in other host range studies, suggests a trend toward modular and nested tropism, but we do not have the strain-level resolution that powered those previous experimental studies.
Finally, it is important to note that our model was built using available full genomes with known interactions, while the experimental datasets resulted in OGUs created from metagenomic shotgun sequence sets, as described above. While this is an informative approach given available data, it is not ideal. We envision future work focusing on training models using metagenomic shotgun sample sets from “mock communities” of bacteria and phages with experimentally validated infectious relationships. This would also be more informative than relying on simulated metagenomic sample sets, whose use would result in models built on simulations and more assumptions instead of empirical data. Together this way the training set can be subjected to the same pre-processing, contig assembly, and OGU binning processes as the experimental data. Furthermore, exciting advances in long read sequencing platforms such as the Oxford Nanopore MinIon system will provide more accurate genomic scaffolds than de novo assembled contigs, allowing for more accurate training and predictions of our models. As discussed above, it is because our current model is susceptible to this noise that we focus our conclusions on comparisons between experimental groups that were both treated the same. This is also why it was important for us to evaluate the susceptibility of our results to noise caused by the less-than-perfect prediction model.
Together our work takes an initial step towards defining bacteria-virus interaction profiles as a characteristic of human-associated microbial communities. This approach revealed the impacts that different human environments (e.g., the skin and gut) can have on microbiome connectivity. By focusing on relationships between bacterial and viral communities, they are studied as the interacting cohorts they are, rather than as independent entities. While our developed bacteria-phage interaction framework is a novel conceptual advance, the microbiome also consists of archaea and small eukaryotes, including fungi and Demodex mites [1, 70]—all of which can interact with human immune cells and other non-microbial community members [71]. Future work will build from our approach and include these additional community members and their diverse interactions and relationships (e.g., beyond phage-bacteria). This will result in a more robust network and a more holistic understanding of the evolutionary and ecological processes that drive the assembly and function of the human-associated microbiome.
A reproducible version of this manuscript written in R markdown and all of the code used to obtain and process the sequencing data is available at the following GitHub repository: https://github.com/SchlossLab/Hannigan_ConjunctisViribus_ploscompbio_2018.
Raw sequencing data and associated metadata were acquired from the NCBI sequence read archive (SRA). Supplementary metadata were acquired from the same SRA repositories and their associated manuscripts. The gut virome diet study (SRA: SRP002424), twin virome studies (SRA: SRP002523; SRP000319), and skin virome study (SRA: SRP049645) were downloaded as .sra files. For clarity, the sample sizes used for each study subset were described with the data in the results section. Sequencing files were converted to fastq format using the fastq-dump tool of the NCBI SRA Toolkit (v2.2.0). Sequences were quality trimmed using the Fastx toolkit (v0.0.14) to exclude bases with quality scores below 33 and shorter than 75 bp [72]. Paired end reads were filtered to exclude sequences missing their corresponding pair using the get_trimmed_pairs.py script available in the source code.
Contigs were assembled using the Megahit assembly program (v1.0.6) [73]. A minimum contig length of 1 kb was used. Iterative k-mer stepping began at a minimum length of 21 and progressed by 20 until 101. All other default parameters were used.
Contig simulations were performed by randomly extracting a string of genomic nucleotides that represented a defined percent length of that genome. This was accomplished using our RandomContigGenerator.pl, which was published in the associated GitHub repository.
Contigs were concatenated into two master files prior to alignment, one for bacterial contigs and one for phage contigs. Sample sequences were aligned to phage or bacterial contigs using the Bowtie2 global aligner (v2.2.1) [74]. We defined a mismatch threshold of 1 bp and seed length of 25 bp. Sequence abundance was calculated from the Bowtie2 output using the calculate_abundance_from_sam.pl script available in the source code.
Contigs often represent large fragments of genomes. In order to reduce redundancy and the resulting artificially inflated genomic richness within our dataset, it was important to bin contigs into operational units based on their similarity. This approach is conceptually similar to the clustering of related 16S rRNA sequences into operational taxonomic units (OTUs), although here we are clustering contigs into operational genomic units (OGUs) [60].
Contigs were clustered using the CONCOCT algorithm (v0.4.0) [75]. Because of our large dataset and limits in computational efficiency, we randomly subsampled the dataset to include 25% of all samples, and used these to inform contig abundance within the CONCOCT algorithm. CONCOCT was used with a maximum of 500 clusters, a k-mer length of four, a length threshold of 1 kb, 25 iterations, and exclusion of the total coverage variable.
OGU abundance (AO) was obtained as the sum of the abundance of each contig (Aj) associated with that OGU. The abundance values were length corrected such that:
A O = 10 7 ∑ j = 1 k A j ∑ j = 1 k L j
Where L is the length of each contig j within the OGU.
To confirm a lack of phage sequences in the bacterial OGU dataset, we performed blast nucleotide alignment of the bacterial OGU representative sequences using an e-value < 10−25, which was stricter than the 10−10 threshold used in the random forest model below, against all of the phage reference genomes available in the EMBL database. We used a stricter threshold because we know there are genomic similarities between bacteria and phage OGUs from the interactive model, but we were interested in contigs with high enough similarity to references that they may indeed be from phages. We also performed the converse analysis of aligning phage OGU representative sequences to EMBL bacterial reference genomes. We ran both the phage and bacteria OGU representative sequences through the Virsorter program (1.0.3) to identify phages (all default parameters were used), using only those in the high confidence identification category “class 1” [44]. Finally, we filtered out phage OGUs that had bacterial elements as described above, but also lacked known phage elements by using the tblastx algorithm and a maximum e-value of 10−25.
Open reading frames (ORFs) were identified using the Prodigal program (V2.6.2) with the meta mode parameter and default settings [76].
The classification model for predicting interactions was built using experimentally validated bacteria-phage infections or validated lack of infections from six studies [36, 45–49]. No further reference databases were used in our alignment procedures. Associated reference genomes were downloaded from the European Bioinformatics Institute (see details in source code). The model was created based on the four metrics listed below.
The four scores were used as parameters in a random forest model to classify bacteria and bacteriophage pairs as either having infectious interactions or not. The classification model was built using the Caret R package (v6.0.73) [77]. The model was trained using five-fold cross validation with ten repeats, and the median model performance was evaluated by training the model on 80% of the dataset and testing performance on the remaining 20%. Pairs without scores were classified as not interacting. The model was optimized using the ROC value. The resulting model performance was plotted using the plotROC R package.
The performance of our model for identifying diverse infectious relationships between bacteria and phages, beyond those that were included in the model creation step, were validated using additional bacterial and phage reference genomes, which could be linked by the records of which phage strains were isolated on which bacteria under laboratory conditions. Viral and bacterial reference genomes were downloaded from the GenBank repository on February 19, 2018 using the viral location ftp://ftp.ncbi.nih.gov/refseq/release/viral/ and the bacterial location ftp://ftp.ncbi.nih.gov/refseq/release/bacteria/. This resulted in the use of 539 complete phages reference genomes (with identified hosts) and 3,469 bacterial reference genomes. We used the same prediction model to predict which phages were infecting which hosts, so as to confirm that the model was capable of identifying interactions in a more diverse dataset. Bacteria interactions were identified at the species level. The random contig iteration analysis was performed using a subset of bacterial reference genomes, for computational performance reasons. Only single representative genomes for each species were used.
The bacteria and phage operational genomic units (OGUs) were scored using the same approach as outlined above. The infectious pairings between bacteria and phage OGUs were classified using the random forest model described above. The predicted infectious pairings and all associated metadata were used to populate a graph database using Neo4j graph database software (v2.3.1) [81]. This network was used for downstream community analysis. Tolerance to false negative and false positive noise within the networks was assessed by randomly removing a defined fraction of network edges before re-running the downstream analysis work flows. This was accomplished using functionality within the igraph R package (v1.0.1) [82].
We quantified the centrality of graph vertices using three different metrics, each of which provided different information graph structure. When calculating these values, let G(V, E) be an undirected, unweighted graph with |V| = n nodes and |E| = m edges. Also, let A be its corresponding adjacency matrix with entries aij = 1 if nodes Vi and Vj are connected via an edge, and aij = 0 otherwise.
Briefly, the closeness centrality of node Vi is calculated taking the inverse of the average length of the shortest paths (d) between nodes Vi and all the other nodes Vj. Mathematically, the closeness centrality of node Vi is given as:
C C ( V i ) = ( ∑ j = 1 n d ( V i , V j ) ) - 1
The distance between nodes (d) was calculated as the shortest number of edges required to be traversed to move from one node to another.
Intuitively, the degree centrality of node Vi is defined as the number of edges that are incident to that node:
C D ( V i ) = ∑ j = 1 n a i j
where aij is the ijth entry in the adjacency matrix A.
The eigenvector centrality of node Vi is defined as the ith value in the first eigenvector of the associated adjacency matrix A. Conceptually, this function results in a centrality value that reflects the connections of the vertex, as well as the centrality of its neighboring vertices.
The centralization metric was used to assess the average centrality of each sample graph G. Centralization was calculated by taking the sum of each vertex Vi’s centrality from the graph maximum centrality Cw, such that:
C ( G ) = ∑ i = 1 n C w - c ( V i ) T
The values were corrected for uneven graph sizes by dividing the centralization score by the maximum theoretical centralization (T) for a graph with the same number of vertices.
Degree and closeness centrality were calculated using the associated functions within the igraph R package (v1.0.1) [82].
We assessed similarity between graphs by evaluating the shared centrality of their vertices, as has been done previously. More specifically, we calculated the dissimilarity between graphs Gi and Gj using the Bray-Curtis dissimilarity metric and eigenvector centrality values such that:
B ( G i , G j ) = 1 - 2 C i j C i + C j
Where Cij is the sum of the lesser centrality values for those vertices shared between graphs, and Ci and Cj are the total number of vertices found in each graph. This allows us to calculate the dissimilarity between graphs based on the shared centrality values between the two graphs.
Differences in intrapersonal and interpersonal network structure diversity, based on multivariate data, were calculated using an analysis of similarity (ANOSIM). Statistical significance of univariate Eigenvector centrality differences were calculated using a paired Wilcoxon test.
Statistical significance of differences in univariate eigenvector centrality measurements of skin virome-microbiome networks were calculated using a pairwise Wilcoxon test, corrected for multiple hypothesis tests using the Holm correction method. Multivariate eigenvector centrality was measured as the mean differences between cluster centroids, with statistical significance measured using an ANOVA and post hoc Tukey test.
|
10.1371/journal.pgen.1000889 | Initial Genomics of the Human Nucleolus | We report for the first time the genomics of a nuclear compartment of the eukaryotic cell. 454 sequencing and microarray analysis revealed the pattern of nucleolus-associated chromatin domains (NADs) in the linear human genome and identified different gene families and certain satellite repeats as the major building blocks of NADs, which constitute about 4% of the genome. Bioinformatic evaluation showed that NAD–localized genes take part in specific biological processes, like the response to other organisms, odor perception, and tissue development. 3D FISH and immunofluorescence experiments illustrated the spatial distribution of NAD–specific chromatin within interphase nuclei and its alteration upon transcriptional changes. Altogether, our findings describe the nature of DNA sequences associated with the human nucleolus and provide insights into the function of the nucleolus in genome organization and establishment of nuclear architecture.
| It is becoming increasingly clear that the nuclear organization and location of genes in metazoan organisms is not random. Functionally related genes are often found next to each other in the linear genome, and distant DNA elements or DNA regions residing on different chromosomes may reside in specific nuclear compartments. The largest nuclear compartment is the nucleolus with its shell of perinucleolar DNA. The nature of the nucleolus-associated DNA, the targeting mechanism, and the cellular function of this subset of genomic DNA are not known. In the present study we report for the first time the high-resolution analysis of a nuclear compartment by sequencing, microarray analysis, and single-cell analysis. We have characterized the nucleolus-associated DNA on sequence level and by 3D microscopy and have determined common elements and the molecular function of this compartment.
| The largest and densest nuclear compartment is the nucleolus with its shell of perinucleolar DNA. The nucleolus is a unique object to study genome activity, since all three RNA polymerases are involved in the highly dynamic and tightly regulated ribosome biogenesis process, which is its main function. High proliferation activity of tumour cells coincides with high ribosome biogenesis activity thus exposing the nucleolus as a promising target in cancer therapy [1]. In addition, cell-type and function-dependent nucleolar localisation of tumour suppressor proteins, such as p53, MDM2 or p14ARF indicates the role of the nucleolus in carcinogenesis [2]–[5]. A number of other biological processes (e.g. senescence, RNA modification, cell-cycle control and stress sensing) are also regulated in the nucleolus and connect it to several functional networks of the cell [2]–[7]. Furthermore, chromatin motion is constrained at nucleoli or nuclear periphery, and disruption of nucleoli increases motility of chromatin domains, indicating the role of the nucleolus in higher-order chromatin arrangement [8]. The nucleolus can therefore be considered as a well-suited model system to investigate functional consequences of genome organisation. It is less well known, however, that alteration in the nucleolus might be linked to multiple forms of human disease, including viral infections. The interaction between viruses and the nucleolus is a pan-virus phenomenon, which is exhibited by DNA viruses, retroviruses and RNA viruses [9],[10]. Moreover, multiple genetic disorders have been mapped to genes that encode proteins located in nucleoli under specific conditions. These include Werner [11], fragile X [12],[13], Treacher Collins [14], Bloom [15], Rothmund–Thomson [16] and dyskeratosis congenita syndromes [17] and Diamond-Blackfan anemia [18].
Nucleoli are easily detectable under the microscope, however, despite the simple methods of nucleolus isolation, their molecular structure is largely unknown. The nucleolar proteome has been recently analysed by high-throughput mass-spectrometry [19], but the nucleic acid composition of nucleoli had not yet been determined. Therefore the aim of our investigations was to construct and characterize the first high-resolution, genome-wide map of NADs. Recent advances in sequencing and microarray technologies provided excellent platforms to subject nucleolus-associated DNA (naDNA) to critical scrutiny. The results presented here help to understand the mechanisms of nuclear information packaging by macromolecular assemblies and the functional compartmentalisation of the nucleus.
Because the nucleolar proteome was analysed in HeLa cells [19], our study started with the purification of nucleoli from this widely used model system (Figure 1A). Enrichment of the nucleolar transcription factor UBF and depletion of nuclear lamina proteins laminA/C from the nucleolar fraction was monitored by Western blot. Nucleolus-associated DNA was then isolated, and ribosomal DNA (rDNA) enrichment was measured by quantitative PCR (Figure S1). To analyse the genomic localisation of purified naDNA at low resolution, we performed 2D FISH experiments. Hybridisation of naDNA on human lymphocyte metaphase spreads shows that it appears predominantly on p-arms of acrocentric chromosomes, the location of the repetitive rDNA, and on centromeres of several chromosomes. The addition of the repetitive Cot1 competitor DNA suppresses binding of the naDNA probe to various chromosomal regions, but not to rDNA-containing nucleolar organiser regions (NORs). The result clearly demonstrates that rDNA, moreover pericentomeric and centromeric repetitive sequences are overrepresented in naDNA compared to other chromosomal regions (Figure 1B). Next, naDNA was analysed using Nimblegen whole genome microarrays at 6,270-bp median probe spacing resolution and compared to genomic DNA by performing two-colour hybridisation (aCGH). The aCGH data reinforced the results of the 2D FISH experiments: p-arm-adjacent regions of the acrocentric chromosomes and pericentromeric regions are enriched in naDNA. More interestingly, many other chromosomal regions are also present in the naDNA fraction (Figure 2A, Figure S2 and S3). For example, a large part of chromosome 19 associates with the nucleolus (Figure 3E). This finding explains the presence of chromosome 19 in central regions of the interphase nucleus [20], being close to the nucleoli. To elucidate NAD-specific sequence signatures in more detail, 454 sequencing was performed. In total 47,378,399 bases were sequenced in 218,030 reads with an average length of 217 bases/read. We used the complementary set of microarray and sequencing data to visualise the genome-wide localisation of NADs. Genome-wide studies are performed almost exclusively using one high-throughput strategy, which limits the quality of the detection. The combination of techniques compensates the inherent mistakes of the different methods. Our results clearly show that certain NADs are detectable only with one of these approaches (Figure S2 and Table S1). It is important to mention that the p-arms of the five acrocentric chromosomes, which contain rDNA and satellite repeats, are not represented in the hg18 genome build and, therefore, were not included in our analysis. In addition to the previously described pericentromeric locations, a significant number of the NADs (nine) localised in sub-telomeric regions. Altogether, 97 chromosomal regions that are associated with nucleoli were identified, encompassing about 4% (126,217,765 bp) of the genome. Our study detected the most frequent nucleolus-associated chromosome domains using stringent cut-off parameters for domain definition (Figure 2A, Figure S2 and S3, Table S1, and Materials and Methods).
After genome-wide NAD identification, sequence and chromatin features were compared to the whole genome and lamina-associated domains (LADs). LADs were recently determined by high-resolution mapping using DamID technology [21]. The size distribution (0.1–10 Mb) and median sequence length (749 kb) of NADs (Figure 2B) were similar to LADs (0.1–10 Mb, 553 kb) suggesting that the architectural units of chromosome organisation within the mammalian interphase nucleus are about 0.5–1 Mb in length.
One thousand thirty-seven genes have been identified within NAD sequences according to the RefSeq gene database, 729 of which were non-redundant (Table S2). Surprisingly, certain gene families were frequently associated with the nucleoli, even though the overall gene density in NADs is about 20% lower than in the whole genome. We observed a 4-fold enrichment of zinc-finger (ZNF) genes in NADs compared to the genome. Olfactory receptor (OR) and defensin genes were enriched in both NADs and LADs, but the enrichment was far greater in NADs (Figure 3A). Moreover, two of the six large clusters of immunoglobulin and T-cell receptor genes [22] overlap with NADs, and one other is juxtaposed to a NAD (Figure S3). The gene families mentioned above have two common features: their members are in large gene clusters, and they are expressed in a tissue-specific manner. This phenomenon suggests that these large chromosomal regions may change their sub-nuclear position with regard to their transcriptional activity. In addition, both immunoglobulin and OR genes exhibit monoallelic expression [23],[24]; therefore, nucleoli may be involved in this type of gene regulation. Though, this has to be tested for each individual gene in specific model systems. Besides the response to other organisms and odour perception, additional biological processes and molecular functions are specifically associated with genes localised in the vicinity of the nucleolus, including tissue development and embryo implantation. (Figure S4 and S5 and Table S3). Carcino-embryonic antigen cell adhesion molecule (CEACAM) genes and pregnancy-specific glycoprotein (PSG) gene clusters, whose protein products regulate implantation, were also found next to and within NADs, respectively. Additionally, a large number (119) of small nucleolar RNA (snoRNA) genes were identified within one NAD on chromosome 15. However, this association may be explained by the close proximity of this cluster to the rDNA repeats (distance of 5 Mb).
RNA genes located within NADs were characterized using the datasets of the ‘RepeatMasker’ and ‘RNA Genes’ databases of the Genome Browser. Both analyses show that 5S and tRNA genes, both of which are transcribed by RNA polymerase III, are specifically enriched in NADs but not in LADs. In contrast, other RNA genes are distributed with a similar frequency in NADs and the rest of the genome (Figure 3B). This finding proofs that RNA polymerase III-transcribed genes co-localise with nucleoli [25]–[27], which is the site of RNA polymerase I transcription. These observations suggest that spatial regulation may play a role in coordinated, well-tuned transcription of the RNA components of the protein translation machinery.
Analysis of the repetitive elements showed a more than 10-fold enrichment of satellite repeats in NADs and depletion of SINE - especially MIR–repeats (Figure 3C). We next performed a detailed quantitative analysis of all major satellite repeat subclasses located within NADs. (Figure S6). Our results demonstrate that the major building blocks of NADs are the alpha-, beta- and (GAATG)n/(CATTC)n-satellite repeats, whereas other types of satellite repeats (e.g. MSR1, D20S16, SATR2) were depleted. These data confirm and extend previous studies [28],[29] that describe nucleolar association of satellite repeats, but do not analyse them in detail. Taken together with the fact that D4Z4 macrosatellite repeats are located on the short arms of acrocentric chromosomes [30] and that ‘RepeatMasker’ does not contain information about low copy number repeats (e.g., segmental duplications or macrosatellites), we extended our investigations to such repetitive elements and showed that these genomic features are enriched in NADs (Figure S3 and Table S4). The presence of low-copy number repeats in NADs underlie the difficulties of alignment-based localisation of naDNA sequences within the genome: segmental duplications and major satellites will be mapped to more than one region [31],[32], thus the nucleolar association of chromosome regions containing such sequences has to be confirmed by neighbouring sequences or in 3D FISH experiments. Enrichment of satellites and segmental duplications in NADs may also explain the assignment of several domains to chromosome Y even though HeLa cells are derived from a female. The Y chromosome has been shown to co-localise with nucleoli in the interphase nucleus [29],[33], indicating that such low-copy number repeats are maybe involved in nucleolar targeting. The detailed map of nucleolus-associated chromosomal regions and genomic features enriched in NADs is shown in Figure 3E for chromosome 19. The complete set of data is shown in Figure S3 and Table S5.
In order to reveal specific chromatin patterns enriched within the nucleolus-associated chromatin domains, we used the genome-wide maps of histone modifications [34]–[36]. Multiple repressive histone marks were specifically enriched, whereas the active histone mark H3K4Me1 was significantly depleted in NADs. As mirrored by the enrichment of repressive histone marks, we observed the reduced global gene expression in NADs (Figure 3D and Table S6). These findings imply that NADs tend to form large inactive chromatin domains in the interphase nucleus. However, nucleolus-associated inactive chromatin differs markedly from lamina-associated inactive chromatin in the kind of repetitive elements and the gene-associated biological processes, suggesting that multiple domains of functionally distinct inactive chromatin exist within the nucleus. Furthermore, the presence of the highly expressed classes of 5S RNA and tRNA genes in nucleolus-associated chromatin indicates that the perinucleolar region is not exclusively transcriptionally silent.
We used 3D immuno-FISH to confirm whether NADs revealed by the high-throughput methods co-localise with nucleoli. Nucleo li were stained with an α-B23/nucleophosmin antibody, and we have chosen 11 genomic loci that were analysed by appropriate BAC clones. Target, negative and positive control regions were selected from different chromosomes (Table S7, Figure S7, and Materials and Methods). The pericentromeric Xq11.1 region and the 5S rDNA cluster at 1q42.13 served as positive controls [26],[37]. The combination of microarray and high-throughput sequencing analysis revealed a high-fidelity list of nucleolus-associated DNA as all of our selected NADs were more frequently associated with nucleoli of HeLa cells than the negative controls. To prove whether the nucleolar association of these chromosomal regions is a cell type specific feature or it is a general property in human cells, IMR90 embryonic lung fibroblasts were analysed. In contrast to HeLa, IMR90 cells possess diploid karyotype and they are not immortal. Except the 5S rDNA cluster on chromosome 1, all selected regions showed similar levels of nucleolar association in IMR90 and HeLa cells (Figure 4A and Figure S8), suggesting that the nucleolar targeting of certain chromosomal regions is a common feature in human cells.
We next addressed the function of transcription in DNA targeting to the nucleolus by monitoring nucleolus association of selected chromosomal domains upon transcriptional inhibition. We used α-amanitin to block transcription by RNA polymerases II and III, whereas the synthesis of the 47S rRNA precursor was repressed by the addition of actinomycin D. We found that the specific inhibition of any of the RNA polymerases results in spatial reorganization of the nucleolus-associated domains (Figure S9 and Table S7), which indicates that the nucleolus forms a functional unit together with the associated perinucleolar chromatin. However, the concomitant partial disruption of nucleolar structures [38] makes the interpretation of such experiments difficult.
In addition to localisation studies of single chromosomal regions, three typical features of the perinucleolar chromatin were visualised. To this end, five-colour immunofluorescence experiments were performed, which allowed direct comparison of the signal distributions of centromere, H3K27Me3 and active RNA polymerase II localisations in the same cell. RNA polymerase II transcription was depleted around nucleoli, furthermore the frequent association of H3K27Me3 and centromere signals with nucleoli reinforced the results of the bioinformatic analysis of NADs. Both HeLa and IMR90 cells showed similar localisation of these nuclear marks and the observed punctuated patterns suggest that functionally distinct chromatin domains co-exist around nucleoli (Figure 4B and 4C and Figure S10).
We report here the mapping and characterization of nucleolus-associated chromatin domains in the human genome. Bioinformatics and statistical analyses reveal that the main building blocks of NADs are certain types of satellite repeats, tRNA and 5S RNA genes and members of the ZNF, OR, defensin and immunoglobulin gene families. Thus, our data suggest that certain type of satellite repeat sequences play an important role in establishing of NADs. Indeed, the internal scaffold of the nucleolus, the rDNA repeats were analysed only by qPCR (Figure S1), but not in our high-throughput studies for several reasons: i) they are not represented in the hg18 genome build, ii) repetitive sequences are not printed on microarrays, iii) the number of 454 sequencing reads depends on the GC content, which is very variable throughout the rDNA repeat (Figure S11).
The findings of a recent publication indicate that centromeric nucleoprotein complexes may be targeted to the nucleolus via an alpha-satellite RNA-mediated mechanism [39], and address the importance of transcription in this process. These data suggest that transcription has a general regulatory role in maintaining the nuclear architecture around the nucleolus. The transcribed RNA may be bound by nucleolar RNA-binding proteins, which sequester NADs to the nucleolar periphery. On the other hand, our results imply that there is not a unique predictor sequence – in addition to certain satellite repeats, other elements e.g. tRNA genes, 5S RNA genes may be sufficient for the nucleolar targeting of individual chromatin domains. The aforementioned DNA elements, together with specific RNA molecules and scaffold proteins like UBF, may coordinate the (at least partial) self-assembly of the nucleolus with its shell. The principles of the assembly might be similar to the ones that were demonstrated recently for the pseudo-NORs [40],[41] and for the Cajal-body [42], where single DNA, protein or RNA scaffolds were able to nucleate the formation of nuclear compartments. Further experiments are required to uncover the molecular steps of transcription-dependent nucleolar targeting of different groups of NADs and to identify the players in this process. The dynamics of nucleolus association during cell cycle and cell differentiation will be addressed in future studies. The functional organisation of the nuclear architecture is studied intensively [43]–[46] and the identification of NADs in the present work provides a basis for the better understanding of the role of nucleoli in the spatial organisation of the human genome.
HeLa cervix carcinoma cells were cross-linked with 1% formaldehyde and nucleoli were isolated as described [47]. rDNA content of equal amounts of naDNA and genomic DNA was quantified in real-time PCR reactions.
Oligonucleotide sequences:
Hr132F: 5′CCTGCTGTTCTCTCGCGC,
Hr155P: 5′FAM-AGCGTCCCGACTCCCGGTGC-TAMRA,
Hr198R: 5′GGTCAGAGACCCGGACCC;
Hr9776F: 5′GCCACTTTTGGTAAGCAGAACTG,
Hr9802P: 5′FAM-CTGCGGGATGAACCGAACGCC-TAMRA,
Hr9840R: 5′CATCGGGCGCCTTAACC.
Numbers indicate rDNA (GenBank Acc. No U13369) position relative to the transcriptional start site. Two rDNA regions were measured in technical triplicates from two biological replicate experiments. UBF and laminA/C protein levels were monitored with the sc-9131 and sc-20681 antibodies (Santa Cruz Biotechnology), respectively.
naDNA was isolated and subjected to 454 sequencing (MWG-Biotech) and microarray analysis on HG18 CGH 385K WG Tiling v1.0 platform (Nimblegen). Genomic features of NADs were analysed using the UCSC Table Browser (http://genome.ucsc.edu/cgi-bin/hgTables) and chromatin features using the Ensembl Database (http://www.ensembl.org) and the GSE12889 NCBI GEO dataset. Genomic features were visualised using Galaxy (http://galaxy.psu.edu/) and the UCSC Genome Browser (http://genome.ucsc.edu/). All analyses were performed on the hg18 genome build. Biological processes and molecular functions associated with NAD-located genes were analysed by using FatiGO [48].
Array CGH, 454 sequencing and subsequent data analysis were performed as follows: naDNA samples from two biological replicate experiments were subjected to microarray analysis on HG18 CGH 385K WG Tiling v1.0 platform. Hybridisation and pre-processing of hybridisation signals were performed at Nimblegen. For each of the samples, regions of increased intensity measurements were considered to be relevant if their mean value was greater than the 85 percentile of the sample distribution at 0.1 Mb running window size. Only the intersection of relevant regions across the microarray replicas was considered as a NAD.
High-throughput sequencing was performed using the Roche GS FLX system. One of the aCGH analysed naDNA samples was taken as template for sequencing. 454 sequence reads were quality filtered and automatically assembled into contigs with the Newbler Assembler software at MWG-Biotech. Contigs were matched against the human genome using BLAT. Repeat masked sequences were used both for 454 data and genome data. For matching a 95% of sequence identity and coverage was requested and a maximum gap size of 3 was permitted. Of the mapped reads, 88% had unique hits. 454 data was widely spread on the genome. Only a few regions had higher intensity, mainly around centromeres. For domain detection, 454 data was first transformed into a binary (1/0) signal indicating presence/absence of mapped reads at chromosome positions defined by 100 nts length segments situated at a 1000 nts inter-spacing. A running mean algorithm was run on these data with a window size of 100 (which implies an actual chromosome window size of 0.1 Mb), to identify chromosomal regions with higher abundance of 454 sequencing hits (red bars in plots). 454 ‘Chip-Seq’ domains were selected as those areas with a running mean value above the 98% of the chromosome percentile. This arbitrary threshold fits well visual evaluation of 454 data as well as aCGH data. Finally, 454 regions were edited and border positions were curated manually. The significance of the 454-based NAD determinations was assessed empirically by comparing the number of reads in each of the detected NADs against the distribution of number of reads in 1000 randomly selected same-chromosome regions of the same size. The significance is then obtained as the quartile position of the NAD reads number in the random distribution.
454 and aCGH domains were merged in one single list of NADs. For merging, overlapping regions from both technologies were fused in one domain. Domain borders were defined following aCGH data unless the absence of array probes at merged borders suggested to use the 454 limits. Furthermore, adjacent regions separated by less than 0.1 Mb were joined to single domains.
Microarray data have been submitted to the ArrayExpress Database (http://www.ebi.ac.uk/microarray-as/ae/) under accession number E-MEXP-2403. 454 sequencing data have been submitted to the Sequence Read Archive (http://www.ncbi.nlm.nih.gov/Traces/sra/) under accession number SRA009887.3.
2D FISH experiments were performed on HeLa and human female lymphocyte metaphase spreads according to standard protocols. naDNA was labelled without amplification. NAD target and control BACs were selected as follows: RP11-434B14 (Xq11.1; ‘X cen’) and RP5-915N17 (1q42.13; ‘5S’) were used as positive controls. Perinucleolar localisations of the X chromosome and the large 5S rDNA cluster on chromosome 1 were reported previously [26],[37]. RP11-90G23 (8q21.2; ‘REXO1’) and RP11-173M10 (13q21.1; ‘7SK’, encompassing a 7SK RNA gene) were selected based on 454 sequencing data. We tested in the latter case if smaller 454 signals, which have not identified NADs could also be associated with nucleoli. RP11-44B13 (19q13.12; ‘27ZNF’) –selected based on our microarray data - marks a chromosomal fragment in FISH experiments where 27 KRAB-ZNF genes are located. The KRAB-ZNF gene cluster at 19q13.12 represents a SUV39H1 and CBX1 binding region. Our 3D FISH results reveal spatial features of this locus, which was formerly characterized at the level of chromatin domain organisation [49]. RP11-89H10 (3p12.3; ‘FRG2C’) and RP11-413F20 (10q26.3; ‘FRG2B’) were selected from combined aCGH/454 and aCGH results respectively. Both chromosomal regions contain D4Z4 major satellite repeats which may have nucleolar targeting potential. RP11-89O2 (3p14.1; ‘FRG2C ctrl’) and RP11-123G19 (10q24.1; ‘FRG2B ctrl’) served as negative controls for the latter two targets. RP11-81M8 (19p13.3; ‘REXO1’) covers a large 2 Mb chromosome fragment. This region contains the REXO1 gene thus having similarity at the primary sequence level to the REXO1L target and serves as its negative control. The negative control of the ZNF gene cluster (RP11-1137G4; 19p13.3-19p13.2; ‘ZNF557’) contains a single ZNF gene.
3D immuno-FISH experiments were performed as described [50]. In localisation experiments α-B23/nucleophosmin (Sigma, B0556), α-H3K27Me3 (Upstate, 07-449), α-active Pol II (Covance, MMS-129R), α-centromere (Antibodies Inc., 15–134) and different fluorescence dye-conjugated secondary antibodies, furthermore BAC clones RP11-90G23, RP11-173M10, RP11-44B13, RP11-89H10, RP11-413F20, RP11-81M8, RP5-915N17, RP11-1137G4, RP11-89O2, RP11-123G19 and RP11-434B14 were used on HeLa cervix carcinoma cells and IMR90 lung embryonic fibroblasts. HeLa cells were treated with 75 µg/ml or 300 µg/ml α-amanitin for 5 hours in order to inhibit RNA polymerase II or RNA polymerases II and III. RNA polymerase I mediated synthesis of the rRNA precursor was impaired by treatment of the cells with 50 ng/ml actinomycin D for 1 hour. Cells were fixed and 3D immuno-FISH experiments were performed. Confocal microscopy and image analysis was performed after 3D FISH experiments as follows: series of optical sections through 3D-preserved nuclei were collected using a Leica TCS SP5 confocal system equipped with a Plan Apo 63×/1.4 NA oil immersion objective and a diode laser (excitation wave length 405 nm) for DAPI, an argon laser (488 nm) for FITC and Alexa 488, a DPSS laser (561 nm) for Cy3, a HeNe laser (594 nm) for Texas Red and a HeNe laser (633 nm) for Cy5. For each optical section, signals in different channels were collected sequentially. Stacks of 8-bit gray-scale images were obtained with z-step of 200 nm and pixel sizes 30–100 nm depending on experiment. The axial chromatic shift was corrected and corresponding RGB-stacks, montages and maximum intensity projections were created using published ImageJ plugins [51]. Positions of FISH signals were assessed by visual inspection of RGB stacks using the ImageJ program.
|
10.1371/journal.pbio.1001379 | Evolutionarily Repurposed Networks Reveal the Well-Known Antifungal Drug Thiabendazole to Be a Novel Vascular Disrupting Agent | Studies in diverse organisms have revealed a surprising depth to the evolutionary conservation of genetic modules. For example, a systematic analysis of such conserved modules has recently shown that genes in yeast that maintain cell walls have been repurposed in vertebrates to regulate vein and artery growth. We reasoned that by analyzing this particular module, we might identify small molecules targeting the yeast pathway that also act as angiogenesis inhibitors suitable for chemotherapy. This insight led to the finding that thiabendazole, an orally available antifungal drug in clinical use for 40 years, also potently inhibits angiogenesis in animal models and in human cells. Moreover, in vivo time-lapse imaging revealed that thiabendazole reversibly disassembles newly established blood vessels, marking it as vascular disrupting agent (VDA) and thus as a potential complementary therapeutic for use in combination with current anti-angiogenic therapies. Importantly, we also show that thiabendazole slows tumor growth and decreases vascular density in preclinical fibrosarcoma xenografts. Thus, an exploration of the evolutionary repurposing of gene networks has led directly to the identification of a potential new therapeutic application for an inexpensive drug that is already approved for clinical use in humans.
| Yeast cells and vertebrate blood vessels would not seem to have much in common. However, we have discovered that during the course of evolution, a group of proteins whose function in yeast is to maintain cell walls has found an alternative use in vertebrates regulating angiogenesis. This remarkable repurposing of the proteins during evolution led us to hypothesize that, despite the different functions of the proteins in humans compared to yeast, drugs that modulated the yeast pathway might also modulate angiogenesis in humans and in animal models. One compound seemed a particularly promising candidate for this sort of approach: thiabendazole (TBZ), which has been in clinical use as a systemic antifungal and deworming treatment for 40 years. Gratifyingly, our study shows that TBZ is indeed able to act as a vascular disrupting agent and an angiogenesis inhibitor. Notably, TBZ also slowed tumor growth and decreased vascular density in human tumors grafted into mice. TBZ’s historical safety data and low cost make it an outstanding candidate for translation to clinical use as a complement to current anti-angiogenic strategies for the treatment of cancer. Our work demonstrates how model organisms from distant branches of the evolutionary tree can be exploited to arrive at a promising new drug.
| Systems biology has shown great promise in providing a better understanding of human disease and in identifying new disease targets. These methods typically leave off once the target is identified, and further research transitions to established paradigms for drug discovery. However, the vast majority of molecular pathways that function in human disease are not specific to humans, but rather are conserved across vertebrates and even to very distantly related organisms. The remarkable growth of genetic data from tractable model organisms implies that most genetic modules relevant to human biology are currently best characterized in non-human species. Such evolutionary conservation, even when the homology of the systems to the human case is distant or perhaps non-obvious, should enable new drug design strategies.
Clearly, identification of deeply conserved gene networks in distant organisms opens the possibility of pursuing drug discovery in those organisms. While traditional methods of drug discovery focus on gene-by-gene rather than network- or system-level similarities, we suggest that phenologs—gene networks that while orthologous may nonetheless produce different phenotypes due to altered usage or organismal contexts [1]—can provide a basis not just for screening against a single protein, but also for simultaneous drug discovery efforts against multiple targets in parallel. Given the key roles that model organisms already play in biomedical research, identification of such deep homologies should also allow us to better leverage the particular strengths of the wide variety of animal models in order to rapidly test candidate drugs found from such an approach.
We recently developed a method for systematically discovering phenologs, and this approach identified a conserved module that is relevant to lovastatin sensitivity in yeast and is also responsible for regulating angiogenesis in vertebrates [1]. Angiogenesis, the process of forming new blood vessels, plays an essential role in development, reproduction, and tissue repair [2]. Because the vascular network supplies oxygen and nutrients to cancer cells as well as to normal cells, angiogenesis also governs the growth of many types of tumors, and is central to malignancy [2]–[5]. The vasculature is thus considered to be a major therapeutic target for drug development. Some cancers, such as the most common and deadly brain neoplasm, glioblastoma multiformae [6], are heavily vascularized, but have not responded to current angiogenesis inhibitors [7],[8]. Because new agents that target the vasculature would increase our arsenal for battling cancers resistant to current therapies [2]–[5], there is a clear clinical need for novel approaches to their identification.
Here, we have exploited data mining of genetic interactions in yeast, in vivo time-lapse imaging in a non-mammalian vertebrate, loss-of-function analysis in cultured human cells, and preclinical xenografts in mice to identify and characterize a novel anti-angiogenic small molecule (Figure 1). Excitingly, this compound is already FDA approved for use in treating certain infections in humans, making it an excellent candidate for rapid translation to the clinic. This research exemplifies a general strategy for exploiting deeply conserved genetic modules for drug screening, characterized by screens focused not on single genes but rather on conserved genetic modules and by a strong reliance on tractable model organisms in order to speed the discovery of therapeutics.
The remarkable conservation of a genetic module that controls lovastatin sensitivity in yeast and angiogenesis in vertebrates ([1]; Figures 2AB and S1; Table S1) led us to test the possibility that small-molecule inhibitors modulating the yeast pathway might also act as angiogenesis inhibitors. Indeed, preliminary evidence suggests that lovastatin itself at least partly inhibits angiogenesis [9],[10] and may even reduce the incidence of melanoma [11],[12]. We therefore devised a strategy to exploit the evolutionary repurposing of this module in order to direct our search (Figure 1). Specifically, we desired to identify compounds in a manner that did not require their mechanism of action or even their biochemical target to match that of lovastatin; we thus employed a genetic strategy in yeast in order to select compounds that genetically interacted with this module. By computationally mining available large-scale chemical sensitivity datasets [13], candidate compounds were prioritized based upon their measured synthetic genetic interactions with yeast genes, using clustering algorithms to identify those compounds with genetic interaction profiles most similar to that of lovastatin (Figure 2C; Table S2; Figure S2). Notably, four out of eight prioritized chemicals were already known to modulate angiogenesis, indicating strong enrichment for angiogenesis effectors (Table S2).
One compound—thiabendazole (TBZ; 4-(1H-1,3-benzodiazol-2-yl)-1,3-thiazole)—stood out because it has already been approved by the U.S. Food and Drug Administration (FDA) for systemic oral use in humans (as an anti-fungal and anti-helminthic treatment). TBZ was initially marketed by Merck as Mintezol, and is now off-patent and issued as a generic under the trade names Apl-Luster, Mertect, Mycozol, Tecto, Tresaderm, and Arbotect. TBZ has been used by humans since its FDA approval in 1967, so its safety has been well-established. In animals, TBZ has no carcinogenic effects in either short- or long-term studies at doses up to 15 times the usual human dose [14],[15]. Moreover, TBZ does not appear to affect fertility in mice or rats, and it is not a mutagen in standard in vitro microbial mutagen tests, micronucleus tests, or host-mediated assays in vivo [14],[15]. Thus, TBZ was an outstanding candidate for further study.
We first tested the effect of TBZ on the expression of vascular-specific genes in developing Xenopus embryos, which provide a rapid, tractable, and accurate model for in vivo studies of angiogenesis [16]–[19]. Using in situ hybridization to either the apelin-receptor (aplnr) or the vascular ETS factor (erg), we found that TBZ treatment severely impaired angiogenesis (Figure 3A–D). This result was confirmed in living embryos in which vasculature was visualized by expression of GFP under control of a kdr enhancer/promoter fragment (Figure 3E–F) [19]. Notably, TBZ also inhibited angiogenesis in a dose-dependent manner in cultured human endothelial cells (HUVECs), suggesting that the activity of TBZ is conserved in vertebrates (Figure 4). We then sought to position the site of TBZ action relative to that of VEGF, as this growth factor is central to both normal and pathogenic angiogenesis [3],[4]. In frog embryos, ectopic VEGF potently induces ectopic angiogenesis [16], and this effect was blocked by TBZ, suggesting that the drug acts downstream of this key regulatory node (Figure S3).
These data implicate TBZ as an effective inhibitor of angiogenesis. Importantly, we observed angiogenesis inhibition in both human cells in vitro and in Xenopus embryos in vivo at a concentration of 100–250 µM. This dose corresponds to 20–50 mg/kg (Figures 3 and 4), which is notable because the oral LD50 of MINTEZOL is 1.3–3.6 g/kg, 3.1 g/kg, and 3.8 g/kg in the mouse, rat, and rabbit, respectively, and the human approved recommended maximum daily dose is 3 grams, corresponding to 50 mg/kg for 60 kg patients. Finally, we note that the overall morphology and patterning of TBZ-treated Xenopus embryos was grossly normal at the stages when the vasculature was severely disrupted (Figure S4). Consistent with this, TBZ has good safety data in humans and model animals at the doses for which we observe a specific inhibition of angiogenesis [14],[15].
We next asked if angiogenesis inhibition may be a general property of benzimidazoles. Examination of commercially available TBZ derivatives showed that this is not the case, with benzimidazole itself inactive at doses up to 1 mM and administration of other benzimidazoles causing diverse developmental defects but not angiogenesis inhibition (Figure S5). These findings are thus significant for demonstrating a high level of precision for this evolutionary approach to drug discovery.
We next sought to better understand the cellular basis for angiogenesis inhibition by TBZ. In the course of our studies, we noted an interesting feature of the vasculature in TBZ-treated embryos: disconnected and scattered arrays of cells in which vascular gene expression persisted (Figures 3B,D and S3). Hypothesizing that such morphological defects in the absence of changes to vascular gene expression may stem from direct impairment of vessel integrity, we tested the ability of TBZ to disrupt pre-existing vasculature by treatments at later stages, when blood vessels were already well formed and patent [20]. TBZ treatment elicited overt breakdown of established vasculature at these stages (Figure S6).
The ability of TBZ to disassemble extant blood vessels was especially significant because such an activity has recently drawn the attention of cancer biologists [4],[21],[22]. A new class of drugs called Vascular Disrupting Agents (VDAs) break down existing vascular structures, thereby disrupting blood flow, particularly within solid tumors [4],[21],[22]. No VDAs have as yet been approved for use in humans, although several such agents are therapeutically promising and are in phase II and III trials [4].
As a direct test of the vascular disrupting activity of TBZ, we performed time-lapse imaging of developing vasculature. Using Kdr-GFP transgenic embryos [19] and time-lapse confocal microscopy [23], we could effectively image developing vasculature in vivo for periods of up to 20 h. During this time, the growth of existing vessels and the sprouting of new vasculature could be easily followed (Figure S7; Movie S1). Treatment with TBZ completely prevented growth and sprouting of vessels, and moreover elicited a striking disintegration of established vessels after ∼90 min of exposure (Figures 5 and S8; Movie S2). Upon longer exposures, endothelial cells scattered and many underwent dramatic rounding (Figures 5A and S8; Movie S2).
These data demonstrate the efficacy of TBZ as a vascular disrupting agent. Previously defined VDAs can act either by targeting endothelial cells for selective cell death (e.g., ASA404 [24]) or by disrupting endothelial cell behaviors (e.g., combrestatin A4 [25]), and so we sought to distinguish between these two possible mechanisms for TBZ action. We noted that treatment with TBZ doses sufficient to severely perturb the vasculature elicited only modest increases in apoptosis in cultured HUVECs (Figure S9). Moreover, vascular gene expression in dispersed, rounded kdr-GFP+ endothelial cells in vivo reliably persisted for up to 17 h after TBZ treatment (Figures 3F, 5, and S8).
These data argue against a role for apoptosis in vascular disruption by TBZ. To test this idea more directly, we performed washout experiments. Compellingly, washout of the drug after overt TBZ-induced endothelial cell dispersal and rounding resulted in significant re-spreading of endothelial cells and re-formation of vessels in living Xenopus embryos assessed by time-lapse imaging (Figure 6; Movie S3). In several cases, widely separated kdr-GFP-positive endothelial cells reconnected into nascent vessels after washout of TBZ (Figure 6).
Finally, we found that treatment with TBZ significantly slowed endothelial cell migration in a scratch wound assay using cultured HUVECs (Figure 7AB). This quantitative in vitro assay with mammalian cells, combined with our in vivo data from Xenopus, demonstrate that TBZ disrupts established vasculature not by eliciting cell death but rather by perturbing endothelial cell behavior.
The effect of TBZ on endothelial cells is striking and rapid. Our in vivo imaging of the vasculature revealed that endothelial cells retract from one another and round up within 2 h of TBZ treatment (Figures 5 and 6). Moreover, we observed that this effect is reversible by washout within a similarly rapid time frame (Figure 6). The rapid time-frames observed here argue that TBZ may act at the level of the cytoskeleton to influence endothelial cell behavior.
We first considered that, while not an assumption of the phenolog approach (see above), TBZ may nonetheless impact the vasculature by the same mechanism as lovastatin. Lovastatin disrupts angiogenesis at least in part by perturbing the geranyl-geranylation of the RhoA GTPase, thereby abrogating its activity [10]. RhoA is a critical regulator of actin-based behaviors in all animal cells [26], and the loss of RhoA signaling in endothelial cells treated with lovastatin is directly linked to cytoskeletal changes and inhibition of angiogenesis [10]. Indeed, inhibition of angiogenesis by lovastatin can be overcome by addition of geranyl-geranyl pyrophosphate (GGPP; [9],[10]). We therefore used the HUVEC scratch-wound closure model to quantitatively assess the effects of GGPP addition on TBZ action. However, we found that addition of GGPP did not reverse the action of TBZ on HUVEC cell motility in this assay (Figure S10). Similarly, while TBZ has been observed to affect the activity of porcine heart mitochondria [27], we detected no differences in mitochondrial mass (measured by MitoTracker Green signal) or mitochondrial membrane potential (measured as the ratio of MitoTracker Red signal to Mitotracker Green signal) (unpublished data), thus ruling out this potential activity as being relevant.
We next considered the possibility that TBZ acted on the vasculature at the level of the microtubule (MT) cytoskeleton, because TBZ has been found to disrupt microtubule assembly and dynamics in a number of cell types (e.g., [28]–[31]), and because several currently-studied VDAs act as MT-disrupting agents [21],[32]. Curiously, TBZ had only a very slight effect on the gross organization of the MT cytoskeleton in HUVEC cells in culture (Figure S11A), but a quantitative analysis using mass-spectrometry revealed a significant reduction in the abundance of several tubulin proteins following treatment of HUVECs with TBZ (Figure S11B).
Many MT-targeting VDAs act via hyper-activation of Rho signaling [25],[33],[34], likely reflecting the key role of MT-binding RhoGEFs [35]. We reasoned, therefore, that TBZ may also act via increased Rho signaling, as the drug elicited several phenotypes known to be associated with dysregulated Rho signaling (e.g., cell rounding, re-distribution of actin filaments, and defects in cell motility; Figure 7). To test this model directly, we asked if disruption of Rho signaling might counteract the effects of TBZ. Indeed, pharmacological disruption of Rho kinase function using the small molecule Y27632 elicited a significant and dose-dependent rescue of the TBZ-induced HUVEC cell motility defect (Figure 7A,B). Together, these data suggest that vascular disruption by TBZ results from reduced tubulin levels and hyper-active Rho signaling.
It is hoped that VDAs may open new therapeutic avenues by complementing the action of currently used angiogenesis inhibitors (e.g., [4]). Moreover, the data above suggest that the mechanism of TBZ action distinguishes it from VDAs such as ASA404, which act by inducing endothelial cell apoptosis [24], but which failed to show efficacy in a recent Phase III clinical trial for treatment of lung cancer [36]. To begin to ask if TBZ may be useful in the arena of cancer therapy, we tested the ability of TBZ to slow the growth of solid vascularized tumors in a mammal. We therefore employed a mouse xenograft model typical of those proven valuable in indicating the effectiveness of anti-angiogenesis therapy [37],[38]. We found that TBZ treatment significantly slowed HT1080 human fibrosarcoma xenograft growth in athymic Cre nu/nu mice [39], as assessed by a time course of tumor size and also by final tumor mass (Figure 8).
Our in vivo data from Xenopus, as well as our human in vitro data, suggest that TBZ likely slows tumor growth by acting at the level of the vasculature (Figures 3 and 4). Consistent with this model, TBZ treatment did not alter the rate of proliferation in HT1080 cells when cultured in vitro but did significantly impair tumor microvessel density in xenografts (Figures 9 and S12). In addition, we noted that treatment with TBZ did not alter the levels of VEGF expressed or secreted by HT1080 cells, consistent with it acting downstream of VEGF in tumor xenografts (Figure S13), as it does in developing Xenopus embryos in vivo (Figure S3). Notably, we employed a TBZ dose of 50 mg/kg for these experiments, which is concordant with the FDA-approved maximum recommended daily dose of TBZ in humans, suggesting the possibility of chemotherapeutic use in humans.
In sum, an analysis of evolutionary repurposing of a genetic module shared from yeast to humans has led directly to the discovery that an orally available drug, thiabendazole, already FDA approved for clinical use in humans, also acts as an angiogenesis inhibitor and vascular disrupting agent. Moreover, these data establish TBZ as the only VDA currently approved for human use (albeit for a different purpose). Our data suggest that, even for antifungal or antihelmintic use, the possibility of side effects related to vascularization should be considered, for example, in patients with cardiovascular disease or to the fetus if administered to pregnant women, for whom TBZ has not been broadly tested. Significantly, while research on VDAs has largely centered on cancer therapy, their use may also provide new therapeutic avenues for non-malignant diseases, such as diabetic retinopathy and macular degeneration [40],[41]. With more than 40 years of human use, the low cost and generic availability of TBZ make it a compelling candidate for translation into the clinic as a VDA.
Finally, this research emphasizes the advantages of an evolutionary approach to drug discovery, in which the natural experimental strengths of various organisms can be exploited to accelerate our understanding of a conserved genetic module. Importantly, this approach proceeded from a gene module-based discovery strategy and proved effective even though the associated organismal phenotypes were entirely unrelated. Curiously, at least two other known antifungal drugs can also act as angiogenesis inhibitors. Itraconazole, an azole antifungal drug otherwise structurally unrelated to TBZ and acting via different mechanisms, was identified as an angiogenesis inhibitor via high-throughput screening [42]. This observation, in addition to the dual anti-fungal and anti-angiogenic properties of lovastatin [9],[10],[43], suggests additional interesting evolutionary connections between the processes of yeast cell wall metabolism and vertebrate angiogenesis. Such evolutionary connections further support the yeast cell wall-relevant activity of TBZ, from among its multiple pharmacological targets, as being the relevant activity for angiogenesis. Thus there is no reason to suspect that a more highly targeted agent could not be repurposed in a similar fashion. Overall, these results suggest that a fundamental understanding of systems biology will prove to be directly relevant to drug discovery, complementing traditional screening approaches to pharmacophore discovery and accelerating both basic and clinical biomedical research.
Compound genetic interaction profiles were downloaded from http://chemogenomics.stanford.edu:16080/supplements/global/download.html. We employed the p values reported for fitness defects in the yeast homozygous deletion collection for all analyses [13]. Candidate angiogenesis inhibitors were prioritized that consistently clustered with lovastatin across different choices of similarity measures and hierarchical clustering algorithms, specifically centered and uncentered correlation, Spearman rank correlation, absolute correlation (centered and uncentered), Euclidean distance, and City-block distance, employing centroid linkage, complete linkage, single linkage, or average linkage clustering. Clustering results were visualized with Cluster 3.0 (http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm) and Java TreeView (http://jtreeview.sourceforge.net/).
Female adult Xenopus were ovulated by injections of human chorionic gonadotropin, and eggs were fertilized in vitro and dejellied in 3% cysteine (pH 7.9) and subsequently reared in 1/3× Marc's modified Ringer's (MMR) solution. For microinjections, embryos were placed in a solution of 2% Ficoll in 1/3× MMR solution, injected using forceps and an Oxford universal manipulator, reared in 2% Ficoll in 1/3× MMR to stage 9, then washed and reared in 1/3× MMR solution alone. For bilateral rab11b knock-down experiments, the posterior cardinal vein and intersomitic veins were targeted by injecting Morpholino antisense oligonucleotides (MOs) into the two ventral cells equatorially at the four-cell stage. For unilateral knockdown, only one ventral cell was injected. MOs were injected at 40 ng per blastomere. For the experiments to see the drug effects, embryos were placed in a solution of each chemical dissolved in 1% DMSO diluted in 1/3× MMR during indicated stages. For bead micro-surgery implantation, Affi-Gel Blue Gel beads (Bio-Rad) were soaked with 0.7 mg/ml recombinant mouse VEGF 164 aa (R&D systems) or BSA as a control. Whole-mount in situ hybridization for erg and aplnr was performed as described [44].
Erg and aplnr cDNAs were obtained from Open BioSystems (erg: IMAGE:5512670, aplnr: IMAGE:8321886). Translation-blocking antisense morpholinos for rab11b were designed based on the sequences from the National Center for Biotechnology Information database (accession number: BC082421.1). MOs were obtained from Gene Tools with the following sequence: 5′-CGTATTCGTCATCTCTGGCTCCCAT-3′.
Human umbilical vein endothelial cells (HUVECs) were purchased from Clonetics, and were used between passages 4 and 9. HUVECs were cultured on 0.1% gelatin-coated (Sigma) plates in endothelial growth medium-2 (EGM-2; Clonetics) in tissue culture flasks at 37°C in a humidified atmosphere of 5% CO2.
HUVECs (104 cells) were seeded in a 96-well plate coated with 50 µl of ECMatrix (Chemicon) or Matrigel (BD Bioscience) according to the manufacturer's instructions. Cells were incubated for 16 h on EGM-2 containing thiabendazole, dissolved in 1% DMSO. Negative control cells were treated with 1% DMSO in the same manner. As a positive control, siRNA versus the human HoxA9 sequence [45] was transfected into HUVECs using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions. Tube formation was observed using an inverted microscope (Nikon, eclipse TS100), and branch points were measured using ImageJ software (http://rsb.info.nih.gov/ij).
HUVECs (1.2×105 cells) were seeded into 24-well plates for 24 h, and the monolayers were wounded identically. Then, cells were washed with PBS and treated with EBM-2 containing 1% DMSO or 250 µM TBZ dissolved in 1% DMSO with a combination of Y27632 or GGPP (Sigma). In the case of Y27632 treatment, cells were preincubated for 2 h before wounding. Cells were photographed at time zero and after 15 h, and the ratios of cell free area [(0 h–15 h)/0 h] were calculated.
Specific pathogen-free athymic Cre nu/nu mice were purchased from Charles River Laboratories. The HT1080 human fibrosarcoma cell line was obtained from the American Type Culture Collection (ATCC). HT1080 cells were cultured in DMEM (Gibco) containing 10% fetal bovine serum (FBS, Gibco) in tissue culture flasks at 37°C in a humidified atmosphere of 5% CO2. In order to generate a mouse xenograft model, a suspension of the HT1080 cells (3×106 in 50 µl PBS) mixed with an equal volume of Matrigel (BD Bioscience) was subcutaneously implanted into the flank region of 7–8-wk-old female mice. Upon establishment of tumors (approx. 40 mm3), mice were given daily intraperitoneal injections of 1 mg thiabendazole (Sigma-Aldrich), suspended in 20 µl DMSO. Mice weighed on average 20 grams; this dose thus corresponded to 250 µM TBZ. As a control, an equal volume of DMSO was injected in the same manner. Tumor growth was monitored by measuring the length and width of each tumor using digital calipers, and the tumor volume in mm3 calculated by the formula: Volume = (width)2×length/2. Upon a tumor reaching the maximum size permitted by the Institutional Animal Care and Use Committee (1.5 cm in diameter), the mouse was sacrificed, and the tumor excised.
Each tumor was fixed with 4% paraformaldehyde in PBS, and cryostat sections were processed. After blocking with 5% goat serum in PBST (0.3% Triton X-100 in PBS) for 1 h at room temperature, sectioned tissues were incubated with anti-mouse CD31 antibody, hamster clone 2H8, 1∶100 (Millipore). After several PBST washes, samples were incubated for 2 h at room temperature with FITC-conjugated anti-hamster IgG antibody, 1∶1,000 (Jackson ImmunoResearch).
In order to determine the effect of thiabendazole on proliferation and apoptosis, 2×105 HUVECs or HT1080 were cultured in 6-well plates and treated with thiabendazole dissolved in 1% DMSO. Control cells received 1% DMSO. For actin and tubulin cytoskeleton analysis, 7×104 HUVECs were seeded. After 24 h, cells were fixed using 4% paraformaldehyde in PBS. Cell membranes were permeabilized with 0.2% Triton X-100 in PBS, and nonspecific immunobinding sites were blocked with 5% goat serum for 1 h at room temperature. Cells were incubated with primary antibodies to Caspase-3 (Abcam), Phospho-histone H3 (Ser10; Millipore), or β-tubulin (Sigma) at 4°C overnight. After washing with PBST, primary antibodies were detected by Alexa Fluor-488 or 555 goat anti-rabbit immunoglobulin (IgG). Alexa Fluor 488 phalloidin (Invitrogen) and/or 4′,6-Diamidino-2-phenylindole (Sigma) were added as needed.
Immunostaining for Xenopus was performed as previously described [46]. Embryos at stage 35–36 were fixed in 1× MEMFA. 12/101 (1∶500; DSHB) and primary antibodies were detected with Alexa Fluor-488 or 555 goat anti-mouse Immunoglobulin (IgG).
Immunohistochemistry experiments and kdr:GFP transgenic Xenopus laevis were imaged on an inverted Zeiss LSM5 Pascal confocal microscope and Zeiss 5-LIVE Fast Scanning confocal microscope. Confocal images were processed and cropped in Imaris software (BITPLANE) and Adobe Illustrator and Adobe Photoshop for compilation of figures.
HUVECs were treated with 1% DMSO or 1% DMSO, 250 µM TBZ for 24 h, and lysed by Dounce homogenization in low salt buffer (10 mM Tris-HCl, pH 8.8, 10 mM KCl, 1.5 mM MgCl2) with 0.5 mM DTT and protease inhibitor mixture (Calbiochem). 2,2,2-trifluoroethanol was added to 50% (v/v) for each sample, and samples were reduced with 15 mM DTT at 55°C for 45 min and then alkylated with 55 mM iodoacetamide at room temperature for 30 min. Following alkylation, samples were diluted in digestion buffer (50 mM Tris-HCl, pH 8.0, 2 mM CaCl2) to a final 2,2,2-trifluoroethanol concentration of 5% (v/v) and digested using proteomics grade trypsin (Sigma) at 1∶50 (enzyme/protein) concentration and incubated at 37°C for 4–5 h. Digestion was halted with the addition of 1% formic acid (v/v), and sample volume was reduced to 200 µl by SpeedVac centrifugation prior to loading on HyperSep C-18 SpinTips (Thermo). Samples were eluted (60% acetonitrile, 0.1% formic acid), reduced to 10 µl by SpeedVac centrifugation, and resuspended in sample buffer (5% acetonitrile, 0.1% formic acid). Tryptic peptides were then filtered through Microcon 10-kDa centrifugal filters (Millipore), and collected as flow-through. Peptides were chromatographically separated on a Zorbax reverse-phase C-18 column (Agilent) via a 230 min 5%–38% acetonitrile gradient, then analyzed by on-line nanoelectrospray-ionization tandem mass spectrometry on an LTQ-Orbitrap (Thermo Scientific). Data-dependent ion selection was performed, collecting parent ion (MS1) scans at high resolution (60,000) and selecting ions with charge >+1 for collision-induced dissociation fragmentation spectrum acquisition (MS2) in the LTQ, with a maximum of 12 MS2 scans per MS1. Ions selected more than twice in a 30 s window were dynamically excluded for 45 s. MS2 spectra were interpreted using SEQUEST (Proteome Discoverer 1.3, Thermo Scientific), searching against human protein-coding sequences from Ensembl release 64 [47]. Search results were then processed by Percolator [48] at a 1% false discovery rate. Protein groups were generated comprising proteins with identical peptide evidence, omitting those proteins whose observed peptides could be entirely accounted for by other proteins with additional unique observations. Differential expression of proteins across TBZ-treated and control samples was quantified from the MS2 spectral count data using the APEX method of relative quantification [49].
HT1080 (2×105 or 4×104cells) were cultured in 6-well plates and treated with 1% DMSO or 1% DMSO, 250 µM TBZ for 24 h. Cells were lysed in cell lysis buffer (Cell Signaling Technology) containing 1 mM PMSF, and analyzed by SDS-PAGE and Western blotting using anti-VEGF (Santa Cruz, A-20) or anti-GAPDH (Cell Signaling Technology) antibodies. The secreted VEGF level in culture medium was determined by enzyme-linked immunosorbent assay (ELISA; R&D) according to the manufacturer's instructions.
|
10.1371/journal.ppat.1006360 | MicroRNAs upregulated during HIV infection target peroxisome biogenesis factors: Implications for virus biology, disease mechanisms and neuropathology | HIV-associated neurocognitive disorders (HAND) represent a spectrum neurological syndrome that affects up to 25% of patients with HIV/AIDS. Multiple pathogenic mechanisms contribute to the development of HAND symptoms including chronic neuroinflammation and neurodegeneration. Among the factors linked to development of HAND is altered expression of host cell microRNAs (miRNAs) in brain. Here, we examined brain miRNA profiles among HIV/AIDS patients with and without HAND. Our analyses revealed differential expression of 17 miRNAs in brain tissue from HAND patients. A subset of the upregulated miRNAs (miR-500a-5p, miR-34c-3p, miR-93-3p and miR-381-3p), are predicted to target peroxisome biogenesis factors (PEX2, PEX7, PEX11B and PEX13). Expression of these miRNAs in transfected cells significantly decreased levels of peroxisomal proteins and concomitantly decreased peroxisome numbers or affected their morphology. The levels of miR-500a-5p, miR-34c-3p, miR-93-3p and miR-381-3p were not only elevated in the brains of HAND patients, but were also upregulated during HIV infection of primary macrophages. Moreover, concomitant loss of peroxisomal proteins was observed in HIV-infected macrophages as well as in brain tissue from HIV-infected patients. HIV-induced loss of peroxisomes was abrogated by blocking the functions of the upregulated miRNAs. Overall, these findings point to previously unrecognized miRNA expression patterns in the brains of HIV patients. Targeting peroxisomes by up-regulating miRNAs that repress peroxisome biogenesis factors may represent a novel mechanism by which HIV-1 subverts innate immune responses and/or causes neurocognitive dysfunction.
| Host cells employ a myriad of antiviral defense systems but most viruses have developed effective countermeasures. Viruses such as HIV that cause lifelong infections are particularly successful in subverting the host antiviral response. While mitochondria have long been known to be critical hubs for antiviral signaling, it has only recently become apparent that peroxisomes are also important for this process. Peroxisomes are small and numerous structures that are best known for their roles in lipid metabolism. New evidence suggests that pathogenic viruses such as West Nile and Dengue viruses block the production of peroxisomes by sequestering and degradation a critical biogenesis factor. In the present study, we report that HIV significantly reduces the number of peroxisomes in infected cells via a completely novel mechanism. Specifically, HIV-infected cells express high levels of microRNAs that inhibit production of proteins required for peroxisome formation. Interestingly, levels of these microRNAs were elevated in the brains of patients with HIV-associated neurocognitive disorders. Thus, as well as affecting antiviral signaling, loss of peroxisomes during HIV infection may contribute to development of neurological disorders. Understanding how pathogenic viruses affect peroxisome biogenesis and cognate antiviral signaling may ultimately lead to novel therapeutic avenues and prevention of long-term sequelae.
| Leukocytes infected by human immunodeficiency virus type 1 (HIV-1) traverse the blood-brain barrier within days of primary infection resulting in subsequent infection of macrophage lineage cells (microglia and perivascular macrophages) and astrocytes in the central nervous system (CNS) [1, 2]. As HIV/AIDS progresses, a subset of infected patients develop a neurological syndrome termed HIV-associated neurocognitive disorders (HAND) [3, 4]. HAND affects approximately 25% of HIV-infected patients despite the availability of effective antiretroviral therapy [3, 5–7]. Some of the proposed mechanisms that contribute to HAND include genetic host susceptibility factors, viral properties [8–11] and altered host immune responses [12, 13]. Moreover, neurotoxic effects of some antiretroviral therapies have been implicated in HAND development (reviewed in [14]). The collective actions of neurotoxic viral proteins and chronic neuroinflammation mediated by cytokines and free radicals culminate in synaptic injury and eventual neuronal death, leading to HAND. There are currently no specific therapies for HAND although antiretroviral therapy can alleviate some neurological defects.
Among the factors suggested to contribute to the development of HAND is altered expression of host cell microRNAs (miRNAs). These small noncoding RNAs can regulate both host and viral gene expression [15] and profiling miRNAs in different pathological conditions has yielded important insights into underlying disease mechanisms [16–18]. To this end, it was recently reported that miRNA profiles in the central nervous systems of HIV-infected patients with HAND, differs from nonHAND patients [19, 20]. Similarly, the miRNA signatures in blood from HIV-infected elite controllers differ from those of viremic patients, HAND and nonHAND patients [21–23]. Importantly, altered expression of host miRNAs may not only contribute to the development of HAND but also could potentially be exploited as diagnostic and prognostic biomarkers for HAND [23]. To further investigate the link between host miRNA expression and HAND development as well as HIV-1 biology, brain miRNA profiles were examined in HIV/AIDS patients with and without HAND. We identified 17 miRNAs that had abnormal expression levels in the brains of HAND patients. Bioinformatic analyses revealed that four of the up-regulated miRNAs target key peroxisome biogenesis factors.
Peroxisomes are ubiquitous and essential subcellular organelles responsible for the catabolism of fatty acids (beta oxidation), amino acids, reduction of free radicals such as hydrogen peroxide and the synthesis of plasmalogens. The latter is critical for myelin formation and brain development [24]. Formation of peroxisomes requires multiple peroxin (PEX)-encoding genes and mutations result in devastating diseases that include defects in brain development (reviewed in [25, 26]). In addition to their roles in cellular lipid metabolism and brain development and function, peroxisomes serve as signaling platforms in antiviral defense [27] further underlying their importance in human health. Activation of peroxisomal-MAVS during RNA virus infections leads to the production of type III interferon (IFN) as well as IFN-stimulated genes (ISGs) [27, 28]. Peroxisomes play a role in sensing the HIV-1 genomic RNA [29] and stimulation of peroxisome proliferator-activated receptor alpha by fenofibrate impairs replication of HIV-1 and flaviviruses in vivo [30, 31]. Consistent with their roles in antiviral defense, a number of recently published reports revealed that during viral infection, peroxisome biogenesis and/or peroxisome-based signaling is disrupted [32–34]. In these cases, viral proteins directly interact with peroxisomes or biogenesis factors to interfere with peroxisome function or formation.
Here, we show for the first time that peroxisomes are depleted during HIV-1 infection via a unique mechanism. While the PEX mRNA targeting miRNAs were initially discovered in the brains of HIV-infected patients with neurocognitive defects, subsequent analyses revealed that their upregulation is a fundamental aspect of HIV infection. Thus as well as potentially blunting the innate immune response during early stages of infection, HIV-1 induced loss of peroxisomes may play a role in development of neurological disorders in AIDS patients.
The development of HAND is dependent on multiple factors including aberrant expression of host-encoded miRNAs. To determine whether there were signature miRNA expression patterns common to HAND patients, we examined a well-defined patient cohort [35–37], focusing on miRNA profiles in brain tissue from HIV/AIDS patients with HAND (n = 20; with encephalitis, n = 10 and without encephalitis, n = 10) to HIV/AIDS patients without HAND or encephalitis (n = 10). To ensure there were sufficient patients in each group and because there were no significant differences in miRNA expression between the two HAND groups, the results from each HAND group were pooled. We found that expression of 17 miRNAs (Fig 1 and Table 1) was consistently dysregulated in the HAND samples. Twelve of the miRNAs were upregulated and five were down-regulated at least 1.5 fold (p < 0.05).
Three algorithms (TargetScan, miRDB and DIANA) were used to predict targets of each miRNA and high-ranking potential targets predicted by at least two out of three algorithms are shown. Notably, peroxisomal genes (PEX2, PEX7, PEX11B and PEX13) that are the predicted targets of 4 up-regulated miRNAs (miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p) are bolded and underlined.
To understand the potential effects of the differentially expressed miRNAs in pathogenesis of HAND and/or HIV-1 biology, it was important to elucidate their cellular targets. Three bioinformatics algorithms (miRDB, DIANA, and TargetScan) were used to predict potential targets of the 17 differentially expressed miRNAs. We first focused on targets that were predicted by at least two of the three algorithms. In keeping with the notion that a single miRNA can affect expression of dozens of mRNAs, we identified hundreds of potential targets. Some of the highest-ranking candidates are listed in Table 1. Interestingly, four of the up-regulated miRNAs (miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p) are predicted to target mRNAs encoding the peroxins PEX2, PEX7, PEX11B and PEX13. These proteins play different but critical roles in biogenesis of peroxisomes. Specifically, PEX2 and PEX13 are required for import of peroxisomal matrix proteins; PEX11B facilitates peroxisomal division and proliferation and PEX7 functions as a receptor for the import of peroxisomal matrix proteins with type 2 targeting motifs (reviewed in [38]).
Peroxisomes have only recently been shown to play roles in antiviral defense [24, 25] but have long been linked to neuroinflammation (reviewed in [39]). As such, we elected to determine if/how the HIV-induced miRNAs affect expression of peroxisomal proteins. In most cases, miRNAs negatively regulate gene expression at the post-transcriptional level through binding to the 3’untranslated regions (UTRs) of mRNAs. Therefore, we first determined whether miR-500a-5p, miR-34c-3p, miR-93-3p, or miR-381-3p affected expression of a reporter gene upstream from the 3’UTRs of PEX2, 7, 11B or 13 mRNAs (Fig 2). The pMIR-REPORT miRNA expression reporter system consists of a firefly luciferase reporter vector (for 3’-UTR cloning) and a β-gal reporter control plasmid (for normalization based on potential differences in cell viability and transfection efficiency). Several controls were included for each experiment. For example, miR-344-3p targets the 3’UTR of KLF4 [40] and therefore, we used this miRNA as the positive control. As a negative control, cassettes encoding the 3’-UTRs for the PEX genes were cloned into the reporter vector in the opposite direction.
Expression of luciferase activity under the control of PEX2, PEX7, PEX11B, or PEX13 UTRs was inhibited by 50–70% in cells transfected with miR-500a-5p, miR-34c-3p, miR-93-3p or miR-381-3p respectively (Fig 2). Conversely, these miRNAs did not affect luciferase activity when the orientations of PEX 3’UTRs were reversed. Together, these data indicate that four of the miRNAs upregulated in the brains of HAND patients efficiently suppress translation of PEX mRNAs.
We next focused on determining whether expression of the PEX mRNA-targeting miRNAs reduced levels of peroxisomal proteins. Immunoblotting was used to quantify the relative levels of peroxisomal proteins in cells transfected with mimics of miR-500a-5p, miR-34c-3p, miR-93-3p, miR-381-3p or a non-silencing miRNA (miR-NS). Data in Fig 3A show that compared to mock and miR-NS-transfected cells, over-expression of miR-500a-5p, miR-34c-3p, miR-93-3p and miR-381-3p resulted in significantly decreased levels of peroxisomal proteins albeit to different extents. Specifically, miR-500a-5p, which targets PEX2 mRNA (Fig 2), reduced levels of PEX2 protein by 35%. Interestingly, PEX7 and PEX11B protein levels were 70% and 69% lower respectively in cells transfected with miR-500a-5p. Similarly, the PEX13-targeting miR-381-3p decreased expression levels of four peroxisomal proteins including PMP70 (a peroxisomal membrane protein), PEX7, PEX13, and PEX2. Unexpectedly, transfection of cells with miR-34c-3p or miR-93-3p mimics did not significantly impact PEX7 or PEX11B protein levels respectively. However, miR-34c-3p expression resulted in loss of both PMP70 and PEX13 proteins. Expression of PEX13 was only slightly decreased by miR-93-3p. Finally, levels of catalase, a peroxisomal matrix protein, were unaffected by expression of the four miRNAs.
There are a number of scenarios in which a single miRNA can affect expression of multiple Pex gene products. One possibility is that miR-500a-5p, miR-381-3p and/or miR-34c-3p inhibit translation of multiple mRNAs that encode PEX proteins. Indeed, miRNAs that target components of a cellular pathway can be synthesized as a common transcript that contains multiple primary miRNAs [41]. However, a search of the miRBase database indicated that genes encoding miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p are located on different chromosomes. Moreover, the initial miRNA target search using miRDB, DIANA, and TargetScan did not indicate that multiple PEX mRNAs are targeted by miR-500a-5p, miR-34c-3p, miR-93-3p, or miR-381-3p. Nevertheless, we employed the luciferase-based reporter assay described above to experimentally determine if any of these miRNAs could target more than one PEX gene. Data in S1 Fig confirmed that the miRNAs only regulated expression of luciferase under the control of 3’UTRs from their predicted PEX mRNA targets. Specifically, miR-500a-5p, miR-34c-3p, miR-93-3p and miR-381-3p downregulated expression of luciferase under the control of the 3’UTRs from PEX2, PEX7, PEX11B and PEX13 mRNAs respectively.
We also used siRNAs to determine if decreasing expression of PEX2, PEX7, PEX11B or PEX13 proteins affected steady state levels of one another. Unlike miRNAs, which are inherently degenerate with respect to mRNA targets, siRNAs are perfectly complementary to their mRNA targets. siRNAs against PEX2, PEX7, PEX11B or PEX13 were transfected into HEK293T cells and levels of proteins were determined by immunoblotting (S2 Fig). These experiments showed that targeted knockdown of a single PEX protein can indeed result in concomitant loss of other PEX proteins. For example, siRNAs against PEX7 not only reduced the level of PEX7 protein, but PEX11B was also markedly lower. Similarly, a PEX13-specific siRNA reduced the levels of PEX13 and PEX7 proteins. Finally, we showed that downregulation of the multifunctional peroxisome biogenesis factor PEX19 using siRNA, effectively reduced levels of PEX19, PEX7, PEX11B and PEX13 proteins. Unfortunately, we were unable to achieve significant reduction of PEX2 protein with siRNAs, despite using at least three different siRNAs.
Next we examined how overexpression of miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p affected peroxisomes. Super-resolution microscopy was used to analyze the morphology, distribution and numbers of peroxisomes in miRNA-transfected cells. Peroxisomes were identified using an antibody to PMP70, a peroxisomal membrane protein involved in membrane assembly [42]. Cells transfected with a non-silencing miRNA (miR-NS) contained hundreds of PMP70-positive puncta throughout the cytoplasm (Fig 3B). While the number of peroxisomes was significantly reduced by expression of miR-500a-5p (which targets PEX2), most striking was the change in morphology and PMP70 staining of the peroxisomes. Specifically, miR-500a-5p over-expression resulted in enlargement and elongation of peroxisomes. Decreasing the intracellular level of PEX2, an E3 ubiquitin ligase that targets PMP70 [43] could certainly explain the higher levels of PMP70 protein (Fig 3A) and increasing staining intensity of anti-PMP70 in miR-500a-5p over-expressing cells (Fig 3B).
It is important to point out that PEX11B is required for peroxisome fission (reviewed in [38]) and as such, the fact that miR-500a-5p expressing cells have lower levels of this protein could result in decreased fission of peroxisomes and concomitant lengthening and enlargement of these organelles. Unexpectedly, the effect of miR-93-3p (which targets PEX11B) on peroxisomes was minimal. Despite evidence showing that the 3’UTR of PEX11B is targeted by this miRNA (Fig 2), PEX11B protein levels were not significantly affected by over-expression of a miR-93-3p mimic (Fig 3A). One possibility is that PEX11B protein is very stable and the cellular pool was not depleted within the time frame of our experiments. Finally, it can be seen that expression of miR-34c-3p and miR-381-3p reduce peroxisome numbers by 65% and 45% respectively (Fig 3B). Notably, this is consistent with the immunoblot data in Fig 3A showing that levels of PMP70 protein were reduced by expression of miR-34c-3p and miR-381-3p.
To determine if peroxisomes were affected by HIV-1 infection, immunofluorescence and immunoblot assays were conducted on infected Hela CD4+ cells and monocyte-derived macrophages respectively. Data in Fig 4A show that similar to what was observed in miRNA-transfected cells (Fig 3B), HIV infection results in significant loss of peroxisomes in Hela CD4+ cells. These cells were used for the microscopy assays because their flat morphology is more conducive for peroxisome quantitation. Peroxisomes were identified using an antibody to the tripeptide Ser-Lys-Leu (SKL), a targeting motif found at the carboxyl termini of many peroxisomal matrix proteins [44] (Fig 4A). Quantification of SKL-positive structures showed that on average HIV-infected cells contained 40% less peroxisomes than mock-treated cells (Fig 4A). Immunoblotting revealed that infection of primary macrophages, a physiologically relevant cell type in HIV patients, resulted in dramatic loss of PEX2, PEX7, PEX13, and to a lesser extent, PEX11B (Fig 4B). However, levels of catalase, a peroxisomal matrix protein were not affected by HIV infection. This indicates that the effects of HIV-1 protein expression on peroxisome-associated proteins are highly specific. Similar results were observed in infected Hela CD4+ cells (S3 Fig).
Next, we used immunoblotting to analyze peroxisomal protein levels and immunohistochemistry to assess peroxisome morphology in frontal lobe brain tissue from HIV/AIDS and uninfected patients. Data in Fig 5A show that PEX13 protein was virtually absent in HIV patients with or without encephalitis or HAND. Levels of PEX7 protein were also significantly (40%) lower in the sample from an HIV patient without encephalitis or HAND, however in three HAND samples, steady state levels of PEX7 protein were lower than those seen in HIV patients without HAND as well as non-infected patients. Finally, levels of PEX2 and PEX11B proteins were reduced (~70–80%) in brain tissue from all of the HIV patients assayed.
As a secondary assay, brain tissue samples from uninfected and HIV/AIDS patients were examined by immunocytochemistry. Immunolabeling of frontal lobe sections showed that the intensity of PEX13 and thiolase immunostaining which was concentrated in astrocytes (arrows), was consistently lower in HIV/AIDS tissue compared to that from uninfected patients (Fig 5B). Although the data are from a small sample size, they suggest that HIV infection contributes to loss of peroxisomal material in brain tissue.
Our data are consistent with a scenario in which the loss of peroxisomes during HIV-1 infection is caused by increased expression of miRNAs that target mRNAs encoding peroxisome biogenesis factors. To address this hypothesis, we first determined if miR-500a-5p, miR-34c-3p, miR-93-3p and/or miR-381-3p were upregulated in HIV-infected macrophages. Human primary macrophages were infected with HIV-1 (MOI = 2) and after 5 days, relative levels of miRNAs were determined by RT-qPCR. Data in Fig 6A show that levels of miR-500a-5p and miR34c-3p were increased almost 2.5 fold whereas miR-93-3p and miR-381p were increased between 1.6 and 2.2 fold. In contrast, levels of miR-483-5p (which does not target PEX mRNAs and was identified as a miRNA whose expression was decreased in brain tissue of HAND patients, Table 1) were slightly decreased in HIV-infected macrophages.
To further investigate the mechanism underlying HIV-associated loss of peroxisomes, we used anti-miRs to block the functions of PEX mRNA-targeting miRNAs during HIV infection. As transfection of primary macrophages can be technically challenging [45], we elected to employ CD4+ Hela cells for these experiments. Data in Fig 6B show that with the exception of miR-93-3p, expression of PEX-targeting miRNAs was significantly elevated in HIV-infected CD4+ Hela cells. Anti-miR-500a-5p had the most dramatic effect in that it completely prevented HIV-induced loss of PEX2, PEX7, PEX11B and PEX13 (Fig 7A). Other miRNA inhibitors had intermediate effects. For example, anti-miR-34c-3p increased levels of PEX13; anti-miR-93-3p increased levels of PEX7 and PEX11B; and anti-miR-381-3p increased levels of PEX11B.
Since miR-500a-5p had the greatest effect on peroxisomal protein expression, we questioned whether blocking the activity of this miRNA could prevent HIV-induced loss of peroxisomes. Results in Fig 7B show that anti-miR-500a-5p abrogated the effect of HIV-1 infection on peroxisomes. Specifically, the average number of peroxisomes in HIV-infected cells containing the inhibitor of miR-500a-5p was not statistically different from that of mock-treated cells.
Because peroxisomes are now recognized to have important roles in antiviral signaling [27, 28], we questioned whether expression of miRNA mimics that target PEX mRNAs would affect innate immune genes. A549 cells were chosen for these experiments because they are human in origin and have been used extensively to study innate immune signaling. Interestingly, three of the miRNA mimics (miR-500a-5p, miR-34c-3p and miR-93-3p) significantly increased mRNA levels for five innate immune genes (Fig 8). MiR-93-3p had the most dramatic affect on expression of antiviral genes. Specifically, in cells transfected with miR-93-3p mimic, expression of IFI6 and viperin mRNAs were increased 14-fold and 50-fold respectively. MiR-500a-5p appeared to modestly increase expression of innate immune genes (2–4 fold) whereas miR-381-3p did not significantly affect expression of viperin, IFI6, IFIT2, IRF1 or OAS1 (Fig 8).
A growing body of research has linked changes in miRNA expression to pathogenesis of neurodegenerative diseases including Alzheimer’s disease, Parkinson’s disease, Huntington’s disease and amyotrophic lateral sclerosis (reviewed in [46]). The original goal of the present study was to identify miRNAs that are differentially expressed in the brains of HIV-infected patients with HAND. Of the 17 miRNAs whose expression levels were commonly deregulated in HAND patients, four (miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p) were shown to regulate expression of the peroxisome biogenesis factors PEX2, PEX7, PEX11B and PEX13. Subsequent analyses revealed that elevated expression of these miRNAs was not specific to HIV-HAND but rather, was a common feature of HIV infection. This is the first report to our knowledge demonstrating that viral infection leads to increased expression of mRNAs that downregulate peroxisomes, possibly as a mechanism to alter early antiviral signaling that emanates from these organelles.
The present study also connects two divergent areas, HIV-1 biology and peroxisome dysfunction in the brain, providing previously unrecognized insights into the pathogenesis of a common neurological syndrome, HAND. Although larger samples sizes and further analyses are required to confirm our findings, in general, the relative loss of PEX proteins in brain tissue appears to be greater in HAND compared to non-HAND HIV. This scenario is consistent with our initial observation that miR-500a-5p, miR-34c-3p, miR-93-3p, and miR-381-3p were expressed at higher levels in brain tissue from HIV-HAND vs HIV-non-HAND patients. Moreover, a large number of studies indicate that peroxisomes are critical for brain function (reviewed in [24]); thus stressing the need for further investigation into peroxisomes and HIV pathogenesis.
Peroxisome-based diseases in humans can be classified into two large groups; peroxisome biogenesis disorders and single peroxisome enzyme deficiencies. The first group includes Zellwegger syndrome spectrum disorders and rhizomelic chondrodysplasia punctata type 1. In cells from patients with peroxisome biogenesis disorders, peroxisomes are absent due to mutations in one or more genes that encode critical biogenesis factors including PEX2, PEX7 and PEX13 (reviewed in [26]). Not surprisingly, these disorders often result in death at an early age and patients suffer from a wide variety of neurological abnormalities including leukodystrophy (inflammatory degeneration of white matter), similar to that observed in advanced HAND, termed HIV-associated dementia [47]. Mutations in genes that encode peroxisomal enzymes also result in severe neurological deficits, again defined by inflammatory degeneration of white matter [48]. These studies underscore the fact that even partially diminished function of peroxisomes can lead to severe neurological disease.
The function of peroxisomes in antiviral signaling is a relatively new discovery [27, 28]. A pool of the mitochondrial antiviral signaling protein (MAVS), an adaptor protein for retinoic acid-inducible gene 1 protein (RIG-I), localizes to peroxisomes. Activation of MAVS-dependent signaling from peroxisomes by different RNA viruses leads to activation of type III interferon, a process that is thought to complement the type I interferon response induced from mitochondria, which occurs later. MAVS signaling from both peroxisomes and mitochondria is required for maximal anti-viral activity. The observation that viruses have evolved strategies to interfere with peroxisome-dependent anti-viral signaling illustrates the importance of this organelle in defending against pathogens. For example, the hepatitis C virus NS3-4A protease has been shown to cleave both peroxisomal and mitochondrial MAVS to suppress RIG-I signaling of immune defenses [33, 49, 50]. Another mechanism used by flaviviruses such as West Nile virus and Dengue virus, involves targeting of PEX19, a critical peroxisome biogenesis factor, for degradation [32]. Flavivirus-infected cells contain significantly lower numbers of peroxisomes, an effect that is mediated in large part through binding of capsid protein to PEX19. As a result of the reduced peroxisome pool, type III interferon expression is dramatically reduced. Finally, the observation that human cytomegalovirus HCMV protein vMIA, which inhibits signaling downstream from mitochondrial MAVS, also localizes to peroxisomes [34], may indicate that peroxisomes play a role in defense against DNA viruses too. While vMIA interacts with peroxisomal MAVS and induces peroxisome fragmentation, disruption of peroxisomal morphology is not essential for this viral protein to inhibit antiviral signaling. Association of vMIA with peroxisomes may require interaction with PEX19. Although it was not further investigated, it is intriguing to note that MAVS is also a predicted target of miR-93-3p (Table 1).
Here we show that infection by HIV-1, a lentivirus, negatively impacts peroxisomes by a novel mechanism. The fact that blocking miRNA function with anti-miRs abrogates HIV-induced loss of peroxisomes suggests that upregulation of miRNAs is the main mechanism by which the virus targets these critical organelles that function in antiviral defense and neuroprotection. Of note, three of the four PEX proteins (PEX2, PEX7 and PEX13 targeted by HIV-induced miRNAs are associated with peroxisome biogenesis disorders [26]. Interfering with peroxisome biogenesis/function by altering miRNA expression appears to be a very efficient mechanism because some of the HIV-induced miRNAs repress expression of multiple PEX proteins. For instance, miR-500a-5p, which targets PEX2, also reduced expression of PEX7 and PEX11B proteins. Similarly, the PEX13-targeting miR-381-3p decreased expression of PMP70, PEX7, PEX13, and PEX2. This was not due to the miRNAs targeting multiple PEX mRNAs but rather, it seems that expression and/or stability of given PEX proteins is often dependent on other PEX proteins. For example, cells transfected with siRNAs against PEX7 not only reduced PEX7 protein but PEX11B was also markedly lower. Similarly, a PEX13-specific siRNA reduced the levels of PEX13 and PEX7 proteins. Consistent with these data, it has been reported that siRNAs against PEX7 also reduce levels of PMP70 protein [51] and knockout of the PEX2 gene in mice negatively impacts PEX14, PEX3, PEX16 and catalase [52]. Finally, PEX11B and PEX13 knockout mice express lower levels of PEX14 protein [53, 54].
Inhibition of peroxisome biogenesis and/or function during virus infection is only a recently discovered phenomenon [32–34]. However, given the roles of these organelles in antiviral defense and nervous system function, understanding how viruses manipulate peroxisomes will undoubtedly reveal pathological mechanisms that underlie multiple viral diseases. In the case of HIV-1, elevated expression of PEX gene-targeting miRNAs was initially detected among patients with HAND. However, the fact that these miRNAs are also upregulated during HIV-1 infection of macrophages and that loss of peroxisomal proteins was observed in the brains of HIV patients without HAND indicate that virus-induced loss of peroxisomes is a fundamental aspect of HIV biology.
Of interest is whether the degree of PEX-specific miRNA expression correlates with disease severity. The higher levels of PEX-specific miRNA expression in HAND patients compared to HIV patients without HAND is consistent with this scenario. Moreover, in a small sample size, we observed that loss of certain PEX proteins (PEX) was more pronounced in brain tissue from HAND patients compared to HIV patients without HAND. However, further investigation is required to determine if this is a ubiquitous phenotype in HAND patients. It is important to point out that only a small subset of permissive brain cells exhibit detectable viral genome or protein expression. Thus, it is plausible that the miRNA changes as well as altered Pex gene expression in HAND brains might be due in part to effects on bystander cells such as astrocytes, which rarely exhibit in vivo productive infection but are the most populous cell type in the brain. Nonetheless, miRNAs are being explored as diagnostic and prognostic biomarkers for various neurological conditions including Alzheimers’s and Parkinson’s diseases as well as HAND (reviewed in [55]).
Presently, there is very little known about aberrant miRNA expression and peroxisomal biogenesis disorders. However, a large number of recent studies have focused on the relationship between miRNA expression and peroxisome proliferator-activated receptors [56–59] and some of the findings have implications for neurological disease. Future studies that further clarify how viruses manipulate miRNAs are likely to reveal novel roles for miRNAs in peroxisome-dependent anti-viral defense, lipid metabolism and neurodegenerative disorders.
Brain tissue from HAND and non-HAND patients was collected at autopsy with informed consent at different geographical locations (Texas, New York, San Diego and Los Angeles) by the National NeuroAIDS Tissue Consortium [36, 37]. The use of autopsied brain tissues (Protocol number 2291) is approved by the University of Alberta Human Research Ethics Board (Biomedical) and written informed consents from all participants were signed before or at the collection times. The protocols for obtaining post-mortem brain samples comply with all federal and institutional guidelines with special respect for the confidentiality of the donor's identity.
Neocortical brain tissue samples from midfrontal gyrus were excised from fresh-frozen brain slices and shipped in dry ice to the Laboratory for Neurological Infection and Immunity Brain Bank at University of Alberta. Samples were stored at -80°C until total RNA extraction including microRNAs (miRNAs) was performed as follows. Briefly, ~100 mg/sample of autopsy-derived brain tissue was aseptically collected using sterile instruments into a 2 ml Lysing Matrix tube (MP Biomedicals, Santa Ana, CA, USA). Tissue samples were homogenized in a FastPrep-24 tissue homogenizer (MP Biomedicals, Santa Ana, CA, USA) after adding 1 ml of Trizol reagent (Invitrogen Carlsbad, CA, USA). Chloroform (200 μl) was added to each homogenate which was then centrifuged 12,000 x g for 15 minutes at 4°C. The aqueous phase was collected and extraction followed as indicated in the manufacturer’s manual (Qiagen, Catalog no. 217004).
Affymetrix miRNA 3.0 GeneChips were used for miRNA analyses. This microarray chip provides comprehensive coverage for mature human miRNAs (1733 probes) and pre-miRNAs (1658 probes). The Affymetrix FlashTag Biotin highly sensitive and reproducible (HSR) RNA Labelling kit was used to label RNA samples for analysis. Equal concentrations of total RNA including microRNAs (800–1000 ng) were poly-A tailed as specified by the manufacturer (Affymetrix) followed by biotin-HSR ligation. Next, samples were treated with T4 DNA ligase before they were hybridized to Affymetrix miRNA 3.0 GeneChip arrays at 48°C for 16 hours. Arrays were then stained and washed on an Affymetrix GeneChip Fluidics 450 following manufacturer’s protocol and then scanned with an Affymetrix GeneChip Scanner 3000 7G System.
Genespring (version 12.6) software (Agilent Technologies) was used to normalize the data and identify differentially expressed miRNAs. The normalization in this software is based on the Robust Multi-array Average (RMA) algorithm, in which data are background-corrected, log2 transformed and quartile normalized. To identify differentially expressed miRNAs, the median of each probe set in the HAND or nonHAND patients was calculated and the non-parametric test Mann-Whitney unpaired test was applied. To select for differentially expressed miRNAs in this analysis, a cut-off fold change (≥ 1.5) in relative miRNA abundance and a p value of <0.05 was considered statistically significant.
Three different bioinformatics algorithms (miRDB, http://mirdb.org/miRDB/index.html; Diana-microT-CDS;http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=microT_CDS/index; and TargetScanHuman v6.2, http://www.targetscan.org/) were used to predict the potential targets of differentially expressed miRNAs. Only mRNA targets that were predicted by at least two of the three algorithms were investigated further.
Complete EDTA-free protease inhibitor cocktail (Roche Diagnostics (Laval, Quebec, Canada); ProLong Gold anti-fade reagent with 4,6-diamidino-2-phenylindole (DAPI), SlowFade Gold reagent mounting media, cell culture media DMEM, RPMI 1640, and fetal bovine serum (FBS) from Invitrogen (Carlsbad, CA) were purchased from the indicated suppliers.
Lipofectamine 2000 and Lipofectamine RNAiMAX were purchased from Invitrogen (Carlsbad, CA); Per-Fectin transfection reagent was from Genlantis (San Diego, CA).
miRIDIAN microRNA mimics including human hsa-miR-500a-5p, hsa-miR-34c-3p, hsa-miR-93-3p, hsa-miR-381-3p; miRIDIAN microRNA Mimic Negative Control #1 and miRIDIAN microRNA mimic mouse mmu-miR-344-3p; miRIDIAN microRNA inhibitors including human hsa-miR-500a-5p-Hairpin Inhibitor, hsa-miR-34c-3p-Hairpin Inhibitor, hsa-miR-93-3p-Hairpin Inhibitor and hsa-miR-381-3p-Hairpin Inhibitor were purchased from GE Healthcare Dharmacon Inc. (Lafayette, CO).
MGC human PEX2 (Clone ID: 3347824), PEX7 (Clone ID: 5176358), PEX11B (Clone ID: 3866690), and PEX13 (Clone ID: 6285875) sequence-verified full-length cDNA clones were purchased from GE Healthcare Dharmacon Inc. (Lafayette, CO).
Reagents for purification and quantitation of miRNAs including MiRNeasy Mini kit, miScript PCR Starter kit, miScript II RT kit, and miScript SYBR Green PCR kit were purchased from Qiagen (Toronto, ON).
Mouse monoclonal antibodies against the peroxisomal membrane protein PMP70 (Sigma, St. Louis, MO), HIV-1 p24 (Abcam, Cambridge, MA), and beta-actin (Abcam, Cambridge, MA) were purchased from indicated suppliers. Rabbit polyclonal antibodies to PEX7, PEX11B, PEX13, PEX19 and catalase were from Abcam (Cambridge, MA); Rabbit polyclonal antibody to PEX2 (PXMP3) was purchased from Pierce (Rockford, IL); Rabbit polyclonal antibody to thiolase (ACAA1) was from MyBioSource (San Diego, CA); Rabbit polyclonal antibody to the tri-peptide SKL were produced as previously described [60].
Donkey anti-mouse IgG conjugated to Alexa Fluor 680, goat anti-rabbit IgG conjugated to Alexa Fluor 680, donkey anti-mouse IgG conjugated to Alexa Fluor 488, donkey anti-rabbit IgG conjugated to Alexa Fluor 488, and donkey anti-mouse IgG conjugated to Alexa Fluor 546 were purchased from Invitrogen (Carlsbad, CA).
The buffy coats used for PBMC isolation were derived from healthy volunteer blood donors. Human monocytes were isolated using Histopaque (Sigma-Aldrich). Briefly, the blood was diluted 1:1 with phosphate-buffered saline (PBS), placed under a layer of Histopaque and centrifuged for 22 min at 1800 rpm in a clinical centrifuge. Cells from the interphase layer were harvested, washed twice with serum-free RPMI, and then resuspended in RPMI1640 with 15% FBS, 1% penicillin and streptomycin (Invitrogen, Carlsbad, CA). The cells (2–4 million per well) were then seeded in 6-well plates that were pre-coated with poly-L-ornithine (Sigma, St. Louis, MO). After 4 hours, the cells were washed three times with warm RPMI medium before adding 2mL Differentiation medium (25 ng/mL macrophage colony-stimulatory factor (M-CSF) (Sigma, St. Louis, MO) in RPMI containing 2mM L-glutamine, 1% penicillin and streptomycin and 15% FBS) to each well. Cells were incubated for 7 days in this media (with media changes every 3 day) to allow differentiation of MDMs.
A549 and HEK293T cells from the American Type Culture Collection (Manassas, VA) were cultured in DMEM (Invitrogen) containing 10% heat-inactivated FBS, 4.5 g/liter D-glucose, 2 mM glutamine, 110 mg/liter sodium pyruvate at 37°C in a 5% CO2 atmosphere. Hela CD4+ (clone 1022) cells (NIH AIDS Reagent Program, Germantown, MD) were cultured in RPMI 1640 supplemented with 10% FBS and 1.0 mg/ml G418 (Geneticin, Gibco).
A549 and HEK293T cells were transfected with the expression plasmids using Lipofectamine 2000 (Invitrogen) and PerFectin (Genlantis) respectively as described by the manufacturers. When using miRNA mimics or anti-miRs, cells were transfected with Lipofectamine RNAiMAX (Invitrogen). HIV-1 infection in Hela CD4+ cells (pYU2, MOI = 10) or primary monocyte-derived macrophages (pYU2, MOI = 2) was performed under biosafety CL-3 conditions.
To test whether miRNA mimics could silence predicted target genes, the entire 3’-untranslated regions (UTRs) of selected target genes were subcloned into the luciferase expression vector pMIR-REPORT-Luc (Ambion). Plasmids were constructed using polymerase chain reaction (PCR) and standard subcloning techniques. Sequence-verified full-length cDNAs of each PEX gene were used as templates to amplify the 3’-UTRs by PCR with primers listed in Table 2. The resulting PCR products were digested with HindIII and then subcloned immediately downstream of the luciferase cassette contained in the reporter plasmid pMIR-REPORT-Luc. The orientation of each 3’-UTR insert was determined by endonuclease digestion and all constructs were then verified by DNA sequencing.
The pMIR-REPORT miRNA Expression Reporter System (Ambion) was used to validate miRNA targets and conduct quantitative evaluations of miRNA function. The assay employs an experimental firefly luciferase-based reporter vector and an associated β-gal reporter control plasmid (pMIR-REPORT β-gal). The pMIR-REPORT Luciferase plasmid contains a firefly luciferase reporter gene upstream of a multiple cloning site for insertion 3’UTRs that contain predicted miRNA-binding sites in its 3’-UTR. By cloning a cDNA fragment with a miRNA target sequence into the pMIR-REPORT plasmid, expression of the luciferase reporter can be negatively regulated by miRNAs. β-galactosidase expression from the pMIR-REPORT β-gal was used to normalize variability due to differences in cell viability and/or transfection efficiency.
After 48 hours, lysates prepared from HEK293T cells transfected with pMIR-REPORT-Luciferase containing 3’-UTRs from different PEX genes (PEX2, PEX7, PEX11B or PEX13), pMIR-REPORT β-gal together with miRNA mimics were subjected to luciferase and β-gal assays. Briefly, growth medium was removed and cells were rinsed once with PBS. A minimal volume of 1X Reporter Lysis Buffer (RLB) (Promega) was added to each well and then the plates were rocked for several times to ensure complete coverage of the cells with RLB. Cells scraped from the wells were transferred to microcentrifuge tubes and placed on ice for 10 minutes. The microcentrifuge tubes were vortexed for 10–15 seconds and then centrifuged at 12,000 x g for 2 minutes at 4°C. The supernatant/cell lysates were transferred to new tubes and used immediately for assays or stored at -70°C.
For luciferase assays, 20 μl of cell lysate and 100 μl of Luciferase Assay Reagent (Promega) were mixed in microcentrifuge tubes and luminescence was measured using a model Synergy 4 Luminometer (BioTek). For β-Galactosidase assays, 150 μl of cell lysate (2:1 dilution to 1X RLB) was mixed with 150 μl of Assay 2X Buffer (Promega) and then incubated at 37°C for 30 minutes or until a faint yellow color had developed. The reactions were terminated with 1M sodium carbonate (500 μl) after which the absorbances were read at 420 nm. The relative luciferase activity was expressed as a ratio of luciferase activity to β-gal activity.
Transfected or HIV-infected cells grown in 6-well plates were washed twice with cold PBS on ice and then lysed with RIPA buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% Triton x-100, 1% Sodium deoxycholate, 0.1% SDS, 1 mM EDTA) containing a cocktail of protease inhibitors. Lysates were incubated on ice for 30 minutes and then centrifuged at 14,000 x g for 15 minutes at 4°C after which protein concentrations in the supernatants were quantified using a Pierce BCA protein assay kit (Thermo Scientific). Equivalent amounts of total protein (20 μg) were resolved by SDS-PAGE and then transferred to polyvinylidene difluoride membranes (EMD Millipore) membranes for immunoblotting.
Membranes were blocked with 3% skim milk powder in PBS containing 0.1% Tween 20 (PBS-T) and then incubated overnight at 4°C or 3 hours at room temperature with appropriate primary antibodies diluted in 3% milk-PBS-T. After washing three times with PBS-T for 10 minutes each, fluorescent secondary antibodies (donkey anti-mouse IgG conjugated to Alexa Fluor 680 or goat anti-rabbit IgG conjugated to Alexa Fluor 680) diluted in PBS-T were used to detect the primary antibodies. After 1-hour incubation with the secondary antibodies, membranes were washed three times with PBS-T for 10 minutes each. Detection and quantification of the protein signals in the immunoblots was performed using a Licor Odyssey Infrared Imaging System (Lincoln, NE) using the protocol posted at http://biosupport.licor.com. Relative levels of PMP70, PEX2, PEX7, PEX11B, PEX13, PEX19, and catalase (normalized to actin) were determined using Odyssey Infrared Imaging System 1.2 Version software.
Hela CD4+ and A549 cells grown on coverslips were processed respectively for confocal or super-resolution microscopy at 48h post-transfection or infection. Cells were washed in PBS containing 0.5 mM Ca2+ and 1.0 mM Mg2+ and then fixed with 3% paraformaldehyde (for confocal imaging) or 1.5% electron microscopy grade paraformaldehyde (for super-resolution imaging) for 30 min at room temperature. Samples were then quenched with 50mM NH4Cl in PBS for 5 minutes at room temperature, washed three times with PBS, and then permeabilized with 0.2% Triton-X-100 for 5 min. Incubations with primary antibodies diluted (1:500–1000) in blocking buffer (3% BSA in PBS) were performed at room temperature for 2 hours followed by three washes in PBS containing 0.1% BSA. Samples were then incubated with secondary antibodies in blocking buffer for 1 hour at room temperature followed by three washes in PBS containing 0.1% BSA. Secondary antibodies were donkey anti-mouse/rabbit IgG conjugated to Alexa Fluor 488 and donkey anti-mouse IgG conjugated to Alexa Fluor 546 (Invitrogen).
For confocal microscopy, coverslips were mounted onto microscope slides using ProLong Gold antifade reagent with DAPI (Invitrogen), and samples were examined using an Olympus 1x81 spinning disk confocal microscope equipped with a 60x/1.42 oil PlanApo N objective. Confocal images were acquired and processed using Volocity 6.2.1 software.
For super-resolution microscopy, coverslips were mounted on slides pre-cleaned with acetone and ethanol using SlowFade Gold reagent mounting media (Invitrogen). Images were acquired using a DeltaVision OMX V4 structured illumination microscope (Applied Precision, GE) equipped with a 60x 1.42 oil PSF (PlanApo N) objective and immersion oil N = 1.514~1.516. Images were analyzed using Volocity 6.2.1 software.
Z-stack images acquired using a confocal microscope were exported from Volocity 6.2.1 as an OEM.tiff file. The exported images were then processed using Imaris 7.2.3 software (Bitplane). Peroxisomes within polygonal areas that excluded the nucleus were quantified (quality and voxel). Within the selected regions, the absolute intensity/region volume of the peroxisomes were determined and then entered into a Microsoft Excel spreadsheet. The data were then analyzed using student’s t-test.
Where indicated, 0.125 μm optical sections acquired using an Applied Precision OMX super resolution microscope (with a 60X/1.42 Oil lens and three CMOS cameras) were also analyzed. The raw data were processed using Deltavision OMX SI image reconstruction and registration software and the final images were imported into Volocity 6.2.1 software as.dv files for quantification. In each cell, peroxisomes were selected based on the absolute pixel intensity in the corresponding channel and their numbers and volumes were then determined. Only those SKL/PMP70-positive structures with volumes between 0.001 and 0.05 μm3 were included for measurement.
Formalin-fixed paraffin-embedded human brain was processed and tissue sections (10 μm) were prepared and labeled as described us previously [13, 61, 62]. Briefly, samples were deparaffinized by incubation for 1 hour at 60°C followed by one 10 min and 2 five min incubations in xylene baths through decreasing concentrations of ethanol to distilled water. Antigen retrieval was performed by boiling in 10mM sodium citrate (pH 6.0) 1 hr. Slides were blocked with HHFH buffer (1 mM HEPES buffer, 2% (v/v) horse serum, 5% (v/v) FBS, 0.1% (w/v) sodium azide in Hank’s balanced salt solution (HBSS)) for 4 hrs at room temperature. Slides were stained with hematoxylin and eosin (H&E). In addition, serial brain sections were immuno-labelled with antibodies to host proteins. Immunocytochemistry was performed with rabbit anti-Iba-1 (Wako Pure Chemical Industries Ltd., Osaka Japan), anti-thiolase or anti-PEX13 with appropriate secondary antibodies. For immunofluorescence studies, slides were incubated with a cocktail of rabbit anti-GFAP (DAKO, Carpenteria CA) or anti-Iba-1 (1:400) and anti-PEX13, overnight at 4°C. The primary antibodies were removed by three 5 min PBS washes and slides were incubated for three min in 0.22 micron filtered 1% (w/v) Sudan black in 70% ethanol and washed an additional 3 times in PBS. A cocktail of 1:500 Alexa 488 goat anti rabbit IgG, Alexa 568 goat anti mouse IgG for two hrs, washed 3 times in PBS stained with DAPI for 10 min, washed 3 times in PBS and mounted with Prolong gold antifade reagent. Slides were imaged with Wave FX spinning disc confocal microscope (Zeiss).
Total RNA including small RNA from HIV-infected Hela CD4+ cells or primary MDMs was purified using the miRNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. Mature miRNAs, certain small nucleolar RNAs and small nuclear RNAs (snoRNAs and snRNAs) were selectively reverse-transcribed into cDNA using miScript HiSpec buffer according to the instructions of miScript II RT Kit (Qiagen). Mature miRNAs, which are polyadenylated, were reverse transcribed into cDNA using oligo-dT primers. The oligo-dT primers included a 3’ degenerate anchor and a universal tag sequence on the 5’ end, allowing amplification of mature miRNA during the real-time PCR step.
The resulting cDNAs served as the template for real-time PCR analysis using miRNA-specific primers (forward primers, from IDT) and the miScript SYBR green PCR kit (Qiagen), which contains the miScript universal primer (reverse primer) and QuantiTect SYBR green PCR master mix. The amplification cycles consisted of an initial activation step at 95°C for 15 min, followed by 40 cycles of 15s at 94°C, 30s at 55°C and 30 s at 70°C. Fluorescence data were collected during the 70°C extension step. The miRNA targets and primers that were used in this study are listed in Table 2. As an internal control, levels of a small nuclear RNA RNU6B (a miScript PCR control provided in the miScript PCR starter kit (Qiagen)) were determined. Relative miRNA expression was normalized to RNU6B levels using the comparative cT (ΔΔcT) method. All miRNA expression studies were conducted using a Mx3005P (Stratagen, LaJolla, CA) thermocycler.
Microarray data were deposited into the NCBI GEO database (Accession number GSE97611).
|
10.1371/journal.ppat.1001148 | Leishmania-Induced Inactivation of the Macrophage Transcription Factor AP-1 Is Mediated by the Parasite Metalloprotease GP63 | Leishmania parasites have evolved sophisticated mechanisms to subvert macrophage immune responses by altering the host cell signal transduction machinery, including inhibition of JAK/STAT signalling and other transcription factors such as AP-1, CREB and NF-κB. AP-1 regulates pro-inflammatory cytokines, chemokines and nitric oxide production. Herein we show that upon Leishmania infection, AP-1 activity within host cells is abolished and correlates with lower expression of 5 of the 7 AP-1 subunits. Of interest, c-Jun, the central component of AP-1, is cleaved by Leishmania. Furthermore, the cleavage of c-Jun is dependent on the expression and activity of the major Leishmania surface protease GP63. Immunoprecipitation of c-Jun from nuclear extracts showed that GP63 interacts, and cleaves c-Jun at the perinuclear area shortly after infection. Phagocytosis inhibition by cytochalasin D did not block c-Jun down-regulation, suggesting that internalization of the parasite might not be necessary to deliver GP63 molecules inside the host cell. This observation was corroborated by the maintenance of c-Jun cleavage upon incubation with L. mexicana culture supernatant, suggesting that secreted, soluble GP63 could use a phagocytosis-independent mechanism to enter the host cell. In support of this, disruption of macrophage lipid raft microdomains by Methyl β-Cyclodextrin (MβCD) partially inhibits the degradation of full length c-Jun. Together our results indicate a novel role of the surface protease GP63 in the Leishmania-mediated subversion of host AP-1 activity.
| Leishmaniasis is a tropical disease affecting more than 12 million people around the world. The disease is caused by the Leishmania parasites that are transmitted to the mammalian host by a sandfly vector when it takes a blood meal. The parasites are able to survive and multiply inside of cells that comprise the primary defence of the host, the macrophages. We have extensively studied the mechanism whereby Leishmania escapes from macrophage microbicidal functions. Herein we report that the parasite can inactivate these cells by decreasing the activity of transcription factors such as Activated Protein-1(AP-1) that are involved in transcription of genes coding for antimicrobial functions of macrophages. In this study, we showed that Leishmania parasites use their most abundant surface protein GP63 to inactivate the AP-1 transcription factor. Furthermore, we found that GP63 enter into the macrophages independently of parasite internalization using lipid rich microdomains localized in the cellular membrane. In addition, GP63 was observed to reach the nuclear compartment where it cleaves AP-1 subunit proteins. Collectively, our findings reveal a novel mechanism used by Leishmania to facilitate its survival and propagation within its mammalian host cells. Better knowledge concerning the mechanisms whereby this pathogen can escape the innate immune response may help to develop new anti-Leishmania therapy.
| Parasites of the Leishmania genus are the causative agent of leishmaniasis; a disease distributed worldwide affecting more than 12 million people in 88 countries [1]. Leishmaniasis is a complex of diseases ranging from self-healing cutaneous lesions to lethal visceral afflictions [2]. In its mammalian host, Leishmania is an obligate intracellular pathogen infecting hematopoietic cells of the monocyte/macrophage lineage. Macrophages are specialized for the destruction of invading pathogens and priming the immune response. In order to survive within these cells, Leishmania has evolved sophisticated mechanisms to subvert macrophage microbicidal functions such as inhibition of nitric oxide (NO) production and cytokine-inducible macrophage functions [3]. This occurs as the direct consequence of parasite-mediated activation of protein tyrosine phosphatases, alteration of signal transduction and inhibition of nuclear translocation and activity of transcription factors such as NF-κB, STAT, CREB and AP-1[4], [5]. Activated Protein-1 (AP-1) is an important transcription factor that mediates gene regulation in response to physiological and pathological stimuli, including cytokines, growth factors, stress signals, bacterial and viral infections, apoptosis, as well oncogenic responses [6], [7]. AP-1 is formed by homodimers of Jun family members (c-Jun, Jun B and Jun D), or heterodimers of Jun and Fos family members (c-Fos, Fos B, Fra 1 and Fra 2). Homodimers within the Fos family do not occur due to conformational repulsion [8].
Previous studies have reported that the AP-1 transcription factor is inactivated by Leishmania infection. For instance, activation of macrophage AP-1 and NF-κB is inhibited by L. donovani promastigotes through an increase in intracellular ceramide concentration, which leads to the down-regulation of classical PKC activity, up-regulation of calcium independent atypical PKC-ζ and dephosphorylation of Extracellular Signal-Regulated Kinases (ERK) [9], [10]. Other studies have shown that Leishmania alters signal transduction upstream of c-Fos and c-Jun by inhibiting ERK, JNK and p38 MAP Kinases, resulting in a reduction of AP-1 nuclear translocation [11], [12]. However, little is known about the molecular mechanism (s) by which Leishmania parasites are able to inactivate this important transcription factor.
Many Leishmania-specific factors such as lipophosphoglycan (LPG), A2 proteins, cysteine peptidases (CPs) and the protease GP63, contribute to Leishmania virulence and pathogenicity. LPG has been implicated in altering phagosome maturation in L. donovani infection [13].The A2 proteins of L. donovani are involved in intracellular amastigote survival [14].The cysteine peptidases of L. mexicana are implicated in facilitating the survival and growth of the parasite [15]. Furthermore GP63, also known as the major surface protease (MSP), has been related to resistance to complement-mediated lysis, among others [16], [17]. GP63 is a metalloprotease which belongs to the metzincin class. It is the most abundant surface glycoprotein of the parasite and accounts for 1% of the total protein content of L. mexicana promastigotes [18]. GP63 of different Leishmania species encode similar amino acid sequences, although slight substrate specificity variations have been reported [19]. Specific characteristics of this class of metalloproteases include a conserved signature motif HEXXHXXGXXH and an N-terminal pro-peptide that serves to maintain the pro-enzyme inactive during translation, which is removed upon protein maturation and activation [20]. The mature GP63 contains 3 domains: 1) N-terminal (bases ∼101-273) which comprises a structure corresponding to the catalytic module of metzincin class zinc protease, 2) central domain (bases ∼274–391) and 3) C-terminal domain containing the site of glycosylphosphatidylinositol (GPI) anchor addition (bases ∼392–577) [17], [18], [20], [21]. We have previously shown that this protease actively participates in the cleavage of NF-κB [5], protein tyrosine phosphatases (PTP) [22] and actin cytoskeleton regulators [23]. In this study we have investigated how GP63 contributes to the inactivation of AP-1 and the degradation of its subunits. Herein, we report that GP63 enters the host cell via lipid raft microdomains, independently of parasite internalization, and for the first time show that it is able to reach the nuclear compartment shortly after infection where it degrades and cleaves c-Jun and other AP-1 subunits.
We have previously studied the effect of Leishmania promastigote infection on the activity of various macrophage transcription factors: STAT-1α degradation is proteasome and receptor-dependent and is mediated through a mechanism involving PKC-α [4], and cleavage of NF-κB subunits upon Leishmania infection is in part dependent on GP63 [5]. As AP-1 is an important transcription factor regulating the expression of many genes involved in the activation of macrophage functions (TNFα, iNOS, and IL-12) [24], [25], [26] critical for the adequate innate immune response against Leishmania infection, we investigated the mechanisms underlying AP-1 inactivation upon Leishmania infection.
To evaluate nuclear translocation and DNA binding activity of macrophage AP-1 upon infection with Leishmania promastigotes, Electrophoretic Mobility Shift Assays (EMSA) were performed. As shown in Figure 1A, AP-1 nuclear translocation was inhibited as early as 30 min post-infection in L. donovani-infected macrophages. Furthermore, we observed that other mammalian pathogenic Leishmania species (L. mexicana and L. major) were able to alter AP-1 DNA binding (Figure 1B). Of interest, we did not observe any effect on macrophages infected with L. tarentolae, whose pathogenicity is limited to reptilian hosts.
In order to better understand the observed decrease in AP-1 activity, we performed Western Blot (WB) analysis in the total cell extracts to evaluate the various AP-1 subunits during infection. Five out of the seven AP-1 subunits (c-Fos, Fra 1, Fra 2, c-Jun and Jun B) showed decreased expression after infection with L. donovani, whereas Jun D presented a slight reduction and Fos B was maintained intact (Figure 2A). To further confirm the presence of these subunits in the AP-1 complex we used super shift analysis. This approach uses the incorporation of specific antibodies to nuclear protein extracts, allowing the visualization of the antibody: protein: DNA complexes by retarding the migration of the specific bands in the gel. As shown in Figure 2B, inclusion of antibodies specific for Fos B, c-Fos, Fra 1, Fra 2, c-Jun, Jun B and Jun D, demonstrated presence of Fra 1, Fra 2, c-Jun, Jun B and Jun D, but not c-Fos or Fos B, within the macrophage nuclear AP-1 complex. Importantly, L. donovani infection clearly affected the AP-1 complex as the bands observed for Fra-1, Fra-2, c-Jun, Jun B and Jun D in the super shift assay was greatly reduced. Whereas the c-Fos protein was not detectable by super shift assay, this protein was still affected by Leishmania infection since less expression was observed by WB (see Figures 2A and 3), suggesting that the amount of c-Fos might not be enough to be detected by super shift assay.
After phosphorylation AP-1 subunits are translocated into the nucleus where they dimerize with another subunit to form an active AP-1 complex [6], [7], [8], [27]. To determine the level of expression of each AP-1 subunit in the different cellular compartments (cytoplasm vs nucleus), we performed WB analysis on separated nuclear and cytoplasmic fractions. As shown in Figure 3, different phenomena can be observed. c-Fos and Fra-1 expression in the cytoplasmic fraction are not altered with Leishmania infection, but their expression in the nuclear fraction is decreased in infected macrophages; Fra-2 and c-Jun have decreased expression in both cytoplasmic and nuclear fractions, and Fra-2 in the nuclear fraction presents a band with less migration than the band observed in the cytoplasmic fractions, possible due to post-nuclear translocation modifications. On the other hand, Jun-B and Jun-D were detected only in the nuclear fraction; however, only Jun-B expression is affected by Leishmania infection. The lower expression of the different subunits in the nucleus could be due to decreased complex formation and/or cleavage and further degradation of the subunits, as it is possible to detect smaller bands (c-Jun and Jun-B).
Leishmania surface molecules such as LPG and GP63, among others, play important roles as virulence factors and modulators of host cell signalling. LPG, for instance, has been implicated in the interference of phagolysosome maturation and inactivation of PKC signalling [13], [28]. GP63 has been related to resistance to complement-mediated lysis, migration of Leishmania parasites through the extracellular matrix by degradation of casein, fibrinogen and collagen [16], [21] and inhibition of JAK/STAT signalling by modulation of PTP activities [22]. To address the role of LPG and GP63 in AP-1 inactivation we performed EMSA with extracts from cells infected with Leishmania mutants for these two surface molecules. As shown in Figure 4A, LPG is not involved in the AP-1 degradation induced by Leishmania infection since DNA binding in macrophages infected with either L. donovani or L. donovani LPG−/− promastigotes was similarly altered. Importantly, however, we observed that cells infected with an L. major strain lacking GP63 (L. major GP63−/−) [29] showed normal AP-1 DNA binding capacity, compared to uninfected controls. This suggests that GP63 but not LPG is highly involved in the mechanism responsible for the inactivation of AP-1 transcription factor.
To further elucidate the role of GP63, we performed WB analysis of all the AP-1 subunits of macrophages infected with L. major, L. major GP63−/− and L. major GP63 Rescued. Results obtained further revealed that in the absence of GP63, no degradation or cleavage of any AP-1 subunit was evident (Figure 4B); supporting the finding that AP-1 activity is unaffected in L. major GP63−/−-infected macrophages.
In addition, to validate the role of GP63 on AP-1 activity we verified the expression of IL-12 transcripts, as it is known that AP-1 regulates its transcription [24]. As shown in the Figure S1, LPS-induced IL-12 expression is fully blocked by all infectious Leishmania species but not by L. major GP63−/− and L. tarentolae. As expected, the JNK/c-Jun inhibitor has completely inhibited LPS-induced IL-12 transcripts.
Leishmania GP63 can be found in three different forms: 1) Intracellular GP63, 2) Surface GPI-anchored GP63 and 3) secreted or released GP63 [21], [30]. For GP63 to target its intracellular macrophage substrates, it needs to gain access to or be internalized by the macrophage. To explore whether the internalization of the parasite is necessary to deliver GP63 inside the cell, murine macrophages were pre-treated with the phagocytosis inhibitor cytochalasin D which inhibits actin polymerization, therefore blocking internalization by phagocytosis (Figure S2A). We used c-Jun as a model protein to evaluate the cleavage and degradation of the AP-1 subunits. WB analysis showed that parasite phagocytosis was not necessary for c-Jun cleavage and less expression (Figure 5A). To confirm this, we incubated macrophages with the culture supernatant of L. mexicana promastigotes, which is rich in soluble GP63 [18], [31]. WB showed that even in the absence of the parasite, c-Jun degradation was observed (Figure 5B).
Since phagocytosis seems not to be completely required in the internalization of GP63 we addressed whether GP63 internalization could be dependent on lipid raft-mediated endocytosis, given the fact that GP63 is an excreted and membrane-GPI anchored protein. On the other hand, lipid raft microdomains are highly dynamic membrane domains rich in cholesterol and sphingolipids, and present high affinity for proteins containing GPI anchors [32], [33], [34]. In order to examine the possible role of host lipid raft microdomains in GP63 internalization, we pre-treated cells with a non-cytotoxic dose (Figure S2B) of the cholesterol chelator and inhibitor of lipid raft integrity methyl-β-clyclodextrin (MβCD) prior to infection. As shown in Figure 5C, full length c-Jun was not degraded in cells infected under these conditions, although interestingly, a cleavage fragment was still observed. Pre-treatment of macrophages with MβCD and subsequent incubation with L. mexicana supernatant showed that lipid raft disruption altered internalization of parasite-free soluble GP63 and also impaired c-Jun degradation (Figure 5D). As shown in Figure S3, confocal microscopy confirmed an interaction between lipid raft microdomains and GP63, since in macrophages infected with L. major, GP63 (green) partially co-localized with the lipid raft marker Choleratoxin B (red). Furthermore, we have previously shown that, pre-treatment with MβCD before infection abrogates GP63 internalization [22]. To determine if MβCD had any effect over the c-Jun expression we performed a time course analysis of macrophages stimulated with MβCD. As shown in Figure S4A, there was no alteration in the expression of c-Jun after 2 hr of incubation of the macrophages with the drug. Together these data strengthen the hypothesis that GP63 uses lipid raft microdomains for internalization independent of parasite entry.
To evaluate the role of the GPI anchor in mediating GP63 internalization via lipid raft microdomains, macrophages were incubated with a GPI-deficient recombinant GP63 (rGP63) and c-Jun degradation was monitored. WB analysis evidenced that neither degradation nor cleavage of c-Jun occurred (Figure 5E) in the presence or absence of MβCD, similarly to what we have previously shown for GP63-mediated PTP cleavage. Moreover, although rGP63 is still internalized in macrophages to a limited extent, perinuclear localization was never detected [22].
To demonstrate that the less expression and cleavage of c-Jun observed in this set of experiments were occurring inside the cells and not as an effect of proteolysis during the preparation of the lysates, we included two experiments as controls; first, we lysed the cells using sample loading buffer 1× and the samples were boiled right after, to stop the proteolysis; second, we added 1 mM of phenanthroline (a Zn chelator [35]) to the lysis buffer to abrogate post-infection GP63 activity. In both experiments, we observed that cleavage of c-Jun under these conditions still occurs, suggesting that the cleavage of c-Jun occurs inside the cell and not during the sample preparation (Figure S4B and S4C). In addition, to establish whether GP63 proteolytic activity is critical for c-Jun cleavage in the macrophage, L. mexicana culture supernatant was treated with the GP63 inhibitor phenanthroline prior to its incubation with macrophages. As shown in the Figure S4D, phenanthroline fully inhibited GP63-mediated c-Jun degradation.
One of the most surprising elements of the evidence presented above was the fact that GP63 is able to act on its substrate proteins within the nucleus of its host cell. In order to further demonstrate that GP63 reaches the nucleus, we separated cytoplasmic and nuclear proteins from macrophages infected with L. major, L. major GP63−/− and L. major GP63 Rescued. WB analysis using an anti-GP63 antibody revealed that this protease is present in both fractions of Leishmania-infected cell extracts. As expected, there was no GP63 in macrophages infected with L. major GP63 −/− or the uninfected control (Figure 6A). Confocal microscopy of Leishmania-infected macrophages confirmed that GP63 reaches the nuclear membrane as early as 1 hr post-infection (Figure 6B). In order to demonstrate the purity of our fractions, we performed WB of the cytoplasmic and nuclear proteins against the lysosomal marker LAMP-1, the ER specific marker (the KDEL protein - Lys-Asp-Glu-Leu endoplasmic reticulum protein retention receptor), histone 2B (nuclear marker), and actin (cytoplasm marker). Figure S5 shows that actin, LAMP-1 and KDEL are only present in the cytoplasmic fraction, in contrast, histone is only detected in the nuclear fraction, and this way we are confident to say that GP63 was present in both protein fractions.
In order to confirm nuclear interaction of GP63 with c-Jun we performed a Co-Immunoprecipitation (IP) assay. c-Jun was immunoprecipitated from nuclear extracts of Leishmania-infected macrophages and subjected to WB analysis of GP63. This result revealed a band around 65 kDa, confirming the interaction between nuclear c-Jun and GP63 in the macrophages infected with L. major and L. major GP63 Rescue, but not with L. major GP63 −/− as is shown in Figure 7.
To further support that degradation of c-Jun could occurs in the nucleus we performed confocal microscopy. As shown in Figure 8A (upper panel), c-Jun (red) is localized inside the nucleus in uninfected cells. However, after 1 hr of infection GP63 was detected in the perinuclear area and the fluorescence intensity of c-Jun was considerably diminished (Figure 8A, lower panel), such reduction in the fluorescence was not observed in macrophages infected either with L. major GP63−/− or L. tarentolae (Figure 8C upper and lower panels, respectively). The upper panel of Figure 8B shows partial co-localization between the nuclear stain (blue) and GP63 (green) in the periphery of the nucleus, giving a light blue signal. Of utmost importance, the panel representing c-Jun (red) versus nucleus (blue) co-localization, clearly reveals that c-Jun is absent from perinuclear area as this one is solely stained in blue (Figure 8B, lower panel). To discard possible unspecific signals in the confocal micrographs we included specific isotype and secondary antibody controls (Figure S6B). Collectively, our results suggest that GP63 reaches the perinuclear area of the cell shortly after macrophage-parasite contact occurred leading to degradation and cleavage of various members of AP-1 subunits, leading to its inability to dimerize and bind DNA and therefore, altering AP-1 transcriptional activity on genes under its regulation.
To further understand the direct effect of GP63 on c-Jun, we used a purified GST tagged-c-Jun protein and incubated it with Leishmania promastigotes of different species (including L. donovani, L. mexicana, L. major, L. major GP63−/− and L. major GP63 Rescued). WB analysis showed that direct contact of parasites expressing GP63 and c-Jun protein is sufficient to induce c-Jun degradation (Figure 9A). This was corroborated by the reduction of c-Jun degradation when incubated with the GP63−/− strain. Figure S7 shows that L. tarentolae has no effect on the degradation of GST-c-Jun.
GP63 recognizes a four amino acid motif in its target protein substrates based on amino acid characteristics: polar/hydrophobic/basic/basic amino acids (P1- P′1-P′2-P′3) [19]. Sequence analysis of AP-1 subunits revealed putative cleavage sites within c-Jun, Jun B, Jun D and c-Fos (Figure 9B). Of interest, one of the sequence-identified cleavage sites of c-Jun was found between the leucine zipper and the DNA binding domain as shown in Figure 9C. In addition, this motif is found at amino acids 271-275, which will generate cleaved fragments with molecular weight similar to the one detected by WB of lysates from infected cells (∼30 kDa).
Collectively, these data indicate that the Leishmania protease GP63 actively participates in altering the DNA binding capacity of AP-1 as a consequence of the diminished expression and cleavage of its subunits. These data further corroborate a mechanism whereby GP63 can enter the cell using lipid raft microdomains, and show for the first time that GP63 reaches the perinuclear area where it proteolytically degrades AP-1 subunits.
Leishmania parasites have evolved many mechanisms to undermine macrophage signalling pathways in order to survive and replicate inside these cells. For instance, parasite-mediated activation of macrophage PTPs leads to protein dephosphorylation resulting in the inactivation of transcription factors controlling the expression of many genes required for the effective activation of the innate immune response [36], and macrophage effector functions such as NO production [37]. We have previously reported that Leishmania promastigote infection induces degradation and inactivation of some transcription factors. For example, STAT 1 is inactivated by a proteasome mediated mechanism [4], and NF-kB activity is altered in a cleavage-dependent fashion [5]. We show that cleavage of p65 generates an active fragment, p35, which is able to translocate into the nucleus, where it dimerizes with p50 to induce specific chemokine gene expression. Interestingly this cleavage event was found to occur in the macrophage cytoplasm in a GP63-dependent mechanism [5].
Along with STAT and NF-κB, AP-1 is responsible for the transcription of iNOS [38]. NO is a by-product of iNOS-mediated conversion of L-arginine to L-citruline and is essential for the control of Leishmania infection [3], [36]. Among other genes regulated by AP-1 in macrophages and known to be affected by Leishmania infection are TNFα, IL-1β and IL-12 [24], [25], [26], [39]. The breadth and importance of the immunological functions of AP-1 highlights how detrimental its degradation is to host defence against Leishmania infection. Previously, Descoteaux and Matlashewski (1989) demonstrated that the c-fos gene, one of the main activators of AP-1, was down-regulated due to abnormal PKC signalling [40]. More recently Ghosh and colleagues (2002) reported Leishmania-dependent inactivation of both AP-1 and NF-κB in a ceramide dependent mechanism, where increased levels of intracellular ceramide conducted to the down-regulation of classical PKC activity and impartment of the phosphorylation of ERK, which results in decreased AP-1 activation [10]. These previous reports have given some indication of AP-1 inactivation by Leishmania. Here we further demonstrated the molecular mechanisms involved in the AP-1 inactivation by Leishmania parasites and its impact on IL-12 expression.
We have found that infection with several Leishmania species alters the DNA binding capacity of AP-1. In particular, we have shown that the parasite metalloprotease GP63 is responsible for this DNA binding alteration and is able to induce the degradation/down-regulation and cleavage of c-Jun, the central component of the AP-1 transcription factor [41], as well of other components including c-Fos, Fra-1, Fra-2 and Jun B. We provide evidence that GP63 exerts its effect by its internalization into the host cell, in a mechanism that is independent of parasite internalization, and induces AP-1 proteolysis within the nucleus or in the nuclear membrane.
The present study corroborates along with previous studies (Ghosh et. al) that AP-1 is down-regulated by Leishmania parasites. Alteration of AP-1 activity varies according to the pathogen, for instance it has been shown, that the hepatitis C virus alters MAP kinases and AP-1 to accelerate the cell cycle progression, helping the development of hepatocellular carcinoma and HCV development [42]. Another example is the Edema toxin produced by Bacillus anthracis; this toxin is able to up-regulate macrophage gene expression, among them genes that are known to be involved in inflammatory responses, regulation of apoptosis, adhesion, immune cell activation, and transcription regulation. Interestingly this up-regulation was found to correlate with induced activation of AP-1 and CAAAT/enhancer-binding protein-beta [43]. In contrast with these reports where different pathogens up-regulate AP-1 to survive inside the host cell, herein we have shown how this transcription factor is down-regulated after Leishmania infection in a cleavage-dependent manner. Whether AP-1 down-regulation is a general mechanism used by different intracellular protozoa requires further investigation.
The AP-1 transcription factor is formed by dimers of Jun and Fos family members. In addition, the Jun proteins can dimerize with other proteins that share the leucine zipper region such as ATF-1 and ATF-2 [8], [27]. Although we did not test other non-classical AP-1 subunits, we demonstrated that at least 5 of the classical subunits belonging the Jun and Fos families are degraded by the parasite within 1 hr of infection. Of interest, c-Jun subunit, one of the main activators of AP-1 along with c-Fos, is cleaved generating a GP63-mediated 30 kDa fragment. The cleaved product would be unable to dimerize and bind DNA, as it has been demonstrated that truncated c-Jun deprived of either the leucine zipper or the DNA binding domain results in only marginal AP-1 transactivation [41], [44], [45]. The generation of c-Jun fragment by GP63 can explain the lower AP-1 binding activity observed in the EMSA experiment. Furthermore, Fos B, which is not cleaved or degraded and also apparently absent in AP-1 complexes (Figure 2B), lacks putative GP63 cleavage sites. Surprisingly Jun D presents two putative sites of cleavage by GP63. However, we did not detect either complete degradation/down-regulation or cleavage products. One possible explanation is that the structural conformation of this protein renders these sites unavailable for GP63-Jun D interaction.
GP63 is known to interact with various substrates. For instance we have recently reported that Leishmania GP63 impacts the stability of cortactin and caspase-3, and negatively regulates p38 kinase activity [23]. Furthermore, we have shown that GP63 cleaves host PTPs resulting in enzymatic activation and leading to JAK 2 dephosphorylation, and inhibition of NO production in IFN-γ primed and infected macrophages [22]. Our current study further supports the important role of GP63 as a negative regulator of host cell functions, actively participating in the pleiotropic effects excreted by Leishmania parasite to suppress the immune response, our results showed that internalization of GP63 by cells from innate immune response is independent of parasite internalization, as our data revealed GP63 proteolytic activity was not affected by inhibition of phagocytosis, but clearly abolished by a lipid raft disruptor, strongly suggesting that lipid rafts microdomains are important for internalization of GP63. Proteins that have a GPI anchor have affinity for lipid rafts, and it has been reported that these rafts recognize these GPI anchors allowing the entrance of GPI-bearing proteins in endocytic vesicles [33], [34]. In addition, Brittingham and collaborators showed interaction of GP63 with the fibronectin receptor (α4β1), that also translocate into the lipid rafts microdomains [46], suggesting that GP63 could have two different ways to get access into the cell: 1) GPI-anchor (native and excreted) and 2) receptor mediated (RGP63). Additionally, we have shown that the GPI anchor is important for the internalization of GP63 since recombinant GP63 (rGP63) lacking the GPI anchor is less internalized [22]. Most importantly, GPI anchor seems to be required for the cellular localization of GP63 since rGP63 is localized inside intracellular compartments whereas GPI-GP63 (native protein) is found within nuclear membrane (see Figure 6). Despite this evidence we have not excluded the possibility that GP63 could use other mechanisms to enter the cells, such as micropinocytosis or classical endocytosis pathways.
One of the main finding of this research is the macrophage nuclear localization of GP63. One plausible mechanism is by the recognition of its GPI domain by the recently described lipid microdomains rich in cholesterol and sphingolipids in the nuclear membrane [47]. Another possible mechanism for the internalization of GP63 inside the nucleus is the presence of a nuclear localization signal (NLS)-like motif (Figure S8) in the GP63 sequence. Nuclear proteins are usually transported inside the nucleus by recognition of a NLS, which usually consist of short chains of basic amino acids with the signature motif K-K/R-X-K/R [48]. These NSLs are recognized by the adaptor molecule importin α, which forms a hetorodimer with the transporter receptor importin β. The importin α/β-NSL-cargo complex is then translocated through the nuclear pore complex [48], [49]. The exact mechanisms of how the nuclear translocation of GP63 occurs are currently under investigation.
In summary, GP63 seems to preferentially target AP-1 subunits within the nuclear membrane, altering its DNA binding capacity. Given the critical role of this transcription factor in the transcription of several genes involved in the innate immune response, alterations in AP-1 activity can dramatically contribute to the down-regulation of innate immune functions observed during the early stages of Leishmania infection. Therefore, this novel mechanism of evasion by Leishmania further demonstrates the complex negative regulatory mechanisms developed by the parasite, which has permitted its adaptation to the harsh intracellular environment leading to its survival and propagation within its mammalian host.
Immortalised murine bone marrow derived macrophages B10R cell line were maintained at 37°C in 5% CO2 in Dulbecco's Modified Eagle medium (DMEM) supplemented with 10% heat inactivated FBS (Invitrogen, Burlington, ON, Canada) and 100 U/ml penicillin 100 µg/ml streptomycin and 2 mM of L-glutamine (Wisent, St-Bruno, QC, Canada). Leishmania promastigotes (L. donovani infantum, L. donovani 1S2D, L. donovani R2D (LPG −/−), L. mexicana, L. major A2 (WT), L. major GP63 −/−, L. major GP63 Rescued [29] and L. tarentolae) were grown and maintained at 25°C in SDM-79 culture medium supplemented with 10% FBS by bi-weekly passage. Macrophages were infected at parasite to macrophage ratio 20∶1 with stationary phase promastigotes for the times specified in each Figure legend. Using this ratio of infection we normally observe around 30% and 60% of infected cells in 1 or 2 hr, respectively. When chemical inhibitors were used, 2 µM cytochalasin D (Sigma-Aldrich, St-Louis MO, USA) and 20 mM Methylβ-cyclodextrin (MβCD) (Sigma-Aldrich, St-Louis MO, USA), cells were treated 1 hr prior to infection and the inhibitor remained throughout the infection time.
B10R macrophages (2×106) were infected, washed three times with Phosphate Buffered Saline (PBS) to remove non-internalized parasites, and processed for nuclear extraction as previously described [4], [50]. Briefly, macrophages were collected in 1 ml of cold PBS, centrifuged and pellets were resuspended in 400 µl of ice-cold buffer A (10 mM HEPES, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT and 0.5 mM of PMSF) and incubated 15 min on ice. Twenty five µl of IGEPAL 10% (Sigma-Aldrich, St-Louis MO, USA) were added, and samples vortexed for 30 sec. Nuclear proteins were pelleted by centrifugation and resuspended in 50 µl of cold buffer C (20 mM HEPES, 400 mM NaCl 1 mM EDTA, 1 mM EGTA 1 mM DTT and 0.5 mM PMSF).
Protein concentrations were determined by Bradford assay (Bio-Rad, Hercules CA, USA). 6 µg of nuclear proteins were incubated for 20 min at room temperature with 1 µl of binding buffer (100 nM Hepes pH 7.9, 8% v/v glycerol, 1% w/v Ficoll, 25 mM KCl, 1 mM DTT, 0.5 mM EDTA, 25 mM NaCl, and 1 µg/µl BSA) and 200 ng/µl of poly (dI-dC), 0.02% bromophenol blue and 1 µl of γ-P32labeled oligonucleotide containing a consensus sequence for AP-1 binding complexes (5′-CGTTTGATGACTCAGCCGGAA-3′) (Santa Cruz Biotechnology Inc, Sta Cruz CA, USA). After incubation, DNA-protein complexes were resolved by electrophoresis in non-denaturing polyacrylamide gel 5% (w/v). Subsequently gels were dried and autoradiographed. Competition assays were conducted by adding a 100-fold molar excess of homologous unlabeled AP-1 oligonucleotide, or the non-specific competitor sequence for SP-1 binding (5′-ATTCGAATCGGGGCGGGGCGAGC-3′).
For supershift assay, 2 µg of nuclear protein extract were incubated for 1 hr at room temperature with binding buffer, poly (dI-dC), 0.02% bromophenol blue, labeled oligonucleotide and 4 µg of individual specific antibodies (α-c-Jun, Jun B, Jun D, c-Fos, Fos B, Fra 1 or Fra 2; Santa Cruz Biotech Inc, Sta Cruz CA, USA). Complexes were resolved on standard non-denaturing polyacrilamide gel 5% (w/v).
Infected and non infected cells (1×106) were washed 3 times with PBS and lysed with cold buffer (50 mM Tris (pH 7), 0.1 mM, 0.1 mM EGTA, 0.1% 2-mercaptoethanol, 1% NP-40, 40 µg/ml aproptinin and 20 µg/ml of leupeptin). Proteins were dosed by Bradford (Bio-Rad, Hercules CA, USA), and 30 µg of proteins were separated by SDS-PAGE, and transferred onto PVDF membranes (GE healthcare, Piskataway NJ, USA). Membranes were blocked in 5% non-fat dry milk, washed and incubated for 1 hr with α-c-Jun (BD Biosciences, San Jose, CA, USA), α-Jun B, α-Jun D, α-c-Fos, α-Fos B, α-Fra 1 and α-Fra 2 (Santa Cruz Biotech Inc. Sta Cruz CA, USA). After washing, membranes were incubated 1 hr with α-mouse or α-rabbit HRP-conjugated antibody (GE healthcare, Piskataway NJ, USA), and developed by autoradiography.
B10R macrophages (10×106) were infected with either L. major A2, or L. major GP63 −/− or L. major GP63 Rescued for 1 hr, and nuclear proteins were extracted as previously described in [5]. c-Jun was immunoprecipitated from pre-cleared nuclear extracts with 2 µg antibody, followed by addition of 12.5 µl (packed volume) of protein A/G PLUS agarose beads (Santa Cruz Biotech Inc. Sta Cruz CA USA). Beads were washed three times and bound proteins were analyzed by WB as described above.
B10R macrophages (0.5×106) were plated ON in glass cover slips. After infection for 30, 60 and 180 min cells were gently washed 3 times with PBS, and then fixed with 4% p-formaldehyde for 30 min at 4°C. Slides were permeabilized for 5 minutes with PBS containing 1% BSA and 0.05% NP-40 and blocked with 5% non-fat dry milk for 1 hr. Incubation with primary antibody α-c Jun or α-c-Fos or α-GP63 mouse monoclonal antibody clone #96 [51] was conducted in humid dark chamber for 1 hr at room temperature. After three washes with PBS, cells were incubated with secondary antibody (Alexa Fluor 488 or 594, from Molecular probes, Burlington ON, Canada) for 1 hr. Nuclei were stained with propidium iodide or DAPI for 10 min and slides were mounted in permaflour medium (Thermo Co. Waltham MA, USA). Images were taken using an Olympus FV1000 confocal microscope and a Zeiss LCS 500.
B10R macrophages (10×106) were infected with either L. major A2, L. major GP63−/−, L. major GP63 Rescued or L. tarentolae for 18 hr or treated with 20 µM of JNK inhibitor SP600125 for 1 hr. Thereafter cells were stimulated with 100 ng/ml of LPS for 6 hr and RNA extracted using TRIzol reagent according to the manufacturer's instruction (Invitrogen Canada). Reverse transcriptase was performed using oligo(dT) primers. Quantitative relative PCR was performed using Invitrogen Platinum qPCR Super-Mixes and 0.4 µM primer in 25 µl and the following parameters: 50°C for 2 min and 95°C for 3 min (95°C for 20 s, 60°C for 30 s, 72°C for 20 s, 80°C (reading step) for 20 s) for 40 cycles followed by a melting curve. Annealing temperature was 60°C. IL-12 primer sequences: 5′-GGA AGCACG GCA GCA GAA TA-3′ and 3′-AAC TTG AGG GAG AAG TAGGAA TGG-5′.
|
10.1371/journal.pntd.0001285 | Efficacy and Safety of Artemether in the Treatment of Chronic Fascioliasis in Egypt: Exploratory Phase-2 Trials | Fascioliasis is an emerging zoonotic disease of considerable veterinary and public health importance. Triclabendazole is the only available drug for treatment. Laboratory studies have documented promising fasciocidal properties of the artemisinins (e.g., artemether).
We carried out two exploratory phase-2 trials to assess the efficacy and safety of oral artemether administered at (i) 6×80 mg over 3 consecutive days, and (ii) 3×200 mg within 24 h in 36 Fasciola-infected individuals in Egypt. Efficacy was determined by cure rate (CR) and egg reduction rate (ERR) based on multiple Kato-Katz thick smears before and after drug administration. Patients who remained Fasciola-positive following artemether dosing were treated with single 10 mg/kg oral triclabendazole. In case of treatment failure, triclabendazole was re-administered at 20 mg/kg in two divided doses.
CRs achieved with 6×80 mg and 3×200 mg artemether were 35% and 6%, respectively. The corresponding ERRs were 63% and nil, respectively. Artemether was well tolerated. A high efficacy was observed with triclabendazole administered at 10 mg/kg (16 patients; CR: 67%, ERR: 94%) and 20 mg/kg (4 patients; CR: 75%, ERR: 96%).
Artemether, administered at malaria treatment regimens, shows no or only little effect against fascioliasis, and hence does not represent an alternative to triclabendazole. The role of artemether and other artemisinin derivatives as partner drug in combination chemotherapy remains to be elucidated.
| Fasciola hepatica and F. gigantica are two liver flukes that parasitize herbivorous large size mammals (e.g., sheep and cattle), as well as humans. A single drug is available to treat infections with Fasciola flukes, namely, triclabendazole. Recently, laboratory studies and clinical trials in sheep and humans suffering from acute fascioliasis have shown that artesunate and artemether (drugs that are widely used against malaria) also show activity against fascioliasis. Hence, we were motivated to assess the efficacy and safety of oral artemether in patients with chronic Fasciola infections. The study was carried out in Egypt and artemether administered according to two different malaria treatment regimens. Cure rates observed with 6×80 mg and 3×200 mg artemether were 35% and 6%, respectively. In addition, high efficacy was observed when triclabendazole, the current drug of choice against human fascioliasis, was administered to patients remaining Fasciola positive following artemether treatment. Concluding, monotherapy with artemether does not represent an alternative to triclabendazole against fascioliasis, but its role in combination chemotherapy regimen remains to be investigated.
| Fascioliasis, a zoonotic disease caused by a liver fluke infection of the species Fasciola hepatica and F. gigantica, is of considerable veterinary and public health importance [1], [2]. Owing to global changes, infections with Fasciola spp. appear to be emerging or re-emerging in several parts of the world [1]. An estimated 91 million people are at risk of fascioliasis, whereas the estimated number of infections shows a large range from 2.4 to 17 million [3]. Severe clinical complications in the chronic phase of a Fasciola infection include cholangitis, cholecystitis, jaundice, and biliary colic [1], [4].
In Egypt, fascioliasis is an important clinical problem, particularly among school-aged children living in rural areas of the Nile Delta [5], [6]. Prevalence rates of Fasciola infections have been reduced in recent years, explained by control measures put forth by the Egyptian governorates, including triclabendazole administration [6]. Indeed, chemotherapy with triclabendazole, a member of the benzimidazole family of anthelmintics, is the current mainstay for morbidity control of fascioliasis [7]. It should be noted, however that triclabendazole is often difficult to obtain, since it is currently registered in only four countries for human treatment [7]. In addition, resistant fluke populations have been reported from several countries [7]–[9]. Unfortunately, no vaccine is currently available for prevention of fascioliasis [10].
There is a need to develop new fasciocidal drugs. Several studies have documented that the artemisinins (e.g., artemether and artesunate), which have become the most important antimalarial drugs, particularly when deployed as artemisinin-based combination therapy (ACT) [11], also possess schistosomicidal [12] and fasciocidal activities [13]. Regarding fascioliasis, complete elimination of worms was achieved in rats experimentally infected with adult F. hepatica when artesunate and artemether were administered at single oral doses (400 and 200 mg/kg, respectively) 8 weeks postinfection [14]. Severe tegumental changes and death of flukes occurred when Fasciola spp. were incubated with an artemisinin derivative (50–100 µg/ml) in vitro [14]–[17]. Artesunate and artemether, given by the intramuscular route, yielded high egg and worm burden reductions in natural F. hepatica infections in sheep [18], [19]. Finally, a study in 100 Vietnamese patients has shown that artesunate might also play a role in the treatment of acute fascioliasis, as patients treated with artesunate were significantly more likely to be free of abdominal pain when compared to triclabendazole-treated patients [20].
The aim of the present study was to assess the efficacy and safety of oral artemether, adhering to two different malaria treatment regimens [21], [22], in patients with a chronic Fasciola spp. infection. The study was carried out in a Fasciola-endemic area of Egypt, where Schistosoma mansoni co-exists, but malaria is absent.
Ethical clearance was obtained from the Theodor Bilharz Research Institute (Giza, Egypt), the Ministry of Health and Population (Cairo, Egypt), and the Ethics Committee of Basel, Switzerland (EKBB, reference no. 54/07). The trial is registered with Current Controlled Trials (reference no. ISRCTN10372301). Written informed consent was obtained from eligible study participants or parents/legal guardians from individuals aged below 16 years.
The study was designed as an interventional, open-label, non-randomized, proof-of-concept trial, consisting of two separate single-arm studies, to evaluate the efficacy and safety of two artemether regimens in the treatment of asymptomatic Fasciola-infected patients. Twenty individuals were assigned to each study, following recommendations for pilot studies of at least 12 patients per treatment [23] and sufficient number of patients who might not comply to follow-up.
The primary end points were cure rate (CR, defined as percentage of patients who became Fasciola egg-negative after treatment, who were egg-positive at study enrollment) and egg reduction rate (ERR, defined as reduction of geometric mean (GM) egg output after treatment divided by the GM of the same individuals before treatment, multiplied by a factor 100) of Fasciola infection, 28 days after the final dosing. Incidence of adverse events, monitored up to 2 days after the final dosing, was used as secondary outcome measure.
Paticipants who remained Fasciola positive following artemether treatment were orally treated with a single 10 mg/kg dose of triclabendazole. Efficacy of triclabendazole was determined in the frame of the second intervention study. Patients who were still found with Fasciola eggs in their stool following 10 mg/kg triclabendazole were treated with 20 mg/kg triclabendazole in two divided doses.
Study 1 was carried out between April and July 2007 in El-Haddad El-Bahary village, Behera governorate, north-east of Delta. El-Haddad El-Bahary village is s a typical rural setting, with canals fed from the Nile River and no access to the Mediterranean. The total population in the village is 8144.
Study 2 was conducted between August 2008 and May 2010 in Abis village, located south-west of Alexandria. It comprises 10 sub-villages, with an estimated total population of 35,000. Abis village is fed by water canals drawn from the Nile River, with no access to the Mediterranean.
Artemether, formulated as 40 mg capsules (study 1) and 50 mg tablets (study 2) was purchased from Kunming Pharmaceutical Cooperation (Artemidine®; Kunming, People's Republic of China). The following two treatment schemes were investigated: (i) 6×80 mg over 3 consecutive days (study 1) and (ii) 3×200 mg within 24 h (study 2). Treatment was supervised by a physician with date and precise time of drug administration recorded. Patients were observed for 1 h to ensure retention of medication. In case of vomiting or any treatment-related adverse events, a second dose of artemether was administered.
Triclabendazole (Egaten® 250 mg tablets, scored tablets) was the product of Novartis (Basel, Switzerland). Patients who failed to become Fasciola egg-negative following artemether administration received 10 mg/kg triclabendazole. The triclabendazole dosage, according to the patients' weight, was calculated in half-tablet increments with a maximum of 2.5 tablets (625 mg). In case of triclabendazole treatment failures (assessed in study 2), patients were provided two doses of 10 mg/kg of triclabendazole given on subsequent days according to manufacturer's instructions.
Several weeks before conducting a parasitological baseline survey, the health directorate of Beheira (study 1) and Alexandria governorate (study 2) were informed about the objectives, procedures, and potential risks and benefits. After written informed consent was obtained, participants were asked to provide a stool sample in order to screen for the presence of F. hepatica and/or F. gigantica eggs. Stool collection containers were labeled with patient's name and a unique identifier (ID). Filled containters were transfered to a laboratory for diagnostic work-up. Two additional stool samples were collected on consecutive days among participants who were found with Fasciola eggs in their feces. In addition, a blood sample was collected before drug administration to examine hematologic parameters, liver, and kidney functions.
At enrollment a full clinical examination was carried out to assess participants' general health status. Exclusion criteria were: (i) age below 5 years, (ii) pregnancy, (iii) major systemic illnesses (e.g., history of chronic illness such as cancer, diabetes, hypertension, chronic heart, liver or renal disease, severe liver disease of other etiology), and (iv) recent history of anthelmintic treatment (e.g., albendazole, bithionol, dehydroemetine, mebendazole, praziquantel, and triclabendazole taken within the past 4 weeks). Patients meeting our inclusion criteria were treated with artemether, which was administered over 3 consecutive days (study 1) or within 24 h (study 2).
Adverse events were monitored on each treatment day and for 24–48 h following the final dosing. Participants were asked to report any potential drug-related signs and symptoms using a standardized questionnaire. Full clinical examinations were performed on all participants. Adverse events were graded (i.e., mild, moderate, severe, and serious) and recorded. Therapy was offered to patients presenting with adverse events, as judged by the study physician.
Five and 28 days posttreatment, blood samples were collected for clinical chemistry analyses. The final parasitological assessment started on day 28 posttreatment: stool samples were obtained from all study participants over 5 consecutive days. Patients found with Fasciola eggs in their stool following artemether administration were treated with 10 mg/kg triclabendazole. In study 2, stool samples were collected from triclabendazole-treated patients 28 days posttreatment over 3 consecutive days and CRs and ERRs were determined. Those patients who remained Fasciola positive were retreated with a double dose of triclabendazole (20 mg/kg given 24 h apart) [24] and efficacy (CRs and ERRs) was assessed 28 days posttreatment, on the basis of three stool samples. In both groups of triclabendazole-treated patients, liver and renal function and hematological parameters were determined pre- and posttreatment (5 and 28 days after drug administration).
For detection and quantification of Fasciola eggs, all stool samples were processed shortly after collection using the Kato-Katz technique [25]. From each stool sample, 3–6 thick smears were prepared on microscope slides. The slides were transported in enumerated boxes to the Theodor Bilharz Research Institute and examined within a maximum of 48 h. The presence of S. mansoni and soil-transmitted helminths (i.e., Ascaris lumbricoides and Trichuris trichiura) was also determined and recorded for each participant individually. Each slide was examined independently in a blind manner by two microscopists. For quality control, several slides were re-examined by a senior staff. For confirmation of Fasciola and other helminth eggs, at baseline the merthiolate-iodine formaldehyde (MIF) concentration technique [26] was employed for one stool sample per participant. Briefly, 2.35 ml of stock MIF solution was added to at least about 0.5 g of each stool sample in a 15 ml centrifuge tube, closed with a rubber stopper, and placed in a refrigerator for subsequent examination. On the next morning 0.15 ml of Lugol's iodine solution was added to each tube. After centrifugation, the upper layers of sedimented feces containing parasite material were examined under a microscope.
Laboratory investigations of blood included total leukocyte count, hemoglobin, eosinophilic count, alanine transpeptidase (ALT), aspartate transpeptidase (AST), alkaline phosphatase (ALP), gamma glutamyl transpeptidase (GGT), total serum bilirubin, blood urea, and serum creatinine. The blood specimens were collected into gel serum tubes (for clinical chemistry variables) and EDTA tubes (for hematology variables). Blood specimens collected into gel tubes were centrifuged at 1800–2000 g for 10–15 min. All blood specimens were analyzed on the day of collection.
Data were entered using EpiData version 6.04 (Epidata Association; Odense, Denmark). CR was calculated as proportion of individuals excreting Fasciola eggs before treatment and absence of eggs at study end. To determine infection intensity, the number of Fasciola eggs per Kato-Katz thick smear (41.7 mg of stool) was multiplied by a factor 24 to obtain eggs per gram of stool (EPG). Fecal egg counts (FECs) of multiple slides per individual were averaged, using the arithmetric mean. To calculate the reduction in infection intensity, individual egg counts were logarithmically transformed (log (count + 1) and the GM expressed as the antilogarithm of the mean. The ERR was calculated as [1 - GM FEC after treatment divided by GM FEC at admission multiplied by a factor 100]. Although infection intensity thresholds are currently lacking for infections with Fasciola [27], we classified infections into two groups: (i) light (1–99 EPG) and (ii) moderate/heavy (≥100 EPG). Of note, a threshold of 100 EPG is also used to distringuish between light and moderate (100–399 EPG) and heavy (≥400 EPG) S. mansoni infection [27]. Fisher's exact test, including 95% confidence intervals (CI), and Mann-Whitney U test were used to compare the outcome of both studies (2-sided P values) assuming no difference in population or sensitivity of the parasite strain. The 2-tailed paired t-test and the Kruskal-Wallis tests were employed to compare the clinical parameters before and after treatment.
Of 584 villagers and 51 school-aged children screened in El-Haddad El-Bahary village (study 1), 22 individuals were found Fasciola-positive. Two patients were excluded (pregnancy, n = 1; age below 5 years, n = 1). Twenty patients (10 females, 10 males; aged 5–70 years with a mean of 24 years) were included in study 1 (Table 1).
In the second study, 631 individuals were examined and 19 Fasciola-positive subjects were identified. Of these, 17 patients (10 females, 7 males; aged 5–26 years, with a mean of 14 years) (Table 1) were included in the study. However, two of the positive cases were excluded because the initial diagnosis by the Ministry of Health and Population could not been confirmed.
The baseline GM Fasciola FECs in the two studies were 28.3 EPG and 29.1 EPG (Table 2). Twenty-six individuals were classified as lightly infected (1–99 EPG), whereas 10 individuals had a moderate/heavy infection (≥100 EPG). Ten participants were concurrently infected with Fasciola spp. and S. mansoni, and one patient was identified with a double infection of Fasciola spp. and Hymenolepis nana.
Data from all patients were included in the analysis, as no patient was lost to follow-up (per-protocol analysis). CRs achieved with 6×80 mg and 3×200 mg artemether were 35% and 6%, respectively (Table 2). Fisher's exact test showed a statistical difference between the CRs obtained with the different treatment schedules (P = 0.048; 95% CI: 0.002–1.15). None of the patients characterized by an infection intensity of 100 EPG and above was cured after artemether administration regardless of the treatment regimen, while CRs documented in patients with a light Fasciola infection were 54% (6×80 mg artemether) and 8% (3×200 mg artemether) (CRs of light infections were significantly higher in study 1 compared to study 2; P = 0.013; 95% CI: 0.001–0.77). Treatment with artemether over 3 consecutive days resulted in ERRs of 63% (67% for light infections and 55% for infections ≥100 EPG). The individual pretreatment and posttreatment FECs are presented in Figure 1. No effect on FECs were observed when artemether was administered on a single treatment day with the exception of a very low ERR of 6% among patients with an infection intensity ≥100 EPG. The overall ERR between the two studies differed significantly (P<0.001).
In each of the two studies, five patients were co-infected with S. mansoni. At treatment follow-up, three out of the five patients in each study were recorded egg-free (CR: 60%).
Sixteen patients who were still found Fasciola-positive after treatment with 3×200 mg artemether were administered a single 10 mg/kg oral dose of triclabendazole. CR and ERR were 69% and 94%, respectively; significantly higher than CR (P<0.001; 95% CI: 3.19–1605.7) and ERR (P<0.001) observed following treatment with 3×200 mg artemether. The infection intensity did not influence the treatment outcome (data not shown). Four out of five patients who were still passing Fasciola eggs following a single triclabendazole dose were provided a double dose of triclabendazole and the respective CR and ERR were 75% and 96%.
While the veterinary importance of fascioliasis cannot be overemphasized, this zoonotic disease is also of considerable and growing public health importance, yet it often remains neglected. A major challenge is that treatment is restricted to a single drug, i.e., triclabendazole, which is registered for human use only in Ecuador, Egypt, France, and Venezuela [7]. Results from a study carried out in Vietnam raised some hope for an alternative; artesunate administered to patients with symptomatic fascioliasis pointed to a potential role of the artemisinins against fascioliasis. Indeed, the authors concluded that it is worthwhile to investigate this drug class in more detail, including additional clinical trials [20]. We now present the first results with artemether in the treatment of chronic fascioliasis in two epidemiological settings of Egypt. Artemether (monotherapy) was administered following the dosing regimen of a commonly used ACT, the 6-dose regimen of artemether-lumefantrine [21], and a previously employed 3-dose malaria treatment schedule administered on a single day [22]. Egypt was selected because of the known fascioliasis endemicity, particularly in the Nile Delta, and the absence of malaria [28], [29]. The prevalence of Fasciola spp. observed in the two study sites (i.e., Behera and Alexandria; prevalence 3–4%) was similar to previous studies in these areas [5], [28], [30], despite frequent community treatment programs with triclabendazole.
Our study failed to extend promising findings obtained with the artemisinins in rats experimentally, and sheep naturally, infected with F. hepatica [31]. Indeed, we found low CRs (6–35%) when artemether was given at two different malaria treatment schedules. Nonetheless, a moderate ERR of 63% was observed following the 6-dose course of artemether. The difference in the ERR between the two artemether treatment schedules (nil vs. 63%) is striking, yet difficult to explain. Since the half life of artemether is very short (<1 h) [32], parasite exposure to the drug might have been insufficient if the drug is given on a single treatment day. However, detailed in vitro drug sensitivity and pharmacokinetic studies are required to further elucidate this issue. It is interesting to note that the CRs (nil vs. 54%) and ERRs (55% vs. 67%) were higher in patients classified as lightly infected compared to moderate/heavy infections in the 6-dose regimen. A similar trend was observed in a recent study, which assessed the efficacy of an artesunate-sulfalene plus pyrimethamine combination in S. mansoni-infected school-aged children in Kenya: significantly higher CRs were observed in children harboring a light S. mansoni infection compared to moderate and heavy infections [33].
In the present study, five participants in each of the two studies were co-infected with Fasciola spp. and S. mansoni. Moderate CRs were observed against S. mansoni (60%) regardless of the selected treatment regimen. This finding is in line with previous studies, which documented low-to-moderate efficacies of an artemisinin monotherapy in the treatment of chronic infections with Schistosoma spp. [34]–[36]. An opposite trend, a CR of 70% and an ERR of 86% was reported following treatment of Nigerian children using two doses of artesunate at 6 mg/kg given 2 weeks apart [37]. In recent years also the effect of ACTs on schistosomiasis has been studied (for a summary of studies, see Utzinger et al. (2010) [38]) Overall, a moderate efficacy was observed using ACTs against the two major schistosome species, S. mansoni and S. haematobium. Although promising results were obtained in small exploratory trials with the artemisinins against schistosomiasis, larger clinical trials could not confirm these findings, and hence praziquantel remains the drug of choice [33], [38], [39].
CRs of 69% and 75% were observed in patients treated with triclabendazole at 10 mg/kg and 20 mg/kg, respectively. The observed efficacy is slightly lower than a calculated overall CR of 83% following 10 mg/kg and reported CRs ranging from 93 to 100% following a double dose of triclabendazole [40]. Additionally, a recent study with 10 mg/kg triclabendazole in Egypt reported a complete cure following triclabendazole (10 mg/kg) [41]. However, care is indicated in these comparisons because of the small sample sizes in the current study, although strain differences in the susceptibility of Fasciola spp. to triclabendazole might play a role in the somewhat lower efficacies observed here compared to previous studies. Participants treated with triclabendazole showed a higher incidence of abdominal pain compared to those treated with artemether, which might be related to the higher efficacy of triclabendazole (dying worms).
In conclusion, significantly higher CRs and ERRs were observed with triclabendazole when compared to artemether, the latter administered following two malaria treatment schedules. Hence, triclabendazole remains the drug of choice against fascioliasis. In view of threatening triclabendazole resistance development, concerted efforts are required, including structure-activity relationships with the synthetic peroxides in F. hepatica-infected rats [42]. Combination chemotherapy is also recognized as a potential strategy for reducing the emergence of drug resistance [43], [44]. Since we have observed synergistic interactions of combinations of triclabendazole (2.5 mg/kg) plus artemether (6.25–100 mg/kg) on adult worm burden in F. hepatica-infected rats [15] further preclinical studies to investigate the efficacy and safety of an artemether-triclabendazole combination are warranted. Combination chemotherapy with artemether and triclabendazole might offer an advantage over triclabendazole monotherapy, in particular in the case of possible future treatment failures with triclabendazole alone.
|
10.1371/journal.pcbi.1002474 | Geometric Theory Predicts Bifurcations in Minimal Wiring Cost Trees in Biology Are Flat | The complex three-dimensional shapes of tree-like structures in biology are constrained by optimization principles, but the actual costs being minimized can be difficult to discern. We show that despite quite variable morphologies and functions, bifurcations in the scleractinian coral Madracis and in many different mammalian neuron types tend to be planar. We prove that in fact bifurcations embedded in a spatial tree that minimizes wiring cost should lie on planes. This biologically motivated generalization of the classical mathematical theory of Euclidean Steiner trees is compatible with many different assumptions about the type of cost function. Since the geometric proof does not require any correlation between consecutive planes, we predict that, in an environment without directional biases, consecutive planes would be oriented independently of each other. We confirm this is true for many branching corals and neuron types. We conclude that planar bifurcations are characteristic of wiring cost optimization in any type of biological spatial tree structure.
| Morphology is constrained by function and vice-versa. Often, intricate morphology can be explained by optimization of a cost. However, in biology, the exact form of the cost function is seldom clear. Previously, for many different natural trees authors have reported that most bifurcations are planar and we confirm this here for branching corals and mammalian neurons. In a three-dimensional space, where bifurcations can have many shapes, it is not clear why they are mostly planar. We show, using a geometric proof, that bifurcations that are part of an optimal wiring cost tree should be planar. We demonstrate this by proving that a bifurcation that is not planar cannot be part of an optimal wiring cost tree, using a very general form of wiring cost which applies even when the exact form of the cost function is not known. We conclude that nature selects for developmental mechanisms which produce planar bifurcations.
| Bifurcations are observed widely in nature [1], such as in dendrites and axons of neurons [2], plants [3], rivers, capillaries [4], [5], [6], bronchi and tracheal systems [7], [8], [9] and octo- and scleractinian corals [10], [11]. Most studies have focused on characterizing tree geometry [7], [12], [13], [14], [15], [16] with an emphasis on the implications for function [17], [18], [19] or in relation to optimization processes [4], [8], [20], [21].
Overall, much less attention has been given to bifurcation properties. While it has occasionally been reported that bifurcations tend to be planar in different natural trees such as in neurons [22], in arterial systems [6], in lungs [9] and in plants [3], this property has not been systematically studied or properly explained [3].
In this paper we first characterize planarity in two very different types of spatial tree: the branches of corals and the dendrites of neurons. The point of doing so is to demonstrate evolutionary convergence, a feature one expects to be exhibited by any true optimization principle in biology [17]. We next investigate whether we can use the theory of Steiner trees to explain this phenomenon. Steiner tree theory is an active research field [23] that studies wiring cost minimization [24], mostly in two dimensions. It has been used as a framework for understanding wiring cost optimization in neurons [25], [26] and other naturally occurring trees [27]. Steiner trees minimize edge costs by allowing the addition of extra nodes (Steiner points) whenever these reduce the total wiring cost. When the costs are defined as the Euclidean distances between nodes, these are Euclidean Steiner trees [28]. Minimal Euclidean Steiner tree bifurcations in space must be planar [29], but they also require 120° angles between adjacent edges [28], [29].
We show that the bifurcations we studied are not compatible with the Steiner tree paradigm and instead propose a new general theoretical framework that has the advantage of not requiring any specific assumptions about wiring cost beyond an increase with branch length. This theory provides for the first time an explanation for flat bifurcations that can be applied to all kinds of natural trees.
We examined whether planarity is a common property of natural structures by quantifying flatness using cone angles [22]. The cone angle of a bifurcation is the aperture or opening angle of a right circular cone which contains the three bifurcation branches within its surface and has its apex at the branching point. A flat bifurcation has a cone angle of 180°. We examined the properties of bifurcations in two very different biological data sets. The first data are corals where planarity has not been demonstrated previously: we measured cone angles in digital reconstructions of four species of the branching coral Madracis (Figure 1A, Table S1). Secondly, we extended the previous observation of flat bifurcations in visual cortex pyramidal neurons [22] to eight different mammalian neuron types (Figure 1B, Table S1). We found that despite their rich and varied spatial morphologies, in all coral species (Figure 1A) and all neuron types (Figure 1B) most dendritic bifurcations (57 to 88%) were close to planar (between 160° and 180°), confirming that planarity is a general property of coral and of neuronal dendritic trees and axons (Figure S1A).
We next asked whether flat bifurcations could be explained by considering these trees to be Steiner trees. We tested whether coral or neuronal dendritic bifurcation angles between any of the 3 branches in a bifurcation tend to be close to 120°, as must be the case for minimal Euclidean Steiner trees. Although the peaks of branching angle distributions of corals were near 120°, only 19 to 22% of branches had branching angles close to 120° (between 110° and 130°) (Figure 1A). There were differences between coral species: while most had a single peak in the distribution, Madracis mirabilis had two separated peaks. Conversely, in neurons all branching angle distributions were bimodal with peaks smaller and larger than 120° (Figure 1B). Consequently only 3 to 15% of angles were close to 120°, as was also shown in other studies [25], [26].
The fact that neurites and coral skeletons are not of Euclidean Steiner tree type leaves unanswered the question of why their bifurcations are mostly planar. Can this be explained by another optimality-based principle?
We first checked the null hypothesis that random bifurcations are not planar. To investigate this, we calculated the probability distribution of cone angles for bifurcations with a random orientation of the 3 branches (Figure 2). For this calculation, the bifurcation point was located at the center of a unit sphere and all other points were located on the sphere surface, but the results apply also to 4 random points in space defining a bifurcation (Text S1). For a bifurcation to be on a cone with cone angle , all non-bifurcation points need to lie on the circular intersection, of the cone with the unit sphere, whose circumference is (Figure 2B). Even taking singular cases into account, the probability distribution for cone angle is in fact (Text S1) given by(1)Since the circumference of the base circle increases with increasing cone angle, the probability distribution of also increases, with a maximum at 180° (Figure 2D). Though the finding that even random bifurcations have a tendency to be flat may be surprising, we found that in corals and neurons the proportion of close to planar bifurcations was always much more pronounced (Figure 2C, p = 10−5).
For the neural data we also excluded histological artifacts as a cause of planarity (Text S1 and Table S2).
Having excluded trivial explanations for the planarity of bifurcations, we returned to the issue of wiring cost minimization and discovered that indeed it is possible to prove that an optimal wiring cost tree, even with varying wiring cost, should have planar bifurcations. We assumed that a number of regions (which may be simply points) containing terminal or target points are given, which are connected by a wiring cost minimizing tree. No causal or teleological relationship between these regions and the growth of the tree is implied. Our proof is essentially a test of optimality of any given spatial tree. Given a tree, which must necessarily have terminal points, we asked whether it could be optimal with respect to those points. Whether or not the terminal regions or points were specified before or during growth, or were simply the terminal points of the tree at the moment we observed it, does not enter in to our mathematical proof. The wiring cost we considered is a sum of individual edge costs, each of which should be continuous and strictly increasing with edge length. This type of wiring cost is general enough to describe not only a total wiring cost in the usual sense, but also the total cost of paths from each terminal point or intermediary target point to a root, or any linear combination of these costs [30], [31], [32]. Moreover, this type of wiring cost can also take into account conflicting cost functions previously proposed in neuroscience such as volume, surface and neural conduction time [26], [33], since all of them increase with wiring length, as well as cost functions that might vary during development or are specific to each branch [34]. See Text S1 for details.
Rather than constructing an optimal solution, which is known to be an NP-hard problem [35], we investigated properties any optimal tree must possess. As in the Steiner tree problem, we allowed additional points to be added. We began by considering an arbitrary bifurcation point () and the three points it connects to () (Figure 3A). These three points define a bifurcation plane. If the bifurcation point is not in the bifurcation plane (i.e. if the bifurcation is not planar) we asked whether it can be part of an optimal tree. We showed that one can always move this bifurcation point, unless explicitly forbidden by an imposed biological necessity or obstruction, onto the bifurcation plane in such a way (normal projection) that the three edges of the bifurcation are all shortened. The fact that three edges of the tree can be shortened, without making any change to the other edges of the tree, means that the entire tree containing the non-planar bifurcation cannot have been optimal. Thus, we could conclude that all bifurcations in any optimal tree must be planar.
What does the shape of optimal bifurcations tell us about the shape of the whole tree? The shape of a tree is determined by its target points, and when these points do not provide directional biases our result suggests that there is no need for any correlation between the orientations of the bifurcation planes. To further test this idea, we measured the angles between consecutive bifurcation planes in the neuronal and coral data (Figure 3B). These experimental distributions clearly approximate a uniform distribution of angles from 0 to 90° for all, except for Purkinje neurons and the skeletons of Madracis mirabilis colonies. Madracis mirabilis is known to be geometrically exceptional within the genus Madracis as a result of its unusually regular branching patterns [16] which are strongly suggestive of developmental constraints overriding purely geometrical optimality considerations. In the case of Purkinje cells, the symmetry breaking induced by the parallel fibers allows only a globally planar solution to be optimal, meaning that the dendritic tree is flat and all the bifurcations lie on the same plane [32], just as the symmetry breaking induced by water currents can result in globally planar fan coral morphologies [36] in octocorals. Similarly developmental controls during lung growth cause the bifurcations which are planar to occur in either the same plane or in an orthogonal plane to the preceding one [9]. These apparent exceptions are important because they strengthen our case for evolutionary convergence between neuronal and octo- and scleractinian coral skeleton morphologies.
We have confirmed that diverse natural trees have a large proportion of flat bifurcations and proposed a general theory that shows that this must be expected for optimal wiring cost trees. Unlike Euclidean Steiner tree theory, in which the cost of an edge is nothing other than its length, our general theory allows each individual edge to have its own cost defined in terms of any continuous, strictly increasing function of length. The fact that we do not need to specify the actual functions any further is a strength of our approach. It is not whether edges make 120° angles which is characteristic of optimal wiring cost [26], but rather the fact that bifurcations are always planar. Our theory extends to bifurcations of the axons of neurons (Figure S1), where planarity has previously been overlooked [26], and planar bifurcations observed in arteries [4], [5], [6], mammalian lungs and invertebrate trachea [9], [37], plants [3], or even nanotubes [38].
Our theory does not suggest specific mechanisms for achieving planar bifurcations and different organisms will choose different strategies. In the case of neurons, dendritic planarity has been attributed to tension forces flattening the dendrite during development [39]. However tension may not suffice to explain dendritic planarity in complex extracellular space. Instead, neuronal dendritic planar bifurcations could arise during growth via mechanisms such as pruning [40], [41], [42] where branches with high wiring cost are eliminated, or with simple growth rules [43] like, repulsion between branches [44], [45], or via genetic control [9], [37], [46].
The rich literature on the optimization of neuronal trees has mostly focused on larger scale tree morphology. Because our theory defines wiring cost optimization in a very general way, it is independent of any particular biophysical model of neuronal tree growth, restructuring or maintenance [39], [40], [47], [48]. Our results are also consistent with different explanations for the appropriate definition of cost, be it minimal energy expenditure of processes [22], flow across a bifurcation [8], [39] involving functional considerations such as connectivity [49], or other models which have been proposed [26], [43], [50].
Conversely, much less attention has been given to the optimization principles underlying tree-structures in coral biology. By some authors it has been hypothesized that the coral is optimizing branch spacing and compactness to maximize the internal flow velocity between the branches of a colony that sustains the mass transfer rate into and out of the colony [51]. There is a strong morphological plasticity in corals due to environmental influences [52]. The wiring cost optimization we observed suggests that the branching structure is formed with a minimum amount of material, whether this structure is also optimizing mass transfer in the colony is still unknown. The branching angles distributions for the 4 related species are clearly species-specific. Usually the classical taxonomy (e.g. [53], [54] for Madracis) in corals is based on corallite morphologies, while the overall colony is described in a rather qualitative and informal way.
Here, we have provided a general, geometric explanation for the planarity of bifurcations. To the best of our knowledge, it is the first proof of the requirement of planarity of bifurcations in a general spatial framework of wiring cost optimization with varying wiring cost. One of the strengths of our contribution is to show that planarity can be understood at a fairly abstract level and a wide variety of branched tree structures in all areas of biology.
|
10.1371/journal.pntd.0002087 | Proteomic Selection of Immunodiagnostic Antigens for Human African Trypanosomiasis and Generation of a Prototype Lateral Flow Immunodiagnostic Device | The diagnosis of Human African Trypanosomiasis relies mainly on the Card Agglutination Test for Trypanosomiasis (CATT). While this test is successful, it is acknowledged that there may be room for improvement. Our aim was to develop a prototype lateral flow test based on the detection of antibodies to trypanosome antigens.
We took a non-biased approach to identify potential immunodiagnostic parasite protein antigens. The IgG fractions from the sera from Trypanosoma brucei gambiense infected and control patients were isolated using protein-G affinity chromatography and then immobilized on Sepharose beads. The IgG-beads were incubated with detergent lysates of trypanosomes and those proteins that bound were identified by mass spectrometry-based proteomic methods. This approach provided a list of twenty-four trypanosome proteins that selectively bound to the infection IgG fraction and that might, therefore, be considered as immunodiagnostic antigens. We selected four antigens from this list (ISG64, ISG65, ISG75 and GRESAG4) and performed protein expression trials in E. coli with twelve constructs. Seven soluble recombinant protein products (three for ISG64, two for ISG65 and one each for ISG75 and GRESAG4) were obtained and assessed for their immunodiagnostic potential by ELISA using individual and/or pooled patient sera. The ISG65 and ISG64 construct ELISAs performed well with respect to detecting T. b. gambiense infections, though less well for detecting T. b. rhodesiense infections, and the best performing ISG65 construct was used to develop a prototype lateral flow diagnostic device.
Using a panel of eighty randomized T. b. gambiense infection and control sera, the prototype showed reasonable sensitivity (88%) and specificity (93%) using visual readout in detecting T. b. gambiense infections. These results provide encouragement to further develop and optimize the lateral flow device for clinical use.
| Human African Trypanosomiasis is caused by infection with Trypanosoma brucei gambiense or T. b. rhodesiense. Preliminary diagnosis of T. b. gambiense infection relies mainly on a Card Agglutination Test for Trypanosomiasis (CATT), which has acknowledged limitations. New approaches are needed, first to identify new diagnostic antigens and, second, to find a more suitable platform for field-based immunodiagnostic tests. We took an unbiased approach to identify candidate diagnostic antigens by asking which parasite proteins bind to the antibodies of infected patients and not to the antibodies of uninfected patients. From this list of twenty-four candidate antigens, we selected four and from these we selected the one that worked the best in conventional immunodiagnostic tests. This antigen, ISG65, was used to make lateral flow devices, where a small sample of patient serum is added to a pad and thirty minutes later infection can be inferred by simple optical read out. This simple prototype device works as well as the CATT test and may be developed and optimized for clinical use in the field.
| Human African Trypanosomiasis (HAT), also known as Sleeping Sickness, is a disease caused by Trypanosoma brucei gambiense and T. b. rhodesiense [1], [2], [3]. The parasites are transmitted in sub-Saharan Africa by the bite from an infected tsetse fly. HAT is of great public health significance, with epidemic outbreaks recorded several times over the past century with, at times, estimates of 300,000 or more infected individuals [4]. Today, the recorded number of new cases has dropped below 10,000 per year, yet HAT still continues to place a large burden on individuals and communities in terms of disability-adjusted life years [5], [6]. The identification of infected individuals is crucial for therapeutic and public health intervention. New tools could aid eradication of this disease when used in coordination with other efforts [5], [7], [8].
Infection with T. b. gambiense or T. b. rhodesiense progresses through two defined stages. The first stage is when trypanosomes are limited to the blood and lymphatic systems. The second stage occurs when the parasites invade the central nervous system [2]. The latter leads to neurological damage, sleep cycle disruption, coma and death if the patient does not receive treatment [9], [10], [11]. The two stages are treated with different drugs, and those used for the second stage have severe toxic side effects [12], [13]. Staging of the infection, to select the appropriate therapeutics and follow up, is currently done by sampling the cerebral spinal fluid to search for the presence of parasites and/or increased numbers of lymphocytes [14]. The view that human trypanosome infections are invariably fatal if not treated has been challenged recently [15], [16] but, nevertheless, early diagnosis is extremely important both for individual patient outcomes and for controlling epidemic spread [17], [18].
The identification of infected individuals relies on dedicated screening teams that visit at-risk communities or patients seeking medical examination [19]. HAT diagnosis in the field faces many difficulties; not least the logistical challenges for the screening teams to attend communities in rural locations. In endemic areas, civil disturbance usually increases the incidence of HAT and decreases the frequency of screening [20], [21], [22]. Once the screening teams are with the communities, they face further challenges to recruit the entire local population into the HAT screening programme, which can lead to under-reporting and under-estimations of infection rates [23], [24], [25], [26].
The current HAT screening regimen uses the Card Agglutination Test for Trypanosomiasis (CATT), a serological test that detects whether antibodies from an individual are able to aggregate a suspension of fixed and stained T. b. gambiense trypanosomes [27], detecting primarily antibodies to the variant surface glycoproteins (VSGs) on the fixed cells. If patients have a positive CATT result, microscopic examination of their blood is carried out to detect trypanosomes. If this is positive, a lumbar puncture is performed to stage the infection. Over the years, the CATT test has been optimised to improve sensitivity, specificity and stability. Such modifications include dilution of the blood samples, the use of multiple trypanosome clones expressing different VSG variants and improvements in thermostability [28], [29], [30], [31]. Despite the usefulness and wide deployment of the CATT test, it has several widely accepted limitations [32], [33], [34], [35]. These include varying degrees of sensitivity and specificity, its inability to detect T. b. rhodesiense infections, the requirement for trained screening personnel to use it and the specialised manufacture which precludes production on a scale necessary to saturate the market [26], [36]. There have also been other post-CATT test diagnostic enhancements. For example, the concentration of trypanosomes from infected blood to improve microscopic detection [37], [38], [39]. Further, the detection of trypanosome DNA in blood by loop-mediated isothermal amplification of DNA (LAMP) [40] methods are under investigation and are summarised in a recent review [41]. However, these diagnostic methods require relatively sophisticated laboratory equipment. In summary, there is well accepted case for developing an extremely simple, low-cost diagnostic device with greater sensitivity and specificity than current field tests [4].
Lateral flow devices are simple tests that can rapidly detect nanogram amounts of antibodies or antigens in finger-prick blood samples without the need for any ancillary equipment [42]. T. b. gambiense infections are characterised by very low parasitemias, often <1000/ml, the equivalent of <5 ng total trypanosome protein/ml blood. Thus, using currently available technology, it is not feasible to directly detect a trypanosome protein and a lateral flow test that detects host antibodies is perhaps more likely to have the required sensitivity and specificity. The manufacture of large numbers of lateral flow devices requires milligram to gram amounts of diagnostic antigen, therefore potential diagnostic antigens for such devices should preferably derive from recombinant or synthetic sources. Recently the Foundation for Innovative New Diagnostics (FIND) has invested in developing new diagnostic tests for human African Trypanosomiasis [43]. With a similar aim in mind, we also set out to identify novel diagnostic antigens, and to create a prototype lateral flow test device, but using a non-biased (proteomics) approach to select potential biomarker antigens. The results of this antigen selection and the performance of a prototype lateral flow device are reported here.
All human serum samples were collected with the informed consent of the patients that they could be used anonymously for diagnostic development. Rodents were used to propagate sufficient T. brucei parasites to make the detergent lysates for immunoaffinity chromatography and proteomics. The animal procedures were carried out according the United Kingdom Animals (Scientific Procedures) Act 1986 and according to specific protocols approved by The University of Dundee Ethics Committee and as defined and approved in the UK Home Office Project License PPL 60/3836 held by MAJF.
Two sets of human serum samples were used, the first was kindly provided by Philippe Büscher (Institute of Tropical Medicine, Antwerp) and consisted of nine sera from T. b. gambiense infected patients and nine from matched non-infected patients. These samples underwent virus inactivation using a procedure that retains antibody reactivity [44]. Briefly, 1% Tri(n-butyl)phosphate (TnBP) and 1% Triton X-45 were added to thawed serum samples and incubated at 31°C for 4 h. Sterile castor oil was added, mixed and the samples were centrifuged (3800× g, 30 min). The oil-extraction was repeated three times and the virus-inactivated sera (lower phases) were aliquoted and stored at −80°C. The second set of 145 patient sera (200 µl aliquots) was obtained from the WHO Human African Trypanosomiasis specimen bank [45]. Serum samples were aliquoted and stored at either −80°C for long-term storage or in 50% glycerol at −20°C when prepared for ELISA analysis. Freeze-thawing was kept to a minimum; samples from P. Büscher and WHO were freeze-thawed three times and twice, respectively, prior to use in ELISA tests.
Following virus inactivation, 125 µl of sera from four infected and four uninfected (control) patients were pooled. Each pool was applied to a 1 ml protein G column (GE Healthcare) equilibrated in phosphate buffered saline (PBS). The columns were washed with 10 ml of PBS and the bound IgG antibodies were eluted with 50 mM sodium citrate pH 2.8, and collected in 1 ml fractions into tubes containing 200 µl of 1 M Tris-HCl, buffer pH 8.5. Peak fractions containing IgG were combined and dialysed for 16 h against coupling buffer (0.1 M NaHCO3, 0.5 M NaCl, pH 8.3).
CNBr-activated Sepharose (GE Healthcare) was hydrated in 1 mM HCl and then equilibrated in coupling buffer. Aliquots (0.75 ml packed volume) were mixed with 7.2 mg of purified infection IgG or purified control IgG in a final volume of 3 ml coupling buffer for 16 h at 4°C. The coupling of IgG was confirmed by measuring the absorbance of the supernatant at 280 nm before and after coupling. The Sepharose-IgG conjugates were centrifuged at 500× g (10 mins, 4°C) and the beads were resuspended in 15 ml 1 M ethanoamine, pH 9, to block remaining amine-reactive sites for 2 h at room temperature. Following this, the IgG-Sepharose beads were washed with three cycles of 0.1 M Tris-HCl, pH 8.0, 0.5 M NaCl followed by 0.1 M sodium acetate buffer, pH 6.0, 0.5 M NaCl and finally washed and stored in PBS containing 0.05% NaN3.
Six BalbC mice were injected with T. b. brucei Lister 427 variant MITat 1.4 cells. After three days, infected mouse blood was harvested with citrate anticoagulant, adjusted to 107 parasites per ml with PBS and aliquots of 0.5 ml were injected into the peritoneal cavity of 12 Wistar rats. The rat blood was harvested after 3 days with citrate anticoagulant and centrifuged at 1000× g for 10 min at 4°C. Plasma was removed and the buffy layer was resuspended in separation buffer plus glucose (SB + glucose; 57 mM Na2HPO4, 3 mM KH2PO4, 44 mM NaCl, 10 g/l glucose) and applied to a DE52 DEAE-cellulose (Whatman) column that had been pre-equilibrated with SB + glucose. The trypanosomes were washed through the column with SB + glucose, counted, centrifuged (900 g, 15 min, 4°C), resuspended in 1 ml PBS and then adjusted to 1×109 parasites/ml in ice-cold lysis buffer (50 mM Na2PO4, pH 7.2, 2% n-octyl β-D-glucopyranoside (nOG) detergent, 1 mM PMSF, 1 mM TLCK, 1 µg/ml aprotinin, 1 µg/ml leupeptin and 1× Roche protease cocktail minus EDTA). The lysate was incubated for 30 min on ice and then centrifuged at 100,000 g for 1 h at 4°C.
Aliquots of T. b. brucei lysate (1010 cell equivalents) were incubated with 0.75 ml packed volume of each of the Sepharose-IgG (infection and non-infection/control) gels, rotating for 3 h at 4°C. The gels were then packed into disposable 10 ml columns and washed with 10 ml of 10 mM Na2PO4, pH 7.2, 200 mM NaCl, 1% nOG, followed by 10 ml of 5 mM Na2PO4 pH 7.2, 1% nOG. The trypanosome proteins were eluted 3 times with 750 µl of 250 mM sodium citrate, pH 2.8, 1% nOG and the eluates were pooled and neutralised with 1.5 M Tris-HCl, pH 9 and further concentrated to 140 µl using a centrifugal concentrator (Millipore, 0.5 ml capacity with 3 kDa MW cut off membrane). To remove eluted IgG, this fraction was mixed with 30 µl PBS-equilibrated Protein G agarose beads (Pierce) and incubated for 10 min and removed by centrifugation. The supernatant, containing the trypanosome proteins, were then transferred to low binding Eppendorf tubes and the proteins were precipitated by adding 1 ml ice-cold ethanol and incubation for 34 h at −20°C.
Following ethanol precipitation, the proteins eluted from the infection IgG and control IgG gels were dissolved in SDS sample buffer, reduced with DTT and run on a precast 4–12% BisTris gradient SDS-PAGE gel (Invitrogen) using the MES running system. The gel was stained with colloidal Coomassie blue and equivalent regions of the infection and control lanes were cut out, reduced and alkylated with iodoacetamide and digested in-gel with trypsin. The tryptic peptides were analysed by LC-MS/MS on a Thermo Orbotrap XL system and MASCOT software was used to match peptides to the predicted trypanosome protein databases (combined GeneDB and UniProt predicted protein sequences).
Trypanosome proteins identified uniquely in the infection IgG immunopurified fractions were considered for recombinant expression. Within these, proteins with high MASCOT scores, likely to be the most abundant, were prioritised for recombinant expression and purification trials. These proteins included Gene Related to Expression Site Associated Gene (GRESAG) 4, Invariant Surface Glycoprotein (ISG) 75, ISG65 and ISG64. The identified protein sequences were used to BLAST search the T. b. brucei predicted protein database, revealing several related protein sequences in each family. CLUSTALW2 alignments were carried out in order to better understand sequence sub-groups within those protein families. Representative gene segments from each protein sub-group that contained the peptide sequences identified by mass spectrometry were amplified from EATRO1125 genomic DNA (for ISG65-1, ISG65-2, ISG64-2, ISG64-3 and ISG75-1) or from stain 427 genomic DNA (for ISG64-1 and GRESAG4) by PCR using the primers described in the Supporting Information (Table S1). In each case, the products of three separate PCR reactions were cloned into a TOPO-TA vector (pCR2.1) for sequencing (DNA Sequencing Service, College of Life Sciences, University of Dundee).
The amplified ISG gene segments were cloned into various pET bacterial expression plasmids that provide a His-tag fused either to the N-terminus or C-terminus of the protein, in some cases via a TEV protease cleavage site, as indicated in (Figure 1). Multiple constructs were designed for GRESAG4 encoding the predicted full-length extracellular domain and several small globular domains based on predictions from GLOBplot software [46] (Figure 1). These constructs were amplified from genomic DNA using the primers described in the Supporting Information (Table S1), cloned into TOPO-TA vector pCR2.1 and verified by DNA sequencing. The constructs were either cloned into the pET15bTEV vector, such that the proteins they encode are fused at the N-terminus to a His tag, or into a pGEX-TEV vector such that the protein is fused at its N-terminus to a glutathione S-transferase (GST) sequence via a TEV cleavage site (Figure 1). The details of protein expression in E. coli and subsequent purification are described in the Supporting Information (Text S1).
White (Costar) un-treated 96 well plates were coated at 50 µl/well for 16 h at 4°C with 2 µg/ml recombinant protein diluted in plating buffer (0.05 M NaHCO3, pH 9.6). Plating solution was removed and wells were blocked with PBS containing 5% BSA, 200 µl/well for 3 h at 22°C or 16 h at 4°C. Plates were stored at 4°C and used within 24 h. Aliquots (50 µl) of serial serum dilutions (see below) were transferred in triplicate by a liquid handling device (Bio-Tek, Precision) to the ELISA plates and incubated for 1 h at room temperature, aspirated and 150 µl ELISA wash buffer was added to each well by the liquid handling device, left for 10 min and aspirated. This wash cycle was performed three times. Biotinylated goat anti-human-IgG (Jackson Immunoresearch) was diluted to 1∶5000 and 50 µl aliquots were applied to each well. After 1 h incubation at room temperature the secondary antibody solution was removed and wells were washed three times, as described above. Horseradish peroxidase (HRP) conjugated to NeutrAvidin (Sigma) was diluted to 1∶4000 and applied to the wells (50 µl/well) for 1 h at room temperature. Wells were washed as before. Finally, chemiluminescent Femto substrate (Pierce) diluted 1∶5 (i.e., 0.5 ml solution A, 0.5 ml solution B with 4 ml PBS) was applied to the wells at 50 µl/well and plates were read using an Envision plate reader after 2.5 min incubation at 22°C.
ELISA measurements were made with both pooled and individual serum samples. Serum pools were made by combining patients sera from; stage 1 T. b. gambiense patients (n = 10), stage 2 T. b. gambiense patients (n = 40) and matched uninfected patients (n = 50); and from stage 1 T. b. rhodesiense patients (n = 5), stage 2 T. b. rhodesiense patients (n = 20) and matched uninfected patients (n = 25). The pooled sera were diluted to 1∶60 in 50% glycerol, PBS and 1% BSA and stored at −20°C. For ELISA assays, the 1∶60 diluted pooled sera were further diluted to 1∶1000 in PBS, 0.1% BSA and then serially diluted (doubling dilutions) to 1∶32,000. For the individual sera, the 1∶60 diluted samples were further diluted to 1∶1000 immediately before use.
Sera were randomised by a member of the University of Dundee Tissue Bank. Forty T. b. gambiense infected patients sera and forty T. b. gambiense uninfected patients sera were randomly selected from the fifty T. b. gambiense infected and fifty uninfected WHO patient sera. These eighty serum samples were then randomised and coded.
Serum aliquots (5 µl) were diluted with 15 µl PBS and applied to the sample pad. Chase buffer (80 µl of PBS, 0.05% Tween 20) was added to the sample pad and the test was allowed to develop for 30 min. The test line was visually scored and the device was opened and the sample pads (at top and bottom of nitrocellulose membrane) were removed to prevent backflow. The lateral flow tests were photographed and scanned using a densitometer (CAMAG TLC scanner 3, CAMAG).
Bar graphs and scatter plots (x by y) were generated by Microsoft Excel. Box plots, Receiver Operator Characteristic (ROC) curves, antigen scatter plots (y axis only) were generated by SigmaPlot 12. Statistical analysis included Mann-Whitney (Rank Sum Test) and Dunn's post-hoc (Analysis of Variance (ANOVA) on rank) in SigmaPlot 12. Data were tested for normality by Kolmogorov-Smirnov test and were further processed by Mann-Whitney or Dunn's post-hoc tests. The P values were recorded for Mann-Whitney with <0.05 set as the cut off for statistical significance.
We took a non-biased proteomics approach to identify proteins that adsorb selectively to pooled infection IgG, and not to pooled control IgG (Figure 2A). Each serum pool contained four individual sera of patients clinically defined as having an infection with T. b. gambiense or as being uninfected. IgG fractions were purified from the pooled sera by affinity chromatography on protein G and then immobilised to cyanogen bromide-activated Sepharose beads. Equal amounts of infection and control IgG-Sepharose were incubated with equal amounts of T. b. brucei detergent cell lysate. Proteins that bound to the IgG columns were eluted by low pH, precipitated with cold ethanol, dissolved in SDS-sample buffer, reduced, separated by SDS-PAGE and stained with colloidal Coomassie blue (Figure 2B). More protein was seen in the eluate from the infection IgG column, consistent with infection-specific anti-trypanosome immune responses. Equivalent sections were cut out from the infection and control lanes, as indicated (Figure 2B). The excised gel pieces underwent in-gel S-alkylation and tryptic digestion, and the tryptic peptides were analysed by LC-MS/MS. Mascot software matched the peptide spectra to proteins in the T. b. brucei predicted protein database and scored the quality of the identifications. Lists of the proteins retained by infection IgG-Sepharose and control IgG-Sepharose were compared in each gel section. Twenty-four proteins with a MASCOT protein score above 50 were found uniquely in the infection IgG eluate and these are described in (Table 1). Several of the infection-specific proteins were defined as ‘hypothetical’, but other hits included known cell surface proteins, such as: Invariant Surface Glycoprotein (ISG) 75, ISG65, ISG64, Gene Related to Expression Site Associated Gene (GRESAG) 4, and the transferrin receptor subunits ESAG 6 and 7. As a starting point, the proteins with high MASCOT scores were prioritised. The rationale for this selection was that, by using an excess of trypanosome lysate in the affinity purification step, the amount of an eluted antigen should reflect, to a first approximation, the relative amount of antigen-specific immobilised IgG. The latter should, in turn, correspond to the immune response to that antigen in infected patients. Using this criterion, the protein antigens selected for study were ISG75, ESAG7, GRESAG4, ISG65, ISG64 and ESAG6 (Table 1). Next, we looked into the likely ease of protein expression of these antigens in E. coli. At this stage, we de-selected ESAG6 and ESAG7 because they form a heterodimer (adding the complication of dual expression) and because successful (but low level) protein expression has only been reported in a eukaryotic baculovirus expression system [47]. On the other hand, E. coli recombinant expression of domains of ISG75, ISG65 and ISG64 had either been reported in the literature [48] or were known to the authors (Mark Carrington, unpublished data). Consequently, we selected all three ISGs for protein expression trials. Finally, we performed expression trials on the predicted extracellular domain of GRESAG4, for which there was no literature precedent.
The selected purified recombinant trypanosome proteins, see Supporting Information (Figure S1), were used to prepare ELISA plates, as described in Experimental Procedures, and these were screened against various pooled human sera. These pools were derived from the 145 individual serum samples provided by the WHO Human African Trypanosomiasis specimen bank. The pooled sera were for stage 1 T. b. gambiense patients (n = 10), stage 2 T. b. gambiense patients (n = 40) and matched uninfected patients (n = 50); and from stage 1 T. b. rhodesiense patients (n = 5), stage 2 T. b. rhodesiense patients (n = 20) and matched uninfected patients (n = 20). The results indicated that both stage 1 and stage 2 T. b. gambiense infection sera have significant antibody titres against all of the rISG64 and rISG65 proteins, compared to pooled non-infection sera (Figure 3A), whereas infection sera titres against rISG75 and GRESAG4a were much closer to those for the control sera. The best performing recombinant protein was ISG65-1, which had the highest infection to control signal. For the T. b. rhodesiense pooled sera, the signals were generally significantly lower, with the stage 2 pooled sera giving a significantly higher signal than the stage 1 pooled sera. There was one exception to this; the T. b. rhodesiense stage 1 pool had the highest antibody titre against rISG75 (Figure 3B). However, as will be described later, the rISG75 result was due to a very high antibody titre in a single individual. From these results, all the rISG proteins were taken forward and screened against the individual sera but GRESAG4a (rG4a) was abandoned at this stage because it had poor infection versus non-infection discrimination.
Recombinant protein ELISA plates that performed well in the pooled sera ELISAs were further screened against all of the individual sera. These antigens included three rISG64 proteins, two rISG65 proteins and one rISG75 protein. In this case, a total of 163 individual serum samples (145 from the WHO HAT specimen bank and 18 from the Institue of Tropical Medicine, Antwerp) were diluted and applied in triplicate to wells coated with single recombinant proteins. T. b. gambiense and T. b. rhodesiense patient sera ELISA results were analysed separately (Figure 4) and (Figure 5), respectively. The data are shown as box plots for each different recombinant antigen ELISA plate (Figures 4A and 5A) to provide a visualisation the range of antibody titres and the heat maps provide a different view of the same data (Figures 4B and 5B). Both views suggest that rISG65 proteins provide the highest detection sensitivity whereas the rISG64-1 may provide slightly greater specificity. The rISG75 protein did not perform as well as the rISG65 or rISG64 proteins by both criteria and, indeed, only the stage 2 sera had statistically significant levels of IgG to rISG75-1 compared to controls (Q = 4.616, P = <0.05). Dunn's post-hoc tests (not shown) demonstrated that, whereas there are significantly higher levels of anti-rISG64 and anti-rISG65 IgG antibodies in both stage 1 and stage 2 sera compared to uninfected controls, there is no statistically significant difference between the stage 1 and stage 2 groups. In other words, relative immunoreactivity to rISG64 or rISG65 antigens cannot be used to stage of the disease. Formal sensitivity (i.e., the proportion of correct positive results) and specificity (i.e., the proportion of correct negative results) parameters for each test were calculated by ROC curve analysis (Figure 4C and 5C) and are collated in (Table 2). The recombinant antigens that best discriminated between T. b. gambiense infected and control patients by ELISA were rISG65-1 and rISG64-1, which had areas under the ROC curve of 0.99 and 0.98 respectively (Figure 4C). The rISG65-1 ELISA antigen had sensitivity of 96.6% (with a 95% Confidence Interval (CI) of 88.3 to 99.6%) and specificity of 93.2% (95% CI of 83.5 to 98.1%), whereas sensitivity and specificity of rISG64-1 antigen was 93.2% (95% CI of 83.5 to 98.1%) and 94.9% (95% CI of 85.9 to 98.9%), respectively (Table 2).
It was more difficult to find a recombinant protein antigen that reliably discriminated T. b. rhodesiense infected patient sera from non-infected sera. The box plots, heat maps (Figure 5A and B) and Dunn's post hoc analyses (not shown) all indicate that, whereas the stage 2 sera show statistically significant immunoreactivity to all the antigens compared to controls, the immunoreactivities of the stage 1 sera are not statistically significant. rISG65-2 was the most sensitive at identifying T. b. rhodesiense infection sera 92% (95%, CI of 74 to 99%), but at a cost to specificity 85% (95%, CI 62.1 to 97%) (Figure 5C and Table 2). As mentioned above, the pooled sera ELISA experiments had indicted that stage 1 T. b. rhodesiense infection sera might have high antibody titres towards rISG75. However this proved not to be the case and was due to a single serum sample with a very high anti-rISG75 titre.
Based on the ROC curve analyses of the performances of the ELISA plates, we selected rISG65-1 (ROC curve area 0.99 for T. b.gambiense sera) for development of a lateral flow prototype. Purified rISG65-1 was supplied to BBInternational (Dundee, www.bbigold.com) a company that specialises in lateral flow technology. The lateral flow approach that was utilised is illustrated in (Figure 6). Thus, rISG65-1 was both immobilised in a band on a nitrocellulose membrane and coupled to colloidal gold that was then localised in the conjugate pad. When the sera and chase buffer are applied to the sample pad, the rISG65-colloidal gold conjugate is resuspended. The absorbent pad at the top of the lateral flow device draws the liquid across the nitrocellulose membrane. During this time, any anti-rISG65 antibody in the serum binds to the rISG65-gold conjugate and when the antibodies reach the rISG65 test band, one Fab arm of the IgG binds to the immobilised rISG65 while the other Fab domain bridges to the rISG65-gold-conjugate. Accumulation of this specific antibody sandwich generates a visible test line. The control line is an internal positive control for the lateral flow test and does not relate to the infection status of the patient but indicates successful test flow. The final reading of this test should be as follows; the appearance of only a control line (upper band) indicates non-infected sera, whereas, the appearance of two lines, a control and test line (upper & lower bands) indicates infected sera, examples are shown in (Figure 6). Absence of a control line (upper band) indicates an invalid test, irrespective of the appearance of the test line and the test should be repeated.
Eighty randomised and coded WHO ‘test’ T. b. gambiense sera, comprising forty infected and forty non-infected sera were applied to the lateral flow prototypes. Each serum sample (5 µl) was diluted with 15 µl of PBS and applied to a lateral flow device sample pad. Within about 30 s, 80 µl of chase buffer was added and the test was left for 30 min, at which point a visual score was recorded. The sample pads were removed to prevent back flow and the visual scores were decoded (Figure 7). Sensitivity and specificity were calculated by ROC curve analysis, and for visual scores a cut off of 2.5 gave 100% sensitivity (95% CI of 91.1 to 100) and 87.5% specificity (95% CI of 73.2 to 95.8%). An analysis of the test lines was also carried out using a densitometer, where an arbitrary cut off at 265.6 RU gave 100% sensitivity (95% CI of 91.2 to 100%) and 92.5% specificity (95% of CI 79.6 to 98.4%) (Table 3) indicating there is potential for separation between infection and non-infected individual scores. Principally the end user will interpret the results visually therefore further optimisation of the test line will be necessary to reduce false positive results due to non-specific binding. A checklist, Supporting Information (Table S2), and flow diagram, Supporting Information (Figure S2), are provided according to the STAndards for the Reporting of Diagnostic accuracy studies (STARD) guidelines.
The overall goal of this project was to develop an immunodiagnostic lateral flow prototype for human African trypanosomiasis that might be developed into a field-based device to replace the CATT screening tool. To do this, we needed to identify potential diagnostic antigen candidates, investigate whether they could be adequately expressed and purified and assess their diagnostic potential with patient sera. We took an unbiased proteomics approach to identify more than twenty potential diagnostic protein antigens, several of which were known cell-surface glycoproteins. This list was filtered pragmatically; first, the proteins with high proteomic MASCOT scores, generally synonymous with their abundance, were selected because, by using an excess of trypanosome lysate in the affinity purification step, the amount of an eluted antigen should reflect the relative amount of antigen-specific IgG in infection sera. The latter should, in turn, correspond to the immune response to that antigen in infected patients. From this list we eliminated ESAG6, since it is known to form a heterodimer with ESAG7 and is, therefore, relatively complicated to express [46]. Our attempts to express parts of the extracellular domain of GRESG4 in E. coli were not very successful, although we were able to isolate the A-domain fused to GST. However, this G4a construct and the ISG75 protein construct did not perform well in the ELISA studies and were removed from this study. Nevertheless, these antigens should not be ignored for diagnostic development as they may simply have been miss-folded in the absence of endoplasmic reticulum folding and quality control components [49]. Indeed, recombinant ISG75 has been shown to have diagnostic potential for T. b. brucei animal infections [47]. Future expression attempts might include bacterial expression systems that target recombinant proteins into the periplasmic space [49] and/or eukaryotic expression systems such as insect cells and Pichia pastoris.
Our data on the diagnostic potential of ISGs 64, 65 and 75 for detecting T. b. rhodesiense infections were somewhat hampered by the small number of sera available for testing. Nevertheless, it is clear from the ELISA data that IgG antibody responses to these antigens are lower than in T. b. gambiense infections. Further, the IgG responses are particularly low in stage 1 T. b. rhodesiense patient sera. This may be due to the differing nature and speed of progression of the infections; T. b. rhodesiense infections are usually acute and progress faster whereas T. b. gambiense infections are chronic and progress over months or years [50], which could in turn lead to a greater amount and diversity of antibodies present in these sera. A previous study also struggled to identify diagnostic antigens for T. b. rhodesiense infections [51]. A good approach may be to repeat the procedures described here using immobilised IgG from T. b. rhodesiense patient sera.
Further research is also required to measure the half-life of antibodies in patients after they have been treated for HAT, as persistent antibodies may lead to false positives. It has been described that antibodies can persist up to 3 years post cure, however it is not known which class of antibodies persist or which antigens they recognise [52]. Ideally, a longitudinal study could be carried out to gain a greater insight into this and how it could affect the diagnostic potential of any future lateral flow test relying on antibodies [32]. Lateral flow tests, whilst having limitations, could potentially be more suitable for use in the field because of their stability and the fact that they can be used by non-specialists [53].
In summary, we report here the selection of ISG65-1 as a potential diagnostic antigen for T. b. gambiense infections and its performance in both conventional ELISA and prototype lateral flow device assays looks promising. The performance of the prototype ISG65 lateral flow device encourages us to further develop and optimize it, perhaps adding an additional antigen or antigens to improve sensitivity and specificity, while aiming for a production cost of <US$1 per unit.
|
10.1371/journal.pcbi.1005040 | The Differential Response of Proteins to Macromolecular Crowding | The habitat in which proteins exert their function contains up to 400 g/L of macromolecules, most of which are proteins. The repercussions of this dense environment on protein behavior are often overlooked or addressed using synthetic agents such as poly(ethylene glycol), whose ability to mimic protein crowders has not been demonstrated. Here we performed a comprehensive atomistic molecular dynamic analysis of the effect of protein crowders on the structure and dynamics of three proteins, namely an intrinsically disordered protein (ACTR), a molten globule conformation (NCBD), and a one-fold structure (IRF-3) protein. We found that crowding does not stabilize the native compact structure, and, in fact, often prevents structural collapse. Poly(ethylene glycol) PEG500 failed to reproduce many aspects of the physiologically-relevant protein crowders, thus indicating its unsuitability to mimic the cell interior. Instead, the impact of protein crowding on the structure and dynamics of a protein depends on its degree of disorder and results from two competing effects: the excluded volume, which favors compact states, and quinary interactions, which favor extended conformers. Such a viscous environment slows down protein flexibility and restricts the conformational landscape, often biasing it towards bioactive conformations but hindering biologically relevant protein-protein contacts. Overall, the protein crowders used here act as unspecific chaperons that modulate the protein conformational space, thus having relevant consequences for disordered proteins.
| Most in vitro and in silico biophysical experiments generally study proteins in an isolated environment, overlooking that their natural environment—the cell cytoplasm—is a solution that is highly populated by proteins. To address this knowledge gap, here we explored how a crowded environment alters the conformational sampling of three proteins, each with a different degree of disorder and flexibility. We simulated a crowded system composed by the three proteins and reaching a cell-like concentration and compared the protein behavior observed with that induced by PEG500, a synthetic crowding agent. Despite some similarities between the environments, protein crowders showed a number of characteristics that raise concerns about the use of diluted solutions or synthetic agents when studying protein behavior.
| Most in vitro and in silico biophysical experiments treat proteins as highly purified entities that act in isolation, overlooking their natural “habitat”, namely the cell cytoplasm. This “habitat” contains between 80 to 400 g/L of several other macromolecules, which together account for 5%-30% of volume occupancy [1]. Among the effects that a crowded environment exerts on protein behavior, volume exclusion is considered the most relevant [2]. Accordingly, crowders behave as inert molecules that do not interact with proteins, and their presence limits accessible space to proteins, thereby reducing the conformational entropy and favoring compact folded forms of the latter [3]. Following this view, most experimental studies on proteins in dense environments have been performed by adding large polymers, such as poly(ethylene glycol) (PEG), Dextran or Ficoll, to the media. These polymers, often referred to as “inert” crowders, are assumed to exclusively mimic the volume-exclusion effect [4]. However, recent experiments show that “inert crowders” exert a complex variety of effects on protein stability, and results largely dependent on the type and size of the crowder involved [3,5,6]. For example, calorimetric analysis concludes that Dextran, glucose and PEG lead to an enthalpic stabilization and an entropic destabilization of the protein; the latter predominant only in presence of PEG [7]. Indeed this synthetic compound appears to be less “inert” than expected due to attractive interactions with proteins, questioning its effectiveness in recreating a pure volume-exclusion effect. Despite so, PEG continues to be used as a reference agent to model macromolecular crowding [8,9]. Regarding the size, intuitively, the volume excluded by inert crowding agents is proportional to the crowder size and consequently small crowders might even help unfolding [5]. Recent studies in cell-like environments have further challenged such a model, suggesting that compacted conformations of proteins may not always be favored in physiological crowded environments [9–15].
Available data suggest that protein crowders have a dual nature. On the one hand, they display the classical volume-exclusion effect and, on the other, they have the ability to form weak and transient (quinary) soft interactions with solute protein [9,11,12]. These effects generates competition between destabilizing and stabilizing forces, the final result of which is difficult to predict [11,13,14]. To further complicate the scenario, crowding might also affect the folding landscape, leading to alternative states not present in dilute solutions and affecting protein functionality [15]. This distortion of the conformational landscape might have a dramatic impact on highly dynamic proteins, such as intrinsically disordered (IDPs) and molten globule proteins (MGPs) [16]. Unfortunately, most crowding studies performed with these proteins have used synthetic polymers and often report only the expected increase in the compactness of the structure [17–21]. Research into IDPs or MGPs in cell-like crowded environments is more rare and provides unclear conclusions [10], [18], [28–33].
A consensus theory—based on experimental data—on the nature of crowding is impeded by the intrinsic limitations of studying highly dynamic systems in which single molecule information is lost within the experimentally detected structural ensemble [20–26]. Theoretical calculations, particularly molecular dynamics (MD), give direct access to atomic information on single molecules in carefully controlled environments, and they are therefore the perfect complement to experimental ensemble-based techniques when addressing crowding effects [21], [35–38]. Here we took advantage of the power of MD simulations to explore the impact of small-sized synthetic (PEG500) and protein crowders (proteins) on the structure, dynamics and interactions of the following three proteins: i) an intrinsically ordered protein (IOP), the 191-residue interferon regulatory transcription factor (IRF-3); ii) a molten-globule conformation (MGP), the 51-residue nuclear coactivator-binding domain of CREB (NCBD); and iii) an intrinsically disordered protein (IDP), the 47-residue activator for thyroid hormone and retinoid receptor (ACTR). These three proteins not only model the three major types of protein conformational landscapes, but also define a specific biological network, with NCBD as the central partner (the hub), able to transiently interact with IRF-3 and ACTR, thanks to its structural promiscuity [39–42]. This is the first study to present calculations of the effect of crowding on proteins of distinct structural complexity that define a biologically relevant crowded microenvironment.
We performed microseconds-long MD simulations of five crowded systems, each composed by eight conformations of the three protein types (6 NCBD, 1 ACTR and 1 IRF-3) at increasing concentrations [from 175 to 300 g/L] (Fig 1). Each conformation was individually simulated in solution with the synthetic crowder PEG500 and in water. The latter condition was used as a control of the behavior of proteins within the selected simulation protocol.
Trajectories in pure water (S1 Fig and Fig 2) showed the expected behavior for the proteins under study. Thus, the intrinsically ordered protein (IOP: IRF3) was stable during the entire trajectory, maintaining the pattern of secondary structure, fold and shape. Native contacts were well preserved, with sizeable movements localized only at the C-terminal helix, in a region with interface contacts in the crystal. A small, but detectable, tightening of the hydrophobic core of the protein occurred. The intrinsically disordered protein (IDP: ACTR) was extremely mobile in water, sampling a wide repertoire of conformations. In this regard, clustering analysis detected more than 250 distinct conformers (most of them compact; see S1 Fig), none of which populated more than 5.5% of the trajectory. The contact map was fuzzy (compare with IRF3 in S1 Fig and Fig 2), suggesting the absence of remote long-lasting contacts, thus hindering the formation of stable folds. Some segments of ACTR tended to form a secondary structure, especially an α-helix at the N-terminal—an observation that is consistent with the results from NMR experiments [32,33]. However, these helical elements were unstable and fuzzy, with local populations rarely above 50% and undefined boundaries, making them unable to nucleate the global structure. Finally, the molten globule protein (MGP: NCBD) showed slow diffusion along the conformational space, with strong memory effects in the trajectories [30,34–36]. When the NCBD trajectory started from the “folded” conformation, significant plasticity was observed (around 100 structural clusters). This plasticity is attributed to the distinct orientation of the three helical motifs (h1, h2 and h3, see below and S3 Fig), which generate a fuzzy contact map with helical arrangements of the prevailing ACTR-binding form, while the helical arrangements required for IRF-3 recognition were rare. When the NCBD starting conformation was “unfolded”, it rapidly collapsed into an amorphous globule; the protein formed many remote and unstable contacts (282 structural clusters), and only small nascent elements of secondary structure (particularly in h1 and h2) were observed. NCBD appears to be a protein that was not evolutionarily designed to collapse into a single well-defined minimum. We conclude that control simulations provide a reasonable picture of the conformational landscape of the three proteins representing IOPs, IDPs and MGPs in water. We can therefore confidently use the same force-field and simulation protocol to explore crowded environments.
As described above, most theoretical and experimental studies on crowding have been performed using polymers (as PEG500) as co-solvents, which theoretically act as “inert” crowders mimicking cellular crowding. However, whether polymers such as PEG500 are truly “inert” crowders and whether they correctly mimic the crowded environment in the cell remain to be confirmed. In order to answer these two questions, we compared the trajectories of the three model proteins in water, and in PEG500-crowding and protein-crowding conditions (using similar crowder concentrations in both cases) (Fig 2).
For IOP (IRF3), the effect of crowding was modest, and neither proteins nor PEG500 induced large changes in the local or global structure of this well-structured protein. Crowding stabilized the secondary structure, including the C-terminal helix, which was fragile in the simulations in water. When compared to water, both types of crowders produced an increase in the size of the protein (see Fig 2 for radius of gyration, and S2 Fig for solvent-accessible surface). This observation is not consistent with the “exclude volume” theory. Only protein crowders were observed to decrease the relative ratio of polar solvent-accessible surface, thereby suggesting that they attenuate the hydrophobic effect compared to water, the latter environment showing a more visible collapse of the core (cartoons in Fig 2 and S2 Fig). Interestingly, the crystal structure of IRF-3 was more similar to the conformations sampled in a crowded environment (especially in the protein media) than to those in dilute aqueous conditions. These findings thus suggest that crystals can, in some cases, mimic physiological conditions better than water.
For IDP (ACTR), crowding agents had a huge impact on the conformational landscape, (Fig 2); however, we were unable to find a pattern of general “crowding” effects, since the changes induced by PEG500 differed from those induced by a protein environment. Thus, PEG500 generated a large expansion of the sampled conformational space, which became dominated by extended conformers showing only a moderate amount of secondary structure. In contrast, protein crowders reduced the conformational space sampled, which was now dominated by relatively compact conformations, with well-defined α-helices localized in those regions required for NCBD binding [37,38]. These results demonstrate the inability of PEG500 to reproduce physiological-like crowded conditions around IDPs and suggest that protein crowding might contribute to IDP folding in the bioactive conformation.
For MGP (NCBD), the behavior of crowders largely depended on the starting conformation, mirroring the “memory effects” detected in the simulations in water and reinforcing the idea that NCBD (and probably other MGPs) moves across a complex and frustrated conformational landscape. In the trajectories starting from folded NCBD, crowders favored more extended conformations than those sampled in water, introducing significant changes in the fuzzy pattern of long-range contacts (Fig 2). The helical fragments were often arranged in the bioactive conformations, sometimes closer to the IRF-3-bound state, which has never been sampled in water (S3 Fig). The bias towards the bioactive state was especially visible for protein crowding, where collected ensembles were on average 0.34 nm closer to the bioactive conformation found in the NCBD-IRF-3 complex than those sampled in water. The effect of crowders was even more dramatic (and complex) for NCBD trajectories starting from an unfolded state. Both PEG500 and protein crowders hindered the hydrophobic collapse observed in water, thus favoring extended conformations (Fig 2) in which native helices—which were hardly distinguishable in water—showed significant populations and well-defined boundaries (especially for helix 1). These observations again support the notion that crowding might help disordered proteins to adopt bioactive conformations. When analyzed in detail, the effects of synthetic (PEG500) and protein (protein) crowding differed significantly (Fig 2), thus again raising concerns about the use of small-sized PEG as a model of physiological crowding.
Overall, compared to water, both synthetic and protein crowders favored open and moderately extended conformations with higher secondary structure content. These results are difficult to explain on the basis of the “excluded volume” hypothesis. The breakdown of the energies between each protein and its surroundings reveals that in presence of both protein and PEG500 the Van der Waals term (Lennard-Jones) increase its weight compared to dilute solutions, at the expense of Coulomb interactions (see S2 Table). However, the percentage of the vdW term in PEG500 is smaller but comparable to the one in protein crowding (~ 21.5% and ~ 19.5% respectively in protein crowding and in PEG500 for NCBD, ~ 16% and ~ 12.5% for ACTR, and ~ 13% and ~ 12.5% for IRF-3), confirming that PEG500 is a non-inert crowder. In general crowding behaves as an unexpected partner, favoring protein binding through the conformational selection paradigm and acting as a chaperon that modulates the conformational space of non-ordered proteins.
The analysis of 5 independent trajectories obtained at protein concentrations from 175 to 296 g/L showed that the conformational landscape of the proteins was relatively robust to moderate changes in the concentration of the protein environment. However, detailed analysis revealed some subtle, but systematic, concentration-dependent changes (see Fig 3, and S4–S6 Figs). For example, a low concentration of protein crowders favored extended conformations, while increasing concentrations favored more collapsed structures (Fig 3). This observation suggests that the effect of protein crowding results from the combination of two opposing contributions: i) soft protein-protein interactions, which favor the exposure of protein moieties and the prevalence of extended conformations; and ii) the “excluded volume” effect, which favors collapsed structures. At low and moderate protein concentrations, the first effect dominates; however, as the number of possible protein-protein contacts is satisfied, the “excluded volume” effect gains relevance, leading to more collapsed structures. The navigation of proteins above their energy landscape can then be fine-tuned by modifying the protein concentration in different cell compartments, thereby creating an additional layer of regulation of protein structure and function.
The results above strongly suggest that soft protein-protein interactions are responsible for the crowding effect generated by a dense protein environment. A key question is whether these contacts correspond to unspecific transient (quinary) or specific interactions, the latter could not be bona fide annotated as crowding. To study this point, we compared the 3 replicas of NCBD (both for the folded and unfolded ones), where NCBD has different protein neighbors. If specific protein-protein interactions play a major role in modulating protein behavior, we can expect the 3 replicas to show distinct behaviors. This was not found to be the case (S3 Table, S4 and S7 Figs); specific interactions can therefore be ruled out as a major guide of the simulations. To further confirm this point, we performed additional trajectories with a 4x larger simulation box (4X CROW; 182 g/L protein concentration), which provided us with several replicas of the different proteins. Again, no remarkable differences were found between the sampling obtained here and the one in smaller simulation boxes (Fig 3, S5 and S8 Figs and Fig 4). Interestingly, the only remarkable exception was one of the copies of ACTR with an N-terminal exposed to a region of low protein density (labeled as A1, in red in Fig 4). There, the lack of protein-protein contacts caused an immediate response (within 100 ns) in ACTR, which underwent structural rearrangements (loss of helicity in the N-tal). These were not achieved when ACTR was surrounded by proteins. In summary, unspecific rather than specific protein-protein contacts appear as a major determinant of the effect of protein crowding.
The crowding shown here had a higher presence of disordered proteins, which were generally characterized by a higher content of charged residues. However, we did not find any significant enrichment in the type of residues located at the contact regions or any dramatic concentration-dependent changes in the inter-protein contacts (Fig 5). Intriguingly, the number of protein-protein interactions and the preference for protein vs. water contacts rose as the intrinsic disorder of the protein increased (Fig 6 and S3 Table). This observation explains why crowding effects are especially dramatic in disordered proteins. In summary, we conclude that our simulations reproduce bona-fide “crowding effects”, which are not contaminated by specific interactions that might occur in a biologically relevant cluster (IRF-3, ACTR and NCBD).
As described above, the presence of a protein environment helps the protein adopt conformations that more closely resemble bioactive ones; however, it also generates contact frustration, as the prevalence of non-specific quinary contacts hinders specific partner recognition. This frustration becomes evident by analyzing the interactions between NCBD (a total of 40 trajectories of NCBD were collected) and its partners (ACTR and IRF-3). All the crowded systems failed to reproduce the contacts observed in the experimentally solved complexes (S9 Fig). When binary complexes (NCBD:ACTR and NCBD:IRF-3) were placed in water, they rapidly adjusted to form new contacts. Remarkably, each complex rebuilt a similar pattern of contacts in most of the simulated copies in water (8 out of 10 for ACTR / F1 and 7 for ACTR / U2, see S10 Fig), thereby suggesting that these intrinsically favored contacts are frustrated in crowded conditions as a result of the presence of many competing interaction partners. Overall, the crowded box appeared as a stagnant system, where contact promiscuity generated a frustrated pattern of interaction that hindered the formation of bioactive conformations.
Protein crowding limited the accessible configurational space both globally (as noted in the number of recognized clusters) and locally (as noted by the fuzziness of the intra-protein contacts) (Fig 2 and S2 Table). The greater the “in-water” intrinsic disorder of a protein, the larger the effect of protein crowding in slowing down protein dynamics (see Fig 6 and for PEG500, where this effect was not observed, S11 Fig), Despite so, the protein underwent frequent but small oscillations that did not produce major conformational changes. As expected, the presence of protein crowders led to a significant increase in viscosity, which was reflected in the reduction of atomic movements. For example, the diffusion of water molecules was slowed down by ~ 25% from a pure aqueous environment (Table 1) [27], while global protein diffusion was reduced to 1/10 of the original value and within the same range as that reported in other studies (below 10X) [40–43]. Note that diffusion values in crowded environments should not be considered as quantitative predictions [35],[44], [51]; however, given that our results confirm those of other studies, we are confident that they still provide a valid qualitative insight. Indeed the impact on diffusion rates and related binding kinetics [8,43] is known to affect the basic functionality of the protein [40].
Our study provides a comprehensive picture of the impact of crowding on the conformational space of three proteins with different structural levels: well-structured, intrinsically disordered, and molten globule. Compared to dilute solution, both small-sized synthetic (PEG500) and protein crowders favored open and moderately extended conformations with higher secondary structure content. However, proteins in PEG500 experience a larger increase in conformational entropy, confirming observation from calorimetric analysis in presence of PEG molecules with similar dimension [7]. Overall we join the concerns regarding its employment in macromolecular crowding: detailed analysis showed that, in general, there were few similarities between the effect of PEG500 and protein crowders. This finding thus argues against the generalized use of PEG of low molecular weight to simulate crowding in physiological environments.
Our results suggest that as previously suggested in previous studies [7,11–14] the protein crowding represented here is a battlefield between two opposing forces, namely the soft protein-protein interactions and the “excluded volume” effects; the outcome depending on the concentration and type of crowder involved. Interestingly, the impact of such protein crowding strongly depends on the intrinsic structural level of proteins (in water). We found that crowding leaves the overall structure of folded proteins (such as IRF-3) almost unaffected but reduces their collapse into the hydrophobic core. This observation would explain the small expansion of the protein and the agreement with the available structures from crystals compared to pure water simulations. The impact of crowding on non-structured proteins was found to be more dramatic and complex, and it typically led to a gain in structure, bringing it closer to the ensemble of bioactive conformations and therefore favoring the conformational selection paradigm of protein binding.
Protein crowders (but not PEG500) limit the exploration of new intra- and inter-protein contacts, leading to a global decrease in conformational entropy, balanced by the enthalpic stabilization cause by quinary contacts, in good agreement again with some of the existing models of crowding [7,11–14]. The prevalence of these soft, transient and non-specific protein-protein contacts slows down solvent diffusion, as well as the global and local dynamics of proteins, thus producing frustration in native contacts, which may slow down functional flexible proteins. Crowding favors bioactive conformations, which may facilitate conformational selection processes, but on the other hand hinders the formation of functional contacts, showing then a dual effect whose impact in functionality is difficult to predict. Although we were unable to detect any bias in the type of contacts formed in the crowded box, we cannot exclude bias in the results caused by the high presence of highly charged disordered proteins. Further studies might address the differential impact as crowders for disordered and ordered proteins and discard eventual IDP-driven artifacts, as these proteins have a unique sequence composition that keep them unfolded under physiological conditions, which might bias their crowding properties. Additional work is also required to evaluate the ability of longer “inert” polymers to simulate protein crowding, as PEG500, which is very convenient to perform MD simulations shows a molecular volume much smaller than that of average proteins. We could expect that longer polymers might simulate better the crowding properties of proteins.
The cell interior differs from an aqueous dilute environment. However, it is far from a “bag full of molecules”and is possibly organized into compartments in which proteins are stabilized or destabilized in response to the specific surrounding environment [45], thereby creating an unexpected extra level of regulation of protein functionality, especially in the case of the ultra-sensitive IDPs, whose structure and dynamics can differ depending on the cellular context. From this perspective, crowding can be regarded as a collective chaperon that modulates protein conformational space.
We mixed NCBD, ACTR, and IRF-3 to obtain five dense (175, 192, 239, 273 and 296 g (of protein)/mL: protein volume fraction 20–30%) protein solutions. A stoichiometry of 6:1:1 (NCBD, ACTR and IRF-3) was used to better reproduce the central protein of the system: NCBD, for which we considered 6 starting conformations (one per copy), three of them taken from a NMR ensemble (PDB: 2KKJ) and corresponding to “folded” states (F1-F3 in the remaining), while the other three were taken from a 50-ns MD simulation at T = 500K, corresponding to a fully “unfolded” protein (U1-U3 in the remaining). The starting conformations for ACTR and IRF-3 were taken from the Protein Data Bank (PDB entry IDs 1KBH and 1ZOQ respectively) without the bounded partner. To remove any bias from the simulations, the starting positions and orientations of the proteins in the simulation boxes were random (see below) and the distance between conformations was increasingly reduced to reach more dense environments. Water molecules were added to fill the box size, calculated with a decreasing distance from the proteins (from 0.5 to 0.1 nm) and the final density was then calculated considering the proportion between water molecules and proteins. See Fig 1 for a map of the simulations performed. All these simulations were extended for at least 3 μs of unbiased dynamics.
Control simulations at a comparable timescale were performed in two environments: eight simulations (1 for ACTR, 6 for NCBD, and 1 for IRF-3) in pure water boxes; and eight additional simulations in a water:PEG500 mixture (200 g/L PEG500 concentration). In order to check for potential biases produced by the finite size of the simulation box and the use of a given set of relative orientations of the proteins, we performed one additional simulation with a ~4-times larger box containing 24 NCBD, 4 IRF-3 and 4 ACTR proteins. This huge system (~277,000 atoms at 182 g/L of concentration) was simulated for 100 ns, allowing us to collect information on each protein copy in many different surroundings.
To address the interaction of NCBD and its partners in a crowded environment, we extracted protein pairs formed by either a folded or an unfolded conformation of NCBD (F1 and U2 with ACTR, F3 and U3 with IRF-3) from the crowding simulation at 273 g/L and used them as starting seeds for multiple simulations in pure water and crowded conditions (273 g/L). For each of the four systems, 10 simulations of 10 ns were performed (reaching a total of 400 ns in water and in protein crowding respectively). These short times allowed us to exclusively study the fast relaxation of the potentially frustrated protein-protein contacts.
All starting structures were titrated, neutralized with monovalent ions, hydrated, minimized, thermalized, and pre-equilibrated using our standard procedure implemented in the MD-Web server [46]. In the case of PEG500 systems, proteins were immersed in a pre-equilibrated box of water/PEG molecules at a concentration of 200 g/L (starting PEG500 conformation PDB ID 4APO); the resulting systems were then pre-equilibrated by relaxing solvent for 10 ns prior to the general MD-Web equilibration procedure [46]. Unless otherwise stated, all the trajectories were collected with Gromacs 4.5 [47] using a time step of 2 fs in the isothermal (300 K) and isobaric (1atm) ensemble with Nose–Hoover thermostat and Berendsen barostat [48–50]. We applied periodic boundary conditions and particle Mesh Ewald corrections [51] for the representation of long-range electrostatic effects with a grid spacing of 1.0 nm and a cut-off of 1.0 nm for Lennard-Jones interactions. Constraints on chemical bonds were solved by the SHAKE algorithm [52]. The Parm99-SB-ILDN force field was used for proteins [53], TIP3P for water molecules [54], and modified TraPPE-UA parameters described by Fischer and colleagues for PEG molecules [55].
Gromacs standard routines and analysis tools in MD-Web [46] were used to analyze the trajectories, with a minimum resolution of 20 ps. We evaluated overall protein compactness using the radius of gyration (Rgyr), the deviation from a reference structure with the root mean square deviation (RMSD), the exposed surface to the outside with the solvent-accessible surface area (SASA), and the movements of each residues with the root means square fluctuations (RMSF). The secondary structure was evaluated by STRIDE [56], while VMD was used to visualize molecules and to analyze contacts [57]. The Coulomb and Lennard-Jones energy terms were calculated using GROMACS energy groups for each protein against the rest of the system. Inter- and intra-protein contacts were defined by a cutoff of 0.8 nm between alpha Carbons (Cα). Intra-protein contacts were defined as “explored” when they were found in more than five frames. Conformations recurrently sampled were detected by a two-step clustering of backbone atoms using the standard GROMOS algorithm [58]: first we reduced the total number of conformations in each trajectory with a cutoff of 0.15 nm, and then, for each protein, the reduced ensembles in WAT, PEG500 and CROW were collected together and subjected to a second clustering with a cutoff of 0.35 nm. Following Knott-Best [34], the relative orientation of the helices of NCBD was calculated by the relative elevation and azimuth between the helix vectors, defined with the axis formed by the Cα atoms in the helix of the PDB structure. The translational mean square displacements (MSD) of the center of mass of molecules were calculated to gain information on intermolecular movements (time windows of 10 and 25 ns were used for water and proteins respectively). Self-diffusion coefficients were determined using the Einstein relation, as described elsewhere, and periodic box corrections were applied [59]. Conformational entropies were approximated at the quasi-harmonic level using the last 1 μs of the simulations [60]. Finally, to detect conformational changes, we clustered the all-atom trajectory using the GROMOS algorithm [58] with a cutoff of 0.15 nm (0.1 nm for IRF-3), labeling any change in the cluster as a large conformational change.
|
10.1371/journal.ppat.1004452 | Detecting Differential Transmissibilities That Affect the Size of Self-Limited Outbreaks | Our ability to respond appropriately to infectious diseases is enhanced by identifying differences in the potential for transmitting infection between individuals. Here, we identify epidemiological traits of self-limited infections (i.e. infections with an effective reproduction number satisfying ) that correlate with transmissibility. Our analysis is based on a branching process model that permits statistical comparison of both the strength and heterogeneity of transmission for two distinct types of cases. Our approach provides insight into a variety of scenarios, including the transmission of Middle East Respiratory Syndrome Coronavirus (MERS-CoV) in the Arabian peninsula, measles in North America, pre-eradication smallpox in Europe, and human monkeypox in the Democratic Republic of the Congo. When applied to chain size data for MERS-CoV transmission before 2014, our method indicates that despite an apparent trend towards improved control, there is not enough statistical evidence to indicate that has declined with time. Meanwhile, chain size data for measles in the United States and Canada reveal statistically significant geographic variation in , suggesting that the timing and coverage of national vaccination programs, as well as contact tracing procedures, may shape the size distribution of observed infection clusters. Infection source data for smallpox suggests that primary cases transmitted more than secondary cases, and provides a quantitative assessment of the effectiveness of control interventions. Human monkeypox, on the other hand, does not show evidence of differential transmission between animals in contact with humans, primary cases, or secondary cases, which assuages the concern that social mixing can amplify transmission by secondary cases. Lastly, we evaluate surveillance requirements for detecting a change in the human-to-human transmission of monkeypox since the cessation of cross-protective smallpox vaccination. Our studies lay the foundation for future investigations regarding how infection source, vaccination status or other putative transmissibility traits may affect self-limited transmission.
| The goal of this paper is to identify epidemiological factors that correlate with either an increased or decreased risk of transmitting a particular disease. We are particularly interested in identifying such factors for diseases that are self-limited (meaning that infections tend to occur in isolated clusters), because targeted control of these diseases can facilitate public health goals for minimizing the risk of disease emergence or promoting disease elimination. For example, we show that there is a significant difference in the transmission of measles between the United States and Canada. In contrast, we find that an observed decrease in the transmission of Middle East respiratory syndrome coronavirus during the latter half of 2013 cannot be ascertained with sufficient confidence. We then quantify the degree to which control was effective in eradicating smallpox in Europe. We also consider how the transmission of monkeypox in humans depends on whether the infection source is an animal or a human. Finally, we demonstrate how our approach can be used by surveillance programs to detect changes in transmission that may occur over time.
| Many infections only occur as isolated cases, short chains of transmission, or as small infection clusters (i.e. intertwined transmission chains). Examples include zoonotic infections with relatively weak human-to-human transmission as well as vaccine-preventable infections in settings of high vaccination coverage [1]–[7]. Even though transmission is limited, these diseases are an important public health concern. For example, zoonotic infections can adapt for increased human-to-human transmission and then cause greater or even pandemic spread [8]–[10]. In addition, decreased voluntary vaccination, difficulty with vaccine delivery or changes in vaccine efficacy can allow growth of the number of individuals susceptible to preventable diseases and thus cause larger outbreaks [3], [11]. Self-limited (or subcritical) transmission also characterizes diseases that are on the brink of elimination such as smallpox during its worldwide eradication campaign or polio today [12]–[14].
Despite a need to monitor disease burden, manage the risk of disease emergence or enhance disease elimination, the surveillance and control of subcritical infections can be challenging. Resource-poor countries, which are home to many zoonoses, have many logistical hurdles that impact the quality of surveillance and control interventions. Meanwhile, even in developed countries, reactive control strategies such as isolation protocols for vaccine-preventable diseases have significant sociological impact beyond the immediate financial costs. Because of these challenges, the overarching goal is to optimize control interventions for the least amount of effort and expense. It is therefore important to gain as much quantitative information about disease transmission as possible from existing surveillance data. This includes monitoring how transmission varies with time, location and other epidemiological characteristics of individual cases. By improving the understanding of mechanisms of disease transmission, finer tuning within the spectrum of intervention strategies becomes possible [15], [16]. Such mechanistic understanding can guide the response to a diverse range of threats that include emerging infections (e.g., Middle East respiratory syndrome coronavirus), vaccine-preventable infections (e.g., measles) and antibiotic resistance [17], [18].
For ethical and logistical reasons, population-level studies of infectious disease transmission in humans typically involve retrospective statistical analysis rather than controlled prospective experimentation. Given this constraint, one approach for evaluating mechanisms underlying transmission patterns is to compare the transmissibility of two distinct, but related populations. In this manuscript, we demonstrate how the strength and heterogeneity of transmission can be compared for two different populations or types of infection sources. We then show how our framework provides insight into the transmission patterns of a variety of subcritical diseases. This analysis builds upon earlier studies that were limited to estimating transmission parameters from chain size distributions and addressing issues of surveillance bias [19], [20].
Mathematically, the transmissibility of a group of infected individuals can be quantified by determining the group's effective reproduction number, . This number represents the mean number of secondary cases caused by an infected case. However, because of the stochastic nature of disease transmission, the realized numbers of secondary infections caused by a given infected individual will vary. is a more general parameter than the oft cited basic reproduction number , which more specifically represents the mean number of secondary cases caused by the first infected case in a completely susceptible population [21]. When , transmission cannot reach epidemic proportions, whereas if there is a potential for epidemic spread. Thus, our focus on subcritical diseases implies that, overall, will be less than one and transmission will be characterized by self-limited clusters of infection. However, our method still permits the possibility that cases can be divided into two groups in which one group has a , and the other group has a .
Our study builds upon the prior success of inferring from the size distribution of observed transmission chains [1], [2], [22]. The same distributions can also be used to infer the degree of transmission heterogeneity, represented by the dispersion parameter, [19], [20], [23]. A high degree of heterogeneity represents a scenario where some individuals are predisposed to spreading infection to a larger number of people (i.e., ‘superspreaders’). When models of chain size distributions incorporate both and , excellent agreement can often be found between observed data and model predictions [19], [20], [23].
Our goal is to evaluate specific hypotheses regarding disease transmission by testing whether and differ between two groups of cases. Our analyses differ from more traditional epidemiological approaches based on case-control studies (and many other study designs) in that we focus on transmissibility instead of individual-level risk factors for disease susceptibility. We demonstrate our methodology by considering four subcritical infections (MERS-CoV, measles, monkeypox and smallpox) and three types of data (size distribution of infection clusters, transmission chain data and infection source classification) to answer four different questions based on published data. For MERS-CoV, we use chain size distributions to determine whether an apparent decrease in during the latter half of 2013 was statistically significant. Assessing temporal trends of has important implications for evaluating the risk of endemic MERS-CoV transmission and the impact of control interventions. For measles, we use chain size distributions to compare two locations (United States and Canada) and test whether there is a significant difference in , which would suggest important differences in vaccine distribution, social connectedness, and/or demographics. For smallpox and monkeypox, we use case series resolved by infection generation to determine whether there are significant differences between the first and subsequent generations of spread [24], [25]. This analysis allows us to assess whether variation in the number of contacts or the timing of control interventions can be linked to changes in . It also allows us to test the validity of a specific ‘random network’ model that relates the contact patterns of primary and secondary cases. We then test whether there is a significant difference between inferred transmission parameters for animal-to-human and human-to-human transmission of monkeypox, which provides insight into the mechanisms of zoonotic spillover. Our analysis of chain size distributions also provides perspective on the surveillance required to detect a change in , such as the expected increase in human monkeypox transmission following the eradication of smallpox. Each of the scenarios considered represents a unique example of how quantitative characterization of transmissibility can provide insight into the effectiveness of control interventions and risk assessment for future spread.
The stochastic nature of infectious disease transmission is particularly important when , as it can result in substantial variation in the size distribution of transmission chains. In this case it is helpful to model transmission as a branching process [26]. In this formulation, the offspring distribution specifies the probability that an infected individual will cause new infections. We specify the corresponding offspring probabilities to be , with . To facilitate likelihood calculations (as seen below), the offspring distribution can be represented as a generating function, , in which the polynomial coefficients are the offspring probabilities [26]–[28].
In line with research demonstrating how the strength and variability of transmission can be modeled [23], we assume the qi's follow a negative binomial offspring distribution with a mean of and a dispersion parameter of . The dispersion parameter represents the degree of transmission heterogeneity, with lower values of corresponding to higher variance. The supplementary methods (Text S1) explains how our simple model of disease transmission can be used to calculate the likelihood for various types of observed data. These likelihood calculations permit inference of the strength and variability of transmission for individual cases, in terms of and . All calculations were conducted with either Matlab or R. Code for all analyses is available at: https://github.com/sbfnk/nbbpchainsizes.
By calculating the likelihood of an observed set of transmission events, we can probe whether there is statistical support for differences in transmission between two pre-specified populations, and . In our general model, the two types of individuals have distinct negative binomial characterizations and thus there are four parameters in total. We label these four parameters , , and with the subscripts corresponding to the type of individual. Five simpler models that are nested within the 4-parameter model can be constructed by assuming , and/or (Figure 1). The specific test case of is chosen for the nested models because this corresponds to a geometric offspring distribution which is the expectation for a traditional SIR or SEIR model. These models assume homogenous mixing with constant infectivity over an exponentially distributed infectious period [29]. For each model, we determine the parameter values (MLE) that maximize the log-likelihood. The 95% confidence intervals and confidence regions shown in the figures were found by profiling on and/or and employing the likelihood ratio test [30]. Model comparison is accomplished via the Akaike Information criterion (AIC) [31].
To identify whether there is statistical support for a difference in for two data sets, the AIC scores were computed for all six aforementioned models. A difference in was deemed statistically significant according to the rule that the model with the best AIC score cannot be within two AIC units of a model that supports identical values of for the two sets of simulations. This rule is in approximate alignment with the commonly used likelihood ratio test for establishing statistical support for the use of an extra parameter with 95% confidence, but we could not employ the likelihood ratio test explicitly because some pairs of models we consider are not nested. We verified the internal consistency of our modeling framework by applying this method to simulated data (Supplementary material, Text S1).
We used parametric bootstrapping to evaluate the type I error and the power for detecting a change in for our analyses. Specifically, for every analysis we simulated 20,000 new data sets. Each simulated data set replicated the two populations involved in the analyses (e.g. MERS-CoV chains before and after June 1, 2013). Two models were simulated. Half of the simulations used two distinct values of and that matched the inferred values of our unrestricted four-parameter model. The other half of the simulations used a single value of and that matched the inferred values of our two-parameter model, which requires both and to be the same for all cases seen in the observed data. Our inferential algorithm for ascertaining a statistically significant difference in the inferred value of was then applied to all simulations. The type I error of an analysis (i.e. the probability that the analysis would falsely claim that is different for the two types of cases considered) was estimated as the proportion of simulations based on the two-parameter model that were found to have a statistically significant difference in for the two types of cases. The parametric bootstrap probability (or power) of detecting a change in was estimated as the proportion of simulations based on the four-parameter model that were found to have significant difference in for the two types of cases.
Data used to generate all results can be found in the supplemental material (Text S2).
Since 2011, there have been over 500 confirmed cases of MERS-CoV, and over 140 associated deaths, suggesting a case fatality rate of 28% [32]. The persistent occurrence of small outbreaks is due to zoonotic spillover [33]–[35]. MERS-CoV may be a new virus, as the most recent common ancestor of viral samples from infected patients was estimated to have occurred after September 2010 [34]. The novelty of this virus and its high case fatality rate underscore the significance of monitoring the transmission of MERS-CoV. Although human-to-human transmission has been relatively limited so far, with likely less than one, there is concern that future adaptation that could lead to spread similar to sudden acute respiratory syndrome (SARS) in 2003. Health authorities have prudently instituted a variety of infection control policies and procedures and a trend towards decreasing has been reported [34]. Since verification of the effectiveness of control has important implications, we reconsidered the evidence for a trend towards decreasing .
To avoid artifacts of assembling multiple data sources, we restricted our analysis to the previously reported chain size distribution for all MERS-CoV cases in the Arabian Peninsula occurring before August 8, 2013 [34]. Previous analysis of these data shows that is 0.74 (95% CI 0.53–1.03) before June 1, 2013 and 0.32 (95% CI 0.14–0.65) after June 1, 2013. Our results replicate the finding that independent evaluation of cases before and after June 1, 2013 results in an estimate of 0.7 and 0.3 for respectively (Figure 2 and Table 1). When our six models are compared, we do not find statistical support for models with different values of before and after June 1, 2013. This is again consistent with the results of prior studies that determined a p-value of 0.07 for change in , but our analysis allows the possibility of a high degree of transmission heterogeneity.
Local elimination of measles is dependent on vaccination programs, and the potential for re-emergence necessitates continued surveillance and re-assessment of vaccination strategy [1], [3], [36]–[38]. Even where elimination has been achieved, there can be sporadic clusters of infection due to a combination of geographic importation and pockets of susceptibility [39]–[41]. Geographical differences in transmission may arise due to differences in cultural practices, public health guidelines, population density and other factors. Methods that delineate whether differences in are statistically significant for two different regions can therefore help to identify key differences in transmission potential and thus pinpoint opportunities for improved control.
Measles data in the United States (1997–1999) and Canada (1998–2001) are reported according to the size of infection clusters [39], [40]. Most infection clusters have a single primary infection, but even when multiple primary infections exist (as in the case of a cluster with six cases in the United States), the likelihood calculation needed for assessing differences in is straightforward (Supplementary Material, Text S1). When the two data sets are compared, the results indicate that for the United States and Canada are significantly different (Figure 3 and Table 2). Meanwhile, the results also confirm previous studies that infer a high degree of transmission heterogeneity in measles transmission [19], [23]. This can be seen from Table 2 since the MLE estimates for and are less than one and the value of the model with is large. On the other hand, there is negligible statistical support for distinct values of in the two countries. The type I error for this situation was estimated to be 4.9% by parametric bootstrapping.
Smallpox is the only human disease to have been eradicated and thus represents a tremendously successful use of control [12]. During the endgame of smallpox eradication in the middle of the 20th century, smallpox cases in Europe resulted in rapid implementation of quarantine and control procedures. Transmission data for smallpox infections in Europe that occurred during this period provide an opportunity to investigate how control interventions impacted the transmissibility of primary cases caused by geographic importation relative to secondary cases resulting from local transmission [12].
Smallpox clusters were tabulated according to the number of cases in each generation of spread [12]. The inference results indicate that secondary cases transmitted significantly less than primary cases (seen by the lack of overlap of contours with the grey line in Figure 4 and by the statistical selection of the non-restricted model in Table 3). In fact, the effectiveness of control procedures can be quantified by looking at the ratio of reproduction numbers for primary and secondary transmission (Figure 4 inset). The ratio of the maximum likelihood values for to suggests that control reduced by 75%. Meanwhile, for both primary and secondary transmission, a high degree of transmission heterogeneity is evident (since the MLE estimates of and are substantially less than one and the value of the model is large). Based on selection of the unrestricted model, and the associated estimates of , there appears to be significantly more heterogeneity of disease transmission for secondary cases than for primary cases. The type I error for this analysis was estimated to be 5.1% by parametric bootstrapping.
Following the eradication of smallpox in 1979, the World Health Organization was concerned that subsequent cessation of smallpox vaccination would allow other diseases to flourish [42]. Monkeypox was of particular concern because exposure to smallpox or smallpox vaccination provided protection against monkeypox. Estimates of , extrapolated from contact tracing data gathered during rigorous surveillance in the Democratic Republic of Congo (formerly Zaire) during 1981–1984, provided re-assurance that endemic transmission would not be sustainable even when population immunity to monkeypox waned [43].
The initial analysis of monkeypox transmission did not quantitatively compare the transmission of primary cases (i.e. those caused by animal-to-human transmission) to the transmission of secondary cases (i.e. those caused by human-to-human transmission). Since the characteristics of these cases differ (i.e. only primary cases required exposure to infected animals), differences in transmission are possible. Increased transmission of secondary cases could also arise from population structure [25], or evolutionary adaptation [8], [10]. For example, network models have proposed that social structure impacts the effective reproduction number of individual cases [44]–[48]. In particular, the random network model that we have considered (Supplementary material, Text S1) predicts that secondary cases transmit more than primary cases since highly-connected individuals are most likely to both acquire and spread infection. If this aspect of the random network model is accurate, the risk of endemic spread as population immunity wanes may be higher than previously expected. This is because for secondary transmission would be expected to increase more than for primary transmission. It is thus important to ascertain whether there is a difference between primary and secondary transmission that is consistent with the random network hypothesis.
As part of the monkeypox surveillance efforts, transmission was tabulated according to the number of cases in each generation of spread [43], [49]. These data can be used to ascertain whether there is a statistically significant difference in primary versus secondary transmission (Figure 5 and Table 4). The results indicate a lack of evidence for a difference between the of primary and secondary cases (seen by noting the overlap of contours with the grey line in Figure 5 and because the preferred model in Table 4 has ). The low values for the maximum likelihood estimates of are consistent with previous studies that infer a high degree of transmission heterogeneity in monkeypox transmission [20], [23].
Animal-to-human transmission of monkeypox is an important contributor to overall disease burden. Determining the factors that allow continual introduction of monkeypox into human populations requires knowledge of how monkeypox maintains itself in reservoir hosts and the mechanisms that allow its transmission to humans [6], [50]. In this section we assess whether an infected animal in contact with humans has a distinct set of inferred transmission parameters than infected humans. The relationship between infection source and transmissibility is an active area of research for many multi-host diseases systems [51]–[55], particularly for zoonotic infections.
Since the infection cluster data for monkeypox contains information on how many primary infections are in each cluster, it can be used to infer the amount of animal-to-human transmission that occurs when infected animals make contact with humans. To accomplish this, we assume that the negative binomial offspring distribution that has been shown to be a good description of human-to-human transmission [23] is also an effective model of animal-to-human transmission. We let represent the average number of primary cases caused by an infected animal that has contact with humans. Our results indicate that the for human-to-human transmission is similar to (Figure 6 and Table 5). There is also evidence that animal-to-human transmission is relatively homogeneous (since the for the preferred model). If one takes the MLEs of and for the preferred model at face value, then we estimate that at least one infection occurs 25% of the time that a infected animal has contact with humans.
Recently, a 20-fold increase in the incidence of monkeypox has been reported in the Democratic Republic of Congo [56], and there is concern that for monkeypox may have increased. The lack of cross-protective immunity to monkeypox from either smallpox vaccination or natural exposure to smallpox provides a mechanism for why would increase [57]. However, land-use changes that impact the potential for animal-human transmission have also been suggested as a cause of an increase in monkeypox incidence [58], [59], and could do so without changing . There are no active interventions in place for monkeypox, so it is important to determine if has changed in order to understand the source of increased incidence.
Due to logistical barriers and the rare nature of the disease, acquiring data on monkeypox is a challenge [42], [56]. In the wake of smallpox eradication, the infrastructure for monkeypox surveillance in 1980–1984 was strong and well funded [42]. The detailed transmission data from this surveillance effort provide an estimate of 0.30 for (95% CI: 0.21–0.42) and 0.33 for (95% CI: 0.17–0.75) [20]. For the 2005–2007 surveillance effort, specific data on cluster sizes and individual-level transmission are unavailable, so an assessment of cannot be made. However, we can quantify the amount of data that would be needed in order to detect a change in relative to 1980–1984 [42], [43], [49]. Simulations show that 200 clusters would provide 70% power to detect an increase in from 0.3 to 0.5 (Figure 7A). As the number of observations increase, smaller changes are more readily noticeable.
Consideration of the relationship between , the number of chains and the number of cases provides perspective on the power of the recent surveillance efforts (2005–2007) to detect a change in [56]. It appears that there is 95% power to detect an increase in from 0.3 to 0.55 with analysis of the 760 observed cases (Figure 7B).
In summary, we have introduced and validated a method for comparing case data grouped into different categories and applied this method to a number of different scenarios. The versatility of the method has been explored through examination of a variety of diseases and data types. By providing quantitative information on transmission, surveillance needs, or the effectiveness of control interventions, each type of analysis has the potential to assist in epidemiological assessments and public health planning.
To reduce the burden of MERS-CoV and reduce the risk of global spread, effective control procedures are of obvious importance. Given the large amount of resources and effort that have already been directed towards the control of MERS-CoV, it would be reassuring to see a statistically significant decrease in . When analyzing data on MERS-CoV cases that presented before Aug 8, 2013, the unrestricted model had the best score. This unrestricted model suggested that because decreased from 0.7 to 0.3, control is over 50% effective. However, there is not enough data to show statistical significance for this result. Meanwhile, our analysis is likely biased by the large outbreak that initiated the observational period for the data, so further studies are needed to more accurately evaluate the impact of control interventions [60].
Unfortunately, the number of recent confirmed MERS-CoV cases remains significant and the overall incidence may be increasing [32]. An increase in the number of cases can be caused by an increased , an increased rate of primary cases, or a combination of these effects [61]. Based on our observation that is more likely to be decreasing after June 2013 than increasing, the paradigm of emergence that is most consistent with the previously published data we have analyzed is that MERS-CoV incidence may be increasing in its non-human reservoir, but that human-to-human transmission remains stable. In fact, sequence data support the possibility of an expanding epidemic in animal hosts of MERS-CoV that could lead to an increased incidence of primary cases [34]. However, other factors, such as seasonal drivers of transmission could also impact the temporal trend of . An increased case load could also be observed if transmission patterns have not changed much, but greater interest in and knowledge of MERS-CoV has led to improved surveillance. This could paradoxically lead to both an increase in the number of observed cases and a decrease in the observed value of because of a greater chance of seeing a larger proportion of smaller outbreaks [19], [62].
Given the relative paucity of cases and uncertainties regarding case observation probability, it would be inappropriate to make a definitive statement concerning the cause of the apparent increase in MERS-CoV incidence at this time. However, as more data on MERS-CoV are reported, the types of analyses presented in this manuscript can be rapidly applied to address hypothesis-driven questions concerning the temporal trends of incidence and the impact of control intervention. In particular there may be concerns that certain subgroups of MERS-CoV cases may have increased transmission, such as those occurring in health care settings where nosocomial transmission is higher or in geographic regions where control interventions are harder to implement. Alternatively, as we have shown with smallpox, there may be a difference in the transmissibility of primary cases versus secondary cases. With more data, our method can help to quantify differences in transmission, and evaluate whether certain population subgroups may have an that exceeds the critical value of one. While it is not necessary for future data to be resolved to the level of individual transmission events, the types of analyses we have presented do require knowledge of chain size distributions rather than aggregate epidemic curve data. Meanwhile, an important gap in the currently available data is a quantitative assessment of the case reporting probability for MERS-CoV cases and whether this is increasing with time. Improved knowledge of the reporting probability would permit adjustments to the likelihood calculations and reduce the bias of imperfect case ascertainment [19].
Our comparison of measles transmission in the United States and Canada provides a framework for elucidating geographic differences in transmission (Figure 3). Interestingly, while our analysis supported a difference in between the two countries, a difference in the degree of transmission heterogeneity (as quantified by the dispersion parameter) was not identified. This apparent disassociation between the strength of transmissibility and the mechanisms of transmission heterogeneity may occur if the heterogeneity is due to intrinsic biological processes such as variability in viral shedding. However, the relationship between the value of dispersion parameter and various mechanisms of transmission heterogeneity is not straightforward so the interpretation of similar values of dispersion is unclear.
There are many reasons why the value of may differ between the United States and Canada. One consideration is a potential difference in the timing of the introduction of two-dose vaccination. The Advisory Committee on Immunization Practices and the American Academy of Pediatrics recommended two-dose coverage in 1989 [63]. Although the coverage in 2004 appeared similar between the United States and Canada [38], it is unclear whether this level of coverage was achieved at the same time in both countries. To assess whether a difference in vaccine coverage explains the difference in observed here, it would be helpful to run a similar analysis on more recent data. Other factors that could contribute to the difference in include a greater tendency in the United States to conduct contact tracing for susceptible cases and vaccinate close contacts, a greater sensitivity in Canada for reporting milder cases of measles, or greater difficulty of detecting isolated cases via passive surveillance in Canada [37], [38]. More detailed information of the impact of contact investigation, stratification of cases based on disease severity, and quantitative comparison of case ascertainment in passive versus active surveillance would provide additional insight.
Smallpox control is already known to have been very effective; however, our analysis of smallpox transmission in Europe around the time of eradication quantifies the impact of interventions for control (Figure 4) showing that there was a reduction of for secondary cases by 75% compared to primary cases. This effect of control may be an underestimate because it does not account for the possibility of late arrival of imported cases during the course of infection. Since the infectious period of imported primary cases may have occurred outside of the country of residence, the actual for primary cases might be higher than seen in the data and thus the effect of control may be even greater than our estimates indicate.
Here we have shown how for each generation can be quantitatively compared, using published transmission data. Our analysis of differences in the transmissibility of cases as an outbreak develops is not unique (see for example [64]). However, previously published methods rely on symptom-onset data to determine at various stages of an outbreak and thus these approaches could not be performed on the smallpox data set.
Aside from the change in , the marked increase in degree of transmission heterogeneity for secondary cases (as evidenced by a decreased in the observed value of ) suggests that control tended to be individual-specific rather than population-wide. Here, individual-specific control refers to an intervention that is completely effective for 75% of cases but not effective at all for the remaining cases, whereas population-wide control refers to an intervention that reduces the transmissibility of each case by 75% [23]. For individual-specific control, a large number of cases become dead ends for infection so the observed degree of heterogeneity increases [19], [20]. In contrast, the observed degree of transmission (as quantified by the dispersion parameter) would not change for population-wide intervention. The support for individual-specific control is highly consistent with the quarantine and ring vaccination methods employed during smallpox elimination efforts [12]. These observations show how understanding the variation in both the strength and heterogeneity of transmission can provide insight into disease dynamics.
Our analysis of monkeypox in the Democratic Republic of Congo demonstrates how our method can be used to inform surveillance planning. In particular, by determining the number of chains that needed to be observed in order to detect various degrees of change in , we provide perspective regarding the extent to which the 760 monkeypox cases observed between 2005 and 2007 [56] can provide enough information to detect increased transmissibility (Figure 7). Based on our power analysis, it appears that a change in due to declining population immunity should be detectable, since is expected to approach [43]. However, this result needs to be interpreted in context because our model assumes that the probability of case observation is high and that distinct infection clusters can be determined. Given the logistical challenges of recent surveillance efforts [56], these assumptions are unlikely to have been met, so the realized power for detecting a change in is probably lower. Nevertheless, this simulation analysis provides perspective concerning the trade-offs of thoroughness in detecting and characterizing cases versus observing cases within a greater catchment area for any future surveillance efforts for which measurement of is of interest.
When we focused on more detailed generation-level data for monkeypox transmission from 1980–1984, we found no support for enhancement of by highly-connected individuals in secondary generations (Figure 5). This suggests that the high degree of transmission heterogeneity may be caused by biological factors, rather than variability in social contact. However, a key assumption of the network model we tested is that primary cases are infected at random relative to their degree (as might reasonably be expected for a zoonotic infection). It may be that high-connected individuals are also more likely to get a primary infection. If this were the case, then highly connected individuals would contribute to heterogeneity of both primary and secondary transmission. Meanwhile, the lack of increased for secondary transmission provides assurance that significant viral adaptation is not occurring, although local depletion of susceptible individuals within small sub-networks such as households could obscure signals of viral adaptation.
We found that humans and animals in contact with humans produce similar numbers of human cases (Figure 6). Moreover, we estimated that 25% of human exposure to an infected animal lead to at least one detected human case. While the truncated negative binomial distribution produces unbiased estimates of transmission parameters, the confidence intervals can be quite large [19]. Furthermore, the a priori specification that the offspring distribution will be characterized by negative binomial distribution is a strong assumption. Thus the inferred proportion of animal-to-human exposures leading to infection deserves cautious interpretation. Nevertheless, this type of analysis could be useful for informing surveillance and detection efforts in wildlife species. In particular, since the overall incidence of monkeypox is quite low (14.42 per 10,000 per year [56]), the observation that there may be only 4-fold more infected animals in contact with humans than the number of observed infection clusters provides perspective on the fact that monkeypox virus has only been isolated from one wild animal (as of 2011) [58]. If contacts with infected animals account for a small proportion of overall human contact with reservoir species, the use of targeted-surveillance strategies that can exploit spatial-temporal data to identify likely hotspots of incidence [58], [59], [65] may be essential to improve detection efforts in wildlife hosts.
As with any model selection or measurement scheme, a small portion of the data, or even a single data point, can have a particularly large influence. For example, the largest transmission chain in the Canadian measles data consists of 155 cases while the second largest chain has just 30 cases. Moreover, the chain with 155 cases was associated with a religious community that resisted immunization, thus it could be argued that this chain is not representative of the population as a whole. If the 155-case chain were excluded from the analysis, our method would no longer find statistical support for a difference in between the United States and Canada (Supplementary material, Text S1).
However, rather than excluding a possible outlier, our preference is to treat the data at face value. From a modeling perspective, it is often unclear whether the mechanism responsible for a purported outlier is absent in the rest of the data. For example, in the case of Canadian measles data set, the second largest chain of 30 cases was also associated with a religious community. In addition, a particularly large chain does not represent a single large transmission event, but rather an entire group of individuals who collectively had relatively high transmission. Mathematically, a high degree of transmission heterogeneity (represented by low values of ) is expected to have a big tail for the distribution of the number of cases that each case causes [23]; thus, a large transmission event or chain in a set of data will increase the estimated value of , but will also decrease the estimated value of . A lower will be associated with a wider confidence interval for and this would make it harder for our analysis to find a statistically significant difference in [19], [20]. Thus our modeling framework has a built-in mechanism that compensates for large transmission events and chains that are consequences of intrinsic population-level or individual-level mechanisms of heterogeneity.
A key caveat of our analyses is that we have assumed perfect observation of cases. Some surveillance programs, such as measles in the United States, have documented evidence of high case observation [36]. However, this level of case ascertainment cannot be expected of all diseases, particularly those such as MERS-CoV that are quite new. Meanwhile, even meticulously collected data are prone to multiple sources of observation bias due to limited surveillance resources, subclinical infections, laboratory error, or other factors.
When the limitations of observation can be quantified, likelihood calculation for observed transmission events can be adjusted appropriately [19], [20]. The challenge is that the limitations of surveillance systems and case ascertainment are often difficult to quantify. An alternative to explicit correction of observation bias is to simply consider what level of observation bias would impact key results. For example, in our analysis of the difference between animal-to-human and human-to-human transmission of monkeypox, it is quite possible that a number of animal-to-human infections are unobserved — particularly if the resulting primary infection is mild and has no further transmission. When we treated observation of an infection cluster as an all-or-none process with an independent probability, , that each case would activate surveillance (thus implying many isolated cases would be unobserved), our preferred model of transmission remained stable even for a of 0.1 (Supplementary material, Text S1). This provides re-assurance that our methodology is not necessarily sensitive to imperfect observation. However, different data sets or a different type of observation bias could yield less stable results.
In our analyses, we have allowed for at most two values of and in a data set rather than permitting additional stratification or a continuous distribution of values. These simplifications are not always valid assumptions. However, modifications to the likelihood calculation can often be made in order to accommodate more complicated data sets so that our framework for detecting a difference in can be utilized. For example, the offspring generating function used for the likelihood calculation can be written in terms of a continuous variable that provides a smooth transition between the extreme limits of classification. In fact this approach has been used to investigate whether there is a temporal trend of measles transmissibility in the United States [61].
Although we have mainly focused on differences in between two populations, our method can also be used to identify whether these populations differ in the observed degree of heterogeneity. Clustering of individuals with higher transmissibility may favor models with two distinct values for whenever two distinct values of are observed. Meanwhile, situations that would favor a model with two distinct values of and one value of could arise if different mechanisms of control were used to maintain below a given threshold, as seen in the smallpox example. Regardless of which model is the preferred model for a given data set, the estimated or assigned value of can be useful to assess the overall degree of transmission heterogeneity and the likely presence of super-spreaders [20], [23]. On the other hand, the specific mechanism of heterogeneity (e.g. differences in transmission potential among cases versus clustering of susceptible individuals) cannot be ascertained from estimation of alone.
Our analysis is focused on determining whether there is statistical support for a difference in for individuals having a specific trait. Also, as exemplified by our direct comparison to the random network model (Figure 5 and Table 4), we can evaluate specific models of transmission. However, in the absence of a mechanistically derived model, our analysis cannot identify the cause of differences in . For-example, population-level factors favoring transmission (e.g. increased human density) cannot be directly distinguished from biological factors (e.g. evolutionary adaptation). Furthermore, the decrease in secondary transmission due to local depletion of susceptibles cannot be directly distinguished from decreases due to control mechanisms. Instead, our method needs to be considered as a tool that can identify differences in transmission (e.g. temporal trends for MERS-CoV, and geographic distinctions in measles) or quantify changes in transmission that are expected to occur (e.g. decreased transmission due to quarantine of smallpox cases or ring vaccination).
By addressing diverse questions within varied data sets, we have demonstrated that a set of inter-related models within a branching process framework allows rigorous statistical assessment of whether particular characteristics of infectious cases impact transmission potential. We have focused on subcritical diseases, in large part because the type of surveillance data gathered for these diseases is most compatible with our computational approach. For MERS-CoV, we evaluated the possibility of a temporal trend towards decreasing that may indicate stronger control, but did not find enough statistical evidence to confirm this finding. For measles, we found evidence of geographic variability that provides potential insight into the effectiveness of surveillance and public health interventions. For smallpox, we identified signatures of effective control by comparing primary and secondary transmission. For monkeypox, we found that the most parsimonious models are ones that incorporate a high degree of transmission heterogeneity, but do not differentiate between animal-to-human transmission, transmission of primary cases, and transmission of secondary cases. In general, the statistical support we observed for models that allow flexible inference of both and reinforces the importance of quantifying both the strength and variability of disease transmissibility. By providing a diverse array of applications and analyses, the method we have demonstrated can increase the value of existing surveillance data and improve strategies for future data collection. Through identifying specific risk factors for transmissibility and by assessing different sources of transmission heterogeneity, we hope that disease monitoring and control interventions can become more targeted and thus more effective.
|
10.1371/journal.ppat.1007959 | SopF, a phosphoinositide binding effector, promotes the stability of the nascent Salmonella-containing vacuole | The enteric bacterial pathogen Salmonella enterica serovar Typhimurium (S. Typhimurium), utilizes two type III secretion systems (T3SSs) to invade host cells, survive and replicate intracellularly. T3SS1 and its dedicated effector proteins are required for bacterial entry into non-phagocytic cells and establishment and trafficking of the nascent Salmonella-containing vacuole (SCV). Here we identify the first T3SS1 effector required to maintain the integrity of the nascent SCV as SopF. SopF associates with host cell membranes, either when translocated by bacteria or ectopically expressed. Recombinant SopF binds to multiple phosphoinositides in protein-lipid overlays, suggesting that it targets eukaryotic cell membranes via phospholipid interactions. In yeast, the subcellular localization of SopF is dependent on the activity of Mss4, a phosphatidylinositol 4-phosphate 5-kinase that generates PI(4,5)P2 from PI(4)P, indicating that membrane recruitment of SopF requires specific phospholipids. Ectopically expressed SopF partially colocalizes with specific phosphoinositide pools present on the plasma membrane in mammalian cells and with cytoskeletal-associated markers at the leading edge of cells. Translocated SopF concentrates on plasma membrane ruffles and around intracellular bacteria, presumably on the SCV. SopF is not required for bacterial invasion of non-phagocytic cells but is required for maintenance of the internalization vacuole membrane as infection with a S. Typhimurium ΔsopF mutant led to increased lysis of the SCV compared to wild type bacteria. Our structure-function analysis shows that the carboxy-terminal seven amino acids of SopF are essential for its membrane association in host cells and to promote SCV membrane stability. We also describe that SopF and another T3SS1 effector, SopB, act antagonistically to modulate nascent SCV membrane dynamics. In summary, our study highlights that a delicate balance of type III effector activities regulates the stability of the Salmonella internalization vacuole.
| Pathogenic bacteria that adopt an intracellular lifestyle must create a specialized niche that supports their replication while avoiding detection and killing by the host. The foodborne pathogen, S. Typhimurium, invades epithelial cells lining the intestine and establishes residence within a host-derived membrane-bound compartment called the Salmonella-containing vacuole (SCV). To create this intracellular niche, S. Typhimurium relies on two type III secretion systems (T3SS). These are specialized needle-like devices that inject bacterial effector proteins into host cells, and thereby manipulate host cell signalling and immune responses. T3SS1 delivers effectors that are required for early events in the infectious cycle. In this study we describe a new T3SS1 effector called SopF that is unique to Salmonella spp. We show that SopF associates with host cell membranes by binding to phosphoinositides, which are specialized lipids present in eukaryotic cellular membranes, and that SopF is required for maintaining the integrity of the nascent SCV membrane. Salmonella has therefore evolved to reside within a membrane-bound compartment by acquiring a unique type III effector whose actions promote vacuole stability.
| Many pathogenic bacteria of public health significance undergo an intracellular cycle as part of their virulence strategy. The ability of these bacteria to direct themselves to a specific intracellular locale is key to their pathogenesis, not only determining their survival and proliferation, but ultimately their virulence. Once internalized, a bacterium can either remain confined within a membrane-bound compartment or lyse its nascent phagosome and colonize the eukaryotic cytosol. The fundamental processes governing intracellular niche selection are poorly understood.
Salmonella enterica serovar Typhimurium (S. Typhimurium), a common cause of foodborne gastroenteritis, is a facultative intracellular pathogen that can colonize epithelial cells, dendritic cells, macrophages and fibroblasts. Within these cells, S. Typhimurium resides within a membrane-bound compartment known as the Salmonella-containing vacuole (SCV) [1], that extensively, but selectively, interacts with the host cell endocytic pathway [2]. A critical virulence determinant for S. Typhimurium is a specialized injection device, the type III secretion system (T3SS). T3SSs are needle-like multi-protein complexes that penetrate host cell membranes to inject bacterial proteins, termed type III effectors, directly into the host cell cytosol. S. Typhimurium translocates ~40 effector proteins [3–6] using two T3SSs, T3SS1 and T3SS2, which are encoded on Salmonella Pathogenicity Island (SPI)-1 and SPI-2, respectively. Based upon their timing of expression, T3SS1 effectors are primarily associated with early events in Salmonella-host cell interactions such as promoting bacterial entry into non-phagocytic cells and establishment of the nascent SCV [7], whereas T3SS2 effectors contribute to later events including maturation of the SCV and bacterial replication and survival within the host cell, particularly phagocytic cells [8,9].
T3SS1 engages the host plasma membrane to translocate five effectors—SipA, SipC, SopB (also known as SigD), SopE and SopE2—that target the actin cytoskeleton [10] to promote the formation of large lamellipodia-like surface projections, known as membrane ruffles, which engulf and internalize S. Typhimurium [11,12]. SopA is also required for the efficient invasion of S. Typhimurium, but only in polarized epithelial cell lines and via an unknown mechanism [13]. The T3SS1 effector SptP manipulates signalling networks to counteract the effects of the “entry” effectors and down-regulate plasma membrane ruffling [14]. Another T3SS1 effector, SopD, cooperates with SopB to promote SCV membrane fission from the plasma membrane [15]. Large macropinosomes are formed early after S. Typhimurium internalization and these eventually fuse with the nascent SCV [16]—the cooperative actions of SopD and SopB are also required for efficient macropinosome formation [15].
In addition to the plasma membrane, T3SS1 also translocates type III effectors across the nascent SCV membrane and these effectors collectively act to modulate SCV-host endocytic pathway interactions. For example, SopE and SopB mediate recruitment of the GTPase, Rab5, to early SCVs to promote their fusion with early endosomes [17,18]. Rab5 in turn recruits Vps34, a phosphatidylinositol 3-kinase (PI3K), to generate phosphatidylinositol 3-phosphate (PI(3)P) on the nascent SCV [18]. PI(3)P then recruits the SNARE protein VAMP8 [19]. Rab5 and Vps34 recruitment, subsequent PI(3)P production, and VAMP8 recruitment are all dependent on the inositol phosphatase activity of SopB [18,19]. SopB also reduces levels of negatively charged lipids on the nascent SCV, specifically PI(4,5)P2 and phosphatidylserine, which alters the surface charge of the vacuole membrane [20] and thereby delays its fusion with late endosomes/lysosomes [21,22]. In support of their important role in post-entry events, numerous T3SS1 effectors (i.e. SopE, SopE2, SipA and SopB) can be detected for many hours post-internalization and/or are required for the efficient replication of bacteria in epithelial cells [22–28].
T3SS needle insertion into bacteria-containing vacuoles is believed to “damage” their membranes. Decoration of damaged vacuoles by host cell galectins acts as a danger signal to target these bacteria for autophagic capture and thus restrict their growth [29–31]. Although typically considered to be a vacuolar pathogen, a sub-population of S. Typhimurium lyse their internalization vacuole and escape into the cytosol of mammalian cells in a T3SS1-dependent manner [28,32–34]. S. Typhimurium in damaged vacuoles are recognized by the autophagy machinery [30,32]. The frequency of nascent SCV lysis and eventual outcome are dependent upon the cell type. More bacteria lyse their nascent vacuole in epithelial cells (10–20% by 90 min post-infection (p.i.)) compared to macrophages (2–6% by 90 min p.i.) [28,32,34–36], possibly due to bacterial entry being largely driven by T3SS1 in non-phagocytic cells. Once in the cytosol, S. Typhimurium hyper-replicate in epithelial cells, whereas in macrophages they do not. This difference might be explained by: (i) incomplete autophagic clearance in epithelial cells, (ii) autophagy-mediated repair of damaged vacuoles in epithelial cells, as has been described for fibroblasts [37], (iii) autophagy supporting cytosolic replication in epithelial cells [38], and/or (iv) higher levels of caspase-1 and caspase-11 inflammasomes in macrophages [36].
If bacterial secretion systems damage vacuole membranes, then why don’t all Gram-negative bacteria that use these injection devices efficiently lyse their vacuoles? We hypothesized that pathogens such as S. Typhimurium possess species-specific factors that limit the overall extent of vacuole membrane damage. In support of this hypothesis, here we describe a Salmonella-specific type III effector, SopF (SL1344_SL1177), that is translocated by T3SS1 to maintain the integrity of the nascent SCV membrane. Compared to wild type S. Typhimurium, a sopF deletion mutant showed increased access to the cytosol and association with galectin-8 (GAL8), a marker of vacuole rupture, and p62 and LC3, two autophagy-associated proteins. SopF targets host cell membranes, whether translocated by S. Typhimurium or ectopically expressed, and also binds phosphoinositides in vitro. We further show that SopF and another T3SS1 effector, SopB, act antagonistically to regulate nascent vacuole membrane dynamics.
In a RNAseq-based screen for S. Typhimurium SL1344 genes that are up-regulated in the cytosol compared to the vacuole during colonization of epithelial cells (T.R. Powers and L.A. Knodler, manuscript in preparation) we identified SL1344_1177 as a gene that is up-regulated in a subset of cytosolic Salmonella at 8 h post-infection (p.i.), a phenotype similar to that described for T3SS1-associated genes [27,33]. SL1344_1177 is regulated by HilA, HilC and HilD [39] and recent CHIP-seq analysis identified that its counterpart in S. Typhimurium 14028s, STM14_1486, is a direct target of InvF binding [40]. SL1344_1177 is therefore part of the SPI-1 regulatory network. SL1344_1177 was recently renamed SopF by Zhou and colleagues [41] (we will adopt this nomenclature henceforth) and is annotated as a “predicted bacteriophage protein”. It is encoded in SPI-11, which is inserted next to the Gifsy-1 prophage and includes a number of genes involved in Salmonella pathogenesis [42]. SPI-11 is one of eight core pathogenicity islands present in Salmonella enterica subspecies enterica (lineage I), which is the subspecies most commonly associated with disease [43,44]. Taken together, this information hinted that SopF could be a candidate T3SS1 translocated effector. To test this, we constructed a fusion of the N-terminal 199 amino acid residues of SopF to the catalytic domain of adenylate cyclase (CyaA) under the control of its native promoter, and electroporated this plasmid into S. Typhimurium wild type and two genetic mutants that are defective for T3SS1 translocation (ΔprgI) or T3SS2 translocation (ΔssaR). We also utilized bacterial growth conditions that either induce or repress each T3SS [45] and J774A.1 mouse macrophage-like cells as an infection model as they are permissive for the entry of T3SS1 mutants [46,47]. As a readout of SopF-CyaA translocation into host cells, cAMP production was quantified by ELISA. Using late log-phase subcultures, which are induced for T3SS1, we detected robust SopF-CyaA translocation into J774A.1 cells by 1 h p.i.; cAMP production was dependent on a functional T3SS1 but not T3SS2 (Fig 1A, upper panel). To test for T3SS2 dependence, bacteria were grown under conditions where T3SS1 was repressed (stationary phase cultures) and infected cell lysates were collected at 8 h p.i.; no cAMP was produced after infection with bacteria harboring SopF-CyaA or SopB-CyaA fusions (Fig 1A, lower panel). As a positive control, we used the T3SS2 effector, SseK1 [48]. Infection of J774A.1 cells with bacteria harboring an SseK1-CyaA fusion raised intracellular cAMP levels substantially in a manner dependent on ssaR (Fig 1A, lower panel). Overall, these data indicate that SopF is a T3SS1 effector, which agrees with a recent study that showed SopF is delivered into host cells [41].
To examine the kinetics of intracellular sopF expression, we infected mammalian cells with ΔsopF pSopF-3xFLAG or ΔsopF pSopF-2xHA bacteria and samples were collected for immunoblotting at various times p.i. Loading of protein samples was normalized to equivalent colony forming units (CFU) over the time course. In epithelial cells (HeLa and HCT116), SopF-3xFLAG was highly produced from 0.5–4 h p.i. and declined thereafter but was still detectable at 8 h p.i (Fig 1B). In J774A.1 macrophage-like cells, the peak of SopF-2xHA production was 0.5–1 h p.i., after which it markedly declined (Fig 1B). These patterns of expression are strikingly similar to that shown for another T3SS1 effector, SopB, in epithelial cells [23] and macrophages [49]. However, unlike what has been shown for SopB [50], we did not detect a role for SopF in the invasion of non-phagocytic cells (Fig 1C). Furthermore, the invasion efficiency of a ΔsopBΔsopF mutant was comparable to a ΔsopB mutant (Fig 1C), indicating that SopB and SopF do not act cooperatively to promote bacterial internalization. The ΔsopF mutant also did not have a replication defect in epithelial cells or macrophages compared to wild type bacteria (S1 Fig). Altogether, these data show that SopF is a T3SS1 effector that is produced early during host cell infection but is not overtly required for the entry of S. Typhimurium into host cells or intracellular replication.
To examine the subcellular localization of bacterially translocated SopF, we infected HeLa cells with ΔsopF pSopF-3xFLAG bacteria (constitutively expressing mCherry), then fixed, permeabilized and immunostained with anti-FLAG antibodies. Under our fixation-permeabilization conditions, only extrabacterial SopF-FLAG will be accessible to antibodies. No specific FLAG signal was detected in infected cells up to 2 h p.i., suggesting that either the FLAG antigen is masked or SopF is translocated in low quantities. In support of the latter, when we amplified the antibody signal by the addition of tyramide, translocated SopF-3xFLAG could be visualized in ~20% of infected cells at 30 min and 1 h p.i. Thirty minutes after infection, SopF primarily localized to plasma membrane ruffles formed on the dorsal surface of cells in close proximity to internalized bacteria (Fig 1D). By 1 h p.i., SopF also concentrated around intracellular bacteria, presumably decorating the SCV (Fig 1D). On rare occasions, membrane tubules emanating from the SCV were also positive for SopF-3xFLAG signal (Fig 1D). From these observations we conclude that translocated SopF targets a distinct subset of host cell membranes, namely those that are co-opted by Salmonella during bacterial internalization and intracellular trafficking.
We next used ectopic expression of SopF to gain insight into its biological function. HeLa cells were transfected with plasmids encoding for EGFP-SopF or FLAG-SopF and steady-state protein localization was monitored by fluorescence microscopy. SopF was distributed throughout the cytosol, and also accumulated at the leading edge of cells and in filopodia-like structures extending out from the plasma membrane (Fig 2A and 2B). At the leading edge, FLAG-SopF and EGFP-SopF partially co-localized with actin-binding proteins such as moesin, which is involved in cell adhesion, membrane ruffling and microvilli formation [51], vasodilator-stimulated phosphoprotein (VASP), which regulates filopodial length and dynamics [52] and lamellipodin, a regulator of lamellipodia protrusion and cell migration [53] (Fig 2C, S2A Fig).
We also used biochemical fractionation to determine the subcellular localization of SopF. Transfected HeLa cells were first treated with saponin, which permeabilizes the plasma and internal membranes and releases their soluble content, followed by Triton X-100 (TX-100), which solubilizes integral and peripheral membrane proteins, and lastly SDS to solubilize all remaining cellular components. Samples were subject to immunoblotting with antibodies against GFP, FLAG, Hsp27 (cytosolic protein), calnexin (integral membrane protein) and lamin A/C (nuclear protein). EGFP-SopF (Fig 2A) and FLAG-SopF (Fig 2B) equally partitioned to the saponin- and TX-100-soluble fractions, indicating that SopF is both cytosolic and membrane-associated when ectopically expressed.
Bacterial effectors often use specialized membrane-localization domains to target to host cell membranes. These include coiled-coil domains, lipidation, ubiquitylation, and phosphoinositide binding [22,25,54–61]. Phosphoinositides are concentrated on the cytosolic surfaces of membranes and the subcellular localization of each phosphoinositide is tightly governed by the combined actions of lipid kinases and lipid phosphatases, giving each cellular membrane a unique and dynamic phosphoinositide signature [62]. To test for a role of phosphoinositides in directing SopF localization, we used a loss-of-function PI kinase screen in the model eukaryotic organism, Saccharomyces cerevisiae [58]. Yeast have six main PI kinases that phosphorylate the 3’, 4’ or 5’ position of phosphoinositides and inactivation of a particular PI kinase, via genetic deletion or conditional repression, affects the generation of a specific phosphoinositide pool (S3A Fig). The tandem pleckstrin homology (PH) binding domain of oxysterol-binding protein homolog 2 (Osh2) from S. cerevisiae was used as a positive control. 2xPH-Osh2 localizes to the Golgi and plasma membranes in yeast via PI(4)P binding [63,64]. SopF was fused to yEGFP and expressed in wild type yeast and the six PI kinase yeast mutants (genetic deletions or Tet-off strains); localization was visualized by fluorescence microscopy of live cells (Fig 3A, S3B Fig) and protein production was monitored by immunoblotting with anti-GFP antibodies (S3C Fig). yEGFP-SopF was produced at equivalent levels in all yeast strains upon galactose induction (S3C Fig). Notably, the steady-state subcellular localization of SopF in yeast (internal membrane sites; Fig 3A, S3B Fig) and mammalian cells (plasma membrane; Figs 2C and 3B) differed, similar to what has previously been reported for some secreted bacterial proteins [58,65]. This may be a consequence of imaging in live (yeast) versus formaldehyde-fixed (HeLa) cells. Localization of yEGFP-SopF was unchanged in fab1Δ, lsb6Δ, vps34Δ, pik1tet-off and stt4tet-off yeast strains (Fig 3A, S3B Fig). However, yEGFP-SopF showed increased plasma membrane targeting and decreased punctate localization in the mss4tet-off strain (Fig 3A, S3B Fig), revealing that the steady-state localization of SopF is dependent on the activity of Mss4. Mss4 is the major PI(4)P 5-kinase in S. cerevisiae [66,67]. It localizes to the plasma membrane where it directs PI(4,5)P2 synthesis and is required for the maintenance of actin cytoskeleton organization and endocytosis [66–68]. While total cellular levels of PI(4,5)P2 are reduced by 65% in a Mss4 mutant, total levels of PI(3)P, PI(4)P and PI(3,5)P2 are unaffected [66,68,69]. At the subcellular level, however, a Mss4 mutant has decreased PI(4,5)P2 and increased PI(4)P accumulation at the plasma membrane [69,70]. Therefore, concomitant with PI(4,5)P2 depletion and/or PI(4)P accumulation at the plasma membrane, SopF redistributes from punctate intracellular compartments to the plasma membrane in yeast.
To investigate the relationship between SopF localization and phosphoinositide pools in mammalian cells, we used fluorescently-tagged phospholipid biosensors and microscopy. HeLa cells were co-transfected with mCherry-SopF and either EGFP or EGFP-PH domain chimeras, then fixed and visualized by confocal microscopy (Fig 3B). Compared to EGFP alone, co-transfection with EGFP-PH-phospholipase C δ1 (PLCδ1), EGFP-2xPH-Osh2 or EGFP-PH-Bruton’s tyrosine kinase (Btk) appeared to promote the plasma membrane localization of mCherry-SopF (Fig 3B). The PH domain of PLCδ1 serves as a lipid biosensor for PI(4,5)P2 and localizes exclusively to the plasma membrane [71]. We found extensive co-localization of SopF and PH-PLCδ1 at the periphery of cells, particularly in filopodia (Fig 3B). PI(4)P-binding PH domain-GFP chimeras concentrate either at the plasma membrane or Golgi in mammalian cells—the PH domain of Osh2 localizes predominantly at the plasma membrane and weakly at the Golgi, whereas the PH domains of oxysterol binding protein (OSBP), four phosphate adaptor protein (FAPP1) and SidC only label PI(4)P pools at the Golgi [64,72,73]. We observed colocalization of SopF with 2xPH-Osh2 at the plasma membrane, particularly in filopodial extensions (Fig 3B), but not with Golgi-targeting PI(4)P-binding probes (PH-FAPP1 or P4C-SidC; S4A Fig). The Btk PH domain specifically binds to PI(3,4,5)P2 [74] and PH-Btk co-localized with mCherry-SopF staining in lamellipodia at the cell periphery (Fig 3B). Lastly, the PI(3)P binding FYVE domain from Hrs (2xFYVE-Hrs), which concentrates on early endosomes [75], did not colocalize with mCherry-SopF (S4A Fig).
The above studies suggested that SopF localization was connected to host cell phospholipids. We therefore tested whether SopF could directly bind phospholipids using a protein-lipid overlay assay. SopF was heterologously produced in Escherichia coli as a glutathione S-transferase (GST) fusion protein, affinity purified and used to probe nitrocellulose-immobilized phosphoinositides and lipids on PIP Strips. Bound protein was detected using anti-GST antibodies. Compared to GST alone, which displayed no binding affinity for phosphoinositides or lipids, GST-SopF bound to phosphatidic acid and a number of phosphoinositides in vitro, specifically PI(3)P, PI(3,5)P2 and PI(3,4,5)P3 (Fig 3C). A preference for SopF binding to PI(3,5)P2 was indicated by PIP Arrays (S4B Fig). Collectively, this data revealed that SopF subcellular targeting in eukaryotic cells is phosphoinositide-dependent and SopF binds phosphoinositides in vitro.
We have shown that SopF is a T3SS1 translocated effector that targets eukaryotic cell membranes and binds phosphoinositides. While SopF is not overtly required for S. Typhimurium invasion into non-phagocytic cells (Fig 1C; [41]) or intracellular replication (S1 Fig), T3SS1 and its dedicated effectors also promote SCV biogenesis, trafficking and lysis. A small but significant proportion of S. Typhimurium lyse their internalization vacuole, but no type III effectors that modulate the frequency of this event have yet been identified. To test whether SopF is involved in vacuole maintenance, we infected HeLa epithelial cells with wild type and ΔsopF bacteria and monitored nascent SCV lysis using three independent assays: (i) the chloroquine (CHQ) resistance assay, (ii) a cytosolic reporter plasmid and (iii) GAL8 recruitment. CHQ selectively kills vacuolar but not cytosolic Salmonella and when used in combination with a gentamicin protection assay it allows for the quantification of the proportion of total bacteria that are present in the host cell cytosol [34,76]. Using the CHQ resistance assay we found that there was a significant increase in the proportion of ΔsopF bacteria present in the cytosol of HeLa cells compared to wild type bacteria at 90 min p.i. (Fig 4A). This phenotype was also observed in colonic epithelial cells, HCT116 (Fig 4A), and mouse macrophage-like cells, J774A.1 (Fig 4A). Complementation of ΔsopF bacteria with plasmid-borne SopF-3xFLAG restored the efficiency of nascent vacuole lysis to wild type levels (Fig 4A).
The PuhpT-gfpova plasmid has previously been used as a biosensor for S. Typhimurium exposure to the mammalian cytosol [77]. Expression of the unstable GFP variant, GFP-OVA, is under the control of the glucose-6-phosphate responsive S. Typhimurium uhpT promoter. Since glucose-6-phosphate is present in the mammalian cytosol and not the lumen of the SCV, GFP-OVA fluorescence is indicative of the sub-population of intracellular S. Typhimurium that are in damaged vacuoles and/or free in the cytosol. HeLa cells were seeded on glass coverslips and infected with S. Typhimurium wild type and ΔsopF bacteria (constitutively expressing mCherry on the chromosome) harboring the PuhpT-gfpova plasmid. At various times p.i., cells were fixed and the number of GFP-positive bacteria scored by fluorescence microscopy. GFP-positive bacteria were not detectable until 90 min p.i., presumably reflecting the time required for GFP-OVA to be produced, fold and fluoresce after initial bacterial exposure to the epithelial cytosol. By 90 min p.i., approximately 20% of wild type bacteria were GFP-positive and this level was maintained at a steady-state until 6 h p.i., whereupon there was a significant increase in the proportion of cytosolic wild type bacteria (Fig 4B), coincident with the initiation of rapid replication in the epithelial cytosol [33,78]. Significantly more ΔsopF bacteria were GFP-positive at 1.5 h, 2 h, 3 h and 4.5 h p.i. (Fig 4B), indicating that upon deletion of sopF there is an increased frequency of S. Typhimurium in damaged vacuoles and/or free in the cytosol over this timeframe. Altogether, data from these independent assays identify SopF as a T3SS1 effector that promotes the integrity of the nascent SCV membrane. Interestingly, despite an increased nascent vacuole lysis for the sopF deletion mutant, an equivalent proportion of wild type and ΔsopF bacteria was present in the epithelial cytosol at later times (≥6 h p.i.) according to the fluorescent biosensor assay (Fig 4B), CHQ resistance assay (49 ± 8.2% cytosolic bacteria for wild type and 46 ± 9.7% for the ΔsopF mutant at 7 h p.i. in HeLa cells, n≥6 independent experiments) and single-cell analysis (S1 Fig).
Galectins are β-galactoside-binding lectins that act as host “danger receptors” by monitoring endosomal and lysosome integrity [30]. Fluorescently-tagged galectin-3 (GAL3) has been used as a marker of vacuole lysis by Shigella flexneri [29,79], Listeria monocytogenes [29], Legionella pneumophila [80] and S. Typhimurium [16]. It localizes to the limiting membrane of damaged bacteria-containing vacuoles [16,29]. GAL8 and galectin-9 (GAL9) also decorate ruptured SCVs [30]. GAL3, GAL8 and GAL9 recruitment to damaged vacuoles is transient in nature because of their disappearance once the membrane is repaired or disassembled. We used endogenous GAL8 as a measure of vacuole integrity over a 3 h time course of infection. HeLa cells seeded on glass coverslips were infected with wild type and ΔsopF bacteria (constitutively expressing mCherry on the chromosome) and at various times, fixed and immunostained for GAL8. The number of GAL8-positive bacteria was scored by fluorescence microscopy. We found that the recruitment of GAL8 by wild type bacteria peaked at 90 min p.i. (13.1±1.2%, Fig 4C), in agreement with a previous study using fluorescently-tagged GAL8 [30]. Significantly more ΔsopF bacteria were decorated with GAL8 at 1 h, 1.5 h and 2 h p.i. (Fig 4C), indicating enhanced SCV disruption. In trans complementation of ΔsopF bacteria with SopF-3xFLAG restored the amplitude and kinetics of GAL8 acquisition to wild type levels (Fig 4C).
GAL8 directly binds NDP52 to direct the autophagy machinery to damaged SCVs [30]. Polyubiquitin coating of cytosolic S. Typhimurium is also sensed by NDP52, along with two other autophagy adaptor proteins, p62/SQSTM1 and optineurin. All three autophagy adaptors promote engulfment of S. Typhimurium by autophagosomes by direct binding to microtubule-associated protein 1A/1B light chain 3B (LC3) [81–84]. By immunofluorescence analysis, we found that the sopF deletion mutant colocalized more frequently with p62/SQSTM1 and LC3 than wild type bacteria (Fig 4D, S5 Fig). Particularly striking was robust LC3 recruitment to ΔsopF bacteria at 1 h and 2 h p.i. Rather than forming a discrete ring around bacteria, like for wild type S. Typhimurium, LC3 accumulation was cloud-like and often tubular in nature around ΔsopF bacteria (Fig 4D). Collectively our GAL8, p62 and LC3 recruitment data indicates that the sopF deletion mutant is subject to enhanced disruption of the nascent vacuole and detection and capture by the autophagy pathway.
SopF is a 374 amino acid protein with a predicted molecular mass of 42 kDa. It contains a domain of unknown function (DUF), DUF3626, spanning residues 178–338. Basic local alignment analyses (e.g. BLAST) of SopF do not yield any significant matches to sequences of proteins of known function, but secondary structure predictions (e.g. PHYRE2, PSIPRED) suggest that SopF has a high content of α-helices (50%). Many of these are in the N-terminus (amino acids 14–23, 31–43, 65–70, 75–116, 121–133, 144–162, 177–190). The N-terminus of type III effectors is required for their translocation [85–87], minimizing the feasibility of making N-terminal SopF deletions and retaining effector translocation. Instead we focused on the C-terminus of SopF for structure-function analysis. Based upon the location of the DUF3626 domain and the prediction by PHYRE2 [88] that amino acids 334–343 and 354–366 of SopF adopt α-helical structures, we created two SopF truncations, SopF(1–367) and SopF(1–345) and assessed whether deleting these regions affected SopF function during S. Typhimurium infection. HeLa cells were infected with wild type, ΔsopF or ΔsopF bacteria complemented in trans with SopF full-length, SopF(1–345) or SopF(1–367) and nascent SCV lysis was assessed via the CHQ resistance assay at 90 min p.i. and GAL8 recruitment at 1 h p.i. (Fig 5A). SopF(1–345) and SopF(1–367) were produced (S6A Fig) and translocated by S. Typhimurium (S6B Fig, S6C Fig), but neither truncation was able to restore vacuole lysis efficiency of the sopF mutant to wild type levels, whereas full-length SopF did (Fig 5A). From these results we can conclude that the C-terminus of SopF is required for its biological function to promote SCV membrane integrity.
We next assessed the effect of C-terminal truncations on the localization of SopF in eukaryotic cells. To do this, we ectopically expressed two SopF truncations—FLAG-SopF(1–345) or FLAG-SopF(1–367)—in HeLa cells and monitored their steady-state localization by immunofluorescence of fixed cells and sequential detergent fractionation. Unlike FLAG-SopF, which was associated with plasma membrane ruffles and filopodia (Fig 2B), the immunostaining pattern of both SopF truncations was primarily cytosolic (Fig 5B). This altered subcellular localization was confirmed by sequential detergent fractionation, where the majority of FLAG-SopF(1–345) and FLAG-SopF(1–367) was solubilized with 0.1% (w/v) saponin treatment (Fig 5C), which releases cytosolic proteins. Comparable results were obtained with EGFP-tagged SopF truncations (S2B Fig, S2C Fig). Therefore, deletion of as few as seven amino acids from the C-terminus of SopF prevents its association with mammalian cell membranes.
We also examined the impact of C-terminal truncations on the subcellular localization of SopF in S. cerevisiae. yEGFP-SopF(1–345) and yEGFP-SopF(1–367) were expressed in wild type and mss4tet-off yeast strains; protein production was monitored by immunoblotting with anti-GFP antibodies and subcellular localization by fluorescence microscopy of live cells. Both SopF truncations were produced in yeast upon galactose induction (S3D Fig). In contrast to yEGFP-SopF, which primarily localized to internal membrane sites, pEGFP-SopF(1–345) and yEGFP-SopF(1–367) appeared predominantly cytosolic in wild type yeast (Fig 5D and 5E). Furthermore, in the mss4tet-off mutant, while full length SopF redistributed to the plasma membrane, neither of the SopF truncations did (Fig 5D and 5E). This narrows down the Mss4-dependent localization phenotype to the carboxyl-terminal seven amino acids of SopF (368-RDCIILY-374). To further refine which residue(s) within this heptapeptide are subcellular localization determinants, we mutated individual amino acids and tested their effect on localization. First, we mutated the Cys370 residue to test whether it might be a lipidation site. The post-translational addition of lipid groups such as palmitate, farnesyl or geranylgeranyl to cysteine residues can facilitate the membrane association of proteins, including bacterial effectors [61,89]. For example, host prenylation of a C-terminal CAAX motif in the Salmonella type III effector, SifA, promotes its association with late endosomal/lysosomal membranes [55] and host S-palmitoylation of several S. Typhimurium and L. pneumophila effectors directs their membrane association upon translocation [90–92]. However, the subcellular localization of the SopF C370S mutant was indistinguishable from SopF in wild type and mss4tet-off yeast strains (Fig 5D and 5E). Individual mutation of the remaining six amino acids (R368A, D369A, I371A, I372A, L373A and Y374A) also failed to affect the localization of SopF in wild type and mss4tet-off yeast strains (S3E Fig, S3F Fig). Therefore, no single amino acid within this heptapeptide defines the subcellular localization of SopF in wild type or Mss4-deficient S. cerevisiae. Collectively, the results from this structure-function analysis indicate that membrane association and biological activity of SopF in eukaryotic cells are dependent on the carboxyl-terminal seven amino acids, implying that SopF must be membrane-associated to fulfil its biological function.
Phosphoinositides are key regulators of host-pathogen interactions, including SCV trafficking and identity [93]. The T3SS1 effector, SopB/SigD, is an inositol phosphatase that modulates the phosphoinositide composition of Salmonella-induced ruffles and the nascent SCV [18–20]. Given that SopB generates specific phosphoinositides during an infection and SopF binds phosphoinositides in vitro, along with the tendency for T3SS1 effectors to act on common cellular pathways [15,50,94,95], we examined whether there was any crosstalk between SopF and SopB biological activities. HeLa cells were infected with the following mCherry-expressing S. Typhimurium strains: wild type, ΔsopF, ΔsopB, ΔsopB pWSKDE (in trans complementation with SopB/SigD and SigE, its type III chaperone), ΔsopB pWSKDE C460S (in trans complementation with the catalytically inactive SopB/SigD C460S and SigE), ΔsopBΔsopF, ΔsopBΔsopF pWSKDE and ΔsopBΔsopF pWSKDE C460S and nascent SCV damage (GAL8) and autophagic capture (LC3) were monitored by immunostaining of fixed cells at 1 h and 2 h p.i. (Fig 6A). Compared to wild type bacteria, more ΔsopF bacteria were decorated by GAL8 at 1 h and 2 h p.i., whereas significantly fewer ΔsopB bacteria were GAL8-positive at 1 h p.i., but not 2 h p.i. (Fig 6A, upper panel). This result suggests that SopB and SopF might have counteracting activities on membrane stabilization during early vacuole biogenesis (≤1 h p.i.). In agreement, ΔsopBΔsopF bacteria were comparable to wild type bacteria for GAL8 acquisition at 1 h and 2 h p.i. (Fig 6A, upper panel). In trans complementation of ΔsopBΔsopF bacteria with SopB (ΔsopBΔsopF pWSKDE) restored levels of GAL8 acquisition to that observed for the sopF deletion mutant, confirming that the altered phenotype for the double deletion mutant was specifically due to the absence of SopB. Mutation of the Cys460 residue of SopB (C460S) abrogates its 4- and 5-phosphatase activity on phosphoinositides in vitro [96,97]. Providing SopB C460S (pWSKDE C460S) in trans failed to complement the proportion of ΔsopBΔsopF bacteria in damaged vacuoles back to the level of ΔsopF bacteria, or ΔsopB bacteria back to wild type levels (Fig 6A, upper panel). The trend for accumulation of the autophagosomal membrane marker, LC3, mirrored that of GAL8, with the exception of the double deletion mutant (Fig 6A, lower panel). The ΔsopBΔsopF mutant showed a significant increase in LC3 recruitment compared to wild type bacteria at 1 h and 2 h p.i., to a level intermediate that of wild type and ΔsopF bacteria (Fig 6A, lower panel). In trans complementation of ΔsopBΔsopF bacteria with SopB, but not SopB C460S, fully restored LC3 accumulation back to ΔsopF levels (Fig 6A, lower panel).
Our structure-function studies (Fig 5) indicated that SopF must associate with host cell membranes in order to modulate the stability of the nascent SCV. One explanation for the different levels of vacuole damage observed for ΔsopF and ΔsopBΔsopF bacteria (Fig 6A) was if translocated SopF depends on SopB’s inositol phosphatase activity for its membrane association. To test this, HeLa cells were infected with ΔsopF pSopF-3xFLAG and ΔsopBΔsopF pSopF-3xFLAG bacteria and at 1 h p.i., monolayers were subject to mechanical lysis followed by differential centrifugation at 6,500 xg and 100,000 xg to obtain three fractions: P (6,500 xg pellet, contains host cell nuclei, intact bacteria and unbroken cells), M (100,000 xg pellet, contains host cell membranes) and C (100,000 xg supernatant, contains host cell cytosol). The content of these fractions was assessed by immunoblotting with anti-DnaK (intact bacteria), anti-Hsp27 (host cell cytosol), anti-LAMP-1 (host cell membranes) and anti-FLAG (SopF-3xFLAG) antibodies. For the ΔsopF pSopF-3xFLAG infection, translocated SopF-3xFLAG was detected in M and C fractions, indicating partitioning to host cell membranes and cytosol, respectively (Fig 6B). Upon infection with ΔsopBΔsopF bacteria, SopF-3xFLAG translocation appeared comparable to ΔsopF bacteria, both in total amount and partitioning profile (Fig 6B). We also compared SopF localization by immunofluorescence. HeLa cells were infected with ΔsopF and ΔsopBΔsopF bacteria harboring pSopF-3xFLAG (both strains constitutively expressing mCherry) and the localization of translocated SopF was examined using tyramide-coupled immunofluorescence with anti-FLAG antibodies. The number of SopF-positive cells was equivalent for ΔsopF and ΔsopBΔsopF bacteria at 30 min and 1 h p.i. (23.0% and 21.6% of infected cells for ΔsopF bacteria at 1 h and 2 h p.i., respectively; 27.7% and 22.9% for ΔsopBΔsopF bacteria at 1 h and 2 h p.i., respectively (mean from 3 independent experiments)). The distribution pattern of translocated SopF staining–ruffles, bacteria and/or tubules–was also comparable for ΔsopF (Fig 1D) and ΔsopBΔsopF bacteria at 30 min and 1 h p.i. (Fig 6C). Altogether, from these results we can conclude that translocated SopF associates with host cell membranes independently of SopB.
While bacterial secretion systems are essential to the virulence of many Gram-negative pathogens, their insertion into the bacteria-containing vacuole membrane also acts as an “Achilles heel” because it damages the membrane, which allows cytosolic surveillance pathways to detect the presence of bacteria in these damaged vacuoles [31]. Bacterial pathogens that are tailored for life in a vacuole, such as S. Typhimurium, must possess strategies to minimize vacuole membrane damage. Exemplifying this, S. Typhimurium has T3SS needle tip proteins that confer relatively poor damaging/lytic ability compared to their counterparts from cytosolic bacteria [35]. Here we identify an additional mechanism used by S. Typhimurium to promote the integrity of its nascent SCV membrane, SopF, a type III effector that is specific to Salmonella enterica and Salmonella bongori spp. The importance of SopF and nascent vacuole integrity to the pathogenesis of S. Typhimurium is illustrated by results from animal models of infection. In a transposon-directed insertion-site sequencing (TraDIS) screen of S. Typhimurium mutants for their relative fitness during intestinal colonization of pigs, cows and chickens after oral infection, a sopF (SL1344_1177) mutant was significantly attenuated in all three animal models [98]. In the same study, a sopF mutant was not attenuated for systemic infection of mice upon intravenous infection [98]. However, upon oral infection of streptomycin-pretreated mice, a sopF mutant is outcompeted by wild type bacteria for colonization of the cecum, spleen and liver at 24 h [41]. sopF is therefore part of a core set of genes required for efficient intestinal colonization by S. Typhimurium, irrespective of the host species.
We report that SopF targets the nascent SCV to promote membrane stability. We and others have previously noted the high prevalence of bacterial effectors that associate with host cell membranes [56,58,99]. SopF does not appear to be stably associated with cell membranes, either when ectopically expressed or bacterially translocated, as only a portion of the total pool partitions to this fraction (Figs 2A, 2B and 6B). This result argues that SopF is a peripheral membrane protein, which is supported by bioinformatics analysis that predicts SopF does not have any transmembrane spanning regions (TMpred program). Our structure-function studies pinpoint that the carboxy-terminus of SopF plays a critical role in its membrane targeting in eukaryotic cells, and subsequently its biological activity (Fig 5). This region comprises only seven amino acids, with four of these being hydrophobic residues, and is predicted by PHYRE2 to adopt a β-strand structure. We predict that this heptapeptide sequence is critical for phosphoinositide and/or protein interactions that promote the membrane association of SopF. Of note, short hydrophobic stretches in YopE and ExoS, type III effectors from Yersinia spp. and Pseudomonas aeruginosa, respectively, function as membrane localization domains in eukaryotic cells [100] and some small GTPases rely on a short stretch of hydrophobic amino acids or polybasic clusters to facilitate their plasma membrane binding [101].
Purified SopF binds a number of phosphoinositides in vitro, hinting that membrane targeting of SopF is mediated by phosphoinositide binding in vivo. This does not preclude that protein-protein interactions are also involved. Alto and colleagues showed that the localization of a number of Salmonella effectors (PipB, PipB2, SopA, SseG, SseJ and SteA) is affected in yeast PI kinase mutants, implying that their subcellular targeting also depends on phosphoinositide-binding [58]. The authors further identified that purified recombinant PipB2, SopA and SteA bound phospholipids in vitro [58]. The findings for SteA corroborated an earlier study that demonstrated that this T3SS2 effector localizes to the SCV membrane via its interaction with PI(4)P [60]. Phosphoinositide-binding is not unique to S. Typhimurium effectors and, in fact, is quite widespread for type III and type IV effectors. DrrA/SidM [102], SidC [102–106], Lpg1101 and Lpg2603 [107] from Legionella spp. bind PI(4)P, LpnE, LtpD, LtpM and RidL from L. pneumophila [108–111] and CvpB from Coxiella burnetii [112] bind PI(3)P, and ExoU from Pseudomonas aeruginosa [113] and VopR from Vibrio parahaemolyticus [57] bind PI(4,5)P2 for example.
The phosphoinositide binding preference of SopF in vivo remains unclear. SopF preferentially binds PI(3,5)P2 in lipid overlay assays but we were unable to assess PI(3,5)P2 binding in eukaryotic cells because there is currently no known specific PI(3,5)P2-binding probe. PI(3,5)P2 is much less abundant than other phosphoinositides in mammalian cells and localizes on early endosomes, late endosomes and lysosomes [114]. Perturbation of PI(3,5)P2 synthesis in yeast (fab1Δ mutant) did not change the subcellular localization of SopF, however (Fig 3A). By contrast, its subcellular localization was affected in yeast defective for proper PI(4,5)P2 synthesis (mss4tet-off strain, Fig 3A). It remains possible that the SopF localization defect in the mss4 mutant is not specifically due to the alteration of phosphoinositide pools at the plasma membrane but rather an indirect effect of aberrant actin cytoskeleton organization, Rho1-mediated signaling or sphingolipid biosynthesis [66,67,115]. Upon ectopic expression in mammalian cells, we observed partial colocalization of SopF with multiple PH domain probes that have specificity for different phosphoinositide pools at the plasma membrane (Fig 3B). The nascent SCV is known to be decorated with PI(4)P [60] but PI(4,5)P2 is absent [20,116]. The direct accumulation of PI(3,5)P2 on the SCV has not been monitored. The association of SopF with host cell membranes is independent of SopB (Fig 6), the S. Typhimurium effector whose inositol phosphatase activity largely governs the phosphoinositide composition of the nascent SCV. Consequently, we propose that host-generated phosphoinositides promote the membrane recruitment of translocated SopF. Our ongoing studies are directed towards identification of which specific phosphoinositide(s) are bound by SopF during infection.
Why does only a subset of S. Typhimurium lyse their internalization vacuole? In an elegant study by Enninga and colleagues, it was shown that fusion, or not, of the nascent SCV (designated the Salmonella-containing compartment (SCC) by the authors) with macropinosomes determines the fate of S. Typhimurium in epithelial cells [16]. By live-cell imaging, the majority of SCCs that fused with macropinosomes remained intact whereas all bacteria that eventually hyper-replicated in the cytosol of epithelial cells had initially failed to fuse with macropinosomes. It is likely that SCC fusion with macropinosomes provides new membrane to allow for vacuole expansion. This study therefore established a direct correlation between SCC-macropinosome fusion and vacuole integrity, which is in contrast to the early events during infection with the professional cytosolic pathogen, S. flexneri. Internalized S. flexneri are surrounded by and come in contact with newly formed macropinosomes, but do not fuse with them, and then rapidly lyse their internalization vacuole [117]. There is a positive correlation between the number of macropinosomes in the vicinity of wild type S. flexneri and the timing of vacuole lysis, however. Infection with a S. flexneri ipgD mutant led to a 60% reduction in the number of macropinosomes surrounding bacteria [117] and a delay in the kinetics of vacuole rupture compared to wild type S. flexneri [79]. It has been shown that SopB, an IpgD homolog, contributes to macropinosome formation following Salmonella infection [15,21]. Similar to a S. flexneri ipgD mutant, here we describe that a S. Typhimurium sopB deletion mutant has a transient delay in nascent vacuole lysis compared to wild type bacteria; it is defective at 1 h p.i., but not 2 h p.i. (Fig 6A). SopB and IpgD both control the phosphoinositide composition of the vacuole membrane and might modulate bacteria-containing vacuole rupture in a similar macropinosome-dependent manner, perhaps via the recruitment of Rab11 [79,117] or sorting nexins [97,118–120].
While SopF is the first bacterial factor reported to promote nascent SCV stability, a number of host factors have been previously implicated in this process; some promote stability and some instability. TANK-binding kinase 1 (TBK1) was shown to maintain the integrity of internalization vacuoles containing Gram-negative and Gram-positive bacteria, including that of the nascent SCV [121]. TBK1 also recruits autophagy-associated proteins to promote clearance of cytosolic S. Typhimurium [84]. The phosphoinositide 3-phosphatase, myotubularin 4 (MTMR4), promotes SCV stability in fibroblasts. It is recruited to the nascent SCV as early as 15 min p.i. [122] and in MTMR4-depleted A431 cells, an increase in GAL8-, LC3- and p62-positive SCVs at 2 h and 3 h p.i., but not 1 h p.i., is observed [122]. Patrick et al. (2018) recently reported that recruitment of the retromer sorting complex stabilizes the nascent SCV. shRNA-mediated knockdown of VPS35, a component of the retromer sorting complex, led to an increase in the proportion of cytosolic S. Typhimurium at 1 h and 2 h p.i. in HeLa cells [123]. They proposed a model whereby VPS35 is initially recruited to the nascent SCV, then displaced from the SCV due to its interaction with SseC, a T3SS2 translocon protein. While the fate of ΔsseC bacteria was not reported in this study, this mutant is T3SS2 translocation-defective and such mutants remain in an intact SCV [124], hinting that the retromer displacement model for vacuole stabilization might be more complex than proposed. The lone host protein known to antagonize nascent SCV membrane stability is the coat protein complex II (COPII). Santos et al. (2015) found that COPII was recruited to the nascent SCV and siRNA mediated knockdown of a COPII component, Sec13, halved the proportion of S. Typhimurium in ruptured SCVs in HeLa cells at 1 h p.i. [125]. COPII recruitment therefore promotes nascent SCV rupture in epithelial cells. Using numerous techniques, we have not identified an interaction between SopF and any of these above-mentioned host proteins, suggesting that SopF is acting in a novel manner to modulate nascent SCV integrity.
Bacterial effectors can act alone, or they can interact with other effectors to modulate host functions; such effector-effector interactions can be categorized as direct or indirect [126]. “Meta-effectors” describes the scenario whereby one effector acts directly on another to modulate its function and is typified by the L. pneumophila type IV effectors LubX and SidH–LubX polyubiquitinates SidH to regulate its levels, and thus activity, inside of host cells [127]. Effectors that act cooperatively or antagonistically on a shared host target or pathway represent indirect effector-effector interactions. One example of cooperativity is the five Salmonella type III effectors—SipA, SipC, SopB, SopE and SopE2—that target the host cell actin cytoskeleton to promote plasma membrane ruffling and bacterial invasion. SopE and SptP from Salmonella are hallmark examples of effectors with antagonistic functions; SopE-mediated activation of Rho GTPases is counteracted by the GTPase-activating activity of SptP [14]. SopF and SopB have opposing effects on nascent vacuole membrane dynamics and provide a new example of antagonistic T3SS1 effectors. Interestingly, other antagonistic effector-effector relationships have been reported for the maintenance of vacuole integrity by S. Typhimurium (SifA/SseJ and SifA/SopD2) and L. pneumophila (SdhA/PlaA). S. Typhimurium has a second type III effector, SifA, that stabilizes the mature SCV membrane. SifA is translocated by T3SS2 and acts much later than SopF during the infection process; a sifA mutant is not defective for maintaining vacuole integrity until ≥6 h p.i. [124,128]. SifA interacts with the host protein SKIP (PLEKHM2) to downregulate the recruitment of kinesin, a molecular motor, to the SCV and affect vacuolar membrane dynamics [129,130]. SseJ is a T3SS2 effector [86] with phospholipase and acyltransferase activity [131,132]. Unlike ΔsifA bacteria, a sifA sseJ deletion mutant is not defective for stability of the mature SCV membrane [132,133] indicating that the loss of vacuolar membrane around ΔsifA bacteria requires SseJ. Similarly, a sifA sopD2 deletion mutant also resides within an intact vacuole [134] highlighting that sopD2 is epistatic over sifA. The biological function of SopD2 is not known. L. pneumophila utilizes the Dot/Icm secretion system to translocate type IV effectors that customize its intracellular niche. It was recently identified that SdhA and PlaA have antagonistic activities regarding stability of the Legionella-containing vacuole (LCV) [80]. In the absence of sdhA, the LCV is unstable, resulting in bacterial release into the cytosol, their autophagic capture and degradation. A suppressor screen identified plaA as being able to rescue the ΔsdhA replication defect. PlaA has phospholipase activity, like its homolog SseJ from S. Typhimurium. The mechanistic basis for how SdhA/PlaA, SifA/SseJ, SifA/SopD2 and SopB/SopF counteract each other’s activities remains undetermined. Regardless, the expanding number of effector proteins identified to have antagonistic activities in regard to bacteria-containing vacuole membrane dynamics reinforces that bacterial control over vacuole integrity is an important virulence strategy.
S. Typhimurium SL1344 was the wild type strain used in this study [135]. The SL1344 ΔssaR, ΔprgI::FRT and ΔsopB strains have been described previously [34,136,137]. An in-frame unmarked deletion of sopF (deletion of amino acids 3–373 of SopF) was constructed using allelic exchange in a counter-selectable suicide vector harboring SacB, pRE112, as previously described [138]. Briefly, a non-polar deletion cassette was amplified using overlap extension PCR from S. Typhimurium SL1344 genomic DNA (prepared using a Bactozol DNA isolation kit (Molecular Research Center, Inc.)) (oligonucleotide sequences are provided in S1 Table), ligated into pRE112 and transformed into E. coli SY327λpir, a donor strain for conjugation into SL1344 wild type and ΔsopB strains. Resulting meridiploids were incubated at 30°C overnight on agar containing 1% (w/v) tryptone, 0.5% (w/v) yeast extract and 5% (w/v) sucrose. Sucrose-resistant clones were screened for deletion of sopF by PCR with primers flanking the recombination region. The resulting strains were designated SL1344 ΔsopF and ΔsopBΔsopF. SL1344 wild type glmS::Ptrc-mCherryST constitutively expresses mCherry (codon-optimized for S. Typhimurium) under the control of the trc promoter and has been described previously [139]. P22 lysate derived from this strain was used to transduce ΔsopF, ΔsopB and ΔsopBΔsopF bacteria to create ΔsopF glmS::Ptrc-mCherryST, ΔsopB glmS::Ptrc-mCherryST and ΔsopBΔsopF glmS::Ptrc-mCherryST strains, respectively, followed by removal of the CmR cassette using pCP20 [140]. For construction of a chromosomal 3xFLAG-tagged derivative of SopF (SL1344 sopF::3xFLAG), recombinational transfer of the 3xFLAG coding sequence to the 3’ end of the sopF coding sequence was achieved using the oligonucleotides SL1177-3xFLAG-for and SL1177-3xFLAG-rev with pSUB11 as a template, following the method described by Uzzau et al. (2001) [141].
For detection of type III effector translocation, a fusion of SopF to adenylate cyclase (CyaA) was constructed. To create pSopF-CyaA, 417 bp upstream of the sopF start codon plus the genetic region encoding for amino acids (1–199) of SopF was amplified from SL1344 genomic DNA, in addition to the catalytic domain of Bordetella pertussis CyaA from pMS107 [142]. These two fragments were then mixed in a second round of PCR, and the resulting amplicon was digested with SmaI/XhoI and ligated into the corresponding sites of pACYC177 (New England Biolabs). The SopB(1–200)-CyaA-SigE (pACYC177 backbone) and SseK1-CyaA (pACYC184 backbone) plasmids have been described previously [48,143]. Alternatively, SopF, SopF(1–367) and SopF(1–345) fusions to TEM1 β-lactamase were constructed in pCX340 [144] to detect type III effector translocation.
For complementation of ΔsopF bacteria, the entire coding sequence of sopF and 417 bp of upstream region were amplified from SL1344 genomic DNA with Sma-SL1177CyaA-F and Xho-SL1177-R, digested with SmaI/XhoI and ligated into SmaI/XhoI-digested pACYC177 to create pSopF. The SopF truncations—pSopF(1–345) and pSopF(1–367)—were constructed similarly. For plasmid-borne expression of 3xFLAG-tagged SopF, we used the oligonucleotides Sma-SL1177CyaA-F and Xho-SL1177FLAG-R with SL1344 sopF::3xFLAG genomic DNA as a template. The resulting amplicon was digested with SmaI/XhoI and ligated into the corresponding sites of pACYC177 to create pSopF-3xFLAG. The pSopF-2xHA plasmid was constructed using SL1344 wild type genomic DNA as an amplification template with Sma-SL1177CyaA-F and Xho-SL11772HA-R oligonucleotides, followed by ligation into SmaI/XhoI-digested pACYC177. The SopB complementation plasmids–pWSKDE and pWSKDE C460S - are in a low copy number plasmid backbone (pWSK29, [145]) and have been described previously [22,146].
SL1344 wild type glmS::Ptrc-mCherryST and ΔsopF glmS::Ptrc-mCherryST strains harboring the PuhpT-gfpova plasmid [77] were used as a biosensor for bacterial exposure to the mammalian cytosol. Expression of the unstable GFP variant, GFP-OVA, is under the control of the glucose-6-phosphate responsive S. Typhimurium uhpT promoter.
For ectopic expression in mammalian cells, SopF was cloned into pFLAG-pcDNA4/TO (pFLAG; [147]), pEGFP-C2 (Clontech) or pmCherry-C1 (Clontech) to create pFLAG-SopF, pEGFP-SopF and pmCherry-SopF, respectively. C-terminal truncations of SopF were cloned into pFLAG-pcDNA4/TO and pEGFP-C2. SopF, SopF truncations and SopF amino acid point mutants were cloned into p413Gal-yEGFP (yEGFP; [58]) for galactose-inducible expression in S. cerevisiae. All constructed plasmids were verified by sequencing.
The following plasmids encoding for yEGFP- or EGFP-labeled lipid-binding domains were used as biosensors of cellular phospholipid pools: yEGFP-2xPH-Osh2 [58], EGFP-2xPH-Osh2 [64], PH-PLCδ1-EGFP [148], PH-Btk-EGFP [149], EGFP-2xFYVE-Hrs [150], P4C-SidC-EGFP [60] and PH-FAPP1-EGFP [151].
HeLa (ATCC CCL-2) human cervical adenocarcinoma epithelial cells, HCT116 (ATCC CCL-247) human colorectal carcinoma epithelial cells and J774A.1 (ATCC TIB-67) mouse macrophage-like cells were purchased from American Type Culture Collection (ATCC) and used within 15 passages of receipt. Cells were maintained in the growth medium recommended by ATCC, all containing 10% (v/v) heat-inactivated fetal calf serum (FCS; Gemini Bio Products). Tissue culture plasticware was purchased from Thermo Scientific Nunc.
Mammalian cells were seeded at the following densities: (1) HeLa epithelial: 5x104cells/well or 6x104 cells/well (glass coverslips) in 24-well plates or 2x105 cells/well in 6-well plates 24 h prior to infection; (2) HCT116 epithelial: 1x105 cells/well in 24-well plates or 3.4x105 cells/well in 6-well plates ~42–44 h prior to infection; (3) J774A.1 macrophage-like: 2x105 cells/well or 2.5x105 cells/well (glass coverslips) in 24-well plates or 8x105 cells/well in 6-well plates 24 h prior to infection. Unless otherwise stated, T3SS1-induced bacterial subcultures were prepared in LB-Miller broth (Difco) [76] and HeLa (MOI ~50), HCT116 (MOI ~20) and J774A.1 (MOI ~10) cells were infected for 10 min with bacterial subcultures as described [76]. Gentamicin protection and CHQ resistance assays (400 μM CHQ for all cell lines) were as described previously [76].
To quantify type III effector-CyaA translocation, J774A.1 cells were seeded in 24-well plates the day prior to infection. S. Typhimurium wild type, ΔprgI::FRT and ΔssaR strains harboring pSopF(1–199)-CyaA, pSopB(1–200)-CyaA-SigE or pSseK1-CyaA plasmids were grown as 3.5 h subcultures (to assess T3SS1-dependent translocation) or as overnight stationary phase cultures (to assess T3SS2-dependent translocation) as described [45,76]. J774A.1 cells were infected with bacterial subcultures for 10 min at an MOI of ~10 (for wild type and ΔssaR strains) or ~20 (for the ΔprgI::FRT strain). Under these conditions, equivalent CFU were internalized for all three strains and <6% macrophage cytotoxicity was observed at 1 h p.i. as determined by CytoTox 96 Non-Radioactive Cytotoxicity Assay (Promega). Monolayers were washed with PBS at 1 h p.i., then lysed and processed for cAMP quantification as described previously [47]. Alternatively, J774A.1 cells were infected with overnight stationary phase cultures (MOI ~50) by centrifugation at 500 xg for 5 min (t0), followed by a further incubation at 37°C in 10% CO2 for 25 min. Cells were washed three times in Hanks’ balanced salt solution (HBSS), then incubated in growth media containing 50 μg/ml gentamicin for 1 h, followed by 10 μg/ml gentamicin until 8 h p.i., when lysates were collected and processed as described above. cAMP was measured using the Amersham cAMP Biotrak enzyme immunoassay system (GE Healthcare BioScience) according to the manufacturer’s instructions for the non-acetylation procedure.
TEM1 β-lactamase translocation assays were performed as follows. J774A.1 macrophage-like cells were seeded in tissue culture treated black 96-well plates (Cellvis for microscope imaging or Corning for fluorescence plate reader detection) at 4x104 cells/well two days prior to infection. Subcultures of S. Typhimurium SL1344 wild type carrying pCX340, pCX340-SopF, pCX340-SopF(1–345) or pCX340-SopF(1–367) were inoculated from overnight cultures at a 1:10 dilution in LB-Miller broth. After 3.5 h growth at 37°C, shaking at 220 rpm, cultures were induced with 1 mM isopropyl β-D-1 thiogalactopyranoside (IPTG; Fisher) for 1 h. J774A.1 cells were infected with bacterial subcultures for 20 min (MOI ~20). Non-internalized bacteria were removed by washing thrice with HBSS, and cells were loaded with CCF2-AM fluorescent substrate in the presence of 2.5 mM probenecid (Biotium) and 1 mM IPTG for 90 min at room temperature according to the manufacturer’s instructions for the LiveBLAzer FRET-B/G loading kit (Thermo). Wells were then washed once in HBSS and incubated in FluoroBrite DMEM Live Cell Fluorescence Imaging Media (Gibco) containing 2.5 mM probenecid. β-lactamase activity was assessed by fluorescence microscopy or in a fluorescence plate reader. Images of live cells from randomly chosen fields were acquired as a single 1 μm section on a Leica SP8 confocal microscope upon excitation at 405 nm and separate collection of fluorescence emissions with either a 420–480 nm (blue fluorescence) or a 520–560 nm bandpass (green fluorescence). Images were assembled using Adobe Photoshop CS6. Alternatively, fluorescence was quantified on a TECAN SPARK plate reader with excitation at 410 nm (20 nm bandpass), and emission was detected via 450 nm (20 nm band pass, blue fluorescence) and 520 nm (20 nm band pass, green fluorescence) filters. Translocation was expressed as the emission ratio at 450/520 nm and normalized to mock-infected cells for each experiment.
Plasmid DNA was purified using the Nucleobond Xtra Midi Plus kit (Macherey-Nagel) according to the manufacturer’s protocol. HeLa cells were transfected with plasmid DNA using the FuGENE 6 transfection reagent (Promega) according to the manufacturer’s protocol. Cells were transfected for 16–18 h with 500 ng DNA/well (24-well plate) or 1 μg DNA/well (6-well plate).
For analysis of the membrane association of SopF when ectopically expressed, transfected HeLa cells were subject to sequential detergent fractionation as described previously [23,56], with minor modifications. HeLa cells were seeded in 6-well plates at 1.4–1.8 x 105 cells/well and transfected for 18–20 h prior to fractionation. Transfected cells were sequentially treated with 0.1% (w/v) saponin, 0.5% (v/v) TX-100, then 2.5% (w/v) SDS (resuspension of the final pellet in 1.5x SDS-PAGE sample buffer). Equal volumes of saponin-, TX-100- and SDS-soluble fractions were analyzed by immunoblotting.
The subcellular association of translocated SopF was determined by mechanical lysis of infected HeLa cells followed by differential centrifugation as previously described [56,152], with minor modifications. HeLa cells were seeded in 10 cm dishes at 1.5 x 106 cells/dish the day prior to infection. Bacterial subcultures of ΔsopF pSopF-3xFLAG and ΔsopBΔsopF pSopF-3xFLAG bacteria were used to infect monolayers for 10 min (MOI ~50) (two dishes per strain). At 1 h p.i., cells were mechanically disrupted by 4–5 passes through a 22-gauge needle, followed by low-speed centrifugation at 6,500 xg for 10 min to pellet nuclei, unbroken cells and intact bacteria. The supernatant was further subject to a high-speed centrifugation at 100,000 xg for 30 min to separate host cell membranes from cytosol. Equal volumes of each fraction were subject to immunoblotting with anti-FLAG, anti-Hsp27 (cytosol), anti-LAMP-1 (membranes) and anti-DnaK (intact bacteria) antibodies.
HeLa cells were seeded on acid-washed, 12 mm glass coverslips (#1.5 thickness, Fisher Scientific) in 24-well plates. Transfected or infected cells were fixed for 10 min at 37°C with 2.5% (w/v) paraformaldehyde (PFA; EMD Millipore). Fixed cells were washed three times with PBS then permeabilized-blocked in PBS containing 10% normal goat serum (NGS; Gibco) and 0.2% (w/v) saponin (Acros) for 20 min. Primary and secondary antibodies were diluted in blocking buffer. Coverslips were incubated with the following primary antibodies for 45 min at room temperature: mouse anti-FLAG M2 affinity isolated (1:500 dilution; Sigma), rabbit anti-VASP (clone 9A2, 1:100; Cell Signaling), rabbit anti-moesin (clone Q480, 1:100; Cell Signaling), rabbit anti-lamellipodin (clone D8A2K, 1:100, Cell Signaling), mouse anti-human galectin-8 (clone 210608, 1:100; R&D Systems), guinea pig polyclonal anti-p62 (1:200; Progen), rabbit polyclonal anti-LC3 (1:300; MBL). Coverslips were washed three times with PBS and incubated for a further 30–45 min at room temperature with Alexa Fluor-conjugated secondary antibodies (1:400 dilution; Life Technologies) in blocking buffer. After three washes in PBS, cells were incubated with Hoechst 33342 (1:10,000; Invitrogen) for 1 min before mounting in Mowiol on glass slides. Samples were visualized using a Leica DM4000 upright fluorescence microscope. Image acquisition was on a Leica SP8 Scanning Point confocal microscope using the sequential acquisition mode through an optical section of 0.3 μm in the z-axis. Images are maximum intensity projections of z-stacks.
To detect translocated SopF, tyramide signal amplification for immunofluorescent enhancement was used. HeLa cells seeded on glass coverslips were infected with ΔsopF glmS::Ptrc-mCherryST or ΔsopBΔsopF glmS::Ptrc-mCherryST bacteria harboring pSopF-3xFLAG. At 30 min and 1 h p.i., monolayers were fixed and blocked/permeabilized as described above. The subsequent staining procedure followed the manufacturer’s instructions for the Alexa Fluor 488 Tyramide SuperBoost kit (Invitrogen). Primary antibody—mouse anti-FLAG M2 affinity isolated (1:2,000 dilution; Sigma)—and poly-HRP-conjugated secondary antibody incubations were for 45 min at room temperature. The tyramide amplification step was for 5 min at room temperature.
Proteins were separated by SDS-PAGE and transferred to 0.2 μm or 0.45 μm pore-size nitrocellulose membranes (GE Healthcare Life Sciences). Membranes were blocked at room temperature for 1 h with Tris-buffered saline (TBS) containing 5% (w/v) skim milk powder and 0.1% (v/v) Tween-20 (TBST-milk), then incubated with the following primary antibodies overnight at 4˚C: mouse anti-FLAG M2 affinity isolated (1:2,000 dilution; Sigma), mouse anti-HA.11 ascites (1:2,000; BioLegend), rabbit polyclonal anti-GFP (1:40,000; Thermo), mouse anti-β-lactamase (clone 8A5.A10, 1:2,000 dilution; Thermo), mouse anti-Hsp27 (clone G31, 1:20,000; Cell Signaling), rabbit polyclonal anti-calnexin (1:40,000; Enzo), rabbit polyclonal anti-lamin A/C (1:5,000; Cell Signaling) or mouse anti-LAMP-1 (clone H4A3, 1:1,000 dilution; Developmental Studies Hybridoma Bank). The hybridoma H4A3 developed by J.T. August and J.E.K. Hildreth (The Johns Hopkins University School of Medicine) was obtained from the Developmental Studies Hybridoma Bank, created by the NICHD of the NIH and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242. Blots were then incubated with anti-rabbit IgG or anti-mouse IgG horseradish peroxidase (HRP)-conjugated secondary antibodies (1:10,000; Cell Signaling) in TBST-milk for 1–2 h at room temperature, followed by Supersignal West Femto Max Sensitivity ECL Substrate (Thermo). Chemiluminescence was detected using a ChemiDoc MP Imaging System (Bio-Rad) and Bio-Rad Image Lab software.
E. coli Rosetta 2(DE3) (Novagen) harboring pGEX-6P-1 and pGEX-6P-1-SopF plasmids were used for recombinant protein production. Overnight bacterial cultures were subcultured to logarithmic phase (OD600 of 0.6–0.8) and GST fusion production was induced overnight at 18˚C by the addition of 1 mM IPTG (AppliChem). Bacteria were harvested by centrifugation at 10,000 xg for 15 minutes at 4˚C. All subsequent steps were performed on ice or at 4˚C. The cell pellet was resuspended in Tris-buffered saline (TBS) (50 mM Tris-HCl pH 8.0, 150 mM NaCl) containing cOmplete EDTA-free protease inhibitor cocktail (Roche), then lysed using an Avestin EmulsiFlex-C3 high-pressure cell homogenizer or a Constant Systems Ltd. CF cell disruptor. Bacterial lysates were centrifuged at 13,000 xg for 30 min, after which the clarified supernatant was loaded into a PolyPrep Chromatography Column (Bio-Rad) pre-packed with equilibrated glutathione sepharose beads (Merck). Beads were then washed once with TBS and bound protein was eluted with TBS containing 10 mM glutathione (Sigma). Eluted protein was dialyzed overnight against TBS and purified protein was aliquoted and stored at -80˚C.
PIP Strips and PIP Arrays (Echelon Biosciences) were used to test lipid binding of SopF according to the manufacturer’s protocol. Briefly, PIP Strips and PIP Arrays were blocked in PBS/0.1% (v/v) Tween-20 (PBST) containing 1% (w/v) skim milk powder (PBST-milk) for 1 h at room temperature. Purified recombinant GST fusion protein was incubated with PIP Strips and PIP Arrays at a concentration of 1 μg/ml in PBST containing 3% (w/v) bovine serum albumin (PBST-BSA) for 1 h at room temperature. PIP Strips and PIP Arrays were washed 3 times with PBST, then probed with rabbit anti-GST antibody (1:2,000; Cell Signaling) in PBST-BSA for 1 h at room temperature. PIP Strips and PIP Arrays were washed 3 times with PBST, then incubated with anti-rabbit HRP-conjugated secondary antibodies (1:3,000; Perkin Elmer) in PBST-milk at room temperature for 1 h. Chemiluminescence signal was detected using Clarity Western ECL Substrate (Bio-Rad) and an Amersham Imager 600 machine.
The following S. cerevisiae strains were used: wild type with genetic background BY4742 (MATa his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0); lsb6Δ, vps34Δ and fab1Δ with genetic background BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0); stt4tet-off, pik1tet-off and mss4tet-off with genetic background R1158 (MATa his3Δ1 leu2Δ0 met15Δ0 URA3::CMVp-tTA). Competent yeast cells were transformed with p413Gal-yEGFP-SopF variants (expression is under the control of a galactose-inducible promoter) using the standard lithium acetate transformation protocol [153]. Transformants were grown for 2–3 days on minimal media agar lacking histidine. Tet-off strains were repressed with doxycycline for 24 h prior to the induction of protein expression in galactose-containing media as described previously [58]. A Leica DM4000 upright fluorescence microscope was used for visualization and image acquisition of yEGFP-tagged protein localization in live yeast.
All experiments were conducted on at least three separate occasions, unless otherwise indicated, and results are presented as mean ± SD. Statistical analyses were performed using one-way analysis of variance (ANOVA) with Dunnett’s post-hoc test or Student’s t-test (GraphPad Prism). A p-value of ≤0.05 was considered significant.
|
10.1371/journal.pcbi.1003109 | A Family of Algorithms for Computing Consensus about Node State from Network Data | Biological and social networks are composed of heterogeneous nodes that contribute differentially to network structure and function. A number of algorithms have been developed to measure this variation. These algorithms have proven useful for applications that require assigning scores to individual nodes–from ranking websites to determining critical species in ecosystems–yet the mechanistic basis for why they produce good rankings remains poorly understood. We show that a unifying property of these algorithms is that they quantify consensus in the network about a node's state or capacity to perform a function. The algorithms capture consensus by either taking into account the number of a target node's direct connections, and, when the edges are weighted, the uniformity of its weighted in-degree distribution (breadth), or by measuring net flow into a target node (depth). Using data from communication, social, and biological networks we find that that how an algorithm measures consensus–through breadth or depth– impacts its ability to correctly score nodes. We also observe variation in sensitivity to source biases in interaction/adjacency matrices: errors arising from systematic error at the node level or direct manipulation of network connectivity by nodes. Our results indicate that the breadth algorithms, which are derived from information theory, correctly score nodes (assessed using independent data) and are robust to errors. However, in cases where nodes “form opinions” about other nodes using indirect information, like reputation, depth algorithms, like Eigenvector Centrality, are required. One caveat is that Eigenvector Centrality is not robust to error unless the network is transitive or assortative. In these cases the network structure allows the depth algorithms to effectively capture breadth as well as depth. Finally, we discuss the algorithms' cognitive and computational demands. This is an important consideration in systems in which individuals use the collective opinions of others to make decisions.
| Decision making in complex societies requires that individuals be aware of the group's collective opinions about themselves and their peers. In previous work, social power, defined as the consensus about an individual's ability to win fights, was shown to affect decisions about conflict intervention. We develop methods for measuring the consensus in a group about individuals' states, and extend our analyses to genetic and cultural networks. Our results indicate that breadth algorithms, which measure consensus by taking into account the number and uniformity of an individual's direct connections, correctly predict an individual's function even when some of the group members have erred in their assessments. However, in cases where nodes “form opinions” about other nodes using indirect information algorithms that measure the depth of consensus, like Eigenvector Centrality, are required. One caveat is that Eigenvector Centrality is not robust to error unless the network is transitive or assortative. We also discuss the algorithms' cognitive and computational demands. These are important considerations in systems in which individuals use the collective opinions of others to make decisions. Finally, we discuss the implications for the emergence of social structure.
| A goal of many of network studies (e.g. [1]–[4]) is to predict the effects of perturbations, such as extinction and predation events, on network structure. Making these predictions requires information about network connectivity (e.g. is the network scale-free, exponential, etc.). When the connectivity is non-uniform, it is also important to quantify variation at the node level in order to identify nodes that, if removed, are likely to impact negatively network stability. This is well recognized and many useful methods have been developed for measuring this variation [1]–[2], [5]–[14] in a range of networks, including the world-wide web [5], food webs describing trophic interactions [1], [2], networks of interactions between genes and proteins [6]–[10], and social networks, in both animal and human societies [11]–[16].
Patterns of connectivity can also influence node function in the larger system of which the network is a part. For example, in previous work on the behavioral causes of multi-scale social structure in primate societies [14], [17]–[22] it was found that group consensus about an individual's ability to win fights – its social power (see Sec. Primate communication network)–is population coded in a status signaling network. In this system, individuals use subordination signals to communicate to adversaries that they perceive themselves to be the weaker opponent. The signals are often repeated and are always unidirectional (emitted by one individual in a pair but not the other). A single signal indicates that the sender perceives the receiver capable of using force successfully against him. The frequency of signals emitted (over some defined time period) indicates the strength of the sender's perception that the receiver can successfully use force against him. In the work cited above it was demonstrated that consensus in the group about individual ability to successfully win its fights can be calculated by quantifying uniformity in the weighted in-degree distribution of signals sent to by its senders and weighting this score by the total number of signals received (this calculation is described in Sec. Shannon consensus). The resulting score for may not be the preferred score for of any specific group member, but can be said to reflect the group's collective view about how good is at winning fights. Correspondingly, the rank order associated with the distribution of scores in the population might not match the preferred rank order of any single individual, but as the outcome of integrating over all of the individual opinions, it can be said to be the consensus social power rank order.
The data indicate that individuals can estimate their own social power and also know something about how others in the group are collectively perceived [17], [20], [23], [24]. Consequently, social power is informative about the likely cost of interaction when interactions are not strictly pair-wise (a common feature of these systems and the reason why a consensus-based definition is important) [17]–[19]. Under heavy-tailed power distributions, in which a few individuals are disproportionately powerful, conflict management mechanisms like third-party policing (a critical social function) can emerge and are performed by nodes in the tail of the power distribution [14], [20]. Policing is an important social function because by controlling conflict it facilitates edge building by nodes in the signaling as well as other social networks [18], [20]. These results suggest that (1) network structure can encode node function and that (2) measures that quantify agreement in node connectivity patterns can be used to decode this population coding of node function. In Table 1 we give several examples of other networks in which node function might also be population coded and consensus estimation could be useful for identifying important nodes.
In principle, consensus about node state or function can be quantified by measuring the uniformity of a node's weighted in-degree distribution [14], as in the above example, by measuring the “flow” into and out of a node (depth), or using simple counts. To capture these competing notions of consensus, we introduce a variety of alternative information theoretic, diffusion, and count algorithms that capture breadth and depth to different degrees, and so serve as hypotheses about how functional variation in nodes is encoded in interaction networks via consensus. The algorithms take an interaction network as input and produce a vector of scores for the nodes in the network as output. We interpret the score of node as the collective opinion, or consensus, about state or its capacity to perform a given behavior. We note that the algorithms only quantify agreement in the connectivity patterns; what the consensus is about– state– depends on the type of interactions in the interaction network. For example, in the work on power in primate societies mentioned above, the interaction matrix contained directed subordination signals. These signals have special properties that allow them to reliably encode information about the ability to win fights, which is the basis of power [19]. We discuss the importance of the interaction matrix for the interpretation of consensus in greater detail in Sec. Background and motivation.
After introducing the algorithms, we compare their mathematical properties, and in a few cases, establish approximate equivalence. We introduce three data sets that we use to empirically evaluate how well the output of the algorithms predicts node function out of sample. We investigate the properties of these algorithms that make them predictive measures of consensus. The data sets include a status communication network in a primate society, a network of collaborating condensed matter physicists from a prominent journal, and a functional linkage network of yeast genes that influence viability and growth. Finally, we assess the sensitivity of the algorithms to systematic error at the node level and strategic manipulation of the network by nodes or small sub-sets of the network.
For all of the algorithms we consider, we begin with a matrix of interactions . (For discussion of data in the interaction matrix, see Secs. Primate communication network, Physics collaboration network, and Functional gene linkage network). We adopt the convention that if the interactions are directed , denotes the number of interactions traveling from individual to individual . Tables 2 and 3 contain alphabetical lists of all matrices and variables used in the text.
The algorithms we compare fall into three classes with respect to consensus: algorithms that accord higher scores to nodes with a more uniform in-degree distribution (breadth algorithms), algorithms that accord higher scores to nodes that are sinks in the network (depth algorithms), and count based algorithms, like the Borda count. For each algorithm, we construct a new matrix that gives the strength of the interactions in a way that depends on what the algorithm is meant to measure. For the breadth algorithms, we calculate the strength of the interaction between nodes and in a way that depends on all of in-edges and reflects the uniformity of the interactions from the whole population to . These measures therefore measure the breadth of consensus.
We then perform matrix operations on the appropriately transformed interaction matrix. The depth algorithms treat events as a diffusion process over the network and weight more heavily interactions with partners who themselves have many interaction partners and so forth. To capture this chain-like property of interactions, we do repeated matrix operations. These measures therefore measure the depth of consensus. We can quantify depth by the number of matrix operations required to perform the algorithm or, equivalently, the length of the “chains” in the network that affect the algorithm. Figure S1 shows a flow chart that summarizes the steps required to derive the distribution of consensus scores under each algorithm. We discuss the computational complexity of the algorithms in Section Cognitive and computational complexity and in the Text S1, Section Computational complexity).
Here we consider one additional algorithm, the Borda count, for computing consensus on networks. The Borda count is an algorithm that is traditionally used to determine the outcome of an election. Each member of a voting population ranks the candidates of the election. This is analogous to each individual in a primate group emitting signals to others in accordance with whom they perceive as more or less likely to use force successfully. The Borda count aggregates these preferences into one ranking over the candidates.
Supposing that there are candidates, each voter gives votes to his highest preference, to his next highest choice, on down to one vote to his least favorite candidate. A voter can rank candidates equally and the candidates' votes in this case are the average of the numbers of votes they would have received were they not tied. A candidate's score is the sum of his votes from each voter. In the signaling case, the receiver of the most signals from a given individual will receive n “votes” and the receiver of the fewest signals from that individual will receive one “vote”. In unweighted networks, each individual “voter” divides the group into nodes with whom he does or does not interact, giving the same number of votes to the individuals in each class. Mathematically, we define a matrix such that is the rank given node by node , where gives rank to its highest preference and rank to its lowest preference, and define the vector is the Borda consensus scores for node .
The Borda count is more coarse-grained than the total frequency of interactions received because information about the number of interactions received is lost and only the ordinal ranking of nodes by the number of interactions received is used. It does, however, convey information about agreement among interaction partners. If we find that a node has a high score under the Borda count, this indicates that many other nodes rank the receiver highly and agree about its relative value to them. Hence like Shannon Consensus it should be intrinsically sensitive to certain kinds of bias in the interaction matrix (see Sec. Empirical Comparison for further discussion).
All of the algorithms we compare provide some measure of consensus in a network about the state of a given node, such that we expect they are positively correlated (these data are presented in Sec. Basics of data set). In fact, we can describe these correlations by deriving mathematical relationships between some of the algorithms. The mathematical relationships between the breadth algorithms are easiest to see. , and are related by the definition of and a simple theorem about :Consider the definitions of and :These definitions make the mathematical relationships and on the one hand and obvious.
If the network is unweighted, then we can write the following algorithms as a function of in-degree, :where is constant across nodes and depends on the total number of edges in the network. In this case, the rankings generated by these algorithms will be the same, although the actual values will be different.
Eigenvector Centrality can be related to the redistribution probabilities and in-degree. Recall that is the stochastic transition matrix where denotes the probability of walking from node to node and Eigenvector Centrality is defined by the equation Since is stochastic, for all and we can choose such that . Since , this gives If we let , then and we can show thatwhere . In the case that for all , these bounds can be combined to giveThis bound gives an indication of how is related to the number of interactions received and the redistribution weights used in the calculation of . As we increase the redistribution weight , the minimum possible Eigenvector Centrality scores increases. In general, nodes that engage in more interactions and that interact with nodes with few other interaction partners will have higher Eigenvector Centrality scores.
Much of the research on consensus aims to determine how a group comes to a single decision, such as which direction to move, who should be president, etc. [30]–[33]. In this study our aim is somewhat different. Our goal is to quantify how much consensus there is in the group (e.g. network) about the state of a node (is it on or off, is it capable of performing a target function, etc.). Hence the interpretation of consensus turns on the meaning of the edges in the network, represented by the data in the interaction matrix, as much as on the algorithm applied to the matrix to compute the consensus scores for the nodes. It is therefore critical that the interaction data used to construct the matrix be chosen carefully.
Below we provide basic details about the three test systems –a primate status communication network, a collaboration network, and a functional gene linkage network. We provide the biological interpretation of the edges in the networks and of node state, and we introduce the functional data used to empirically evaluate the algorithm's performance. We note that the mechanistic basis for consensus as an important network measurement is best understood for the primate communication network, and this fact is reflected in that section's length.
In Table 1 we provide interpretations of consensus scores for several different kinds of networks in addition to those we describe below.
We are using a primate communication network in a large captive social group of pigtailed macaques (Macaca nemestrina) to measure social power, operationalized as group consensus about individual ability to win fights.
We are using a collaboration network to measure reputation, defined operationally as group consensus about whether to work with a given scientist.
We are using a network of functional linkages between genes to measure gene importance, defined operationally as group consensus about whether to be functionally linked with a given gene (this “decision” could be made in either developmental or evolutionary time).
Each algorithm produces as output a vector of scores for nodes in the network. In Table S1 in Text S1 we present the correlations between these outputs for each network . The distribution of consensus scores for each of the networks, according to each of the algorithms considered in this paper, is presented in Figures S2, S3 and S4. Within each data set, most algorithms suggest roughly similar distributions. In the case of the signaling network, these distributions look heavy-tailed, which is consistent with the distributions of functional data. Additionally, for the signaling network, two of the algorithms– Shannon Consensus and Weighted Simple Consensus– produce distributions that are not significantly different than normal after log transform, indicating they are consistent with the log-normal distribution.
Our predictor variables are the social power indices produced by the consensus algorithms. Dependent variables include: support solicited – requests for support received by a third-party to a fight from fight participants (should be positively correlated with power); intervention cost – operationalized as the intensity of aggression received by an intervener in response to its interventions into fights among group members (should be negatively correlated with); and intensity of aggression used by an intervener during its intervention (should be negatively correlated) (these variables are defined and the data collection methods are described in Section Methods and in [14]). These dependent variables are corrected for underlying variation in tendency to fight (see Section Methods).
All algorithms are significantly correlated with the dependent variables (, Table 4). The best predictor of the dependent variables is Weighted Simple Consensus, followed closely by Shannon Consensus, and Eigenvector Centrality. The worst predictors are David's Score, Borda Count, and Simple Consensus. The most highly predictive algorithms have very similar values, so it is hard to differentiate between them based on their predictive power alone. However, as we discuss below, these algorithms vary in their sensitivity to source biases and in their computational and cognitive complexity.
In this social system, there are a few individuals in the tail of the power distribution who are disproportionately powerful [14], [17]. This is borne out in our data, as the correlation between the algorithm scores and the dependent variables is substantially higher for the top quartiles than the bottom quartiles (Figure 1).
Our predictor variables are the reputation indices produced by the consensus algorithms. The dependent variable is total amount of grant money awarded to a PI or CoPI by the National Science Foundation (see Section Methods). Of all the algorithms we consider, only Eigenvector Centrality is significantly correlated with this external variable (, Table 4). Two reasons, one mathematical and one sociological, appear to account for this result.
First, Eigenvector Centrality can distinguish between nodes that have identical local neighborhoods. In-degree can only take integer values and there is presumably an upper bound on the number of possible collaborators given time and other constraints. In this network, the highest in-degree observed is so that there are possible values, , a node's in-degree can take. As Eigenvector Centrality can take any value between and , it can give different scores to nodes with the same in-degree. In other words, Eigenvector Centrality uses global information to differentiate between nodes that are locally identical. This effect is not as noticeable in the subordination signaling network because there are only individuals in the signaling network and therefore less degeneracy in the in-degree distribution.
Second, it is perhaps not surprising that for this kind of network Eigenvector Centrality is more predictive of the dependent variable than the breadth algorithms – although physicists involved in the process of awarding grants to others are expected to recuse themselves when confronted with an application from one of their own collaborators, they may be more likely to award grants to collaborators of their collaborators. Therefore, having many collaborators may not be that helpful in receiving grant money, but scientists whose collaborators have many collaborators may have an advantage.
Our predictor variables are the importance indices produced by the consensus algorithms. The dependent variables are the viability and competitive fitness of organisms with mutated versions of the gene. For each of our algorithms, the importance scores for essential genes are significantly higher than the importance scores for non-essential genes (, Table 4). Similarly, for each algorithm, the importance scores are significantly negatively correlated with the competitive fitness variable (,Table 4). The most predictive algorithms are, in order, Eigenvector Centrality , Simple Consensus , the Borda count, and Shannon Consensus . In differentiating between essential and non-essential genes, Eigenvector Centrality is marginally better than the other algorithms. In predicting competitive fitness, the four most predictive algorithms perform equally well.
With both external variables, the test statistics are noticeably smaller for the Graph Laplacian than for the other algorithms. As we showed above, on unweighted networks,On both the collaboration and linkage networks, nodes with high in-degree tend to interact with many other highly connected nodes. For both networks, we find high correlations between in-degree, , and the sum of the in-degrees of a node's neighbors, (, , for the collaboration network and , , for the linkage network). Nodes that have many interactions with other highly connected nodes receive low Graph Laplacian scores, a counterintuitive result that suggests the Graph Laplacian is not a robust measure of consensus. We summarize the predictive performance of the algorithms on the three data sets in Table 5.
An important question in evaluating the performance of a consensus algorithm is how sensitive the algorithm is to deficiencies in the data in the interaction matrix. Aspects of this question have been addressed in previous work. Ghoshal et al. [57] showed that in scale-free networks of sufficient size, if all edges in the network are shuffled but the in-degrees maintained, the ranking of the nodes according to eigenvector centrality is not severely perturbed. This type of shuffle allows the researcher to simulate the effects of missing or noisy data in the interaction matrix on an algorithm's output. We are particularly interested in the effects on the algorithm's output of nodes systematically making errors in their assessments of the states of other nodes or nodes attempting to manipulate social structure by “loading the deck” or inflating the consensus scores of nodes by, for example, manipulating the weighted degree distribution. (One way to manipulate the weighted degree distribution is to inflate a node's weighted in-degree by sending many signals.) Capturing this kind of “deficiency,” which we call source bias requires a different kind of shuffle.
First, we measure in our interaction matrices the correlation between a node's Shannon entropy (as defined in Sec. Shannon consensus) and the total frequency of interactions it receives (weighted in-degree or in-degree) (see Table S1 in Text S1). If entropy and in-degree were poorly correlated, we could independently evaluate the effects of receiving many interactions from receiving interactions from many individuals. However, this is not the case on the data sets we consider. We break the correlation by systemically shuffling the data in the matrices such that we create matrices with strong source biases but conserve the total number of interactions (e.g. signals) received. We now have two matrices –the original, unshuffled matrix, and the shuffled matrix, . We then compute consensus scores for the nodes using the unshuffled and shuffled matrices and assess how much the rank order changes under the shuffle.
More specifically, for a given pair of interaction partners in the network, say nodes and , we construct a matrix in which the target node, receives all of its interactions from partner node, . If the original network is directed, we hold constant the out-edges of in addition to holding constant weighted in-degree. If the original network is undirected, we maintain the symmetry. The subordination signaling network is small enough so that we can perform this shuffle for every pair of partners. However, the collaboration network and the functional linkage networks are too large to exhaust every pair of partners, so we choose of the nodes that are also represented in the functional data sets. Partner nodes are chosen at random from the target node's neighbors.
An algorithm is said to be sensitive to source bias if the rank order of the shuffled matrix, differs from the rank order of the original matrix, . Large changes in the rank order indicate that the test algorithm tends to give higher scores to nodes that interact with many neighbors than to nodes that interact strongly with just one other node and is an indication that the algorithm is sensitive to source biases.
We find that Shannon Consensus , and the Graph Laplacian, , tend to be quite sensitive to source biases (Figure 2). This is expected, as and depend on the entropy of the receiving distribution, which is by design in the shuffled matrices.
By definition, in-degree, , is maximally insensitive to source bias as we hold it constant in our shuffle. Eigenvector Centrality, , is also fairly insensitive, but the explanation why is initially counter-intuitive. As can be seen in Figure 2, Eigenvector Centrality appears to be particularly insensitive to the shuffle for the subordination signaling network, as the rank order for the shuffled and unshuffled matrices for that network is very similar. The reason for this is that in the subordination signaling network individuals who receive many signals receive some of these signals from partners who themselves receive many signals. In addition individuals who receive many signals send very few signals. Hence there is information about breadth encoded in the second and third order connections (and so forth) in the network. Even if we shuffle the matrix so that all of an individual's signals come from a single other node, as long as we hold constant the out-edges of , the in-edges to are likely to be from an individual who itself receives relatively many signals. Eigenvector Centrality, by emphasizing paths through the network, takes these second and third order connections into account. It is consequently likely to get the rank order right, even after we reduce the diversity or breadth in the first order or direct connections, as long as the second and third order connections in the shuffled matrices encode information about the first order connections in the unshuffled matrices. See Text S1, Section Sensitivity of eigenvector centrality on transitive networks for more discussion of the relationship between transitivity and sensitivity to source bias.
This suggests that measures of consensus that emphasize depth– paths through the networks –also implicitly measure consensus breadth when there is some degree of either assortativity or transitivity in the network, and work well because of these features. In the absence of transitivity, or when transitivity is very low, depth measures like Eigenvector Centrality, should not perform well as measures of consensus, unless, as in the case of the Graph Laplacian, they explicitly incorporate Shannon information. In the Text S1 we provide details on an additional analysis we performed to evaluate algorithm sensitivity to source bias.
The algorithms we investigate take as an input a matrix of physical or social interactions. These interactions encode “perceptions” or “opinions” about a target node's state or abilities. The algorithms produce as an output scores that quantify agreement or consensus in the population about this state or ability. We find that the best performing algorithms (in terms of prediction and robustness) are those that capture the breadth of agreement among a node's population of partners (see Table 5). Capturing the breadth of agreement requires quantifying the diversity of the target node's population of partners and weighting this by the number of interactions. Our analyses suggest that algorithms with these properties are robust to source biases in the interaction matrix arising from efforts to manipulate output or when nodes make systematic errors in assessing the state of a target node.
We find that in a primate status communication network, the most predictive algorithm is Weighted Simple Consensus, which is also one of the more computationally minimal algorithms we consider. In contrast to the class of algorithms that make use of Shannon entropy, Weighted Simple Consensus is not particularly sensitive to source bias. This suggests a tradeoff between computational complexity and sensitivity to source bias.
In the physicist collaboration network indirect measures of an author's (the target node) state are important, as decisions about the target node's reputation can be based on the reputation of its associates. Depth-based algorithms like Eigenvector Centrality capture these indirect effects. One caveat is that these algorithms only work as measures of consensus when networks are characterized by an elevated degree of transitivity or assortativity.
We conclude that in general the uniformity based algorithms are preferable but that Eigenvector Centrality is suitable if the network is transitive or assortative and if there is a mechanistic reason to believe that it is important to take second and third order connections into account. We discuss these issues in greater detail below.
The results reported in this paper and elsewhere([17], see also [60]–[62]) suggest that at least in social networks nodes may be making strategic decisions about social interactions using knowledge of how they are perceived by the group. For example, the individuals in the primate study group appear to estimates of their relative power to make decisions about whether to intervene in conflicts [17]. This requires that they have some knowledge of moments or properties of the distribution of power (e.g. approximate variance). An important question is how individuals extract this information [22], [63]. More generally, what do animals know about social structure and collective dynamics, how precise are their estimates, and what heuristics might they use to make calculations [64]?
It would be useful, for example, to be able to quantify the algorithmic complexity of each algorithm so that we could rank calculations by some measure of computational difficulty (see also [65]). Ideally, we would also like to know how sensitive each algorithm is to the input data. (e.g. is the exact number of signals received by individual critical, or will a rough estimate do?) for the output distribution of power to be a useful predictor of out of sample data. Addressing this robustness question would help to determine how much room there is to relax the mathematical requirements of a given algorithm, and find a heuristic simple enough for this study species.
Ranking the algorithms by their algorithmic complexity is a long way off, if achievable at all. As is illustrated in Figure S1, we can only crudely rank the algorithms given what we know about the minimum number of steps each requires in order to estimate critical quantities from an empirical perspective – the absolute power of individual , the relative power of (e.g. where it falls in a power distribution of a given type), and the moments of the power distribution. In most circumstances it seems unlikely that we, or the animals, would be interested in an isolated individual's score. This is because it is not her power value that is important, but rather where an estimated value falls in a distribution of power scores. Yet calculation of absolute and relative power require different computational approaches and a preliminary assessment suggests that the difficulty of these steps varies across algorithms. We discuss these issues in greater detail in Text S1, Section Computational complexity.
In addition to approaching the problem of complexity mathematically, we can approach it empirically by asking how sensitive the algorithms are to imperfect information in the input matrices. For example, perhaps the individuals in our system cannot discriminate based on identity and can only remember classes of individual (e.g. male or female, or matriline x or y, etc.), signals or signalers, or an interaction history of length . By coarse-graining the input data, it is in principle possible to test how sensitive the algorithms are to this kind of imperfect information resulting from various cognitive or spatial constraints. Aspects of this question have been addressed in previous work, as discussed in Section Sensitivity of the algorithms to source biases. However, many questions remain open for future work.
If node function in many different systems is collectively encoded in interaction networks and this information is decodable by quantifying the agreement in network connectivity patterns, this would suggest that consensus formation is at the core of sociality. Consider the primate society used as a model system in this paper.
Power in our primate study group is a critical social variable. However power is not a simple variable. The distribution of power does not map directly onto a distribution of body sizes or even a distribution of fighting abilities. Rather it consolidates as multiple interacting individuals learn about fighting abilities and signal about this to reduce social uncertainty [14], [19], [21], [22], [63]. When the statistics used to operationalize an aggregate social property, like power structure, are more than simple counts over strategies, and when the inputs are not simply individual traits but network data, we need to worry explicitly about the mappings between behavioral strategies and decision-making at the microscopic level and social organization [65]–whether we are working with the social organization of primates or of cells forming a tissue. A central question becomes, How do strategies get collectively combined by multiple components to produce macroscopic social properties? How much degeneracy characterizes this mapping? Once we can describe the developmental dynamics giving rise to an aggregate social property, we will be in a position to study how the social processes producing power and other kinds of social structure have evolved in a wide range of systems.
The data set, collected by J.C. Flack, is from a large, captive, breeding group of pigtailed macaques that was housed at the Yerkes National Primate Research Center in Lawrenceville, Georgia.
The physicist collaboration network was collected by Mark Newman, as described in [38], and is available at http://www-personal.umich.edu/~mejn/netdata/. The data were initially collected from the Los Alamos e-Print Archive, now the arXiv at http://arxiv.org. Since initial publication in 2001, the network has been updated with collaborations from the arXiv through 2005. scientists are represented in the network and the collaborations occurred between January, 1995 and March, 2005.
The National Science Foundation makes the data about awarded grants publicly available at http://www.nsf.gov/awards/about.jsp. For each scientist in the collaboration network, we searched this database for any grant concerning condensed matter physics on which the scientist was one of the investigators. If the scientist was awarded more than one grant, we summed the total amount of grants awarded him or her. Grant data was available for of the scientists in the collaboration network. The grants were awarded between September, 2008 and September, 2012, with one grant starting in September, 2004.
The functional linkage network was constructed by Lee et al., as described in [56] and [51], and is available at http://www.yeastnet.org/. Functional linkages between genes are associations that “represent functional constraints satisfied by the cell during the course of the experiments” [56]. Evidence of a functional linkage between two genes was provided by mRNA coexpression levels, the results of protein interaction experiments, phylogenetic profiles, and the co-occurrence of the two genes in a scientific paper [51], [56]. Lee et al. combined these data to calculate the log-likelihood that two genes are involved in a similar function. In our analyses, we say an edge is present if its log-likelihood score is greater than and is absent otherwise. The resulting network has nodes.
The Saccharomyces Genome Database maintains information about the phenotypic effects of genes in the yeast genome at www.yeastgenome.org. Two phenotypic effects reflect a gene's overall importance. One measure is the viability of organisms with a mutant version of the gene and a second measure is the competitive fitness of organisms with a mutant version of the gene. The viability measure is binary: a mutation to a gene can lead to either a viable or an inviable organism. An inviable organism is one that is unable to grow under standard growth conditions for S. cerevisiae, defined as glucose-containing rich medium (YPD) at C. A gene's competitive fitness is given by the relative growth rate of an organism with a mutated version of the gene compared to one with the normal genotype. Greater competitive fitness is indicated by a relative growth rate of greater than . These experiments can be performed in various media: we only used those performed in minimal medium to standardize our comparisons. More information is available at http://www.yeastgenome.org/help/function-help/phenotypes. These phenotype data are available for of the genes in the linkage network.
|
10.1371/journal.pgen.1000436 | A Common Variant Associated with Dyslexia Reduces Expression of the KIAA0319 Gene | Numerous genetic association studies have implicated the KIAA0319 gene on human chromosome 6p22 in dyslexia susceptibility. The causative variant(s) remains unknown but may modulate gene expression, given that (1) a dyslexia-associated haplotype has been implicated in the reduced expression of KIAA0319, and (2) the strongest association has been found for the region spanning exon 1 of KIAA0319. Here, we test the hypothesis that variant(s) responsible for reduced KIAA0319 expression resides on the risk haplotype close to the gene's transcription start site. We identified seven single-nucleotide polymorphisms on the risk haplotype immediately upstream of KIAA0319 and determined that three of these are strongly associated with multiple reading-related traits. Using luciferase-expressing constructs containing the KIAA0319 upstream region, we characterized the minimal promoter and additional putative transcriptional regulator regions. This revealed that the minor allele of rs9461045, which shows the strongest association with dyslexia in our sample (max p-value = 0.0001), confers reduced luciferase expression in both neuronal and non-neuronal cell lines. Additionally, we found that the presence of this rs9461045 dyslexia-associated allele creates a nuclear protein-binding site, likely for the transcriptional silencer OCT-1. Knocking down OCT-1 expression in the neuronal cell line SHSY5Y using an siRNA restores KIAA0319 expression from the risk haplotype to nearly that seen from the non-risk haplotype. Our study thus pinpoints a common variant as altering the function of a dyslexia candidate gene and provides an illustrative example of the strategic approach needed to dissect the molecular basis of complex genetic traits.
| Dyslexia, or reading disability, is a common disorder caused by both genetic and environmental factors. Genetic studies have implicated a number of genes as candidates for playing a role in dyslexia. We functionally characterized one such gene (KIAA0319) to identify variant(s) that might affect gene expression and contribute to the disorder. We discovered a variant residing outside of the protein-coding region of KIAA0319 that reduces expression of the gene. This variant creates a binding site for the transcription factor OCT-1. Previous studies have shown that OCT-1 binding to a specific DNA sequence upstream of a gene can reduce the expression of that gene. In this case, reduced KIAA0319 expression could lead to improper development of regions of the brain involved in reading ability. This is the first study to identify a functional variant implicated in dyslexia. More broadly, our study illustrates the steps that can be utilized for identifying mutations causing other complex genetic disorders.
| Dyslexia, or reading disability (RD), is a condition that affects an individual's ability to read and spell in the absence of any obvious sensory or neurological impairment and despite adequate intelligence and educational opportunity [1]. RD is one of the most common learning disabilities in school-aged children, with a prevalence ranging from 5% to 17.5% [2],[3]. Although the specific causes of the disorder have yet to be elucidated, it is generally accepted that RD has a strong genetic component [4],[5]. Family studies have estimated a high heritability of RD, reporting an incidence of about 40% in siblings of affected individuals [6],[7]; twin studies have shown a concordance rate of 68% in monozygotic twins versus 38% in dizygotic twins [8].
Numerous candidate genes have emerged from genetic association studies and the characterization of chromosomal translocations in individuals with RD, including DYX1C1 on 15q21 [9]–[11], ROBO1 on 3p12 [12], DCDC2 [13],[14] and KIAA0319 [15]–[19] on 6p22, and MRPL19 and C2ORF3 on 2p12 [20]. Several of these genes have been implicated in brain development [21]. In particular, RNAi-knockdown studies suggest that DYX1C1 [22]–[24], DCDC2 [13],[25], and KIAA0319 [26] play a role in neuronal migration during the development of the rat cortex. Interestingly, altered neuronal migration has been implicated in RD based on the only post-mortem anatomical study conducted to date [27]; specifically, the brains of dyslexic individuals were found to have subtle structural anomalies consistent with defective neuronal migration.
We previously detected an RD-associated ‘risk haplotype’ through an association analysis of candidate genes residing at the chromosome 6p22 locus, which is one of the most consistently identified candidate regions by linkage studies [15]. Single-nucleotide polymorphisms (SNPs) were selected within brain-expressed genes and used for subsequent genetic analyses of RD. A 77-kb region of high inter-marker linkage disequilibrium (LD) that includes the first four exons of KIAA0319, all of TTRAP, and the region immediately upstream of THEM2 (Figure 1A) showed significant associations with RD. Three SNPs captured most of the genetic variation and described the most common haplotype in the 77-kb region. One of these haplotypes, which was effectively tagged by the rs2143340 marker, was found to be significantly associated with RD. The association between this risk haplotype and reading-related traits was detected in two independent family-based sample sets of U.K. and U.S. origin [15]. Association with the same region was reported in a completely independent study [16]. Most recently, we replicated the association between the risk haplotype and reading-related phenotypes in an unselected sample of more than 6,000 children from the Avon Longitudinal Study of Parents and Children (ALSPAC) [19]. Using a quantitative allele-specific gene expression assay, we showed that there is reduced expression of KIAA0319 (but not TTRAP and THEM2) from the risk haplotype in both lymphoblastoid and neuronal cell lines [26]. These data are consistent with the findings of a comprehensive association study, which tested an identical set of SNPs within the chromosome 6p22 locus in two independent U.K. sample sets [17]. The strongest association with RD was found with SNPs near the first exon of KIAA0319 in both sample sets. Taken together, these data suggest that the risk haplotype might harbor a regulatory variant that alters KIAA0319 transcription.
Here, we report additional genetic and functional characterization of the risk haplotype, specifically focusing on variants within the putative regulatory element(s) immediately upstream of KIAA0319. Our results implicate one variant as the likely cause of reduced KIAA0319 expression from the risk haplotype. More broadly, these findings are relevant for further understanding the role of KIAA0319 in RD and brain development as well as for establishing the role of non-coding mutations in complex genetic diseases.
Thirteen human bacterial artificial chromosomes (BACs) spanning the 77-kb RD-associated region were obtained either from Children's Hospital Oakland Resource Institute or the California Institute of Technology. The BACs were genotyped for the three risk haplotype-tagging SNPs (rs4504469, rs2038137, and rs2143340 [15]) using the Sequenom platform, according to the manufacturer's instructions. BACs RP11-195J19 [containing the risk haplotype (‘risk BAC’); GenBank accession number CR925830] and RP11-948M1 [containing a non-risk haplotype (‘non-risk BAC’); GenBank accession number CR942205] were chosen for their similar genomic coverage. Both BACs were sequenced at the Wellcome Trust Sanger Institute. Variants were detected by pair-wise comparisons using AlignX in the Vector NTI Advance 9 program (Invitrogen).
The collection of families used for quantitative trait association has been extensively described [15]. Briefly, all probands and siblings from our complete Oxford set of 264 unrelated nuclear families were identified from the dyslexia clinic at the Royal Berkshire Hospital (Reading, U.K.) and were administered a battery of psychometric tests. The following reading-related measures were used for statistical analyses: orthographic coding using irregular words (OC-irreg), phonological decoding ability (PD), orthographic coding assessed by forced word choice test (OC-choice), single-word reading ability (READ), spelling ability (SPELL), phonemic awareness (PA), and measures of IQ [verbal (SIM) and nonverbal (MAT)]. The scores were adjusted for age and IQ, and then standardized against a normative control data set as described [28],[29].
SNP genotyping was performed using either the MassARRAY hME or iPLEX system (Sequenom), according to the manufacturer's instructions (all primer sequences are available upon request). Marker-trait association was evaluated using the “total” association model with the QTDT package [30]. Variants were initially tested for association with the reading-related traits in a sample set consisting of 89 U.K. families previously described by Francks et al. [15], referred to as ‘sample 1.’ The LD among SNPs in this sample was determined using Haploview version 4.0 (http://www.broad.mit.edu/mpg/haploview) [31]. SNPs showing significant associations were tested in the entire sample of 264 families, referred to as ‘entire U.K. set,’ as well as in a phenotypically severe sample subset consisting of 126 families described previously [15],[17] and referred to as ‘severe U.K. subset.’ Briefly, the severe U.K. subset was chosen based on scores >0.5 SD below a composite mean score of the PD and OC-irreg traits, the two measures that contribute to the greatest degree to the chromosome 6p22 linkage peak [15].
Genomic sequences orthologous to the interval between TTRAP and KIAA0319 were obtained from publicly available databases (http://genome.ucsc.edu for chimpanzee, orangutan, macaque, marmoset, dog, mouse, and rat; http://www.ncbi.nlm.nih.gov/blast/Blast.cgi for horse, pig, and elephant). A multi-sequence alignment of these sequences was generated with MultiPipMaker (http://pipmaker.bx.psu.edu/pipmaker) using the sequence of the non-risk BAC as the human reference [32].
The genomic segment immediately upstream of KIAA0319 [−4,028 bp to +77 bp relative to the transcription start site (TSS)] from the non-risk BAC was cloned into the luciferase-expressing pGL3-Basic vector (Promega) using BAC recombineering [33]. Specifically, the pGL3-Basic vector was linearized with the restriction enzyme KpnI (New England Biolabs) and gel-purified (Qiagen). PCR amplification (Bioxact Long, Bioline) was performed using the linearized pGL3-Basic vector as the template and appropriate recombineering primers (see Text S1 for sequences). Electrocompetent cells containing the non-risk BAC were generated as described [34]. Column-purified (Qiagen) PCR product (2 µg), consisting of linearized pGL3-Basic vector flanked by homologous sequence to the non-risk BAC, was electroporated into 25 µl of temperature-induced SW102 E. coli [34] containing the non-risk BAC, and the cells were plated onto LB agar containing 100 µg/ml ampicillin and incubated at 32°C for 30 hours.
Constructs harboring various deletions were engineered by removing the segment between the restriction sites for EcoRV (−4,026 bp from the TSS) and the following: PmlI (−2,802 bp), BstXI (−2,185 bp), NsiI (−1,728 bp), PvuII (−940 bp), StuI (−544 bp), Bpu10I (−216 bp), Bsu36I (−97 bp), and BssHII (−24 bp). Site-directed mutagenesis of the full-length construct was performed using the QuikChange XL Site-Directed Mutagenesis kit (Stratagene) according to the manufacturer's protocol (primer sequences used for the mutagenesis are provided in Text S1). All mutated constructs were sequenced to ensure the absence of unwanted additional mutations.
SHSY5Y, SK-N-MC, and HEK293T cell lines were grown according to ECACC guidelines at 37°C with 5% CO2. All three cell lines were grown in 96-well plates at a concentration of 2.4×104 cells/well for SHSY5Y, 3.5×104 cells/well for SK-N-MC, and 1.5×104 cells/well for HEK293T. After 24 hours, the cells were co-transfected with 0.05 pmol of the pGL3-derived construct (e.g., containing the non-risk haplotype, deletions, or mutations; note that a promoter-less pGL3-Basic construct was used as a negative control) and 2 ng of pRL-CMV with 20 µl Lipofectamine 2000 (Invitrogen). At 3–4 hours post-transfection, the medium was replaced. At 48 hours post-transfection when the cells had reached approximately 90% confluency, cells were lysed, and the luminescence was assayed using the Dual Luciferase Reporter Assay (Promega). The luminescence of 20 µl of lysis product was measured using a microplate luminometer (Luminoskan Ascent, Thermo Fisher Scientific).
The transfection efficiencies were normalized to the level of pRL-CMV renilla luciferase activity, and the results reflected as ‘relative luciferase activity’ (RLA). The RLA for each transfection were scaled so that the pGL3-Basic construct (in the case of constructs harboring deletions) or the full-length non-risk haplotype-containing construct (in the case of mutagenized constructs) yielded a 1.0 RLA. All transfections were performed in quadruplicate and repeated at least three times (twelve biological replicates in total). An unpaired two-sided t-test was used to compare the RLAs between the non-risk haplotype and mutagenized constructs.
To create double-stranded EMSA probes carrying risk and non-risk alleles of the RD–associated SNPs, complementary oligonucleotides (see Text S1 for sequences) were annealed, end-labeled with [γ-32P]ATP (PerkinElmer) using 10 units of T4 polynucleotide kinase (Promega), and column-purified (GE Healthcare). Equal amounts of nuclear extract from the SHSY5Y cell line, prepared using a nuclear extraction kit (Cayman Chemical), were pre-incubated with or without an unlabeled double-stranded ‘competitor’ DNA in the presence of DNA-binding buffer (Promega) for 10 minutes at room temperature, and then incubated with the relevant 32P-labeled probe (17.5 fmol/sample) for 20 minutes at room temperature. For the ‘supershift EMSA’ [35], 2 µg of appropriate EMSA-grade concentrated antibody [OCT-1 (octamer-1), sc-232x and CRX (cone-rod homeobox), sc-30150x; Santa Cruz Biotechnology] was then added, and the sample was incubated at 4°C overnight. DNA-protein complexes were electrophoretically separated on a 6% polyacrylamide 0.5× TBE DNA retardation gel (Invitrogen) at room temperature, dried at 80°C for 1 hour, and visualized using a Fujifilm FLA-5000 image analyzer.
SHSY5Y cells, chosen because of their heterozygosity for the RD-associated risk haplotype, were reverse-transfected with corresponding siRNA cocktails or with Lipofectamine only (‘mock-transfected’). Briefly, SHSY5Y cells (4×105 cells/well) were plated in 24-well plates just before transfection and mixed with pre-incubated siRNA cocktails. For the cocktails, siRNAs for OCT-1 (sc-36119, Santa Cruz Biotechnology), a positive control [GAPDH (glyceraldehyde 3-phosphate dehydrogenase), AM4605, Ambion], or a scrambled negative control (AM4636, Ambion) were diluted with Opti-MEM media and pre-incubated with Lipofectamine 2000 (Invitrogen). Two concentrations of siRNA were used for all experiments: 1.5 and 3.0 µM. The results were consistent with both concentrations, but there was less variation and the results were more statistically significant with the 1.5 µM concentration. After incubation at 37°C with 5% CO2 for 24 hours, the medium was replaced, and the cells were incubated for an additional 24 hours. All siRNA transfections were performed in 6 biological replicates for each concentration of siRNA. Subsequently, total RNA from the cells was prepared with Trizol reagent (Invitrogen) and the RNeasy miniprep kit (Qiagen). cDNA was synthesized from 1 µg of total RNA using the Superscript III First Strand Reverse Transcriptase Kit and random hexamers (Invitrogen).
Effects on gene expression by OCT-1 and GAPDH siRNAs, compared to scrambled siRNA, were evaluated by quantitative real-time PCR (qRT-PCR) with TaqMan expression assays (4333764F for GAPDH and HS00231250_m1 for OCT-1, Applied Biosystems). Expression was measured in siRNA-transfected and mock-transfected samples, and normalized to the level of expression of endogenous B2M (ß2-microglobulin), which is not affected by siRNA transfection (assay HS00187842_m1, Applied Biosystems). For each sample, expression was measured in 4 technical replicates, and average values were used for analysis.
Allele-specific expression in cDNA samples from different transfections was measured in quadruplicate by use of allele-discriminating TaqMan genotyping assays for SNPs rs807541 and rs4504469 (C___3073667_1_ and C___390135_10, respectively; Applied Biosystems). Both SNPs are located within coding sequence of KIAA0319, and therefore both alleles could be detected in cDNA. The alleles of these SNPs represent the risk and non-risk haplotypes: the risk haplotype allele of rs4504469 was established previously [26], while the risk haplotype allele of rs807541 was established by sequencing cloned cDNA derived from SHSY5Y cells. For each assay, a standard curve consisting of 10 dilutions of two HapMap DNA samples homozygous for either the risk or non-risk haplotype allele was generated (rs4504469: NA10847, NA12761; rs807541: NA10847, NA18858). The standard curve was used to validate the assay quality and to generate a regression equation necessary for determining the relative allelic ratio in the experimental samples. The relative ratio of the two alleles (A and B) was measured as the ratio between VIC and FAM fluorofores, which were attached to the two different corresponding allele-specific probes in each case. Specifically, the Ct (cycle at threshold) values were averaged between technical replicates, and the differences between the two alleles were calculated as ratio(A/B) = ratio(VIC/FAM) = Ct(VIC)−Ct(FAM) = dCt. The ratios of known dilutions of the HapMap DNA samples were plotted relative to dCt, and a linear regression model fitted to the data. The allelic ratios for the experimental samples were calculated using dCt in the regression equation. An unpaired two-sided t-test was used to compare the means between groups of samples.
Pair-wise sequence comparison of the risk and non-risk BAC sequences revealed eight variants within the 4-kb region between TTRAP and KIAA0319: one simple repeat and seven SNPs [designated SNP 1 through SNP 7 (Figure 1A)]. These seven KIAA0319 promoter region SNPs were genotyped in sample 1 (see Methods) and tested for association with various reading-related traits. In addition to the previously reported associations with rs3212236 (SNP 4) [17] and rs9467247 (SNP 5) [15], we found that rs9461045 (SNP 2) is significantly associated with many reading-related traits (Figure 1B and Table S1). Specifically, the minor alleles of these SNPs are most significantly associated with OC-irreg (P = 0.0002, SNP 2; P = 0.0002, SNP 4; P = 0.0001, SNP 5). An evaluation of LD across the region (Figure 1C) showed that all SNPs but rs28501680 (SNP 1) are in strong LD with rs2143340, the previously implicated risk haplotype-tagging SNP (residing within the TTRAP gene) [15]. SNPs 2 and 5 are in perfect LD with each other, sharing the same minor allele frequency of 0.19 (i.e., the minor alleles of both SNPs always occur together), with the slight differences in association P-values likely reflecting different genotyping success rates (Table S2). We followed up these findings by genotyping SNPs 2, 4, and 5 in both the entire U.K. set (Table 1 and Table S3) and the severe U.K. subset (Table 1 and Table S4). SNPs 2 and 5 show the strongest association detected so far with these samples. Both SNPs are most significantly associated with OC-irreg (P = 0.0046, SNP 2; P = 0.0025, SNP 5) in the entire U.K. set. Additional significant associations were found with the severe U.K. subset for OC-irreg (P = 0.0006, SNP 2; P = 0.0003, SNP 5), OC-choice (P = 0.0003, SNP 2; P = 0.0001, SNP 5), and READ (P = 0.0003, SNP 2; P = 0.0002, SNP 5).
We also performed comparative analyses of the genomic region 4 kb upstream of the KIAA0319 TSS using sequences from 11 vertebrate species (Figure 1D). Multi-species sequence comparisons can reveal genomic segments under evolutionary constraint due to their functional importance [36]–[39], such as serving a role in transcriptional regulation. Overall, there is little conservation of this upstream region across species, evident by a paucity of multi-species conserved sequences identified on the UCSC Genome Browser (http://genome.ucsc.edu) [40] ‘Most Conserved’ track (Figure S1), which compares orthologous sequences from 12 different species. The most pronounced conservation across species extends from the TSS to approximately 1 kb upstream of KIAA0319. This region includes SNPs 5, 6, and 7, and likely encompasses the promoter and perhaps other upstream elements important in regulating KIAA0319 expression. Examination of this region using the UCSC Genome Browser reveals a predicted CpG island, a DNase I hypersensitive site, a FirstEF-predicted promoter, and evidence for sequence conservation in certain species (Figure S1). SNP 5 is the only variant within this conserved region showing association with RD; analysis of SNP 5 reveals that the nucleotide on the non-risk haplotype (G allele) is conserved across primates only (Figure 1D). In the case of the other two associated variants (SNPs 2 and 4), the SNP 2 nucleotide on the non-risk haplotype (G allele) is conserved across all species examined except marmoset and horse, while the SNP 4 nucleotide on the non-risk haplotype (A allele) is conserved across all species examined except pig. Note that the sequences encompassing SNPs 2 and 4 could not be aligned with orthologous mouse or rat sequences.
We generated a series of luciferase-expressing constructs containing progressively smaller segments of the genomic region immediately upstream of KIAA0319 (derived from the non-risk BAC), and tested each construct in two neuronal cell lines, SHSY5Y and SK-N-MC (Figure 2A). Neuronal cell lines were chosen based on the strong expression of KIAA0319 in the developing human brain [26]. These studies indicated the presence of promoter activity between −24 and −97 bp of the KIAA0319 TSS. TRANSFAC [41] analysis of this interval revealed predicted binding sites for the transcription factors RFX1 (regulatory factor X, 1) and ETF (epidermal growth factor receptor transcription factor); we also identified the same RFX1-binding site using the UCSC Human Genome Browser (Figure S1). Site-directed mutagenesis of the RFX1- or ETF-binding site significantly reduced luciferase expression (Figure 2B), although neither mutated site was associated with a complete loss of promoter activity. Interestingly, ETF is known to drive transcription from promoters that are GC-rich and lack a TATA box [42]; this is the case for the putative promoter of KIAA0319, which includes an in silico-predicted CpG island (Figure S1). None of the seven SNPs we identified between TTRAP and KIAA0319 reside in this putative promoter region.
Additionally, transcriptional silencing activity appeared to be associated with the region from −97 to −216 bp of the KIAA0319 TSS, an interval in which TRANSFAC predicted a Pax-6 (paired box gene 6) binding site. While SNP 7 falls within this region, it does not interrupt this predicted binding site or show strong association with any reading-related traits.
We next investigated the effect of the three variants highly associated with reading-related traits (SNPs 2, 4, and 5) on luciferase expression using mutagenized versions of the above-described non-risk haplotype construct (Figure 3A). In these studies, we directly compared the non-risk versus risk allele for each SNP, measuring luciferase expression in SHSY5Y and SK-N-MC cells as well as in HEK293T, a human embryonic kidney cell line; this allowed examination of promoter activity in neuronal as well as non-neuronal cell lines. Introduction of the SNP 2 risk variant significantly reduced luciferase expression (by 35–57%) in all three cell lines. The SNP 4 risk variant increased luciferase expression in SHSY5Y cells, but not in SK-N-MC or HEK293T cells; the SNP 5 risk variant had a negligible effect on luciferase expression in these cell lines. These findings suggest that SNP 2 may contribute to the reduced KIAA0319 expression seen from the risk haplotype.
EMSAs were performed to investigate the potential role of SNP 2 in modulating transcription factor binding. A probe corresponding to the risk (but not the non-risk) allele of SNP 2 binds nuclear protein(s) in an EMSA (Figure 3B). No allele-specific nuclear protein binding was detected by EMSA for either SNP 4 or 5 (data not shown). In silico analysis of the sequence encompassing SNP 2 using TRANSFAC revealed that the risk variant creates a putative binding site for CRX and OCT-1. Accordingly, we performed an EMSA in the presence of unlabeled competitors containing known binding sites for human CRX and OCT-1, respectively. Both competitors ablated binding of the nuclear protein(s) to the probe containing the SNP 2 risk variant (Figure 3C). We also performed a supershift EMSA (see Methods) with anti-CRX or anti-OCT-1 polyclonal antibody, and found that the presence of the anti-OCT-1 (but not anti-CRX) antibody decreased the observed binding (Figure 3D). These data provide in vitro evidence of a functional mechanism by which the SNP 2 risk allele contributes to the reduced KIAA0319 expression through creation of a binding site for the transcription factor OCT-1.
Using an siRNA, we knocked-down OCT-1 expression in SHSY5Y cells, which Paracchini et al. [26] previously showed express KIAA0319, and are heterozygous for the risk haplotype. This siRNA reduced OCT-1 expression by about half (versus transfection with a scrambled siRNA, P = 0.0002). We then measured the effect of OCT-1 knock-down on KIAA0319 expression using allele-specific qRT-PCR assays for two heterozygous coding SNPs residing within KIAA0319 (rs807541 and rs4504469). These SNPs showed mean allelic ratios significantly lower than 1.0 (Figure 4) after transfection of a scrambled siRNA (negative control); in particular, the results for both SNPs indicate that KIAA0319 expression from the risk haplotype is lower than from the non-risk haplotype, with risk:non-risk allelic ratios of Xrs807541 = 0.48±0.13 and Xrs4504469 = 0.57±0.08 (in agreement with values previously reported by Paracchini et al. [26]). Following OCT-1 knock-down, the allelic ratios were significantly closer to 1.0 (Xrs807541 = 0.81±0.06 and Xrs4504469 = 0.85±0.04), consistent with an increase in KIAA0319 expression from the risk haplotype.
In this study, we sought to identify a variant(s) on the RD-associated risk haplotype [15] that decreases expression of KIAA0319. Our experimental data consistently indicate that the minor allele of rs9461045 (SNP 2) is likely to be functionally relevant for the development of RD. Specifically, we have shown that the risk allele of rs9461045: (1) is one of the markers most significantly associated with RD in our set of families; (2) decreases gene expression in luciferase-based assays; and (3) creates a binding site for a nuclear protein(s), likely to include the transcriptional silencer OCT-1. Moreover, the role of OCT-1 was further supported by the increase in KIAA0319 expression from the risk haplotype upon siRNA-mediated knock-down of OCT-1.
The chromosome 6p22 risk haplotype is a well-established genetic risk factor for reading problems in populations of European descent, showing association in at least two sets of families with RD [15] and a large unselected set of additional individuals [19]. The data we present here help to provide an explanation for previous contradictory reports that failed to replicate an RD-association with the risk haplotype. Specifically, Luciano et al. [18] detected an opposite trend of association, showing that the same haplotype was associated with good (as opposed to poor) reading skills in an unselected Australian sample set. A different LD structure of the region in the population examined in this latter study can explain these apparently divergent findings, as previously suggested [19]. HapMap samples were analyzed for both markers, rs2143340 (the risk haplotype-tagging SNP) and rs9461045 (SNP 2), as shown in Figure S2; the detected LD differs among populations. LD is strong in the CEPH population (European descent), implying that haplotypes containing the minor allele of rs9461045 will also harbor the minor allele of rs2143340; LD between these two markers is not seen in three other HapMap populations. As such, the two markers will be present in all the possible haplotypes within these other populations, which makes it possible that, by chance, the minor allele of rs9461045 will appear more frequently in combination with the major allele of rs2143340. This scenario can explain why we see conflicting association results between studies using different populations, as is often the case in replication analyses of disease/trait associations [43]. This could certainly be the case for the Australian sample set, which is at least partially admixed. Thus, our study provides an empirical explanation for apparently contradictory complex trait-related genetic associations.
The precise function of KIAA0319 has yet to be elucidated, but it appears to play a role in neuronal migration during brain development, similar to other RD candidate genes [44] and as evidenced by its specific pattern of expression in the developing human and mouse neocortex [26]. Additionally, KIAA0319 is strongly expressed in human adult brain, specifically in the superior parietal cortex, primary visual cortex, and occipital cortex [13], areas thought to be important in reading [45]. Our studies identified two regions that may contribute to this expression specificity (Figure 2A). First, the KIAA0319 promoter has a potential binding site for RFX1, a protein shown to regulate differentiation of ciliated sensory neurons in C. elegans [46] and Drosophila [47]. Second, the region implicated as a likely silencer element contains a predicted binding site for Pax-6, a transcription factor known to play a major role in regulating cortex development [48]. It is notable that the Pax-6 and KIAA0319 genes have similar expression patterns in the developing mouse and human brains [26], consistent with their potential transcriptional regulatory interactions.
The rs9461045 risk variant creates potential binding sites for CRX and OCT-1 transcription factors, although we could only find evidence for OCT-1 binding to the risk haplotype (Figure 3D). Both CRX and OCT-1 contain DNA-binding homeobox domains with similar recognition sites [49],[50]; it is thus possible that OCT-1 was able to bind both CRX and OCT-1 competitors, which would explain the observed ablation of risk probe-binding in the EMSA with either competitor (Figure 3C). OCT-1, also known as POU2f1 (POU domain, class 2, transcription factor 1), is a ubiquitously expressed member of the POU domain factor family [51]. This protein is involved in many biological processes, and has been shown to play a role in the formation of radial glia, the cells that provide a scaffold structure for neuronal migration [52]. OCT-1 can act as a transcriptional silencer by binding to an 8-bp AT-rich target (‘octamer’) near a promoter [53]. Notably, rs9461045 falls in a 120-bp AT-rich genomic region that has relatively higher sequence identity with the orthologous regions in the horse, pig, and elephant genomes compared to the surrounding region (Figure 1D). Further, it has been shown that such AT-rich regions are important for unzipping DNA during transcription [54] and are susceptible to binding by nuclear matrix attachment proteins, such as OCT-1 [53]. While the specific region encompassing rs9461045 is not highly conserved across mammals, recent findings suggest that upwards of 50% of authentic transcription factor-binding sites are not heavily conserved, at least not based on the methods used to date for identifying multi-species sequence conservation [55]. Since the rs9461045 risk variant appears to create a human-specific transcription factor-binding site that reduces gene expression, this site may not be under evolutionary constraint.
The studies reported here provide for the first time strong evidence implicating a specific variant to be functionally relevant for RD. Our findings provide new insights for understanding the role of KIAA0319 in RD and brain development as well as for establishing the role of non-coding mutations in complex genetic diseases. A growing body of evidence suggests that variants residing in transcriptional regulatory elements (as opposed to coding regions) underlie many such disorders [56],[57]. Therefore, the experimental strategies described here more broadly illustrate a general approach that can be used for investigating the molecular basis of genetically complex diseases. Our findings also provide the first example, to our knowledge, of using siRNA to define the functional basis of allele-specific effects of genetic variants, and highlight the different approaches needed to implicate functional variants in complex genetic diseases.
|
10.1371/journal.ppat.1000655 | A Quantitative RNAi Screen for JNK Modifiers Identifies Pvr as a Novel Regulator of Drosophila Immune Signaling | Drosophila melanogaster responds to gram-negative bacterial challenges through the IMD pathway, a signal transduction cassette that is driven by the coordinated activities of JNK, NF-κB and caspase modules. While many modifiers of NF-κB activity were identified in cell culture and in vivo assays, the regulatory apparatus that determines JNK inputs into the IMD pathway is relatively unexplored. In this manuscript, we present the first quantitative screen of the entire genome of Drosophila for novel regulators of JNK activity in the IMD pathway. We identified a large number of gene products that negatively or positively impact on JNK activation in the IMD pathway. In particular, we identified the Pvr receptor tyrosine kinase as a potent inhibitor of JNK activation. In a series of in vivo and cell culture assays, we demonstrated that activation of the IMD pathway drives JNK-dependent expression of the Pvr ligands, Pvf2 and Pvf3, which in turn act through the Pvr/ERK MAP kinase pathway to attenuate the JNK and NF-κB arms of the IMD pathway. Our data illuminate a poorly understood arm of a critical and evolutionarily conserved innate immune response. Furthermore, given the pleiotropic involvement of JNK in eukaryotic cell biology, we believe that many of the novel regulators identified in this screen are of interest beyond immune signaling.
| Innate immunity is the sole immune response in the overwhelming majority of multicellular organisms and drives the sophisticated antigen-specific adaptive defenses of vertebrates. Defective regulation of immune signal transduction pathways has disastrous consequences for affected individuals and can result in life-threatening conditions that include cancer, autoimmune and neurological conditions. Thus, there is a major need to identify the regulatory circuits that govern activation of critical innate immune response pathways. The genetically accessible model organism Drosophila melanogaster is an ideal springboard for such studies, as core aspects of innate immune pathways are evolutionarily conserved and novel discoveries in Drosophila often inspire subsequent developments in the characterization of biomedically relevant mammalian pathways. Drosophila responses to certain bacterial invaders proceed through the IMD pathway, which contains partially overlapping signal transduction JNK and NF-κB arms. While substantial efforts have illuminated much of the NF-κB arm, there is a considerable paucity of information on the regulation of the JNK arm. We conducted a survey of the entire Drosophila genome for novel regulators the Imd/dJNK pathway. In this study, we uncovered a novel link between the proliferative Pvr pathway and the IMD pathway.
| The adaptive immune response is a recent evolutionary acquisition by vertebrates. In contrast, the innate immune response is highly conserved among metazoans and is the first line of defense against invading pathogens [1]. Drosophila melanogaster is a powerful model for the study of innate immune signaling events owing to the high degree of evolutionary conservation of signal transduction pathways [2]. For example, pioneering studies in Drosophila led to the characterization of Toll as an essential element of invertebrate immune armories, which prompted the search for and characterization of Toll homologs in humans [3],[4]. The identification of the mammalian Toll-like Receptor (TLR) family revolutionized the study of innate immunity in humans and continues to have a profound impact on our understanding of the complexities of vertebrate responses to infectious microbes.
Characterization of a mutation in the immune deficiency (imd) gene uncovered a distinct immune response to gram-negative bacterial infections in Drosophila [5]. Imd is a death-domain containing protein with similarity to the Receptor Interacting Protein (RIP) of the mammalian Tumor Necrosis Factor (TNF) pathway [6]. Drosophila immunity to gram-negative bacteria requires an intact IMD signaling pathway, which shares many other similarities with the TNF pathway. Engagement of the IMD pathway requires recognition of diaminopimelic acid-containing peptidoglycan (PGN) by the PGN Receptor Protein (PGRP-LC) [7],[8],[9],[10],[11]. PGRP-LC coordinately activates the Drosophila c-Jun N-terminal Kinase (dJNK) and the NF-κB transcription factor family member Relish (Rel). The Rel arm of the IMD pathway is well characterized thanks to a number of individual studies and complementary genetic and cell culture RNA interference (RNAi) screens. Essentially, Rel activation requires the activities of Imd, the caspase-8 ortholog Dredd, dFADD, dTAB2, dIAP2 and the MAP3 kinase dTAK1 [12],[13],[14],[15],[16],[17],[18],[19],[20],[21].
Active dTAK1 drives the subsequent activation of the I-Kappa Kinase (IKK) components Kenny (Key) and Ird5 [22],[23],[24],[25]. Rel is a p105 ortholog with an N-terminal Rel domain and a C-terminal ankyrin repeat domain [26],[27]. While the exact mechanism of Rel activation requires clarification, a recent report identified two distinct aspects to the generation of an active Rel [28]. Signal transduction through the IMD pathway results in the endoproteolytic cleavage of Rel of the N-terminal Rel domain from the inhibitory ankrin repeat domain. At the same time, activation of IKK activation drives the phosphorylation and transcriptional activation of Rel. The Rel domain translocates to the nucleus and initiates the transcription of a large number of genes, such as the antimicrobial peptides (AMPs) attacin (att) and diptericin (dipt).
IMD pathway activation of dTAK1 also stimulates a kinase cascade through the MAP2Ks dMAPKK4/7 that leads to dJNK phosphorylation [29],[30]. Phosphorylated dJNK typically activates the nuclear translocation of the AP-1 transcription factor subunits dJun and dFos, which initiate the transcription of dJNK depended gene products [31]. dJNK activation is a transitory event in the IMD pathway [30]. pJNK protein levels are downregulated through the combined activities of Rel and dJNK-responsive transcripts such as the phosphatase Puckered [30],[32],[33]. Mutations in djnk are lethal due to defective epithelial sheet sealing in the dorsolateral axis of the developing embryo [34],[35]. The developmental requirement for dJNK and other components of the dJNK arm of the IMD pathway has hampered the study of dJNK signaling events in innate immune signaling. Thus, the processes that regulate dJNK phosphorylation in the IMD pathway are poorly understood and many of the mechanisms that regulate dJNK signaling remain unknown.
Drosophila tissue culture cells provide an ideal environment to study these events, as PGN-induced activation of the IMD pathway induces a transient dJNK activation that is easily quantified. To understand the regulation of PGN-induced dJNK phosphorylation in the IMD pathway, we performed a high-throughput, quantitative RNAi screen for modulators of dJNK phosphorylation. To this end, we treated the embryonic macrophage-like S2 cell line with 15,683 individual dsRNAs that cover all annotated genes in the Drosophila genome. In contrast to previous RNAi screens of the IMD pathway, our assay did not rely on indirect reporter constructs. Instead, we used phospho-JNK specific monoclonal antibodies in a quantitative plate-based assay to directly quantify the impact of each dsRNA on the extent of PGN-induced dJNK phosphorylation. In this manner, we identified enhancers and suppressors of dJNK activation. As a testament to the accuracy of this screen, we unambiguously identified fifteen established IMD pathway components as modifiers of dJNK activation. In addition, we identified numerous novel regulators of dJNK activation. Given the involvement of dJNK in cellular events as diverse as development, cell migration, immune signaling and cell death, we believe that many of the regulators identified in this screen will be of broad interest to the study of metazoan cell biology.
We present a comprehensive analysis of a novel regulator of dJNK in IMD pathway signaling – the PDGFR and VEGFR receptor (Pvr) tyrosine kinase. Pvr is primarily known for its role in the guidance of cellular movements [36],[37],. We uncover a novel inhibitory circuit in the IMD pathway, where dJNK drives the expression of the Pvr ligands, Pvf2 and Pvf3, which subsequently contribute to the downregulation of dJNK activity via a Pvr/dERK signal transduction cassette. We also demonstrate that Pvr attenuates the expression of Rel-responsive transcripts by regulating the extent of Rel phosphorylation. We confirm a regulatory role for Pvr in the IMD pathway with data that loss of Pvr in adult Drosophila enhances the infection-induced expression of att. These data indicate that the Pvr/dERK signal transduction pathway constitutes a novel negative regulator of the Drosophila IMD pathway.
Engagement of the IMD pathway leads to transient dJNK phosphorylation; PGN-induced dJNK phosphorylation peaks at 5min and returns to basal levels by 60min in S2 cells (e.g. Figure 1A, B). We developed a quantitative high-throughput dsRNA screen to identify novel regulators of dJNK signaling in the IMD pathway (Figure 1C). To this end, we treated Drosophila S2 cells with a library of 15,683 dsRNAs that cover all annotated genes in the Drosophila melanogaster genome and we monitored the subsequent extent of PGN-induced dJNK phosphorylation by In Cell Western (ICW) analysis. We used monoclonal antibodies specific for phosphorylated JNK (P-JNK) and fluorescently labeled secondary antibodies to directly visualize PGN-induced dJNK phosphorylation. We simultaneously monitored filamentous actin (f-actin) levels with fluorescently labeled phalloidin as a control measure of total cell numbers. We then quantified the ratio of P-JNK:f-actin for each well to determine the relative extent of dJNK phosphorylation in each sample. To identify genes that modulate the intensity and duration of dJNK phosphorylation, we screened the entire genome at fifteen and sixty minutes. We reasoned that depletion of gene products that are required for optimal PGN-induced dJNK phosphorylation will decrease dJNK phosphorylation at fifteen minutes and we defined such gene products as enhancers of dJNK phosphorylation. Likewise, we reasoned that depletion of gene products involved in dJNK dephosphorylation will increase the relative intensity and/or duration of dJNK phosphorylation at fifteen and/or sixty minutes and we defined such gene products as suppressors of dJNK phosphorylation.
A representative 96-well plate from the screen is shown in Figure 1D and the corresponding quantification of the P-JNK:f-actin levels are shown in Figure 1E. Consistent with a previous report [43], we identified Dredd as an enhancer of dJNK phosphorylation. In addition, we identified the dJNK signaling pathway element Cka as a suppressor of PGN-induced dJNK phosphorylation. As expected, we identified Act79B as a regulator of f-actin levels and the gene product Clk as essential for S2 cell viability. To eliminate dsRNAs that negatively affected cell viability or cell adherence, we excluded dsRNAs that greatly reduced cell numbers as determined by an absence of f-actin and P-JNK fluorescence from subsequent analyses. We then calculated the P-JNK:f-actin z-score for all remaining wells to determine the statistical significance of dsRNA-treatment on PGN-induced dJNK phosphorylation and to allow for inter-plate comparisons. By these criteria, we successfully identified Cka and Dredd as statistically significant modifiers of dJNK phosphorylation with z-scores of 7.70 and -3.48, respectively (Figure 1F). These data indicate that the ICW assay is an effective method to detect modifiers of PGN-induced dJNK phosphorylation in S2 cells.
We then measured the PGN-induced P-JNK:f-actin levels and determined the z-score for all non-lethal dsRNA treatments. We graphed all the z-scores from highest to lowest for both fifteen and sixty minutes PGN-exposures (Figure 2A,B). dsRNA-mediated depletion of enhancers or suppressors of PGN-dependent dJNK phosphorylation resulted in reduced or elevated P-JNK z-scores, respectively. The z-scores for all dsRNAs are available in Table S1. We disregarded the P-JNK enhancers at sixty minutes PGN-exposures because the level of PGN-induced dJNK phosphorylation was not sufficiently elevated over background P-JNK levels. We identified Key as the strongest suppressor of dJNK phosphorylation at both fifteen and sixty minutes with z-scores of 9.05 and 9.23, respectively. Conversely, we identified dTAK1 as the strongest enhancer of dJNK phosphorylation at fifteen minutes PGN-exposure with a z-score of −5.7. As the Key/Rel axis of the IMD pathway attenuates dJNK activation and dTAK1 is essential for dJNK phosphorylation, these data are consistent with the known roles of Key and dTAK1 in the IMD pathway. We grouped all suppressors of dJNK phosphorylation with z-scores above 2.58 at fifteen and sixty minutes and all enhancers of dJNK phosphorylation with z-scores below −2.58 at fifteen minutes according to their known biological functions (Figure 2C). We identified many genes involved in innate immune signaling, in addition to a large number of genes with previously uncharacterized functions (Tables S2, S3, S4). As a testament to the saturation of this screen, we identified fifteen IMD pathway components as modulators of PGN-induced dJNK phosphorylation with z-scores above 1.96 or below −1.96 (Figure 2D). We note that in each case the z-score is consistent with the established role of the fifteen genes as either suppressors or enhancers of dJNK phosphorylation.
To test the validity of the dsRNA screen, we selected a representative cohort of three enhancers and eight suppressors of PGN-induced dJNK phosphorylation for secondary analysis. We monitored the effect of dsRNA treatment for all genes in the cohort on dJNK phosphorylation relative to f-actin at zero, fifteen and sixty minutes PGN-exposure. We compared the eleven putative modifier dsRNAs to two dsRNAs (CG11318 and Toll) that had no effect on dJNK phosphorylation in the primary screen. Secondary dsRNA analysis was consistent with the screen results as nine of the eleven dsRNAs significantly modified dJNK phosphorylation relative to f-actin compared to control dsRNA (Figure 3A). Even though we excluded actin modifiers from our primary data analysis, we considered the possibility that a fraction of the phenotypes observed may be indirectly caused by effects on f-actin, as opposed to dJNK phosphorylation. To test this hypothesis, we depleted each gene in the cohort and monitored PGN-induced dJNK phosphorylation relative to total dJNK by ICW (Figure 3B). We observed that the P-JNK:JNK analysis essentially mirrored the P-JNK:f-actin analysis for each gene in the cohort. Thus, we have confidence that our screen primarily identified regulators of PGN-dependent dJNK phosphorylation. To map relationships between the identified modulators of PGN-induced dJNK phosphorylation, we mined known genetic and physical interaction databases to develop an interaction network for all hits in our primary screen. We restricted the interaction network to direct physical or genetic interactions between genes identified as modifiers of dJNK phosphorylation. Within this direct interaction network we identified a branch with a high density of interactions that spanned the IMD and the dJNK signaling pathways (Figure 3C). The Drosophila PDGF/VEGF Receptor (Pvr) homolog appeared as a major node within this branch.
To confirm Pvr as a suppressor of dJNK phosphorylation in the IMD pathway, we depleted S2 cells of Pvr with two independent non-overlapping dsRNAs and monitored relative dJNK phosphorylation upon exposure to PGN at zero, fifteen and sixty minutes. We confirmed that both dsRNAs deplete Pvr by monitoring Pvr protein levels relative to actin in S2 cell lysates using Pvr specific antibodies (Figure 3D). Treatment of S2 cells with Pvr dsRNA 1 or 2 reduced relative Pvr protein levels to 1.6% and 15.6% of the control, respectively. In addition, depletion of Pvr by either dsRNA significantly increased PGN-induced dJNK phosphorylation at fifteen minutes (Figure 3E). Thus, we conclude that Pvr suppresses PGN-dependent dJNK phosphorylation.
While a previous dsRNA screen hinted at a role for Pvr in the IMD pathway [14], Pvr is primarily known for its role in Drosophila ERK signaling and cell migration. To investigate the involvement of the Pvr pathway in attenuation of dJNK activation, we determined the dJNK:f-actin z-score for each member of the Pvr/dERK axis in the primary screen. As a comparison, we also determined the dJNK:f-actin z-scores for members of the wingless pathway – a signal transduction pathway with no know interaction with the IMD/dJNK module. As expected, our data do not indicate any major interactions between the wingless and IMD/dJNK pathways. In contrast, our data consistently indicate that the Pvr/dERK pathway negatively regulates dJNK activation (Figure 4A). Ablation of the Pvr ligands Pvf2 and Pvf3; Pvr; established dERK adaptors; Ras and dERK resulted in considerably increased PGN-mediated dJNK phosphorylation.
We then asked if IMD pathway activation results in expression of Pvr ligands. Treatment of S2 cells with PGN resulted in a minor decline in the expression of pvf1 and significant increases in the levels of pvf2 and pvf3 expression (Figure 4B). Induction of pvf2 and pvf3 reached maximal levels within one hour of PGN treatment and reverted to basal levels by six hours. These expression patterns are reminiscent of other IMD/dJNK-responsive transcripts. To confirm that pvf2 and pvf3 are dJNK-responsive transcripts, we pre-incubated S2 cells with the dJNK inhibitor SP600125 and monitored the subsequent levels of pvf2 and pvf3 expression in response to PGN. Our data showed that SP600125 completely blocked the PGN-dependent expression of pvf2 and pvf3 (Figure 4C). Likewise, we observed a significant reduction in PGN-dependent pvf2 induction in cells depleted of PGRP-LC (Figure 4D) or dMKK4/dMKK7 (Figure 4E), confirming a requirement for the IMD/dJNK cassette in pvf2 induction by PGN. In summary, these data show that activation of the IMD pathway results in the dJNK-dependent expression of the Pvr ligands Pvf2 and Pvf3 and that the Pvr/dERK pathway attenuates dJNK activation.
Given that Pvr suppresses dJNK signaling in the IMD pathway, we asked if Pvr also modulates Rel signaling events. To determine if Pvr depletion affects Rel signaling in the IMD pathway, we depleted S2 cells of Pvr with two independent non-overlapping Pvr dsRNAs and monitored PGN-induced AMP expression. Specifically, we monitored expression of the Rel-responsive AMPs dipt and att. Depletion of Pvr by either dsRNA profoundly strengthened PGN-induced expression of att and dipt in comparison to control S2 cells (Figure 5A, B). Additionally, Pvr depletion significantly increased the basal expression levels of both att and dipt, in the absence of PGN stimulation. In fact, the basal levels of att or dipt expression in cells treated with Pvr dsRNA are approximately equal to the PGN-induced expression levels in cells treated with GFP control dsRNA. These data show that loss of Pvr in S2 cells results in an increase in both the uninduced and the PGN-induced expression of AMPs.
To confirm that the increased AMP expression observed upon Pvr loss proceeds through Rel, we then examined the expression of att in S2 cells that were simultaneously treated with Pvr and Rel dsRNA. As expected, depletion of Pvr increased the PGN-mediated expression of att (Figure 5C). In contrast, PGN-mediated expression of att was greatly reduced in cells treated with a combination of Rel and Pvr dsRNA. Thus, our data indicate that the bulk of Pvr RNAi-dependent increases in att expression proceed through the IMD/Rel module. In agreement with a role for the Pvr pathway in reducing att expression, we also observed increased att induction in cells treated with Ras85D dsRNA (Figure 5D). As Pvr loss leads to enhanced Rel-mediated AMP expression, we then asked if Pvr affects Rel cleavage or Rel phosphorylation. Whereas depletion of Pvr greatly sensitized S2 cells to PGN-dependent induction of dJNK phosphorylation (e.g. compare lanes 5 and 11, Figure 5E), we did not detect alterations in the pattern of PGN-induced Rel cleavage in S2 cells treated with Pvr dsRNA (Figure 5E). In contrast, we consistently detected prolonged and increased PGN-responsive phosphorylation Rel (P-Rel) in S2 cells treated with Pvr dsRNA (Figure 5F). These data indicate that Pvr negatively regulates the PGN-induced phosphorylation of both dJNK and Rel in the IMD pathway.
Given our findings that Pvr depletion increases AMP expression, we asked if activation of Pvr suppresses the IMD pathway. We monitored dERK phosphorylation to visualize Pvr signaling, as Pvr engagement results in activation of dERK in S2 cells. Previous reports demonstrated that Pvr ligands in conditioned medium (CM) from the Drosophila KC167 cell line activates Pvr signaling in S2 cells [44]. Likewise, we observed a requirement for Pvr in KC167 CM-induced dERK phosphorylation in S2 cells (Figure 6A). Quantification of relative dERK phosphorylation levels showed that Pvr dsRNA treatment decreased CM-induced dERK phosphorylation 21 fold (Figure 6B). To examine the effect of Pvr signaling on AMP expression, we treated S2 cells with GFP or Pvr dsRNA and monitored PGN-induced att expression levels 6h after exposure to CM (Figure 6C). Consistent with the role of Pvr as a suppressor of Rel signaling, we found that CM significantly decreased PGN-induced att expression. The phenotype is not an indirect effect of CM on PGN or other aspects of the IMD pathway, as dsRNA-mediated depletion of Pvr from S2 cells abrogated the suppressive effects of CM on att expression (Figure 6C). Thus, we conclude that activation of Pvr blocks PGN-responsiveness in S2 cells.
As Pvr signaling often proceeds through dERK and the bulk of the Ras/dERK pathway yielded Pvr-like phenotypes in our primary screen, we then tested if dERK phosphorylation is required for CM suppression of PGN-induced att expression. Treatment of S2 cells with the MEK1 inhibitor PD98059 decreased CM-induced dERK phosphorylation 3.2 fold relative to S2 cells treated with CM alone (Figure 6D, E). To test the effect of dERK inhibition on CM-mediated suppression of att expression, we pretreated S2 cells with PD98059 prior to exposure to PGN and CM (Figure 6F). CM suppressed the PGN-induced expression of att by 7.7 fold. However, we detected significant restoration of PGN-induced att expression in S2 cells treated with CM and PD98059. These data indicate that signal transduction through a Pvr/dERK axis attenuates activation of the IMD pathway.
We then asked if Pvr suppresses IMD pathway activity in vivo. To reduce Pvr activity in whole animals, we expressed Pvr dsRNA hairpin constructs (Pvr-IR) in adult flies. We then compared the immune response of infected wild type flies to flies that express Pvr-IR. Specifically, we monitored the expression of the Rel-responsive transcript att in uninfected flies (control) and flies that were pricked with a needle coated in E. coli (infection). Strikingly, we noticed that in vivo depletion of Pvr significantly enhanced infection-mediated att expression in three separate experiments in two separate Pvr-IR fly lines (Figure 7). These data indicate that depletion of Pvr from adult flies results in increased IMD pathway activity in vivo and support a role for Pvr as a negative regulator of Imd pathway activity.
Signal transduction through the JNK family of MAP kinases is a central element of vertebrate and invertebrate innate immune responses to infectious microbes. In addition, JNK activation contributes to the regulation of essential cellular processes, such as differentiation, apoptosis and directed cell movements [45],[46],[47]. The pleiotropic developmental and homeostatic requirements for JNK activity combined with functional redundancies among JNK pathway member isoforms hampered large-scale evaluations of JNK in model systems. In this study, we present the first whole-genome RNAi screen for modifiers of JNK activation to be performed in any metazoan. We specifically addressed the regulation of JNK activation in the context of innate immunity. We believe that Drosophila S2 cells present an ideal system for the study of the JNK signal transduction pathway, as S2 cells are readily accessible to large-scale RNAi screens, reproduce key elements of the Drosophila innate immune response and serve as a convenient gateway for whole animal studies in the genetically tractable Drosophila melanogaster. Given the evolutionary conservation of the JNK signal transduction pathway, we believe that our studies are of direct relevance to JNK activity in the immune response of higher organisms.
We also consider it likely that we have serendipitously identified general regulators of the JNK pathway with roles that extend beyond immune signaling. For example, we identified core elements of the JNK activation cassette such as misshapen (msn, M4K ortholog), hemipterous (hep, MKK4 orthologs) and dMKK7 (MKK7 ortholog) as required for activation of dJNK in the IMD pathway. A recent RNAi-based survey of four hundred eighty two Drosophila genes identified seventy seven core JNK pathway regulators [48]. Specifically, the authors detected gene products that modified basal dJNK phosphorylation levels in a number of genetically compromised backgrounds. In our assay, we excluded six of these JNK modifiers from analysis as they caused a significant depletion of f-actin. Of the remaining seventy one gene products, twenty three were significant modifiers of PGN-mediated dJNK phosphorylation (Figure S1). Thus, despite the large differences between both screens, we noticed a considerable overlap in our identification of dJNK modifiers.
We consider the false negative rate for IMD pathway members a more pertinent measure of the success of our screen. In contrast to previous RNAi screens of signal transduction pathways, our assay did not rely on indirect reporter assays. Instead, we measured the contribution of each annotated gene within the fly genome to the IMD-responsive phosphorylation of dJNK. We believe that the direct quantitative nature of our assay combined with the ease of RNAi in S2 cells greatly minimizes the likelihood of false negatives in the primary screen. Indeed, preliminary analysis of our primary screen data identified the bulk of the IMD signal transduction pathway (PGRP-LC, Imd, dFADD, Dredd, Pirk, dTAB2, dIAP2, dTAK1, dMKK4/7, dJNK, dFos, Key, Ird5 and Rel) as essential modifiers of JNK activation in the IMD pathway. In each case, the phenotype was consistent with the established molecular function of the respective IMD pathway element as either negative or positive modifiers of JNK activation. Thus, we are satisfied that false negatives do not obfuscate interpretation of our data in any meaningful manner. Ironically, the only anticipated hit we failed to identify was dJun [49].
The Drosophila receptor tyrosine kinase Pvr shows considerable similarity to members of the mammalian PDGF and VEGF receptor families and Pvr is considered an evolutionary ancestor of PDGF/VEGF receptors [38]. Pvr is activated in a partially redundant manner by three PDGF/VEGF-type ligands, Pvf1-3 [37],[38],[42],[50]. Initial studies implicated Pvr as a guidance receptor for cell migratory cues in embryonic hemocyte migration, oocyte border cell migration, thorax closure and dorsal closure of male terminalia [36],[37],[38],[39],[42]. The molecular basis for Pvr-mediated cell movements requires clarification. While functional redundancies appear to exist between individual Pvf ligands, several studies indicate a potential preference for Pvf-1 in the guidance of cell migration [40],[42]. In thorax closure and border cell migration, migratory cues proceed through the Pvr adaptor proteins Mbc, Ced-12 and Crk [36],[39]. In the case of thorax closure and dorsal closure of male genitalia it appears that Pvr induces the corresponding cell movements through the JNK pathway. Thus, Pvr appears to be a positive regulator of JNK activity in the context of cell movements. This is logical given the extensive involvement of JNK in the coordination of cell migration during development. However, our data strongly indicate that Pvr is a negative regulator of JNK activity during immune signaling. We did not detect any requirements for Mbc, Ced-12 or Ckr in the regulation of innate immune signaling. These data suggest that distinct adaptor molecule configurations may discriminate between the impacts of Pvr on immune responses and cell migration.
In addition to requirements for Pvr in cell migration, a parallel body of literature indicates a distinct function for Pvr in the regulation of hemocyte proliferation. The disruptions to embryonic hemocyte migration in pvr mutants were originally interpreted to indicate that Pvr detects migratory guidance cues in hemocytes [37]. More recent studies demonstrated that expression of the anti-apoptotic p35 molecule in the hemocytes of pvr mutants rescues the majority of the migratory phenoptye [51]. Further studies confirmed that the bulk of the pvr hemocyte phenotype is the result of cell death and that there are only minor guidance requirements for Pvr in hemocyte migration. Pvr activates the dERK pathway, which induces hemocyte proliferation [51],[52]. Consistent with a role for Pvr in hemocyte proliferation, overexpression of Pvf2 drives massive hemocyte proliferation in vivo and incubation of embryonic mbn-2 hemocytes with Pvr antibodies blocks cellular proliferation in a dose-dependent manner [50]. In contrast, overexpression of Pvf-1 did not substantially alter hemocyte proliferation in vivo and a recent study indicated that proliferative signals for hemocytes are preferentially provided by Pvf2 and Pvf3 [52]. In this context, we consider it particularly striking that our data reveal that signal transduction through the IMD pathway results in dJNK-mediated expression of Pvf2 and Pvf3.
Our study reveals a novel role for the Pvr/dERK pathway in the attenuation of the IMD pathway and illuminates our understanding of the network of regulatory checks and balances that fine tune the level of IMD/dJNK activity. Our data are most consistent with a model whereby activation of the IMD pathway results in dJNK-dependent expression of the Pvr ligands Pvf2 and Pvf3. Pvr then signals through dERK to negatively regulate the IMD pathway. On a molecular level, our data show that Pvr signaling dampens the dTAK1-dependent phosphorylation of dJNK and Rel. However, we believe that our data may also uncover an additional physiological role for Pvr. We speculate that the infection-driven production of Pvf2 and Pvf3 engages Pvr receptors on hemocytes and thereby stimulates the Ras/dERK-responsive proliferation of hemocytes. Such an increase in hemocytes numbers would provide a timely measure for the phagocytic elimination of invading extracellular microbes at early stages of infection.
We find it intriguing that proliferative signals inhibit activation of immune pathways. It may be that both processes require major metabolic commitments and that hemocytes preferentially reserve resources for proliferation. An alternative and non-exclusive hypothesis reflects the primary role of Drosophila hemocytes in immunity. Hemocytes are the major phagocytic cell type in Drosophila and are ideally suited for the engulfment of extracellular microbes. We consider it possible that induction of immune responses drives Pvr-mediated proliferation of hemocytes to facilitate rapid neutralization of extracellular microbes through phagocytosis. In this situation, it is advantageous for proliferative signals to suppress JNK activation, as hyper or prolonged activation of JNK in Drosophila often results in cell death. Preliminary data in our lab suggest that links between Pvr and immune signaling may be evolutionarily conserved, as we detected suppression of NF-κB activity through the PDGF receptor superfamily member c-Kit in human cell culture assays (Anja Schindler and Edan Foley, unpublished).
Drosophila S2 cells and KC167 cells were cultured at 25°C in HyQ TNM-FH medium (HyClone) supplemented with 10% heat inactivated fetal bovine serum (Invitrogen), 50U/ml of penicillin and 5 µg/ml of streptomycin (GIBCO). Serum-free S2 cells were incubated in SFX-INSECT medium (HyClone) supplemeted with 50U/ml of penicillin and 5 µg/ml of streptomycin (GIBCO). PGN-dependent dJNK activation was inhibited in 106 S2 cells in 1ml of culture media with the addition of 25µM SP600125 for 1h prior to PGN-exposure.
The dsRNA library employed in this screen is an extension of a partial-genome library described previously [53]. The remainder of the library was purchased from Open Biosystems. In-Cell Western quantitative analysis was carried out as described in [54]. Briefly, S2 cells were incubated at 1.5×105 cells/well in 96 well plate in 20% conditioned media and 80% serum-free culture media with 10µg/ml dsRNAs at 25°C for three days. Cells were exposed to 50µg/ml LPS (Sigma) containing contaminating amounts of PGN for 15 or 60 min. Cell were washed with PBS, fixed in PBS + 3.7% formaldehyde, permeablized in PBS + 0.1% Triton-X 100 and blocked in blocking buffer (LI-COR Biosciences). Cells were probed with mouse anti-active-JNK (Cell Signaling) and washed with PBS + 0.1% Tween-20. P-JNK staining was detected with fluorescently labeled goat anti-mouse secondary antibodies and f-actin was stained with fluorescently labeled phalloidin (Invitrogen). Cells were washed in PBS + 0.1% Tween and P-JNK and f-actin levels were quantified with an Aerius automated imaging system (LI-COR Biosciences) following the manufacturers recommendations. In secondary ICW analyses P-JNK was monitored relative to JNK by replacing phalloidin staining with rabbit anti-JNK (Santa Cruz Biotechnology) and fluorescently labeled goat anti-rabbit antibodies.
For the RNAi screen, the raw fluorescent trimmed mean level was determined for P-JNK and f-actin channels in each well and the relative P-JNK:f-actin value was calculated. We applied z-score analysis to normalize P-JNK:f-actin values across the entire screen. Z-scores were calculated by subtracting the sample value by the plate median value and dividing by the plate standard deviation. The z-score assumes normal distribution and represents the standard deviation of every P-JNK:f-actin value from the plate median for each dsRNA treatment. Z-scores above 2.58 or below −2.58 represent the 99% confidence interval and z-scores above 1.96 or below −1.96 represent the 95% confidence interval. The f-actin z-scores were also calculated for every well on each plate and dsRNA treatments resulting in f-actin z-scores below −2.58 (99% CI) were excluded from further analysis to eliminate actin modifiers and lethal dsRNAs. We considered dsRNAs that modified P-JNK:actin z-scores outside the 95% confidence interval as hits in the screen. To identify genetic or physical interactions among hits from our screen, all hits were probed in the Drosophila interactions database [55] and visualized with the IM browser (http://www.droidb.org/IMBrowser.jsp). For analysis of att expression in the infection model, the ΔCt values were standardized to an internal control between qRt-PCR runs. The triplicate 0h ΔCt values were averaged and the ΔΔCt values were calculated relative to these values. The fold change was calculated for each sample and the 0h time point was set to one for each fly line. The SEM was calculated for each time point. Statistical significance of experimental values was expressed as p-values of less than .01 (**) or .05 (*), as calculated by a Student's t-test. We performed two-tailed Student's t-tests with two-samples of equal variance to calculate a p-value of experimental values relative to control values.
Western blot analysis was performed on 106 cells lysed in sample buffer, vortexted and incubated at 95°C for 5min. Proteins were separated by SDS-PAGE electrophoresis and were transfer to nitrocellulose membrane by semidry transfer. Membranes were blocked in blocking buffer (LI-COR Biosciences) and probed with mouse anti-active-JNK (Cell Signailing), rabbit anti-JNK (Santa Cruz Biotechnology), rabbit anti-pan-actin (Cell Signaling), mouse anti-actin (Sigma), rabbit anti-active MAPK1/2 (Upstate), mouse anti-HA (Sigma) or rat anti-Pvr. Western blot analysis of P-Rel and Rel cleveage was performed as described in [28]. All secondary antibodies were purchased from Invitrogen. Proteins levels were quantified with an Aerius automated imaging system (LI-COR Biosciences) following the manufacturers recommendations.
Antimicrobial peptide production was monitored in S2 cells and flies by qRT-PCR. Total RNA was extracted from 106 S2 cells or 10 adult flies using Trizol (Invitrogen) following the manufacturers instructions. cDNA was created from 2µg of RNA using Superscript III (Invitrogen) and oligo-dT primers (Invitrogen), according to the manufacturers instructions. We monitored transcript amplification with a Realplex 2 PCR machine (Eppendorf) using SYBR green as a detection reagent (Invitrogen). We used the following primers to monitor the expression of the corresponding gene products; actin forward 5′-TGCCTCATCGCCGACATAA-3′, actin reverse 5′-CACGTCACCAGGGCGTAAT-3′; att forward 5′-AGTCACAACTGGCGGAC-3′, att reverse 5′-TGTTGAATAAATTGGCATGG-3′; dipt forward 5′-ACCGCAGTACCCACTCAATC-3′, dipt reverse 5′-ACTTTCCAGCTCGGTTCTGA-3′; pvf1 forward 5′-GCGCAGCATCATGAAATCAACCG-3′, pvf1 reverse 5′-TGCACGCGGGCATATAGTAGTAG-3′; pvf2 forward 5′-TCAGCGACGAAACGTGCAAGAG-3′, pvf2 reverse 5′-TTTGAATGCGGCGTCGTTCC-3′; pvf3 forward 5′-AGCCAAATTTGTGCCGCCAAG-3′, pvf3 reverse 5′- CTGCGATGCTTACTGCTCTTCACG-3′. All transcript expression values were normalized to actin and were quantified relative to a control using the ΔΔCt method.
We depleted Pvr from S2 cells with two non-overlapping dsRNAs. We designed the following primers to amplify the associated dsRNA template DNA in a two step PCR using 5′-GGGCGGT-3′ as an anchor sequence; Pvr1 forward 5′-GGGCGGGTGATGACTACATGGAGATGAGCC-3′, Pvr1 reverse 5′-GGGCGGGTATACCTTCGTTGCTCCTTCTCG-3′; Pvr2 forward 5′-GGGCGGGTCTCCTGATTTTGCGGATCTC-3′, reverse 5′-GGGCGGGTGTCTTGGGATCGGTTCTTGA-3′; GFP forward 5′-GGGCGGGTACGTAAACGGCCACAAG-3′, GFP reverse 5′-GGGCGGGTCTCAGGTAGTGGTTGTC-3′. We performed a second PCR amplification of anchor-tagged template DNA with the T7 promoter containing primer 5′-TAATACGACTCACTATAGGGAGACCACGGGCGGGT-3′. dsRNA was amplified from template DNA using T7 RNA polymerase at 25°C for 6h and annealed by cooling from 90°C to 30°C. S2 cells were depleted of Pvr using Pvr1 dsRNA unless stated otherwise. 106 S2 cells were treated with dsRNA for 4 days in 1ml culture media to deplete Pvr. The Pvr pathway was activated in S2 cell using 1∶1 dilution of fresh culture media in conditioned media (CM) collected from 4 day cultures of KC167 cells. Pvr dependent dERK phosphorylation was inhibited in 106 S2 cells in 1ml of culture media with the addition of 50µM PD98059 for 1h prior to CM exposure.
Drosophila strains were cultured on standard cornmeal medium (http://flystocks.bio.indiana.edu/Fly_Work/media-recipes/bloomfood.htm) at 25°C. hs-gal4 flies were obtained from Dr. Sarah Hughes and uas-PvrIR flies were obtained from the Vienna Drosophila RNAi Center. For in vivo knock down of Pvr, UAS-PvrIR flies were crossed with hs-gal4 flies or w118 flies. 1 day old flies were heat-pulsed eight times at 37°C for 1h to initiate the expression of the RNAi construct and returned to 25°C for 5h over 48 hours. Infection was monitored in flies that were either uninjured (control), or pricked with a tungsten needle dipped in a pellet of DH5α E. coli bacteria (infection).
|
10.1371/journal.pcbi.1004360 | Optimal Prediction of Moving Sound Source Direction in the Owl | Capturing nature’s statistical structure in behavioral responses is at the core of the ability to function adaptively in the environment. Bayesian statistical inference describes how sensory and prior information can be combined optimally to guide behavior. An outstanding open question of how neural coding supports Bayesian inference includes how sensory cues are optimally integrated over time. Here we address what neural response properties allow a neural system to perform Bayesian prediction, i.e., predicting where a source will be in the near future given sensory information and prior assumptions. The work here shows that the population vector decoder will perform Bayesian prediction when the receptive fields of the neurons encode the target dynamics with shifting receptive fields. We test the model using the system that underlies sound localization in barn owls. Neurons in the owl’s midbrain show shifting receptive fields for moving sources that are consistent with the predictions of the model. We predict that neural populations can be specialized to represent the statistics of dynamic stimuli to allow for a vector read-out of Bayes-optimal predictions.
| Many behaviors require predictive movements. Predictive movements are especially important in prey capture where a predator must predict the future location of moving prey. How sensory information is transformed to motor commands for predictive behaviors is an important open question. Bayesian statistical inference provides a framework to define optimal prediction and Bayesian models of the brain have received experimental support. However, it remains unclear how neural systems can perform optimal prediction in time. Here we use a theoretical approach to specify how a population of neurons should respond to a moving stimulus to allow for a Bayesian prediction to be decoded from the neural responses. This provides a novel theoretical framework that predicts properties of neural responses that are observed in auditory and visual systems of multiple species.
| Predicting the future position of an object in the environment is a common and critical component of many tasks that involve reaching or orienting toward moving targets [1–4]. To execute these prediction tasks successfully, motor plans must extrapolate beyond accumulated sensory input to account for delays in sensory and motor processing, as well as for the future movements of the object. The ability to make accurate predictions of the location of a moving target is especially critical in prey capture.
Prey capture for moving targets has been studied at the behavioral and neural levels for animals that rely on visual [5–9] and auditory [10–12] information. For example, salamanders use visual input to predict the position of moving prey, make a head orienting movement toward the target, and then generate a ballistic movement of the tongue to capture the prey [7]. Barn owls also visually track their prey when possible [13], but are additionally able to use auditory information to capture moving prey [10]. After estimating a sound source’s trajectory, the owl makes a head orienting movement to localize a moving target before preparing to bring its feet forward to strike the prey [10]. Interestingly, salamanders and barn owls have neurons with similar specialized receptive fields that shift in time to mediate predictive prey capture [12,6,8]. These specializations occur in the fast-OFF retinal ganglion cells of the salamander [6,8] and the auditory spatially selective neurons in the optic tectum (OT) of the barn owl [12]. The receptive fields of these neurons shift toward a moving source, where the amount of shift is sufficient to account for delays in sensory and motor processing. Furthermore, it has been shown in the salamander that it is possible to read out the predicted location of a moving target from the fast-OFF retinal ganglion cells using a population vector average (PV) [8]. Here, we use the PV to model the computations performed by the barn owl as it tracks a moving sound source and address how such a neural circuit may approach optimal performance.
These studies open several questions about the neural basis of predictive behaviors. What information is represented in these populations of neurons? Is the observed neural representation an optimal solution to the prey capture problem faced by each species? An optimal solution to the prediction problem would take into account the source dynamics, sensory statistics, and prior information to guide the solution. This approach to an optimal solution can be formulated as Bayesian prediction [14]. There is support for Bayesian models of perception and behavior in diverse tasks across multiple species [15–18]. Additionally, there have been multiple proposals for how neural systems can implement Bayesian inference [19–23,16,24–26]. In particular, several studies have addressed the problem of inference in time in the context of hidden Markov models [20,27,28] and tracking using the Kalman filter [29,19,22,30,31]. However, it remains unknown how a neural system can perform Bayesian prediction.
Here we specify how a population of neurons should respond to a moving stimulus to allow for a Bayesian prediction to be decoded from the neural responses. We approach this question in the context of auditory-based prey capture by the barn owl. The Bayesian prediction problem we consider is that of predicting a sound source’s future direction, given a sequence of sensory observations and a prior distribution for direction and angular velocity. It has been shown that the owl’s sound localization for brief sounds is consistent with a Bayesian model [24]. Here, evolutionary pressure for optimality may be expected, given the dependence of owls on successful sound localization during hunting.
The success of the PV in decoding predictive movements of visual targets in the salamander [8] and dragonfly [32] makes this a viable candidate mechanism for implementing Bayesian prediction. It has been shown that the owl’s map of auditory space decoded by a PV is consistent with the owl’s localization behavior for brief stationary stimuli [24]. More generally, it has been shown that a population code can encode the statistical properties of the environment to allow a PV to match a Bayesian estimator [23,26,24]. This model for the neural implementation of Bayesian inference is attractive because it matches the common observation that population codes are adapted to natural statistics [33]. However, the applicability of the PV model to Bayesian inference in time is unknown. Here, we determine the conditions under which a population of neurons with spatial-temporal receptive fields can perform Bayesian prediction for moving sound sources.
We consider the problem of predicting the future direction of a moving source from a temporal sequence of auditory observations. Specifically, the prey capture problem is that of predicting the direction of a moving sound source a short time in the future based on the sequence of interaural time difference (ITD) measurements from the sounds reaching the left and right ears (Fig 1). ITD is the difference in the arrival time of sounds at the two ears and is a primary cue for localization in the horizontal dimension [34,35].
The Bayesian filtering approach to predicting at time k the direction at a point n time steps in the future θk+n, given the sequence of observations up to time k, ITD1:k = [ITD1, ITD2, … ITDk], is to compute an estimate from the posterior distribution pk+n(θ, ω|ITD1:k). The form of the posterior distribution is determined by a model for the dynamics of the moving target and the statistical relationship between the state of the target and the ITD observations.
The temporal dynamics of the horizontal direction of the moving target are modeled as
θk=θk−1+Δtωk−1+ηk
ωk=ωk−1+νk
where θk is the target direction, ωk is the angular velocity, ηk is a zero-mean circular Gaussian noise process, νk is a zero-mean Gaussian noise process, k is the current time step, and Δt is the time step duration. The sensory information ITD is modeled as a sinusoidal function of direction that is corrupted by noise:
ITDk=Asin(2πfθk)+ξk
where ξk is a zero-mean Gaussian noise process with standard deviation 12.5 μs and the amplitude A and frequency f are determined by the shape of the owl’s head and facial ruff [24]. The sinusoidal mapping between direction and ITD is based on direct measurements of ITD for the barn owl [36]. All noise processes are assumed to be mutually independent and uncorrelated across time.
The noise process νk influencing the prey velocity depends on the type of behavior displayed by the prey. A large standard deviation of the noise corresponds to irregular fleeing behavior displayed by prey under close attack when there is no place to hide [37]. A small standard deviation produces a smoother trajectory for the prey, which corresponds to escape toward cover [37]. Here we use a velocity-noise standard deviation of 0.125 deg/s corresponding to mouse escape behavior under close-distance owl attack where prey trajectories are smooth [37]. This parameter value has the effect of keeping the velocity roughly constant over a short period of time.
The prior depends on both the natural prey behavior and the owl’s bias as determined by the behavioral cost function [24]. Here we assume that the prior emphasizes directions at the center of gaze [24] and slow source velocities. We also assume that there is a weak negative correlation between direction and velocity such that there is a bias for sources moving into the center of gaze [38,39]. The form of the prior is a Gaussian with zero mean for both direction and velocity. The standard deviation for direction is 23.3 deg [24], the standard deviation for velocity is 50 deg/s, and the correlation between direction and velocity is -0.05. The parameter values for the velocity standard deviation in the prior (σv0=50 deg/s) and during movement (σvk=0.125 deg/s, k ≥ 1) describe a situation where the initial velocity can take on a wide range of values, but the velocity will be roughly constant over a short period of time.
The Bayesian prediction at time k of the direction at a point n steps in the future, θk+n given the sequence of observations ITD1:k is computed as the mean of the posterior n steps in the future pk+n(θ, ω|ITD1:k) Because we are estimating a circular variable, the Bayesian prediction is the direction of the Bayesian prediction vector, defined as the vector that points in the direction of the mean value of the direction n steps in the future:
BVk=∫u(θ)pk+n(θ|ITD1:k)dθ,
where u(θ) is a unit vector pointing in direction θ (Methods). Solving the Bayesian prediction equations may be computationally difficult for nonlinear or non-Gaussian models [40]. If the system is linear with Gaussian noise, then the Kalman filter can be used for Bayesian prediction [41]. Our model includes Gaussian noise but the mapping from direction to ITD is nonlinear. We found that relationship between direction and ITD is nearly linear for sound sources in the frontal hemisphere (Fig 2). The root-mean-square (RMS) error between the measured ITD and the linear approximation ITD = 2.67 μs / deg × θ was 15.1 μs for directions between -100 deg and 100 deg. We therefore used the Kalman filter to perform Bayesian prediction for computational simplicity (Methods).
The Bayesian model successfully predicts future directions of the prey for smoothly moving sources (Fig 3). We chose the prediction time step n in order to predict the source direction 100 ms in the future [12]. Initially the Bayesian prediction is dominated by the prior distribution, which emphasizes central directions (Fig 3A). Because of the influence of the prior, the posterior does not initially lead the source direction. However, after a short delay the posterior pk+n(θ, ω|ITD1:k) predicts the future direction of the source (Fig 3A and 3B). Note that the performance of the Bayesian prediction differs from the Bayesian tracking estimate. Whereas the tracking algorithm seeks to place the center of posterior at the current source position (Fig 3C), the prediction algorithm seeks to place the center of posterior at the future position of the source. Also, the predictive posterior (Fig 3A) is wider than the posterior for tracking (Fig 3C) because uncertainty increases as the time window for prediction increases beyond the current time where observations are available.
It has been shown that the owl’s map of auditory space decoded by a PV is consistent with the owl’s localization behavior for brief stationary sounds [24]. Here we investigate conditions on a population of neurons with spatial-temporal receptive fields under which the PV will match the Bayesian prediction in time. The PV at time k is given by an average of weighted preferred direction vectors:
PVk=1N∑j=1Na(ITD1:k|θ(j),ω(j))u(θ(j))
where the preferred directions θ(j) are defined by the motor output. The PV at time k depends on the sequence of past ITD measurements and predicts the future direction of the target. By associating each neuron with a fixed preferred direction θ(j), we are making the assumption that the motor neurons that the OT neurons ultimately influence are fixed. This assumption means that the effect of a given level of response for an OT neuron on the motor output stays constant. The rate function a(ITD1:k|θ(j), ω(j)) is the firing rate of the jth neuron in response to the sequence of ITD values ITD1:k. We now state our main result, which specifies sufficient conditions so that the PV will approximate the Bayesian prediction estimate.
The first prediction derived from our result is that neurons implementing Bayesian prediction using this type of population code will have receptive fields that shift in time towards the moving source (Fig 4A–4D). This is the type of shift that is necessary to compensate for delays and allow for the owl to capture the moving source [6,12,8]. These delays include signal processing in the brain as well as motor delays, and total approximately 100 ms [12]. While the receptive fields shift in time, there is a delay to the onset of the shift of the receptive field. This delay in the shift occurs in the Bayesian model because the response is initially dominated by the prior before sufficient sensory information has been accumulated. Therefore, the predictive posterior initially lags behind the source direction (Fig 3A). It is only after a delay that the predictive posterior leads the current source direction. The model also predicts that receptive fields get sharper with time. The sharpening of the receptive fields follows the sharpening of the posterior as more sensory information is collected (Fig 3A). Additionally, the model predicts that the shift of the receptive field depends on the speed of the moving source. Faster source velocities lead to larger shifts, while slower source velocities correspond to smaller shifts of the receptive field (Fig 5A). This prediction follows from the fact that the posterior shifts faster for faster sources.
The receptive field shifts predicted by the model are consistent with experimental results in the barn owl [12] (Figs 4 and 5). Neurons in the owl’s OT that are involved in generating head orienting movements show shifting receptive fields for moving sources [12] (Fig 4E and 4F). The receptive field shifts in the owl are consistent with the Bayesian prediction model in that the shift toward the source is not instantaneous, but occurs after a delay (Fig 4E and 4F). Receptive fields of midbrain neurons also get sharper in time, as predicted by the model [12,43]. Additionally, the size of the shift varies with the speed of the moving source (Fig 5B). The time course and magnitude of the observed shifts correspond well to the predicted shifts in the model.
The model predicts an asymmetry in the shifts of the receptive fields for sounds moving into and out of the center of gaze that increases with the eccentricity of the receptive field (Fig 6). For neurons with receptive fields at the center of gaze, the shifts for clockwise and counterclockwise sources are mirror images (Fig 6A–6C). For neurons with more peripheral receptive fields, the shifts for clockwise and counterclockwise moving sources are asymmetric (Fig 6D–6I). For neurons with peripheral receptive fields, the initial shift of the receptive field for sources moving into the center of gaze is in the opposite direction than one would expect based on the idea that receptive fields should move towards the source. This occurs because of the effect of the prior on the performance of the posterior (Fig 3A). Initially, the posterior is dominated by the prior and thus at stimulus onset is not leading the source by the desired 100 ms.
The asymmetry of the receptive field shifts for peripheral OT neurons has not been investigated in the owl. However, neurons in the owl’s external nucleus of the inferior colliculus (ICx) do have an asymmetry in their direction selectivity for sounds moving into and out of the center of gaze, which may be related to asymmetric shifts [38,39]. Testing this prediction will require further study.
The prediction of asymmetry in the receptive field shift for clockwise moving and counterclockwise moving sources depends on the presence of a prior that emphasizes central directions. We found that predicted receptive field shifts were symmetric for clockwise moving and counterclockwise moving sources in both central and peripheral neurons when the prior in the model was uniform (Fig 7). As noted above, the asymmetry is caused by the initial dominance of the prior on the location of the peak in the posterior. When the prior is uniform, this effect is removed and the posterior can quickly lead the source direction for motions both into and out of the center of gaze.
The receptive field shifts predicted by the model were robust to parameter variation (Fig 8). We examined the receptive field shifts for different standard deviations of the noise terms and different prior standard deviations for direction and velocity. The model predicted similar magnitudes of shifts for the chosen values (center column) and when each parameter was halved (left column) or doubled (right column). Changing the standard deviation of the noise corrupting ITD had the greatest effect on the receptive fields (Fig 8A–8C). This parameter influences the width of the posterior and therefore influences the width of the receptive field.
The net effect of the receptive field shifts is that the activity moves across the population so that it predicts the future direction of the moving source (Fig 9A). It is this activity that must be decoded by the PV to approximate the Bayesian prediction. To test the PV implementation of Bayesian prediction, we constructed a model of 5000 Poisson neurons with receptive fields that shift according to the posterior (Methods). The PV matched the Bayesian prediction closely for different stimulus conditions (Fig 9). The PV approximated the Bayesian prediction to within 3 degrees (root-mean-square (RMS) error) for velocities up to 125 deg/s (Fig 9B). The RMS error in the approximation of the Bayesian prediction by the PV depended strongly on the fraction of time the predicted source direction was in the frontal hemisphere (spearman rank correlation = 0.92; Fig 9C). Since all of the preferred directions of the model neurons are in the frontal hemisphere, the model will necessarily fail when the posterior is localized at source directions behind the head. We also computed the RMS error using a population of deterministic neurons to determine the contribution of the Poisson variability of the neurons to the error (Fig 9D). The Poisson variability increased the RMS error for many trajectories (mean ± s.d. ratio of RMS error for deterministic neurons to RMS error for Poisson neurons 0.43 ± 0.23). However, the largest errors in the approximation are primarily due to the limited range of preferred directions of neurons in the population. The pattern of error as the initial direction and velocity of a moving source varied is explained by larger errors occurring when the predicted source trajectory spends more time behind the head.
We showed that the PV can read out the Bayesian prediction in time from a population of neurons. The PV will approximate the Bayesian prediction when the population has specialized responses with shifting receptive fields. The types of shifting receptive fields predicted by our analysis are observed in the OT of the owl [12] and the retina of the rabbit [6] and salamander [6,8]. This result shows that with the appropriate encoding of the stimulus, a simple decoding algorithm can perform complex computations [44,19,8].
Our work provides a theoretical framework in which to interpret observations about circuits underlying prediction. Previous work identified neurons in the OT [12] and retina [6,8] with shifting receptive fields that account for delays in neural processing. Leonardo and Meister (2013) further showed that decoding a population of such responses with a PV can predict a moving target position. Our work shows that this type of network computation can be optimal and capture the statistics of a dynamic target.
This work shows that a non-uniform population code model with a PV decoder can implement Bayesian inference for stationary and moving sources. The non-uniform population code model proposes that a prior distribution is encoded in the distribution of preferred stimuli and that the statistics of the sensory input are encoded by the pattern of neural responses across the population [23,24]. Here we extend this model to show that the dynamics of a population code can represent the statistics of a dynamical system. This is an important extension of the non-uniform population code model due to the dynamic nature of ethologically relevant stimuli.
We make several predictions about the receptive field shifts necessary for optimal prediction. First, we predict that neurons have receptive fields that shift towards a moving source where the shift increases with the source velocity. This prediction is consistent with observations in the OT [12] and retina [6,8]. We also predict that the shift is sluggish when a non-uniform prior is present. This is consistent with responses of OT neurons [12]. Our analysis also leads to several predictions that have not been tested in the auditory or visual systems. In particular, we predict an asymmetry in the shifts of receptive fields for sources moving into and out of the center of gaze when a prior emphasizes the center of gaze (Fig 6). We also predict that for noisier stimuli, the magnitude of the shift will decrease and the receptive fields will become wider (Fig 8A–8C). Finally, we predict that receptive fields should become narrower over time to reflect the accumulation of sensory information. Studies of neurons thought to support predictive behaviors have not yet investigated all of these response features predicted by our model.
Bayesian theories of perception propose that neural systems represent statistical models of the environment, where the models may contain many parameters. The parameters of these models may be learned by an animal over multiple time scales. For the owl, information about the prior and the basic relationship between sound localization cues and source directions is primarily due to a combination of genetic changes over an evolutionary time scale and learning over the life of the animal [45]. There is evidence, however, that the owl adjusts to the noise level of the stimulus on a trial-to-trial basis [46]. We therefore predict that the noise-level parameter of the model is learned rapidly, leading to wider and more slowly shifting receptive fields in high noise environments. Future work is required to determine how the parameters of the model are learned in the owl’s auditory system.
Previous studies have shown that a cascade model with a gain control component can produce the experimentally observed shifting receptive fields [6,12,8]. This model involves a negative feedback loop, causing the neural response at each time step to be influenced by its predecessors. This model is phenomenological, but it suggests that a recurrent network within the OT is sufficient to generate the receptive field shifts necessary for Bayesian prediction. However, neurons upstream from OT in ICx show direction selectivity [39,47] and it is therefore possible that shifting receptive fields originate in ICx. Furthermore, the asymmetric direction selectivity observed in ICx may possibly be explained by single-cell adaptation [39] rather than by a network effect. Therefore, the mechanism underlying receptive field shifts in OT remains an open question.
Previous work has addressed inference in time using the Kalman filter [19,22,30,48]. While we determine how a population of neurons should respond to a moving stimulus but did not specify a mechanism for implementing the responses, these studies constructed networks to represent the Kalman filter estimate and variance as a function of time. One type of model produces a population code where the estimate of the target location is at the peak of a symmetric population response [22,30]. This is accomplished through a nonlinear encoding model involving divisive normalization. It is possible to read out the estimate using a center-of-mass decoder, but the model is limited to Gaussian distributions. Another model encodes the target estimate and variance using a linear probabilistic population code [48]. This model also relies on divisive normalization to implement the Kalman filter, but requires a nonlinear decoder to determine the estimated location from the activities. The model of Eliasmith and Anderson (2003) utilizes nonlinear responses and linear decoding. However, unlike the preferred direction vectors in the PV, the linear decoders are not in general equal to the preferred directions and are obtained using supervised learning. These models may be extended to consider the case of prediction, but the responses of neurons performing prediction in these schemes has not been investigated. Our model differs from the previous models in that the preferred direction at the peak of the population activity profile will not in general equal the PV estimate (Fig 9). This occurs because our model includes a non-uniform population, whereas previous models use a uniform population. An additional distinction between our model and previous models is that our predictions apply to nonlinear and non-Gaussian models.
It has previously been shown that the PV performs poorly when decoding arm movements from motor cortical responses [49]. The work presented here does not conflict with this previous finding. We show that the PV will perform well in tracking and prediction when the receptive fields of the neurons encode the state dynamics with shifting receptive fields. This is not a general-purpose decoder, but rather must be used to read out the activity of a specialized population with shifting receptive fields such as those in the OT. Experimental evidence suggests that populations of neurons with response properties that are adapted to the natural statistics are important for perception and behavior. The work presented here shows how network properties tailored to the dynamics of moving prey allow for optimal Bayesian prediction by a population of neurons.
The Bayesian prediction at time k of the direction at a point n steps in the future θk+n, given the sequence of observations ITD1:k is computed from the posterior n steps in the future pk+n(θ, ω|ITD1:k). To construct the posterior at time k+n we first compute the posterior at the current time step pk(θ, ω|ITD1:k), then predict n steps in the future using the transition probability density pk+n|k(θk+n, ωk+n|θk, ωk). Using the dependence relationships between direction, velocity, and ITD indicated in Fig 1, the posterior at time k+n is given by
pk+n(θ,ω|ITD1:k)=∬pk+n|k(θ,ω|θk,ωk)pk(θk,ωk|ITD1:k)dθkdωk.
The Bayesian prediction of the direction of the sound source at time k+n conditioned on the observations ITD1:k is the mean of the predictive posterior over direction pk+n(θ|ITD1:k). This posterior is found by marginalizing pk+n(θ, ω|ITD1:k) over the angular velocity ω. Because we are estimating a circular variable, the Bayesian prediction is the direction of the Bayesian prediction vector, defined as the vector that points in the direction of the mean value of direction n steps in the future:
BVk=∫u(θ)pk+n(θ|ITD1:k)dθ,
where u(θ) is a unit vector pointing in direction θ.
We used the Kalman filter to compute the Bayesian prediction for simulations where the linear approximation to the relationship between direction and ITD was valid. The Kalman filter computes the mean and covariance of the posterior when the system is linear with Gaussian noise [41]. Given that the relationship between azimuth and ITD is nearly linear for the frontal hemisphere, a linear model is a reasonable approximation to our system. The dynamical system for the moving source can be described as:
xk=Axk−1+ςk
where the state vector consists of the direction and angular velocity
xk=[θkωk], the matrix A=[1Δt01] describes the state dynamics, and the noise vector contains the noise for direction and velocity ςk=[ηkνk]. The noise at time k ≥ 1 is Gaussian with zero mean and covariance matrix Q and is uncorrelated across time.
The output of the system is a linear approximation to the mapping from direction to ITD plus noise:
ITDk=Cxk+ξk
where C = [2.67 0] and ξk is a Gaussian noise process with zero mean and variance R that is referred to as the observation error.
The Kalman filter is used to compute the mean and covariance of the posterior at each time. Define x^i|j and Σi|j to be the mean and covariance, respectively, of the posterior at time i given observations up to time j. The mean of the posterior distribution is computed recursively through a process of prediction and updating. The prediction one step ahead in time is computed as
x^k|k−1=Ax^k−1|k−1
Σk|k−1=AΣk−1|k−1AT+Q.
Updating the estimate with a new observation is computed as
x^k|k=x^k|k−1+Lk[ITDk−Cx^k|k−1]and
Σk|k=(I−LkC)Σk|k−1
where the Kalman gain is
Lk=Σk|k−1CT[CΣk|k−1CT+R]−1.
When an estimate has been made for the state x^k|k, it is possible to use that estimate as a basis for predicting future states at time k+n. This requires the estimate at time k to be multiplied by the state transition matrix n times:
x^k+n|k=Anx^k|k.
The covariance of the posterior at time k+n is computed as
Σk+n|k=∑m=1nAm−1Q(Am−1)T+AnΣk|k(An)T.
We used a particle filter to compute the Bayesian prediction for simulations where the linear approximation to the relationship between direction and ITD was not valid. Particle filtering algorithms are sampling-based approaches to approximating the posterior distribution that are valid for nonlinear and non-Gaussian models [40]. The particle filter algorithm we used was adapted from [49]. The algorithm is given by the following steps:
The neural population model consists of 5000 Poisson neurons with receptive fields that shift according to the prediction given in the proposition proved in the results. The preferred directions of the neurons were drawn from the prior Gaussian distribution with mean zero and standard deviation 23.3 deg. These preferred directions match the model of Fischer and Peña (2011). To generate the neural responses to a sequence of ITD inputs we first computed the predictive posterior pk+n(θ, ω|ITD1:k) as described above. We then used our main result specifying that the activities are proportional to the ratio of the posterior and prior to generate the spiking probabilities for the population of neurons. We scaled the ratio of the posterior to the prior so that firing rates would be approximately 10 spikes/s for neurons with peak responses. Spike counts were generated for the population at each time step using independent Poisson neurons with the specified rate. The direction of the PV was used to estimate the predicted source direction at each time.
The PV was tested for counterclockwise source trajectories with initial directions covering -180 deg to 180 deg in 10 deg steps and angular velocities ranging from 0 deg/s to 150 deg/s in 25 deg/s steps. We calculated the RMS error between the PV estimate θPVA(t) and the Bayesian prediction θBayes(t) to quantify the approximation error where
RMS=1T∫0T(θBayes(t)−θPVA(t))2dt.
|
10.1371/journal.pgen.1002196 | A Rice Plastidial Nucleotide Sugar Epimerase Is Involved in Galactolipid Biosynthesis and Improves Photosynthetic Efficiency | Photosynthesis is the final determinator for crop yield. To gain insight into genes controlling photosynthetic capacity, we selected from our large T-DNA mutant population a rice stunted growth mutant with decreased carbon assimilate and yield production named photoassimilate defective1 (phd1). Molecular and biochemical analyses revealed that PHD1 encodes a novel chloroplast-localized UDP-glucose epimerase (UGE), which is conserved in the plant kingdom. The chloroplast localization of PHD1 was confirmed by immunoblots, immunocytochemistry, and UGE activity in isolated chloroplasts, which was approximately 50% lower in the phd1-1 mutant than in the wild type. In addition, the amounts of UDP-glucose and UDP-galactose substrates in chloroplasts were significantly higher and lower, respectively, indicating that PHD1 was responsible for a major part of UGE activity in plastids. The relative amount of monogalactosyldiacylglycerol (MGDG), a major chloroplast membrane galactolipid, was decreased in the mutant, while the digalactosyldiacylglycerol (DGDG) amount was not significantly altered, suggesting that PHD1 participates mainly in UDP-galactose supply for MGDG biosynthesis in chloroplasts. The phd1 mutant showed decreased chlorophyll content, photosynthetic activity, and altered chloroplast ultrastructure, suggesting that a correct amount of galactoglycerolipids and the ratio of glycolipids versus phospholipids are necessary for proper chloroplast function. Downregulated expression of starch biosynthesis genes and upregulated expression of sucrose cleavage genes might be a result of reduced photosynthetic activity and account for the decreased starch and sucrose levels seen in phd1 leaves. PHD1 overexpression increased photosynthetic efficiency, biomass, and grain production, suggesting that PHD1 plays an important role in supplying sufficient galactolipids to thylakoid membranes for proper chloroplast biogenesis and photosynthetic activity. These findings will be useful for improving crop yields and for bioenergy crop engineering.
| Photosynthesis is carried out in chloroplast, a plant-specific organelle. Photosynthetic membranes in chloroplasts contain high levels of glycolipids, and UDP-galactose is a dominating donor for glycolipid biosynthesis. Although glycolipid assembly of photosynthetic membranes has been characterized at the genetic and enzymatic level, the mechanism of substrate supply of UDP-galactose for the glycolipid biosynthetic pathway remains obscure. By genetic screening of rice mutants that are impaired in photosynthetic capacity and carbon assimilation, we identified PHD1 as a novel nucleotide sugar epimerase involved in a process of glycolipid biosynthesis and participating in photosynthetic membrane biogenesis. PHD1 was preferentially expressed in green and meristem tissues, and the PHD1 protein was targeted to chloroplasts. We revealed that UDP-galactose for glycolipid biosynthesis catalyzed by the new enzyme was generated inside chloroplasts, and the reduced amounts of glycolipids in the mutant led to decreased chlorophyll content and photosynthetic activity. Overexpression of this gene lead to growth acceleration, enhanced photosynthetic efficiency, and finally improved biomass and grain yield in rice. These results suggest that PHD1 has significant economic implications in both traditional crop improvement and bioenergy crop production.
| Plants possess a sophisticated sugar biosynthetic machinery comprised of families of nucleotide sugars that can be modified at their glycosyl moieties by nucleotide sugar interconversion enzymes to generate different sugars [1], [2]. UDP-glucose 4-epimerase (also UDP-galactose 4-epimerase, UGE; EC 5.1.3.2) catalyzes the interconversion of UDP-D-glucose (UDP-Glc) and UDP-D-galactose (UDP-Gal) [3], [4]. UGE is essential for de novo biosynthesis of UDP-Gal, a precursor for the biosynthesis of different carbohydrates, glycolipids, and glycosides. Genes encoding UGE have been cloned from a range of different organisms including bacteria, yeast, and human [5]–[7], and the crystal structures have also been obtained [8]–[10].
The original biochemical and genetic analyses of UGE in plants was described by Dörman and Benning [11]. To date, five UGE isoforms have been identified in Arabidopsis [2], [12], three in barley [13], and a family of four putative UGE isoforms exist in rice. In Arabidopsis, global co-expression analysis revealed that UGE2, -4, and -5 preferentially act in the UDP-Glc to UDP-Gal directions, whereas UGE1 and UGE3 might act in the UDP-Gal to UDP-Glc directions [14]. Reverse genetic studies demonstrated that UGE2 and UGE4 influence vegetative growth and cell wall carbohydrate biosynthesis, that UGE3 is specific for pollen development, and that UGE1 and UGE5 act in stress situations [15], [16]. Compared to 4-day-old seedlings, UGE expression increased 5-fold in roots of 3-week-old pea plants, suggesting that increased UGE expression correlated with the copious secretion of pectinaceous mucigel in older seedling roots [17]. To date, all UGEs identified from plants lack transmembrane motifs and signal peptides and appear to exist as soluble entities in the cytoplasm.
Photosynthetic reactions in higher plants depend on the well-developed chloroplast thylakoid membrane system. Chloroplast thylakoid assembly and maintenance require a continuous supply of membrane constituents. Galactose-containing glycerolipids are predominant lipid components of photosynthetic membranes in plants, algae, and cyanobacteria. The two most common galactolipids are mono- and digalactosyldiacylglycerol (MGDG and DGDG), which account for about 50 and 25 mol% of total thylakoid lipids, respectively [18], [19]. About 80% of all plant lipids are associated with photosynthetic membranes, and MGDG is considered to be the most abundant membrane lipid on earth. Recent studies have demonstrated that galactolipids play an important role in not only the organization of photosynthetic membranes, but also in their photosynthetic activities [20], [21]. Arabidopsis mutants with a lower amount of these galactolipids have a reduction in chlorophyll content and photosynthetic activity, alterations in chloroplast ultrastructure, and impairment of growth [22]–[25].
In plants, MGDG is synthesized in two unique steps: (i) the conversion of UDP-Glc into UDP-Gal by an UGE, and (ii) the transfer of a galactosyl residue from UDP-Gal to diacylglycerol (DAG) for synthesis of the final product by MGDG synthase (MGD1) [26], [27]. Although MGD1 has been characterized at both genetic and enzymatic levels, the UDP-Gal supply mechanisms for the MGDG biosynthetic pathway remain obscure. MGD1 is localized in the inner chloroplast envelope membrane [26], [27] and uses UDP-Gal as a substrate. However, the concentration of UDP-Gal in chloroplasts is considered to be very low [28], suggesting that the UDP-Gal source is imported from the cytosol or generated inside chloroplasts.
To gain insight into genes controlling photosynthetic activity and carbon assimilation in plants, a rice stunted growth mutant (phd1) with decreased photoassimilate and yield production was selected for further study from a large-scale screening of our T-DNA mutant population. Interestingly, PHD1 encoded a chloroplast-localized UDP-Glc epimerase involved in UDP-Gal supply for chloroplast galactolipid biosynthesis during photosynthetic membrane biogenesis. Its homologs are highly conserved in the plant kingdom, and the gene was preferentially expressed in various young meristems where plastid proliferation actively occurred. Most strikingly, overexpression of PHD1 increased photosynthetic activity and enhanced rice growth. The important roles of PHD1 in photosynthetic capability and carbon assimilate homeostasis are discussed.
To identify genes affecting photosynthetic activity and carbon assimilation, a large-scale screening of our rice T-DNA insertion mutant population (Oryza sativa var. Nipponbare background) [29] was carried out. Of 480 mutant lines with altered carbohydrate levels in vegetative organs, photoassimilate defective1 (phd1) with both low carbohydrate contents and stunted growth was selected for further characterization. Scanning electron micrograph of culms demonstrated that fewer starch granules were deposited in parenchyma cells of the phd1 mutants (data not shown). During the young seedling stage, both shoots and primary roots of the mutant were shorter and lighter than those of the wild type (Figure 1A). After internode elongation, the phd1 mutant exhibited a semi-dwarf, less grain-filling, retarded vegetative growth, later flowering, and less tillering phenotype (Figure 1B–1E). In addition, although the grain number per panicle was not altered between the mutant and wild type, the seed-setting ratio of the phd1 mutant was significantly decreased, which finally led to a significant reduction of grain yield (Figure 1F, 1G). Compared to wild type, mature leaves of the mutant had somewhat reduced sucrose (Figure 1H) and rather low starch levels (Figure 1I) at all time-points taken during the light/dark cycle, while hexose levels were a little higher in the mutant (Figure S1).
Genetic analysis indicated that the phd1 phenotype was controlled by a single recessive gene that did not co-segregate with the T-DNA insertion, and hence map-based cloning was carried out. The PHD1 locus was physically delimited to a 72-kb region on the short arm of chromosome 1. This region contains six annotated genes, and sequencing of these genes from phd1-1 identified a single nucleotide transition (G-to-T) in exon 2 of Os01g0367100, leading to a premature translational termination. The identity of Os01g0367100 as PHD1 was confirmed by analysis of two other phd1 alleles with similar phenotypes isolated from the same genetic screen, for which a single nucleotide substitution (A-to-T) in exon 7 in phd1-2 and a 13-bp insertion between exon 3 and exon 4 in phd1-3 were found (Figure 2A). Almost no PHD1 mRNA was detected in any of the three allelic mutants (Figure S2). The phd1 phenotype was complemented by transgenic expression of wild type Os01g0367100 in the phd1-1 mutant background (Figure 2B, 2C), confirming that the nonsense mutation of Os01g0367100 was responsible for the presumed null mutant phenotype.
Database searches revealed that PHD1 has similarity to proteins from Thalassiosira pseudonana (XP_002290295), Phaeodactylum tricornutum (XP_002178225), Chlamydomonas reinhardtii (XP_001699105), Micromonas pusilla (EEH60780), Ostreococcus tauri (CAL54696), Physcomitrella patens (XP_001767242), Ricinus communis (XP_002516868), Arabidopsis thaliana (AT2G39080), Populus trichocarpa (XP_002311843), Vitis vinifera (XP_002276706), Zea mays (NP_001131736), and Sorghum bicolor (XP_002457832), with 27 to 75% amino acid identity (Figure S3). Phylogenetic analysis between PHD1 and its 16 putative homologs indicated that PHD1 is closely related to Sb03g014730 from sorghum and LOC100193101 from maize (Figure 3). PHD1 homologs are only found in the plant kingdom, suggesting that these proteins are evolutionally conserved across plant species. However, none of the homologous genes have been functionally characterized. Analysis of the conserved domain demonstrated that PHD1 and its homologs contain the consensus WcaG domain, featured in nucleoside-diphosphate sugar epimerases (Figure S3). One of the best characterized nucleotide sugar epimerases is UDP-Glc epimerase, which catalyzes the interconversion of UDP-Glc and UDP-Gal. Hence, PHD1 and its homologs may function as novel plant specific UDP-Glc epimerases.
To validate PHD1's biochemical function as an UDP-Glc epimerase, the mature PHD1 protein lacking the putative N-terminal 62-aa transit peptide was expressed in E. coli and UGE activity was examined. The result showed that PHD1 could catalyze the conversion of UDP-Gal to UDP-Glc, and curve fitting indicated that UDP-Gal binding followed a simple Michaelis-Menten kinetics with a Km value of 0.84 mM at 30°C (Figure S4A). To examine whether PHD1 had UDP-Glc epimerase activity in vivo, the mature PHD1 under the control of the yeast glyceraldehyde-3-phosphate dehydrogenase promoter was used to complement the auxotrophic phenotype of a yeast gal10Δ mutant which cannot grow on a medium containing D-galactose as sole carbon source. The complementation results demonstrated that PHD1 also had UDP-Glc epimerase activity in vivo (Figure S4B).
RNA gel blot analysis revealed that PHD1 was present in all green tissues, with highest abundance in leaf blades and leaf sheaths, then flowers and culms, but only at very low levels in roots (Figure 4A). mRNA in situ hybridization using an antisense probe revealed that PHD1 was expressed predominantly in leaf primordia and shoot apical meristems (Figure 4B), the mesophyll cells surrounding the vascular bundles of young leaves (Figure 4C), inflorescence primordia (Figure 4D), and axillary buds (Figure 4E). In contrast, hybridization with a PHD1 sense probe showed no signal (Figure 4F).
PHD1 encodes a 340 aa protein with a putative 62-aa chloroplast transit peptide at the N-terminus. To confirm chloroplast localization of PHD1, the full-length PHD1 was fused to the green fluorescent protein (GFP) reporter gene under the control of the cauliflower mosaic virus (CaMV) 35S promoter and subsequently transformed into rice shoot protoplasts. Figure 5A shows that GFP fluorescence co-localized with the red chlorophyll autofluorescence, confirming that PHD1 was a chloroplast-localized protein and the predicted transit peptide was functional.
To further investigate the subcellular localization of PHD1, we performed western blot experiments using purified plastid subfractions (Figure 5B). Several antibodies were used as specific markers for the different chloroplast subfractions. Tic 40 was used as a specific envelope marker, and Rubisco, the major stroma protein, as a marker of this chloroplast subfraction. PsbA, one of the components of photosystem II (PSII), was used as a marker to validate the thylakoid membrane fraction, and HSP82 was used as a cytosol specific marker. As shown in Figure 5B, the PHD1 protein was detected mainly in the stroma fraction and was absent from the cytoplasmic compartment, thus confirming that PHD1 was a chloroplast-targeted protein. To complete the subcellular localization study and to obtain additional information about the distribution of PHD1 in different chloroplast subcompartments, we further performed immunocytochemical analysis on ultrathin sections of rice tissues using polyclonal PHD1 antiserum. The positive signal of PHD1, visualized as black dots, was found specifically in the chloroplasts (Figure 5C and 5D). In contrast, sections treated with a preimmune serum (Figure 5E and 5F) showed no signal. The overall data thus strongly indicated that PHD1 is targeted to chloroplasts in rice.
Intact chloroplasts were isolated from leaves of wild type and phd1-1 mutant plants, and the UGE activity in isolated chloroplasts was measured (Figure S5). Compared to the wild type, a severe decrease (ca. 50%) in UGE activity was observed in isolated chloroplasts from the phd1-1mutant compared with the wild type, suggesting that PHD1 was responsible for a major part of the UGE activity in chloroplasts. Moreover, levels of the UGE substrates UDP-Glc and UDP-Gal in isolated chloroplasts were also determined (Figure 6). While compared to wild type and complemented mutant an overabundance of UDP-Glc was found in chloroplasts isolated from the phd1-1 mutant, almost no amount of UDP-Gal was detected in the mutant. The levels of nucleotide sugars in whole leaves were also determined, which showed that the amount of UDP-Gal was slightly higher in phd1-1 than in wild type plants, and the UDP-Glc amount was significantly higher (Figure S6). Hence, the ratio of UDP-Glc to UDP-Gal in phd1-1 leaves was also higher than in wild type plants. These results suggested that PHD1 dysfunction may trigger an accumulation of substrates and disturb the balance of interconversion between the two sugar nucleotides.
Chloroplast membranes contain high levels of glycolipids, and UDP-Gal is a dominant substrate for glycolipid biosynthesis. To examine the effect of PHD1 dysfunction on membrane lipid homeostasis, the composition of total lipids extracted from phd1-1, wild type, and PHD1-complemented plants was analyzed (Figure 7). In the phd1-1 mutant, the mol% amount of MGDG was reduced by 19% compared to wild type and the complemented plants, indicating that PHD1 is involved in MGDG biosynthesis. In contrast, only a slight decrease (2.5%) in DGDG content was observed in the phd1-1 mutant, demonstrating that PHD1 may not be required for DGDG synthesis and suggesting that the UDP-Gal substrate for DGDG formation was presumably supplied from the cytosol. Reduced abundance of MGDG in phd1-1 was accompanied by an increase in the abundance of other major membrane lipids such as phosphatidylcholine (PC), phosphatidylglycerol (PG), and phosphatidylinositol (PI), while the mol% levels of sulfoquinovosyldiacylglycerol (SQDG) and phosphatidylethanolamine (PE) were not altered significantly in the phd1-1 mutant (Figure 7A). Because the two galactolipids and SQDG are major components of thylakoid membrane lipids, this result suggests that the mutant had an overall lower amount of chloroplast membrane lipids than wild type plants. Focusing on the exclusive chloroplast lipid MGDG, the fatty acid composition was also investigated (Figure 7B). MGDG of the phd1-1 mutant contained considerably decreased levels of stearic acid (18∶0) compared with the wild type and elevated levels of linoleic acid (18∶2) and linolenic acid (18∶3). The levels of other fatty acids were similar to those observed in wild type plants. Hexadecatrienoic acid (16∶3), which is typically found in the plant prokaryotic pathway, was not detected in all the rice plants, suggesting that rice entirely relies on endoplasmic reticulum (ER)-derived lipids for thylakoid galactoglycerolipid biosynthesis.
Noninvasive chlorophyll fluorescence measurements indicated that the maximum quantum yields for photosystem II photochemistry (Fv/Fm) were similar for phd1-1 and wild type (Table 1). The effective quantum yield of photochemical energy conversion in photosystem II (ΦPSII) was slightly but significantly reduced in the mutant (Table 1). Pigment content was also reduced in the phd1-1 mutant (Table 1). Interestingly, chloroplasts of 2-month-old phd1-1 plants were significantly smaller than those of wild type plants (wild type, 5.0±0.4 µm; phd1-1, 3.0±0.5 µm), and starch grains were also either absent or reduced in size and/or number in the mutant (Figure S7). These data indicated that a reduced amount of galactolipids in chloroplasts and perhaps a smaller size of chloroplasts due to a decrease in membrane lipid content might lead to reduced photosynthetic capability of higher plants.
UDP-Gal is the activated form of galactose in biosynthetic reactions, but a galactose salvage pathway exists in eukaryotic organisms. To assess expression of genes involved in the Leloir salvage pathway, the expression levels of three key genes of this pathway, GalM, GalK, and GalT, were analyzed in both phd1-1 and wild type. The expression of all three genes was significantly upregulated in the phd1-1 mutant, suggesting an activation of the whole salvage pathway (Figure 8A). β-Lactase is involved in the generation of free β-D-Gal from polysaccharide breakdown, and UDP-Glc pyrophosphorylase (UGP) catalyzes the formation of UDP-Glc from Glc-1-P. The expression levels of genes encoding β-lactase and UGP3 were also upregulated in phd1-1. More strikingly, the expression levels of OsUGE1 and OsUGE4 encoding for putative cytoplasmic isoforms of UGE in rice were more than two-fold higher in phd1-1 than in wild type plants, indicating an upregulation of de novo UDP-Gal biosynthesis in the cytoplasm. These results suggested that PHD1 may be responsible for a majority of the UGE function in chloroplasts, and appears to be involved in the generation of UDP-Gal from UDP-Glc to supply building blocks for galactolipid biosynthesis required for proper chloroplast membrane composition.
Because the phd1-1 mutant exhibited a dramatic decrease of carbon assimilate levels, we determined whether transcript levels of several key genes involved in the synthesis, transport, and cleavage of starch and sucrose were altered in mature leaves of wild type and phd1-1 plants. Interestingly, while the expressions of starch biosynthesis genes such as AGPL2, SSI, SSIIIa, GBSS, BE, and BT1, were suppressed in the phd1-1 mutant (Figure 8B), expression levels of genes participating in sucrose cleavage, such as INV1/3 and SuSy1, were all increased (Figure 8C). Meanwhile, the GPT gene encoding a glucose-6-phosphate/phosphate translocator was upregulated in phd1-1, implying an enhanced export of hexose-phosphates from chloroplasts to the cytosol. In addition, increased expression level of UGP2, a gene involved in UDP-Glc synthesis, was correlated with increased UDP-Glc accumulation and a higher UDP-Glc/UDP-Gal ratio in the phd1-1 mutant.
Since a mutation in PHD1 affected photosynthesis and growth rate, we further investigated whether biomass and grain yield could be improved by PHD1 overexpression. When grown in paddy fields, transgenic rice plants overexpressing PHD1 showed a significant increase in tillering (branching) and photosynthetic rate (Figure 9A, Table S1) in lines that overexpressed the PHD1 protein (Figure 9B). Compared to non-transgenic control plants, grain yield per plant of transgenic lines S3, S5, and S8 increased 10.7, 15.5, and 18.3%, respectively (Figure 9C). In addition, the growth rate of transgenic plants accelerated at the seedling stage and dry material accumulation was enhanced 12.5% to 22.4% at the mature stage compared to non-transgenic plants (Figure 9D, Table S1). These results demonstrated that PHD1 overexpression in rice is positively correlated with an increase in biomass production and grain yield.
To date all UGE genes coding for UDP-Glc epimerases isolated from plants are localized to the cytosol, where their substrates UDP-Glc and UDP-Gal are present at high levels [30]. As a precursor for the synthesis of the galactolipid MGDG in chloroplasts, UDP-Gal is widely assumed to be mobilized from the cytosol, because the UDP-Gal concentration is relatively low within plastids [28] and MGDG synthase (MGD1) is associated with the inner envelope membrane [26], [27]. However, a labeling experiment in which radioactively labeled UDP-Gal was applied to isolated Arabidopsis chloroplasts revealed that radioactivity was not efficiently incorporated into MGDG [23], raising the question of how UDP-Gal is transported into the chloroplasts. In this study, we found that a mutation in PHD1, which encodes a novel rice plastidial UGE involved in the biosynthesis of chloroplast galactolipids, lead to disturbed carbon assimilation homeostasis and impaired photosynthetic efficiency. Our work revealed that PHD1 codes for an active epimerase that is targeted to chloroplasts, and, therefore, that the UDP-Gal substrate for MGDG biosynthesis can be generated in situ in chloroplasts (Figure 10). The novel finding that this UGE is chloroplast-targeted was supported by three independent lines of evidence (Figure 5). First, PHD1-GFP fusion products were found exclusively in chloroplasts. Second, Western blot analyses of fractionated chloroplasts showed that PHD1 was highly enriched in the stroma fraction of chloroplasts. And third, immunocytochemistry indicated that PHD1 was concentrated inside the chloroplast stroma, most likely associated with the thylakoid surface. This striking result provides a well-defined genetic and biochemical framework to study the novel functional mechanism of this UGE in plastids, and to evaluate the role of galactolipids in photosynthetic activity of rice.
Of MGDG synthases that are primarily important for thylakoid membrane biogenesis, MGD1 is considered to be the major isoform [24]. In Arabidopsis, two more MGDG synthases, MGD2 and MGD3, are targeted to the outer chloroplast envelope where substrates can be recruited from the cytosol [27]. MGDG generated by them can move from the outer to the inner envelope and to the thylakoids. Here we show that compared to wild type, the relative amount of the major galactolipid MGDG was reduced by 19% in the phd1-1 mutant, whereas that of DGDG was only slightly decreased by 2.5%. We observed a slight increase in the mol% amount of the thylakoid lipid phosphatidylglycerol, which may compensate for a fraction of the galactolipids lost in the phd1-1 mutant. Meanwhile, the relative amount of several extraplastid phospholipids was found to be slightly but significantly higher in the phd1-1 mutant, suggesting that compared to extraplastidic membranes, the overall amount of plastid membranes might have decreased. These results are consistent with the hypothesis that the amounts of glycolipids and phospholipids are reciprocally controlled in plants to maintain a proper balance of lipids in the ER and plastid membrane systems [20], [31]. It has been shown previously that osmotic stress induced variations in membrane fluidity that correlated with the physical properties of membrane lipids [32]. Due to an overabundance of UDP-Glc observed in chloroplasts and entire leaves of the phd1-1 mutant, hyperosmotic stress might occur, and an increased production of 18∶3 could affect hyperosmotic stress tolerance in the mutant chloroplasts. This would be in agreement with earlier observations that transgenic enhancement of fatty acid unsaturation rendered cells and whole plants more tolerant to sorbitol-induced osmotic stress in tobacco [33].
Most galactolipids are restricted to plastid membranes during normal growth and development, however, DGDG can also be found in extraplastidic membranes following phosphate (Pi) starvation [34], [35]. Importantly, x-ray crystallographic analyses of photosynthetic proteins in cyanobacteria revealed that MGDG is associated with the core of the reaction centers of both photosystems I and II (PSI and PSII) [36], [37], which suggest that these lipids are required not only as bulk constituents of photosynthetic membranes, but also for the photosynthetic reaction itself. Consistent with this, we found that the effective quantum yield of photochemical energy conversion in photosystem II (ΦPSII) was reduced in the phd1-1 mutant. Seedlings lacking MGDG were previously shown to have disrupted photosynthetic membranes, leading to a complete impairment of photosynthetic ability and photoautotrophic growth [22], [24]. In agreement with this, a possible reduction of thylakoid membrane amount and a changed galactolipid to phospholipid ratio in chloroplast membranes in the phd1-1 mutant might have led to the dramatic phenotype of retarded growth, reduced photosynthetic capability, and decreased photoassimilate accumulation. Taken together, this strongly suggests that the stunted growth phenotype of phd1-1 mutants is due to an insufficient provision or slower production of membrane building blocks to support chloroplast proliferation during plant growth, which is also consistent with the reduced numbers of thylakoid stacks and sizes of chloroplasts observed in mutant plants.
In plants, starch acts as a depository for reduced carbon produced in leaves during the day, and as a supply of chemical energy and anabolic source molecules during the night [38]. Pyrophosphate (PPi) is produced during the upregulation of UGP3 (Figure 10), and hydrolyzed by very high pyrophosphatase (PPase) activity in plastids [39]. Moreover, inorganic phosphate (Pi) released during PPi hydrolyzation is an inhibitor for key regulatory starch biosynthesis enzymes such as AGP [40]. In the phd1-1 mutant, expression levels of starch biosynthesis genes such as AGP, SS, GBSS, and BE, were significantly downregulated in source leaves, leading to a sharp decrease of starch content. However, the reduced starch did not result in increased sucrose levels, because activation of sucrose cleavage genes SuSy1 and INV1/3 resulted in reduced sucrose and increased hexose-phosphate and UDP-Glc levels. Therefore, sucrose as the main transport form of photoassimilate produced in source organs was not able to export efficiently to the sink organs. Moreover, a large amount of UDP-Glc catalyzed by SuSy1 or UGP2 would be converted to UDP-Gal by cytosolic OsUGE1/4 and transported into chloroplast as galactosyl donors of chloroplast glycolipids to compensate for the loss of PHD1 activity in the phd1-1 mutant. In contrast, PHD1 overexpression in rice, which enhanced PHD1 activity in chloroplasts (Figure S5), might increase the relative amount of MGDG and increase the effective quantum yield of photochemical energy conversion in thylakoid membranes, resulting in increased photosynthetic efficiency and growth rate, implicating a key role of PHD1 for the photosynthetic system in rice. These improvements of both biomass production and grain yield have significant economic implications in both traditional crop improvement and bioenergy crop production.
The rice (Oryza sativa L.) phd1 mutant is in the Nipponbare (ssp japonica) background. F2 mapping populations were generated from a cross between the rice phd1 mutant and MH63 (ssp indica). Rice plants were cultivated in the experimental station of the Institute of Genetics and Developmental Biology (IGDB) in Beijing in natural growing seasons. For analysis of diurnal changes of starch and sugars, rice plants were kept in a growth chamber at 28°C and 70% relative humidity under a photoperiod of 12 h light/12 h darkness, with a light intensity of 200 µmol quanta m−2 s−1.
Genomic DNA was isolated from seedlings of the selected plants with the mutant phenotype. For fine mapping of PHD1, STS markers were generated based on the polymorphisms between Nipponbare and MH63. The molecular lesion of phd1-1 was identified by PCR amplification of the PHD1 genomic region from wild type and phd1-1 mutant plants and comparison of their sequences. The candidate gene was mapped between the 2 new STS markers S221 (5′-AGAGCTAGGGGGTAAAAA-3′ and 5′-GTGCAGAACAGTGGAATG-3′) and S246 (5′-AACCCTATCCTTCCTCACCA-3′ and 5′-TTGTCCCTCCGCCTGCTTCC-3′).
PHD1 homologs were detected by BLASTp using the entire amino acid sequence of PHD1 as a query in the National Center for Biotechnology Information database (http://www.ncbi.nlm.nih.gov/BLAST). Multiple alignment of the homologs was performed by Clustal X version 2.0 with the default parameters [41] and manually adjusted. For constructing phylogenetic trees, the neighbor-joining method of the MEGA 4.1 software [42] was used, and a bootstrap analysis with 1 000 replicates was performed to test the confidence of topology.
The BAC clone BAC53 containing the entire PHD1 fragment was digested with Sac I and Pst I to generate a 7.96 kb genomic DNA fragment. The DNA fragment was ligated to the Sac I and Pst I digested pCAMBIA1300 vector (CAMBIA), to generate the pSCL construct for complementation analysis. The full-length PHD1 cDNA was PCR amplified using primers 5′-GATCCGATCCCCTCACCTC-3′ and 5′- TTCTCTGGCCGAAACCATT-3′, and subcloned into the pCAMBIA2300-35S binary vector, between the cauliflower mosaic virus 35S promoter and nopaline synthase (nos) terminator, to generate the pSOL construct for overexpression analysis. Transgenic rice plants were generated according to Agrobacterium tumefaciens-mediated transformation methods [43], [44]. The transgenic plants were then transferred to the field at the IGDB experimental station for normal growth and seed harvesting.
PHD1 cDNA was amplified by primer sets 5′-TGATGATACAGGGGTCAAGATG-3′ and 5′-ACTGTCAAGACCAAGGAATTCT-3′ and cloned into the Xma I and Xho I sites of pGEX-4T-1 (GE Healthcare Life Sciences) and expressed in E. coli strain BL21 (DE3). Recombinant PHD1 protein was affinity-purified through glutathione Sepharose resin (Amersham Pharmacia Biotech) and used for antibody production [45].
Total RNA was prepared with an RNeasy kit (Qiagen). In the RNA gel blot analysis, 5 µg of total RNA was electrophoresed on a 1.2% (w/v) agarose gel and transferred to a nylon membrane, and mRNA was detected by a digoxigenin labeling system (Roche Diagnostics). For quantitative RT-PCR, 15 ng of cDNA and SYBR Green SuperMix (Bio-Rad) were used in 15 µL qRT-PCR reactions with a CFX96 96-well real-time PCR detection system (Bio-Rad) and CFX96 software to calculate threshold cycle values, and rice 18S ribosome RNA was used as an internal control. Oligonucleotide primers are given in Table S2. The 2−ΔΔCT method was adopted to calculate the relative expression levels for the phd1 and wild type samples, and a two-tailed t test used to compare the ratios and determine statistical significance [46].
Freshly collected rice tissues were fixed in FAA solution (50% ethanol, 5% acetic acid, 3.8% formaldehyde) at 4°C overnight, dehydrated with ethanol solution from 50% to 100%, cleaned by a series of xylene washes from 25% to 100%, and embedded in paraffin (Paraplast Plus, Sigma-Aldrich) at 54–56°C as described in [47]. 8 to 12 µm sections were cut with a microtome (Leica RM2265), and mounted on RNase-free glass slides and photographed.
RNA in situ hybridization was performed as described previously with minor modification [48]. Briefly, the 420-bp region of PHD1 was amplified by gene-specific primers with T7 or SP6 promoters 5′-TAATACGACTCACTATAGGGCCCCTTCTCCGTCAACCT-3′ and 5′-AACGAAAGAGCCTTCACCA-3′ or 5′-CCCCTTCTCCGTCAACCT-3′ and 5′-ATTTAGGTGACACTATAGAACGAAAGAGCCTTCACCA-3′ in front of the reverse primer (for making anti-sense probe) or forward primer (for making sense probe). Digoxigenin-labeled RNA probes were prepared using a DIG Northern Starter Kit (Cat. No. 2039672, Roche) according to the manufacturer's instructions. The hybridization signals were observed using bright field imaging with a microscope (Olympus BX51) and photographed with a Micro Color CCD camera (DVC Co. Austin, USA).
A binary vector containing GFP fused with full-length PHD1 was constructed as follows. The PCR product amplified with primers 5′-ACCTCCGTCCCTGCTTCCTC-3′ and 5′-GGGCTCCCAACCAATCTCA-3′ was subcloned into the CaMV 35S::GFP vector to generate CaMV 35S::PHD1-GFP. The binary vector was transformed into rice protoplasts using the polyethylene glycol method [49]. After overnight incubation in the dark, the protoplasts expressing GFP were imaged by a confocal laser scanning microscope (LSM510, Zeiss, Germany) using 488 nm excitation and 500–530 nm emission pass-filters. Chlorophyll autofluorescence was detected with 570 nm excitation and 640 nm emission pass-filter [50]. Composite figures were prepared using Zeiss LSM Image Browser software.
PHD1 and its derivative cDNAs were amplified by PCR using the primers 5′- ATGATACAGGGGTCAAGATGG-3′ and 5′-ACTGTCAAGACCAAGGAATTCT -3′, and inserted into the vector pDBLeu (Invitrogen). The Euroscarf S. cerevisiae strain BY4742 (Matα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 gal10::kanMX4) was transformed using a lithium acetate procedure and tested on 1% (w/v) galactose medium (1% (w/v) yeast extract (Duchefa), 2% (w/v) Bacto-peptone (BD Bio- sciences), 1% (w/v) galactose (Sigma), 2% (w/v) Micro agar (Duchefa)).
Individual samples (leaves of circa 500 mg fresh weight) were harvested and frozen rapidly in liquid N2. The frozen samples were homogenized and extracted with perchloric acid. Glucose, fructose, sucrose, and starch were measured enzymatically for the neutralized supernatant (sugars) and the insoluble pellet (starch) [51]. Determination of UDP-Glc and UDP-Gal were performed as described [11].
Total lipids were extracted from 2-month-old phd1-1, wild type, and the PHD1-complemented plants as described [52]. For quantitative analysis, individual lipids were separated by two-dimensional thin-layer chromatography and used to prepare fatty acid methyl esters. The methyl esters were quantified by gas-liquid chromatography as described [53]. A 1 µl sample was applied for GC-MS (Agilent 7890A GC coupled to 5975C MS) analysis at a 10∶1 split ratio. The GC-MS program started with 80°C for 1 min, then ramped at 8°C/min to 300°C and held for 5 min; injector and inlet temperatures were set at 250°C and 280°C, respectively. Separation was performed on a HP-5 MS column (30 m×0.25 mm×0.25 µm) with a constant flow of 1.1 ml/min helium. The MS scan range was from 50 to 500 m/z. The quantification of fatty acid methyl esters was performed by the external standard method.
UGE activity was measured using a NADH-coupled assay developed by Wilson and Hogness [54] with some minor modifications. The 1 ml assay mixture consisted of 100 mM glycine buffer (pH 8.7), 1 mM β-NAD+ (Sigma), and 0.8 mM UDP-Gal (Sigma). The reaction was started by adding 10 µl of epimerase (140 µg/ml) in 50 mM Tris·Cl (pH 7.6), 1% (w/v) bovine serum albumin, 1 mM dithiothreitol, 1 mM EDTA, and 1 mM β-NAD+, and stopped by incubation for 10 min at 100°C. The UDP-glucose produced was determined by addition of 0.04 unit of bovine UDP-glucose dehydrogenase (Calbiochem) and incubation for 10 min at 30°C, and the increase in absorbance due to NADH formation was then measured at 340 nm. Km values were determined by varying the UDP-Gal concentration between 0.4 mM and 3.2 mM. The experiment was conducted in triplicate.
All isolation procedures were carried out at 4°C. Batches of 50 g rice leaves were cut to little pieces and homogenized in 250 ml of isolation buffer (50 mM HEPES/KOH, pH 7.8, 0.33 M sorbitol, 2 mM EDTA, 1 mM MgCl2, 1 mM MnCl2, 0.1 M Na-ascorbate, 0.2% (w/v) bovine serum albumin) using a Waring blender. The chloroplast suspension was passed through four layers of Miracloth and centrifuged at 4 000 g for 4 min. The pellet was gently suspended in the isolation buffer and layered onto a discontinuous density gradient consisting of 10, 40, and 80% (v/v) Percoll in the isolation buffer. The gradient was centrifuged at 8 000 g for 10 min. Intact chloroplasts distributed around the 40/80% Percoll interface were isolated and reapplied to the Percoll gradient centrifugation.
Chloroplasts were lysed by resuspension to 0.5 mg chlorophyll ml−1 in 10 mM HEPES/KOH (pH 8.0), 5 mM MgCl2, for 20 min on ice, and the lysate was fractionated into envelope, stroma, and thylakoids by differential centrifugation as described by Skalitzky et al [55]. All solutions contained a cocktail of protease inhibitors. To verify recovery and purity of the sucrose density fractions, several antibodies against specific marker proteins were used: Tic40 was used as an envelope marker, RbcL as a stromal marker, and PsbA as a thylakoid membrane marker.
Immunoelectron microscopy experiments were carried out as previously described [56]. Briefly, nickel grids carrying ultrathin leaf sections prepared from two-week-old wild type seedlings were sequentially floated in 0.01 M sodium phosphate buffer (PBS, pH 7.2) containing 5% (w/v) bovine serum albumin (BSA) for 5 min, then for 1 h at 37°C in PBS containing diluted anti-PHD1 antibody. After several washes in PBS, ultrathin sections were incubated for 1 h at 37°C in PBS containing goat anti-rabbit IgG antibody conjugated to 10-nm colloidal gold (1∶40, Sigma-Aldrich, St. Louis, MO, USA). After 5 washes with PBS, ultrathin sections were washed with distilled water, air dried, counterstained with 2% uranyl acetate, and examined with a FEI Tecnai G2 20 transmission electron microscopy at an accelerating voltage of 120 kV. Negative controls were performed using the same procedure with the exception of substituting the anti-PHD1 antibody with preimmune serum.
|
10.1371/journal.pgen.1003631 | Genome-scale Co-evolutionary Inference Identifies Functions and Clients of Bacterial Hsp90 | The molecular chaperone Hsp90 is essential in eukaryotes, in which it facilitates the folding of developmental regulators and signal transduction proteins known as Hsp90 clients. In contrast, Hsp90 is not essential in bacteria, and a broad characterization of its molecular and organismal function is lacking. To enable such characterization, we used a genome-scale phylogenetic analysis to identify genes that co-evolve with bacterial Hsp90. We find that genes whose gain and loss were coordinated with Hsp90 throughout bacterial evolution tended to function in flagellar assembly, chemotaxis, and bacterial secretion, suggesting that Hsp90 may aid assembly of protein complexes. To add to the limited set of known bacterial Hsp90 clients, we further developed a statistical method to predict putative clients. We validated our predictions by demonstrating that the flagellar protein FliN and the chemotaxis kinase CheA behaved as Hsp90 clients in Escherichia coli, confirming the predicted role of Hsp90 in chemotaxis and flagellar assembly. Furthermore, normal Hsp90 function is important for wild-type motility and/or chemotaxis in E. coli. This novel function of bacterial Hsp90 agreed with our subsequent finding that Hsp90 is associated with a preference for multiple habitats and may therefore face a complex selection regime. Taken together, our results reveal previously unknown functions of bacterial Hsp90 and open avenues for future experimental exploration by implicating Hsp90 in the assembly of membrane protein complexes and adaptation to novel environments.
| Hsp90 is a chaperone protein that aids the folding of many other proteins (clients), which tend to be signal transduction proteins. Hsp90 is particularly important when organisms are under environmental or mutational stress (e.g. in cancerous cells). Although Hsp90 is well-studied in eukaryotic species from yeast to humans, little is known about its counterpart in bacteria. To address this challenge, we analyzed the presence and absence of thousands of genes across numerous bacterial species and identified genes that co-evolved with Hsp90. These genes provide insights into potential functions of Hsp90 in bacteria. We found that Hsp90 co-evolves with membrane-associated protein complexes such as the flagellum and that Hsp90 is associated with a preference for inhabiting multiple habitats. We extended our analysis to identify genes that exhibit evolutionary dynamics characteristic of Hsp90 clients. Many of the putative clients were involved in flagellar assembly, suggesting a crucial role of Hsp90 in the regulation of bacterial motility. We experimentally confirmed that E. coli Hsp90 interacts with selected candidates and demonstrated Hsp90's role in flagellar motility and chemotaxis. The computational approach described here, identifying novel functions and specific clients of bacterial Hsp90, further provides exciting starting points for research in bacterial chaperone biology.
| In eukaryotes, the universally conserved and essential chaperone Hsp90 aids the folding of key proteins in development and responses to environmental stimuli [1]–[3]. In yeast, up to 10% of all proteins are estimated to be Hsp90 clients under standard culture conditions [4]. Hsp90 function is even more important under stressful conditions that challenge protein folding, such as increased temperature [5]. The activity of eukaryotic Hsp90 is further modulated by various co-chaperones, which confer substrate specificity and alter protein folding kinetics [2], [5]. Depletion of eukaryotic Hsp90 in vivo increases phenotypic variation, reveals ‘cryptic’ heritable variation, and increases penetrance of mutations [6]–[9]. Accordingly, eukaryotic Hsp90 enables organisms to maintain a stable phenotype in the face of environmental and genetic perturbation and to correctly interpret environmental stimuli.
In stark contrast, in prokarya, Hsp90 is not essential [10] and many bacterial genomes lack Hsp90 altogether [11]. Among Archaea, only very few species contain Hsp90, and those are thought to have gained Hsp90 horizontally from bacteria [11], [12]. This fragmented phylogenetic pattern likely results from multiple independent gains and losses, though phylogenetic reconstructions are confused by ancient Hsp90 paralogy [11], [12]. At the amino acid level, the Escherichia coli Hsp90 (High-temperature protein G or HtpG) is 42% identical to its human homolog, suggesting strong stabilizing selection consistent with functional conservation [13]. Indeed, E. coli Hsp90 appears to retain generic protein chaperone activity [14] and homologous Hsp90 mutations cause chaperone defects in both the prokaryotic E. coli and eukaryotic yeast [15]. However, there are no identified obligate Hsp90 co-chaperones in bacteria, adding to the uncertainty regarding the extent of its client spectrum and specificity.
To date, only three proteins have been implicated as Hsp90 clients in bacteria, with non-overlapping functions in ribosome assembly, the assembly of light-harvesting complexes, and the CRISPR/Cas immunity system [16]–[18]. Several other proteins have been shown to physically interact with the chaperone [19], [20]. Together with our knowledge of eukaryotic Hsp90 function, these data have given rise to the speculation that Hsp90 may facilitate the assembly of oligomeric protein complexes in bacteria, much like it does in eukaryotes [21]. Unlike in eukaryotes, however, further exploration of Hsp90's functional role in bacteria has proven challenging because there are no pleiotropic Hsp90-dependent phenotypes.
To address this challenge, we used a genome-scale co-evolutionary ‘guilt-by-association’ approach [22], [23] to explore the spectrum of conserved Hsp90-associated genes, functions, and organismal traits. Hsp90-associated genes tended to function in flagellar assembly, chemotaxis, and secretion. Consistent with these functions, Hsp90-associated organismal traits included the ability to inhabit multiple environments. To add to the sparse list of known bacterial Hsp90 clients, we further developed a statistical method to predict putative Hsp90 clients, which included flagellar, ribosomal, and chaperone proteins. We validated our predictions experimentally, focusing on two candidates functioning in motility and chemotaxis. Indeed, both the flagellar protein FliN and the kinase CheA were found to be Hsp90 clients in vivo. Our findings demonstrate the power of co-evolutionary inference to correctly identify substrates and functions of conserved genes like bacterial Hsp90.
Our method for inferring the function of bacterial Hsp90 is based on the analysis of its distribution across the bacterial phylogeny. However, this analysis is complicated by the existence of multiple ancient Hsp90 paralogs in bacteria. These paralogs may be older than existing phyla in bacteria [11], [12], and may have evolved distinct functions on this enormous time scale. To address this issue and to identify each paralog, we first clustered bacterial Hsp90s by sequence identity. We identified 897 bacterial Hsp90 protein sequences in the KEGG database [24] and built a neighbor-joining gene tree of bacterial Hsp90s (Figure S1A–B). We observed two well-supported long-branching clades as well as several less confident divisions in the tree (Figure S1B). These two long-branching clades contain sequences corresponding to the ‘hsp90B’ and ‘hsp90C’ paralogs that were described previously [11], [12]. All other branches correspond to ‘hsp90A’ [11], which is the largest of the Hsp90 families in bacteria (Figure S1C, Text S1). Notably, hsp90A is the lineage out of which all eukaryotic Hsp90s (excluding mitochondrial and chloroplast Hsp90s) are derived. Moreover, the E. coli gene htpG belongs to the hsp90A family, and its gene product is the best-studied bacterial Hsp90 protein. For these reasons, we restricted our analysis to hsp90A.
We set out to identify orthologous groups whose presence and absence profiles across bacterial species are associated with the presence and absence profile of hsp90A. To avoid spurious associations, any such comparative analysis must go beyond a naïve comparison of presence/absence patterns across genomes and incorporate phylogenetic information [25]. To this end, we used BayesTraits [26]–[28], a computational framework for phylogenetic analysis of character evolution. Given the states (e.g., presence/absence) of two characters across some set of species and a phylogenetic tree relating these species, BayesTraits evaluates the likelihood of various evolutionary models throughout the tree. This approach can be utilized, for example, to determine whether these two characters evolve in a mutually dependent vs. an independent fashion.
We used BayesTraits to detect associations between hsp90A and 4646 other orthologous groups in bacteria (which hereafter we shall refer to as ‘genes’ for simplicity). We used the tree constructed by Ciccarelli et al. [29] as a model phylogeny (Figure 1). In this initial analysis, we tested for any kind of dependency between hsp90A and other genes, and did not make specific assumptions about the nature of the relationship between hsp90A and the genes in question [28]. Specifically, we compared a model in which the rate of gain and loss of a given gene is independent of the rate of gain and loss of hsp90A (independent evolution) vs. a model in which the rate of gain and loss of this gene is affected by the presence or absence of hsp90A or vice-versa (co-evolution).
In total, we found 327 genes that co-evolve with hsp90A (Dataset S1). We will refer to this set as hsp90A-associated genes. These hsp90A-associated genes were significantly enriched for annotations related to the flagellum and to bacterial secretion systems (Table 1). Moreover, out of the 16 hsp90A-associated bacterial secretion genes, 10 were part of the non-flagellar Type III secretion system, suggesting that hsp90A is associated specifically with this system rather than with secretion systems in general. Using a different and markedly more extensive phylogeny [30] provided similar results (see Text S1, Table S1), as did a pruned Ciccarelli tree without the species containing the hsp90B or hsp90C (see Text S1).
The associations of hsp90A with other genes identified above are agnostic to the specific nature of the dependency between hsp90A and the gene in question. For example, our initial analysis could not distinguish between a positive association (i.e. genes tend to be gained and lost together) and a negative association (i.e. genes tend not to co-occur in genomes). Similarly, this analysis did not distinguish between genes whose gains and losses are affected by the presence of hsp90A (but that do not themselves affect hsp90A evolution) and genes that exhibit mutually dependent dynamics with hsp90A. Without a quantitative estimate of the effects that hsp90A and its co-evolving partners have upon one another, inference of Hsp90A function and its relationship with other genes is challenging.
To characterize the specific nature of the dependency between hsp90A and hsp90A-associated genes, we therefore examined rates of gain and loss inferred by BayesTraits. We focused on the two major non-overlapping hsp90A-associated functional categories, flagellar assembly and bacterial secretion. Considering, for example, fliI, a representative flagellar gene, we found that its gain and loss was strongly affected by the presence of hsp90A. Specifically, in the presence of hsp90A, fliI was often gained and rarely lost, whereas it was rarely gained and often lost when hsp90A is absent (Figure 2A). This pattern was common to all hsp90A-associated flagellar genes (Figures 2C, S2), suggesting a positive association between hsp90A and flagellar genes throughout evolution. In contrast, the co-evolutionary relationship between hsp90A and yscN, a representative nonflagellar type III secretion system gene, was markedly different, with yscN presence strongly affecting the gain and loss of hsp90A (Figure 2B). Specifically, the presence of yscN was associated with a large increase in the rates of gain and (even more dramatically) loss of hsp90A relative to these rates in its absence. Again, this pattern was common to all hsp90A-associated bacterial secretion genes (Figures 2D, S3, S4), suggesting a negative association between hsp90A and nonflagellar secretion genes throughout evolution.
To further validate the fundamentally distinct co-evolutionary dynamics of these two groups of genes, we considered four different co-evolutionary models: (1) hsp90A and the gene in question are independent (null); (2) hsp90A and the gene in question are mutually dependent; (3) hsp90A is dependent on the gene in question but not vice versa, and (4) the gene in question is dependent upon hsp90A but not vice versa (Methods). We used the Akaike Information Criterion (AIC [31]) to determine which of these 4 models best fit the co-evolutionary dynamics of each hsp90A-associated gene. As expected, none of the hsp90A-associated genes fit the independent model. Of the 27 hsp90A-associated flagellar genes, 25 were classified as being dependent on hsp90A but not vice-versa (model 4). Of the 16 hsp90A-associated secretion system genes, 10 genes were classified as mutually dependent with hsp90A (model 2; 6 of which were Type III secretion system genes), whereas 6 were classified as affecting the evolution of hsp90A (model 3). Furthermore, considering all hsp90A-associated genes, we found that genes that best fit each of the evolutionary dependency models above (models 2, 3, and 4) were enriched for different functions (Table 1). Specifically, among genes dependent on hsp90A, flagellar motility was strongly enriched, whereas among genes mutually dependent on hsp90A, secretion system components were enriched. Taken together, these patterns suggest that flagellar genes and secretion system genes had markedly different regimes of co-evolution with hsp90A.
Although many genes exhibited distinct patterns of co-evolution with hsp90A, these patterns could be the result of indirect evolutionary relationships rather than the outcome of a direct interaction with Hsp90A. We therefore aimed to predict specific genes that encode putative hsp90A clients. Our method is based on the assumption that strong, conserved clients should be heavily dependent on Hsp90A, and thus should be found only rarely in the absence of hsp90A throughout evolution. To estimate the expected frequency of each hsp90A-associated gene with and without hsp90A, we used the inferred BayesTraits rates to calculate the steady-state probabilities of each of the 4 possible two-gene presence/absence states (Methods). These probabilities represent the proportion of the time that some arbitrary bacterial lineage will spend in each of the presence/absence states throughout evolution. From these probabilities we calculated a Putative Client Index (PCI) for each hsp90A-associated gene to evaluate how often it was present without hsp90A throughout evolution, compared to a null expectation (see Methods). This index is close to zero for genes that were infrequently present without hsp90A and were hence likely to be Hsp90A clients. We defined the genes with the lowest PCI values as putative clients (Table 2; see also Text S1).
Consistent with our prior analysis, several flagellar genes behaved as potential clients (Table 2). In particular, our set of putative clients included several genes (fliH, fliI, fliN) whose products had been previously shown to physically interact with Hsp90A in E. coli [19]. The products of these genes are cytoplasmic components of the flagellar rotor and export apparatuses. In contrast, nonflagellar type III secretion genes were all absent from the list of potential clients. In fact, nonflagellar type III secretion system components were rated as some of the least likely clients by our index (Figure 3). This disparity in predicted client status mirrors the different evolutionary relationships of these complexes with hsp90A (Figure 2).
Chaperone/proteases (e.g. ClpA and PpiD) also ranked high in our list of potential clients. Hsp90A is known to collaborate with other chaperone systems such as DnaK [14], [32] but to date no obligate co-chaperones have been described. The identified chaperone/proteases may represent such co-chaperones or collaborating chaperone systems, since our index cannot discriminate between Hsp90 clients and Hsp90 co-chaperones (or other collaborating proteins). Alternatively, these observed associations could simply indicate that components of the cytoplasmic stress response are dependent upon Hsp90A.
We also found several unexpected putative clients, such as the 3-hydroxybutyryl-CoA dehydrogenase PaaH and the transcription termination factor Rho, which we predict to be the two strongest clients. Further study will be necessary to understand these associations and the underlying cause of the co-evolutionary association between these genes and hsp90A.
Our putative clients and the predicted chaperone role of Hsp90A in flagellar assembly are consistent with previous observations. Specifically, the deletion of E. coli hsp90A, also known as htpG, resulted in reduced surface swarming movement [33]. We also previously observed physical interactions between the HtpG protein and certain flagellar proteins [19]. Yet, these observations lacked a clear demonstration of client status or mechanism, and E. coli swarming is a complex behavior that depends on numerous factors in addition to flagellar function [34]. We therefore set out to test our hypothesis that Hsp90A is physiologically important for flagellar assembly and function and that flagellar components are indeed Hsp90A clients.
We examined the swimming motility phenotype of ΔhtpG E. coli strains on soft-agar plates (Methods). In contrast to surface swarming, swimming is a less complex behavior, in which bacteria use functional flagella and chemotaxis components to swim from an inoculation point through agar pores, following nutrient gradients that are created by nutrient depletion within the colony. The soft-agar assay is routinely used to assay bacterial swimming motility and chemotaxis. To enhance our ability to detect differences between wild-type and ΔhtpG cells, the assays were performed competitively. Competitive assays emphasize small differences between strains and reduce experimental error, thereby increasing the sensitivity of the assay. After mixing equal amounts of YFP-labeled WT and CFP-labeled ΔhtpG strains, this mixture was inoculated in the center of a soft-agar plate and incubated at 34°C for 8 hrs. We then counted cells of each strain in the plate center vs. the outer edge using fluorescence microscopy (Figure 4A). ΔhtpG mutants migrated less efficiently to the plate's outer edge relative to WT, confirming that they are partially deficient in their motility and/or chemotaxis (Figure 4B). This defect is apparently subtle, since little difference between WT and ΔhtpG cells was observed in a non-competitive assay (Figure S5), but it could be revealed due to strong selection for cells with optimal motility and chemotaxis at the outer edge of the spreading bacterial population.
We also tested the phenotype of the HtpG(E34A) mutant, which has reduced rates of ATP hydrolysis and is deficient in substrate refolding [14], [35]. Since HtpG ATPase activity is necessary for release of clients, HtpG(E34A) is less efficient at releasing clients [36]–[38]. Indeed, this mutant showed stronger motility/chemotaxis defects than the ΔhtpG strain (Figure S5), presumably due to sequestration of its client proteins. We therefore employed the HtpG(E34A) mutant in all subsequent assays as a more sensitive test of HtpG involvement. Taken together, our observations suggest that the motility defect may be due to the improper function or sequestration of HtpG clients.
To further investigate the in vivo interaction of HtpG with flagellar components, we used htpG-yfp and htpG(E34A)-yfp constructs expressed in WT cells to perform acceptor photobleaching FRET between HtpG and FliN-CFP over an E. coli growth curve. Motility of E. coli is known to increase at the transition from the early exponential to post-exponential phase of growth [39], and this experimental design enabled us to examine the HtpG-FliN interaction in the context of the flagellar assembly process. If HtpG is indeed involved in the assembly process of these structures, the interaction of HtpG with FliN should correspond temporally to the timing of flagellar assembly. Indeed, we found that the interaction with FliN peaked at OD600 = 0.2 (Figure 5A) and correlated well with the onset of cell motility in wild-type cells (Figure 5B). Moreover, the interaction of HtpG(E34A) with FliN was stronger and delayed compared to the binding of wild-type HtpG. Correspondingly, the onset of motility was delayed in cells expressing HtpG(E34A) (Figure 5B). This is consistent with the delayed release of clients by HtpG(E34A), suggesting that HtpG's role in motility derives from a direct involvement in flagellar complex assembly.
Given that both bacterial and eukaryotic Hsp90s are known to collaborate with Hsp70 in refolding proteins [14], [40]–[42], we considered the possibility that this was also the case for bacterial flagellar assembly. We previously showed that some flagellar motor components interact with DnaK, the E. coli Hsp70 homolog [19]. Therefore, we repeated the FRET experiments testing for interactions between HtpG or HtpG(E34A) and FliN in a ΔcbpAΔdnaJ background. CbpA and DnaJ are DnaK co-chaperones and are essential for DnaK–dependent refolding activity [14]. DnaK should not be able to pass substrates to HtpG in this mutant background. Indeed, we found that FRET interactions with FliN disappear for both HtpG proteins in this background (Figure S6A), suggesting that DnaK-dependent remodeling precedes HtpG action in flagellar complex assembly.
Since a recent high-throughput assay showed kinases to be overrepresented among eukaryotic Hsp90 clients [43], [44], we next examined whether the HtpG-dependent defects in chemotaxis may also be due to defective chemoreceptor kinase activity. Although no chemotaxis proteins were found in our list of the strongest putative clients, we did observe a significant enrichment of these components in the hsp90A-associated set (Table 1). We thus tested interactions between six chemoreceptor cluster components and HtpG(E34A) using, as before, acceptor photobleaching FRET (Table S4). We observed a strong interaction of HtpG(E34A) with the chemoreceptor kinase CheA. Our results suggest that the FliN/HtpG and CheA/HtpG interactions are direct and do not depend on other flagellar or chemotaxis proteins, since these interactions are robust to deletion of flhC, which ablates expression of all endogenous flagellar and chemotaxis genes (Table S4) [19]. Moreover, the CheA dimerization domain was required for association with HtpG, supporting the hypothesis that HtpG aids oligomerization of its clients [17], [45]. Testing HtpG interactions with other chemotaxis proteins of E. coli revealed an additional strong interaction with the dimeric phosphatase CheZ but not with other proteins (Table S4).
We again examined the temporal dynamics of these interactions. Due to the hierarchical order of flagellar and chemotaxis gene expression [39], [46], the assembly of chemoreceptor clusters is delayed compared to the assembly of flagellar motors as non-motile cells transition into motile cells. Indeed, the interaction of HtpG with CheA peaked at OD600 = 0.3, after the FliN peak (Figure 5A). Just as for FliN, the interaction of HtpG(E34A) with CheA was stronger and delayed compared to wild-type HtpG, and the HtpG-CheA interaction disappeared in a ΔcbpAΔdnaJ background (Figure S6B). Collectively, these findings suggest that HtpG plays an important role in the assembly of both the flagellar motor and chemoreceptor clusters through separate client interactions.
Given the role of HtpG in chaperoning proteins that mediate interactions with the environment, and the known role of eukaryotic Hsp90 in phenotypic robustness, we finally examined whether hsp90A directly co-evolved with certain bacterial organismal traits. We considered several organismal traits, including aerobism, thermophilicity, halophilicity, the ability to form endospores, pathogenicity, motility, and habitat preferences (see Methods). We used BayesTraits and the Ciccarelli tree to identify traits that co-evolve with hsp90A. Out of the 11 analyzed traits, 4 exhibited significant associations with hsp90A (p<0.05; Table S5), with the strongest association observed between hsp90A and the capacity to inhabit multiple habitats. Moreover, examining the gain and loss rates obtained, we found that hsp90A is gained and lost at significantly higher rates in organisms that inhabit multiple habitats (with no gains inferred in single habitat organisms), suggesting that a preference for multiple habitats imposes a different selection regime on hsp90A (Figure 6). We also tested whether the co-evolutionary dependency between hsp90A and multiple-habitat preferences was unidirectional, as we observed for some hsp90A-associated genes. Comparing the four co-evolutionary models described above and applying AIC to identify the best-fitting model, we found that hsp90A gain and loss depended on habitat preference, but not vice versa. This observation suggests that in organisms inhabiting multiple environments hsp90A is subjected to dynamically shifting selective pressures, potentially alternating between selection for and against hsp90A.
We set out to discover Hsp90 functions conserved throughout the bacterial tree of life. We found that hsp90A, the most common paralog of bacterial Hsp90, bore strong signatures of co-evolution with several hundred genes and with specific life history traits, shedding light on its function and impact on evolutionary history. Most notably, we found that hsp90A co-evolved with membrane protein complexes such as flagella and other Type III secretion (T3S) systems. Our results suggest that Hsp90's role in sensing and responding to environmental stimuli is conserved between bacteria and eukaryotes.
Similar to verified eukaryotic Hsp90 clients [5], our predicted putative Hsp90A clients were a diverse group of proteins (e.g. the flagella protein FliN, the chaperone ClpA, and the ribosomal protein RluB; see Table 2) that tended to belong to specific functional categories (e.g. flagellar proteins, chaperones, and ribosomal components). As our methods can only infer associations between genes that are frequently gained and lost, we may substantially underestimate the number of hsp90A-associated genes and clients. However, the non-essentiality and frequent loss of hsp90A throughout bacterial diversity argues that genes not captured in our analysis (since they are not frequently gained and lost) are unlikely to be strongly dependent on the chaperone throughout bacteria.
The subtlety of the bacterial Hsp90 mutant phenotypes that we (and others) report implies that Hsp90's role in cellular physiology has diverged between eukaryotes and prokaryotes [17], [45], [47]. In other words, either essential pieces of cellular physiology changed, or Hsp90 function changed. We favor the first hypothesis, because Hsp90 is well-conserved among bacteria, archaea, and humans at the sequence level [13], and retains a similar quaternary structure [48] and biochemical activity [15], [37], [44]. In contrast, bacterial and archaeal cells differ significantly from eukaryotic cells. Eukaryotic cells have higher cell compartmentalization, longer and multifunctional proteins with multiple domains [49], and increased protein interactome complexity [50]. Together with the existence of many eukaryotic Hsp90 co-chaperones, all these features may contribute to the greater essentiality of Hsp90 in eukaryotes.
The dependence of HtpG-client interactions upon the DnaK chaperone system, as observed by us and by others [14], [15], argues that Hsp90A is well-integrated with other chaperone systems. Our putative clients included ClpA, the substrate adaptor for the ClpAP/ClpAXP chaperone/protease complexes, and PpiD, a periplasmic chaperone [51]. Like HtpG, PpiD is necessary for optimal swarming motility [33], suggesting that it may participate in flagellar assembly. We speculate that these proteins act as Hsp90A co-chaperones in some bacteria; alternatively, their dependence on Hsp90A may represent an example of collaborating chaperone systems.
The best-characterized Hsp90 client in bacteria is the structural ribosomal protein L2 [15], [18], which is near-universally conserved throughout life (and hence not detectable by our method). In addition to L2, other ribosomal proteins were found to interact with HtpG in large-scale proteomics analyses. In agreement with these observations, we found the ribosomal proteins RlmE and RluB among the predicted hsp90A clients.
Although these chaperone and ribosomal proteins were predicted to be stronger clients than flagellar proteins, our experimental validation focused on the latter as their client status was suggested by previous observations [19], [33]. We present four lines of evidence for HtpG client status for the flagellar protein FliN and the chemoreceptor kinase CheA, including direct interactions with HtpG, physiologically relevant timing of HtpG-FliN/CheA interactions, phenotypic consequences of reduced HtpG function in CheA/FliN-dependent traits, and dependence of CheA/FliN interactions with HtpG upon the Hsp40-Hsp70 pathway. The identification of FliN and CheA as HtpG clients is consistent with the hypothesis that bacterial Hsp90 facilitates the assembly of large membrane-associated protein complexes [17], [45].
Curiously, whereas the flagellar T3S system contained Hsp90A clients, the nonflagellar T3S system is predicted to have an antagonistic relationship with Hsp90A. Nonflagellar T3S systems and the flagellar T3S systems are closely related (NF-T3SS and F-T3SS) [52], [53]. 9 NF-T3SS components are directly homologous to flagellar components, of which 8 were found to co-evolve with hsp90A in our analysis. Yet, these 8 genes are predicted to co-evolve antagonistically with hsp90A (Figure 3), whereas their flagellar homologs are mostly predicted to be clients (for instance, the fliI and yscN genes shown in Figure 2 are homologous). This result suggests that some relationship with Hsp90A is conserved between the two T3S systems, but with apparently opposite effects in each system. This result may reflect the fact that each of these systems is an adaptation to different ecological challenges. Specifically, we have shown that Hsp90A is important for flagellum-enabled motility and chemotaxis in E. coli. This mode of motility is strongly adaptive in certain physical environments [34], [54], [55], and thus Hsp90A is likely to be associated with fitness in these environments through flagellar assembly. The presence of NF-T3SS is likewise an adaptation to certain biotic environments [55], [56]. Our observation that organisms inhabiting multiple habitats experience fluctuating selection for hsp90A is also consistent with competing selection pressures. Representative genes of these homologous T3S families were not significantly associated with habitat preferences, arguing that hsp90A's association with habitat preferences is not a byproduct of associations with T3S systems. Nonetheless, we suggest that these two T3S systems constitute a link between Hsp90A and phenotypic robustness across different environments.
Inferring function from evolutionary associations has some caveats. For instance, F-T3S systems can be found in genomes that lack hsp90A. If F-T3S systems include Hsp90A clients, then what may render Hsp90A-dependent stabilization dispensable in some bacteria? Experimental validation will be necessary to answer such questions, and to distinguish true client relationships from indirect co-evolutionary associations. As discussed before, our method is subject to gene set bias, in that only genes that are gained and/or lost frequently will have enough statistical power to reject the null hypothesis. Similarly, as our method assumes that relationships are maintained throughout the analyzed phylogeny, we cannot reliably detect genes that are associated with hsp90A in some organisms but not in others.
Although much work remains to articulate the precise mechanistic relationships between hsp90A and its co-evolving genes, our results highlight the tremendous potential of evolutionary inference for guiding experimental research. More generally, our study provides a successful example of how evolutionary perspectives and phylogenetic analyses can inform and advance the study of complex biological systems and the inference of elusive biological functions.
We downloaded all Hsp90 amino acid sequences (including all paralogs) for bacteria with full KEGG genome annotations from the KEGG database [24], [57]. We aligned these sequences using ClustalO [58], and used the PHYLIP package [59] to construct neighbor-joining trees and assess their phylogenetic support through bootstrapping. We assigned Hsp90 families to branches according to bootstrap support for the branch and previous classifications [11], [12].
We acquired presence/absence patterns of genes across organisms from the KEGG database release 60.0 (in the form of KEGG Orthology/KO profiles) [57], and functional annotations from KEGG Class. Genes that were either present in fewer than five species or absent in fewer than five species in the tree of interest were dropped from our analysis, as these genes are unlikely to show meaningful signatures of co-evolution by this method.
We obtained the tree constructed by Ciccarelli et al. (Ciccarelli tree) [29] and pruned it to 148 bacterial species for which KEGG genome data was available. We also obtained the LTP104 version of the 16S/23S rRNA tree from the All-Species Living Tree Project (Yarza tree) [30], [60]. We used ARB [61] to prune this tree to bacterial species for which KEGG genome data was available. We further pruned this tree to omit clades placed paraphyletically at the taxonomic levels of phylum, class, order, and family. This filtered tree included 797 bacterial species. As BayesTraits cannot process trees with zero-length branches, all branch lengths equal to zero were replaced with a negligible branch length (0.00001, approximately an order of magnitude smaller than the next smallest branch length in each tree).
We acquired organismal trait data from the NCBI Entrez genome project, November 2011 [62]. We recoded all traits into presence/absence patterns for the trait in question. For instance, an organism found to be pathogenic towards any other organism was coded as ‘1’ for the trait of pathogenicity, whereas an annotated organism that was never found to be pathogenic was coded as ‘0’. Similarly, we coded both thermophilic and hyperthermophilic organisms as ‘1’ for the trait of thermophilicity, whereas all other annotated organisms were coded as ‘0’; anaerobic organisms were coded as ‘0’ for the trait of aerobicity, whereas all other annotated organisms were recoded as ‘1’. We define as inhabiting multiple habitats any organism that inhabits more than one of NCBI's habitat categories. For BayesTraits analysis, the tree was pruned to include only species annotated for the trait in question (each trait analysis was accordingly performed on a slightly different set of species; see Table S5 for details on species number for each analysis).
A complete description of the BayesTraits (v1.0) framework can be found elsewhere [26]. Briefly, consider a character with 2 states, 0 and 1. If a species has 2 such distinct characters, it can occupy 4 possible states: 1:(0,0), 2:(0,1), 3:(1,0), and 4:(1,1). Specifically, if these 2 characters represent the presence or absence of two genes, hsp90A and gene X, these four states correspond to (hsp90A−, X−), (hsp90A+, X−), (hsp90A−, X+), and (hsp90A+, X+). Evolution is then the process by which these genes are gained and lost over time. Consider accordingly an evolutionary process where only one character can change state at a time. Such a process can then be described by 8 parameters for the rates of transition per unit time between these 4 states: Q = [q12, q13, q21, q31, q24, q34, q42, q43], where qxy is the rate of transition from state x to state y. BayesTraits implements this model of evolution as a continuous-time Markov process and estimates each of these rate parameters by maximum-likelihood (ML). We further validated that these ML-based rates are consistent with reversible-jump Markov chain Monte Carlo-derived estimates (Methods; Text S1). This estimation is based on a phylogeny and on the states of the two characters at the tips of the phylogeny. Having estimated these rates, BayesTraits additionally calculates the likelihood of the model based on the character states at the tips of the phylogeny.
We can further compare different models of evolution by forcing certain parameters to be equal. We specifically considered the following 4 models:
We used discrete from the BayesTraits package [26]–[28] to infer associations between hsp90A and other bacterial genes and between hsp90A and various organismal traits. We first tested for an evolutionary association with hsp90A by comparing model 1 to model 2 above with a likelihood ratio test (LRT), as previously described [28]. In our likelihood-ratio tests, the 2Log(LR) approximates a χ2 test statistic for rejecting the independent model as a null hypothesis, and is calculated as twice the difference of the log-likelihoods of a co-evolutionary model and a model of evolutionary independence. The set of genes for which model 2 is preferred (i.e., model 1 is rejected as a null hypothesis) have an evolutionary association with hsp90A. Since different runs of the BayesTraits maximum likelihood method can potentially produce different parameter values, we repeated this procedure 100 times, each potentially resulting in a different gene set. We validated that these sets are similar and the choice of gene set does not substantially affect downstream analysis (Text S1). Any gene that was found to be associated with hsp90A in at least 90 runs was defined as hsp90A-associated gene. See Text S1 for more details.
We selected 10 genes at random from the hsp90A-associated set and used the BayesTraits implementation of reversible-jump Markov chain Monte Carlo to estimate the rate parameters for their gain and loss in concert with hsp90A [63]. For each of these 10 genes, we used an exponential rate prior with mean and variance equal to 30, and ran the chain for 150 million iterations while sampling every 100 iterations. We discarded the first 75 million iterations as burn-in and used the remaining iterations as a posterior distribution of rate parameter estimates. We used Tracer v1.5 [64] and previously described criteria to evaluate chain convergence in this remaining sample [65]. For each rate, we used the median of its posterior distribution in this sample as a point estimate.
To provide an accurate description of the co-evolutionary dynamics of hsp90A-associated genes, we further applied BayesTraits to these genes, estimating the likelihood of each of the four models described above. We identified the best fit model for each gene using the Akaike Information Criterion (AIC) [31], taking into account both the likelihood score and the number of parameters in each model. We again repeated this procedure 100 times and classified a gene into a specific co-evolutionary model only if it fit this same model in at least 90 runs (see Text S1 for more details). This two stage scheme, first identifying associated genes and then selecting a model that best describes their evolutionary relationship with hsp90A, provides a more stringent test of co-evolution and supports a simple approach for multiple testing correction.
We used BayesTraits-derived evolutionary transition rates under the fully unrestricted model to estimate residence times in specific states (for instance, the proportion of time spent by bacteria in a state where both hsp90A and some other gene are present, vs. the time when only the other gene is present) under steady state dynamics. For a given gene, the probability of being in one of the four states, A: (hsp90A absent, Gene absent), B: (hsp90A present, Gene absent), C: (hsp90A absent, Gene present), D: (hsp90A present, Gene present) at a very small increment of time Δt after time t is given by:We can differentiate this to obtain the instantaneous change in each probability:At steady state dA/dt = 0, dB/dt = 0, etc., and therefore:This set of linear equations can be solved for A, B, C, and D, with the requirement that A+B+C+D = 1. We replaced 0 rates with the smallest nonzero rate in the model multiplied by 0.001 to allow transitions between all states. The positive nonzero solution for A, B, C, and D can then be conceived as the expected residence times along some arbitrary bacterial lineage. We used these residence times to estimate a Putative Client Index, PCI, denoting the normalized residence time in state C:Notably, if Hsp90A and the gene's product have no client relationship, the proportion of time spent in state C is expected to be equal to (C+D)(A+C), so a PCI close to 1 indicates that the observation does not differ from the expectation. Smaller values of PCI therefore indicate that a gene is observed less frequently than expected without hsp90A, and is thus more likely to be a client. Since no obvious threshold value can be defined, we considered the 20 genes with the lowest PCI values as putative clients (Figure 3 and Table 2; Methods). To account for variation in rates between BayesTraits runs we repeated this procedure 100 times and defined as putative clients those that were identified as clients in at least 90 of these runs (see Text S1). PCI scores shown in Table 2 and Figure 3 are averages across all runs.
We used a hypergeometric test to assess whether each KEGG Class functional annotation is overrepresented in the various Hsp90-associated gene sets. As a background set in each case we used the entire set of genes analyzed. Any annotation present in less than 4 copies in the background set was not considered. We accepted enrichments at a 5% FDR.
Escherichia coli K-12 strains and plasmids used in this study are listed in Table S2. Cells were grown in tryptone broth (TB; 1% tryptone and 0.5% NaCl) and when necessary supplemented with ampicillin, chloramphenicol and/or kanamycin at final concentrations of 100, 35 and 50 µg/ml, respectively. Overnight cultures, grown at 30°C, were diluted 1∶100 and grown at 34°C for about 4 h, to an OD600 of 0.45–0.5. All expression constructs for YFP and CFP fusions were constructed as described previously [19], [66], [67]. Induction levels for protein expression were 1 µM IPTG (pHL24, pHL35, pVS129 and pVS132), 20 µM IPTG (pVS64 and pVS99), 25 µM IPTG (pDK36, pDK90 and pDK91), 50 µM IPTG (pDK19 and pVS18), 0.005% arabinose (pHL13, pVS108 and pVS109) and 0.01% arabinose (pHL52, pHL70, pDK14, pDK29, pDK30 and pDK49). Cells were harvested by centrifugation (4,000 rpm, 5 min), washed once with tethering buffer (10 mM potassium phosphate, 0.1 mM EDTA, 1 mM L-methionine, 67 mM sodium chloride, 10 mM sodium lactate, pH 7) and resuspended in 10 mL tethering buffer prior to FRET measurements.
TB soft agar plates were prepared by supplementing TB with 0.3% agar (Applichem) and when necessary with 100 g/mL ampicillin and 1 µM IPTG. Equal amounts of cells from different overnight cultures, adjusted depending on their optical density to the equivalent of 2.5 µl of culture with OD600 of 2.0, were inoculated and allowed to spread at indicated temperatures for indicated times. Following incubation, photographs of plates were taken with a Canon EOS 300D (DS6041) camera. Images were analyzed with ImageJ (Wayne Rasband, NIH, http://rsb.info.nih.gov/ij/) to determine the diameter of the rings of spreading colonies.
For analysis of motility at different growth stages (indicated by OD600 value), percentages of motile cells were estimated from the microscopy movies of swimming cells. The experiment was performed with the RP437 strain, which is non-motile above 37°C. Cells were grown overnight in TB medium at 37°C to completely inhibit their motility. After dilution in fresh TB medium to OD600 0.01, cells were grown at 34°C for measurements.
For microscopy, cells were taken from the soft-agar plates and applied to a thin agarose pad (1% agarose in tethering buffer). Fluorescence imaging was performed on a Zeiss AxioImager microscope equipped with an ORCA AG CCD camera (Hamamatsu), a 100× NA 1.45 objective, and HE YFP (Excitation BP 500/25; Emission BP 535/30) and HE CFP (Excitation BP 436/25; Emission BP 480/40) filter sets. Each imaging experiment was performed in duplicate on independent cultures. All images were acquired under identical conditions. Images were subsequently analysed using ImageJ software.
FRET measurements by acceptor photobleaching were performed on a custom-modified Zeiss Axiovert 200 microscope as described before [66]. Briefly, cells expressing YFP and CFP fusions of interest were concentrated about tenfold by centrifugation, resuspended in tethering buffer and applied to a thin agarose pad (1% agarose in tethering buffer). Excitation light from a 75 XBO lamp, attenuated by a ND60 (0.2) neutral-density filter, passed through a band-pass (BP) 436/20 filter and a 495DCSP dichroic mirror and was reflected on the specimen by a Z440/532 dual-band beamsplitter (transmission 465–500 and 550–640 nm; reflection 425–445 and 532 nm). Bleaching of YFP was accomplished by a 20 sec illumination with a 532 nm diode laser (Rapp OptoElectronic), reflected by the 495DCSP dichroic mirror into the light path. Emission from the field of view, which was narrowed with a diaphragm to the area bleached by the laser, passed through a BP 485/40 filter onto a H7421-40 photon counter (Hamamatsu). For each measurement point, photons were counted over 0.5 s using a counter function of the PCI-6034E board, controlled by a custom-written LabView 7.1 program (both from National Instruments). CFP emission was recorded before and after bleaching of YFP, and FRET was calculated as the CFP signal increase divided by the total signal after bleaching. ΔflhC strains were used to define direct interactions between HtpG and flagellar and chemotaxis components. In this background expression of endogenous flagellar and chemotaxis genes is inhibited, thus eliminating indirect interactions that may result from concomitant binding of HtpG and tested protein to a third flagellar or chemotaxis protein.
|
10.1371/journal.pcbi.1002711 | Coding and Decoding with Adapting Neurons: A Population Approach to the Peri-Stimulus Time Histogram | The response of a neuron to a time-dependent stimulus, as measured in a Peri-Stimulus-Time-Histogram (PSTH), exhibits an intricate temporal structure that reflects potential temporal coding principles. Here we analyze the encoding and decoding of PSTHs for spiking neurons with arbitrary refractoriness and adaptation. As a modeling framework, we use the spike response model, also known as the generalized linear neuron model. Because of refractoriness, the effect of the most recent spike on the spiking probability a few milliseconds later is very strong. The influence of the last spike needs therefore to be described with high precision, while the rest of the neuronal spiking history merely introduces an average self-inhibition or adaptation that depends on the expected number of past spikes but not on the exact spike timings. Based on these insights, we derive a ‘quasi-renewal equation’ which is shown to yield an excellent description of the firing rate of adapting neurons. We explore the domain of validity of the quasi-renewal equation and compare it with other rate equations for populations of spiking neurons. The problem of decoding the stimulus from the population response (or PSTH) is addressed analogously. We find that for small levels of activity and weak adaptation, a simple accumulator of the past activity is sufficient to decode the original input, but when refractory effects become large decoding becomes a non-linear function of the past activity. The results presented here can be applied to the mean-field analysis of coupled neuron networks, but also to arbitrary point processes with negative self-interaction.
| How can information be encoded and decoded in populations of adapting neurons? A quantitative answer to this question requires a mathematical expression relating neuronal activity to the external stimulus, and, conversely, stimulus to neuronal activity. Although widely used equations and models exist for the special problem of relating external stimulus to the action potentials of a single neuron, the analogous problem of relating the external stimulus to the activity of a population has proven more difficult. There is a bothersome gap between the dynamics of single adapting neurons and the dynamics of populations. Moreover, if we ignore the single neurons and describe directly the population dynamics, we are faced with the ambiguity of the adapting neural code. The neural code of adapting populations is ambiguous because it is possible to observe a range of population activities in response to a given instantaneous input. Somehow the ambiguity is resolved by the knowledge of the population history, but how precisely? In this article we use approximation methods to provide mathematical expressions that describe the encoding and decoding of external stimuli in adapting populations. The theory presented here helps to bridge the gap between the dynamics of single neurons and that of populations.
| Encoding and decoding of information with populations of neurons is a fundamental question of computational neuroscience [1]–[3]. A time-varying stimulus can be encoded in the active fraction of a population of neurons, a coding procedure that we will refer to as population coding (Fig. 1). Given the need for fast processing of information by the brain [4], population coding is an efficient way to encode information by averaging across a pool of noisy neurons [5], [6] and is likely to be used at least to some degree by the nervous system [7]. For a population of identical neurons, the instantaneous population firing rate is proportional to the Peri-Stimulus Time Histogram (PSTH) of a single neuron driven repeatedly by the same stimulus over many trials.
When driven by a step change in the input, the population of neurons coding for this stimulus responds first strongly but then adapts to the stimulus. To cite a few examples, the activity of auditory nerve fibers adapt to pure tones [8], cells in the retina and the visual cortex adapt to contrast [9], [10] and neurons in the inferior temporal cortex adapt to higher order structures of images [11]. Adaptation is energy-efficient [12] but leads to a potentially ambiguous code because adapting responses generate a population activity which does not directly reflect the momentary strength of the stimuli [13]. Putting the case of sensory illusions aside, the fact that our perception of constant stimuli does not fade away indicates that the adapting responses can be efficiently decoded by the brain areas further down the processing stream. In fact, illusions such as the motion after-effect are believed to reflect errors in decoding the activity of neuronal populations [14]. But what is the correct rule to decode population activity? What elements of the population history are relevant? What are the basic principles?
Synapse- and network-specific mechanisms merge with intrinsic neuronal properties to produce an adapting population response. Here we focus on the intrinsic mechanisms, commonly called spike-frequency adaptation. Spike-frequency adaptation appears in practically all neuron types of the nervous system [15]. Biophysical processes that can mediate spike-frequency adaptation include spike-triggered activation/inactivation of ion-channels [16]–[18] and a spike-triggered increase in the firing threshold [19]–[22]. Neurons adapt a little more each time they emit a spike, and it is the cumulative effect of all previous spikes that sets the level of adaptation. The effect of a single spike on future spiking probability cannot be summarized by a single time constant. Rather, the spike-triggered adaptation unfolds on multiple time scales and varies strongly across cell-types [22], [23].
Mean-field methods were used to describe: attractors [24]–[28], rapid-responses [6], [29] and signal propagation [30]. While adaptation is important to correctly predict the activity of single neurons [22], [31]–[33], it is difficult to include it in mean-field methods. A theory relating spike-frequency adaptation to population dynamics should be general enough to encompass a variety of different spike-triggered adaptation profiles, as observed in experiments. In the literature we find six main approaches to the population coding problem. The first and most simple one formulates the rate of a neuronal population (or the time-dependent rate in a PSTH) as a linear function of the stimulus. This phenomenological model relates to the concept of receptive fields [34] and can be made quantitative using a Wiener expansion [35]. Yet, early experimental tests showed that linear filtering must be complemented with a non-linear function [35], [36]. The linear-non-linear model can thus be considered as the second approach to population coding. In combination with a Poisson spike generator it is called the LNP model for Linear-Nonlinear-Poisson. It makes accurate predictions of experimental measurements for stationary stimulus ensembles, but fails when the stimulus switches either its first or second order statistics. Neural refractoriness is in part responsible for effects not taken into account in this linear-nonlinear model [37]–[40]. In a third approach proposed by Wilson and Cowan [41] the population activity is the solution to a non-linear differential equation. Unfortunately this equation has only a heuristic link to the underlying neuronal dynamics and cannot account for rapid transients in the population response. The fourth approach formulates the population activity in terms of an integral equation [6], [41], [42] which can be interpreted as a (time-dependent) renewal theory. While this renewal theory takes into account refractoriness (i.e. the effect of the most recent spike) and captures the rapid transients of the population response and PSTH, neither this one nor any of the other encoding frameworks mentioned above consider adaptive effects. To include adaptation into previously non-adaptive models, a common approach is to modify the effective input by rescaling the external input with a function that depends on the mean neuronal firing rate in the past [15], [43], [44]. This forms the fifth method. For example, Benda and Herz [15] suggested a phenomenological framework in which the linear-non-linear approach is modified as a function of the past activity while Rauch et al. [43] calculated the effective rate in integrate-and-fire neurons endowed with a frequency-dependent modification of the input current. Finally, there is also a sixth method to determine the population activity of adapting populations. Inspired by the Fokker-Planck approach for integrate-and-fire neurons [27], this last approach finds the population activity by evolving probability distributions of one or several state variables [45]–[49]. Isolating the population activity then involves solving a non-linear system of partial differential equations.
The results described in the present article are based on two principal insights. The first one is that adaptation reduces the effect of the stimulus primarily as a function of the expected number of spikes in the recent history and only secondarily as a function of the higher moments of the spiking history such as spike-spike correlations. We derive such an expansion of the history moments from the single neuron parameters. The second insight is that the effects of the refractory period are well captured by renewal theory and can be superimposed on the effects of adaptation.
The article is organized as follows: after a description of the population dynamics, we derive a mathematical expression that predicts the momentary value of the population activity from current and past values of the input. Then, we verify that the resulting encoding framework accurately describes the response to input steps. We also study the accuracy of the encoding framework in response to fluctuating stimuli and analyze the problem of decoding. Finally, we compare with simpler theories such as renewal theory and a truncated expansion of the past history moments.
To keep the discussion transparent, we focus on a population of unconnected neurons. Our results can be generalized to coupled populations using standard theoretical methods [3], [6], [27].
How does a population of adapting neurons encode a given stimulating current ? Each neuron in the population will produce a spike train, denoted by , such that the population can be said to respond with a set of spike trains. Using the population approach, we want to know how the stimulus is reflected in the fraction of neurons that are active at time , that is, the population activity (Fig. 1). The population activity (or instantaneous rate of the population) is a biologically relevant quantity in the sense that a post-synaptic neuron further down the processing stream receives inputs from a whole population of presynaptic neurons and is therefore at each moment in time driven by the spike arrivals summed over the presynaptic population, i. e. the presynaptic population activity.
Mathematically, we consider a set of spike trains in which spikes are represented by Dirac-pulses centered on the spike time : [3]. The population activity is defined as the expected proportion of active neurons within an infinitesimal time interval. It corresponds, in the limit of a large population and small time interval, to the number of active neurons in the time interval divided by the total number of neurons and the time interval [3]:(1)The angular brackets denote the expected value over an ensemble of identical neurons. Experimentally, the population activity is estimated on a finite time interval and for a finite population. Equivalently the population activity can be considered as an average over independent presentations of a stimulus in only one neuron. In this sense, the population activity is equivalent to both the time-dependent firing intensity and the Peri-Stimulus Time Histogram (PSTH).
Since the population activity represents the instantaneous firing probability, it is different from the conditional firing intensity, , which further depends on the precise spiking history, or past spike train . Suppose we have observed a single neuron for a long time (e.g. 10 seconds). During that time we have recorded its time dependent input current and observed its firing times . Knowing the firing history for and the time-dependent driving current for , the variable describes the instantaneous rate of the neuron to fire again at time . Intuitively, reflects a likelihood to spike at time for a neuron having a specific history while is the firing rate at time averaged on all possible histories (see Methods):(2)
Ideally, one could hope to estimate directly from the data. However, given the dimensionality of and , model-free estimation is not feasible. Instead we use the Spike Response Model (SRM; [6], [50]–[52]), which is an example of a Generalized Linear Model [53], in order to parametrize , but other parametrizations outside the exponential family are also possible. In particular, can also be defined for nonlinear neuron models with diffusive noise in the input, even though explicit expressions are not available. The validity of the SRM as a model of neuronal spike generation has been verified for various neuron types and various experimental protocols [22], [31], [32]. In the SRM, the conditional firing intensity increases with the effective input :(3)where is the total driving force of the neuron:(4)where ‘’ denotes the convolution, is the input current convolved with the membrane filter, encodes the effect of each spike on the probability of spiking, is a scaling constant related to the instantaneous rate at the threshold with units of inverse time (see Methods for model parameters). The link-function can take different shapes depending on the noise process [3]. Here we will use an exponential link-function since it was shown to fit the noisy adaptive-exponential-integrate-and-fire model [54] as well as experimental data [22], [32], [55]. The exponential link-function: corresponds to after absorbing the scaling parameter in the constant and and in the functions and to make these unit-free.
To see that the function can implement both adaptation and refractoriness, let us first distinguish these processes conceptually. The characteristic signature of refractoriness is that the interspike interval distribution for constant input is zero or close to zero for very short intervals (e.g. one millisecond) - and in the following we use this characteristic signature as a definition of refractoriness. With this definition, a Hodgkin-Huxley model (with or without noise) or a leaky integrate-and-fire model (with or without diffusive noise) are refractory, whereas a Linear-Nonlinear-Poisson Model is not. In fact, every neuron model with intrinsic dynamics exhibits refractoriness, but Poissonian models do not.
While refractoriness refers to the interspike-interval distribution and therefore to the dependence upon the most recent spike, adaptation refers to the effect of multiple spikes. Adaptation is most clearly observed as a successive increase of interspike intervals in response to a step current. In contrast, a renewal model [56], where interspike intervals are independent of each other, does not exhibit adaptation (but does exhibit refractoriness). Similarly, a leaky or exponential integrate-and-fire model with diffusive noise does not show adaptation. A Hodgkin-Huxley model with the original set of parameters exhibits very little adaptation, but addition of a slow ion current induces adaptation.
Conceptually, contributions of multiple spikes must accumulate to generate spike frequency adaptation. In the Spike Response Model, this accumulation is written as a convolution: . If for and vanishes elsewhere, the model exhibits absolute refractoriness of duration but no adaptation. If for and with ms, then the model exhibits adaptation in addition to refractoriness. In all the simulations, we use with and , With this choice of we are in agreement with experimental results on cortical neurons [22], but the effects of adaptation and refractoriness cannot be separated as clearly as in the case of a model with absolute refractoriness. Loosely speaking, the long time constant causes adaptation, whereas the short time constant mainly contributes to refractoriness. In fact, for and equal to the membrane time constant, the model becomes equivalent to a leaky integrate-and-fire neuron [3], so that the neuron is refractory and non-adapting. In the simulations, is longer than the membrane time constant so that, for very strong stimuli, it may also contribute to adaptation. We note that the formalism developed in this paper does not rely on our specific choice of . We only require (i) causality by imposing for and (ii) so that the effect of a past spike decreases over time.
The effects described by can be mediated by a dynamic threshold as well as spike-triggered currents [22]. Throughout the remainder of the text we will refer to as the effective spike after-potential (SAP). It is, however, important to note that has no units, i.e. it relates to an appropriately scaled version of the experimentally measured spike after-potential. A depolarizing (facilitating) SAP is associated with , while a hyperpolarizing (adapting) SAP is associated with .
In a population of neurons, every neuron has a different spiking history defined by its past spike train where is the most recent spike, the previous one and so on. To find the population activity at any given time, we hypothesize that the strong effect of the most recent spike needs to be considered explicitly while the rest of the spiking history merely introduces a self-inhibition that is similar for all neurons and that depends only on the average firing profile in the past. Thus for each neuron we write the past spike train as where is the time of the last spike. Our hypothesis corresponds to the approximation , i.e. the last spike needs to be treated explicitly, but we may average across earlier spike times. This approximation is not appropriate for intrinsically bursting neurons, but it should apply well to other cell types (fast-spiking, non-fast-spiking, delayed, low-threshold). According to this hypothesis, and in analogy to the time-dependent renewal theory [3], [42] we find (derivation in Methods):(5)Unfortunately Eq. 5 remains insolvable, because we do not know . Using Eqs. 3 and 4 we find:(6)As mentioned above, we hypothesize that the spiking history before the previous spike merely inhibits subsequent firing as a function of the average spiking profile in the past. In order to formally implement such an approximation, we introduce a series expansion [57] in terms of the spiking history moments (derivation in Methods) where we exploit the fact that is a moment generating function:(7)The first history moment relates to the expected number of spikes at a given time . The second history moment considers the spike-spike correlations and so on for the higher moments.
We truncate the series expansion resulting from Eq. 7 at the first order (). We can then write Eq. 6 as:(8)We can insert Eq. 8 in Eq. 5 so as to solve for as a function of the filtered input . The solutions can be found using numerical methods.
We note that by removing the integral of from Eq. 8 we return exactly to the renewal equation for population activity (). Adaptation reduces the driving force by an amount proportional to the average spike density before , that is, the average spiking density before the most recent spike. In other words, instead of using the specific spike history of a given neuron, we work with the average history except for the most recent spike which we treat explicitly. We call Eqs. 5 and 8 the Quasi-Renewal equation (QR) to acknowledge its theoretical foundations. It is renewal-like, yet, we do not assume the renewal condition since a new spike does not erase the effect of the previous history (see Methods).
Let us now assess the domain of validity of the QR theory by comparing it with direct simulations of a population of SRM neurons. To describe the single neurons dynamics, we use a set of parameters characteristic of L2–3 pyramidal cells [22]. The SAP is made of two exponentials: one with a short time constant (30 ms) but large amplitude and another with a long time constant (400 ms) but a small amplitude. The results presented here are representative of results that can be obtained for any other physiological set of parameters. For details on the simulation, see Methods.
The response to a step increase in stimulating current is a standard paradigm to assess adaptation in neurons and used here as a qualitative test of our theory. We use three different step amplitudes: weak, medium and strong. The response of a population of, say, 25,000 model neurons to a strong step increase in current starts with a very rapid peak of activity. Indeed, almost immediately after the strong stimulus onset, most of the neurons are triggered to emit a spike. Immediately after firing at , the membrane potential of theses neurons is reset to a lower value by the contribution of the SAP; . The lower membrane potential leads to a strong reduction of the population activity. Neurons which have fired at time are ready to fire again only after the SAP has decreased sufficiently so that the membrane potential can approach again the threshold . We can therefore expect that a noiseless population of neurons will keep on oscillating with the intrinsic firing frequency of the neurons [6]; however, due to stochastic spike emission of a noisy population the neurons in the population gradually de-synchronize. The damped-oscillation that we see in response to a strong step stimulus (Fig. 2C) is the result of this gradual de-synchronization. Similar damped oscillations at the intrinsic firing frequency of the neurons have also been observed for a Spike Response Model with renewal properties [6], i.e., a model that only remembers the effect of the last spike.
In contrast to renewal models (i.e., models with refractoriness but no adaptation), we observe in Fig. 2C that the population activity decays on a slow time scale, taking around one second to reach a steady state. This long decay is due to adaptation in the single-neuron dynamics, here controlled by the slow time constant ms. The amount of adaptation can be quantified if we compare, for a given neuron its first interspike interval after stimulus onset with the last interspike interval. The mean first interspike interval (averaged over all neurons) for the strong step stimulus is 93 ms while the last interval is nearly twice as long (163 ms), indicating strong adaptation. For smaller steps, the effect of refractoriness is less important so that adaptation becomes the most prominent feature of the step response (Fig. 2A). An appropriate encoding framework should reproduce both the refractoriness-based oscillations and the adaptation-based decay.
The QR equation describes well both the damped oscillation and the adapting tail of the population activity response to steps (Fig. 2). Also, the steady state activity is predicted over a large range (Fig. 2D). We note that an adaptation mechanism that is essentially subtractive on the membrane potential (Eq. 4) leads here to a divisive effect on the frequency-current curve. Altogether, we conclude the QR theory accurately encode the response to step stimulus.
Step changes in otherwise constant input are useful for qualitative assessment of the theory but quite far from natural stimuli. Keeping the same SAP as in Fig. 2, we replace the piecewise-constant input by a fluctuating current (here Ornstein-Uhlenbeck process) and study the validity of QR over a range of input mean and standard deviation (STD), see Fig. 3. As the STD of the input increases, the response of the population reaches higher activities (maximum activity at STD = 80 pA was 89 Hz). The prediction by the QR theory is almost perfect with correlation coefficients consistently higher than 0.98. Note that the correlation coefficient is bounded above by the finite-size effects in estimating the average of the 25,000-neuron simulation. Over the range of input studied, there was no tendency of either overestimating or underestimating the population activity (probability of positive error was 0.5). There was only a weak tendency of increased discrepancies between theory and simulation at higher activity (correlation coefficient between simulated activity and mean square error was 0.25).
Decoding the population activity requires solving the QR equation (Eq. 5 and 8) for the original input (see Methods). Input steps can be correctly decoded (Fig. 4A–C) but also fluctuating stimuli (Fig. 4D–E). Again, the input mean does not influence the precision of the decoding (Fig. 4E). The numerical method does not decode regions associated with population activities that are either zero or very small. Accordingly, the correlation coefficient in Fig. 4E is calculated only at times where decoding could be carried out. Note that unless one is to estimate the statistics of the input current and assume stationarity, it is impossible for any decoder to decode at times when . If the size of the population is decreased, the performance of the QR decoder decreases (Fig. S1). Finite size effects limit decoding performance by increasing the error on the mean activity (as can be seen by comparing the effect of filtering the average population activity (Fig. S1A and B)). Another finite-size effect is that at small population sizes there is a greater fraction of time where an estimate of the activity is zero and the decoding cannot be performed (Fig. S1D–F). Also, decoding errors are larger when is close to zero (Fig. S1C). Nevertheless, for an input with STD = 40 pA and a population of 250 neurons, QR decoding can be performed 55% of the times with a correlation coefficient of 0.92. If the filtering of the population activity is on a longer time scale (20 ms instead of 2 ms) then decoding is possible 82% of the times and the accuracy is roughly the same (Fig. S1).
We will consider two recent theories of population activity from the literature. Both can be seen as extensions of rate models such as the Linear-Nonlinear Poisson model where the activity of a homogeneous population is where is a linear filter and some nonlinear function. First, we focus on adaptive formulations of such rate models. For example Benda and Herz [15] have suggested that the firing rate of adapting neurons is a non-linear function of an input that is reduced by the past activity, such that the activity is where is a self interaction filter that summarizes the effect of adaptation. Second, we compare our approach with renewal theory [3], [42] which includes refractoriness, but not adaptation. How does our QR theory relate to these existing theories? And how would these competing theories perform on the same set of step stimuli?
To discuss the relation to existing theories, we recall that the instantaneous rate of our model depends on both the input and the previous spike trains. In QR theory, we single out the most recent spike at and averaged over the remaining spike trains : . There are two alternative approaches. One can keep the most recent spike at and disregard the effect of all the others: . This gives rise to the time-dependent renewal theory, which will serve as a first reference for the performance comparison discussed below. On the other hand, one can average over all previous spikes, that is, no special treatment for the most recent one. In this case(9)The right-hand side of Eq. 9 can be treated with a moment expansion similar to the one in Eq. 7. To zero order, this gives a population rate , that is, an instantiation of the LNP model. To first order in an event-based moment expansion (EME1) we find:(10)Therefore, the moment expansion (Eq. 7) offers a way to link the phenomenological framework of Benda and Herz (2003) to parameters of the SRM. In particular, the nonlinearity is the exponential function, the input term is and the self-inhibition filter is . We note that Eq. 10 is a self-consistent equation for the population activity valid in the limit of small coupling between the spikes which can be solved using standard numerical methods (see Methods). A second-order equation (EME2) can similarly be constructed using an approximation to the correlation function (see Methods).
We compare the prediction of EME1, EME2 and renewal theory with the simulated responses to step inputs (Fig. 5). All the encoding frameworks work well for small input amplitudes (Fig. 5A). It is for larger input steps that the different theories can be distinguished qualitatively (Fig. 5C). Renewal theory predicts accurately the initial damped oscillation as can be expected by its explicit treatment of the relative refractory period. The adapting tail, however, is missing. The steady state is reached too soon and at a level which is systematically too high. EME1 is more accurate in its description of the adapting tail but fails to capture the damped oscillations. The strong refractory period induces a strong coupling between the spikes which means that truncating to only the first moment is insufficient. The solution based on EME2 improves the accuracy upon that of EME1 so as to make the initial peak shorter, but oscillates only weakly. We checked that the failure of the moment-expansion approach is due to the strong refractory period by systematically modifying the strength of the SAP (Fig. S2). Similarly, when the SAP is weak, the effect of will often accumulate over several spikes and renewal theory does not capture the resulting adaptation (Fig. S2).
Fluctuating input makes the population respond in peaks of activity separated by periods of quiescence. This effectively reduces the coupling between the spikes and therefore improves the accuracy of EME1. The validity of EME1 for encoding time-dependent stimulus (Fig. S3) decreases with the STD of the fluctuating input with no clear dependence on the input mean.
Decoding with EME1 is done according to a simple relation:(11)where the logarithm of the momentary population activity is added to an accumulation of the past activity. The presence of the logarithm reflects the non-linearity for encoding (the link-function in Eq. 3) and leads to the fact that when the instantaneous population activity is zero, the stimulus is undefined but bounded from above: . Fig. S4 shows the ability of Eq. 11 to recover the input from the population activity of 25,000 model neurons. We conclude that Eq. 11 is a valid decoder in the domain of applicability of EME1.
In summary, the EMEs yield theoretical expressions for the time-dependent as well as steady-state population activity. These expressions are valid in the limit of small coupling between the spikes which corresponds to either large interspike intervals or small SAP. Renewal theory on the other hand is valid when the single-neuron dynamics does not adapt and whenever the refractory effects dominate.
The input-output function of a neuron population is sometimes described as a linear filter of the input [41], as a linear filter of the input reduced as a function of past activity [58], [59], as a non-linear function of the filtered input [60], or by any of the more recent population encoding frameworks [47], [48], [61]–[65]. These theories differ in their underlying assumptions. To the best of our knowledge, a closed-form expression that does not assume weak refractoriness or weak adaptation has not been published before.
We have derived self-consistent formulas for the population activity of independent adapting neurons. There are two levels of approximation, EME1 (Eq. 10) is valid at low coupling between spikes which can be observed in real neurons whenever (i) the interspike intervals are large, (ii) the SAPs have small amplitudes or (iii) both the firing rate is low and the SAPs have small amplitudes. The second level of approximation merges renewal theory with the moment-expansion to give an accurate description on all time-scales. We called this approach the QR theory.
The QR equation captures almost perfectly the population code for time-dependent input even at the high firing rates observed in retinal ganglion cells [55]. But for the large interspike intervals and lower population activity levels of in vivo neurons of the cortex [66], [67], it is possible that the simpler encoding scheme of Eq. 10 is sufficient. Most likely, the appropriate level of approximation will depend on the neural system; cortical sparse coding may be well represented by EME, while neuron populations in the early stages of perception may require QR.
We have focused here on the Spike Response Model with escape noise which is an instantiation of a Generalized Linear Model. The escape noise model, defined as the instantaneous firing rate given the momentary distance between the (deterministic) membrane potential and threshold should be contrasted with the diffusive noise model where the membrane potential fluctuates because of noisy input. Nevertheless, the two noise models have been linked in the past [51], [54], [68]. For example, the interval-distribution of a leaky integrate-and-fire model with diffusive noise and arbitrary input can be well captured by escape noise with instantaneous firing rate which depends both on the membrane potential and its temporal derivative [51]. The dependence upon accounts for the rapid and replicable response that one observes when an integrate-and-fire model with diffusive noise is driven in the supra-threshold regime [68] and can, in principle, be included in the framework of the QR theory.
The decoding schemes presented in this paper (Eq. 11 and 45) reveal a fundamental aspect of population coding with adapting neurons. Namely, the ambiguity introduced by the adaptation can be resolved by considering a well-tuned accumulator of past activity. The neural code of adapting populations is ambiguous because the momentary level of activity could be the result of different stimulus histories. We have shown that resolving the ambiguity requires the knowledge of the activity in the past but to a good approximation does not require the knowledge of which neuron was active. At high population activity for neurons with large SAPs, however, the individual timing of the last spike in the spike trains is required to resolve the ambiguity (compare also Fairhall et al. [13]). Unlike bayesian spike-train decoding [55], [69], [70], we note that in our decoding frameworks the operation requires only knowledge of the population activity history and the single neuron characteristics. The properties of the QR or EME1 decoder can be used to find biophysical correlates of neural decoding such as previously proposed for short term plasticity [71], [72], non-linear dendrites [73] or lateral inhibition [74]. Note that, a constant percept in spite of spike frequency adaptation does not necessarily mean that neurons use a QR decoder. It depends on the synaptic structure. In an over-representing cortex, a constant percept can be achieved even when the neurons exhibit strong adaptation transients [75].
Using the results presented here, existing mean-field methods for populations of spiking neurons can readily be adapted to include spike-frequency adaptation. In Methods we show the QR theory for the interspike interval distribution and the steady-state autocorrelation function (Fig. 6) as well as linear filter characterizing the impulse response function (or frequency-dependent gain function) of the population. From the linear filter and the autocorrelation function, we can calculate the signal-to-noise ratio [3] and thus the transmitted information [1]. The autocorrelation function also gives an estimate of the coefficient of variation [76] and clarifies the role of the SAP in quenching the spike count variability [49], [77], [78]. The finite-size effects [27], [79]–[81] is another, more challenging, extension that should be possible.
The scope of the present investigation was restricted to unconnected neurons. In the mean-field approximation, it is straight-forward to extend the results to several populations of connected neurons [6]. For instance, similar to EME1, a network made of inter-connected neurons of cell-types would correspond to the self-consistent system of equation:(12)where is the scaled post-synaptic potential kernel from cell-type to cell-type (following the formalism of Gerstner and Kislter [3]), is an external driving force, each subpopulation is characterized by its population activity and its specific spike after potential . The analogous equation for QR theory is:(13)where is:(14)Since the SAP is one of the most important parameter for distinguishing between cell classes [22], the approach presented in this paper opens the door to network models that take into account the neuronal cell-types beyond the sign of the synaptic connection. Even within the same class of cells, real neurons have slightly different parameters from one cell to the next [22] and it remains to be tested whether we can describe a moderately inhomogeneous population with our theory. Also, further work will be required to see if the decoding methods presented here can be applied to brain-machine interfacing [82]–[84].
This section is organized in 3 subsections. Subsection A covers the mathematical steps to derive the main theoretical results (Eqs. 2, 5 and 7). It also presents a new approach to the time-dependent renewal equation, links with renewal theory and the derivation of the steady-state interspike interval distribution and auto-correlation. Subsection B covers the numerical methods and algorithmic details and subsection C the analysis methods.
All simulations were performed on a desktop computer with 4 cores (Intel Core i7, 2.6 GHz, 24 GB RAM) using Matlab (The Mathworks, Natwick, MA). The Matlab codes to numerically solve the self-consistent equations are made available on the author's websites. The algorithmic aspects of the numerical methods are discussed now.
When assessing the accuracy of the encoding or the decoding, we used the correlation coefficient. The correlation coefficient is the variance-normalized covariance between two random variables and :(52)where the expectation is taken over the discretized time.
|
10.1371/journal.pntd.0001013 | Leprosy among Patient Contacts: A Multilevel Study of Risk Factors | This study aimed to evaluate the risk factors associated with developing leprosy among the contacts of newly-diagnosed leprosy patients.
A total of 6,158 contacts and 1,201 leprosy patients of the cohort who were diagnosed and treated at the Leprosy Laboratory of Fiocruz from 1987 to 2007 were included. The contact variables analyzed were sex; age; educational and income levels; blood relationship, if any, to the index case; household or non-household relationship; length of time of close association with the index case; receipt of bacillus Calmette-Guérin (BGG) vaccine and presence of BCG scar. Index cases variables included sex, age, educational level, family size, bacillary load, and disability grade. Multilevel logistic regression with random intercept was applied. Among the co-prevalent cases, the leprosy-related variables that remained associated with leprosy included type of household contact, [odds ratio (OR) = 1.33, 95% confidence interval (CI): 1.02, 1.73] and consanguinity with the index case, (OR = 1.89, 95% CI: 1.42–2.51). With respect to the index case variables, the factors associated with leprosy among contacts included up to 4 years of schooling and 4 to 10 years of schooling (OR = 2.72, 95% CI: 1.54–4.79 and 2.40, 95% CI: 1.30–4.42, respectively) and bacillary load, which increased the chance of leprosy among multibacillary contacts for those with a bacillary index of one to three and greater than three (OR = 1.79, 95% CI: 1.19–2.17 and OR: 4.07–95% CI: 2.73, 6.09), respectively. Among incident cases, household exposure was associated with leprosy (OR = 1.96, 95% CI: 1.29–2.98), compared with non-household exposure. Among the index case risk factors, an elevated bacillary load was the only variable associated with leprosy in the contacts.
Biological and social factors appear to be associated with leprosy among co-prevalent cases, whereas the factors related to the infectious load and proximity with the index case were associated with leprosy that appeared in the incident cases during follow-up.
| Leprosy is an infectious disease that can lead to physical disabilities, social stigma, and great hardship. Transmitted from person to person, it is still endemic in developing countries, like Brazil and India. Effective treatment has been available since 1960, but early diagnosis of the disease remains the most effective way to stop the transmission chain and avoid late diagnoses and subsequent disabilities. Knowledge of the risk factors for leprosy can facilitate early detection; therefore, our study aimed to investigate the factors presented by leprosy patients and their contacts, who are considered at highest risk of contracting the disease. We studied 6,158 contacts of 1,201 patients under surveillance from 1987 to 2007 in a Public Health Care Center in the City of Rio de Janeiro, Brazil. We evaluated the ways patient and contact demographics and epidemiological characteristics were associated with the detection of leprosy. Statistical analyses took into account both individual and group characteristics and their interrelationships. The main characteristics facilitating the contraction of leprosy among contacts were shown to be consanguinity and household association. Conversely, the bacillary load index of leprosy patients was the principle factor leading to disease among their contacts.
| The primary aim of all disease control measures is to reduce the incidence, prevalence, morbidity and/or mortality rates to the lowest level possible in a given population. However, once control program objectives have been met, continuous interventions are necessary to maintain these minimal rates [1].
In 2007, the Brazilian Ministry of Health adopted new case detection rates for all ages and for children under 15 years of age as indicators of the effectiveness of leprosy control measures in the country. Because detection of leprosy in those under 15 years of age is considered indicative of recent Mycobacterium leprae (ML) transmission, evaluating these cases for epidemiologic markers was especially important [2].
In early 2009, the global prevalence of leprosy was approximately 213,000 cases; however, the annual detection rate of leprosy worldwide has declined. In 2002, more than 620,000 cases were detected; whereas, in 2008, there were approximately 249,000 cases. In Brazil, in 2008, there were 38,914 new leprosy cases detected. Nevertheless, there seems to be a tendency for the detection rates to stabilize in Brazil at somewhat higher levels in the North, Midwest and Northeast regions of the country. In the state of Rio de Janeiro, there is a clear decreasing trend from 1990 to 2008. For instance, detection rates ranged from 27.30 cases per 100,000 population in 1997 to 11.84 cases per 100,000 population in 2008. The detection rates in Rio de Janeiro for children less than 15 years old in the period 2001–2008 had very high ratings (6.00/100,000 population to 2.69/100,000 population).
In addition to the administration of multidrug therapy (MDT) to patients diagnosed with leprosy, disease control strategies in Brazil include early new case detection, routine clinical examinations, and Bacillus Calmette-Guérin (BGG) vaccination of the patient's contacts, which is a group considered to be at high risk to develop the disease [3].
One activity of early detection of leprosy is contact surveillance, which aims to interrupt disease transmission and prevent the development of disabilities [4].
The notion that group-level factors are important in understanding the risk of disease has long been present in infectious disease epidemiology, because the risk of an individual contracting an infectious disease depends not only on his or her own risk behavior and biological and socio-economic factors, but also on his or her population group. With regard to scientific validity and the practical implications for disease prevention, the growing consensus is that investigations into the causes of disease must include factors defined on multiple levels, such as the individual and communities. In infectious disease epidemiology, multilevel analysis can be used to examine how both group- and individual-level factors are related to individual-level infectious disease outcomes and how factors on both levels affect group differences in the risk of disease. The application of multilevel analysis has only recently begun to emerge in the infectious disease literature [5], [6].
Several potential risk factors associated with individual features of leprosy patients and their contacts have been suggested but, to date, these factors' effects have yet to be evaluated. In-depth investigation of these factors may allow for the simultaneous examination of group-level and individual-level factors, assessment of the demonstrable interaction between contacts- and index case-level constructs, and exploration of how factors at multiple levels contribute to differences in disease risk.
The aim of the present study was to identify potential risk factors of the index cases and their contacts on development of leprosy among contacts.
Since 1987, the Leprosy Outpatient Clinic, a National Reference Center at the Oswaldo Cruz Foundation in Rio de Janeiro, RJ, Brazil, has conducted routine clinical examinations of the contacts of leprosy patients diagnosed at the Clinic. The Clinic provides health care recommendations to leprosy patients and their families at diagnosis and during treatment.
The study population consisted of 6,158 contacts of 1,201 newly-diagnosed leprosy patients of the cohort treated at the Leprosy Outpatient Clinic from 1987 to 2007. The average duration of follow-up of contacts was 16.9 years. Among the patients, 454 had paucibacillary leprosy, and 747 had multibacillary leprosy.
After confirmation of the leprosy diagnosis, patients were given educational information about the disease, and medical visits were scheduled for their close contacts (within and outside of the household). During the initial visits, contacts answered a questionnaire regarding socio-economic status (income and education level) and type of contact with the index case. The contacts were examined by specialized dermatologists and neurologists to confirm a leprosy diagnosis and the existence of a BCG scar.
The Brazilian Ministry of Health recommends that all leprosy contacts receive the BCG vaccine [3]. Between 1987 and 1991, all contacts were instructed to attend the Clinic at least once a year. From January 1992 throughout December 2007, they were also requested to return to the Clinic if and when symptoms and/or skin lesions appeared. Follow-up visits included medical consultations with specialized dermatologists and neurologists. Those presenting signs or symptoms that were suggestive of leprosy were assessed through bacteriological, histopathological, and immunological examinations.
In September 2009, the Brazilian Disease Notification System (SINAN), covering December 1987 to September 2009, was searched to locate the healthy contacts to ascertain whether any leprosy cases had been missed in contact follow-up procedures. SINAN records were matched to the database of the study group with respect to the variables present in both: name of contact, date of birth and mother's full name. Contacts that had not been identified as leprosy patients in SINAN by September 2009 were considered healthy.
Co-prevalent cases were the contacts diagnosed with leprosy at the first examination after the index case was diagnosed. Incident cases were apparently leprosy-free contacts at the time of index case diagnosis but developed the disease at some point during follow-up.
Household contacts were defined as individuals who had lived in the same dwelling during the five-year period prior to the index case diagnosis. Non-household contacts were defined as those indicated by the index case as having had other types of contact, such as next-door neighbors, blood relatives, friends and/or co-workers, etc., during the five-year period prior to the index case diagnosis.
Variables that described the contact included sex; age; educational and income levels; blood relationship, if any, and type (household and non-household) and length of time of close association with the index case. With regard to BCG vaccination, contacts were examined to verify the presence or absence of a BCG scar, which was considered the first dose. Once a leprosy diagnosis is excluded, the BCG vaccine is administered to a healthy contact, and this vaccination corresponded to the second dose. For the index cases, the variables included sex, age, educational level, family size, bacillary index (BI) from the slit skin smear test at the beginning of treatment and disability grade.
The patients were classified as paucibacillary, based on a zero BI, or multibacillary, based on an above-zero BI.
We classified the initial disability/impairment grade according to the present World Health Organization classification system [7], which consisted of three grades (0, 1 and 2). Grade 0 indicates no loss of sensation or visible deformity, grade 1 is defined by the loss of sensation without visible deformity, and grade 2 indicates the presence of a visible deformity. All disability grade evaluations were conducted by specialized professionals.
A two-level logistic model with a random intercept was used, and the contacts were considered first-level units and grouped with their respective index cases, who were considered second-level units. For the empty models, the Variance Partition Coefficient (VPC) was calculated according to the simulation method proposed by Goldstein et al [8]. The total number of simulations was 5,000.
Initially, a bivariate analysis was conducted separately for the co-prevalent and incident cases. The association between the occurrence of leprosy disease and a set of independent variables was assessed using the crude odds ratio (OR) and the associated 95% confidence interval (CI).
The second step of the analysis involved adjusting the multilevel logistic regression model for all the contact and index case variables (full model – data not shown.)
The final model consisted of all the variables that were statistically significant after adjustment for all other factors related to the contacts and their respective index cases. Additional variables in the final included those recognized for epidemiological relevance or were frequently regarded as confounding variables, such as age of contact, sex, and contact and index case educational levels.
The estimated measure of association was the OR. The OR associated with incident-case risk factors may be interpreted as a relative risk (RR) when the disease frequency is low, as in the present study. The OR of prevalence cases also estimates the RR if the disease duration among the exposed and unexposed is the same [9].
The software MlWin 2.10 was used to perform the multilevel statistical analysis. The estimation method of Penalized Quasi-Likelihood, second order, was adopted throughout the analysis.
All contacts who returned to the clinic for examination were eligible for the study. All adult participants and the guardians or parents of the children that were included in the study provided written consent. This study was approved by the Ethics Research of the National School of Public Health.
This study included 6,158 contacts of 1,201 leprosy patients, with an average 5.12 contacts per patient. Of the contacts studied, 57.6% (3546/6158) were female. The mean age was 25.6 (±17,8) years. Of the index cases, 63.9% (767/1201) were male, and the mean age was 38.2 (±16,9) years.
Among the contacts, 452 (7.3%) new cases of leprosy were diagnosed. The first contact examination found 319 (5.2%) co-prevalent cases, and during the follow-up, 133 (2.3%) incident cases were diagnosed. Among the incident cases, this study found an incidence rate of 3.32 cases per person-year. The average period for the incident cases of leprosy diagnosis was 4.1 years after the index case diagnosis.
Among the contacts diagnosed with leprosy, 89.4% (404/452) had multibacillary leprosy, 74.5% (337/452) had paucibacillary leprosy, and 65.8% of them (222/337) had borderline-tuberculoid leprosy.
Table 1 shows the numbers and proportion of contacts with leprosy according to the clinical classification of index cases.
The VPCs were approximately 18% and 13% for the co-prevalent and incident cases, respectively, i.e., the proportion of the outcome variability due to the determinants on the first level was somewhat greater in incident patients than in co-prevalent patients.
The frequencies and the bivariate analyses for the contacts and index cases, for the co-prevalent and incident cases are shown in Table S1.
A significant association was observed between the contacts diagnosed with leprosy at the initial examination (co-prevalent cases) and several of the variables under study; these included few years of schooling (OR = 1.50, 95% CI: 1.03–2.19), a monthly family income under three minimum wages (OR = 1.85, 95% CI: 1.35–2.54 and OR = 2.18 95% CI: 1.50–3.17), consanguineous relationship with (OR = 1.50, 95% CI: 1.15–1.96) and close proximity to the index case for a minimum five-year period (OR = 2.64, 95% CI: 1.75–3.98). Household contacts were more likely than non-household contacts to present with leprosy, for both co-prevalent cases (OR = 1.44, 95% CI: 1.11–1.86) and incident cases (OR = 2.05, 95% CI: 1.35–3.11). Having received a neonatal BCG vaccine was a protective factor in both co-prevalent and incident cases. In addition, the application of the BCG vaccine, as recommended by the Ministry of Health, was also a protective factor in the follow-up.
Among the index case variables, some were associated with a leprosy diagnosis in co-prevalent cases; these included up to 4 years of schooling (OR = 3.31, 95% CI: 1.87–5.58), between 4 to 10 years of schooling (OR = 2.53, 95% CI: 1.37–4.64), monthly family income up to two minimum wages (OR = 2.17, 95% CI: 1.34–3.52), having an income between two and three minimum wages (OR = 2.31, 95% CI: 1.44–3.70), and a disability grade = 2 (OR = 1.50, 95% CI: 1.04–2.16). The contacts who were 15 years and older had an increased odds ratio (OR = 8.37, 95% CI: 1.12–62.4) of contracting leprosy, compared with those who were under 15, only among incident cases. Contacts of male index cases were more likely to have leprosy than contacts of female index cases. This was true for both prevalent and incident leprosy cases among contacts. BIs of index cases over three was significantly associated with the diagnosis of co-prevalent leprosy cases (OR = 4.37, 95% CI: 2.95–6.46). BIs of one to three (OR = 4.30, 95% CI: 2.12–8.71) and more than three (OR = 7.31, 95% CI: 3.63–14.75) were associated with incident leprosy cases, considering as reference a negative BI.
Table 2 summarizes the results of the multivariate analysis. In the final model for co-prevalent cases, the variables that remained associated with leprosy between contacts were household contact (OR = 1.33: 95% CI: 1.02–1.73) and consanguinity with the index case (OR = 1.89, 95% CI: 1.42–2.51). With respect to the index case model, the variables associated with leprosy included up to 4 years of schooling and 4 to 10 years of schooling (OR = 2.72, 95% CI: 1.54–4.79 and 2.40, 95% CI: 1.30–4.42, respectively), and bacillary index, which increased the risk of leprosy among contacts for those with index cases with BI of one to three and greater than three (OR = 1.79, 95% CI: 1.19–2.70 and OR: 4.07, 95% CI: 2.73–6.09, respectively).
In the multilevel model for incident cases, household exposure was associated with leprosy in the incident case contacts, with OR = 1.96 (95% CI: 1.29–2.98). The consanguinous relationship of contacts with their index case was also a significant risk factor for contracting leprosy (OR = l.54, 95% CI: 1.00–2.37). In connection with index case variables, an elevated bacillary load was the only variable whose association was maintained after adjusting for the other variables under consideration.
The presence of a BCG scar showed a highly statistically significant protective effect in both models for co-prevalent and incident cases, with OR = 0.28 (95% CI: 0.21–0.37) and 0.45 (95% CI: 0.30–0.68), respectively. The contacts who received the BCG vaccine also demonstrated significant protection against the disease: OR = 0.44 (95% CI: 0.29–0.64).
There were no statistically significant differences in the odds between male and female contacts in either incident or co-prevalent cases.
Finally, the presence of overdispersion in the final models was not detected. The overdispersion parameter in the model for co-prevalent cases was 0.89 and that for incident cases was 0.94.
In this study, we found that the major risk factor among contact incident cases was proximity to the index case. Among the characteristics of the index cases, bacillary load was the only risk factor associated with developing leprosy. A BCG scar and the application of the vaccine after index case diagnosis independently contributed as protective factors. However, among co-prevalent cases, the variables most strongly associated were a consanguinous and household relationship with the index case. Furthermore, a BCG scar contributed independently as a protective factor. Factors related to the index cases included up to 4 years and between 4 to 10 years of schooling and bacillary load, both associated with leprosy among their contacts at the first examination.
Although men make up most of the leprosy cases in Brazil, our study did not find any gender differences in the risk of contracting the disease among contacts, suggesting that the gender differences in the detection rates for the general population may be due to differences in their exposure. These findings are in agreement with those of other studies that likewise did not observe any gender differences in the likelihood of acquiring leprosy [10]–[12] Nevertheless, Ali et al. [13], in a prospective contact study and two other retrospective studies, found that the attack rate was, in fact, lower among women [14], [15]. Conversely, Fine et al. [16] reported a significantly higher attack rate among men.
In the present study, contact age was not associated with leprosy among either co-prevalent or incident cases. Our decision to categorize the age of minors and those over 15 years to conform to the indicator adopted by the Brazilian Leprosy Control Program may be an explanation for this lack of association. Other studies have shown that among contacts the risk of leprosy is significantly higher for those younger than 14, particularly for contacts of multibacillary index cases [11], [13], [17]. Likewise, Moet et al. [12] reported a bimodal distribution according to age: the risk increased for those between 5 to 15 years of age, reached a peak for those aged 15 to 20, decreased for those aged 20 to 29, and gradually increased after a 30-year lag.
Leprosy has traditionally been associated with lower socio-economic status. An ecological study recently conducted in Brazil by Kerr et al. [17] showed an association between social inequality, population growth and a high prevalence of leprosy. Population-based studies have also described an increased risk of leprosy associated with fewer years in school, poor housing and low income [18]. Our findings suggested an association between level of education and leprosy. However, in our study, poor schooling was associated with disease duration in index case patients and with a higher prevalence of leprosy among their close contacts (co-prevalence). Poor schooling among index case patients is likely to be a proxy for lower socio-economic status and could be associated with late diagnosis of leprosy, allowing for longer periods of exposure among their contacts. This finding is most certainly related to both the unavailability and inaccessibility of health care facilities, making it more difficult for individuals to maintain good health and prevent disease. In the present study, the lack of association between socio-economic markers and the risk of disease could be understood in light of the homogenous distribution of these markers along the study sample; everyone involved in this study was from the same socio-economic strata.
From the moment of the index case diagnosis, the consanguinous relatives had a higher risk of developing leprosy (OR = 1.89, 95% CI: 1.42–2.51). Most likely due to their increased vulnerability, genetic susceptibility, and type of immune response, these contacts were more likely to become ill. In turn, the confidence interval of the probability of association with incidence cases was just above the cut-off probability of 0.05. A cross-sectional study on determinants of the transmission of leprosy showed that consanguinous relatives had a 2.8 higher risk than non-consanguineous contacts [19]. Similarly, Moet et al. [12], in the initial evaluation of a contact cohort, calculated that consanguinous contacts had an increased odds (OR = 1.65, 95% CI: 1.05, 2.57), regardless of physical distance from their index case.
As expected, contact/index case co-habitation was shown to be a key risk factor in developing leprosy. However, the strength of this association was different for both co-prevalent and incident cases. Household contacts had a higher risk for leprosy in the follow-up. Among incident cases, the risk of household contacts developing the disease was twice that among non-household contacts, which also corroborated findings of aforementioned studies. To reiterate, a number of reports have indicated that household contacts are at the highest risk, compared with the general population [11], [14], [20] and non-household contacts [16], [21].
As in other, similar studies, the most important association determining leprosy disease among contacts was the bacillary load of the index case. These findings were in agreement with the literature that demonstrates that multibacillary patients are primarily responsible for ML transmission in endemic areas [10]–[13], [15], [20], [22]. In the follow-up, index cases with BIs over three were eight times more likely to transmit leprosy to their contacts (incident cases) than were paucibacillary patients. The contacts of multibacillary index cases also had a four-fold higher chance of being diagnosed with leprosy (co-prevalent cases) than did the contacts of index cases with a negative BI. A previous study conducted in Brazil demonstrated that a high familial bacillary index and the presence of more than one source of contamination in the family at the time of first examination of contacts were associated with greater risk of developing leprosy, especially among those younger than 15 years [23].
Again, in the present study, the BCG vaccine administered in infancy was shown to effectively protect against leprosy in 72% [(1-OR)×100] of all co-prevalent cases and 55% of incident cases. During the follow-up, the protective rate conferred by the BCG vaccine applied after index case diagnosis was 56% [(1-OR)×100)]. Other Brazilian studies have confirmed the significant impact of neonatal BCG on the incidence and transmission of leprosy [24], [25]. In our study, of the contacts vaccinated who developed leprosy in the follow-up period, 89% have presented with the paucibacillary form of the disease, indicating the protective effect of BCG vaccine against the development of multibacillary forms, consistent with other studies that point to the role of vaccine in the interruption of leprosy transmission [26], [27].
In summary, socio-economic factors appear to be more strongly associated with leprosy among the contacts found to be ill at the first examination (co-prevalent cases), compared with the association among incident cases. This finding among co-prevalent cases may be secondary to the difficulties that patients with lower educational level have in finding adequate health care facilities and information. With regard to the incident cases, bacillary load factors, i.e., intensity of transmission, increased the likelihood of contracting leprosy, in comparison with other social and biological factors. Moreover, incident cases developed the disease even when the associated co-prevalent and index cases were undergoing treatment, had neurological and skin examinations, and received the BCG vaccination.
A major strength of this study was the multilevel approach in analyzing the data, which allowed for the simultaneous observation of the effects of the predictor variables on both the group (index case) and individual levels (contacts). Importantly, inter-group-dependent observations were taken into account, which highlighted and did not disregard the dependency of leprosy as an infectious disease. According to the evaluation of VPC in Goldstein [8], with regard to empty models, the leprosy variance among contacts that can be attributed to the differences among index cases was 18% of the co-prevalent and 13% of the incident cases. Moreover, we observed that outcome variability at the superior hierarchical level was sufficient to justify the use of this model. We also found that the VPC evaluation of the final models indicated that only 2.4% and 4.0% of the explained variables continued to be attributable to the index cases, whereas the model appears to be well fitted for both the co-prevalent and incident cases.
The ability to accurately identify contacts of leprosy patients who are at high risk of disease is of utmost importance for leprosy control. Surveillance and appropriate health education of household contacts should be strongly reinforced and extended to all close contacts of index case-patients, including their consanguineous relatives. In our study, however, we have identified a group of contacts who, despite all appropriate intervention measures, acquired leprosy. Therefore, household contacts of MB index case-patients, especially those with high bacillary load at diagnosis, should be considered for chemoprophylaxis in addition to immunoprophylaxis with BCG vaccination, once the efficacy of chemoprophylaxis is proven.
|
10.1371/journal.pntd.0003730 | Suppression of the Insulin Receptors in Adult Schistosoma japonicum Impacts on Parasite Growth and Development: Further Evidence of Vaccine Potential | To further investigate the importance of insulin signaling in the growth, development, sexual maturation and egg production of adult schistosomes, we have focused attention on the insulin receptors (SjIRs) of Schistosoma japonicum, which we have previously cloned and partially characterised. We now show, by Biolayer Interferometry, that human insulin can bind the L1 subdomain (insulin binding domain) of recombinant (r)SjIR1 and rSjIR2 (designated SjLD1 and SjLD2) produced using the Drosophila S2 protein expression system. We have then used RNA interference (RNAi) to knock down the expression of the SjIRs in adult S. japonicum in vitro and show that, in addition to their reduced transcription, the transcript levels of other important downstream genes within the insulin pathway, associated with glucose metabolism and schistosome fecundity, were also impacted substantially. Further, a significant decrease in glucose uptake was observed in the SjIR-knockdown worms compared with luciferase controls. In vaccine/challenge experiments, we found that rSjLD1 and rSjLD2 depressed female growth, intestinal granuloma density and faecal egg production in S. japonicum in mice presented with a low dose challenge infection. These data re-emphasize the potential of the SjIRs as veterinary transmission blocking vaccine candidates against zoonotic schistosomiasis japonica in China and the Philippines.
| Schistosomiasis affects over 200 million people globally. An anti-schistosome vaccine is not currently available. Schistosome eggs play a critical role in host pathology and the transmission of schistosomiasis; thus a vaccine targeting parasite fecundity and/or egg viability represents a realistic strategy for blocking transmission, promoting disease control in endemic areas. Based on our previous studies on the insulin receptors (SjIRs) of Schistosoma japonicum, as potential vaccine candidates, we have now further investigated the impact of insulin signaling on the growth, development, sexual maturation and egg production of adult schistosomes. Protein binding assays and RNAi strongly support our hypothesis that the insulin pathway in schistosomes is activated by the binding between host insulin and the parasite IRs, regulating the transcription of downstream genes integrally involved in glucose uptake and fecundity in these parasites. This feature was reflected in the striking decreased glucose levels of worms when the SjIRs were knocked down. Furthermore, the importance of the SjIRs in the growth and fecundity of adult S. japonicum was further demonstrated in murine vaccine trials using a low dose cercarial challenge which resulted in depressed female growth and faecal egg production in mice vaccinated with the recombinant L1 subdomains of SjIR1 and SjIR2.
| Schistosomiasis is a major public health problem in many developing countries in the tropics and sub-tropics where it affects 200 million people, and is directly or indirectly responsible for many thousand deaths annually [1]. Despite the existence of the highly effective antischistosomal drug praziquantel (PZQ), schistosomiasis continues to spread into new areas [2], and although regarded as the cornerstone of current control programs, PZQ chemotherapy does have limitations. Important shortcomings of PZQ are its relative inactivity against migratory juvenile worms [3], its inability to prevent reinfection and the possibility that drug resistant parasites might evolve [2,4]. Consequently, an anti-schistosome vaccine, combined with chemotherapy and other interventions, can provide an important component of a sustainable, integrated package strategy for the future control of schistosomiasis [4]. Bovines, especially water buffaloes, contribute up to 90% of the environmental egg contamination for Schistosoma japonicum infection in China, and recent evidence suggest bovines are also important reservoirs of infection in the Philippines [5,6]. Mathematical modelling has predicted that a transmission blocking vaccine that can reduce the faecal egg output of bovines by 45% in endemic areas, in combination with PZQ treatment and other interventions, could lead to a significant reduction in schistosomiasis transmission almost to the point of elimination [7]. The currently available evidence thus emphasises the relevance of developing a transmission blocking veterinary vaccine for use in bovines to reduce S. japonicum infection (or reinfection) and to decrease the force of transmission by interrupting female worm egg production [8].
As well as being involved in transmission, schistosome eggs also play a key role in the pathology of schistosomiasis and it has recently been shown that receptor tyrosine kinase (RTK) signalling, triggered by host growth factors or hormones, is an active process in the reproductive tissues of schistosomes, regulating sexual maturation and egg production in adult females parasites [9]. Furthermore, glucose, the major nutritional source exploited by these blood flukes from the mammalian host, is also essential to fuel their growth and fecundity [10]. Four glucose transporter proteins have been identified in Schistosoma mansoni (SGTP1, 2, 3 and 4) although only SGTP1 and SGTP4, when expressed in Xenopus laevis oocytes, were shown to exhibit glucose transport activity [11]. Due to the importance of RTK signaling and glucose metabolism in schistosome biology, two types of insulin receptors have been isolated from S. mansoni (SmIR1 and SmIR2) [12] and S. japonicum (SjIR1 and SjIR2) [13], which were shown to bind human insulin in two-hybrid analysis and to be highly expressed in adult worms and schistosomula. However, it is not clear whether schistosomes utilise the insulin receptor to modulate glucose transport via the insulin signalling pathway in a similar manner to that observed in Caenorhabditis elegans [14].
Our previous microarray analysis demonstrated that host insulin appears to play an important role in insulin signalling in schistosomes by stimulating glucose metabolism through up-regulation of the phosphoinositide-3-kinase (PI3K) sub-pathway and in worm fecundity by activation of the mitogen-activated protein kinases (MAPK) sub-pathway [15]. Our earlier data also showed that SjIR1 is located in the tegument basal membrane and intestinal epithelium of adult worms of S. japonicum, while SjIR2 is located in the vitelline glands of female worms, where it likely plays an important role in oogenesis and egg production [13]. Of note, the Venus kinase receptor, which has an intracellular tyrosine kinase domain similar to that of the insulin receptor, is highly expressed in the reproductive organs of adult S. mansoni and plays an essential role in fecundity by activating the PI3K/AKt/P70s6k and Erk/MAPK sub-pathways [16], similar to the mechanism found in the insulin signalling pathway.
Results of vaccination/challenge trials we undertook in mice have provided further support for the important role played by the insulin receptor in worm fecundity in S. japonicum, indicating its potential value as a transmission blocking vaccine candidate; we obtained highly significant reductions in faecal eggs (56–67%), stunting of adult worms (12–42%), and a reduction in the numbers of mature intestinal eggs (75%) in animals vaccinated with the L1 subdomain of SjIR-2 (SjLD2) recombinantly expressed in Escherichia coli compared with controls [17]. Despite these promising data, the precise biological functions of the SjIRs have yet to be fully elucidated in schistosomes. In this study, we employed protein interaction assays and RNA interference (RNAi) to explore the functional roles of SjIR1 and SjIR2 in adult S. japonicum. To determine whether rSjLD1/2 could generate a longer period of repression of worm growth and fecundity, we carried out vaccine trials in mice with both proteins using a challenge with low numbers of cercariae to further establish the value of the insulin receptors as vaccine targets.
The conduct and procedures involving animal experimentation were approved by the Animal Ethics Committee of the QIMR Berghofer Medical Research Institute (project number A0108-054). This study was performed in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health.
Oncomelania hupensis, naturally infected with S. japonicum, were obtained from an endemic area in Anhui Province, PR China, and transported to the Brisbane laboratory in Australia. Cercariae were shed from the infected snails and collected as described [18].
Characterisation of SjIR1 and SjIR2 gene function was carried out using RNAi, an approach now feasible for schistosomes, in light of recent advances in knocking down schistosome genes [19,20]. BALB/c mice (females, 6 weeks old) were challenged percutaneously with 30 S. japonicum cercariae and, 6 weeks post-infection, mice were euthanised humanely and adult worms obtained by portal perfusion using 37°C RPMI 1640 medium. Adult worms were incubated in high glucose (4.5g/L) DMEM (Invitrogen, Carlsbad, CA, USA) medium, supplemented with 20% (v/v) heat-inactivated fetal calf serum, 100 IU/ml penicillin and 100 mg/ml streptomycin, at 37°C in an atmosphere of 5% CO2 in air overnight [13]. dsRNAs were transcribed in vitro from template PCR products using gene-specific primers tailed with the T7 promoter sequence. Luciferase dsRNA (dsLUC) was used as a negative control, as reported in all other RNAi studies with schistosomes [21–23]. SjIR dsRNAs were synthesised from S. japonicum cDNA using gene-targeted primers containing T7 promoter sequences as shown:
The L1 subdomain of SjIR-1 [13] (GenBank Accession No. GQ214553), termed SjLD1 (1059bp):
(F: 5′- GCATATGGACTGTTCCGGACGTTTACTGAATTTACGT -3′;
R: 5′- CGGAGCTCGTTCACAATCACGAATACTAATAAGGATTG -3′).
The L1 subdomain of SjIR-2 [13] GenBank Accession No. GQ214554), termed SjLD2 (1474bp):
(F: 5′- GGGATCCGCGTTGCACTGTCATAGAAGG -3′;
R: 5′- GCTCGAGTCACCAATTACAATAAGCTAAATCTCCATTTGT-3′).
dsRNA was synthesised and purified using a Megascript RNAi kit (Ambion, Foster City, CA, USA). For each group, 20–25 pairs in total were treated, with five pairs at a time electroporated in 50 μl electroporation buffer, containing 12.5 μg dsRNA, in a 4 mm cuvette by applying a square wave with a single 20 ms impulse at 125 v [21]. The four treatment groups included: SjIR1 dsRNA, SjIR2 dsRNA, SjIR1+IR2 dsRNA (6.25μg SjIR1 dsRNA+6.25μg SjIR2 dsRNA) and luciferase dsRNA. RNAi experiments were repeated independently three times. Following electroporation, parasites were transferred to 400 μl pre-warmed (37°C) complete, high glucose DMEM. After overnight culture, 600 μl fresh complete DMEM medium was added to each well. Two days after electroporation, each pair of worms was put in one well with 1ml fresh low glucose (1g/L) DMEM. Cultured medium was collected (20 μl) on days 4 and 6 after electroporation and the glucose concentration measured using a glucose assay kit (which can detect glucose in the range 1–10000μM) (Biocore, Gaithersburg, MD, USA), according to the manufacturer’s instructions. Untreated worms without electroporation were cultured under the same conditions as other groups. Cultured worms were collected on days 4 and 6 after electroporation for total RNA extraction and real time PCR analysis. Worms were also collected on day 6 after electroporation for extraction of proteins subsequently used for western-blot analysis as described below.
Total RNA was extracted from worms collected from the four different dsRNA-treated groups on days 4 and 6 as outlined above, followed by cDNA synthesis [15]. The cDNA was subsequently used as template in real time PCR analysis to determine the transcriptional levels of important genes in the insulin signalling pathway, including SjIR1, SjIR2, CBL E3 ubiquitin protein ligase (CBL), PI3K, SHC transforming protein 3 (SHC), glycogen synthase (GYS), GTP4 and GTP1. Real time PCR amplicons of SjIR1 and SjIR2 were located in the specific regions for schistosomes in the FnIII-3 domain of SjIR1 and in the tyrosine kinase (TK) domain between sub-domain IV and V of SjIR2, respectively, downstream of the regions used for dsRNAi in the ligand domains of the SjIRs [13]. Der1-like domain member 1 (AY814165) (contig8577) was used as reference gene which we have demonstrated has a stable transcription level in male and female worms of S. japonicum treated with or without insulin [15]. Primers were designed using the Primer 3 software (http://frodo.wi.mit.edu/) employing a target product size of 150–200 bp and a primer melting temperature of 60–65°C. Unique primers for each gene were designed to span exon-exon junctions (intron splice-sites) in the target mRNA. Primer design excluded outputs containing stretches of four or more identical nucleotides that might interfere with binding specificity. The specificity of each primer sequence was confirmed by BLAST analysis. All primer sequences used are shown in supplementary S1 Table. The % reduction in transcription of each target gene was calculated as: (1-copy number of target gene/copy number of luciferase) x 100.
Binding assays with recombinant SjLD1 and SjLD2, produced in the Drosophila S2 system, were performed in 96-well microplates at 25°C using the Octet Red system (ForteBio, Menlo Park, CA, USA) [26]. Human insulin was biotinylated using a NHS-PEO4-biotin kit (Thermo scientific, Rockford, IL, USA). Assays were carried out by placing the Streptavidin Biosensors (ForteBio) in the wells and measuring changes in layer thickness (in nanometers, nm) over time (in seconds). Firstly, a duplicate set of sensors were rinsed in kinetic buffer (1mM phosphate, 15mM Nacl, 0.1mg/ml BSA, 0.002% Tween-20) for 300 s which served as the baseline. Secondly, sensors were immobilized for 600s with 200μl culture containing biotinylated human insulin (12.5–50μg/ml). Thirdly, sensors were washed in kinetic buffer for another 600 s. Lastly, sensors were exposed to different samples run in 200 μl volumes in the same assay. These samples included rSjLD, negative control (BSA) and a positive control, a synthesised peptide from the alpha subunit of the HIR (α655–670) exhibiting specific binding activity with insulin [27]. To obtain reliable accurate equilibrium dissociation constants (KD), a dilution series of rSjLD1 (3.5–56 μg/ml) and rSjLD2 (1.78–56 μg/ml) were used in the association step. The association of rSjLD1 or SjLD2 with insulin was monitored for 1000s followed by dissociation in kinetic buffer for 1000s. Data analysis from the FortéBio Octet RED instrument included a double reference subtraction. Sample subtraction was performed using the BSA negative control and sensor subtraction was performed on all samples automatically with Octet User Software 7 [28].
SjLD1 and SjLD2, expressed in E. coli and purified, were used in two independent vaccination-challenge trials in mice. There were mice used in vaccine trial 1 and 2, respectively. Three groups of female CBA mice (6–8 weeks old, 6–10 mice/group; in total, 46 animals in trial 1 and 59 animals in trial 2) were used in each trial. Mice in the vaccinated groups were immunised intraperitoneally (i.p.) with 25 μg of rSjLD1 or 25 μg rSjLD2 protein in 0.1 ml of PBS homogenised with 20 μg of Quil A adjuvant (Superfos, Denmark) [17]. The mice were boosted i.p. twice at 2 week intervals with the same vaccine regimen. The control mouse group received PBS formulated with Quil A for the primary and two adjuvant boosts by the i.p. route. All mice were challenged with 14 ± 1 S. japonicum cercariae, counted under a microscope, by the abdominal skin route 2 weeks after the third injection. Serum samples were collected at 0, 2, 4, 6, 8, 10, 12 and 14 weeks after the first immunisation, to assess antibody responses by ELISA, using HRP-conjugated sheep anti-mouse IgG, IgG1, IgG2a, IgG2b and IgG3, IgE (1:2000 dilution) (Invitrogen) as primary antibody and HRP-conjugated sheep anti-mouse IgG as secondary antibody [17].
Worm numbers and egg burdens in livers, intestines and in faeces were determined as described [17] in the control and vaccinated mice at six and eight weeks post cercarial challenge to evaluate the vaccine efficacy of rSjLD1 and rSjLD2. Adult S. japonicum worm numbers were counted and all adult worms from each mouse were fixed and used for length measurement as described [17]. For each mouse, the left anatomical lobe of each liver and a 2 cm section of the small intestine were fixed in 4% (v/v) formalin. Paraffin-embedded sections of these samples were prepared and stained with haematoxylin and eosin (H&E). Slides were digitised using an Aperio Slide Scanner (Aperio, Germany) and the liver and intestine pathology was quantified by measurement of the volume density of granulomatous lesions using Aperio ImageScope v11.1.2.760 software. Faecal samples were obtained on 4 separate occasions from pooled mice (2–3 mice/faecal sample) from the vaccinated or control groups on the 2 days prior to perfusion to quantify schistosome egg output in faeces as described [17].
All data are presented as the mean ± SE. Differences between groups were assessed for statistical significance using the t-test. A statistically significant difference for a particular comparison was defined as a P value ≤ 0.05 (GraphPad Prism software (Version 6.05) was used for all statistical analyses.
To determine whether knock down of SjIR1 and SjIR2 resulted in gene silencing in adult S. japonicum via the RNAi pathway, parasites were treated with dsRNA of both gene targets. Both SjLD1 and SjLD2 are critical in the binding of SjIR1 and SjIR2 with insulin [17]. Parasites were electroporated with dsRNAs for SjIR1, SjIR2 or a mixture of both. Control parasites were treated with an irrelevant dsRNA (luciferase). Parasites were then cultured and collected on day 4 and 6 after electroporation for real time PCR analysis targeting SjIR1, SjIR2 and other S. japonicum genes implicated in glucose uptake and glycogen synthesis in the insulin signalling pathway. These genes included PI3K, GYS, SHC, CBL, GTP4 and GTP1. Fig 1 shows the transcript levels of these 8 genes on day 4 and 6 after electroporation. We found adult worms treated with dsRNA targeting SjLD1 alone exhibited a decrease in SjIR1 [of 41% (p = 0.001) on day 4 and 61% (p = 0.04) on day 6] and SjIR2 [of 38% (p = 0.03) on day 4 and 41% (p = 0.04) on day 6] gene expression. When adult worms were treated with dsRNA targeting SjLD2 alone, a decrease for both SjIR1 [of 60% (p = 0.001) on day 4 and 65% (p = 0.001) on day 6] and SjIR2 [of 46% (p = 0.01) day 4 and 42% (p = 0.02) on day 6] gene expression was evident. Parasites treated with a mixture of dsRNAs targeting both SjLD1 and SjLD2 exhibited a more pronounced reduction in SjIR1 (90%, p = 0.04) or SjIR2 (84%, p = 0.03) gene expression on day 6 compared with day 4.
On day 4 and 6 post-electroporation, there were 56% (p = 0.02) and 61% (p = 0.03) reductions in transcript levels of PI3K, respectively, when worms were treated with dsRNA targeting SjLD1 alone, and a 76% (p = 0.04) and 56% (p = 0.001) reduction, respectively, when worms were treated with dsRNA targeting SjLD2 alone. There was a consistent reduction (83–85%, p = 0.004) when both SjIR1 and SjIR2 were knocked down on day 4 and 6 after electroporation. The transcript level of GYS was reduced by 27% (p = 0.03) when SjIR1 was knocked down on day 6, but there was no effect by day 4. The knockdown of GYS expression by SjIR2 dsRNA dropped from 75% (p = 0.004) to 32% (p = 0.002) from day 4 to 6, indicating GYS responded earlier to a reduction in the level of SjIR2 and also suggested potentially different transcriptional kinetics of SjIR1 and SjIR2. When both SjIR1 and SjIR2 were knocked down in adult worms, the inhibition of GYS expression increased from 33% (p = 0.04) on day 4 to 97% (p = 0.04) on day 6. Both SHC and CBL genes exhibited an early response to the combined SjIR1 and SjIR2 knockdown and had a higher reduction [83% (p = 0.0001) and 95% (p = 0.0001), respectively] on day 4 compared with day 6.
Notably, the transcription of SGT4 increased on days 4 and 6 by 70–79% (p = 0.05) when SjIR1 was knocked down and by 281–299% (p = 0.04) when SjIR2 was knocked down, whereas there was no significant change in the transcription level of SGT1 on either day when either SjIR1 or SjIR2 was knocked down, both of which when compared to the irrelevant luciferase RNAi negative control. When worms were treated with dsRNA targeting both SjIR1 and SjIR2, SGT4 and SGT1 expression increased by 71% (p = 0.0001) and 157% (p = 0.02), respectively, on day 4 after treatment but expression of SGT4 and SGT1 was decreased by 24% (p = 0.0001) and 57% (p = 0.03), respectively, on day 6.
Glucose uptake in SjIR-suppressed adult worm pairs was compared with luciferase knockdown and unsuppressed worm controls. There was no difference in consumed glucose in worms treated with 0.25μg/μl SjIR1 or SjIR2 dsRNAs or a combination of SjIR1 and SjIR2 dsRNAs after 4 days incubation compared with the control group (Fig 2A). However, at 6 days post-treatment, the glucose consumed by each pair worm decreased significantly by 44% (p = 0.01), 46% (p = 0.023) and 38% (p = 0.013) in the SjIR1, SjIR2 and the combination of SjIR1 and SjIR2 suppressed worms, respectively, compared with the luciferase knockdown control group (Fig 2A). There was no difference in consumed glucose in untreated worms and in worms treated with luciferase dsRNA on days 4 and 6 (Fig 2). Of note, there was a significant 2.6-fold increase in glucose uptake by the luciferase knock down worms but there was no significant change in glucose uptake in the SjIR1, SjIR2 or the combined SjIR-knock down groups on day 6 compared with day 4 (Fig 2A). However, striking decreases in glucose uptake by worms were evident in the SjIR-knockdown groups over the last 2 days (days 4–6) of culture; glucose consumed by worm pairs decreased significantly by 60% (p = 0.03), 83% (p = 0.005) and 86% (p = 0.001) in the SjIR1, SjIR2 and the combined SjIR-knock down groups, respectively (Fig 2B).
To determine whether the knockdown of the SjIR1 and SjIR2 dsRNAs was mirrored at the protein level, we performed Western blot analysis using extracts of adult worms obtained 6 days post-treatment with dsRNA and anti-SjLD1 and anti-SjLD2 antisera. Both SjIR1 and SjIR2 were readily detected in an extract of luciferase knocked down control parasites, although the band recognised by the anti-LD1 antibody was weaker due to the fact there is lower expression of SjIR1 compared to SjIR2 in adult S. japonicum [13]. Markedly decreased levels of protein expression were evident in adult worms treated with SjIR1, SjIR2 and both SjIR1 and SjIR2 dsRNA compared with control worms (Fig 3A). The levels of the control Sm-Pmy protein expression did not change in any of the test or control groups, demonstrating that comparable levels of protein were present in each lane (Fig 3A).
To determine whether there was any cross reactivity between the SjLDs and HIR, we undertook further Western blot analysis using rHIR and rSjLDs and antibodies prepared against the recombinant proteins (Fig 3B). Anti-HIR antibodies only recognised rHIR at the expected molecular size of 135 kDa, with no cross reactivity with rSjLD1 or rSjLD 2, while neither the anti-SjLD1 nor the anti-SjLD2 antisera bound to rHIR. Both the anti-SjLD1 and anti-SjLD2 antibodies reacted with rSjLD1 and rSjLD2, expressed in E. coli (Fig 3B), at the expected molecular sizes of 45 kDa and 61 kDa, respectively.
As the binding between insulin and its insulin receptor is a structure-based interaction [29], it was necessary to obtain the rSjLD1 and rSjLD2 proteins in soluble form for the insulin binding assays. The Drosophila S2 cell system was thus used to produce both proteins in secreted, near native form; although protein expression was ~100 times lower than in the E. coli system. The sizes of the expressed rSjLD1 and rSjLD2 were the same (45 kDa and 61 kDa, respectively) as the two proteins expressed in the E. coli system. Real-time analysis using the Octet-RED system showed that there was specific interaction in vitro between human insulin and rSjLD1 or rSjLD2 expressed in the Drosophila S2 cell system (Fig 4). Increasing the concentration of the rSjLD proteins increased the binding response (Fig 4) and the dissociation phase revealed a slowly decreasing response, indicative of a specific interaction during the association phase. The KD values for the binding capacity of human insulin with rSjLD1 and rSjLD2 were 6.44E-08 and 8.78E-09, respectively, with stronger binding evident with the latter protein.
Both rSjLD1 and rSjLD2 generated solid anti-SjLD IgG antibody responses in vaccinated mice shown by ELISA after the second injection and these peaked after the third. The serum antibody titres dropped prior to perfusion of the mice 12 weeks after the first vaccination (Fig 5A and 5B). IgG1 and IgG2a antibodies were dominant (Fig 5C) whereas IgE was at background level throughout the two trials with rSjLD1 and rSjLD2. The sera collected from mice 6 weeks after the first immunisation with either rSjLD1 or SjLD2 recognised both SjLDs at the same titre (1:128,000), re-emphasising the sharing of similar epitopes by the two antigens as shown above and previously [17].
Parasitological data for the rSjLD1 and rSjLD 2-vaccinated and control mice challenged with 14±1 S. japonicum cercariae in two independent vaccine trials are presented in Tables 1 and 2. There were no significant (P>0.05) changes in the mean total worm burdens of the rSjLD-vaccinated and control groups in both trials. There were, however, significant (P ≤0.05) reductions in the vaccinated groups compared with controls for the following important parameters:
(i) The mean lengths (mm) of S. japonicum adult worms from the rSjLD1- and rSjLD2-vaccinated mice were reduced significantly at 8 weeks post-challenge (Tables 1 and 2). In trial 1, the length of male worms in the rSjLD1-vaccinated group decreased by 16%, while there was no significant change in the length of males in the rSjLD2-vaccinated group at 6 weeks post-challenge compared with males in the control group. At 8 weeks post-challenge, female worms were reduced in length by 14% and 15%, respectively in the rSjLD1 and rSjLD2 vaccinated groups; males were reduced in length by 17% in the rSjLD2 group. In trial 2, the length of both males and females decreased after 8 weeks post-challenge but no changes were observed at 6 weeks post challenge. Female worm lengths were reduced by 19% and 17% in the rSjLD1 and rSjLD-2 vaccinated groups, respectively, whereas males were reduced in length by 13% and 10.5% in the rSjLD-1 and rSjLD-2 vaccinated groups, respectively.
(ii) There were consistent and significant reductions in faecal eggs in the rSjLD1 (47–50%) and rSjLD2 (38–40%) vaccinated mice in the two trials (Tables 1 and 2). Importantly, in trial 1, the faecal egg burden in control mice increased 22-fold from 6 to 8 weeks post- challenge (Fig 6). In contrast, the number of faecal eggs in the rSjLD2-vaccinated mice was only increased 5.7-fold whereas there was no significant change in the faecal egg burden of mice vaccinated with SjLD1 during the 6 to 8 week period post-challenge. In the repeated trial (trial 2), during the 6 to 8 week period post- challenge, faecal egg numbers in control mice increased 7.8-fold, while there was only a 2.4 and 2.9 fold increase in egg burden in the SjLD1 and SjLD2 vaccinated groups, respectively, over the same period.
(iii) There was no difference in intestinal granuloma density between the SjLD-vaccinated and control mice at 6 weeks post-challenge but intestinal granuloma density was reduced significantly by 56% and 65% in the rSjLD1- and rSjLD2- vaccinated mice at 8 weeks post-challenge, respectively, compared with the control groups (Fig 6 and Tables 1 and 2).
To elaborate on previous studies and to further demonstrate that schistosomes exploit host insulin for growth and development, we have shown that human insulin can strongly bind the recombinant L1 subdomains (SjLD1 and SjLD2) of SjIR1 and SjIR2 expressed in Drosophila S2 cells. Schistosomes are reportedly unable to synthesize insulin [30], although an insulin-like peptide (Sjp_0020480, http://www.genedb.org) has been putatively annotated in the S. japonicum genome; however a function for this peptide has yet to be demonstrated and a comprehensive investigation of its characteristics is needed. The interaction between host insulin and the IRs of schistosomes implies it could be the first step in activating the insulin signalling pathway in these bloodflukes in a similar manner to that observed in mammalian cells. In addition, to further understand the potential roles for the IRs and their involvement in the insulin pathway in schistosomes, we have now shown that RNAi knockdown of SjIR1 or SjIR2, has an indirect effect in the down regulation of a number of key downstream genes in insulin signalling. We found that after knock down of both SjIR1 and SjIR2 in adult worms, the transcriptional suppression of both SjIR1 and SjIR2 was more pronounced in parasites with either the gene for SjIR1 or SjIR2 suppressed on day 6 post-treatment with dsRNA. We also found that only when both SjIR1 and SjIR2 were knocked down, there was significant regulation of SGT1 on day 4 and day 6 after treatment, further suggesting synergy between the two proteins. While the suppression of either SjIR1 or SjIR2 reduced the transcription of the other, this was not unexpected due to the relatively high level of sequence identity (35% at the nucleotide level) between SjLD1 and SjLD2, which are the specific domains in SjIR1 and SjIR2 dsRNAs that were respectively targeted. This cross suppression suggests SjIR1 and SjIR2 may share a similar function in insulin binding or in activating the insulin signalling pathway. This was also reflected in the immunological cross reactivity between SjLD1 and SjLD2 (Fig 3B) at the protein level, although the differential location of SjIR1 and SjIR2 suggests distinct and specialized functions. In addition, worms subjected to both SjIR1 and SjIR2 suppression showed a pronounced reduction in the transcript levels of the insulin signalling pathway-associated genes phosphoinositide-3-kinase (PI3K) and glycogen synthase (GYS) on day 6 post-treatment with dsRNA. This observation further suggests that knockdown of both insulin binding sites for SjIR1 and SjIR2 is more effective in blocking downstream signal transduction in the insulin pathway in different cellular locations given the tissue specific expression of SjIR1 is distinct to SjIR2, as observed in immunolocalisation studies [13]. This result also further supports our previous microarray analysis showing that PI3K is required for insulin-stimulated glucose transport in schistosomes [15]. Previous studies have demonstrated that chemical inhibition of PI3K, which plays an essential role in glucose uptake and GTP4 translocation, is intimately involved in cell growth and proliferation, and can completely block the stimulation of glucose uptake by insulin in mammalian systems [31,32]. Similar functions for insulin receptors have been demonstrated in the cestode Echinococcus multilocularis, where insulin was shown to stimulate the activation of the PI3K/Akt-pathway, leading to increased glucose uptake from the host and enhanced phosphorylation of the Echinococcus insulin signalling pathway components [33]. Furthermore, elevated expression of E. multilocularis IR1 (functionally related to SjIR2) in the cestode’s glycogen storage cells emphasised the important role of IR in regulating glycogen levels [33]. GYS is located downstream to PI3K in insulin signalling and is the key enzyme responsible for glycogen synthesis, catalysing the rate limiting step of UDP-glucose incorporation into glycogen [34]. When SjIR1 and SjIR2 were both subjected to knockdown by dsRNA in adult worms of S. japonicum, the expression level of GYS was reduced by 33% on day 4 and by 97% on day 6 post-treatment, suggesting that glycogen synthesis was highly suppressed over these 2 days, when control worm cultures experienced a surge in glucose consumption (Fig 2). This feature that was reflected in the striking decreased glucose consumption by each sample of SjIR1 and SjIR2 suppressed adult worms on days 4–6 (Fig 2B). These results are supported by previous studies of ours and those of others showing that the IR inhibitors HNMPA and tyrphostin AG1024 significantly decreased glucose uptake in schistosome worms [13,35,36].
As early response genes, SHC and CBL were reduced by 83% and 95%, respectively, on day 4 when both SjIR1 and SjIR2 were suppressed in S. japonicum worms. In mammalian cells the insulin receptor (IR) binds insulin, the activated receptor interacts with SHC and then the N-terminal PTB domain of SHC binds to the NPXY motif of the IR [37], which has also been shown present in SjIR2 [13]. SHC has been shown to compete with insulin receptor substrate (IRS) which mediates downstream signalling leading to glycogen synthesis as the substrate of the insulin receptor [37]. When CBL is phosphorylated by the IR, the translocation of the GPT4 protein can be elicited through the CAP/Cbl/TC10 pathway [38]. The decreased transcript levels of SHC and CBL as early responses (on day 4 post-treatment with dsRNA) after gene knock down of SjIR1 or SjIR2 suggest the same mechanism of signal transduction may occur in schistosomes as in mammalian cells.
It is noteworthy that we found the expression of SGTP4 was increased considerably when either SjIR1 or SjIR2 were knocked down on days 4 and 6 post-treatment with dsRNA. It is not surprising that the GTP4 expression levels increased as we have shown that incubation of adult S. japonicum with the IR-specific inhibitor HNMPA increased the level of GTP4 transcription [13]. Overall, these observations suggest that glucose uptake in schistosome parasites is dependent on phosphorylation processes that could be regulated by the activation of the IR. Knocking down either SjIR1 or SjIR2 strongly stimulated S. japonicum worms to express more SGTP4 as a mechanism to allow the acquisition of more host glucose. However, knock down of both SjIR1 and SjIR2 resulted in increased transcript levels of SGTP4 and SGTP1 on day 4 post-treatment as an early response, although the levels started to decrease by day 6. This modification in gene expression may have significantly impaired the ability of the S. japonicum worms to effectively consume glucose, thereby further emphasising the effect of the activated IR on glucose transport in schistosomes, although the precise mechanism involved remains to be determined. In this respect, it has been shown that the rate of glucose transport in schistosomes is also altered by acetylcholine interaction with tegumental acetylcholine receptors and acetylcholinesterase [39], suggesting that glucose uptake in these parasites may be modulated or regulated by multiple genes outside of the central insulin signalling pathway, although further investigation is required to determine how this gene regulatory network might operate.
Disruption of the insulin pathway in schistosomes would likely result in reduced glucose uptake which would in turn logically result in the starvation and stunting of worms with reduced egg output. So, we next moved to corroborate the important role of the SjIRs in the growth and fecundity of adult S. japonicum by using a modification of a vaccine strategy we have used previously [17], whereby we used a relatively low dosage of cercariae (14 instead of the usual 34) to challenge vaccinated mice. Reducing the cercarial challenge was a means to ensure a patent infection (ie at least one pair of egg-producing worms) in the mouse host, while minimising pathology so the animal would not succumb too early after infection. The fact that schistosome prevalence has been reduced to low levels in certain schistosome-endemic areas, as is occurring in China [40], requires a reappraisal of the use of low dosages of cercariae for challenge in vaccine/challenge experiments, so as to provide basic protective efficacy data on vaccines that will be used subsequently in the field.
Vaccination with either E. coli expressed rSjLD1 or rSjLD2 resulted in similar levels of growth retardation in females and males at 8 weeks post challenge infection and a consistent reduction in the number of faecal eggs. We previously reported a significant reduction in faecal eggs and in the number of mature intestinal eggs in rSjLD2-vaccinated mice after 6 weeks using the higher challenge dose of 34 cercariae [17]. In the current study, the reduction in faecal egg numbers was delayed until week 8 post-infection when vaccinated mice received the lower parasite challenge dose. The significant faecal egg reduction was, thus, not evident until the number of intestinal eggs accumulated to a certain level prior to release into the intestinal lumen. Further, the highly significant increased faecal egg numbers from weeks 6 to 8 after challenge in the control group compared to those observed in mice vaccinated with the SjLDs further demonstrates the effect that the SjLDs have in reducing the number of fecal eggs produced. Finally, the decreased intestinal granuloma density in mice receiving the SjLD vaccines implies a reduction in egg-induced pathology, which may be due to fewer mature eggs being produced in these vaccinated animals, a feature we have observed previously [17]. Encouragingly, recent studies on the insulin signalling pathway in E. multilocularis provide a mechanism whereby insulin signalling promotes parasite development [41]. The authors suggest that insulin signalling can stimulate the growth of juvenile or developing parasite by acting, in the case of Echinococcus, via stem cells. Stem cells, which have been identified in adult schistosomes [41], can differentiate into many cell types and likely play important roles in promoting asexual maturation of the juvenile [33]. However, the direct role of insulin signalling in this developmental mechanism has yet to be determined.
In conclusion, the evidence we present from protein binding assays, RNAi and vaccine/challenge experiments are strongly supportive of adult schistosomes having an insulin signalling transduction system. The insulin pathway in schistosomes is presumed to be first activated by the binding between host insulin and the parasite IRs with this binding then regulating the transcription of downstream genes, such as PI3K, GYS, SHC, CBL and GTPs, which are integrally involved in glucose metabolism in these blood flukes. RNAi showed that in vitro the glucose level of worms decreased when SjIR1 and SjIR2 were knocked down for 6 days compared with control parasites. The results from the complementary vaccine/challenge trials in mice strongly suggest that both rSjLD1 and rSjLD2 were able to induce a significant retardation in the growth of adult worms, presumably due to reduced glucose uptake, and a reduced faecal egg output. The eggs produced by the starved adult worms from mice vaccinated with the SjLDs, also lead to decreased intestinal granuloma density (Tables 1 and 2). These results are further supported by our previous study [17] showing that the poorly developed intestinal eggs in rSjLDs vaccinated mice, were less likely to be able to pass through the host intestinal wall into the intestinal lumen and reach the faeces. That may result in fewer viable eggs reaching the external environment and reducing parasite transmission.
In order to develop a safe, stable and effective vaccine based on SjIR1 and SjIR2, it is critical to further characterise the functionality of these proteins and to determine their precise biological importance to the parasite. Such investigations will greatly help in devising a specific and strongly protective vaccine effective against schistosomiasis. The protective effect of the SjIRs as vaccine antigens could be increased by the use of other adjuvants [42] and/or their co-immunisation with other key schistosome components as multi-epitope constructs [43]. Further, combination vaccine of SjLD1 and SjLD2 may improve protective efficacy by inducing antibodies that block both insulin binding sites on the SjIRs. Furthermore, in order to investigate the feasibility of the SjLDs as potential transmission blocking vaccines, it will be necessary to test their vaccine efficacy in bovines, notably water buffaloes, which are the major reservoirs for zoonotic schistosomiasis in Asia being responsible for 75% of disease transmission in S. japonicum-endemic areas.
|
10.1371/journal.pbio.1000054 | Stepwise Development of MAIT Cells in Mouse and Human | Mucosal-associated invariant T (MAIT) cells display two evolutionarily conserved features: an invariant T cell receptor (TCR)α (iTCRα) chain and restriction by the nonpolymorphic class Ib major histocompatibility complex (MHC) molecule, MHC-related molecule 1 (MR1). MR1 expression on thymus epithelial cells is not necessary for MAIT cell development but their accumulation in the gut requires MR1 expressing B cells and commensal flora. MAIT cell development is poorly known, as these cells have not been found in the thymus so far. Herein, complementary human and mouse experiments using an anti-humanVα7.2 antibody and MAIT cell-specific iTCRα and TCRβ transgenic mice in different genetic backgrounds show that MAIT cell development is a stepwise process, with an intra-thymic selection followed by peripheral expansion. Mouse MAIT cells are selected in an MR1-dependent manner both in fetal thymic organ culture and in double iTCRα and TCRβ transgenic RAG knockout mice. In the latter mice, MAIT cells do not expand in the periphery unless B cells are added back by adoptive transfer, showing that B cells are not required for the initial thymic selection step but for the peripheral accumulation. In humans, contrary to natural killer T (NKT) cells, MAIT cells display a naïve phenotype in the thymus as well as in cord blood where they are in low numbers. After birth, MAIT cells acquire a memory phenotype and expand dramatically, up to 1%–4% of blood T cells. Finally, in contrast with NKT cells, human MAIT cell development is independent of the molecular adaptor SAP. Interestingly, mouse MAIT cells display a naïve phenotype and do not express the ZBTB16 transcription factor, which, in contrast, is expressed by NKT cells and the memory human MAIT cells found in the periphery after birth. In conclusion, MAIT cells are selected by MR1 in the thymus on a non-B non-T hematopoietic cell, and acquire a memory phenotype and expand in the periphery in a process dependent both upon B cells and the bacterial flora. Thus, their development follows a unique pattern at the crossroad of NKT and γδ T cells.
| White blood cells, or lymphocytes, play an important role in defending the body from infection and disease. T lymphocytes come in many varieties with diverse functions. Mucosal-associated invariant T (MAIT) cells constitute a subset of unconventional T lymphocytes, characterized by their invariant T cell receptor (TCR)α chain and their requirement for the nonpolymorphic class Ib (MHC) molecule, MR1. MAIT cells are extremely abundant in human blood and mucosae. Contrary to mainstream T cells, their development requires B cells and commensal microbial flora. To shed light on the little-understood MAIT cells, we used new tools, including an antibody that we recently developed to detect human MAIT cells, and we were able to show that MAIT cell development is a stepwise process, with an intra-thymic selection followed by peripheral expansion. We show that thymic selection is MR1 dependent but requires neither B cells nor the commensal flora, which are both necessary for the expansion in the periphery. In contrast with the other evolutionarily conserved invariant subset, the natural killer T (NKT) cells, we found that MAIT cells exit the thymus as “naïve” cells before becoming antigen-experienced memory cells and expanding in number to represent a significant 1%–4% of peripheral T cells in human blood. In mice, we found that MAIT cells remain naïve and do not expand substantially. We conclude that MAIT cell development follows a unique scheme, where, unlike NKT cells, MAIT cell selection and expansion are uncoupled events that are mediated by distinct cell types in different compartments.
| Unconventional T cells include several T cell receptor (TCR)αβ+ major histocompatibility complex (MHC) class Ib restricted, as well as TCRγδ+ subsets in both mice and humans [1,2]. They usually display tissue-specific location and are endowed with “natural memory” phenotype and functions, such as the ability to mount rapid responses in the face of a pathogenic challenge. Several pathways of development have been described for these T cell subsets, which show various degree of dependency from the thymus as well as variable requirement for their restricting MHC molecule. At one end of the spectrum, natural killer T (NKT) cells are selected, expand, and acquire their innate-like phenotype and functions in the thymus [3,4]. On the other hand, unconventional (type b) intra-epithelial lymphocytes (IEL) ontogeny can show minimal dependency upon the thymus, as they can escape the thymus at a very early stage and migrate into the gut mucosa where they achieve maturation [5]. They may even develop directly from bone marrow-derived precursors in specific intestinal lymphoid aggregates called cryptopatches [6,7]. Interestingly, both NKT and type b IEL development, phenotype, and functions are seemingly independent from any exogenous antigens such as those derived from the intestinal bacterial flora [8–10]. Finally, it has been recently demonstrated that mouse T10/T22-restricted γδ T cells are able to mature and exit the thymus in the absence of their selecting element, but that thymic selection endows them with a memory phenotype and new effector functions [11].
Among unconventional T cells, only two subsets display both a TCR and selecting MHC class Ib molecules highly conserved between species, the NKT cells and the mucosal-associated invariant T (MAIT) cells (reviewed in [3,4,12]). Indeed, these two populations express highly restricted TCR repertoires consisting of an invariant TCRα chain (mouseVα14/humanVα24-Jα18 for NKT cells; and mVα19/hVα7.2-Jα33 for MAIT cells). Both subsets are selected by hematopoietic cells expressing evolutionarily conserved nonpolymorphic MHC class Ib molecules, encoded in humans in a MHC paralogous region on Chromosome 1: CD1d for NKT cells [13] and MHC-related molecule 1 (MR1) for MAIT cells [14]. NKT cells accumulate in the liver and the spleen, independently of the presence of any exogenous stimuli such as the normal bacterial flora [8], while MAIT cells accumulate primarily in the intestinal lamina propria (LP), in a process dependent on the commensal flora [14]. MAIT cell function is unknown, but the unusual features of these cells and their evolutionary conservation suggest an important role at the interface between innate and adaptive immunity, probably in the intestinal immune system.
For NKT cells, the molecular basis of their specific development begins to be unraveled: contrary to mainstream T cells and γδ T cells, NKT cell development requires the integrity of the SLAM/SAP/fyn pathway with the involvement of homotypic interactions of SLAM receptor family members and CD1d expression on CD4+/CD8+ (DP) thymocytes [15]. Consequently, NKT cells are absent in mice and humans deficient in the molecular adaptor SAP [16–19]. In addition, mouse and human NKT cells specifically express the ZBTB16 transcription factor very early on during their development and the absence of this factor in luxoid mice leads to a quasi-disappearance of the NKT cells as only a very few number of naive NKT remains [20,21].
Little is currently known about MAIT cell development. They are absent from nude mice [22], but MAIT cells are not in sufficiently large number in the thymus for detection by highly sensitive reverse transcription-(RT)-PCR methods in either humans or wild-type (wt) mice ([22] and unpublished data). Previous studies on MAIT cell selection have been based on the measure of the number of these cells in the periphery (in the mesenteric lymph-node [MLN] and in the LP) [14]. MAIT cell development does not require expression of their selecting element on thymus epithelial cells, but the presence of MR1 on B cells is probably necessary for accumulation of MAIT cells in the gut LP [14]. It is therefore unclear whether MR1-dependent selection of MAIT cells really occurs in the thymus, and if so, whether B cells and/or the bacterial flora are required for MAIT cell development in the thymus or peripheral accumulation in the gut.
Herein, we investigated MAIT cell development in both humans and mice. In humans, we used an anti-Vα7.2 monoclonal antibody (mAb) generated in our laboratory to track normal unmanipulated MAIT cells. MAIT cells are rare and display a naïve phenotype both in human thymus and cord blood, while they are abundant and show a homogeneous memory/effector-like phenotype in the blood of all healthy subjects tested. In mice, the over-expression of one of the TCR chains was sufficient to increase the frequency of MAIT cells in central and peripheral lymphoid organs in an MR1-dependent manner. Fetal thymic organ culture (FTOC) and phenotypic studies of the thymus in these mice demonstrated that MAIT cells are selected in the thymus in the absence of B cells and bacterial products. However, most mouse MAIT cells remained naïve in the periphery. The complementary data obtained from humans and mice indicate that MAIT cells are exported from the thymus as naïve cells and subsequently become memory and expand, following interactions with B cells and the commensal flora, in an SAP-independent manner. Thus, MAIT cells use a developmental pathway distinct from both known unconventional T cell subsets and mainstream T cells, showing that they represent a unique subset with its own features.
As MAIT cells are defined by the use of an invariant iVα7.2-Jα33 TCRα chain, we developed a monoclonal antibody (3C10) recognizing the TCRα Vα7.2 segment by immunizing Balb/c mice with a recombinant Vα7.2-Jα33/Vβ13 recombinant protein (see Methods). We then used this antibody to measure the number of cells expressing the Vα7.2 segment in blood T cells from healthy donors by assessing the proportion and phenotype of 3C10+ cells in the CD4, CD8β and DN subsets of CD3+/TCRγδneg lymphocytes (Figure 1A). Most 3C10+ DN (which also includes CD8αα cells) T cells were also found to display strong expression of the CD161 (NKRP1A) marker, and the percentage of 3C10+/CD161hi T cells in the DN + CD8αα subset (40.3 ± 21%, m ± SD; range = 1–98%) was very high. In CD8β T cells, the proportion of 3C10+/CD161hi cells was variable but significant (7.8 ± 5.5%; 0.2–26.3), while within CD4 T cells, the proportion of 3C10+/CD161hi cells was low (0.34 ± 0.34%; 0–1.4), with only a minority of 3C10+ T cells also expressing CD161. Reverse transcription (RT)-PCR and polyclonal sequencing of Vα7.2-Cα amplicons obtained from the different FACS-sorted subsets showed that the canonical iVα7.2-Jα33 sequence segregated with the 3C10+/CD161hi cells (Figure S1 and unpublished data). The presence of 3C10+/CD161neg CD4 and CD8β T cells with no canonical sequence expression demonstrates that 3C10 monoclonal antibody (mAb) is not a clonotypic antibody, but is instead specific for the Vα7 segments, which can be used by mainstream T cells. We were surprised to observe that a significant proportion of CD8α+/3C10+/CD161+ T cells also expressed the CD8β molecule, in conflict with our previous findings [22]. However, a careful analysis showed that the MFI for CD8β in 3C10+/CD161hi T cells was only one-quarter to one-sixth that in 3C10−/CD161− T cells (Figure S2), suggesting that we missed this subset of cells in our previous FACS-sorting experiments. Interestingly, no such difference in MFI was seen for CD8α (Figure S2), suggesting that at least some of the CD8αβ+/3C10+/CD161+ MAIT cells also express the CD8αα molecule. These results show that MAIT cells can be identified as 3C10+/CD161hi T cells, and confirm that they are abundant within the DN and CD8αα+ subsets, and scarce in the CD4+ T cell subset. However, a large number of these cells may also express an intermediate level of CD8αβ.
The high numbers of DN+CD8αα and CD8βMAIT cells were similar in 104 blood donors, whereas the CD4 subset contained only one-tenth as many such cells (Figure 1B). In adult blood, these cells displayed a homogenous CD45RAneg/RO+/27+/62Llo/CCR7−/CXCR6+ phenotype (Figure 1C and unpublished data) suggesting that blood MAIT cells are memory cells that can migrate into the intestinal tissues.
We used the aforementioned staining strategy to assess the presence of MAIT cells in the human thymus, and observed some 3C10+ DN or single-positive CD8β and CD4 T cells (Figure 2A, upper panels). However, by contrast to what was observed with blood (Figure 2A, lower panels), most of the 3C10+ cells in the thymus did not express CD161. To assess the identity of these cells, we quantified Vα7.2-Jα33 amplicons in sorted CD161-positive and CD161-negative DN, CD4 and CD8β 3C10+ thymocytes, using CD161neg/CD4+ and CD161+/DN blood 3C10+ T cells as negative and positive controls, respectively. In two of the three thymuses studied we found a small excess of Jα33 usage (8%–12%) in the CD161neg 3C10+ thymocytes (Figure 2B). By contrast, Jα33 was used by 50% to 100 % of the few 3C10+/ CD161hi DN thymocytes in the two thymuses containing 3C10+/CD161hi T cells. Almost half of this small number of 3C10+/CD161hi cells had a naïve (CD45ROlo, CD45RA+, CD27+) phenotype (Figure 2C) ruling out the possibility of blood contamination or recirculating cells. The presence of a few naïve T cells expressing the canonical iVα7.2-Jα33 TCRα chain in the thymus suggests that MAIT cells develop within the human thymus, although this remains to be definitively demonstrated. We therefore used mouse models for the further characterization of MAIT cell ontogeny.
MAIT Cells Ontogeny in Mice Transgenic for the iVα19 TCRα, MAIT Cell Vβ6 TCRβ Chains or Both
Mice have very few MAIT cells and the lack of an antibody specific for the invariant mVα19 TCRα chain precludes the direct analysis of MAIT cells in vivo or ex vivo. We therefore generated mice expressing an invariant mVα19-Jα33 TCRα chain transgene (Tg). A detailed analysis of these mice and their phenotype are provided as supporting information (see Figure S3A and Text S1). As also shown by others [23], the transgenic over-expression of the invariant TCRα chain induces a clear MR1-dependent bias in Vβ chain segment usage towards Vβ6 and Vβ8, reproducing the MAIT cell repertoire of our iVα19 T-T hybridoma [22]. Comparison of Vβ6/Vβ8 bias in the presence or absence of MR1 provided us with an estimate of T cell selection by MR1: at least 50% of the T cells in these mice are MR1 restricted (see Figure S3B, S3C, and Text S1). As shown by others [23] and in the Supporting Information (Figure S3C) MR1-restricted T cells are found in the thymus of iVα19 Cα−/− Tg mice as indicated by a MR1 dependent Vβ6 and/or Vβ8 bias observed in all CD4/CD8/DN T cell subsets.
The forced transgenic expression of a TCRα chain greatly modifies T cell ontogeny [24,25]. We therefore used an alternative strategy based on the transgenic expression of a Vβ6+ TCRβ chain from one of the iVα19 T-T hybridoma. As shown previously for mainstream T cells [26] and NKT cells [27], the frequency of the iVα19-Jα33 TCRα chain was increased in these mice in a MR1-dependent manner (see Figure S3D and Text S1).
To further increase the proportion of MAIT cells and to decrease the unwanted TCR specificities due to the endogenous TCRβ chains, we crossed the TCRα Tg and TCRβ Tg mice together in the presence or absence of MR1. The thymuses of the resulting mice were a little smaller than those of wt, in both backgrounds (53 mean [M] ± 26, ± standard deviation [SD], n = 15). In the presence of MR1, the mature thymocytes (TCRhi/HSAlo) were in low numbers in comparison with B6 mice and were mostly DN and CD8 with a concomitant reduction in the CD4 T cell subset (Figure 3A). In the absence of MR1, the frequency and number of mature (TCRhi/HSAlo) thymocytes decreased by a factor of six to eight, indicating that MR1-restricted MAIT cells are present in the thymus of the MR1+ TCRαβ double Tg mice. Importantly, peripheral T cells from these mice strongly up-regulated CD69 expression when cocultured with murine MR1-transfected fibroblasts, confirming that these cells are indeed reactive towards MR1 (S. Huang, EM, S. Kim, L. Yu, CS, et al., unpublished data).
In all the mice studied above, the mature thymocytes were in very small numbers and could be recirculating T cells generated elsewhere. We therefore studied FTOC from iVα19 Tg Cα−/− mice in the presence or not of MR1 to demonstrate formally the intra-thymic development of MAIT cells. Negligible numbers of CD4 T cells were generated whereas the proportion of CD8 and DN T cells was very high among the mature (TCRβhi/HSAlo) thymocytes (Figure 3B), similarly to the phenotype found in the adult thymus of mice of the same genetic background although the number of DN cells was lower in adult thymus (Figure S4). Both DN and CD8 T cells displayed a Vβ6 and Vβ8 bias that was dependent on MR1 expression as shown by the significant decrease in Vβ6 (p < 0.023 and 0.002 for DN and CD8, respectively) and Vβ8 (p < 0.0001) expression in the absence of MR1 (Figure 3C). This MR1 dependent Vβ6/8 usage bias indicates a MR1 dependent selection of many T cells in these FTOC of iVα19 Tg mice. Altogether, these experiments formally demonstrate the existence of an intra-thymic development pathway for MAIT cells.
We have previously shown that accumulation of MAIT cells in the periphery depends on the presence of B cells [14]. In contrast with the presence of around 1% B cells in adult thymus, no CD19+ cells were detected in the FTOC (unpublished data), suggesting that B cells may not be necessary for thymic MAIT cell selection. To further address the role of B cells in MAIT cell ontogeny, we crossed the iVα19 TCRα and Vβ6 TCRβ Tg lines onto a RAG−/− background, thereby generating mice producing monoclonal MAIT cells in the absence of B cells. In the presence of MR1, the thymus was extremely small (4–6 × 105 cells) with very few mature (TCRβhi/HSAlo) cells (Figure 4A, left upper panel), which were either DN or CD8, and only a minimal number of CD4 T cells (Figure 4A, left lower panel). As T cells do not develop at all in the absence of the restricting MHC molecule in mice transgenic for classical TCRs [28,29], this result is compatible with MR1 expression on cells other than B cells. In accordance with this hypothesis, RAG−/− TCRαβ double TCR transgenic mice harbored no mature thymocytes in the absence of MR1 (Figure 4A, right upper and lower panels). Thus, MAIT cells can develop intra-thymically in a MR1-dependent but B cell-independent fashion.
The periphery of MR1−/− RAG−/− TCRαβ double Tg mice was almost completely devoid of mature T cells (1.5 ± 0.5 × 103, m ± standard error [SE], n = 4) (Figure 4B, upper left panel, and 4C), whereas in the presence of MR1 only a small subset of T cells was found in the MLN (1.3 ± 0.5 × 105, n = 7), blood or spleen (Figure 4B upper right panel, 4C, and unpublished data). This suggests that in the absence of B cells, selected MAIT cells are not able to expand and accumulate in the periphery. To directly assess this issue, we performed the following “add-back” experiment: we injected splenocytes devoid of T cells from MR1+ or MR1−/− CD3ε−/− mice intravenously into MR1+ or MR1−/− RAG−/− TCRαβ double Tg mice, and counted the T cells in the MLN after 2 wk. As shown in Figure 4B (lower panel), the splenocytes transfer allowed the same level of reconstitution of B cells in both MR1+ and MR1−/− background. The transfer of either MR1− or MR1+ B cells into MR1 deficient hosts allowed a clear expansion of the T cells, which however remains in low numbers (5.5 ± 1.3 × 104, n = 3 and 5.4 ± 1.8 × 104, n = 5, respectively) (Figure 4B, lower middle and right panels, and Figure 4C). This expansion, which was similar with MR1− and MR1+ B cells, could either be related to some trophic effect of the B cells or to the modifications of the lymphoid organs they induced. By contrast, when MR1+ B cells were transferred into MR1+ RAG−/− αβTCR double Tg mice, we found much higher number of T cells (1 ± 0.26 106, n = 10) in the periphery (Fig 4B, 4C, and unpublished data). However, MR1 deficient B cell transfer also induced some T cell expansion (3.4 ± 0.8 × 105, n = 8) in the periphery of the MR1+ TCRαβ double Tg mice. Thus, noncognate interactions (or MR1-independent cognate interactions with membrane molecules exclusively expressed by B cells) between MAIT and B cells might be sufficient to allow MAIT cell expansion. Alternatively, the MR1neg B cells may have captured some MR1 molecules from the MR1+ hosts [30] as they are in a “sea” of MR1+ cells. However, the much larger number (8-fold with MR1+ B cells versus 2.7-fold with MR1− B cells) of MAIT cells found after the transfer of B cells into MR1+ RAG−/− αβTCR double Tg mice suggests that cognate interactions are likely involved.
Therefore, in mice where MAIT cells are selected in the thymus but do not expand in the periphery, the adoptive transfer of MR1+ B cells is sufficient to promote their accumulation. Interestingly, in all cases MAIT cells acquired a memory (CD44hi) phenotype after transfer of B cells (unpublished data). The acquisition of a memory phenotype might be related to a “homeostatic” expansion due to the sudden provision of selecting niches. In any cases, although it could be argued that the splenocyte mixture we injected contains both B cells and other cell types, such as dendritic cells or macrophages, the latter are already present in the host mice before transfer and are obviously not able to induce the peripheral accumulation of MAIT cells. We can therefore formally conclude that B cells are necessary to allow MAIT cells accumulation in the MLN of these monoclonal Tg mice, either through survival, expansion, and/or addressing of MAIT cells to the intestinal territory.
The accumulation of murine MAIT cells in the intestinal compartment also requires the presence of the commensal flora [14]. The generation of MAIT cells in FTOC in the absence of exogenous ligand suggests that the commensal flora is necessary not for thymic selection but for the migration and/or peripheral expansion of MAIT cells. We investigated this issue in humans by studying the presence and the phenotype of MAIT cells in immunologically naïve cord blood. The number of CD161hi/3C10+ MAIT cells was small but measurable in cord blood DN and CD8β T cells (Figure 5A). Polyclonal sequencing of sorted CD161hi/3C10+ DN cord blood lymphocytes confirmed the presence of the canonical sequence (Figure S6). Strikingly, unlike their adult counterparts, these cord blood MAIT cells displayed a naïve (CD45RAhi/CD27hi/CD45ROlo) phenotype (Figure 5B), making the possibility of a materno-fetal transfusion highly unlikely and demonstrating the presence of naive MAIT cells in human cord blood. Altogether, these results suggest that MAIT cells acquire CD161 expression right before (or concomitantly to) their exit from the thymus, and remain naïve until birth. Colonization with commensal bacteria after birth induces the expansion and/or final maturation of MAIT cells in the periphery.
We next addressed the question of the naïve/memory status of mouse MAIT cells, which are in much lower number than human MAIT cells. Indeed, in the different Tg mice studied above, the DN and CD8 mature thymocytes are mostly naïve (CD44lo/CD122lo), while the low numbers of CD4 T cells are constantly CD44hi (Figure S7). As shown and discussed in Figure S5, some of the Vβ6+ CD4 T cells found in iVα19 Tg mice are not MR1 restricted but classical MHC class II dependent. To avoid this confounding variable and to increase the proportion of MR1 restricted T cells, we crossed our Tg mice to TAP−/−Ii−/− double KO mice, which, being devoid of classical MHC molecules, would select lower number of mainstream T cells. In the absence of classical MHC molecule (Cα−/−TAP−/− Ii−/−), the MLN T cells found in the iVα19/Vβ6 double Tg mice displayed a naive (CD44lo) (Figure 5B) and CD122lo (unpublished data) phenotype. In the absence of MR1, the proportion and the number of CD8 T cells greatly decreased (Figures 5C and S8A) and some increase in CD44 expression was observed with a concomitant increased usage of endogenous Vβ segments (Figure S8B). These results suggest that, in the absence of MR1, pairing of the iVα19 TCRα chain with endogenous TCRβ chains allows selection and peripheral expansion. The proportion and the phenotype of the few CD4 T cells found in these mice remained unchanged in MR1+ or MR1− mice indicating that many of these cells were not MR1 dependent.
To further address the question of the naïve/memory phenotype of MAIT cells, we sorted CD44hi and CD44lo/int T cell subsets from the MLN of TAP−/−/Ii−/− (to enrich in MAIT cells) and Vβ6 Tg mice on a MR1+ or MR1−/− background and quantified the amount of iVα19 transcripts. In both strain of mice, we found that the iVα19 transcripts largely segregated in the CD44lo/int fraction, confirming that wt polyclonal MAIT cells indeed display a naïve phenotype in mice (Figure 5D).
Altogether, these results indicate that murine MAIT cells, like their human counterparts, are selected by MR1 in the thymus without acquiring a memory phenotype. However, they remain naïve in the periphery, even in the presence of B cells and the commensal flora. The difference in number and phenotype of MAIT cells between humans and mice might then be related to the absence of peripheral expansion in the latter.
Contrary to mainstream T cells, NKT cells express the ZBTB16 transcription factor from the first stage of their differentiation in the thymus. In mice deficient for this transcription factor, NKT cells remain naïve, do not expand, and do not colonize the effector organs such as the liver but are found in small numbers in the LN [20,21]. The transgenic forced expression of ZBTB16 controlled by a CD4 promoter induces an effector phenotype in conventional CD4 T cells, indicating that ZBTB16 expression is probably the result of the productive interaction of the iVα14 TCR with CD1d and is responsible for the effector/memory phenotype of NKT cells. The only T cell subset expressing the ZBTB16 transcription factor in addition to the NKT cells was the human MAIT cells [21]. However, we found that mouse MAIT cells did not express ZBTB16 in the periphery (Figure 6A), in accordance with their naïve phenotype. Thus, in contrast with NKT cells, which express ZBTB16 at the early stages of their thymic selection/expansion, ZBTB16 expression in MAIT cells is not related to their selection process. The expression of this transcription factor in MAIT cells is probably linked to an activation step followed by the acquisition of a memory phenotype and peripheral expansion.
Although both NKT cells and MAIT cells are selected on hematopoietic cells, the former subset expands and acquires a memory phenotype within the thymus, whereas MAIT cells seemingly show a more conventional selection process. We sought to investigate the molecular basis for this divergence by studying the SAP-dependency of MAIT cells development. Indeed, CD1d-restricted NKT cells ontogeny involves signaling through the SAP/Fyn/NF-κB signaling pathway, which is triggered by homotypic interactions between SLAM family molecules expressed both on developing NKT cells and selecting cortical thymocytes [15]. We therefore measured the number of MAIT cells in five SAP-deficient patients: whereas no Vα24 NKT cells could be found in these patients, as previously described [16–18,31], MAIT cells were present in normal numbers (Figure 6B and 6C), indicating a probable lack of involvement of the SLAM/SAP/Fyn pathway in MAIT cell development. Interestingly, ZBTB16 was normally expressed by MAIT cells from SAP-deficient patients, confirming that the SLAM/SAP/Fyn pathway is not involved in the induction of this transcription factor expression (Figure 6D). Thus, MAIT cell development clearly diverges from CD1d-restricted NKT cells, despite their common selection on hematopoietic cells [14].
MAIT cells constitute a new subset of MHC class Ib-restricted T lymphocytes conserved among mammals. In this study, we provide for the first time, to our knowledge, phenotypic data on human MAIT cells. We show that they can be tracked by costaining with an anti-Vα7 antibody (3C10) and CD161, allowing us to assess their frequency in the peripheral blood of healthy subjects. Blood MAIT cells are numerous (at least one order of magnitude higher than NKT cells), in accordance with our previous estimates [22]. As previously published, a majority of MAIT cells display a CD4−CD8− (DN) or CD8αα phenotype, but few of them are CD4+. We now show that MAIT cells can express intermediate levels of CD8αβ, which were missed in our first characterization. The low level of CD8αβ expression may allow developing CD8αβ+ MAIT cells to escape negative selection. These data imply that selection of MAIT cells is independent of coreceptors, but that an Ag-driven process induces differential expansion of the different subsets, as also suggested by the differences in Vβ usage between DN/CD8 and CD4 T cells in the MLN and LP of iVα19 TCRα Tg mice.
Two other groups have generated iVα19 TCRα Tg mice [23,32,33]: in only one of these studies [23], the comparison was made between MR1+ and MR1-deficient backgrounds allowing the distinction between selection-driven features and those related to the transgenic artifact. Our results are similar in terms of organ cellularity, CD4/CD8/DN subset distribution, and Vβ6/8 bias. The careful study of the phenotype of our mice in the different background allowed us to show that MAIT cells are DN and CD8 and not much CD4. We also observed no significant expression of NK1.1, CD25, CD69, or ICOS on the T cells of iVα19 or Vβ6 single or double Tg mice (unpublished data). Moreover, we studied the phenotype of wt murine MAIT cells and found that NK1.1 is expressed on <30%–50 % of MAIT cells from the LP of wt C57Bl/6 mice (unpublished data). Most significantly, unmanipulated human MAIT cells express none of these markers, given that CD161 (NKR-P1A) is not the human ortholog of NK1.1 (NKR-P1B/C in B6 mice). The differences observed between our mice and those studied in previous investigations are probably due to experimental procedures, such as differences in the transgenic vectors used or to differences in the commensal flora of different animal facilities as MAIT cells require commensal flora to expand [14]. In any case, we believe that these markers, including NK1.1 in particular, cannot be used to reliably track MAIT cells in vivo.
We have previously shown that MAIT cells accumulate in the MLN and the gut LP. However, MAIT do not accumulate in sufficient numbers in the thymus of wt mice to be detectable by sensitive reverse transcription (RT)-PCR methods, precluding the possibility to discriminate between intra-thymic selection per se and peripheral expansion. Moreover, it has been suggested recently that the development of some unconventional intra-epithelial lymphocyte (IEL) T cell populations requires a functional thymus but actually takes place in the gut mucosa [5,7], while T10/T22-restricted γδ T cells can mature and migrate in the periphery in the absence of thymic selection [11]. Herein, we demonstrate for the first time, to our knowledge, that MAIT cells are indeed selected in the thymus in an MR1-dependent manner, but that B cells and bacteria are not required for their initial intra-thymic generation, whereas both are necessary in the periphery for MAIT cell expansion. Therefore, MAIT use a unique stepwise developmental pathway that is distinct from both NKT and conventional T cells.
Our data clearly demonstrate that B cells are not the selecting cells in the thymus, but are necessary for the accumulation of MAIT cells in the MLN and LP, by promoting either expansion, survival, and/or addressing to the gut compartment. We have previously shown that MAIT cells are selected on non-T hematopoietic cells [14]; therefore our data implicate most probably myeloid cells such as macrophages or dendritic cells in intra-thymic MAIT cell selection.
Recent data from Bendelac and coworkers and from our own lab show that MAIT cells and NKT cells exclusively share the expression of the transcription factor PZLF (ZBTB16) [21], emphasizing a close lineage relationship between these two distinct subsets. However, the data presented in this paper nonetheless describe profound differences during the ontogeny of these two populations. We clearly show that MAIT cell selection is not accompanied by activation (as is the case with NKT cells [34,35]), since it does not lead to local expansion and acquisition of a memory phenotype. Accordingly, ZBTB16 was not expressed by mouse MAIT cells (Figure 6A), which display a naïve phenotype while human MAIT cells acquired ZBTB16 together with a memory phenotype after birth. It has been also shown recently that the ontogeny of NKT cells is heavily dependent upon homotypic interactions between SLAM family members and the SAP signaling pathway [15]. We show here that MAIT cells development is apparently unaffected by SAP mutations in humans, and this might be related to the fact that postselection MAIT cells display a naïve phenotype in both humans (Figure 2C) and mice (Figure 5C). In agreement with the mouse data showing that SAP deficiency does not prevent the expression of the ZBTB16 molecule at the early stages of NKT ontogeny, ZBTB16 expression in MAIT cells was unaffected in SAP deficient patients.
The absence of ZBTB16 expression by mouse MAIT cells could be related to the cleanness of the animal facilities, which would not provide the ligand or inflammatory context necessary for MAIT cell expansion. Alternatively, one key genetic component may be missing in mice because of the genetic bottleneck that laboratory mouse strains have been through without selection pressure by the putative function mediated by the MAIT cells. Finally, how ZBTB16 expression is acquired in the thymus by NKT cells and in the periphery by human MAIT cells is an open question. It is probably related to the context of the interactions between the iTCR and the selecting element.
The acquisition of a memory phenotype by human MAIT cells after birth is most likely linked to the colonization of the gut (and other mucosae) with the commensal flora. We speculate that B cells provide the link between the gut bacterial flora and MAIT cells. Direct or indirect (via epithelial cells or dendritic cells [36,37]) interactions between B cells and bacteria in the gut could either induce MR1 expression and/or endow B cells with specific costimulatory properties, akin to the SLAM-mediated costimulation of developing NKT cells. However, the absence of ZBTB16 expression, and the concomitant naïve phenotype of mouse MAIT cells implicate that some other signals are required for their full innate-like differentiation. The dependency upon bacteria of MAIT cell expansion provides another example of the strong mutualism between the commensal flora and the development of the mammalian immune system [38].
In conclusion, we show here that besides their similarities, MAIT cell ontogeny is clearly different from NKT cells. Postselection NKT thymocytes already display innate-like properties, with a high frequency and a memory phenotype, whereas postselection MAIT cells are still naïve and need contacts with both B cells and bacteria to expand, acquire a memory phenotype in human, and accumulate in the gut. In this respect, they may be compared to Vδ2+ T cells, which appear in blood as naïve but become memory soon after birth [39,40]. The striking high frequency of MAIT cells in the blood of healthy subjects suggest that they play a prominent role in various diseases, either of infectious, tumoral, or auto-immune origin.
CD3ε-deficient B6 mice, Cα deficient mice (N9 to C57Bl/6 [B6]) and Ii−/− (B6/129) mice were obtained from the CNRS CDTA central animal facility (Orléans, France). TAP−/− mice in a B6/129 background were obtained from the Jackson laboratory. Cα−/−, Ii−/−, and TAP−/− mice were intercrossed to obtain double and triple deficient mice. MR1−/− mice have been described elsewhere [14]. MR1−/− mice, kindly provided by S. Gilfillan, were backcrossed for more than ten generations onto B6 mice. They were also crossed to TAP−/−/Ii−/− mice. All the TAP−/−/Ii−/− mice had a mixed B6/129 background. iVα19-Jα33 and Vβ6 transgenic mice were obtained as previously described [41] by cloning PCR products obtained from a Vα19/Vβ6 T-T hybridoma cDNA into the TCR alpha or beta shuttle vector [26]. The TCRα and TCRβ constructs were injected into B6/DBA2 F1 and B6 eggs, respectively, at the CNRS transgenic facility (C. Goujet, Villejuif, France). Three iVα19 TCRα founder lines were backcrossed for at least ten generations onto Cα−/− B6 mice before further crossing. Most of the results were obtained with two founder lines expressing six and eight copies of the transgene. At this stage, these two founder lines were also crossed with the Vβ6 TCRβ Tg lines. They were also crossed with the RAG2−/− CD45.1 B6 line and then with the Vβ6 TCRβ Tg B6 lines to obtain iVα19/Vβ6 RAG−/− CD45.1 B6 lines. The CD45.1 allotype marker was also introduced into one of the iVα19 Cα−/− B6 lines. The two TCRα Tg lines were also crossed with Cα−/−/TAP−/−/Ii−/−. The MR1-deficient allele in a B6 background was introduced into all lines. All mice were housed in our SPF colony and genotyped by PCR or FACS staining, as appropriate. Live animal experiments were done in accordance with the guidelines of the French Veterinary Department.
The canonical TCRα chain and the associated Vβ13 TCRβ chains were cloned (ET and OL), from an iVα7.2-Jα33 T cell clone (J.F. Davodeau, ET, M. Bonneville, and OL, unpublished data). Using this template, L. Teyton (SCRI) generated a recombinant heterodimeric protein, which was used to immunize Balb/c mice. After fusion with SP2/0, hybridoma specifically staining a high proportion of CD3+ DN and a proportion of CD8α but few CD4 T cells were selected and cloned. The 3C10 hybridoma was chosen for further characterization because, according to quantitative PCR, the Vα7.2-Jα33 TCRα chain segregated with the 3C10 positive fraction in DN T cells. When transfectants expressing different mouse TCRα and TCRβ chains and a chimeric human iVα7.2-mouse Cα were stained with the 3C10 supernatant, only transfectants harboring the Vα7.2 segment displayed positive staining demonstrating that 3C10 was not an anti-TCRβ chain antibody (unpublished data). Quantitative PCR analysis of the 3C10+ and 3C10− fractions using primers for all Vα segments, demonstrated the absence of significant cross-reactivity between the 3C10 mAb and other Vα segments (unpublished data). The 3C10 antibody was biotinylated and detected with streptavidin PE-Cy7 (BD Pharmingen).
Cell suspensions were prepared from thymus, spleen, peripheral or mesenteric lymph nodes, and gut LPL as previously described [14].
Flow cytometry was performed with directly conjugated antibodies (BD Pharmingen) according to standard techniques with analysis on FACS Aria and LSRII flow cytometers (Becton Dickinson). DAPI and a 405-nm excitation were used to exclude dead cells. The following antibodies, mostly from BD Pharmingen or eBiosciences, were used in mice: anti-CD45.2-Fitc (104), anti-CD3ε-PC7 (145–2C11), anti-CD5-APC (Ly-1), anti-Vβ6-PE (RR4–7), anti-CD44-PE or APC (IM-7), anti-CD45.1-PE (A20), anti-βTCR-PC5 (H57–597), anti-CD19-PE, Fitc or APC-Cy7 (1D3), anti-CD8α-APC-Cy7 (Ly-2), and anti-CD4-PE-Texas Red (L3T4).
For human stainings, the following were used: anti-CD4-APC-Cy7 (RP4-T4), anti-CD3ε-Alexa 700 (UCHT1), anti-TCRγδ-PC5 (IMMU510), anti-CD8β-PE Texas Red (2ST8.5H7), anti-CD45RO-Fitc (UCHL1), anti-CD45RA-PE (HI100), anti-CD27-Fitc (M-T271), anti-CCR7-PE (150503), anti-CD62L-PE (Dreg 56), anti-CXCR6-PE (56811), anti-CD161-APC (DX12), and anti-CD8α-PE Alexa700 (RPA-T8).
Fragments of human thymuses were operating residues from children undergoing cardiac surgery. Thymuses were cut into small pieces in cold 0.5% BSA PBS. The resulting cell suspension was centrifuged to exclude aggregates. The supernatant was recovered and the cells were washed again with 0.5% BSA PBS. Thymocytes were then counted and labeled.
Blood samples were obtained from healthy donors from the blood bank in accordance with institutional regulations.
Spleens from CD3ε−/− MR1+ mice were harvested and 107 splenocytes were IV injected into MR1+ or MR1− TCRαβ double Tg Rag−/− mice. 2 wk later, the blood, spleen, and MLN were harvested and analyzed by FACS.
iVα19 Tg Cα−/− MR1−/− or MR1+/+ female mice were caged one night with the respective genotype male mice. Pregnant female mice were sacrificed at day 14. Uterine corns were harvested in CO2-independent medium supplemented with 5% FCS and penicillin/streptomycin (Invitrogen). Fetal thymuses were harvested and cultured in 300 μl of IMDM medium supplemented with 10% FCS, 50 μM beta-mercaptoethanol, 10 mM Hepes, 1 mM Sodium pyruvate, 2 mM L-glutamine, and penicillin/streptomycin in transwell plates from Costar (0.4 μm, 12 wells). The medium was changed every 3 d over a period of 6–8 d.
Molecular biology methods: RNA extraction, reverse transcription, TCR primers, PCR methods, and quantitative polyclonal PCR and polyclonal sequencing were carried out as previously described [14,22].
|
10.1371/journal.pgen.1004819 | A Massively Parallel Pipeline to Clone DNA Variants and Examine Molecular Phenotypes of Human Disease Mutations | Understanding the functional relevance of DNA variants is essential for all exome and genome sequencing projects. However, current mutagenesis cloning protocols require Sanger sequencing, and thus are prohibitively costly and labor-intensive. We describe a massively-parallel site-directed mutagenesis approach, “Clone-seq”, leveraging next-generation sequencing to rapidly and cost-effectively generate a large number of mutant alleles. Using Clone-seq, we further develop a comparative interactome-scanning pipeline integrating high-throughput GFP, yeast two-hybrid (Y2H), and mass spectrometry assays to systematically evaluate the functional impact of mutations on protein stability and interactions. We use this pipeline to show that disease mutations on protein-protein interaction interfaces are significantly more likely than those away from interfaces to disrupt corresponding interactions. We also find that mutation pairs with similar molecular phenotypes in terms of both protein stability and interactions are significantly more likely to cause the same disease than those with different molecular phenotypes, validating the in vivo biological relevance of our high-throughput GFP and Y2H assays, and indicating that both assays can be used to determine candidate disease mutations in the future. The general scheme of our experimental pipeline can be readily expanded to other types of interactome-mapping methods to comprehensively evaluate the functional relevance of all DNA variants, including those in non-coding regions.
| With rapid advances in sequencing technologies, tens of millions of DNA variants have now been discovered in the human population. However, there are currently no experimental methods available for examining the impact of DNA variants in a high-throughput fashion. As a result, we have no functional data on the vast majority of these variants, which is a major roadblock to generating novel biological insights and developing new disease prevention therapeutic strategies. To address this issue, we have successfully developed the first massively-parallel site-directed mutagenesis approach, Clone-seq, to leverage the power of next-generation sequencing to generate a large number of mutant alleles in a fast and cost-effective manner. In conjunction with Clone-seq, we established a high-throughput comparative interactome-scanning pipeline to experimentally elucidate the effect of variants on protein stability and interactions. Additionally, Clone-seq can be used to generate clones for all DNA variants, including those in non-coding regions.
| Owing to rapid advances in next-generation sequencing technologies, tens of thousands of disease-associated mutations [1] and millions of single nucleotide polymorphisms (SNPs) [2], [3] have been identified in the human population. With the large number of ongoing whole-exome and whole-genome sequencing projects [2], [3], hundreds of thousands of new SNPs are now being discovered every month. Hence, there is an urgent need to develop high-throughput methods to sift through this deluge of sequence data and rapidly determine the functional relevance of each variant. Here, we focus on coding variants, firstly because trait- and disease-associated SNPs are significantly over-represented in nonsynonymous sites [4], and secondly because the vast majority of disease-associated mutations identified to date reside within coding regions [1]. We evaluate the functional impact of coding variants by examining their effects on corresponding protein-protein interactions, because most proteins carry out their functions by interacting with other proteins [5].
Recent studies have begun to use large-scale protein interaction networks to understand human diseases and their associated mutations [5], [6]. By integrating structural details with high-quality protein networks, we created a 3D interactome network where the interface for each interaction has been structurally resolved [7]. Using this 3D network, we demonstrated that in-frame disease mutations (missense mutations and in-frame insertions/deletions) are significantly enriched at the interaction interfaces of the corresponding proteins [7]. Our results indicate that alteration of specific interactions is very important for the pathogenesis of many disease genes, highlighting the importance of 3D structural models of protein interactions in understanding the functional relevance of coding variants. However, many important questions still remain unanswered – for example, what fraction of protein-protein interactions is altered by disease mutations to cause the corresponding disorders? Furthermore, do structural details of the interacting proteins, especially the position of the mutation relative to the interaction interface, affect the ability of a given disease mutation to alter a specific interaction?
To address these questions, we decided to focus on proteins with known disease mutations that participate in interactions with available co-crystal structures in the Protein Data Bank (PDB) [8]. To detect the alteration of the interactions by disease mutations, it is necessary to first detect the interactions of the wild-type proteins using an assay of choice. This turns out to be a major bottleneck because all high-throughput interaction-detection assays have very limited sensitivity [9], [10]. Our assay of choice is Y2H because there are over 16,000 human protein interactions detected by our version of Y2H that can serve as the reference interactome for comparison [11], [12], [13], [14], the largest for any assay performed to date (Figure S1). In total, there are 217 interactions detected by our version of Y2H with available co-crystal structures; 51 of these also have known missense disease mutations on corresponding proteins in the Human Gene Mutation Database (HGMD) [1] and the corresponding interactions for the wild-type proteins can be detected in our experiments with strong Y2H-positive phenotypes (Figure S2; Materials and Methods). Here, we focused on missense mutations because they are intrinsically more likely to generate interaction-specific disruptions [6]. We established a high-throughput comparative interactome-scanning pipeline to clone disease mutations and examine their molecular phenotypes (Fig. 1). The methodologies established here can be readily applied to any non-synonymous variant in the coding region, including nonsense mutations.
The first step of our pipeline is a massively parallel approach, termed Clone-seq, designed to leverage the power of next-generation sequencing to generate a large number of mutant alleles using site-directed mutagenesis in a rapid and cost-effective manner. Current protocols for site-directed mutagenesis require picking individual colonies and sequencing each colony using Sanger sequencing to identify the correct clone [15]. This standard approach is both labor-intensive and expensive; therefore, it does not scale up to genome-wide surveys. In Clone-seq, we put one colony of each mutagenesis attempt into one pool (Fig. 1a; in other words, each pool contains one and only one colony for each desired mutation) and combine multiple pools through multiplexing for one Illumina sequencing run [16]. Colonies for generating different mutations of the same gene can be put into the same pool, which can be easily distinguished computationally when processing the sequencing results. This is true even for mutations occurring at the same site (Fig. 2a, Text S1).
For the 51 selected interactions, we chose 27 disease-associated mutations of residues at the interface (“interface residue”), 100 mutations in the rest of the interface domain (“interface domain”) and 77 mutations away from the interface (“away from the interface”; Fig. 3a,b). These interfaces were determined using solvent accessible surface area calculations as previously described [17], [18] on 7,340 co-crystal structures (Materials and Methods). To set up our Clone-seq pipeline, we first started with 39 mutations from these 204 and picked 4 colonies for each mutation. As a reference, we also pooled together all the wild-type alleles in our human ORFeome library to be sequenced together with the 4 pools of the mutagenesis colonies. In total, there were 40.1 million Illumina HiSeq 1×100 bp reads for our Clone-seq samples (Text S1) for an average of >2,500× coverage on all desired mutation sites. Therefore, our Clone-seq pipeline has the capacity to generate >3,000 mutations in one full lane of a HiSeq run with 1×100 bp reads, drastically improving the throughput and decreasing overall sequencing costs by at least 10-fold (Text S1).
Fig. 2a presents a schematic of the criteria we use to determine which clones contain the desired mutation and can be used for subsequent steps. For example, in pool 1, all reads (ignoring sequencing errors) confirm that genes I and II each contain the desired mutation – T116A and G298T, respectively. For gene III, we want to generate two separate clones with two separate mutations – IIIA41T and IIIC194T. Since half the reads contain T41 (instead of A41) and the other half contain T194 (instead of C194), and we normalize DNA concentrations across all samples, we can infer that both mutant clones were generated successfully. In contrast, for gene IV, we see that while half the reads contain A511 (instead of G511), all the reads are wild-type at C74. Thus, we infer that while the IVG511A clone is successfully generated, the IVC74T clone is not. For gene V, although both mutant clones are successfully generated, half the reads contain an additional mutation, C436G. Since it is impossible to know which of the two clones for V contains this unwanted mutation, neither clone is usable. Similarly, we can determine mutant clones IT116A, IIIA41T, IIIC194T, IVC74T, IVG511A, VT53G, and VG272A as usable clones in pool n. Based on these criteria, we developed the S score calculation and used it to determine successful mutagenesis attempts (Materials and Methods). Out of 156 colonies for 39 mutations, 125 of them contain the desired mutations (S>0.8), an overall 80% PCR-mutagenesis success rate. In fact, we were able to pick correct clones for all 39 mutant alleles using only the first two pools in Clone-seq. All 78 clones from the first two pools, from which the correct ones were selected for use in subsequent steps, were also Sanger sequenced for verification. 55 Clone-seq positive results with S>0.8 were all confirmed and there is a clear separation in the S scores between the successful and failed mutagenesis attempts (Fig. 2b).
One major advantage of our Clone-seq pipeline is that it allows us to carefully examine whether other unwanted mutations have been inadvertently introduced during PCR-mutagenesis in comparison with the corresponding wild-type alleles, since we obtain reads spanning the entire gene. We found that there are on average 4–5 unwanted mutations introduced in each pool of 39 colonies. This corresponds to a 0.013% PCR error rate (Materials and Methods), in agreement with previous studies [19]. The detection of unwanted mutations, especially those distant from the mutation of interest, is achieved in traditional site-directed mutagenesis pipelines by Sanger sequencing through the gene of interest. This is costly and labor-intensive, especially because multiple sequencing runs are needed for one long gene. However, since Clone-seq yields reads spanning the entire gene, we were able to determine which of the generated clones definitely do not have unwanted mutations in the full length of their sequences as illustrated in Fig. 2a (Materials and Methods), and we pick only these clones for subsequent assays.
To further test our Clone-seq pipeline, we applied it to generate clones for 113 SNPs on 66 genes from the recently published Exome Sequencing Project dataset [3]. Using the same approach as described above, we sequenced 4 colonies each for the 113 alleles of interest using one third of a 1×100 bp MiSeq run. We obtained 4.7 million reads for these 113 alleles. With a threshold of S>0.8, we were able to determine that 370 out of the 452 colonies (82%) contain the desired mutation, in perfect agreement with the PCR-mutagenesis success rate obtained earlier. We were able to choose colonies that contain only the desired mutation for all 113 alleles. Because the whole MiSeq run produced 17.7 million reads and we only used 4.7 million for generating the 113 mutant clones, the capacity of our Clone-seq pipeline using one full lane of a 1×100 bp HiSeq run is estimated to be >3,000, exactly the same as our previous assessment (Text S1).
Finally, we generated the remaining 165 disease mutations (of the 204) and 717 other coding variants from the Exome Sequencing Project and the Catalog of Somatic Mutations in Cancer [20] using a full 1×100 bp HiSeq run, including 40 mutations on a single gene – MLH1. Using 111.2 million reads for these 882 alleles, we found that 2,958 of the 3,528 colonies (84%) contain the desired mutation, again in excellent agreement with our previously obtained PCR-mutagenesis success rate. There was at least one colony with only the desired mutation for all 882 alleles, including all 40 MLH1 mutations (Table S1). Therefore, our Clone-seq pipeline can generate a large number of mutations (>40) even for a single gene. In fact, to generate even more mutations for one gene, we can implement a two-round barcoding approach: generate groups of 40 mutations and barcode them differently for one HiSeq run (Figure S3). Ten such groups will enable us to generate ∼400 mutations for a single gene (Text S1). Since the average coverage of these 882 alleles is >300×, the capacity of our Clone-seq pipeline using one full lane of a 1×100 bp HiSeq run is estimated to be >3,000, again in agreement with our previous two estimates (Text S1).
Overall, our pipeline has been significantly optimized to make it very efficient. We established a web tool (http://www.yulab.org/Supp/MutPrimer) to design mutagenesis primers both individually and in batch. MutPrimer can design ∼1,000 primers for ∼500 mutations in one batch in less than one second. All of the 2,068 primers for the 1,034 mutations in this study were generated by MutPrimer. All mutagenesis PCRs are performed in batch using automatic 96-well procedures. Since single colony picking after bacterial transformation of mutagenesis PCR product is a rate-limiting step, we rigorously optimized this step and found that adding 10 µL mutagenesis PCR products to 100 µL competent cells and plating 50 µL transformed cells give the best transformation yield and well-separated single colonies. Furthermore, rather than individually streaking transformed cells onto agar plates one sample at a time, we were able to significantly increase throughput by spreading colonies using glass beads onto four sector agar plates which are partitioned into four non-contacting quadrants (Materials and Methods). In this manner, a 96-well plate of transformed bacteria can be plated out onto 24 four-sector agar plates in ∼15 minutes. Traditional site-directed mutagenesis pipelines require miniprepping each of the selected colonies and sequencing them separately by Sanger sequencing. To drastically improve the throughput of our Clone-seq pipeline, we pooled together the bacteria stock of a single colony for each mutagenesis attempt to perform one single maxiprep, which makes the library construction step much more efficient and amenable to high-throughput (Text S1). Furthermore, existing variant calling pipelines [21] cannot be applied to our Clone-seq results because the expected allelic ratios built into these pipelines are a function of the ploidy of the organism. However, in our Clone-seq pipeline there is no concept of ploidy. We pool together many mutations for one gene in the same pool (e.g., 40 mutations for MLH1) and different genes often have different numbers of mutations. Our S score calculation and unwanted mutation detection pipeline was designed according to our pooling strategy (Materials and Methods).
In total, we have used the novel Clone-seq pipeline successfully to generate 1,034 (39+113+882) mutant clones without any additional unwanted mutations, confirming the scalability, accuracy, and throughput of our Clone-seq pipeline.
For the 204 mutations on proteins with co-crystal structures, we first examined whether the mutant proteins can be stably expressed in human cells. To do this, we tagged every wild-type and mutant protein with GFP at the C-terminus using high-throughput Gateway cloning (Fig. 1b). The GFP constructs were transfected into HEK293T cells and fluorescence intensities were measured by a plate reader (Fig. 3c; Materials and Methods). All fluorescence intensity readings were also confirmed manually under a microscope. Compared with the corresponding wild-type proteins, the expression levels of 3 of the 27 “interface residue” mutants, 8 of the 99 “interface domain” mutants and 6 of the 77 “away from the interface” mutants are significantly diminished (Fig. 3c; Materials and Methods; S2 Table). To validate these findings, we also performed Western blotting for 8 random mutants that are stably expressed and 8 random mutants with significantly diminished expression levels (Fig. 4a). Western blotting results confirm our GFP intensity readings.
Next, we investigated whether these mutations could affect protein-protein interactions using Y2H (Fig. 1c; Materials and Methods). We found that 21 of the 27 (78%) “interface residue” mutations, 57 of the 100 (57%) “interface domain” mutations, and only 22 of the 77 (29%) “away from the interface” mutations disrupt the corresponding interactions, thereby demonstrating a clear difference (Fig. 4b; P = 3×10−6 between “interface residue” and “interface domain” and P = 8×10−10 between “interface domain” and “away from the interface”) in terms of ability to interfere with protein-protein interactions between mutations at different structural loci within the same protein. Furthermore, comparing with the GFP results, we found that all destabilizing mutations were shown to disrupt the corresponding interactions in our Y2H experiments. By considering only the mutations that do not affect protein expression based on the GFP experiments, we found the same difference: 13 out of 18 (72%) “interface residue” stable mutations, 42 out of 83 (51%) “interface domain” stable mutations, and only 9 out of 52 (17%) “away from the interface” stable mutations disrupt the corresponding interactions (Fig. 4b; P = 2×10−5 between “interface residue” and “interface domain” and P = 9×10−13 between “interface domain” and “away from the interface”; Table S2). Since these interfaces are obtained from actual co-crystal structures, our results suggest that accurate structural information can help determine the functional impact of mutations on protein-protein interactions. Wild-type proteins corresponding to 113 of the 153 stably expressed mutant proteins also interact with other proteins as determined by our Y2H experiments (114 interactions in total, termed “other interactions”); however, for these interactions, there are currently no co-crystal structures available in the PDB. Using these other interactions, we calculated the likelihood of a given mutation disrupting a specific interaction without any structural information to be 32% (Fig. 4b).
We then analyzed whether the molecular phenotypes measured by our high-throughput GFP and Y2H assays are correlated with corresponding disease phenotypes. We first examined how mutation pairs on the same gene affect protein stability and its relationship to their corresponding diseases. We find that pairs of mutations that are either both stable or both unstable cause the same disease in 68% and 70% of cases, respectively. However, pairs comprising one stable and one unstable mutation cause the same disease in only 30% of cases (P = 6×10−9 and 8×10−10, respectively, Fig. 5a). For example, we find that the mutations R727C and L844F on the spindle checkpoint kinase Bub1b both cause the protein to become unstable and lose all its interactors. These mutations are both associated with the same disease, mosaic variegated aneuploidy, an autosomal recessive disorder that causes predominantly trisomies and monosomies of different chromosomes [22], [23]. Since our GFP assay shows that these two mutations cause loss of protein product, our results are consistent with Matusuura et al.'s finding that a more than 50% decrease in Bub1b activity leads to abnormal mitotic spindle checkpoint function and mosaic variegated aneuploidy [24].
We then examined whether mutation pairs on the same gene disrupt the same set or different sets of interactions (i.e., their interaction disruption profiles) and investigated whether their disruption profiles correlates with disease phenotypes. We found that mutation pairs with the exact same disruption profile are significantly more likely to cause the same disease than those with different profiles (70% and 61% respectively, P = 3×10−5, Fig. 5b). For example, we found that two mutations on Smad4, R361C and Y353S, disrupt its interactions with Smad3 and Smad9 while leaving the interactions with Lmo4 and Rassf5 unaltered (Fig. 5c). These two mutations both cause juvenile polyposis coli [25], [26], a disease is known to be caused by disruption of the core Smad/Bmp signaling pathways [27]. Our Y2H results clearly demonstrate that the R361C and Y353S mutations disrupt the Smad4-Smad3 and Smad4-Smad9 interactions (Fig. 5c) leading to disruption of core Smad signaling pathways. However, the mutation N13S on Smad4 does not disrupt any of these interactions (Fig. 5c) and is associated with a different disease, pulmonary arterial hypertension. Our results agree with Nasim et al.'s finding that the N13S mutation does not alter downstream Smad signaling [28]. Our findings provide support for the hypothesis that the N13S mutation either impacts pathways outside the core Smad signaling network or are pathogenic only when combined with other environmental and genetic factors [29].
Overall, these results show that mutation pairs with similar molecular phenotypes in terms of both protein stability and interactions are significantly more likely to cause the same disease than those with different molecular phenotypes. This confirms that the molecular phenotypes measured by our high-throughput GFP and Y2H assays are biologically relevant in vivo. Furthermore, by comparing the molecular phenotypes, in particular the protein interaction disruption profiles, of mutations/variants to those of known disease mutations, potential candidate mutations for a variety of diseases can be identified.
While we use only those interactions that are supported by co-crystal structures to estimate the fraction of interactions that are disrupted by mutations at different structural loci, the described procedures can also be applied to interactions with predicted interfaces and structural models [30], [31], [32], [33]. This is of particular importance because over 90% of known interactions do not currently have corresponding co-crystal structures [33], [34]. For example, Mlh1 is known to interact with Pms2, both of which are well-studied DNA mismatch repair genes frequently mutated in hereditary nonpolyposis colorectal cancer [35]. Although the structural basis of the Mlh1-Pms2 interaction still remains unknown, both our previous 3D reconstruction of the human interactome network [7], [32] and the newly-established Interactome3D [33] database suggest that the HATPase_c domain is part of the interface for Mlh1's interaction with Pms2. Previous work has shown that a point mutation (I107R) on the HATPase_c domain of Mlh1 is associated with colorectal cancer and disrupts the Mlh1-Pms2 interaction [7], [35], [36]. First, using Y2H, we were able to confirm the disruption (Figure S4). Next, we developed a high-throughput-amenable mass spectrometry pipeline using Stable Isotope Labeling by Amino acids in Cell culture (SILAC) [37], [38], which was designed to reveal both lost/weakened and gained/enhanced interactions of the target proteins (Fig. 1d) [39]. We added an HA-tag to the N-terminus of both wild-type and mutant Mlh1, as well as to GFP as a control, and performed four SILAC experiments: wild-type Mlh1 (heavy) vs. GFP control (light), mutant Mlh1 (heavy) vs. GFP control (light), wild-type (heavy) vs. mutant (light) Mlh1, and mutant (heavy) vs. wild-type (light) Mlh1 (Fig. 6a; Materials and Methods). Interactors of wild-type/mutant Mlh1 are defined as those that bind wild-type/mutant Mlh1 more than 2× stronger than GFP control (Materials and Methods). For a lost/weakened interaction, we required that the interaction be more than 2× stronger with wild-type Mlh1 than with mutant Mlh1 as confirmed both in wild-type (heavy) vs. mutant (light) and in mutant (heavy) vs. wild-type (light) experiments; we further required that the interaction be detected in the wild-type vs. control experiment (Fig. 6a; Materials and Methods). For a gained/enhanced interaction, we required that the interaction be more than 2× stronger with mutant Mlh1 than with wild-type Mlh1 as confirmed both in wild-type (heavy) vs. mutant (light) and in mutant (heavy) vs. wild-type (light) experiments; we further required that the interaction be detected in the mutant vs. control experiment (Fig. 6a; Materials and Methods). We were able to detect Pms2 as the only specifically weakened interactor caused by the mutation (Figs. 6b,c; E = −1.77; P = 3×10−4), in agreement with our Y2H results and previous studies [7], [36]. Additionally, we were able to detect Hspa8 as the only specifically enhanced interactor of the mutant protein (Figs. 6b,c; E = 2.71; P = 7×10−8). Two other known interactors of Mlh1, Pms1 (Figs. 6b,c; E = −0.32; P = 0.21) [40] and Brip1 (Fig. 6b,c; E = 0.18; P = 0.32) [41], were also detected, although their interactions with Mlh1 are not affected by this particular mutation (Materials and Methods).
Hspa8 was not previously known to interact with Mlh1 and the impact of the Mlh1 I107R mutation on its interactions with Pms1 and Brip1 has not been reported in the literature. To verify our SILAC results, we performed in vivo co-immunoprecipitation using HA-tagged wild-type and mutant Mlh1 and tagged Hspa8 and Brip1 with V5 (Materials and Methods). Our co-immunoprecipitation results confirm that Hspa8 only weakly interacts with wild-type Mlh1, but the interaction is dramatically enhanced by a single amino acid substitution (I107R) (Fig. 6d, lanes 3 and 4), whereas the interaction between Mlh1 and Brip1 is not affected by this mutation (Fig. 6d, lanes 6 and 7; Materials and Methods). Hspa8 is a constitutively expressed member of the heat shock protein 70 family [42]. It functions as a chaperone to facilitate protein folding [42] and also functions as an ATPase in the disassembly of clathrin-coated vesicles during membrane trafficking [43]. A recent study reported that Hspa8 is specifically recruited to reovirus viral factories, independent of its chaperone function [44]. Our Western blotting results demonstrate that the expression level of Mlh1 is not affected by the I107R mutation (Figure S5). Therefore, our SILAC results suggest that Hspa8 may play an important role in colorectal cancer and that its function could be independent of its role as a chaperone.
We have successfully developed the first massively parallel site-directed mutagenesis pipeline, Clone-seq, using next-generation sequencing. Our Clone-seq pipeline is entirely different from previously described random mutagenesis approaches [45], [46], [47], [48]. Clone-seq is used to generate a large number of specific mutant clones with desired mutations; each individual mutant clone has a separate stock and different clones can therefore be used separately for completely different downstream assays. In random mutagenesis, a pool of sequences containing different mutations for one gene is generated using error-prone PCR or error-prone DNA synthesis. Therefore, it is not possible to separate one mutant sequence from another and the whole pool can only be used for the same assay(s) together. Furthermore, it is not possible to control which or how many mutations are generated on each DNA sequence. In fact, to improve coverage, most random mutagenesis pipelines generate on average two or more mutations on each DNA sequence [45], which makes it impossible to distinguish the functional impact of each individual mutation on the same sequence. Site-directed mutagenesis and random mutagenesis are designed for different goals: if one wants to generate all possible mutations for a certain protein without the need to separate different clones, it would be more favorable to use random mutagenesis; whereas if one needs to have separate clones for each mutation, site-directed mutagenesis is required. As a result, the two approaches are complementary and not comparable.
While there are highly efficient methods for random mutagenesis [45], [46], [47], [48], current protocols for site-directed mutagenesis are low-throughput and become prohibitively expensive if a large number of clones needs to be generated. Clone-seq directly addresses the necessity for a high-throughput site-directed mutagenesis pipeline. It is a robust, cost-effective and efficient method that can be used to generate a total of ∼3,000 distinct mutant clones in one full lane of a 1×100 bp HiSeq run. Clone-seq is suitable both for generating mutations across many genes as well as a large number of mutations on a few genes. The former situation is applicable when one wants to generate many mutations/variants from large-scale studies (e.g., whole-genome or whole-exome sequencing) since they typically identify mutations/variants on a large number of genes [49], [50]. The latter situation usually arises in a study focused on a single pathway with a few genes of interest (e.g., an alanine-scanning mutagenesis to determine functional sites on a gene of interest [51]).
Integrating with Clone-seq, we also established a comprehensive comparative interactome-scanning pipeline, including high-throughput GFP, Y2H, and mass spectrometry assays, to systematically evaluate the impact of human disease mutations on protein stability and interactions. We examine each mutation individually, rather than looking at their combinatorial effects because these inherited germline disease mutations are extremely rare. Therefore, the probability of having even two of these in the same individual becomes infinitesimally small. Our results reveal that the overall likelihood of a given disease mutation disrupting a specific interaction is 32%. Accurate structural information of these interactions obtained from co-crystal structures greatly improves our understanding of the impact of disease mutations: 13 out of 18 (72%) “interface residue” stable mutations, 42 out of 83 (51%) “interface domain” stable mutations, and only 9 out of 52 (17%) “away from the interface” stable mutations disrupt the corresponding interactions, unveiling a clear dependence of the molecular phenotypes of disease mutations on their structural loci. These estimates are not affected by the false negative rate of our Y2H assay as we only use those interactions for which we can detect the wild-type interaction with strong Y2H phenotypes. Thus, any observed disruption is due to the mutation of interest and not an assay false negative. Furthermore, our Y2H pipeline has been shown to be of high quality and has an experimentally measured false positive rate of ∼5% or lower in different organisms [9], [12], [52], [53]. In addition, the interactions used to understand the relationship between molecular phenotypes and structural loci of disease mutations are all supported by co-crystal structures, therefore these interactions are not assay false positives. We also find that the molecular phenotypes detected by our GFP and Y2H assays correlate with known disease phenotypes, confirming the in vivo biological significance of our measurements.
Moreover, as shown by the Mlh1 example (Fig. 6), our comparative interactome-scanning pipeline can also be used with predicted structural models [30], [31], [32], [33]. The consequent experimental results will clearly be affected by the quality of these predictions, which is not part of our pipeline. In fact, our experimental interactome-scanning pipeline can be applied to evaluate or improve these predicted models by testing mutations at different loci of a protein of interest and examining how these mutations disrupt different interactions of this protein.
Our comparative interactome-scanning pipeline described and validated here can be applied to experimentally determine in a high-throughput fashion the impact on protein stability and protein-protein interactions for thousands of DNA coding variants and disease mutations, which can directly lead to hypotheses of concrete molecular mechanisms for follow-up studies. Furthermore, the elucidation of molecular phenotypes of disease mutations is also vital for selecting actionable drug targets and ultimately for making therapeutic decisions. Finally, the general scheme of our pipeline can be readily expanded to other interactome-mapping methods, particularly other protein-protein [10], protein-DNA [54], [55], protein-RNA [56], and protein-metabolite interaction assays [57], to comprehensively evaluate the functional relevance of all DNA variants, including those in non-coding regions.
To calculate atomic-resolution interaction interfaces, we systematically examined a comprehensive list of 7,340 PDB co-crystal structures. To define the interface, we used a water molecule of diameter 1.4 Å as a probe and calculated the relative solvent accessible surface areas of the interacting pair as well as the individual proteins involved in the interaction. Residues whose relative accessibilities change by more than 1 Å2 are considered as potential interface residues, because amino acids at the interface reside on the surfaces of the corresponding proteins, but will tend to become buried in the co-crystal structure as the two proteins bind to each other [58]. So, for these residues, there should be a significant decrease in accessible surface area when we compare the bound and unbound states of the protein chains.
To identify interface domains, we required at least one of the following criteria to hold:
We then identified the subset of these interactions that contain at least one disease mutation and are amenable to our version of Y2H [11], [12], [13], [14]. Subsequently, we performed a pairwise retest of all these interactions and selected the ones that yield strong Y2H phenotypes, because subsequent steps involve detecting a significant decrease in these phenotypes.
Primers for site-directed mutagenesis were selected based on a customized version of the protocol accompanying the Stratagene QuikChange Site-Directed Mutagenesis Kit (200518). The following criteria are used:
For cases where no primer satisfies all three criteria simultaneously, we relaxed criterion 2 to GC content ≥30%.
We established a supplementary web tool (http://www.yulab.org/Supp/MutPrimer) to design mutagenesis primers individually or in bulk.
All wild-type clones were obtained from the human ORFeome v8.1 collection [61]. To generate mutant alleles, sequence-verified single-colony wild-type clones and their corresponding mutagenic primers were aliquoted into individual wells of 96-well PCR plates. Mutagenesis PCR was then performed as specified by the New England Biolabs (NEB) PCR protocol for Phusion polymerase (M0530L), noting that PCR was limited to 18 cycles. The samples were then digested by DpnI (NEB R0176L) according to the manufacturer's manual. After digestion, samples were transformed into competent E. coli. Since single colony picking after bacterial transformation of mutagenesis PCR product is a rate-limiting step, we rigorously optimized this step. First, we tried different volumes of competent cells for transformation and found that single colony yields peak when ∼100 µL of competent cells are used. It is also necessary to use ∼10 µL of mutagenesis PCR product: any lower volume of PCR product results in significantly reduced colony yields, while higher volumes of PCR product do not increase yield. Finally, colony picking was done using four-sector agar plates (VWR 25384-308) that are partitioned into four non-contacting quadrants with glass beads poured onto each plate quadrant. Each bead-filled quadrant was inoculated with ∼50 µL of transformed bacteria. This was then spread by lightly shaking the four-sector agar plate. Our optimized transformation protocol results in a large number of well-separated single colonies that can be easily picked the next day. Upon recovery, single colonies from each quadrant were then picked and arrayed into 96-deepwell plates filled with 300 µL of antibiotic media. Four colonies per allele were picked for next-generation sequencing.
DNA library preparation was performed using NEBNext DNA Library Prep Master Mix Set for Illumina (NEB E6040S) according to the manufacturer's manual. Briefly, 5 µg of pooled plasmid DNA (∼100 µL, all samples were normalized to the same concentration) was sonicated to ∼200 bp fragments. The fragmented DNA was first mixed with NEBNext End Repair Enzyme for 30 mins at 20°C. Blunt-ended DNA was then incubated with Klenow Fragment for 30 mins at 37°C for dA-Tailing. Subsequently, NEBNext Adaptor was added to dA-Tailed DNA. Adaptor-ligated DNA (∼300 bp) was size-selected on a 2% agarose gel. Size-selected DNA was then mixed with one of the NEBNext Multiplex Oligos (NEB E7335S) and Universal PCR primers for PCR enrichment. At each step, DNA was purified using a QIAquick PCR purification kit (Qiagen 28104). Multiplexed DNA samples were combined and analyzed in one lane of a 1×100 bp run by Illumina HiSeq 2500.
The mutant colonies were barcoded and pooled as shown in Fig. 1a. The multiplexed colonies were then run on an Illumina sequencer (2 HiSeq runs and 1 MiSeq run) to give 1×100 bp reads. These reads were then de-multiplexed and mapped to the genes of interest using the BWA “aln” algorithm [62]. For each allele, we identified all reads that mapped to the position of the mutation of interest (Rall) and those that actually contained the desired mutation (Rmut). We then calculated a normalized score (S) that quantifies the fraction of reads containing the desired mutation:where k is the number of different mutations for the same gene.
For 39 mutations, we Sanger sequenced two mutant colonies per mutagenesis attempt to quantify the correlation between S and observation of the desired mutation. We found that all clones with S>0.44 are confirmed to be correct via Sanger sequencing with a clear separation between those that are correct and those that are not (Fig. 2b). However, to further ensure that the clones we picked were correct, we require S>0.8 for a colony to be scored as containing the desired mutation.
One major advantage of our Clone-seq pipeline over traditional site-directed mutagenesis protocols using Sanger sequencing [15] is that we can now carefully examine whether there are other unwanted mutations inadvertently introduced during the PCR process, in comparison with the corresponding wild-type alleles. It is essential to use clones with no unwanted mutations for downstream experiments, as the presence of these will make it impossible to determine whether the observed disruption is due to the desired or other undesirable mutation(s).
We use samtools “mpileup” [63] to obtain read counts for different alleles at each nucleotide for all the clones. We calculate the background sequencing error rate by calculating the average fraction of non-reference alleles across all nucleotides where we did not attempt to introduce a mutation. Any site that has a significantly higher fraction of non-reference alleles (using a P value cutoff of 0.2 from a cumulative binomial test) is considered to have an unwanted mutation. A lenient P value cutoff (0.2 as opposed to the more traditionally used 0.05 or 0.01) implies more stringent filtering in this case because we want to eliminate type II errors i.e., we want to identify all unwanted mutations at the cost of discarding a few clones that actually do not have any unwanted mutations.
We identified an average of 4–5 unwanted point mutations per pool. The overall per-base point mutation rate of Phusion polymerase was calculated to be ∼10−4. NEB's advertised error rate for Phusion polymerase varies from 4.4–9.5×10−7 per PCR cycle. Since we perform 18 PCR cycles, the expected overall error rate is ∼10−5. Our calculated mutation is within an order of magnitude of this advertised error rate. It is slightly higher than the advertised rate as we use stringent filtering criteria as described above.
All wild-type and mutant clones were moved into the pcDNA-DEST47 vector with a C-terminal GFP tag using automated Gateway LR reactions in a 96-well format. After bacterial transformation, minipreps were prepared on a Tecan Freedom Evo 200, and DNA concentrations were determined by OD 260/280 with a Tecan Infinite M1000 plate reader in 96-well format. A 100 ng aliquot of each expression clone plasmid was used for transfection into HEK293T cells in 96-well plates using Lipofectamine 2000 (Invitrogen 11668019) according to the manufacturer's instructions. At approximately 48 hrs post-transfection, cells were processed with Tecan M1000. Fluorescence intensities were measured at 395 nm for excitation and 507 nm for emission, according to Invitrogen's manual. As negative controls, the fluorescence intensities corresponding to cells transfected with the empty vector were measured. The normalized fluorescence intensity was calculated as:where I corresponds to the measured intensity and Ibackground corresponds to the average intensity of the empty vector controls for each plate. All Inorm values greater than K are considered to correspond to stable protein expression. K corresponds to the range (maximum – minimum) of background fluorescence intensities of the empty vector controls for each plate. For this study, all fluorescence intensity readings were also confirmed manually under a microscope. All transfection and GFP experiments were repeated 3 times.
Y2H was performed as previously described [7]. All wild-type/mutant clones were transferred by Gateway LR reactions into our Y2H pDEST-AD and pDEST-DB vectors. All DB-X and AD-Y plasmids were transformed individually into the Y2H strains MATα Y8930 and MATa Y8800, respectively. Each of the DB-X MATα transformants (wild-type and mutants) were then mated against corresponding AD-Y MATa transformants (wild-type and mutants) individually using automated 96-well procedures, including inoculation of AD-Y and DB-X yeast cultures, mating on YEPD media (incubated overnight at 30°C), and replica-plating onto selective Synthetic Complete media lacking leucine, tryptophan, and histidine, and supplemented with 1 mM of 3-amino-1,2,4-triazole (SC-Leu-Trp-His+3AT), SC-Leu-His+3AT plates containing 1 mg/l cycloheximide (SC-Leu-His+3AT+CHX), SC-Leu-Trp-Adenine (Ade) plates, and SC-Leu-Ade+CHX plates to test for CHX-sensitive expression of the LYS2::GAL1-HIS3 and GAL2-ADE2 reporter genes. The plates containing cycloheximide select for cells that do not have the AD plasmid due to plasmid shuffling. Growth on these control plates thus identifies spontaneous auto-activators [64]. The plates were incubated overnight at 30°C and “replica-cleaned” the following day. Plates were then incubated for another three days, after which positive colonies were scored as those that grow on SC-Leu-Trp-His+3AT and/or on SC-Leu-Trp-Ade, but not on SC-Leu-His+3AT+CHX or on SC-Leu-Ade+CHX. Disruption of an interaction by a mutation was defined as at least 50% reduction of growth consistently across both reporter genes, when compared to Y2H phenotypes of the corresponding wild-type allele as benchmarked by 2-fold serial dilution experiments. All Y2H experiments were repeated 3 times.
Wild-type MLH1, HSPA8, and BRIP1 entry clones are from the human ORFeome v8.1 collection [61]. Using Gateway LR reactions, wild-type MLH1, mutant MLH1 (I107R), and GFP were transferred into the pMSCV-N-FLAG-HA-PURO vector [65]; HSPA8 and BRIP1 were transferred into the pcDNA-DEST40 vector that contains a C-terminal V5 tag (Invitrogen 12274-015).
HEK293T cells were grown in SILAC media comprising SILAC DMEM (Thermo Scientific) and 10% dialyzed FBS (JR Scientific) supplemented with either 0.1 mg/ml L-lysine and L-arginine (light media) or 0.1 mg/ml L-lysine 13C6, 15N2 and L-arginine 13C6, 15N4 (heavy media). Heavy- or light-media cultured HEK293T cells were transfected using Lipofectamine 2000 (Invitrogen) in three 10 cm plates. 48 hrs after transfection, cells were washed three times in cold PBS and then resuspended in 5 ml RIPA buffer [1% NP-40, 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 5 mM EDTA, 1× EDTA-free Complete Protease Inhibitor tablet (Roche)]. Cells were lysed for 30 mins on ice before centrifuging at 13,000 rpm for 10 mins. Cell lysates were incubated with 60 µL EZview Red Anti-HA Affinity Gel (Sigma-Aldrich) for 3 hrs. After 3 washes with RIPA buffer, bound proteins were eluted with 3 resin volumes elution buffer (100 mM Tris-HCl pH 8.0, 1% SDS). Eluted proteins from light and heavy media were mixed together, reduced with 5 mM DTT, alkylated with 15 mM of iodoacetamide, and then precipitated with 3 volumes PPT solution (50% acetone, 49.9% ethanol, 0.1% acetic acid). Proteins from pull-down experiments were solubilized with 50 µL Urea/Tris solution (8 M Urea, 50 mM Tris-HCl pH 8.0) and 150 µL NaCl/Tris (50 mM Tris-HCl pH 8.0, 150 mM NaCl) followed by the addition of 1 µg Trypsin Gold (Promega). Protein digestion was performed overnight at 37°C after which trifluoroacetic acid and formic acid were added to a final concentration of 0.2%. Peptides were de-salted with Sep-Pak C18 columns (Waters Corporation), dried in a speed-vac, and reconstituted in 85 µL of a solution containing 80% acetonitrile and 1% formic acid. Samples were fractionated by Hydrophilic Interaction LIquid Chromatography (HILIC) using a TSK gel Amide-80 column (Tosoh Bioscience). HILIC fractions were dried in a speed-vac, reconstituted in 0.1% trifluoroacetic acid, and analyzed by LC-MS/MS using a 125 µM ID capillary column packed in-house with 3 µm C18 particles (Michrom Bioresources) and a Q-Exactive mass spectrometer (Thermo Fisher Scientific) coupled with a Nano LC-Ultra system (Eksigent). Xcalibur 2.2 software (Thermo Fischer Scientific) was used for the data acquisition and Q-Exactive was operated in the data-dependent mode. Survey scans were acquired in the Orbitrap mass analyzer over the range of 380 to 2000 m/z with a mass resolution of 70.000 (at m/z 200). Up to the top 10 most abundant ions with a charge state higher than 1 and less than 5 were selected within an isolation window of 2.0 m/z. Selected ions were fragmented by Higher-energy Collisional Dissociation (HCD) and the tandem mass spectra were acquired in the Orbitrap mass analyzer with a mass resolution of 17.500 (at m/z 200). The fragmentation spectra were searched by using the SEQUEST software on a SORCERER system (Sage-N Research) and a human database downloaded from the International Protein Index (version 3.80). In all database searches, trypsin was designated as the protease, allowing for one non-tryptic end and two missed-cleavages. The following parameters were used in the database search: a mass accuracy of 15 ppm for the precursor ions, differential modification of 8.0142 Daltons for lysine and 10.00827 Daltons for arginine. Results were filtered based on probability score to achieve a 1% false positive rate. The Xpress software, part of the Trans-Proteomic Pipeline (Seattle Proteome Center), was used to process the raw data and quantify the light/heavy peptide isotope ratios. Results were also manually inspected.
We performed four SILAC experiments using both wild-type and mutant Mlh1, as well as GFP as a control: wild-type (heavy) vs. control (light) [WT_Control]; mutant (heavy) vs. control (light) [Mutant_Control]; wild-type (heavy) vs. mutant (light) [WT_Mutant]; and mutant (heavy) vs. wild-type (light) [Mutant_WT].
We use the following variables and define four ratios for all subsequent calculations. In the WT_Control experiment, the relative abundance of protein p pulled down by wild-type Mlh1 to protein p pulled down by GFP (WTp) is quantified by the inverse of the geometric mean of rwc reads with Xpress values Xi. In the Mutant_Control experiment, the relative abundance of protein p pulled down by mutant Mlh1 (I107R) to protein p pulled down by GFP (Mutp) is quantified by the inverse of the geometric mean of rmc reads with Xpress values Yi. In the WT_Mutant experiment, the relative abundance of protein p pulled down with mutant Mlh1 (I107R) to protein p pulled down by wild-type Mlh1 is quantified by the geometric mean of rwm reads with Xpress values Pi. The amount of mutant Mlh1 (I107R) to wild-type Mlh1 is quantified by the geometric mean of twm reads with Xpress values Cj. In the Mutant_WT experiment, the relative abundance of protein p pulled down with mutant Mlh1 (I107R) to protein p pulled down by wild-type Mlh1 is quantified by the inverse of the geometric mean of rmw reads with Xpress values Qj. The amount of mutant Mlh1 (I107R) to wild-type Mlh1 is quantified by the inverse of the geometric mean of tmw reads with Xpress values Di.where both FCwm and FCmw denote the fold change in protein abundance as the normalized ratio of the amount of protein pulled down with mutant Mlh1 to that with wild-type Mlh1.
To identify interactors that are lost/weakened due to the I107R mutation, we required the following criteria to hold simultaneously:
The first criterion ensures that the protein identified is a true interactor of wild-type Mlh1. The second criterion ensures that the loss of interaction is significant and reliably observed across both WT_Mutant and Mutant_WT experiments.
Similarly, to identify interactors that are gained/enhanced due to the I107R mutation, we required the following criteria to hold simultaneously:
The first criterion ensures that the protein identified is a true interactor of the I107R mutant of Mlh1. The second criterion ensures that the gain of interaction is significant and reliably observed across both WT_Mutant and Mutant_WT experiments.
We also identify interactors of Mlh1 that are unaffected by the I107R mutation using the following criteria:
Integrating both WT_Mutant and Mutant_WT experiments, we calculated a weighted average of the individual fold changes:P values are calculated using a two-sided Kolmogorov-Smirnov test (with bootstrapping).
HEK293T cells were maintained in complete DMEM medium supplemented with 10% FBS. Cells were transfected with Lipofectamine 2000 (Invitrogen) at a 6∶1 (µL/µg) ratio with DNA in 6-well plates and were harvested 24 hrs after transfection. Cells were gently washed three times in PBS and then resuspended using 200 µL 1% NP-40 lysis buffer [1% Nonidet P-40, 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1× EDTA-free Complete Protease Inhibitor tablet (Roche)] and kept on ice for 20 mins. Extracts were cleared by centrifugation for 10 mins at 13,000 rpm at 4°C. 15 µL EZview Red Anti-HA Affinity Gel (Sigma-Aldrich) and 100 µL protein lysate were used for each co-immunoprecipitation reaction. The samples were rotated gently at 4°C for 2 hrs. HA beads were then washed three times with protein lysis buffer, treated with 6× protein sample buffer, and subjected to SDS-PAGE. Proteins were then transferred from the gel onto PVDF (Amersham) membranes. Anti-HA (Sigma H9658), anti-V5 (Invitrogen 46-0705), anti-β-tubulin (Promega G7121), and anti-GFP (Santa Cruz sc-9996) antibodies were used at 1∶3,000 dilutions for immunoblotting analysis.
|
10.1371/journal.pgen.1002652 | Context-Dependent Dual Role of SKI8 Homologs in mRNA Synthesis and Turnover | Eukaryotic mRNA transcription and turnover is controlled by an enzymatic machinery that includes RNA polymerase II and the 3′ to 5′ exosome. The activity of these protein complexes is modulated by additional factors, such as the nuclear RNA polymerase II-associated factor 1 (Paf1c) and the cytoplasmic Superkiller (SKI) complex, respectively. Their components are conserved across uni- as well as multi-cellular organisms, including yeast, Arabidopsis, and humans. Among them, SKI8 displays multiple facets on top of its cytoplasmic role in the SKI complex. For instance, nuclear yeast ScSKI8 has an additional function in meiotic recombination, whereas nuclear human hSKI8 (unlike ScSKI8) associates with Paf1c. The Arabidopsis SKI8 homolog VERNALIZATION INDEPENDENT 3 (VIP3) has been found in Paf1c as well; however, whether it also has a role in the SKI complex remains obscure so far. We found that transgenic VIP3-GFP, which complements a novel vip3 mutant allele, localizes to both nucleus and cytoplasm. Consistently, biochemical analyses suggest that VIP3–GFP associates with the SKI complex. A role of VIP3 in the turnover of nuclear encoded mRNAs is supported by random-primed RNA sequencing of wild-type and vip3 seedlings, which indicates mRNA stabilization in vip3. Another SKI subunit homolog mutant, ski2, displays a dwarf phenotype similar to vip3. However, unlike vip3, it displays neither early flowering nor flower development phenotypes, suggesting that the latter reflect VIP3's role in Paf1c. Surprisingly then, transgenic ScSKI8 rescued all aspects of the vip3 phenotype, suggesting that the dual role of SKI8 depends on species-specific cellular context.
| The production and turnover of messenger RNAs (mRNAs) are conserved processes in eukaryotes, from single-cell organisms to plants and mammals. To some degree, this is also true for modulators of these processes, such as the Paf1 and SKI complexes. One particular protein, SKI8, has been described to have a role in the SKI complex, which influences mRNA stability, both in yeast and in mammals. Moreover, in yeast SKI8 has an additional role in meiotic recombination, whereas in humans it influences mRNA production through association with the Paf1 complex. This functional divergence is commonly thought to arise from differences in protein sequence between the yeast and mammalian SKI8 homologs. Here we show that the conserved SKI8 homolog of the model plant Arabidopsis acts in the SKI complex as well as the Paf1 complex, similar to human. However, using an Arabidopsis ski8 mutant as a tool, we show that yeast SKI8 can fulfill all roles of Arabidopsis SKI8 if introduced into Arabidopsis cells. Thus, it appears that the functional divergence of SKI8 homologs might a priori be related to species-specific cellular context rather than divergence in protein sequence.
| Production and turnover of eukaryotic mRNAs are highly conserved processes, which are mainly driven by RNA polymerase II (RNAPolII) and the 3′ to 5′ exosome (exosome), respectively [1], [2]. Regulation of transcription initiation by RNAPolII through promoter sequence-specific transcription factors is a major topic in developmental biology, since it is considered the prime mechanism for differential, cell and organ type-specific gene expression [3]. However, generic accessory factors, which are typically heteromultimeric protein complexes, exist as well. Compared to the RNAPolII machinery, they are less conserved but have been found in all uni- and multicellular eukaryotes investigated so far. In line with their lower conservation, these factors are generally not essential. However, loss of function mutations in their subunits typically result in pleiotropic phenotypes with varying degrees of severity. An example is the Mediator complex, which typically comprises more than 15 subunits and interacts with the C-terminal domain of the largest RNAPolII subunit [4], [5]. In yeast (S. cerevisiae), Mediator is associated with constitutively transcribed genes [6] and yeast Mediator mutants are typically viable but display impaired growth [4]. In multicellular organisms, the composition of Mediator is even more complex and individual subunit loss of function can lead to rather specific phenotypes. For instance, in the model plant Arabidopsis (A. thaliana), in which several additional Mediator subunits have been identified [7], respective mutants display such diverse phenotypes as increased cell proliferation, shifts in embryonic patterning or early flowering [7], [8], [9].
Screens for early flowering mutants also identified Arabidopsis subunit homologs of another conserved multimeric regulator of transcription, the RNAPolII-associated factor 1 complex (Paf1c) [10], [11], [12]. In yeast, Paf1c consists of five subunits [13], whose Arabidopsis homologs are VERNALIZATION INDEPENDENCE (VIP) 4, VIP5, EARLY FLOWERING (ELF) 7, VIP6/ELF8 and PLANT HOMOLOGOUS TO PARAFIBROMIN (PHP) [10], [14], [15], [16]. Among the respective loss of function mutants, php mutants only flower early, whereas vip4, vip5, elf7 and vip6/elf8 mutants all display additional pleiotropic growth defects and aberrant flower development (e.g., variable floral organ number). The early flowering phenotype of vip/elf mutants has been linked to down-regulation of the central flowering time regulator, FLOWERING LOCUS C (FLC), via an epigenetic mechanism, consistent with a role of Paf1c in chromatin modification through changing histone methylation patterns [10], [11], [14]. The latter could also explain the phenotypes in flower development, which can be altered by mutation in epigenetic regulators [17].
Another mutant with dwarf, early flowering and aberrant flower development phenotypes is vip3. VIP3 encodes a WD40 repeat protein, which is the putative Arabidopsis homolog of the yeast Superkiller (Ski) 8 gene [12]. SKI8 is part of the cytosolic SKI complex, which is thought to positively regulate exosome activity [1], [18], [19]. The SKI complex consists of a SKI8 dimer and the SKI2 RNA helicase, which are connected by their mutual interaction with the scaffold protein SKI3 [20]. Interestingly, human hSki8 as well as VIP3 also associate with Paf1c [11], [21], which is not the case for yeast ScSki8 [21], [22]. Rather, ScSki8 has a SKI complex-independent nuclear function in meiotic recombination [23]. This feature is not conserved in VIP3 [24], suggesting that Ski8 activity in plants might be functionally closer to mammals than unicellular eukaryotes.
Compared to its well documented role in Paf1c, the potential role of VIP3 in the SKI complex has not been characterized. Notably, although VIP3 is the top hit in a homology search of the Arabidopsis proteome using ScSki8 as a query, the next best hits are nearly equally significant with better overall coverage and represent structurally similar WD40 repeat proteins. Conversely, if the yeast proteome is queried with VIP3, more than two dozen hits score markedly better than ScSki8. Interestingly, the top hits, like PRP4 or TUP1, have been described as modifiers of pre-mRNA processing or chromatin modifications, respectively [25], [26], which would also be consistent with existing experimental data on VIP3 activity. However, reciprocal BLAST searches with higher eukaryotes clearly identify the respective SKI8 homologs as best hits. Still, experimental evidence for VIP3 involvement in the Arabidopsis SKI complex and the facets of the vip3 phenotype that could be attributed to this role is missing. In this study, we investigate this question by a combination of biochemical, genetic and high throughput techniques.
Analysis of natural genetic variation has become a common tool for isolation of allelic variants in Arabidopsis, facilitated by availability of collections of wild strains, so-called accessions. In the Slavice-0 (Sav-0) accession, we found largely infertile dwarf plants segregating at low frequency when grown in permanent light conditions and low humidity (∼40%) (Figure 1A). Moreover, careful investigation of the segregating population revealed a substantial fraction of non-viable, seedling lethal individuals. At higher humidity (∼60%) and long day conditions, the fraction of seedling lethals decreased considerably, whereas the ratio of dwarfs increased to near Mendelian (typically>20%) proportion, suggesting that the two classes represent the phenotypic spectrum of the same underlying genetic cause. The infertility of the dwarf plants, which rarely produced seeds and if so, very little (Figure 1E–1G), could be overcome to some degree by out-crossing with pollen from wild type looking plants. This allowed us to generate a segregating F2 population derived from a cross to the standard lab accession, Columbia-0 (Col-0). Genetic mapping revealed that the dwarf and infertility phenotypes segregated as a recessive single Mendelian locus on the lower arm of chromosome 4, which we named ZWERGERL (ZWG, Bavarian for “dwarf”).
A detailed analysis of the zwg phenotype revealed various floral defects. These included a low penetrance aberrant floral organ number phenotype (Figure 1B–1C) and shorter sepals (Figure 1D). Shorter anthers with few viable pollen accounted for the decreased fertility. The floral phenotypes were accompanied by early flowering, which was evident both in terms of age and rosette leaf number (Figure 1H). By following the development of individual seedlings from germination in tissue culture onwards, we could also detect reduced root elongation in zwg plants (Figure 1I). Radial growth of all organs was affected as well, as exemplified by the dramatically reduced diameter of the main inflorescence stem (Figure 1J–1M). Since the relative decrease in cell number (∼75% of wild type) (Figure 1J) was not as strong as the overall decrease in diameter (∼50% of wild type) (Figure 1K–1M), the reduced organ size in zwg mutants likely represents a combination of impaired cell proliferation and expansion.
Complementing the morphological characterization, we also analyzed the zwg transcriptome by hybridization of CATMA microarrays [27] with cDNA prepared from aerial tissues. Based on four replicate hybridizations, statistically solid expression changes (≥2-fold; p≤0.05) were found for 173 genes that were up-regulated and 425 genes that were down-regulated in zwg as compared to wild type (Table S1). These gene lists did not point to any specific defect in zwg mutants, such as mis-regulation of a particular hormone pathway. Rather, the genes represented an overall balanced sample across functional categories as illustrated by their gene ontology analysis (Figure 1N–1O). The only consistently over-represented category in both the up- and down-regulated sets was response to stress. In summary, our morphological as well as molecular characterization suggests that a general growth defect is responsible for the panoply of zwg mutant phenotypes.
To identify the molecular cause of the zwg mutation, we sequenced the genomes of Sav-0 wild type and zwg individuals with short reads [28]. Mapping of the reads onto the Col-0 reference genome revealed an extended region of heterozygosity on the lower arm of chromosome 4 in Sav-0 that encompassed the ZWG locus. The sequence information was exploited to generate polymorphic molecular markers that allowed mapping of the zwg mutation in the zwg x Col-0 population (Figure 2A). Within the zero recombination mapping interval, the sequence reads indicated the presence of a homozygous 7 bp deletion in the coding sequence of At4g29830, previously described as VIP3, in zwg but not in wild type (Figure 2B), which was confirmed by Sanger sequencing of respective PCR fragments. Analysis of a cross between zwg and a vip3 null mutant (SALK_083364) [29] indicated non-complementation, confirming that zwg is indeed a new vip3 allele, which we thus named vip3zwg. The deletion in vip3zwg encompasses nucleotides 861–867 of the open reading frame of the mRNA, which is expressed at similar levels in vip3zwg and wild type. Conceptual translation predicts that the deletion causes a frameshift to produce a 36 kDa instead of a 32 kDa protein with a modified and extended C-terminus, thereby disrupting the last of the five WD40 repeats of VIP3 (Figure 2C). Because of its phenotypic resemblance with the knock out allele, including the down-regulation of FLC expression (Table S1), and its recessive behavior, vip3zwg can be considered a null allele.
To clarify whether VIP3 is indeed the functional Arabidopsis SKI8 homolog, we sought to determine its subcellular localization. To this end, we created a binary construct for expression of a GFP-VIP3 fusion under control of the constitutive 35S promoter. This transgene was introduced into Sav-0 wild type-looking plants that were heterozygous for the zwg mutation as determined by genotyping of the 7 bp deletion on high resolution agarose gels. Western analysis revealed variable expression of GFP-VIP3 fusion protein of the expected size in several independent lines (Figure 2D), within one order of magnitude of the level of endogenous VIP3 as judged from qPCR. In the progeny of these plants, the zwg phenotype segregated in a proportion close to 1/16th rather than 1/4th and was significantly different from the segregation in the parallel grown non-transgenic mother line (Chi-square = 11.54 for df = 1, significant at p<0.001). None of the plants with a zwg phenotype carried the transgene as determined by genotyping. All other plants appeared wild type, suggesting that the fusion protein is functional and rescues all zwg phenotypes. Confocal microscopy showed both cytoplasmic and nuclear (but not nucleolar) localization of GFP-VIP3, in differentiated as well as proliferating cells, in both root and shoot tissues (Figure 2E–2G). Matching the dual subcellular localization, analysis of protein extracts by gel filtration detected the presence of GFP-VIP3 in at least two peaks, one in the ∼690 kDa and another in the 300 kDa range (Figure 2H). Moreover, substantial amounts were observed in smaller (100–200 kDa) fractions. To determine whether any of these fractions could represent the SKI complex or its sub-components, we collected three distinct sets of fractions after gel filtration and performed immunoprecipitations with anti-GFP antibody. Subsequent MALDI-TOF identified peptides of the Arabidopsis SKI3 homolog (At1g76630) in the pool of the smaller fractions (Figure 2I). Notably, protein homology searches unequivocally identify At1g76630 and ScSki3 as unique reciprocal and highly significant hits, suggesting that At1g76630 represents indeed the Arabidopsis SKI3 homolog. No other SKI complex or Paf1c components were identified, which might have resulted from our stringent conditions combined with previous gel filtration. Direct immunoprecipitation from total protein extract using the same conditions indeed not only identified AtSKI3, but also AtSKI2 (At3g46960, see below) and the Paf1c component PHP (Figure 2J). Thus, our analyses suggest that VIP3 is not only part of Paf1c in the nucleus, but also of the cytoplasmic SKI complex and likely represents the true SKI8 homolog.
To corroborate the consequent notion that VIP3 should have a role in mRNA turnover, we applied a high throughput sequencing strategy to RNA samples isolated from vip3zwg and wild type. Because we aimed to sequence both full length mRNAs and mRNAs undergoing (3′ to 5′) degradation, cDNA from these samples was produced by random-primed rather than poly-T-primed synthesis. Prior to this, mRNA was enriched by removing the bulk of ribosomal RNA with the help of capture columns. The cDNA was then size-fractionated and the 200 bp fraction was used for preparation of the library, which was sequenced to produce single reads of 75 bp (21.3 mio. for wild type; 25.2 mio. for vip3zwg). The reads were mapped onto the Col-0 reference transcriptome, including the 5′ and 3′ UTRs, with relaxed stringency to accommodate nucleotide polymorphisms between Sav-0/vip3zwg and Col-0 [28]. Parallel mapping onto the Col-0 reference genome placed the large majority of reads in exons (80.2% in wild type; 84.5% in vip3zwg), confirming that our sequence data represent RNA molecules and that genomic contamination, if any, is negligible (Table S2). For the follow up analyses, we concentrated on the reads that mapped onto mRNA (27.0% in wild type; 13.8% in vip3zwg), and in particular on the nuclear encoded genes (16.7% of reads in wild type; 6.6% in vip3zwg).
In total, of 26’598 transcripts interrogated, 14’228 were covered by at least one read in both the wild type and vip3zwg sample. In order to obtain a parameter that would allow us to estimate the steady state abundance of full length versus degrading mRNAs, we calculated the ratio between the number of reads mapping onto the 5′-most 20% of a transcript versus those mapping onto the 3′-most 20%. After removal of nonsense values (i.e. 0 or ∞ because of absent coverage of one end) and outliers with extreme values (resulting from excess read abundance combined with obvious mis-mapping, e.g. reads covering the flanking region of a gypsy-like retrotransposon [At4g06477]), the distribution of this 5′ to 3′ coverage index was skewed towards values >1. To some degree this likely represents a technical bias [30], but could also reflect a dominant role of 3′ to 5′ degradation in mRNA turnover. Interestingly, the 5′ to 3′ coverage index was generally higher in the wild type than in the vip3zwg sample (Figure 3A). To verify that this was not a technical artifact, we compared the relative proportion of the accumulated reads in 1% bins along the 10% most highly expressed nuclear encoded transcripts of wild type. The respective profiles for the wild type and vip3zwg sample were similar (Figure 3B), suggesting that the RNA sequencing data from the two samples are comparable.
In order to remove statistically doubtful 5′ to 3′ coverage index values that were due to low transcript abundance, we only considered the 6’500 transcripts for which at least 50% of sequence was covered in both the wild type and vip3zwg samples in follow up analyses. Remaining outliers with index values ≥10 or ≤0.1 were removed as well. From this set, we extracted the group of 5’617 nuclear encoded transcripts that were not strongly affected by depletion of exosome activity [31], as well as 34 chloroplast-encoded transcripts and 68 nuclear-encoded transcripts that were significantly stabilized upon exosome depletion and can be considered prime exosome targets [“the hidden transcriptome”; 31] (Table S3). The 5′ to 3′ coverage index value distribution in the nuclear control group confirmed the earlier picture of higher overall values in wild type as compared to vip3zwg, which is for instance also evident in the ratios between the respective averages or medians (Figure 3C). While no significant difference was found in the chloroplast transcripts, this trend was amplified in the prime exosome targets, which displayed higher average and median index values in wild type and lower ones in vip3zwg as compared to the nuclear control group (Figure 3C).
To evaluate the robustness of the difference between the nuclear control group and the exosome targets, we determined the index value distribution for 1’000 random sets of 68 genes extracted from the nuclear control group. These analyses confirmed the trend towards higher values in wild type, underlined by the finding that the wild type to vip3zwg ratio of averages and medians was nearly always >1 (Figure 3D). Notably, even the maximum ratios observed within the 1’000 sets did not or barely reach the values observed in the exosome target set (1.50 versus 1.53 for the average, 1.40 versus 1.39 for the median). Conversely, within 59 random sets of 10 transcripts extracted from the exosome targets, the trend towards higher values in wild type and lower ones in vip3zwg including the ratios was always evident (Figure 3D). We confirmed this finding by an independent method with independent, triplicate RNA preparations for a set of five randomly chosen genes. For each gene, oligonucleotide pairs for qPCR detection of the respective 5′ and 3′ mRNA ends were designed. The reverse primers for each fragment were used to prime separate cDNA synthesis reactions. Subsequent qPCR allowed quantification of the 5′ and 3′ end abundance for each gene in the replicate samples of the two genotypes. With one exception, the ratio between the 5′ and 3′ end abundance was always higher in wild type than in vip3zwg mutants, as expressed by the ratio between those ratios being greater than 1 (Figure 3E). In summary, these analyses suggest that nuclear encoded mRNAs in general and prime exosome targets in particular are stabilized in vip3zwg mutants.
To determine which aspects of the phenotype spectrum of vip3 mutants are due to its involvement in Paf1c or the SKI complex, respectively, we sought to characterize mutants in other SKI subunit homologs of Arabidopsis. Whereas knock out mutants in the SKI3 homolog were not available in reverse genetic collections [29], a line segregating a T-DNA insertion in exon 9 out of 23 of At3g46960 (SALK_118579) was available. Similar to At1g76630 and ScSki3, reciprocal homology searches between Arabidopsis and yeast using ScSki2 or At3g46960 as a query identified each other as the uncontested top hits, suggesting that At3g46960 represents the unique ScSki2 homolog in Arabidopsis (AtSKI2). This notion is also supported by a phylogenetic analysis (Figure 4A; Text S1). Analysis of the SALK_118579 line revealed that it segregates up to ∼25% of dwarf plants (Figure 4B–4C). This phenotype co-segregated perfectly with homozygosity of the T-DNA insert and absence of full length AtSKI2 mRNA. With the caveat that residual RNA production 3′ from the T-DNA insertion site has been reported previously [32], the SALK_118579 line therefore might represent the Atski2 null mutant phenotype. Contrary to the vip3 mutants however, the dwarf phenotype was neither accompanied by a flower development nor an early flowering phenotype (Figure 4D–4E). In line with the latter observation, FLC expression was strongly diminished in vip3zwg, but not in Atski2 mutants (Figure 3F). In summary, these observations suggest that the flower development and early flowering phenotype of vip3 mutants could reflect VIP3's role in Paf1c rather than the SKI complex.
Considering that ScSki8 has not been found to associate with Paf1c, we sought to corroborate this notion by testing whether transgenic ScSki8 could rescue the dwarf phenotype of vip3zwg mutants. To this end, the ScSki8 open reading frame was cloned into a binary construct for constitutive expression under control of the 35S promoter. Again, the transgene was introduced into Sav-0 wild type plants that were heterozygous for the vip3zwg mutation. Genotyping of the 7 bp deletion and the transgene in the segregating progeny identified several homozygous vip3zwg mutants carrying the 35S::ScSki8 transgene. These plants developed either as dwarf or as wild type, and this was correlated with transgene expression (Figure 4F–4G). Thus, transgenic expression of ScSki8 could rescue the dwarf phenotype of vip3zwg mutants. Moreover, surprisingly both the early flowering and flower development phenotypes were also rescued (Figure 4H). Therefore, our data suggest that once introduced into Arabidopsis, ScSki8 can fulfill all functions of VIP3, including those not normally encountered in yeast itself.
In this study, we present experiments that lead to four main conclusions: First, we show that VIP3 is the bona fide SKI8 homolog of Arabidopsis; second, we demonstrate that next generation sequencing of random-primed RNA samples with short reads can be used to estimate the turnover of mRNA transcripts; third, we show that the phenotypic aspects of VIP3 function in Paf1c and the SKI complex can be separated; and fourth, we provide evidence that the dual role of SKI8 homologs in Paf1c and the SKI complex appears to depend on the species-specific cellular context.
Our interest in VIP3 originates from the discovery of the zwg mutant that segregated in the Arabidopsis Sav-0 accession. It seems unlikely that the 7 bp deletion in vip3zwg represents indeed an allelic variant recovered from a natural environment because of its detrimental phenotypic consequences. A haplo-insufficient beneficial effect of vip3zwg could explain maintenance of the allele by balancing selection, however, we did not observe any obvious phenotypes in the heterozygous plants that would support this idea. Rather, it appears likely that vip3zwg is a spontaneous allele that has arisen during the propagation of the Sav-0 accession in stock centers starting in the 1960s over several decades [33], [34].
At the outset of our study, it was still unclear whether VIP3 is indeed the Arabidopsis SKI8 homolog. While its role in epigenetic regulation of FLC transcription through association with the Paf1c complex had been well documented [11], [14], its potential role in the SKI complex had not been characterized. Because of the evolutionary distance between higher plants, yeast and mammals this could not be considered a given, in particular as VIP3 and SKI8 fall into an abundant class of structurally similar WD40 repeat proteins. This was underlined by the finding that ScSki8 is by far not the closest VIP3 homolog in yeast. For instance, position-specific iterated BLAST identifies more than two dozen yeast proteins that are more homologous to VIP3 than ScSki8 (e.g., a 93.2 score, 69% coverage and 5×10−24 e-value for PRP4 as compared to 48.9 score, 40% coverage and 5×10−8 e-value for ScSki8). It is only our functional analyses that suggest that VIP3 is indeed the bona fide ScSki8 homolog. Consistent with a potential role in both Paf1c and the SKI complex, we found that VIP3 is present in both the nucleus and cytoplasm, and in at least two protein complexes of distinct size. The larger peak fractions around 690 kDA could represent Paf1c, whereas the peak around 300 kDa could represent the SKI complex [21]. A third peak around even smaller (100–200 kDa) size fractions could represent partial components of these complexes or VIP3 dimers, which might accumulate in excess as the GFP-VIP3 transgenes were typically expressed at higher levels than endogenous VIP3. Interestingly, immunoprecipitation of GFP-VIP3 after gel filtration identified association with the Arabidopsis SKI3 homolog, but not the SKI2 homolog. This might mean that the SKI complex dissociates into sub-components during gel filtration and/or that SKI2 is lost during immunoprecipitation washes. Alternatively, it could reflect the fact that SKI8 interaction with SKI3 is direct, while interaction with SKI2 is indirect [20]. However, when directly immunoprecipitated from total protein extract, AtSKI3 as well as AtSKI2 was pulled down in our stringent conditions, underlining that VIP3 is indeed part of the SKI complex.
The notion that VIP3 is a functional subunit of the SKI complex is supported by our genome-wide analysis of mRNA stability in vip3zwg mutants. To estimate mRNA turnover was foremost a technical challenge, because it meant that standard cDNA synthesis using poly-T oligonucleotides directed against the 3′ poly-A tail of mRNAs could not be applied. This also abolished the inherent selection of the mRNA fraction for sequencing from the much larger amount of ribosomal or transfer RNAs. Instead, to also capture mRNAs undergoing 3′ to 5′ degradation, cDNA was synthesized with random-priming, and the mRNA fraction was enriched by removing ribosomal RNAs through capture columns. High throughput sequencing of the cDNA samples and subsequent read mapping onto the reference transcriptome revealed that our method efficiently enriched the mRNA fraction, which generally represents 1–2% in total RNA samples, about 5 to 10-fold. The relative read abundance along transcripts is to some degree determined by technical biases, such as the directionality of cDNA synthesis [30]. However, it should also reflect the steady state equilibrium between mRNA synthesis and breakdown considering that primers were not limiting in cDNA synthesis and that poly-A tails provide priming sites but are not included in the sequence analysis. Generally, the coverage profiles displayed a decrease from 5′ to 3′, suggesting that exosome-mediated 3′ to 5′ degradation is the main driver of mRNA breakdown [18], [35]. To quantify the stability of individual transcripts, we defined a 5′ to 3′ coverage index, which was generally >1, consistent with the overall profile. The comparison of the 5′-most 20% of a transcript versus its 3′-most 20% was designed to avoid skewed values in the case of poorly covered transcripts, and indeed comparatively few outliers were observed. In some cases, these reflected obvious mismappings because of repetitive or redundant sequences (e.g. retrotransposon borders), while in others mismapping might have occurred because of the relaxed stringency that was required to map mRNA sequences from a divergent accession onto the reference transcriptome [28]. Overall, the patterns as well as the quantitative difference between the wild type and vip3zwg samples were robust, even if more selective criteria were applied or if other indexes were considered, such as linear fitting of read coverage. Thus, the index values suggest that in the vip3zwg sample the relative abundance of intact 3′ ends as compared to 5′ ends is higher, pointing to a shifted steady state equilibrium between mRNA transcription and degradation. This finding is consistent with the generic role of the SKI complex in exosome activation [18] and was particularly evident in the group of the most prominent exosome targets, termed the “hidden transcriptome” [31]. In summary, our data support the idea that VIP3 is a SKI complex component that affects mRNA stability and that random-primed RNA-Seq is a valid approach to estimate mRNA turnover.
The implication of VIP3 in the SKI complex suggests that the vip3 phenotype should reflect the combination of VIP3 function in both Paf1c and the SKI complex. The availability of a mutant in the AtSKI2 gene, which can be unequivocally identified by homology searches, enabled us to disentangle the two activities. Interestingly, Atski2 plants displayed dwarfism, but neither early flowering nor aberrant flower development. Thus, the latter aspects of the vip3 phenotype should primarily result from impaired Paf1c function. It is noteworthy however that the Atski2 dwarf phenotype is not as severe as in vip3, and that growth defects have also been observed in mutants of other Paf1c components. It thus appears likely that the SKI complex-related growth defects in vip3 are aggravated by the additionally impaired Paf1c activity.
To clarify more directly which portions of the vip3 phenotype are attributable to impaired Paf1c or SKI complex function, we sought to exploit the fact that ScSki8 does not associate with Paf1c in yeast [21], [22] and presumably also not in Arabidopsis. However, to our surprise ScSki8 was able to fully rescue all aspects of the vip3 phenotype. Thus, it appears that in the cellular context of Arabidopsis, ScSKI8 can fulfill VIP3's role in Paf1c. This could mean that other factors determine whether SKI8 is recruited to Paf1c or not, and that in this sense Arabidopsis is closer to mammals than yeast. Indeed we also tried to complement vip3zwg by constitutive expression of the mouse SKI8 homolog, WDR61. However, for unknown reasons, we never managed to recover transgenic plants in repeated transformation attempts, which could mean that WDR61 expression is poisonous for Arabidopsis. Thus, while cellular context must play an important role, SKI8 function might to some degree also depend on inherent features. Future experiments to determine the interaction patterns of different SKI8 homologs and derivative point mutants of interest are a promising avenue to clarify this issue in detail.
Molecular biology and genetics standard procedures, such as plasmid construction, genomic DNA isolation, qPCR, genotyping, or gene mapping were performed as described [36], [37].
The Sav-0 accession used in this study has been described previously [28]. T-DNA insertion lines were obtained from the Nottingham Arabidopsis Stock Centre and the insertions in lines SALK_083364 [knock out of VIP3 (At4g29830)] and SALK_118579 [knock out of AtSKI2 (At3g46960)] were confirmed by PCR analysis. For propagation and analysis of lines, seeds were germinated on half-strength Murashige & Skoog media in tissue culture and transferred to soil at 10–12 days after germination. Plants were then grown in either permanent light and ∼40% humidity, or 16 hr light–8 hr dark cycles and ∼60% humidity at 22°C. The latter conditions were used for characterization of the phenotypes displayed in the figures. For determination of flowering time, seeds were germinated directly on soil and the number of days or of rosette leaves was scored on the first day when the inflorescence meristem became visible. For root growth measurements, seedlings grown vertically in tissue culture were scored at 9 days after germination using ImageJ software and then transferred onto soil to determine wild type or mutant phenotype in the adult shoot.
For transverse sections, stem segments encompassing the first internode were cut with a razor blade and embedded in 6% agarose. From these samples, 85 µm sections were obtained using a Leica-VT 1000S vibratom and photographed using a Leica Diaplan 3 microscope. Subcellular localization of GPF-VIP3 was determined in shoots and roots of 6 day old seedling using a Zeiss LSM 510 confocal microscope. For the propidium iodide staining the roots were incubated for 2–5 min in a 50 µg/ml solution.
For expression of VIP3 or ScSki8 under control of the 35S promoter, the respective open reading frames amplified from cDNA samples were cloned into vectors pMDC43 or pMD32 [38], respectively. Construct integrity was verified by Sanger sequencing before transfer into Agrobacterium and transformation of Arabidopsis plants by the floral dip method. Constructs were introduced into Sav-0 wild type looking plants that were heterozygous for vip3zwg as determined by genotyping. For genotyping, PCR was performed on genomic DNA using oligonucleotides GAG CTG CGA TTC AGA CAA TGA G and GCC CGG ACA CCG GTT CCA C. The PCR products of 87 bp from the wild type or 80 bp from the vip3zwg allele were resolved on 4% agarose gels. Transformants were selected on hygromycin and homozygous vip3zwg plants among the transformants were selected by genotyping. Transgenic GFP-VIP3 or ScSki8 expression levels were determined by quantitative real time or semi-quantitative RT-PCR, respectively, and normalized compared to the EF1 gene as described [36].
For microarray analysis, a mix of equivalent amounts of aerial tissues (rosette leaves, stems, cauline leaves, inflorescences) from 4 week old adult plants was collected and frozen in liquid nitrogen before total RNA was prepped using the QIAGEN RNeasy Plant Mini Kit. cDNA synthesis, labeling, hybridization onto CATMA microarrays [27] and data analysis was then performed as described previously [36].
For RNA-Seq, rosette leaves were harvested from 25 day old phenotypically wild type or vip3zwg plants. Genomic DNA was isolated from one part of each sample to verify genotypes, whereas total RNA was prepped from the remaining tissue using a QIAGEN RNeasy Plant Mini Kit. Ribosomal RNA was subsequently largely removed by treating 10 µg of total RNA with Invitrogen RiboMinus Plant Kits following the manufacturer's instructions. The enriched mRNA samples were then subjected to random-primed cDNA synthesis, amplification, size selection and high throughput sequencing with 75 bp single reads on an Illumina instrument as described [28]. Read mapping onto the Col-0 reference genome or transcriptome (main gene models, TAIR 9.0 release) was performed using the BWA program [39] with a seed length of 50 bp and up to 5 mismatches or gaps allowed.
Total protein was extracted from 35S::GFP-VIP3 or 35S::GFP transgenic plants as described before fractionation by gel filtration using an Amersham Superdex 200 10/300 GL FPLC column with a buffer flow rate of 0.5 ml/min [40]. Consecutive 0.5 ml fractions were collected, concentrated and subjected to 10% SDS–PAGE followed by protein immunoblot analysis. Fusion protein was detected using an anti-GFP antibody (dilution 1∶3000) (Living colors, Clontech). For co-immunoprecipitation of GFP-VIP3, three consecutive sets of gel filtration fractions were pooled and incubated for 90 min. at 4°C with 50 µl of μ-magnetic beads conjugated to anti-GFP antibody (μMACS anti-GFP MACS, Miltenyi Biotec). The slurry was passed through a magnetic column, washed 5 times with protein extraction buffer before elution of proteins with hot protein loading buffer. Samples were analyzed by immunoblot (anti-GFP) and silver staining prior to MALDI-TOF analysis. For MALDI-TOF, samples were migrated on a 12% mini polyacrylamide gel for about 2.0 cm, and rapidly stained with Coomassie blue. Entire gel lanes were excised into 5 equal regions from top to bottom and digested with trypsin (Promega) as described [41], [42]. Data-dependent LC-MS/MS analysis of extracted peptide mixtures after digestion with trypsin was carried out on a hybrid linear trap LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific) interfaced to a nanocapillary HPLC equipped with a C18 reversed-phase column (Agilent Technologies). Collections of tandem mass spectra for database searching were generated from raw data with Mascot Distiller 2.3.2, and searched using Mascot 2.3 (Matrix Science) against the 2011_03 release of the UNIPROT database (SWISSPROT+TrEMBL, www.uniprot.org), restricted to Arabidopsis thaliana taxonomy (50’756 sequences after taxonomy filter). Mascot was searched with a fragment ion mass tolerance of 0.50 Da and a parent ion tolerance of 10 ppm. The digestion enzyme trypsin was specified with one missed cleavage. Iodoacetamide derivative of cysteine was specified as a fixed modification. N-terminal acetylation of protein, deamidation of asparagine and glutamine, and oxidation of methionine were specified as variable modifications. The software Scaffold (version Scaffold_3.0.9, Proteome Software Inc.) was used to validate MS/MS based peptide (minimum 90% probability [43]] and protein [min 95% probability [44]) identifications, perform dataset alignment as well as parsimony analysis to discriminate homologous hits.
To determine the abundance of 5′ and 3′ ends of selected mRNAs, total RNA was prepared from three independent wild type and vip3zwg replicate samples. Separate cDNA syntheses were performed for each individual gene fragment, followed by qPCRs that were performed as described [36] to detect the respective 5′ and 3′ ends of the transcripts. The following oligonucleotides were used: AT1G01010: GAC AGC TCA ACA CTT TTC CAC TTC and CTT TTA TCC TAA ACA AGA CCC GTA AAG (5′ end); GAA CGA AGC ATG TTT GAT TTA TCA TTG and TTG TTG GTG GTT CAT TGG AGT ACA (3′ end); AT1G24706: GTT CCT CTC CCT TTT CAT CTT ATC G and CAT CTT AAA CCC CTT TCG TGT GTA T (5′ end); GAT TTG CAG ATC CTT TGG TTT GTT C and GCT ATG AAT ATA TCT GAA GTC TGG CAA G (3′ end); AT1G64570: CCA TTT ATC GAT TCT TCA CAG ACA CG and GAT TTC ATG ACT CAA ATT AGG GTT CCA (5′ end); GAT GCT GAG GAT GAG TAA GTT CCT TC and GCT AGT AAT CTG CAT TCA AAC AGC ACT A (3′ end); AT3G02830: CAC TAC CTC TCA CCT CTC TGT TTA CAC and CCA TAG ACG TGA AGA GGA AGA ATG (5′ end); GAA GAA ACA AAG GAA GAA GAA GAA GAG and CCA TAG ACG TGA AGA GGA AGA ATG T (3′ end); AT5G56860: ATT GAT GAG ATA AAC AAA TGA AGA CAC AAA G and CCA TGT GTG TTT GGC TCG TGT C (5′ end); GTT GAT CAG ATC ATC ACA ATA TCC TCA TTA C and GCT ATT AAT TAT CAT ATT AAA CTC TCA CAC ACT CT (3′ end); For detection of FLC expression in relation to the EF1 housekeeping gene, qPCR was performed as described [36] using the same oligonucleotide pairs.
|
10.1371/journal.pgen.1000176 | Somatic Pairing of Chromosome 19 in Renal Oncocytoma Is Associated with Deregulated ELGN2-Mediated Oxygen-Sensing Response | Chromosomal abnormalities, such as structural and numerical abnormalities, are a common occurrence in cancer. The close association of homologous chromosomes during interphase, a phenomenon termed somatic chromosome pairing, has been observed in cancerous cells, but the functional consequences of somatic pairing have not been established. Gene expression profiling studies revealed that somatic pairing of chromosome 19 is a recurrent chromosomal abnormality in renal oncocytoma, a neoplasia of the adult kidney. Somatic pairing was associated with significant disruption of gene expression within the paired regions and resulted in the deregulation of the prolyl-hydroxylase ELGN2, a key protein that regulates the oxygen-dependent degradation of hypoxia-inducible factor (HIF). Overexpression of ELGN2 in renal oncocytoma increased ubiquitin-mediated destruction of HIF and concomitantly suppressed the expression of several HIF-target genes, including the pro-death BNIP3L gene. The transcriptional changes that are associated with somatic pairing of chromosome 19 mimic the transcriptional changes that occur following DNA amplification. Therefore, in addition to numerical and structural chromosomal abnormalities, alterations in chromosomal spatial dynamics should be considered as genomic events that are associated with tumorigenesis. The identification of EGLN2 as a significantly deregulated gene that maps within the paired chromosome region directly implicates defects in the oxygen-sensing network to the biology of renal oncocytoma.
| Together, renal oncocytoma and chromophobe renal cell carcinoma (RCC) account for approximately 10% of masses that are resected from the kidney. However, the molecular defects that are associated with the development of these neoplasias are not clear. Here, we take advantage of recent advances in genetics and computational analysis to screen for chromosomal abnormalities that are present in both renal oncocytoma and chromophobe RCC. We show that while chromophobe RCC cells contain an extra copy of chromosome 19, the renal oncoctyoma cells contain a rarely reported chromosomal abnormality. Both of these chromosomal abnormalities result in transcriptional disruptions of EGLN2, a gene that is located on chromosome 19 and is critical for the cellular response to changes in oxygen levels. Defects in oxygen sensing are found in other types of kidney tumors, and the identification of EGLN2 directly implicates defects in the oxygen-sensing network in these neoplasias as well. These findings are important because the chromosomal defect present in renal oncocytomas may also be present in other tumor cells. In addition, deregulation of EGLN2 reveals a unique way in which perturbations in oxygen-sensing are associated with disease.
| Cellular adaptation to changes in oxygen tension is vital for the integrity, maintenance and survival of cells. Hypoxia-inducible factor (HIF), the major transcription factor of the ubiquitous oxygen-sensing pathway, is a heterodimer composed of α and β subunits [1]. While HIFβ is constitutively expressed and stable, HIFα is oxygen-labile by the virtue of the oxygen-dependent degradation (ODD) domain, which undergoes rapid oxygen-dependent ubiquitin-mediated destruction [2]–[5]. Thus, the stability of HIFα dictates the transcriptional activity of HIF [6]. Critical regulators of HIFα stability are the prolyl-hydroxylase domain-containing enzymes (PHD/EGLNs) that hydroxylate HIFα on conserved prolines within the ODD domain in the presence of oxygen [7],[8]. Hydroxylated HIFα is recognized by the von Hippel-Lindau (VHL) protein. VHL is the substrate-conferring component of an E3 ubiquitin ligase called ECV (Elongins/Cul2/VHL) that specifically polyubiquitinates prolyl-hydroxylated HIFα for subsequent destruction via the 26S proteasome.
Deregulation of HIFα regulatory proteins has been strongly associated with cancer development. Germline inheritance of a faulty VHL allele on chromosome 3p25 is the cause of VHL disease, characterized by a high frequency of clear cell renal cell carcinoma (RCC), cerebellar hemangioblastoma, pheochromocytoma, and retinal angioma [9]. Inactivation of the remaining wild-type VHL allele in a susceptible cell leads to tumor formation. Somatic biallelic inactivation of VHL is also responsible for the development of sporadic clear-cell RCCs, the predominant form of adult kidney cancer [10]–[12]. Cells that are devoid of functional VHL show elevated expression of numerous hypoxia-inducible genes due to a failure to degrade HIFα. In addition to VHL, deregulation of the PHD/EGLN family of prolyl-hydroxylases have also been associated with abnormal cell growth. Development of erythrocytosis, characterized by an excess of erythrocytes, has been associated with inactivating germline mutations in PHD2/EGLN1 [13],[14]. Pheochromocytoma, a neuroendocrine tumor of the medulla of the adrenal glands, is linked with deregulation of PHD3/EGLN3 [15].
While biallelic inactivation of VHL is found in the majority of clear cell RCCs, kidney cancer is a heterogeneous disease that can be divided into several subtypes based on morphological and cytogenetic features [16],[17]. Chromophobe RCC and renal oncocytoma are two related kidney tumors that together account for approximately 10% of all renal masses. In contrast to clear cell RCC, VHL mutations and/or increased expression of hypoxia-inducible genes are not found in these tumor subtypes and molecular genetic defects that are associated with tumor development remain unclear. Identification of molecular genetic defects in renal oncocytoma is particularly challenging as these cells are often described as karyotypically normal and the presence of cytogenetically abnormal regions in which to search for tumor modifying genes is rare in this tumor subtype.
To identify molecular defects associated with renal tumor development, we analyzed gene expression data from a variety of kidney tumors. This analysis revealed that renal oncocytoma and chromophobe RCC have a striking transcriptional disruption along chromosome 19. While in chromophobe RCC the disruption reflected a chromosome 19 amplification, in the renal oncocytoma cells the disruption reflected the close association, or pairing, of chromosome 19q in interphase. EGLN2 located within the paired region was dramatically overexpressed in renal oncocytoma cells and was associated with the deregulation of numerous hypoxia-inducible genes including a pro-death BNIP3L. Thus, chromosome 19q pairing in renal oncocytoma unveils a unique mechanism of disrupting oxygen homeostasis via altering the expression of EGLN2.
Gene expression profiling data derived from renal oncocytomas and chromophobe RCCs was scanned for regional increases or decreases in RNA production, which often indicate the presence of chromosomal amplifications or deletions [18]–[24]. Consistent with previous cytogenetic studies, the renal oncocytoma cells were largely devoid of transcriptional abnormalities that would reflect a DNA amplification or deletion. In contrast, losses of chromosomes 1, 2, 6, 10, and 17 are frequently found in chromophobe RCC. In our chromophobe RCC samples, these well-established chromosomal losses were strongly reflected in the gene expression profiling data (Figure 1A). In addition, a transcriptional abnormality involving genes mapping to chr 19 was frequently identified in both the renal oncocytomas and the chromophobe RCCs but not other subtypes of RCC (Figure 1A and Figure S1). In renal oncocytomas, the transcriptional abnormality primarily involved the q arm of chromosome 19, while in chromophobe RCC the abnormality involved the entire chromosome (Figure 1A,B).
Regional increases in overall RNA production often indicate the presence of an underlying DNA amplification. As gain of chromosome 19 has not been previously reported as a recurrent abnormality in either renal oncocytoma or chromophobe RCC, DNA copy number analysis was performed on a subset of these samples using high-density single nucleotide polymorphism (SNP) arrays. From the SNP data, an amplification of the entirety of chromosome 19 was detected in the chromophobe RCC samples (Figure 1C,D). This whole-chromosome amplification was confirmed by fluorescence in-situ hybridization (FISH) using locus-specific probes that mapped to the p and q arms of chromosome 19 (Table S1). In contrast, no change in DNA copy number was detected in the renal oncocytoma samples (Figure 1C,D). As a positive control for the DNA copy number analysis, only oncocytoma (ON) samples derived from female patients were examined, and a relative gain of the X chromosome was clearly detected in these samples (Figure 1C).
To determine the status of chromosome 19 in more detail in the renal oncocytoma cells, this chromosome was evaluated further using a panel of FISH probes. Two distinct and well-separated FISH signals, typical of diploid cells in interphase, were frequently observed when probes specific to the chr 19p arm were used (Figure 2 and Table S2). In contrast, a single, large FISH signal (singlet) or two FISH signals that were in close proximity (proximal doublet) were frequently observed when probes specific to the chr 19q arm were used. Approximately 35% of cells examined contained the singlet signal, while an additional 18% of cells contained proximal doublets (Table S2 and data not shown).
Semi-quantitative image analysis was used to examine the characteristics of the large FISH singlet (Figure 2B). This analysis demonstrated that the size of the singlet FISH signal was on average 1.5-fold larger than the size of two well-separated 19q FISH signals (P = 0.02). This large signal was observed using multiple probes directed against the q arm of the chromosome, including centromeric and telomeric probes (Figure 2C,E). The large FISH singlet had striking similarities to the FISH signals observed in studies of somatically paired chromosomes [25]–[27]. Somatic pairing refers to the close association of homologous chromosomes and is typically associated with chromosomes in meiotic prophase. However, somatic pairing has also been observed in interphase in normal human cells and some tumor cells [26], [28]–[32]. The presence of a large FISH singlet reflects the overlapping FISH signals generated from two chromosomal regions in very close proximity [26],[27]. The lack of evidence for a DNA copy number change coupled with the presence of large FISH singlets and proximal doublets using multiple locus-specific probes, suggested that chr 19q was somatically paired.
To confirm that the q arms of chr 19 were somatically paired in the renal oncocytoma cells, the p and q arms of chr 19 were visualized simultaneously using whole-arm chromosome painting (WCP). Using this approach, two distinct p arms, typical of diploid cells in interphase, were frequently observed in renal oncocytoma cells (Figure 2G,H and Table S2). However, the majority of cells contained a single q-arm signal that was located proximal to the two p-arm signals. While the diffuse nature of the WCP prevented the quantification the fluorescence signal, this pattern is consistent with the locus-specific FISH analysis and further indicates that the q arms of the chromosomes are in close proximity or are paired in these cells.
The changes in gene expression that accompanied the somatic pairing suggested that deregulation of a gene, or multiple genes, associated with tumor development mapped within the paired chr 19q region. As deregulation of the oxygen-sensing network is a common event in other types of sporadic renal cell carcinomas, genes associated with HIF regulation and that mapped to chr 19q were identified from the Entrez Gene database and tested for expression defects (see Materials and Methods). We also identified additional genes that were related to kidney-cancer via additional literature searching (Table S3). Both analyses identified EGLN2/PHD1 as a possible candidate gene in this region. To verify that the prolyl-hydroxylase EGLN2/PHD1 was significantly deregulated in renal oncocytoma cells, the level of EGLN2 protein was evaluated in these tumors (Figure 3A,B). Analysis of matched oncocytoma-normal tissue pairs revealed a dramatic increase in the level of EGLN2 in the oncocytoma tumors versus the level observed in corresponding normal tissue. Higher expression of EGLN2 was also observed in 2 of 3 chromophobe RCCs examined (Figure S2). These results are in contrast to the EGLN2 levels found in clear cell RCC. Consistent with the gene expression data, virtually no EGLN2 protein was detected in patient-derived clear cell RCC samples, while low basal amounts of EGLN2 were visualized by Western blot analysis in the matched normal samples (Figure 3 A,C).
EGLN2 is one of three prolyl-hydroxylases known to post-translationally modify HIFα, which is required for VHL-mediated destruction of HIFα. To address whether increased expression of EGLN2 influenced the binding and ubiquitination of HIF-1αODD via VHL, in vitro translated 35S-labeled HA-VHL and in vitro translated unlabeled Gal4-HA-HIF-1αODD were mixed in extracts in which EGLN2 was enriched (see Materials and Methods). Enrichment of EGLN2 led to an increased association of VHL to the wild-type ODD, but not to a mutated ODD in which a proline residue critical for VHL binding was changed to an alanine (P546A) (Figure 3D). In addition, an in vitro HIF-1αODD ubiquitination assay was performed to determine whether the increased VHL-HIF-1αODD association led to increased HIF-1αODD ubiquitination. Increased levels of EGLN2 resulted in a dose-dependent increase in VHL-mediated HIF-1αODD ubiquitination (Figure 3E). These results suggest that overexpression of EGLN2 in oncocytoma could further decrease the level of HIFα below the level observed in normal tissue.
In clear cell RCC, an increase in HIFα due to functional inactivation of VHL induces a transcriptional program that mimics cellular exposure to hypoxic conditions. In contrast, in the renal oncocytoma, the functional effects of increased expression of EGLN2 would be to decrease HIFα levels. To examine the cellular effects of decreased HIFα, we re-evaluated previously published data that measured HIF-1 DNA-binding activity, HIF-1α protein levels, and HIF-1β protein levels in cells exposed to hypo- and hyper-oxygenated conditions [6]. Normoxic conditions in the kidney cortex is estimated to be 3–5% oxygen [6]. Induction of a hypo-oxygenated condition was associated with a significant increase in HIFα and HIF activity levels (Figure 4A). Specifically, a six-fold decrease in oxygen concentration (3% to 0.5% oxygen) resulted in approximately a four-fold increase in HIF-1α levels (2.5 to 9.8 densitometry units). Further, we noted that HIF-1α levels change in an analogous manner upon induction of hyper-oxygenated conditions: a six-fold increase in oxygen concentration (3% to 18% oxygen) results in greater than a three-fold decrease in HIF-1α levels (2.5 to 0.75 densitometry units). The association between decreased HIF-1α and hyper-oxygenated conditions is easier to evaluate if the HIF dose-response data is plotted on a log-log scale rather than a linear-linear scale (Figure 4B). The log-log transformed data follow a straight line, indicating that HIFα level and oxygen concentration follow a power-law relationship (i.e., f(x) = axk), rather than an exponential relationship (i.e., f(x) = kax). The biological implications of the power-law relationship is that an n-fold change in oxygen concentration leads to a proportional n-fold change in HIF-1α levels and HIF activity (Figure S3). Moreover, these results demonstrate that while increases in HIF-1α are associated with hypo-oxygenated conditions, decreases in HIF-1α are associated with hyper-oxygenated conditions.
To determine whether EGLN2 overexpression is inducing a HIF-mediated hyperoxic cell response in the renal oncocytoma cells, the expression pattern of several known HIF target genes were examined in the renal oncocytoma cells and, for comparison, in clear cell RCC [33]. Consistent with VHL defects present in the clear cell RCC, gene set enrichment analysis revealed a significant up-regulation of the HIF-1 target genes in clear cell RCC (P = 0.0001; Figure 4C). Notable up-regulated genes included carbonic anhydrase IX (CA9), ferroxidase (CP), vascular endothelial growth factor A (VEGFA), and glucose transporter (GLUT1). However, in the renal oncocytoma cells, a distinct population of HIF-target genes were significantly down-regulated (P = 0.01; Figure 4D). Specifically, the HIF-target genes heme oxygenase 1 (HMOX1), enolase 1 (ENO1), and Cbp/p300-interacting transactivator (CITED2) were significantly down-regulated, but genes such as CA9, VEGFA, and GLUT1 were not. In addition, the recently identified tumor suppressor BNIP3L is downregulated three-fold in the renal oncocytoma cells (Figure 4E). BNIP3L is an oxygen-regulated member of the Bcl-2 family (Figure S4). BNIP3L is a pro-death gene (induces features of apoptosis, necrosis and autophagy) and knockdown of this gene is sufficient to convert non-tumorigenic cell lines into tumorigenic lines in xenograft studies [34]–[36]. In support, while hypoxia mimetic treatment significantly induced the expression of BNIP3L, HMOX1, ENO1, and CITED2 (Figure 5A, right panel and Figure S5), ectopic transient expression of EGLN2 under physiologic hypoxia (cyclical 0–7% oxygen exposure [37]) was associated with reduced level of expression of these genes in comparison to cells transfected with empty plasmid (Figure 5 and Figure S5). These results demonstrate that over expression of EGLN2 can downregulate HIF1 responsive factors, such as BNIP3L. Moreover, while up-regulation of HIF-target genes such as VEGFA are associated with the development of clear cell RCC, these results suggest that down-regulation of distinct subset of HIF-target genes are associated with the development of renal oncocytomas.
A proper oxygen-sensing response is vital to the maintenance of normal cellular functions. Deregulation of HIF, the principal driver of the adaptive response to hypoxia, is associated with the pathogenesis of several diseases, including cancer. While the hypoxic tumor microenvironment - by the virtue of the ubiquitous oxygen-sensing pathway - results in modulation of HIF activity, loss-of-function mutations in a growing list of tumor suppressor genes also can affect HIF function. Mutations in PTEN, PML, TSC, and VHL have been identified in tumor cells that result in the deregulation of HIF via multiple distinct mechanisms involving Akt/PI3K, mTOR and the ubiquitin pathway. Emerging evidence now implicates cancer-causing mutations that directly impinge on EGLNs. For example, mutations in succinate dehydrogenase (SDH) result in the cytosolic accumulation of succinate, which inhibits EGLNs, leading to the stabilization and activation of HIF-1α [38],[39]. Inactivating germline mutations in EGLN1 have been identified to cause erythrocytosis [13],[14] and deregulation of EGLN3 has been linked to the development of pheochromocytoma, a neuroendocrine tumor of the adrenal glands [15].
In this study, we reveal somatic pairing of chr 19q as a recurrent cytogenetic abnormality in renal oncocytoma that results in dramatic changes in transcription within the paired region. The functional consequence of chromosome joining is formally unknown but it is may disrupt chromatin structure causing the juxtaposition of cis and trans regulatory regions that modulate the transcription of a large set of genes. The identification of EGLN2 as a significantly deregulated gene that maps within the paired chr 19q region directly implicates defects in the oxygen-sensing network to the pathobiology of renal oncocytoma. These results suggest that in addition to numerical and structural chromosomal abnormalities, somatic pairing should be considered as a chromosomal event that associates with tumorigenesis.
Although the loss of EGLN2 does not lead to decreased HIF1α accumulation, perhaps due to the compensatory activity of EGLN3, the data from this study suggest that overexpression of ELGN2 leads to decreased HIF1 levels. More recently, an E3 ubiquitin ligase called Siah2 was identified to target EGLN2 for ubiquitin-mediated destruction and thereby revealing another level of HIF regulation [40]. The activity of Siah2 is induced under physiologic hypoxia (<10% oxygen), resulting in reduced levels of EGLN2 and stabilization of HIF-1α. The present findings suggest that the overexpression of EGLN2 via somatic pairing is sufficient to counteract the suppressive activity of Siah2 under physiologic hypoxia. Under hyper-oxygenated conditions (21% oxygen; frequently used as experimental normoxia), Siah2 activity is attenuated via a yet-defined mechanism, resulting in the increased abundance of EGLN2 and concomitant reduction in the level of HIF-1α [40]. The ectopic expression of EGLN2 under 21% oxygen did not result in further diminution of HIF-target gene expression (data not shown), which is likely due to the fact that endogenous EGLN2 is highly abundant or that every available EGLN2 is already activated under hyper-oxygenated conditions.
HIF-regulated genes are involved in many physiological processes including angiogenesis, metabolism, cell proliferation, survival, and apoptosis. As such, disruption in the regulation of HIF may affect several regulatory pathways that contribute to the transformation of normal cells into cancer cells. Evasion of apoptosis is one of the hallmark features of cancer cells and represents a key oncogenic event. BNIP3L is a regulator of p53-dependent apoptosis and silencing of BNIP3L has been associated with enhanced tumorgenicity and reduced apoptotic response [36]. We show here that BNIP3L is one of several HIF-responsive genes governed, in part, by EGLN2. Therefore, we propose that the downregulation of BNIP3L is the result of chromosome-pairing induced upregulation of EGLN2 and that downregulation of BNIP3L contributes to the inhibition of apoptosis to facilitate oncocytoma cell survival and growth.
The disruption of HIF activity has been associated with kidney cancer related to VHL disease, sporadic clear cell RCC, and hereditary papillary RCC [38],[41],[42]. The present study reveals deregulation of the oxygen-sensing response in renal oncocytoma, as well as chromophobe RCCs (which display DNA amplification mediated up-regulation of EGLN2) and thereby supporting the dysfunction of HIF pathway as a common and perhaps central theme in the pathogenesis of kidney cancer.
Single-color expression profiles were generated using the HG-U133 Plus 2.0™ chipset (Affymetrix, Santa Clara, CA) from renal oncocytoma (n = 10), chromophobe RCC (n = 10), and nondiseased kidney (n = 12) samples as described [43]. The gene expression data can be obtained at the Gene Expression Omnibus (GSE8271 and GSE7023). Analysis was performed using BioConductor version 2.0 software. Data preprocessing was performed using the RMA method as implemented in the affy package and using updated probe set mappings such that a single probe set describes each gene [44],[45],[46]. Chromosomal abnormalities were predicted using the comparative genomic microarray analysis (CGMA) method as implemented in the reb package [47]. Briefly, for each measured gene, the gene expression value was normalized such that the average gene expression value in the nondiseased samples was subtracted from the tumor-derived gene expression value. A Welsh's t-test was applied to the relative gene expression values that mapped to each chromosome arm. For the smoothing curve, the normalized expression values derived from genes mapping to chromosome 19 were replaced by a summary score that comprised a running two-sided t-test statistic using window sizes of 61, 245, and 611 (representing 5%, 20%, and 50% of the length of the chromosome). The results of the three smoothing curves were averaged. To identify HIF-interacting genes, the Entrez Gene database (http://www.ncbi.nlm.nih.gov/sites/entrez) was searched using the search string ‘(“HIF” or “VHL”) and “19”[chr] and “homo sapiens”[orgn]’. Differentially expressed genes were identified using a two-sided t-test. For HIF target gene analysis, 36 known HIF-responsive genes identified in Maynard et al. were isolated [33]. Enrichment of up- and down-regulated genes in the HIF target gene set was performed by comparing differences in the expression level ranks between HIF target gene set to the results of 10,000 randomly generated 36-gene sets. Ranks were based on tumor versus normal expression comparisons as implemented in the limma package [48].
SNP allele calls were generated using the GeneChip Mapping 100 K Set™ (Affymetrix, Santa Clara, CA) according to the manufacturer's supplied protocol. Image quantification was performed with a GeneChip Scanner 3000 and the resulting data was processed using GCOS 1.4 (Affymetrix, Santa Clara, CA) with default analysis settings. Allele calls were generated using GTYPE 4.0 (Affymetrix, Santa Clara, CA) with a confidence threshold set at 0.25. Raw copy numbers in log2-transformed format (non-paired reference and test samples) were exported from the CNAG version 2.0 (Affymetrix, Santa Clara, CA) software using normal references downloaded from Affymetrix (http://www.affymetrix.comccnt_reference_data). DNA copy number changes were visualized by data smoothing in which raw copy number values were replaced by a summary score that comprised a running 1-sided t-test statistic with window size set to 31, where each SNP probe along with 15 5′ SNPs and 15 3′ SNPs were included in the window. DNA copy number data can be obtained at the Gene Expression Omnibus (GSE8271).
Bacterial artificial chromosomes (BACs) RP11-157B13 (19p12), RP11-1137G4 (19p13.3), RP11-15A1 (19q13.31) were obtained from the Children's Hospital Oakland Research Institute (http://bacpac.chori.org) and BAC CTC-429C10 (19q13.41) was purchased from Invitrogen (Invitrogen Corporation, Carlsbad, CA). These clones were labeled with either SpectrumGreen or SpectrumOrange (Abbott Molecular Inc, Des Plaines, IL) by nick translation and applied to tissue touch preps of oncocytoma samples as described [49], with the exception that slides were counterstained with VECTASHIELD (Vector Laboratories, Inc. Burlingame, CA) anti-fade 4′,6-diamidino-2-phenylindole (DAPI). Telomere-specific DNA probes, the chr 1,5,19 alpha satellite probe, and the arm-specific paints were purchased from Q-BIOgene (MP Biomedicals, Solon, OH). FISH was performed using these probes according to the manufacturer's supplied protocol. As the alpha satellite probe cross-hybridizes to chromosome 1 and chromosome 5, in all studies chromosome 19 was co-labeled with a probe that maps distal to the centromere, RP11-157B13 (19p12). In addition, analysis of the centromeric probe on the metaphase spreads of control cells revealed that hybridization to chromosome 1 resulted in a significantly brighter signal (data not shown). These hybridization characteristics allowed the discrimination between chr 1 and 5 cross-hybridization.
For image quantification, three separate photomicrographs containing five, six, and three cells, respectively, in which the 19q31.31 FISH signals were in the same image plane were obtained. Photomicrographs were processed using the rtiff package for the R environment [50]. The fluorescent FISH signals were automatically segmented from background using the method of Ridler and Calvard [51], individual spots were identified using the connected component algorithm [52], and the number of pixels per feature were calculated. Twelve doublet FISH signals and eight singlet FISH signals were compared. Differences in size were evaluated using a one-sided Student's t-test.
U2OS osteosarcoma cell and CAKI renal clear-cell carcinoma cell lines were obtained from the American Type Culture Collection (Rockville, MD) and maintained in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum (Sigma, Milwaukee, WI) at 37°C in a humidified 5% CO2 atmosphere. Cyclic hypoxia treatment of cells were performed in humidified chambers at 37°C and flushed with 5% CO2 balance N2 for 30 min, followed by 5% CO2 and 7% O2 balance N2 for 30 min as one cycle. Cells were grown in these chambers for 16 hours [53].
Polyclonal anti-EGLN2 and anti-BNIP3L antibodies were obtained from Bethyl Laboratories (Montgomery, TX) and Sigma (Milwaukee, WI), respectively. Polyclonal HIF1α and monoclonal HIF2α antibodies were obtained from BD Biosciences (San Jose, CA) and Novus (Littleton, CO), respectively. Monoclonal anti-vinculin antibody was obtained from Abcam (Cambridge, MA).
Mammalian expression plasmids pcEglN2 was generated by PCR from Flag-EglN2, a kind gift from Dr. Mircea Ivan, using primers 5′-GACGACGGATCCATGGACAGCCCGTGCCAGC-3′ and 5′-GACGACGAATTCCTAGGTGGGCGTAGGCGGC -3′. The PCR product was then ligated into the BamHI and EcoRI sites in pcDNA3(+). Plasmid was confirmed by direct DNA sequencing.
Western blotting were performed as described previously [54].
For first-strand cDNA synthesis, 1 µl of oligo(dT)23 primer (Sigma) was incubated with 5 µg of RNA and distilled H2O (total reaction volume of 20 µl) for 10 min at 70°C in a thermal cycler (MJ Research, Boston, MA). The mixture was cooled to 4°C, at which time 4 µl of 5× first-strand reaction buffer, 2 µl of 0.1 M DTT, 1 µl of a 10 mM concentration of each deoxynucleoside triphosphate, and 1 µl of Superscript II reverse transcriptase (Invitrogen) were added. cDNA synthesis was performed for 1.5 h at 42°C, followed by 15 min at 70°C in the thermal cycler. Human genomic DNA standards (human genomic DNA was obtained from Roche, Mannheim, Germany) or cDNA equivalent to 20 ng of total RNA were added to the quantitative PCR (qPCR) reaction mixture in a final volume of 10 µl containing 1× PCR buffer (without MgCl2), 3 mM MgCl2, 0.25 units of Platinum Taq DNA polymerase, a 0.2 mM concentration of each deoxynucleoside triphosphate, 0.3 µl of SYBR Green I, 0.2 µl of ROX reference dye, and a 0.5 µM concentration of each primer (Invitrogen). Amplification conditions were as follows: 95°C (3 min), 40 cycles of 95°C (10 s), 65°C (15 s), 72°C (20 s), and 95°C (15 s). qPCR was performed using the ABI Prism 7900HT Sequence Detection System (Applied Biosystems, Foster City, CA). Gene-specific oligonucleotide primers designed using Primer Express (Applied Biosystems) were as follows: BNIP3L primer set (5′- CTGCACAAACTTGCACATTG-3′ and 5′- TAATTTCCACAACGGGTTCA-3′), HMOX1 primer set (5′-GAATTCTCTTGGCTGGCTTC-3′ and 5′- TCCTTCCTCCTTTCCAGAGA-3′), ENO1 primer set (5′- CAGCTCTAGCTTTGCAGTCG-3′ and 5′-GACACGAGGCTCACATGACT-3′), CITED2 primer set (5′-ACTGCACAAACTGCCATCTC-3′ and 5′-CAGCCAACTTGAAAGTGAACA-3′), beta-actin primer set (5′- GGATCGGCGGCTCCAT-3′ and 5′- CATACTCCTGCTTGCTGATCCA-3′), GLUT-1 primer set (5′- CACCACCTCACTCCTGTTACTT-3′ and 5′-CAAGCATTTCAAAACCATGTTTCTA-3′). SYBR Green I fluoresces during each cycle of the qPCR by an amount proportional to the quantity of amplified cDNA (the amplicon) present at that time. The point at which the fluorescent signal is statistically significant above background is defined as the cycle threshold (CT). Expression levels of the various transcripts were determined by taking the average CT value for each cDNA sample performed in triplicate and measured against a standard plot of CT values from amplification of serially diluted human genomic DNA standards. Since the CT value is inversely proportional to the log of the initial copy number, the copy number of an experimental mRNA can be obtained from linear regression of the standard curve. A measure of the relative difference in copy number was determined for each mRNA. Values were normalized to expression of beta-actin mRNA and represented as the mean value experiments performed in triplicate±standard deviations.
Extracts containing enriched EGLN2 were purified from rabbit reticulocyte lysate as previously described [8]. Briefly, approximately 1 L of rabbit reticulocyte lysate (Green Hectares, Oregon, WI) was diluted to 5 L in 50 mM Tris-HCl (pH 7.4), 0.1 M KCl, and 5% (vol/vol) glycerol and then was precipitated with 0.213 g/ml (NH4)2SO4. After centrifugation at 16,000×g for 45 min at 4°C, the resulting supernatant was precipitated with an additional 0.153 g/ml (NH4)2SO4. After centrifugation at 16,000×g for 45 min at 4°C, the pellet was resuspended in Buffer A (40 mM HEPES-NaOH [pH 7.4] and 5% (vol/vol) glycerol), dialyzed against Buffer A to a conductivity equivalent to Buffer A containing 0.2 M KCl, and applied at 0.5 L/h to a 0.5 L phosphocellulose (Whatman, P11) column equilibrated in Buffer A containing 0.2 M KCl. The phosphocellulose column was eluted stepwise at 1 L/h with Buffer A containing 0.5 M KCl, and 100-ml fractions were collected. Proteins eluting in the phosphocellulose 0.5 KCl step were pooled and precipitated with 0.4 g/ml (NH4)2SO4. After centrifugation at 16,000×g for 45 min at 4°C, the pellet was resuspended in 4 ml of Buffer A. Following centrifugation at 35,000×g for 30 min at 4°C, the resulting supernatant was applied at 2 ml/min to a TSK SW3000 HPLC column (Toso-Haas, Montgomeryville, PA; 21.5×600 mm) equilibrated in Buffer A containing 0.15 M KCl. The SW3000 column was eluted at 2 ml/min, and 4 ml fractions containing enriched EGLN2 were collected.
An in vitro binding assay was performed as described previously [3]. TNT reticulocyte lysate (Promega) translation products were synthesized in the presence or absence of 35S-methionine. HIF1α-(ODD) translation products were incubated with cellular extract fractions containing enriched EGLN2, where indicated, for 30 min at 37°C. Gal4-HA-HIF-1α (10 µl) and HA-VHL (10 µl) translation products were incubated with the indicated antibodies and protein A-Sepharose in 750 µl of EBC buffer (50 mM Tris [pH 8], 120 mM NaCl, 0.5% Nonidet P-40). After five washes with NETN buffer (20 mM Tris (pH 8), 100 mM NaCl, 0.5% Nonidet P-40, 1 mM EDTA), the bound proteins were resolved on SDS-PAGE and detected by autoradiography.
An in vitro ubiquitylation assay was performed as described previously [3]. [35S]Methionine-labeled reticulocyte lysate Gal4-HA-HIF1α(ODD) (4 µl) were incubated in RCC 786-O S100 extracts (100–150 µg). Reactions were supplemented with an increasing titration of EGLN2-enriched cellular fraction where indicated. Additional reaction supplements include 8 µg/µl ubiquitin (Sigma), 100 ng/µl ubiquitin-aldehyde (BostonBiochem, Inc., Cambridge, MA), and an ATP-regenerating system (20 mM Tris [pH 7.4], 2 mM ATP, 5 mM MgCl2, 40 mM creatine phosphate, 0.5 µg/µl of creatine kinase) in a reaction volume of 20–30 µl for 1.5 h at 30°C.
Figure 5B from the Jiang et al. article [6] was obtained in Portable Document Format (PDF, Adobe Systems), imported into Canvas 9 (ACD Systems), and the x- and y-graphic device coordinates of each data point, the x-axis ticks (oxygen concentration), and the y-axis ticks (densitometry) were extracted. Linear interpolation was used to convert the graphic device coordinates to protein densitometry measurements and oxygen concentrations. Based on comparisons between the extracted oxygen concentrations (0.5, 1.9, 2.9, 3.9, 4.8, 5.8, 7.9, 9.9, 11.9, 13.9, 19.9) and the actual oxygen concentrations (0.5, 2, 3, 4, 5, 6, 8, 10, 12, 14, 20), the extracted data varied on average less than 2% from the original data. The densitometry and oxygen concentration data were log2-transformed and linear model fit to the transformed data. The best-fit power-law equation is HIF1α = 22.61O−0.85, where HIF1α represents HIF-1α protein levels and O represent oxygen concentration.
|
10.1371/journal.pgen.1004536 | Translational Regulation of the DOUBLETIME/CKIδ/ε Kinase by LARK Contributes to Circadian Period Modulation | The Drosophila homolog of Casein Kinase I δ/ε, DOUBLETIME (DBT), is required for Wnt, Hedgehog, Fat and Hippo signaling as well as circadian clock function. Extensive studies have established a critical role of DBT in circadian period determination. However, how DBT expression is regulated remains largely unexplored. In this study, we show that translation of dbt transcripts are directly regulated by a rhythmic RNA-binding protein (RBP) called LARK (known as RBM4 in mammals). LARK promotes translation of specific alternative dbt transcripts in clock cells, in particular the dbt-RC transcript. Translation of dbt-RC exhibits circadian changes under free-running conditions, indicative of clock regulation. Translation of a newly identified transcript, dbt-RE, is induced by light in a LARK-dependent manner and oscillates under light/dark conditions. Altered LARK abundance affects circadian period length, and this phenotype can be modified by different dbt alleles. Increased LARK delays nuclear degradation of the PERIOD (PER) clock protein at the beginning of subjective day, consistent with the known role of DBT in PER dynamics. Taken together, these data support the idea that LARK influences circadian period and perhaps responses of the clock to light via the regulated translation of DBT. Our study is the first to investigate translational control of the DBT kinase, revealing its regulation by LARK and a novel role of this RBP in Drosophila circadian period modulation.
| The CKI family of serine/threonine kinase regulates diverse cellular processes, through binding to and phosphorylation of a variety of protein substrates. In mammals, mutations in two members of the family, CKIε and CKIδ were found to affect circadian period length, causing phenotypes such as altered circadian period in rodents and the Familial Advanced Sleep Phase Syndrome (FASPS) in human. The Drosophila CKI δ/ε homolog DOUBLETIME (DBT) is known to have important roles in development and circadian clock function. Despite extensive studies of DBT function, little is known about how its expression is regulated. In a previous genome-wide study, we identified dbt mRNAs as potential targets of the LARK RBP. Here we describe a detailed study of the regulation of DBT expression by LARK. We found that LARK binds to and regulates translation of dbt mRNA, promoting expression of a smaller isoform; we suggest this regulatory mechanism contributes to circadian period determination. In addition, we have identified a dbt mRNA that exhibits light-induced changes in translational status, in a LARK-dependent manner. Our study is the first to analyze the translational regulation of DBT, setting the stage for similar studies in other contexts and model systems.
| The Drosophila doubletime (dbt, a.k.a. discs overgrown, dco) gene encodes a protein homologous to human casein kinase I isoforms (CKI) [1], [2], in particular CKIδ and CKIε [3]. It is known that the DOUBLETIME (DBT/CKIδ/ε, hereafter referred to as “DBT”) kinase regulates cell proliferation, differentiation and cell polarity by functioning in Wnt [4], [5], Hedgehog [6]–[9], Fat [10]–[13] and Hippo signaling [14], [15] pathways. Those studies demonstrated roles of DBT in growth, development, organ size determination, and tumor suppression. The kinase is also well known for its role in the core molecular mechanism of the circadian clock ([1], [2], reviewed in [16]–[18]).
The molecular oscillator regulating locomotor activity rhythms is comprised of a transcription-translation feedback loop wherein accumulation of clock proteins regulates clock gene transcription and protein production. Transcriptional mechanisms are common to the circadian clocks of organisms ranging from cyanobacteria and fungi to plants and animals [18]–[22], although recent studies have indicated that conserved non-transcriptional clocks mediate certain types of circadian rhythms [23]. Casein kinase I (CKI) is required for period determination in vertebrates as well as insects. For example, in hamster and mouse, a gain-of-function mutation of CKIε (CKIεtau), causes shortening of circadian period [24], [25] whereas inhibition of CKIδ kinase activity in zebrafish disrupts circadian rhythmicity in locomotor activity [26]. In humans, a mutation in the key clock protein PERIOD 2 perturbs its phosphorylation by CKIε and is associated with Familial Advanced Sleep Phase Syndrome (FASPS), as a result of an abnormally short circadian period [27]–[31]. Interestingly, mutations in CKIδ were also found to cause FASPS in humans [32].
In Drosophila, the role of DBT in circadian period determination has been studied extensively. DBT was first shown to regulate PER accumulation [2], introducing a cytoplasmic lag into the circadian molecular loop. It was later established that DBT promotes progressive phosphorylation of PER, which facilitates interaction between PER and Slimb, an F-box/WD40-repeat protein that helps target PER for degradation in the proteasome [33]–[35]. Many DBT phosphorylation sites in the PER protein have been mapped [36]–[38]. Phosphorylation of residues in the so called “short-period domain” by DBT, gated by phosphorylation of a key residue by another kinase called NEMO/NLK, affects progression of the molecular cycle [39]. Phosphorylation of an N-terminal serine residue (S47) by DBT was identified as a key step in controlling the speed of the clock [40]. DBT is also required for phosphorylation of CLOCK (CLK), another key component of the Drosophila molecular clock [41], [42], although it was later found that DBT does not phosphorylate CLK directly but rather plays a non-catalytic role in CLK phosphorylation [43].
Despite extensive studies of DBT function, the mechanisms regulating expression of this protein are largely unknown. In a previous genome-wide study we identified dbt mRNA as a potential target of the LARK RBP, which has been implicated in translational control and clock function [44]–[52]. This suggested the possibility that dbt might be translationally regulated by LARK. Here we describe a detailed study of DBT regulation by LARK. We demonstrate that LARK can bind to and enhance translation of different transcript isoforms of dbt in clock cells of the adult fly head. The effect is most prominent with dbt transcripts RC and RE. Translation of dbt-RC undergoes circadian changes in free-running conditions, whereas translation of dbt-RE is light inducible. Consistent with the known role of DBT in circadian period determination, altered LARK expression in the PDF neurons affects period length, and this effect can be modified by dbt mutations. The role of LARK in modulating circadian period through DBT is further supported by the observation that increased LARK expression delays nuclear degradation of the PERIOD clock protein. Our study is the first to examine translational regulation of the DBT kinase and it supports a role of LARK in the modulation of circadian period.
In a previous genome-wide study, we showed that dbt mRNA, but not other clock mRNAs, was associated with LARK in vivo [44]. The dbt gene produces multiple alternatively spliced transcripts. Earlier versions of genome annotation provided by FlyBase (up to Release 5.30) show three splice variants – dbt-RA, dbt-RB, and dbt-RC – that share protein-coding and 3′UTR sequence but differ at the 5′UTR (Figure S1A). However, the most recent annotation (release R5.49) included a fourth transcript, dbt-RD, that appears to be identical to dbt-RB but with a longer 3′UTR (Figure S1B). This difference is presumably based on recent genome-wide RNA sequencing data that includes sequence reads mapping to regions that extend beyond the previously annotated 3′UTR. However, we do not believe there is sufficient evidence to distinguish transcript D from transcript B; i.e., there may be only one transcript with a long 3′UTR. Thus we did not treat dbt-RD as an independent transcript, but instead focused our studies on the dbt RA, RB and RC transcripts. In addition, we found EST evidence suggesting the existence of an unannotated transcript with a unique 5′UTR, likely resulting from an alternative transcription start site. Two ESTs (GenBank gi 49381530 and gi 103690325) align perfectly to the 5′ region of the gene in a manner distinct from all previously annotated transcripts. We named this previously unannotated transcript dbt-RE. Studies described below demonstrate the expression of this novel transcript.
To determine if dbt transcripts are associated with LARK, in vivo, we quantified RNAs that co-immunoprecipitated specifically with LARK from head tissue lysates of adult flies. Quantitative Real-Time PCR (Q-RTPCR) using primers specific to each isoform demonstrated that dbt transcripts were enriched after anti-LARK immunoprecipitation (IP). Enrichment values, relative to transcript abundance after IP with an unrelated antibody (anti-EGFP) were 7.7, 4.5, 6.2 and 10.2 fold, respectively, for dbt-RA, RB, RC, and RE (Figure 1A). These results demonstrate an association between LARK and all dbt alternative transcripts in vivo.
These IP results do not distinguish between direct binding by LARK versus indirect association because of the presence of the RNA binding protein (RBP) and dbt mRNAs in the same complex. To test whether LARK can directly bind dbt mRNAs, we conducted UV cross-linking assays [53] using radio-labeled dbt transcripts produced by in vitro transcription (see Material and Methods) and a purified recombinant LARK protein containing both RNA Recognition Motifs (RRMs) [48]. This analysis showed that LARK binds to dbt mRNAs in a concentration-dependent manner and at concentrations as low as 100 nM (Figure 1, B and C). In contrast, LARK binding to an unrelated mRNA (GlutR2) was barely discernible at a concentration of 1 µM protein, indicative of specificity (Figure 1B). Thus, LARK can directly bind dbt mRNAs.
To test the hypothesis that LARK regulates translation of the DBT protein, we examined the effect of altered LARK expression on DBT abundance. To our surprise, pan-neuronal overexpression of LARK (in elav-gal4; uas-lark/+ flies) revealed a novel immunoreactive DBT band that was of lower molecular weight than the previously described protein (Figure 2A). To our knowledge, such a DBT immunoreactive protein has not previously been reported. In our experiments, however, the novel DBT band was consistently observed in all LARK overexpression (OE) samples but never in control (OC) samples. Furthermore, the band was detected at three different zeitgeber times (ZTs): ZT2, ZT7 and ZT14. We note that higher molecular weight bands are also detected by the DBT antibody (Figure S6) with LARK or DBT OE (seen with DBT OE on a longer exposure). As these bands are too big to represent single proteins encoded by dbt mRNAs and only seen with LARK or DBT OE, we think they must represent aggregates of DBT (see Discussion).
It is possible that the novel smaller DBT band represents an isoform that, in the absence of increased LARK expression, is normally present at a low undetectable level. To test this idea, we examined head tissue lysates of elav-gal4; uas-dbt/+ flies, which overexpress DBT in all neurons. We found that the novel protein was revealed by DBT overexpression (Figure 2B), indicating that it may represent a rare isoform of the protein. Interestingly, this novel isoform exhibits a diurnal oscillation: in LARK OE flies, it is more abundant at ZT2 than at ZT14 (Figure 2A). Similarly, in DBT overexpressing flies, it can be detected at ZT2 but not ZT14 (Figure 2B). In contrast to LARK OE, LARK knockdown (KD) does not produce a detectable effect on DBT protein level when assayed by Western analysis (Figure 2A). We attempted to show that the novel DBT band corresponded to a previously uncharacterized isoform of the kinase by examining null dbt mutants that survive to larval and early pupal stages (adult null mutants do not survive). However, LARK overexpression at these stages did not induce the novel band (Figure S7). Thus, it may represent an adult-specific form of DBT.
To directly assess the effect of altered LARK expression on translation of DBT, we used the Translating Ribosome Affinity Purification (TRAP) technique to isolate Ribosome bound RNAs from LARK OE, KD and the respective control flies (OC and KC). The TRAP technique was originally developed in mouse [54]. We and others have adapted the technique for use in Drosophila by constructing transgenic flies carrying a uas-EGFP-L10a construct that expresses EGFP-tagged ribosomes in target tissues when crossed to a GAL4 line; this permits isolation of translating mRNAs from target tissues [55], [56]. As LARK is known to have a pan-neuronal expression pattern in the adult head [4], we first generated flies with altered LARK expression in all neurons using elav-gal4 in combination with uas-larkRNAi (for KD) or uas-lark (for OE). As indicated previously, knockdown or overexpression of wild-type LARK using these UAS constructs is associated with altered circadian behavioral rhythmicity [49], [51]. We included the uas-EGFP-L10a transgene in the OE or KD flies to allow isolation of translating mRNAs from all neurons. We found that LARK OE or KD did not significantly affect translation of dbt-RA, RB or RE. However, translation of dbt-RC was significantly increased in these experiments (Figure 2C, left) by LARK OE. Based on the knowledge that LARK and DBT both have circadian functions, we next examined the effect of altered LARK level on the translation of dbt transcripts in clock cells. In these experiments, we expressed uas-lark and uas-EGFP-L10a in clock cells using the tim-uas-gal4 driver [57]. In contrast to pan-neuronal LARK OE, overexpression specifically in clock cells promoted translation of all dbt transcripts, with the effect on dbt-RC being the most dramatic (8 fold increased; Figure 2C, right). LARK KD caused a small but statistically significant decrease in the translation of all transcripts. To test whether the translational changes result from altered abundance of dbt transcripts or changes in translational status, per se, we examined dbt transcript levels in total RNA extracted from control and LARK OE flies. We found that overexpression of LARK in all clock cells of the fly head did not significantly affect the abundance of RA, RB or RE in total RNA samples. However, there was an approximate 2.6 fold increase in RC abundance (Figure S2). Such an increase in abundance cannot account for the observed 8.3 fold increase in translation of RC (Figure 2C, right). Thus, it is likely that LARK OE results in changes in dbt-RC translational status.
Taken together, the results of these experiments demonstrate that LARK promotes translation of DBT, in particular a previously unidentified DBT isoform. The observation that LARK expression in clock cells had more dramatic effects on dbt than pan-neuronal expression of the protein suggests that regulation of dbt translation by LARK may occur predominantly in clock neurons. An alternative but less likely explanation is that tim-uas-gal4 drives higher expression of LARK than elav-gal4. However, we observed a similar level of expression for the two drivers when they were used with a uas-GFP reporter transgene.
In wild-type flies, LARK shows a circadian oscillation in abundance; the level of LARK is high during the day and low at night [47]. If LARK promotes translation of DBT, then the translational profile of DBT might also display a circadian rhythm. To test this hypothesis, we sampled the translational profiles of the four different dbt transcripts at 4-hour intervals under entrained conditions (LD 12∶12) and during the first 2 days of free-running conditions (DD). We emphasize that the endogenous LARK level was not manipulated in these experiments. We found that translation of dbt-RA displayed a low-amplitude rhythm in LD (peak to trough change is ∼2 fold), whereas dbt-RB and dbt-RC did not display rhythmic changes in translation. In contrast, dbt-RE displayed robust diurnal changes, with an 8-fold difference between trough-to-peak levels in LD (Figure 3, left panel; p = 0.036). The rhythms of RA and RE were greatly damped or eliminated when flies were released into free-running conditions (DD1 and 2). Interestingly, translation of dbt-RC appeared to begin cycling in DD, with a trough-to-peak change of about ∼2–3 fold (Figure 3, right panel; DD1, p = 0.036, DD2, p = 0.0003). dbt-RB translation did not exhibit significant rhythmic changes in LD or DD (Figure 3). Previous studies of total RNA extracted from whole adult head did not find significant circadian cycling of the dbt messages [1], [58], although Abruzzi et al. reported a low-amplitude cycling of RC in LD that did not reach their cutoff (1.4 fold change) for statistical significance. In agreement with those studies, we did not find significant cycling of the dbt-RC transcript in DD1 or dbt-RE in LD when abundance of these transcripts was examined in total RNA extracted from the same head lysate used in the TRAP assay (Figure S3). We conclude that RE and RC exhibit translational cycling in LD and DD, respectively.
The observations that translation of dbt-RE displays a robust cycle under LD but not DD, and that peak translation occurs shortly after lights-on suggest that its translation might be induced by light. To test this hypothesis, we entrained tim-uas-gal4; uas-EGFP-L10a flies for 4 days under LD 12∶12 conditions and then released them into constant darkness (DD) on the fifth day. During the first day of DD, the flies were divided into two groups; at CT12 (i.e. the beginning of subjective night) one group received light stimulation while the other was maintained in darkness. We then performed TRAP analysis using head tissues from the two groups of flies and examined translation of dbt-RE at 0.5, 1, 2, 3, 4 and 5 hours after CT12. As shown in Figure 4, translation of dbt-RE steadily increased, peaking at 4 hours following light exposure. In contrast, translation of dbt-RE remained relatively unchanged in the control group not exposed to light (Figure 4A). Statistical significance of the result was verified by a two-way ANOVA, which revealed light exposure as a factor influencing changes in translational level (p = 2.91×10−5). Together with the observation that dbt-RE abundance does not cycle in total RNA, this experiment strongly suggests that translation of dbt-RE is induced within clock cells of the adult head by light exposure.
We next examined whether the light-induced translation of dbt-RE is affected by altering LARK expression. We asked this question by comparing differences in ribosome-bound dbt-RE levels between flies receiving light stimulation at CT12 (the beginning of subjective night) and those maintained in constant darkness. Ribosome-bound RE transcript was examined in LARK knockdown, LARK OE and control flies at CT12 and CT 16, with or without light stimulation. Relative to controls and LARK OE, LARK knockdown flies had significantly decreased light-induced RE translation (Figure 4B). These results support a role for LARK in the light-induced regulation of dbt-RE.
The DBT kinase regulates PER phosphorylation and period of the circadian clock. Mutations that affect DBT level or its kinase function are known to alter period length of locomotor activity rhythms [1], [2], [59]. Given the observed effects of LARK expression on dbt, we tested whether alterations of LARK affect circadian period. We employed fly strains carrying a uas-larkRNAi transgene [51] for selective knockdown of LARK in specific subsets of neurons. This transgene was expressed throughout development, because we have not been successful in producing an adult-specific knockdown of LARK [50]. In order to achieve a more effective knockdown, the RNAi transgene was expressed in a background heterozygous for lark1, a null mutation of the gene [45]. As shown in Figure 5 (A and B) and Table S1, knockdown of LARK in the PDF neurons – important circadian pacemaker cells – caused an approximate 0.85 h shortening of circadian period. This effect is caused by specific knockdown by LARK, because the introduction of a uas-lark transgene into the LARK KD background reverted the period shortening (Figure S4, Table S1). Further, the effect is likely to be mediated by DBT because the period shortening was also corrected by introducing a uas-dbt transgene (Figure S4, Table S1). Predictably, conditional, adult-specific overexpression of LARK had the opposite effect, causing a 1.5 h lengthening of period (Figure 5, A, C, Table S1). It is of interest that LARK overexpression in this experiment caused period lengthening, because a previous study showed that conditional, high-level LARK overexpression, achieved using two copies each of pdf-gal4 and uas-lark (Figure 5E, panel d), caused arrhythmic behavior [50]. We note that the present study utilized a “milder” level of LARK overexpression, achieved using only one copy each of pdf-gal4 and uas-lark, revealing an effect on period. In addition, overexpression of LARK in this study was conditional and restricted to the adult stage, in contrast to a previous study which showed that mild overexpression of LARK throughout development caused increased arrythmicity rather than a lengthened period [49]. In the current study, the different levels of LARK OE and the effectiveness of LARK KD were validated by immunohistochemistry using anti-LARK antibody (Figure 5E). In contrast to wild-type LARK OE, a mutant LARK protein lacking function RRM domains [48], did not cause lengthening of period when overexpressed by pdf-Gal4 (Figure 5, A, D, Table S1). We note that a previous study demonstrated that the UAS-wild-type and UAS-mutant lark transgenes are expressed at similar levels when driven by the same Gal4 driver [48]. These results indicate that the RNA-binding activity of LARK is required for the observed effects on behavior.
To confirm an effect on circadian period in LARK OE and KD flies, we looked at the cycling of PERIOD protein in the PDF neurons in conditions of constant darkness (DD). Abundance and localization of the PERIOD protein were examined every 4 hours for a 24-hour period by immunohistochemistry and confocal imaging. Because the period altering effects are small, especially in the case of LARK KD, we allowed the effect to accumulate for 4 days in DD. On day 4, the phase of the oscillator should have advanced by almost 4 hours in LARK KD flies, allowing the difference to become detectable when sampling every 4 hours. Indeed, we found that the phase of PER cycling is advanced in KD flies and delayed in OE flies (Figure S5), consistent with results of the behavioral analyses.
To further test the possibility that LARK influences period length by modulating expression of DBT, we investigated genetic interactions between altered LARK expression and chromosomal dbt mutations including dbtL, dbtS, dbtP and dbtAR. We found that overexpression of LARK lengthened period in all the dbt mutant backgrounds tested. Interestingly, the period lengthening effect of LARK OE varied in different mutant backgrounds. The effect was more dramatic in mutants with short period than in mutants with long period. For example, overexpression of LARK caused a lengthening of ∼2.5 hour and ∼2.6 h, respectively, in the dbtS/+ and dbtP/+ backgrounds. In contrast, it caused only 1.1 and 0.63 h period lengthening in dbtL and dbtAR backgrounds (Figure 6, A and C). Such non-additive effects suggest a genetic interaction between lark and dbt. Similarly, knockdown of LARK caused period shortening in all dbt mutant backgrounds, with the effect being most prominent in a long-period background (dbtAR/+; Figure 6, B and D).
Our previous research found that high level LARK overexpression, using two copies each of pdf-gal4 and uas-lark, resulted in complete arrythmicity [50]. Research by others has shown that overexpression of a wild-type form of DBT in clock cells has a minimal effect on period but causes a reduction in rhythmicity [60]. We asked whether the arrhythmic behavior caused by high-level LARK expression is mediated through DBT. To address this question, we generated pdf-gal4/+; uas-lark/uas-dbt flies that carry a single copy of each responder transgene. Such flies were arrhythmic compared to controls that only expressed the uas-dbt or uas-lark transgenes (Figures 5 and 7), indicative of an interaction between the genes. This interaction required DBT kinase activity, as overexpression of LARK and DBTD132N, a mutant form of DBT devoid of kinase activity [4] did not cause significant arrhythmicity (Figure 7). In contrast, overexpression of DBTD132N suppressed the period-lengthening effect of mild LARK OE, possibly due to a dominant-negative effect caused by competition of the kinase-dead protein with wild-type protein. The average period for flies overexpressing LARK alone and flies overexpressing both LARK and DBTD132N were 25.1±0.06 hours and 22.67±0.11, respectively (Table S1).
We note that a previous study by Muskus et al. (2007) showed that expression of a different kinase-dead mutation of DBT (DBTK38R) in clock cells caused a lengthened period or arrythmicity [60]. Thus, it is surprising that expression of DBTD132N alone did not have obvious effects on period length or rhythmicity in our experiments (Table S1). However, Muskus et al drove expression of DBTK38R in all clock cells throughout development using a tim-gal4 driver. In this study we used the pdf-gal4 driver to direct expression of DBTD132N only in LNvs. More importantly, to avoid effects caused by potential developmental defects, we used the TARGET method [61] to confine expressing of DBTD132N to adulthood. These factors may explain the differences between our observations and those of Muskus et al (2007).
DBT kinase is involved in multiple steps of the sequential phosphorylation of PERIOD, priming the clock protein for ubiquitin-mediated degradation (reviewed in [18]). PER degradation rate is a key determinant of circadian period length (reviewed in [18]). To test the possibility that LARK modulates period length by regulating DBT-dependent PER degradation, we monitored PER nuclear degradation in the PDF-positive large ventral lateral neurons (l-LNvs) by immunohistochemistry and confocal imaging. We found that LARK OE caused a reduced rate of PER degradation during the initial 2.5 hours after lights on in an LD cycle (Figure 8). This result suggests that LARK modulation of DBT results in altered PER degradation.
Despite many studies of DBT function in cellular signaling pathways and circadian period determination, little is known about the regulation of DBT itself. In this study we show that translation of dbt transcripts are regulated by a clock-controlled RBP called LARK. We provide direct evidence that LARK promotes the translation of dbt transcripts in clock cells. Western Blot analyses reveal a previously undescribed smaller isoform of DBT promoted by LARK overexpression (Figure 2). Although we could not examine this smaller protein in null mutants (see Results) - to show specificity of the DBT antibody - three observations suggest that it corresponds to a novel DBT isoform. First, LARK can bind to dbt transcripts and overexpression of the RBP promotes the appearance of the novel DBT immunoreactive band. Second, overexpression of dbt, similar to LARK, results in the appearance of the novel protein. Finally, the novel protein shows circadian changes in abundance that are in phase with those of LARK. Together, these findings indicate the existence of a novel DBT isoform, encoded by one or more dbt transcripts that are regulated by LARK.
As previously mentioned, LARK or DBT OE are associated with the appearance of higher molecular weight DBT immunoreactive bands in addition to the novel short isoform. (Figure S6). Individual proteins of these size classes cannot be encoded by known dbt mRNAs. Therefore, they likely represent aggregates of DBT. Their formation might be facilitated by interaction with the short isoform, which we postulate may act as a scaffold due to its lack of a kinase domain. Although we hypothesize that the short isoform is responsible for the period altering effect, our results do not rule out the possibility that these higher molecular weight complexes contribute to the observed phenotypes.
As demonstrated by Western analysis, the novel isoform has a slightly lower molecular weight compared to the known isoform of DBT, indicating a shorter amino acid sequence. Since the four alternative transcripts encode the same Open Reading Frame (ORF) and differ only in their 5′UTR, it is possible that binding of LARK promotes translation from an AUG, or an unconventional initiation sites such as CUG, GUG, UUG, or ACG, downstream of the conventional start codon. It is known that translation of another target of LARK, E74A, utilizes at least three alternative initiator codons: two minor forms of the protein are initiated at a CUG and an AUG, while the most abundant form initiates at a CUG [62]. Similar to DBT, our previous studies of E74A show that LARK overexpression dramatically increases E74A protein abundance, changing the level from barely detectable to very high [44]. Of note, the mammalian homolog of LARK, RNA Binding Motif Protein 4 (RBM4), is known to promote cap-independent, internal ribosome entry site (IRES)-mediated translation when phosphorylated by the p38 MAPK pathway [63]. It is possible that the smaller isoform of DBT results from IRES-mediated translation.
At present, we do not know which dbt transcript expresses the short DBT isoform although all four transcripts are capable of encoding it. We also note that our results do not rule out an alternative but unlikely possibility that LARK OE results in DBT proteolytic cleavage resulting in the smaller isoform. However, the observations that LARK binds dbt RNA and promotes ribosome association of dbt transcripts without causing a significant change in abundance of the larger DBT isoform indicates that LARK may promote translation of the small isoform.
As the conserved kinase domain of DBT starts close to the 5′ terminus at amino acid 15, any alternative initiation site downstream of the original AUG is likely to affect kinase activity. Thus, it is possible that the short DBT isoform has no kinase activity but rather plays a structural role. A non-catalytic role of DBT has been suggested by others in a recent study. Yu et. al. (2009) found that PER-DBT binding, but not DBT catalytic activity, is required for CLK hyperphosphorylation and transcriptional repression and proposed a model in which DBT plays a novel, noncatalytic role in recruiting additional kinases that phosphorylate CLK, thereby repressing transcription [36]. Our results indicate that both the LARK-induced short isoform and full length wild-type DBT are required to exert the period lengthening effect, as co-expressing a kinase-dead form of full length DBT abolishes the period-lengthening effect of LARK OE (Figure 7). These results suggest that the short-isoform and full-length kinase may interact to set the speed of the clock. A plausible hypothesis is that the short DBT isoform serves as a non-catalytic subunit which modulates full-length DBT kinase. Thus, the ratio of short to full-length DBT may be important for modification of PER.
In a previous genome-wide study we identified many mRNAs that are associated with LARK in vivo [44]. Among these LARK-associated mRNAs, only three others encode proteins that are known to be involved in circadian function: flapwing (flw), no receptor potential A (norpA), and dunce (dnc). We did not detect association of LARK with canonical clock mRNAs (per, tim, clk, cyc, etc.). Thus it seems likely that the effect of LARK on period is mediated by DBT.
How might LARK regulate DBT and circadian period? As already indicated, RBM4 (mammalian LARK) is activated and shuttles to the cytoplasm to regulate IRES-dependent translation in response to p38 phosphorylation [64]. Interestingly, evidence suggests that p38 may have roles in circadian clock function [65], [66], and it is known to mediate circadian output and/or clock responses to light in several systems [67], [68]. Thus, the known clock regulation of LARK [47] may, in part, depend on p38-mediated phosphorylation of the protein. In turn, changes in LARK amount or activity might regulate DBT translation, as suggested by our study.
Alterations in DBT expression are predicted to modulate circadian period, by affecting either the accumulation or degradation of PER. Our results show that PER degradation in clock neurons is prolonged, in vivo, by increased LARK expression (Figure 8).
PER degradation requires binding of SLIMB, an F-box protein that helps target proteins to the ubiquitin–proteasome degradation pathway [34], [35]; SLIMB binding to PER requires a series of sequential phosphorylation events on PER [40]. These include phosphorylation at S661 and residues within a so-called “per-short domain”, spanning amino acids S585 to Y601, to which mutations that shorten period have been mapped (perS and perT; [69]–[74]). Chiu et al. (2011) have shown that phosphorylation of the per-short domain by the NEMO and DBT kinases (including S589, a DBT target residue) slows down phosphorylation of PER S47, a critical event for binding of SLIMB and PER degradation [40]. Lack of per-short domain phosphorylation leads to faster degradation of PER and short-period behavioral rhythms [40]. These results are consistent with a previous study suggesting that the per-short domain regulates the activity of DBT against PER [75]. Thus, enhanced or prolonged phosphorylation of this domain may lengthen period. We postulate that increased LARK expression and production of a short, non-catalytic DBT isoform leads to delayed PER degradation and lengthened circadian period by altering the timing of DBT-mediated phosphorylation of the per short domain. The observation that dbtP, which is a hypomorphic allele of dbt, enhances the period lengthening effect of LARK OE (compare Figure 6 with Figure 5, also see Table S1) suggests that alteration of the short to full-length DBT ratio may be responsible for period lengthening. Interestingly, a mutant form of DBT (DBTAR) that was suggested to play a non-catalytic, auxiliary role – similar to our proposal for the DBT short isoform – also causes period lengthening in heterozygotes [75].
Our analysis of DBT regulation revealed a dbt transcript showing light-inducible translation that is affected by LARK levels (Figure 4). This novel transcript, dbt-RE, shows a translational oscillation that is in phase with LARK abundance in LD conditions and it can be induced by light in dark conditions. Together with the observation that LARK abundance is highest at the beginning of the day [47], these results suggest that this RNA-binding protein may be light inducible in addition to showing circadian variation. In LD conditions, the light-induced increase in LARK level may up-regulate translation of dbt-RE. Based on the observation that dbt-RE represents an extremely small fraction of all ribosome-associated dbt transcripts (∼0.56%) captured by TRAP, it is possible that such a light-induced event occurs only in a small number of adult head clock cells, perhaps only in cells that mediate the light response. Although a role for LARK and DBT in pacemaker light sensitivity has not been reported, our study suggests it may be of interest to explore this possibility.
The following stocks were obtained from the Bloomington Stock Center (stock number in parenthesis): w1118 (5905), elav-gal4 (458), uas-dbt (26269 and 26274) dbtP (12164) and uas-dicer2 (24650). uas-lark, uas-larkRRM and uas-larkRNAi were described previously [49], [51]. uas-EGFP-L10a is a transgenic line generated in our lab that carries a UAS construct for expressing EGFP-tagged mouse ribosomal protein L10a [55]. tim-uas-gal4 was obtained from Dr. Justin Blau, pdf-gal4 was obtained from Dr. Patrick Emery, dbtL, dbtS, dbtAR were provided by Dr. Paul Hardin, uas-dbtD132N was provided by Dr. Marek Mlodzik. Flies were raised in incubators set at 25°C and 60% humidity and a light-dark cycle consisting of 12 hours of light and 12 hours of dark (LD 12∶12) unless specified otherwise.
For Western Blot (Figure 2) experiments, genotyppes are: KD, elav-gal4(/+); uas-dicer2/+; uas-larkRNAi/+. KC, elav-gal4(/+); uas-dicer2/+. OE, elav-gal4(/+); uas-lark/+. OC, elav-gal4(/+). DBT overexpression, elav-gal4(/+); uas-dbt/+. Control for DBT overexpression, elav-gal4(/+); +/+. Note that “elav-gal4(/+)” denotes the fact that male flies are hemizygous for elav-gal4 and female flies are elav-gal4/+.
For TRAP experiments, genotypes for examining the effect of altered LARK expression in all neurons are: KD, elav-gal4(/+); lark1 uas-larkRNAi/uas-EGFP-L10a. C, elav-gal4(/+); uas-EGFP-L10a/+. OE, elav-gal4(/+); uas-lark/uas-EGFP-L10a. Genotypes for examining the effect of altered LARK expression in all clock cells are: KD, w1118; tim-uas-gal4/+; lark1 uas-larkRNAi/uas-EGFP-L10a. C, w1118; tim-uas-gal4/+; uas-EGFP-L10a/+. OE, w1118; tim-uas-gal4/+; uas-lark/uas-EGFP-L10a (Figure 2). The genotype for examining circadian (figure 3) or light-induced (Figure 4) translation of dbt transcripts is w1118; tim-uas-gal4/+; uas-EGFP-L10a/+.
For locomotor behavior assays, genotypes are: KD, w1118; pdf-gal4 uas-dicer2/+; lark1 uas-larkRNAi/+. KC, w1118; pdf-gal4 uas-dicer2/+. OE, w1118; pdf-gal4/+; Tub-gal80ts uas-lark/+. OC, w1118; pdf-gal4/+; Tub-gal80ts/+. OERRM, w1118; pdf-gal4/uas-larkRRM. pdf>dbt alone: pdf-gal4/+; uas-dbt/+. pdf>dbt with LARK OE: pdf-gal4/+; uas-dbt/Tub-gal80ts uas-lark. pdf>dbtD132N alone: pdf-gal4/+; uas-dbtD132N/+. pdf>dbtD132N with LARK OE: pdf-gal4/+; uas-dbtD132N/Tub-gal80ts uas-lark. To prevent developmental effects known to be caused by LARK OE, the crosses and progeny were reared at 23°C until the time of experiment, when they were transferred into 30°C to deactivate the protective effect of Tub-gal80ts and allow OE to be achieved. To examine genetic interaction between LARK OE or KD and various chromosomal mutations of dbt, virgin females from either the w1118; pdf-gal4; uas-lark Tub-gal80ts strain (for OE) or the w1118; pdf-gal4 uas-dicer2; lark1 uas-larkRNAi/TM2 Ubx strain (for KD) were crossed to males of the dbtL, dbtS, dbtP, or dbtAR, respectively, and male progeny of the crosses were used for the behavioral analyses.
Polyclonal rabbit anti-LARK antibodies [47] were used for IP of LARK protein. A mono-clonal mouse anti-EGFP (clone 19C8 from MACF), was used as a control for unspecific bindings of RNAs to antibody-coupled Dynabeads. The antibodies were coupled to Dynabeads (Invigrogen) according to manufacturer's instruction. Flies of the w1118 strain were entrained to LD 12∶12 for 3 days and then flash frozen in liquid nitrogen at ZT2. Heads were harvested and homogenized in a mild lysis buffer containing 100 mM KCl, 5 mM MgCl2, 10 mM HEPES PH 7.0, 0.5% Ipegal-CA630, 1 mM DTT, 1 mM PMSF, and 10 µg/ml protease inhibitor cocktail (Sigma). The homogenates were incubated on ice for 5 minutes and centrifuged at 14,000× g for 20 minutes at 4°C. Cleared lysates were incubated with antibody coupled Dynabeads at 4°C for 1 hour. Following incubation, the supernatants were removed and the beads were washed 6 times using a buffer containing 20 mM HEPES-KOH (pH 7.4), 5 mM MgCl2, 350 mM KCl, 1% IGEPAL-CA630, and 0.5 mM DTT. RNAs were extracted from the immunoprecipation using the Trizol LS reagent (Invitrogen) and reverse transcribed into cDNA using Superscript II reverse transcriptase (Invitrogen) with random hexamers. The various dbt transcripts in the anti-LARK immunoprecipitated and anti-EGFP immunoprecipitated samples were analyzed by Q-RT-PCR using primers specific to each transcript (see below).
RNA transcripts used in the UV cross-linking assays were synthesized in vitro using 32P-UTP and the MEGAscript Kit (Ambion). The cDNA template for dbt was obtained from the Drosophila Genomics Resource Center (EST clone LD 27173) and for GluR2 was obtained from Dr. Joel D. Richter. A LARK N-terminal GST fusion protein containing the N-terminal RNA-binding domains (two RRM domains and one RTZF) was synthesized and purified using the Pierce GST Purification Kit. RNA-protein binding reactions were carried out according to [53]. Briefly, 1×105 cpm of in vitro synthesized RNA transcript and varying amounts of LARK-GST fusion protein were added to 2X GR buffer (20 mM HEPES, pH 7.6, 100 mM KCl, 2 mM MgCl2, 0.2 mM ZnCl2, 20% glycerol, 2 mMDTT), 10 ng t-RNA, 1.2U Rnase OUT (Life Technologies), and 1 mM DTT and incubated on ice for 10 min. followed by RT for 10 min. 50 mg of heparin was added to the mixture followed by UV exposure at 440 mJ for 3 min. RNase A (10 ng) was added and incubated for 30 min at 37°C. The products were resolved by SDS-PAGE and binding was detected using a Typhoon Phosphoimager (GE Healthcare).
Flies of designated genotypes were raised at 25°C under standard conditions. Newly emerged adult flies were transferred into an incubator and entrained to LD 12∶12 at 30.5°C for 3 full days and then flash froze in liquid Nitrogen at the appropriate zeitgeber times on day 4. Heads of the frozen flies were harvested and ground into fine powder in liquid Nitrogen. The frozen powder was mixed with a mild lysis buffer (100 mM KCl, 5 mM MgCl2, 10 mM HEPES PH 7.0, 0.5% IGEPAL-CA630, 1 mM DTT, 1 mM PMSF, and 10 µg/ml protease inhibitor cocktail (Sigma), incubated on ice for 5 minutes, and centrifuged at 14,000× g for 20 minutes at 4°C. Cleared tissue lysate was obtained after the centrifugation and the concentration of total protein was determined. Approximately 10 ug samples of total protein were loaded onto 12% polyacrylamide gels. Electrophoresis and western blotting were carried out according to standard protocols. The DBT proteins were detected using anti-DBT antibodies provided by Dr. Jeffrey Price (University of Missouri-Kansas City).
Flies carrying the uas-EGFP-L10a construct [55] were crossed to appropriate gal4 lines to express GFP-tagged ribosomes in desired cell types. Details of the TRAP method are described in [55]. Briefly, fly tissues were homogenized in a buffer containing 20 mM HEPES-KOH (pH 7.4), 150 mM KCl, 5 mM MgCl2, 10 µg/ml protease inhibitor cocktail (Sigma), 0.5 mM DTT, 20 unit/µl SUPERase.In RNase inhibitor (Invitrogen), and 100 µg/ml cycloheximide. Thirty mM DHPC and 1% IGEPAL-CA630 were added to the cleared tissue lysates. The mixtures were incubated on ice for 5 minutes and cleared again by centrifuging at 14,000× g for 20 minutes. The cleared lysates were applied to magnetic beads covered by purified anti-EGFP antibodies and incubated at 4°C with gentle rotating for 1 hour. After the IP, the beads were washed with a buffer containing 20 mM HEPES-KOH, pH 7.4, 5 mM MgCl2, 350 mM KCl, 1% IGEPAL-CA630, 0.5 mM DTT and 100 µg/ml cycloheximide. RNAs were extracted from the beads using the Trizol-LS Reagent (Invitrogen).
Total RNA samples were treated with DNase I (Invitrogen) to eliminate potential contamination with genomic DNA. RNAs isolated from TRAP experiments were used directly since these RNAs usually do not carry genomic DNA contamination. Treated total RNAs or TRAP RNAs were primed with random hexamers (Ambion) and reverse transcribed into cDNAs using the Superscript II reverse transcriptase (Invitrogen). Quantification of the relative abundance of specific transcripts in the cDNA samples was conducted by Q-RT-PCR using 2X SYBR green PCR Master Mix (Applied Biosystems) and specific primers. Data were collected with Strategene Mx3000 or Mx4000. A pair of primers specific for the Ribosomal Protein 49 (Rp49) gene, which is known to be transcribed and translated at a constant rate throughout the circadian cycle (Huang and Jackson, unpublished observation), was used as an internal reference to account for variation in the input cDNA amount. Sequences for specific primers were: Rp49-F: GCCCAAGATCGTGAAGAAGC, Rp49-R: CGACGCACTCTGTTGTCG, dbt-RA-F: GATGCAAAACAACCCTTCGAATAC, dbt-RA-R: CCCAGGCGATATTTGTTACC, dbt-RB-F: AACGTAAGTGTCGAATTAGAAG, dbt-RB-R: CTGGCACTGTCCTTTCGTCT, dbt-RC-F: GCGACTGTGGCAACTACAAC, dbt-RC-R: CTGGCACTGTCCTTTCGTCT, dbt-RE-F: CGCTGCAGATGCGATAAAAA, dbt-RE-R: GATTTGCGTTGCCTTTCTGG.
Locomotor activity was assayed using 2- to 3-day-old males and the Drosophila Activity Monitoring (DAM) system (Trikinetics, Waltham, MA). Flies were loaded into activity monitors and placed in incubators set at either 30°C (for flies carrying Tub-gal80ts) or 23°C (for flies not carrying Tub-gal80ts), they were entrained to LD 12∶12 for 4–5 days and then released into constant darkness (DD) for an additional 7–10 days. Visualization of actograms and the analysis of rhythmicity and period length were performed using a signal processing toolbox [76] within the MATLAB software package (MathWorks). The toolbox analyzes circadian rhythmicity of fly locomotor activity by applying an autocorrelation analysis. The Rythmicity Index (RI) is defined as the height of the third peak in the correlogram resulting from this analysis (counting the peak at lag 0 as the first peak). Period length is determined by Fourier analysis [76]. Flies were considered rhythmic if they had a high RI value (generally greater than 0.2) as well as obvious rhythmicity by visual inspection of the actogram.
To visualize PER cycling in the PDF neurons, adult flies were harvested at appropriate circadian times and fixed in 4% paraformaldehyde solution. Brains were dissected from the heads and washed in PBS and PBS-T (0.05% Triton X-100). For assessing LARK abundance in PDF neurons, adult flies were harvested at ZT 2 and brains were dissected prior to fixation. After dissection, the brains were fixed in 4% paraformaldehyde solution and then washed in PBS and PBS-T. Immunohistochemistry was carried out according to standard procedure for staining whole mount fly brains. Primary antibodies were used at the following dilutions: Rabbit anti-PER (1∶10000, R. Stanewsky), mouse anti-PDF (1∶10, DSHB), Rabbit anti-LARK (1∶1000, [47]). Secondary antibodies, goat anti-mouse IgG (Alexa-488 conjugated, Molecular Probes) and goat anti-rabbit (Cy3 conjugated or Alexa-488 conjugated, Molecular Probes) were used at a dilution of 1∶300 and an incubation time of at least 5 hours. Confocal images were acquired from brain whole mounts using a Leica TCS SP2 AOBS microscope within the Tufts Center for Neuroscience Research (CNR) Imaging Core. Blind scoring for PER nuclear versus cytoplasmic localization in the s-LNvs was accomplished by using the following scoring system: 0 = no staining in nuclei, 1 = mixture of nuclear and cytoplasmic staining, and 2 = nuclear staining only. To assess the time course of PER degradation in the nuclei of l-LNvs, a custom ImageJ macro program was used to quantify PER immunoreactivity. All l-LNvs in a brain hemisphere of a particular animal were imaged as a 3D stack with optical sections in 1 µm steps under a 63× oil lens objective. The section with the largest cell diameter, i.e. the middle section of the cell, was identified and an ROI was drawn manually outlining the nucleus. Average pixel intensity within the ROI was calculated for each individual l-LNv cell in a brain hemisphere. The value obtained for individual cells were then further averaged among all cells in a same brain hemisphere to get a value for each individual animal.
|
10.1371/journal.pntd.0000034 | Optimization of Topical Therapy for Leishmania major Localized Cutaneous Leishmaniasis Using a Reliable C57BL/6 Model | Because topical therapy is easy and usually painless, it is an attractive first-line option for the treatment of localized cutaneous leishmaniasis (LCL). Promising ointments are in the final stages of development. One main objective was to help optimize the treatment modalities of human LCL with WR279396, a topical formulation of aminoglycosides that was recently proven to be efficient and safe for use in humans.
C57BL/6 mice were inoculated in the ear with luciferase transgenic L. major and then treated with WR279396. The treatment period spanned lesion onset, and the evolution of clinical signs and bioluminescent parasite loads could be followed for several months without killing the mice. As judged by clinical healing and a 1.5-3 log parasite load decrease in less than 2 weeks, the 94% efficacy of 10 daily applications of WR279396 in mice was very similar to what had been previously observed in clinical trials. When WR279396 was applied with an occlusive dressing, parasitological and clinical efficacy was significantly increased and no rebound of parasite load was observed. In addition, 5 applications under occlusion were more efficient when done every other day for 10 days than daily for 5 days, showing that length of therapy is a more important determinant of treatment efficacy than the total dose topically applied.
Occlusion has a significant adjuvant effect on aminoglycoside ointment therapy of experimental cutaneaous leishmaniasis (CL), a concept that might apply to other antileishmanial or antimicrobial ointments. Generated in a laboratory mouse-based model that closely mimics the course of LCL in humans, our results support a schedule based on discontinuous applications for a few weeks rather than several daily applications for a few days.
| When initiating the cutaneous disease named cutaneous leishmaniasis (CL), Leishmania parasites develop within the parasitophorous vacuoles of phagocytes residing in and/or recruited to the dermis, a process leading to more or less chronic dermis and epidermis-damaging inflammatory processes. Topical treatment of CL could be a mainstay in its management. Any improvements of topicals, such as new vehicles and shorter optimal contact regimes, could facilitate their use as an ambulatory treatment. Recently, WR279396, a third-generation aminoglycoside ointment, was designed with the aim to provide stability and optimal bioavailability for the molecules expected to target intracellular Leishmania. Two endpoints were expected to be reached: i) accelerated clearance of the maximal number of parasites, and ii) accelerated and stable repair processes without scars. A mouse model of CL was designed: it relies on the intradermal inoculation of luciferase-expressing Leishmania, allowing for in vivo bioluminescence imaging of the parasite load fluctuation, which can then be quantified simultaneously with the onset and resolution of clinical signs. These quantitative readout assays, deployed in real time, provide robust methods to rapidly assess efficacy of drugs/compounds i) to screen treatment modalities and ii) allow standardized comparison of different therapeutic agents.
| Of the 350 million people exposed to the risk of Leishmania parasite inoculation and further development, 2 million each year experience the discomfort and potential complications of cutaneous leishmaniasis (CL). Many active lesions are disfiguring, and remain so when healing as inesthetic scars that expose patients to social stigma, sometimes for life [1],[2]. The demand for improved CL therapy has been fueled for decades by the lack of an efficient, affordable, easy-to-apply drug/schedule, as well as by the risks associated with the use of parenteral antiparasitic drugs such as pentavalent antimonial drugs or pentamidine [3],[4]. Topical therapy of CL is a promising approach [5],[6]. The aminoglycoside paromomycin is the most well studied compound as a potential topical treatment for CL [7]. First and second generation paromomycin-based ointments were either reasonably efficient [8],[9] but too irritant (first generation paromomycin-Methyl benzo chloride, “Leshcutan”) [10],[11] or well-tolerated but not efficient enough when first tested in humans (second generation paromomycin-urea “WHO formulation”) [12],[13]. WR279396, a third-generation aminoglycoside ointment that contains 15% paromomycin formulated in a hydrophilic vehicle as well as a second aminoglycoside, 0.5% gentamicin, was designed to be effective but non-irritative. This new formulation was recently shown to be efficient and safe for the treatment of L. major localized cutaneous leishmaniasis (LCL) (Ben Salah, Buffet et al.,submitted and [14]). Although very encouraging, this result is only one step toward a simple and easily applicable therapy for this neglected disease. Various parameters such as frequency and duration of application or application in the presence or absence of an occlusive dressing-may markedly influence the efficacy or safety of topically applied formulations [12],[13],[15],[16]. For example, once-a-day applications of Leshcutan are associated with less frequent and less severe local reactions than a twice-a-day application schedule [17]. Though still suboptimal, a 28-day schedule of paromomycin-urea (WHO formulation) is significantly more efficient than a 14-day schedule [12]. These 2 examples show that optimizing application parameters through clinical trials, the most reliable approach, takes years. Also, for obvious ethical reasons, there is usually no untreated control group in clinical trials, making interpretation of the mechanisms of drug action more difficult.
In order to more rapidly and accurately identify important parameters that influence the efficacy of WR279396, we designed and used a mouse model of CL that mimics important features of the natural sand fly dependent-transmission of parasites to mammal. A relatively low (104) inoculum of L. major metacyclic promastigotes was injected in the C57BL/6 ear dermis [18]. As in a majority of patients with L. major CL [19], the development of localized dermal lesions in C57Bl6 mice is followed by spontaneous healing over the course of weeks to months [18]. Because luciferase transgenic parasites were used in this model, the kinetics of parasite load could be established without killing the mice: indeed, a linear correlation between bioluminescence values and parasite loads assessed by the reference limiting dilution technique has been previously established [20].
Female C57BL/6 (5 week old) and Swiss nu/nu mice were purchased from Charles River (Saint Germain-sur-l'Arbresle, France) and were housed under institutional guidelines of the A3 Animal facility at Institut Pasteur (Paris, France).
A 1.66 kbp firefly luciferase coding region was cut from pGL3 basic (Promega, Madison WI) using NcoI/EagI and subsequently cloned into the Leishmania expression vector pF4X1.HYG (Jenabioscience, Jena, Germany) with a marker gene for selection with Hygromycin B (Cayla, Toulouse, France) which was previously cut with NcoI/NotI. In this construct, 18s rRNA flanked the luciferase and HYG genes. Following linearization with SwaI, luciferase and HYG genes were integrated into the 18s rRNA locus of the nuclear DNA of Leishmania. Transfections were realized by electroporation with the following conditions: 25 µF, 1500 V, in 4 mm cuvette; 3.75 kV/cm [21]. Following electroporation, cells were incubated 24 h in media without drug and plated on semisolid media containing 100 µg/ml of hygromycin B [20].
Transgenic luciferase L. major strain NIH173 (MHOM/IR/-/173) amastigotes were isolated from infected Swiss nude mice. Briefly, the promastigote developmental stage was grown at 26°C in M199 media supplemented with 10% FBS, 25 mM Hepes pH 6.9, 12 mM NaHCO3, 1 mM glutamine, 1×RPMI 1640 vitamin mix, 10 µM folic acid, 100 µM adenosine, 7.6 mM hemin, 50 U/ml of penicillin and 50 µg/ml of streptomycin [21]. Infective-stage metacyclic promastigotes were isolated from stationary phase cultures (6 day old) using density gradient centrifugation, as previously described [22].
C57BL/6 mice were anaesthetised by intraperitoneal administration of a mixture of Ketamine (120 mg/kg−1 Imalgene 1000, Merial, France) and Xylazine (4 mg kg−1; Rompun 2%, Bayer, Leverkusen, Germany). Ten thousand metacyclic promastigotes per 10 µl of Dulbecco's phosphate buffered saline (PBS) were injected in the right ear dermis. Images of ketamine-xylasine anaesthetised mice were captured each day bioluminescence analyses were performed. The clinical features of parasite-loaded ear were examined based upon three phases: 1) early, leucocyte infiltrate-free inflammatory, processes, 2) leucocyte infiltrates-positive inflammatory processes and 3) late repair processes could be distinguished. Only one name, “lesion”, was used to designate these different processes. The “lesion” size measurement (mm2) was approximated from the picture by fit within a rectangle.
Luciferin (D-Luciferin potassium salt, Xenogen, California), the luciferase substrate, was intra-peritonealy inoculated into mice at a concentration of 150 mg/kg 25 minutes before bioluminescence analysis. Mice were anaesthetised in a 2.5% isoflurane atmosphere (Aerane, Baxter SA, Maurepas, France) for 5 minutes and maintained in the imaging chamber for analysis. Emitted photons were collected by 1 minute acquisition with a charge couple device (CCD) camera (IVIS Imaging System 100 Series) using the high resolution (small bining) mode. Analysis was performed after defining a region of interest (ROI) that delimited the surface of the entire ear. The same ROI was applied to every animal at every time point. Total photon emission from the ventral image of each mouse ear was quantified with Living Image software (Xenogen Corporation, Almeda, California), and results are expressed in number of photons/sec/ROI. The photon signal from the ear is presented as a pseudocolor image representing light intensity (red = most intense and blue = least intense) and superimposed on the grey scale reference image. Of note, the lower threshold bioluminescence value indicates a parasite load of ≥5000 parasites per ear, precluding any detection of persisting parasite population that oscillates between 500 and 1000 parasites.
Forty to 60 animals per experiment were inoculated with transfected L. major and the total photon emission of each ear was quantified 11 days later. Mice were monitored and distributed in groups according to an equal median bioluminescence value (1×106–5×106 photons/sec/ROI) and standard deviation. Each experimental group contained 7 to 10 mice, each individually ear-tagged (the contralateral ear with respect to the inoculation site).
Topical formulations were prepared at the Walter Reed Army Institute of Research (Washington DC). WR279396 consists of paromomycin sulphate (15%) plus gentamicin (0.5%) in a vehicle as previously described [23].
From day eleven post-L. major inoculation, topical ointments were applied to parasite-loaded ears once every two days for 10 days or once everyday for 5 days. Each formulation was applied using a sterile tip directly onto the ears to form a thin layer. Control groups were treated with the vehicle used in the medication without any of the active ingredients, i.e., the paromomycin and gentamicin. The ointment was either left open without dressing or covered with an occlusive dressing. The occlusive dressing was an adhesive polyurethane membrane (Tegaderm; 3M Health Care, St Paul, USA) that keeps water but is permeable to both water vapour and oxygen. Then two independent leaflets of 3M Micropore Surgical Tape (3M Health Care) were directly applied to the Tegaderm. This tape permitted maintenance of Tegaderm and formulation in contact with the ear during the two days.
We estimated the number of parasites present in parasite-loaded ears as previously described [24]. Ears were cut off. The dorsal ear half was separated from the cartilage-containing ventral ear half, cut into small pieces and ground in HOSMEM-II culture medium using a glass tissue homogenizer. Tissue/organ homogenates were serially diluted in HOSMEM-II culture medium and then dispensed into 96-well plates containing semi-solid agar (Bacto-Agar, Difco, Detroit, MI) supplemented with 10% sterile rabbit blood collected on heparin. Plates were incubated for ten days and each well was then examined and classified as positive or negative according to whether or not viable promastigotes were present. Limiting dilution analysis was then applied to the data to estimate the number of viable parasites, expressed in limiting dilution assay units (LDAU) [25]. Statistical analysis of the results was based on the maximal likelihood method [26],[27].
Lesion size or log transformed parasite loads were analyzed with a two-way analysis of variance (ANOVA). The two factors examined were the treatment (untreated, vehicle, drug, drug with occlusion) and the period of observation (treatment, post-treatment and final) in the statistical environment R. The assumption of homoscedasticity and normality were tested with the Bartlett and Kolmogorov-Smirnov test, respectively. If the interaction term was significant, pair wise comparisons using t tests were realized for each combination of factors. A probability level of p<0.05 was accepted for the purpose of declaring statistically significant treatment effects.
The first objective of this study was to design and validate standardized readout assays for assessing different drug regimens using C57BL/6 mice inoculated with Leishmania major. To carry out these experiments, 104 luciferase-expressing L. major metacyclic promastigotes were inoculated intradermally into the mouse ear. Parasites produced a significant bioluminescent signal in situ allowing parasite load expansion and reduction to be monitored non-invasively. The development and outcome of parasite burden and parasite-loaded ear features were examined simultaneously over a period of 3 months. The relationship between bioluminescence and the clinical features of the ear were respectively assessed by quantifying the number of photons per second per ear and measuring the “lesion” area.
Figures 1A and 1B illustrate the real-time bioluminescent images and clinical signs displayed by the L. major-inoculated ear from a representative C57BL/6 mouse (untreated group). The first post-inoculation phase (days 0–11) was characterized by a sharp increase of the bioluminescent signal at the inoculation site (from 7×103 to 4.4×106 photons/sec/ear at day 11; Figure 1B, C). By day 7, mouse ears displayed no clinically detectable sign (Figure 1A). However, a leukocyte infiltrate-free tiny red spot (5 mm2) was observed at day 11 (Figure 1A, C). Thus, during the first stage of parasite development no significant correlation was found between the bioluminescence value at the inoculation site and the clinically detectable features. By day 22, the parasite load peaked (Figure 1B and 1C) with a median value of 1.5×107 photons/sec/ear which was associated with the first bona fide cutaneous clinical signs (Figure 1A and 1C). The next phase of L. major-driven processes was characterized by a relatively sharp decrease in bioluminescence followed by healing of the ear lesion (Figure 1A, 1B and 1C). Following the complete and stable healing of this dermal lesion, no more bioluminescent signal was detected in the ear tissue (Figure 1B and 1C). We are aware that any persisting parasite load with a population size value ≤5000 per ear is not detectable using bioluminescence: thus, between days 80–96 post inoculation at the time of mouse sacrifice, mouse ears were recovered in the control and treated group. Using the LDA readout assay, these ears were monitored for the presence of persisting parasites. 40% of the ears were positive in all groups (≤500 parasites per ear) and these percentages were obtained from two independent experiments (data not shown).
In contrast, the persistent presence of a low number of parasites as measured by LDAU was noted in the inoculation site (3 positive mice out of 7) for up to 80 days post-inoculation.
These measurements helped us to define the onset of the first topical ointment application (WR279396 vehicle or WR279396 ointment with or without dressing). We decided to initiate treatment at day 11 post-inoculation for three reasons. First, at this time point, a high parasite load (bioluminescence values>1×106 photons/sec/ear) was reproducibly measured. Secondly, these values were observed in the median part of acute-phase load, allowing for monitoring of either an increase in parasite load in the absence of any topical application or a decrease in treated groups. Thirdly, the last topical ointment application in the group of mice treated with WR279396 was coincident with the highest parasite load measured in the control group.
By day 11 post-parasite inoculation, C57BL/6 mice were distributed in different groups on the basis of equal median bioluminescence values. WR279396 was applied topically to the L. major-inoculated ear every two days for 10 days. Occlusive dressing was performed by covering the L. major-loaded ears with adhesive polyurethane dressing (Tegaderm) and a surgical tape to maintain the formulation for 2 days (Figure 2). An evaluation of the effect of WR279396 with an occlusive dressing was monitored by measuring the bioluminescence and ear “lesion” area (Figures 3B and 3C). Three periods of observation have been defined i) the 10-day treatment period ii) the post-treatment period, which ends with the absence of any bioluminescence signals in the control group and iii) the late period. As controls, three groups were analysed. In the first group, ears were left untreated. In the second group, the paromomycin- and gentamicin-free vehicle was applied to L. major-inoculated ears that remained uncovered after application. In the third group, the WR279396 vehicle was applied and ears were immediately covered with an occlusive dressing. In all control groups, parasite load as well as lesion onset development and healing were simultaneously assessed. No statistical differences in parasite loads and lesion area were observed in any period between untreated and ointment vehicle-treated groups regardless of the period under study with respect to the measurement (not shown).
Monitoring of bioluminescence values showed that topical treatment with WR279396 (without a dressing) accelerated the decrease of both the parasite load (Figure 3A) and “lesion” area (Figure 3B and 3C). Two-way ANOVA analysis indicated a significant effect of treatment (P-value<9.2×10−6) and period (p-value<2.2×10−6) on parasite load for the whole experimental group, and there was a significant interaction between treatment and period effect (p-value<0.008). The parasite load (grey line; Figure 3A) decreased rapidly after the fourth application (day 18; Figure 3A). Median values of bioluminescence indicated that parasite loads in the group of mice left without dressing were significantly lower than the control group during the treatment period (Figure 3A and 3D; grey line vs blue line and box plot-: p-value = 0.0033).Of note, during the post-treatment period, a rebound pattern of parasite load was observed in mice treated with ointment without occlusive dressing (grey line; Figure 3A) no statistical difference between groups being noted (Figure 3D).
For mouse ears that were covered with WR279396 under an occlusive dressing, mean parasite loads (Figure 3A; brown lines) decreased earlier than those with WR279396 left without any dressing. One day post the last application, bioluminescence values reached threshold bioluminescence values (Figure 3A) in 80% of mice (8 out of 10). During the post-treatment period, statistical analyses indicated that parasite load in the group of mice with an occlusive dressing was significantly lower than in the group of mice treated without a dressing (Figure 3B, p-value = 2.2×10−8). The higher significant therapeutic effect of the drug in the presence of an occlusive dressing during this post-treatment period is illustrated in figures 3B, 3C (day 28: p-value = 0.00055). Furthermore, no rebound of parasite load was observed in this group. In conclusion, the decrease in parasite loads and the healing process occurred earlier in mice treated with WR279396 under an occlusive dressing. Among this group of mice, neither clinical relapse-as measured by leucocyte infiltrate-related “lesion” area-nor rebound of parasite load was detected.
Parasite rebound was observed in some mice given WR279396 without occlusive dressing. The individual follow-up of parasite-loaded mouse ears in real time indicated a clear dichotomy in the patterns of parasite load outcome (Figure 4; same experiment as Figure 3; n = 10). In the majority of treated mice, parasite load decreased faster than in control mice with a bioluminescence value lower than 1×106 photons/sec/ear (Figure 4A; 6 out 10 mice depicted in green) at day 33. Furthermore, no rebound was detected in this group of mice (Figure 4B). In contrast, bioluminescence values for the four remaining mice were higher than 1×106 photons/sec/ear at day 33 (Figure 4A and 4B; 4 mice depicted in red): post treatment, either the parasite load reduction pattern followed the same profile (1/4 mice; Figure 4B) as the one displayed by mice of the control group (grey area) or relapsed occurred at day 22 (3/4 mice; Figure 4B). The parasite load in this latter group of 3 mice remained higher than the parasite load of the control group from days 30 to 60. Interestingly, bona fide lesion area values did not assess any obvious clinical failure except in 1 mouse which harboured the highest parasite load (see arrow in Figure 4B) and displayed a somewhat slower healing process (Figure 4C; see arrow in Figure 4C). These data suggest that i) the rebound pattern of parasite load, which was observed in mice treated in the absence of occlusive dressing, had a clinical impact in a minority of mice and ii) the bioluminescence imaging data provided relevant information on parasite load fluctuations that were not provided by careful clinical monitoring.
We also monitored application regimens of WR279396 to determine which one might have a superior therapeutic index against the parasite. The experimental protocol shown in figure 5 was as previously described except for a different schedule of drug ointment application. At day 11, the topical ointment was applied on parasite-loaded ears either daily for 5 days or once every two days for 10 days. A control group (no medication) was used in parallel for determining comparability and efficacy of the different topical therapy regimens. As previously described, all parasite-loaded ears exposed to five applications for 10 days were healed by day 21 (Figure 5A) without relapse by day 64. All lesions (7/7) treated with WR279396 daily for 5 days had healed at day 21 (end of the topical therapy), but 71% (5/7) and 14% (1/7) of mice relapsed at day 50 and day 60, respectively. The clinical aspect of the lesions at day 36 post-inoculation (Figure 5B) confirmed the greater efficiency of the 5 every two days application over a 10 day schedule. By two-way ANOVA, it was shown that the difference between treatments depends on the observation periods considered (significant interaction term, p-value<0.05 ; Figure 5C). Pair wise comparisons using t-tests for each combination of factors are shown in figure 5D.
The integrative analysis of parasite load evolution in 4 experiments, involving mice receiving 5 applications for 10 days either with (n = 31) or without (n = 23) an occlusive dressing, shows that 74% of mice treated without a dressing controlled parasite loads (17/23) without relapse. Of the 6 remaining mice, 2 were unresponsive, as shown by parasite load values similar to untreated mice. The other mice initially controlled parasite loads and lesion size during the treatment period, but relapsed by day 30 as shown by parasite load values similar to or higher those measured from the ears of untreated mice. In contrast, 94% of mice treated with an occlusive dressing healed (29/31) by day 30. Only 6% (2/31) had detectable-though very low-parasite load (bioluminescence level<1×106 photons/sec/ear) during follow-up. These results allow us to establish the greater parasitological efficacy of the schedule using an occlusive dressing, with a trend toward a prophylactic effect on relapse after a successful course of WR279396.
WR279396, a third-generation aminoglycoside-based ointment, was efficient on L. major-induced localized cutaneous lesions (LCL) in C57BL/6 mice. Five applications for 10 days under occlusion induced a 94% healing rate by day 30, without re-expansion of parasite loads. This high cure rate, as well as the general evolution profile in both treated and control mice, is strongly reminiscent of what has been observed in clinical trials (Ben Salah, Buffet et al.-submitted and [14]), providing a strong validation of this new model for drug-testing purposes.
The adjuvant use of an occlusive dressing significantly enhanced control of parasite loads. Several non-mutually exclusive mechanisms may account for these effects. First, the dressing prevented removal of the ointment from the lesion by protecting the skin from scratching, rubbing and scraping. These latter observations have been made in patients treated without occlusion, an important proportion of the ointment being wiped off by clothes during the day, sheets during the night or even attracted to a “protective” gauze put on the top of ulcerated lesions. Second, occlusion on burns or wounds favours epidermal regeneration (ie, ulceration closure). Finally, water retention by semi-permeable occlusive dressing (like the polyurethane film used here) results in hydration of the ointment application zone [28] and likely improves the penetration and diffusion of hydrophilic antiparasitic compounds into the dermis [15], where intracellular amastigotes multiply. The aminoglycosides paromomycin and gentamicin, the active ingredients in the WR279396 ointment, are OH-rich hydrophilic compounds. Whether the dominant mechanism of the adjuvant effect is merely mechanical (enough ointment maintained on the lesion) or linked to dermal diffusion issues, the occlusive dressing enhanced the healing process induced by active ingredients, and prevented persisting parasite loads to re-expand. Now that this adjuvant effect is established, future studies should be set up for dissecting its fine mechanisms.
Apart from a mild difference in thickness, LCL lesions in mice resemble human lesions both clinically and histologically. As opposed to systemic drug testing, topical drug testing in mice will be relevant since potential pharmacokinetic differences between mouse and human skin are expected to be minor and easily tractable to further analyses. These observations, along with the careful validation of the model, support the assumption that our results are likely to apply to human therapy. To our knowledge, occlusion has never been fully validated as an adjuvant for the topical therapy of human cutaneous lesions driven by invasive microorganisms, but several case reports have proposed this approach both in CL [29],[30] and non-infectious dermatologic conditions [15]. We provide here a strong validation of a concept that might apply to other antileishmanial or antimicrobial ointments. Ointments are usually painless, their application requires no sophisticated expensive device or local anaesthesia and they can be applied easily to both children and adults by a primary care health provider with minimal training. Excluding those L. braziliensis. foci where the incidence of mucosal extension can be high, ointment therapy of cutaneous lesions otherwise declared as neglected diseases should be favoured since the potential adjuvant effect of occlusion might help some ointment formulations to reach the required efficacy for development.
The duration of treatment is another important determinant of reaching a stable cure. A very short 5-day daily application schedule under occlusion led to a “rebound” pattern similar to that displayed in mice treated for 10 days without occlusion. In other words, too short of an application period may lead to parasite load rebound, this latter risk being partially controlled by an occlusive dressing. The ability to perform individual mouse follow-up revealed a dichotomic pattern of parasite load evolution (“sustained control” versus “unstable control with parasite rebound”), pervasive over many weeks post transient topical ointment application. Interestingly, these patterns were displayed over several weeks, i.e., well beyond the treatment application period. Parasite load level at the end of applications was not a good predictor of further evolution (Figure 4). So, not only parasite killing but also some modification of parasite environment determined the long-term outcome of tissue damage and repair processes. It is then very likely that, during the treatment application period, some integrated programs are triggered that will be the dominant determinant of evolution. Those results fit well with observations in human CL, such as the low prognosis value of parasitological tests at the end of therapy or the efficacy of therapeutic schedules stopped before lesion healing [4],[31]. Taken together, these experiments show that parasites must be exposed to the drug for>5 days to drive evolution toward long term sustained control of parasite loads and clinical healing. Duration of drug exposure was a stronger determinant of outcome than the total amount of drug used. Intracellular pharmacokinetics of aminoglycosides helps understand the mechanism leading to this observation. In eukaryotic cells exposed to aminoglycosides in vitro, a slow (2–4 days) lysosomal accumulation is observed, followed, when aminoglycosides are removed from the extracellular medium, by an even slower (2–5 days) release [32]. Interestingly, the lysosome is the only subcellular compartment in which aminoglycosides accumulate, an important feature of their antileishmanial efficacy [32],[33]. So, provided that appropriate concentrations of aminoglycosides are reached in the dermal intercellular space, relatively discontinuous applications would probably suffice to allow intracellular killing of replicating amastigotes and long term sustained control of parasite loads. Taken together, our observations will help select the most efficient ointment application schedules for implementation, in the context of the therapy of this neglected disease, by health care providers with little resources and heavy duties. Even relatively discontinuous applications for a few weeks should be preferred to many daily applications for a few days.
Our model offers relevant preclinical readout assays i) of the efficacy of a topical ointment delivered under occlusion or not ii) for establishing the proper regimen/schedule that allows sustained parasite load reduction and lesion healing during post-treatment period features. This luciferase-based imaging study might be useful for pre-clinical evaluation of novel formulations containing molecules that target parasite-loaded cells residing in the dermis as well as molecules that contribute to damaged skin-repair processes. Next challenges will be to screen molecules expected to act on the amastigote population that persist in the dermis or in distant sites [34] and to investigate the acquisition in real time of long-term protective immunity. |
10.1371/journal.pgen.1007089 | The pea branching RMS2 gene encodes the PsAFB4/5 auxin receptor and is involved in an auxin-strigolactone regulation loop | Strigolactones (SLs) are well known for their role in repressing shoot branching. In pea, increased transcript levels of SL biosynthesis genes are observed in stems of highly branched SL deficient (ramosus1 (rms1) and rms5) and SL response (rms3 and rms4) mutants indicative of negative feedback control. In contrast, the highly branched rms2 mutant has reduced transcript levels of SL biosynthesis genes. Grafting studies and hormone quantification led to a model where RMS2 mediates a shoot-to-root feedback signal that regulates both SL biosynthesis gene transcript levels and xylem sap levels of cytokinin exported from roots. Here we cloned RMS2 using synteny with Medicago truncatula and demonstrated that it encodes a putative auxin receptor of the AFB4/5 clade. Phenotypes similar to rms2 were found in Arabidopsis afb4/5 mutants, including increased shoot branching, low expression of SL biosynthesis genes and high auxin levels in stems. Moreover, afb4/5 and rms2 display a specific resistance to the herbicide picloram. Yeast-two-hybrid experiments supported the hypothesis that the RMS2 protein functions as an auxin receptor. SL root feeding using hydroponics repressed auxin levels in stems and down-regulated transcript levels of auxin biosynthesis genes within one hour. This auxin down-regulation was also observed in plants treated with the polar auxin transport inhibitor NPA. Together these data suggest a homeostatic feedback loop in which auxin up-regulates SL synthesis in an RMS2-dependent manner and SL down-regulates auxin synthesis in an RMS3 and RMS4-dependent manner.
| Plant shoot branching results from the precise regulation of bud growth versus dormancy. Positive and negative feedback mechanisms are likely involved in the dynamic control of this highly plastic trait. Strigolactones, the most recently discovered class of plant hormones, play a key role in controlling shoot branching. Negative feedback control of strigolactone biosynthesis has been observed in several species and was shown in pea to be mediated by a shoot-to-root signal that is RAMOSUS2 (RMS2)-dependent. The chemical nature of this feedback signal has been extensively discussed. Here, we demonstrate that the RMS2 protein belongs to the small family of auxin receptors and confirm that it behaves as an auxin receptor. Strigolactones decrease stem auxin levels by rapidly repressing transcript levels of auxin biosynthesis genes, thereby forming a long-distance feedback loop between auxin and strigolactones for the precise regulation of shoot branching in plants.
| Feedback signals are an essential component of dynamic biological systems to enable robustness and plasticity in development. While negative feedback can attenuate signals, positive feedback can amplify or prolong them [1,2]. Several positive and negative feedback mechanisms are likely involved in the control of shoot branching, a sequential and life-long regulated process in plants. Shoot branching patterns are derived from axillary bud activation and branch growth. Axillary buds, located in the axils of most leaves, integrate a multitude of external and endogenous signals resulting in the decision to grow or remain dormant [3,4]. Negative feedback loops can limit excessive branching that may be detrimental to the plant and positive feedback loops can stimulate sustained bud outgrowth or maintain dormancy.
Strigolactones (SL) play a major role in regulating shoot branching and also act as rhizospheric signals [5–7]. Homeostasis of most plant hormones is achieved by feedback control of the biosynthetic pathway by the end-product, via the hormone signaling pathway [8–11]. Evidence for such negative feedback control of SL biosynthesis has been observed in several species, as highly branched SL-defective mutants possess increased transcript levels of SL biosynthesis genes and SL application can reduce these transcript levels [5,12–18]. In contrast with other hormones where this negative feedback is mediated by components of the hormone signaling pathway, for SL at least some of the feedback appears to be indirect [15].
The pea SL synthesis genes RAMOSUS1 (RMS1) and RMS5 encode two members of the CAROTENOID CLEAVAGE DIOXYGENASE family (PsCCD8 and PsCCD7 respectively, MORE AXILLARY GROWTH4 (MAX4) and MAX3 in Arabidopsis) [19,20]. These CCDs act downstream of the DWARF27 (D27) isomerase and together they catalyse the synthesis of carlactone, a key intermediate in SL biosynthesis [21]. Downstream of the two CCDs, different enzymes including the cytochrome P450 (MAX1) and LATERAL BRANCHING OXIDOREDUCTASE (LBO) are involved in the synthesis of bioactive SL or SL-like compounds [22–25]. Carlactone-derived compounds with a butenolide ring (D ring) connected to a tricyclic lactone (ABC rings) via an enol-ether bridge are defined as canonical SLs. The pea RMS3 and RMS4 genes, required for SL response, encode the SL receptor (AtD14 in Arabidopsis) and an F-box protein (MAX2 in Arabidopsis), respectively [20,26]. The SL receptor hydrolyses SL to form a complex with the D-ring product. This complex undergoes a conformational change and binds to the MAX2/RMS4 F-box protein, a subunit of an Skp-Cullin-F-box (SCF) E3 ubiquitin ligase complex [26,27]. In the SL signalling pathway, the ubiquitination and proteasome-mediated degradation targets of this D14/SCFMAX2 complex include the SL repressor proteins D53 in rice and SMXL6-SMXL8 in Arabidopsis [28–31]. These proteins can function as transcriptional repressors by recruiting the corepressors TOPLESS and TOPLESS-RELATED (TPR) [29–31], although SMXL7 retains significant function when the TPL interaction domain is deleted [32]. There is good evidence that one transcriptional target for the SMXLs in the control of shoot branching is inhibition of transcription of the TEOSINTE BRANCHED1, CYCLOIDEA, PCF (TCP) transcription factor family member BRC1 [30,31]. Expression of BRC1, localized in axillary buds, is upregulated by SLs in some species [33,34]. In pea and Arabidopsis, the shoot branching and dwarf phenotypes of the brc1 mutant are less pronounced than those of SL deficient (max3/rms5, max4/rms1) and SL response (max2/rms4, Atd14/rms3) mutants, suggesting other systemic functions for SL [33,35–37]. In Arabidopsis, SL can repress the main stem polar auxin transport stream (PATS) via rapid removal of the PIN FORMED1 (PIN1) auxin efflux protein from the basal plasma membrane of xylem parenchyma cells [38–42]. The SMXL6-SMXL8 proteins appear to be involved in the SL regulation of PIN1 accumulation at the plasma membrane by an unknown mechanism which is unlikely to be transcriptional [40].
In pea, grafting studies and hormone quantifications of highly branched rms mutants (rms1 to rms5) led to a model for shoot branching control involving two novel, long-distance, graft-transmissible, signals [43–46]: a root-to-shoot branching inhibitor, now identified as SL [5,6] and an unknown shoot-to-root feedback signal dependent on RMS2 [47]. This RMS2-dependent feedback signal was proposed to positively regulate SL synthesis gene transcript levels and to negatively regulate xylem-sap cytokinin (X-CK) export from roots, as SL synthesis and signalling mutants all possess greatly increased RMS1 transcript levels and reduced X-CK levels, whereas rms2 mutants have low levels of RMS1 and RMS5 transcripts and increased X-CK export [20,44]. The additive branching phenotype of rms1 rms2 double mutants in comparison with single mutants supported this model where RMS1 and RMS2 controlled two different long-distance signals [48]. Based on grafting studies demonstrating movement of the RMS2-dependent feedback signal in a shoot-to-root direction, the feedback control of SL was proposed to be mostly indirect because SL can only move in a root-to-shoot direction [15,43,46,49]. Feedback regulation of SL biosynthesis gene transcripts was also found to occur in SL mutants of Arabidopsis [15], rice [17,18,50], petunia [16,51,52], maize [13] and the moss Physcomitralla patens [12]. Application of the synthetic SL, GR24, can down-regulate the transcript levels of SL biosynthesis genes [5,12,13]. The Arabidopsis max1 to max4 mutants also display a strong reduction in X-CKs, with reciprocal grafting experiments between WT and max2 (rms4) indicating that X-CK exported from roots is mostly shoot-regulated, as shown in pea [49,53].
The chemical nature of the RMS2-dependent feedback signal has been extensively discussed [54,55]. In pea, two feedback signals were proposed in branching control: a branch-derived signal, very likely auxin, and the RMS2-dependent feedback signal [55]. Since the rms2 mutant has high IAA levels and is able to respond to IAA, it was also suggested that the RMS2-dependent feedback signal was auxin-independent, although auxin and the feedback signal share similar characteristics [54]. In pea and Arabidopsis, treatments that decrease stem auxin levels (decapitation, IAA polar transport inhibitors, defoliation etc.) also reduce transcript levels of the SL biosynthesis genes in the same tissues [15,44]. Auxin application to the decapitated stump or to intact plants results in an increase in transcript abundance of CCD7 (RMS5/MAX3/HTD1) and CCD8 (RMS1/MAX4/D10) in pea [20,44], Arabidopsis [56], rice [17,57] and maize [13]. Auxin is also known to rapidly reduce CK biosynthesis [58]. In particular, decapitation rapidly increases the transcript levels of CK biosynthesis genes in pea stem nodes [59] and X-CK levels in bean [60], whereas IAA applied to the decapitated stump prevents these augmentations. In Arabidopsis, it was proposed that IAA up-regulates the SL biosynthesis gene, CCD7 (AtMAX3) via the AXR1-dependent pathway in the basal inflorescence stem [15], and in the hypocotyl [56]. AXR1 functions in the activation of SCF complexes by rubinylation [61] and mutations in AXR1 confer auxin resistance [62]. In the basal stem of Arabidopsis axr1 max2 double mutants, MAX3 transcript levels are considerably reduced in comparison to max2, but not completely restored to WT levels [15]. These results are similar to analyses of RMS1 transcript levels in the epicotyl of rms1 rms2 double mutants [48] and altogether strongly suggest the involvement of auxin in feedback regulation of SL biosynthesis gene expression.
Here we show that the RMS2 gene encodes an F-box protein of the small family of auxin receptors including the TRANSPORT INHIBITOR RESPONSE1/AUXIN-SIGNALING F-BOX (TIR1/AFB) proteins, with RMS2 belonging to the AFB4/AFB5 clade. We demonstrate that transcript levels of IAA biosynthesis genes are rapidly down-regulated by SL application and propose a model whereby SL and IAA regulate each other’s metabolism, highlighting the importance of homeostatic systems in shoot branching control.
The RMS2 gene had been mapped previously to linkage group (LG) I of the pea genetic map in a large region containing the classical markers ENOD40, sym19 and PsU81288 [63–65]. These three markers were also found to be linked in Medicago truncatula (Mt) where chromosome 5 corresponds to pea LGI. We looked for candidate genes located in this region that were likely to play a role in hormone signaling, particularly auxin signaling, and plant architecture [54]. Taking advantage of the good conservation of synteny between Mt chromosome 5 and pea LGI, we identified pea genetic markers in the vicinity of these candidate genes and mapped them in an F2 pea mapping population of 528 individuals derived from a cross between K524 (rms2) and JI281 [66,67]. Three markers (FG5363261, AM161737, FG535768) corresponding to Medtr5g065010, Medtr5g065860, and Medtr5g065440, respectively, and located near Medtr5g065490, a putative auxin receptor of the TIR1/AFB family, were tightly linked to rms2 in pea (Fig 1A). The sequence of the pea orthologue of Medtr5g065490, PsCam045205, and of other pea homologues of the TIR1/AFB family of auxin receptors were obtained from the pea RNA-Seq gene atlas (http://bios.dijon.inra.fr/FATAL/cgi/pscam.cgi); [68]. PsCam045205 has been mapped on LGI using different pea mapping populations [69]. Phylogenetic analysis indicated that PsCam045205 belongs to the Arabidopsis AFB4/AFB5 clade. The pea Unigene set described in [68] represents most of the expressed genes of pea and was derived from several cDNA libraries. Therefore it is very likely that PsCam045205 is the only pea AFB homologue in this clade (Fig 1B). PsAFB4/5 was sequenced in the two available rms2 mutants and in their respective wild-type lines. Mutations were found for each rms2 mutant. The rms2-1 mutation (line K524 from Torsdag) leads to the replacement of glutamic acid by lysine at position 532 and the rms2-2 mutation (line W5951 from Parvus) leads to the replacement of glycine by arginine at position 117 (Fig 1C and S1 Fig). Both mutations affect amino acids located close to those residues forming the IAA binding pocket of the TIR1 homologue (S1 Fig, [70]). Taken together, these data demonstrate that RMS2 likely corresponds to PsAFB4/5.
An in vitro stem segment assay was used to investigate IAA responses in rms2 as it was previously shown that transcript levels of the SL biosynthesis gene RMS1 are increased in isolated stem segments treated with IAA [44]. Internode 4–5 of 16-d-old plants harvested from WTTérèse, rms1-10, rms2-1 and rms1-10 rms2-1 double mutant plants were treated with a 10 μM IAA solution for 3 h. Transcript levels of RMS1 and RMS5, together with the pea homologue of the rice D27 gene PsD27 were analyzed. Phylogenetic analysis indicated that PsD27 is in the same clade as the rice D27 and the two proteins share 59% identity. Increased transcript levels of all three SL biosynthesis genes were observed in WT and rms1-10 internodes treated with IAA compared to mock controls (S2 Fig). In contrast, the increase in RMS1, RMS5 and D27 transcript levels in response to IAA was either abolished or attenuated in mutants containing the rms2-1 mutation. These results suggest that transcript levels of SL biosynthesis genes are stimulated by IAA and that this induction is impaired in plants containing the rms2 mutation.
To investigate whether pea RMS2 and Arabidopsis AFB4/5 perform similar functions in shoot branching regulation, we analysed the branching phenotypes of single and double Arabidopsis afb4 and afb5 mutants, as well as max mutants, and examined SL biosynthesis gene transcript and auxin levels which are known to be altered in rms2 [66]. The pea rms2 mutants display increased shoot branching, particularly at basal nodes [71]. Single and double afb mutants had levels of rosette branching that were intermediate between WT and the highly branched max2-1 mutant (Fig 2A). The afb4-8 afb5-5 double mutant had a similar number of axillary branches as the SL deficient max4-1 mutant. Interestingly, the classification of axillary branches into three groups according to their length showed that afb4-8, afb5-5 and afb4-8 afb5-5 mutants possess a larger proportion of small branches (< 5 mm) in comparison to WT and the max mutants. This particular branching phenotype is also found in the pea rms2 mutants, which displays long basal branches and small branches at upper nodes, whereas SL mutants have long branches at most nodes [71].
In pea, RMS2 was proposed to play a role in the feedback regulation of RMS1 (PsCCD8) expression because RMS1 transcript levels are greatly up-regulated in all rms mutants except for rms2 [44]. To test if afb mutants have similarly low levels of SL biosynthesis genes, MAX3 (AtCCD7) expression was quantified in adult basal stems of the afb and max mutants. MAX3 transcript levels were increased in max2-1 and max4-1, but were similar or lower than WT in the single and double afb mutants (Fig 2B). Thus, both rms2 and afb mutants possess reduced SL biosynthetic gene transcript abundance. Another physiological trait of pea rms2 mutants is the increased stem level (up to 5 fold higher than WT) of the predominant auxin indole-3-acetic acid (IAA) [66]. IAA levels were also found to be higher in afb4-8 and afb5-5 single mutants and up to 4 fold higher than WT in afb4-8 afb5-5 double mutants (Fig 2C).
It was previously reported that the Arabidopsis afb5 mutant, and to a lesser extent afb4, show specific resistance to the herbicidal auxin picloram (4-amino-3,5,6-trichloropicolinic acid) [72,73]. The resistance of the rms2 pea mutants to this synthetic picolinate auxin was therefore investigated. A foliar spray of 0.83 mM picloram was applied to 20-d old plants of WTTérèse (Térèse background), rms2-1, and rms4-3 mutants. After 10 days, depigmentation was observed in all genotypes and severe auxin-related symptoms including stem curvature and foliar curling were observed in all genotypes except for rms2-1which exhibited limited foliar curling (Fig 3A). To quantify picloram resistance, the chlorophyll content was estimated with a Soil Plant Analysis Development (SPAD) chlorophyll meter in WTTérèse, rms1-10, rms2-1, rms4-3 mutants and rms1-10 rms2-1 double mutants 8 days after treatment with picloram (0.83 mM or 2.07 mM). A strong picloram dose-dependent decrease in chlorophyll content was observed for all genotypes except for rms2-1 and rms1-10 rms2-1 double mutants, which were resistant even at the higher dose (Fig 3B). The picloram resistance of the rms2-1 (Torsdag background) and rms2-2 (Parvus background) mutant alleles were also confirmed (S3 Fig). These results demonstrate that picloram resistance is conferred by the two pea rms2-1 and rms2-2 mutations, similar to that observed for the Arabidopsis afb5-1 mutant, and to a lesser extent afb4-8.
Auxin perception and signaling by TIR1/AFBs require the binding of TIR1/AFBs to ASK1 (ARABIDOPSIS SKP1 HOMOLOG1), a core component of the SCF complex [74]. Interaction between the Arabidopsis ASK1 and pea RMS2 proteins (from WT, rms2-1 and rms2-2 mutants) was therefore tested in a yeast two-hybrid (Y2H) system. ASK1 was shown to interact with both WT RMS2 and mutant rms2-1 proteins, but not rms2-2, likely due to the location of the rms2-2 mutation near the F-box domain (Fig 4 and S1 Fig). Thus, RMS2 can interact with ASK1 and can presumably form an SCF complex. To investigate whether pea RMS2 can function as an auxin co-receptor, the Y2H system was used to test for interactions between pea RMS2 proteins (from WT, rms2-1 and rms2-2 mutants) and Arabidopsis IAA7 and IAA3 proteins in the presence or absence of IAA. We chose these two Aux/IAA proteins because IAA7 is known to interact with Arabidopsis AFB5 and other auxin receptors, whereas IAA3 does not interact with AFB5 [75]. TIR1 interactions were assessed as a positive control. Similar to TIR1, RMS2 interacted with IAA7, even when IAA was not present. The addition of IAA appeared to increase the binding of both TIR1 and RMS2 to IAA7. In our experiment, some interaction was observed in the absence of IAA between IAA3 and TIR1 or RMS2, but not AFB5. This interaction was strongly enhanced in the presence of IAA. For both rms2-1 and rms2-2 mutant proteins, no interaction with IAA7 or IAA3 was detected in either presence or absence of auxin (Fig 4 and S4 Fig). The iaa7m protein has three substitutions in the degron sequence and did not interact with any of the AFBs. These results indicate that pea RMS2 can bind Aux/IAA proteins in an IAA-dependent manner, though no specificity for the IAA3 or IAA7 co-receptor partner was observed for proteins in the AFB4/5 clade.
If RMS2 encodes an auxin receptor, the best candidate for the shoot-to-root RMS2-dependent feedback signal is auxin [54]. IAA is well known for its role in repressing CK biosynthesis [58,59,76] and stimulating SL biosynthetic gene expression [15,20,44]. Previous physiological characterization of the rms branching mutants showed that rather than being depleted in IAA levels, they often contained elevated IAA levels [48,66,77]. Therefore, a model can be proposed where the lack of SL response in the rms SL-biosynthesis and signalling mutants stimulates the synthesis of IAA, which controls CK levels in the xylem sap and SL biosynthesis gene expression via RMS2 (and possibly via other TIR1/AFB proteins). To test this model, we investigated whether SL treatment can repress IAA levels using the pea SL rms mutants. The rms1-2 mutant (Torsdag background) was grown in a hydroponic system in the presence or absence of the synthetic SL analogue (±)-3'-Me-GR24. This analogue is more stable than the classical SL analog GR24 because of two methyl groups on the D-ring and strongly inhibits shoot branching in pea [78]. Analogues with this D-ring structure were shown to act via RMS3 with the same mechanism of perception as analogues with the canonical D-ring structure (present in natural SLs) with one methyl group at the 4′ position [26]. Levels of IAA were quantified in stem segments at upper, middle and basal nodes 6 h and 24 h after SL treatment. For all stem segments, rms1-2 had higher IAA levels than WT (Fig 5A). IAA levels were reduced in the different stem segments within 6 h of (±)-3'-Me-GR24 application and were significantly decreased by 24 h (Fig 5A).
In a second experiment, IAA levels were quantified in stem segments at upper and middle nodes after 24 h of (±)-3'-Me-GR24 application in rms1-2, rms2-1, rms3-2 and rms4-1 in the Torsdag background. IAA levels were not reduced by (±)-3'-Me-GR24 application in both rms3-2 and rms4-1 SL response mutants but were decreased significantly in rms1-2 and rms2-1 (Fig 5B). A small IAA increase was observed in rms4-1 upper stem segments after SL application, an opposite response to SLs regularly observed for max2/rms4 that is not yet well understood [14,26]. These results indicate that SLs can repress IAA levels in the stem via RMS3 and RMS4, and RMS2 is not required for the regulation of IAA levels by SLs. Furthermore, rms2 mutants can respond to SL (Fig 5B) to regulate both shoot branching and stem IAA content ([55]; S5 Fig). Therefore, the high IAA levels observed in the stems of SL biosynthesis (rms1, rms5) and response (rms3, rms4) mutants are at least in part due to impaired down-regulation of IAA biosynthesis in these mutants. The rms2 mutant also contains very high levels of IAA in stems, particularly at upper nodes ([48,66]; Fig 5B and S6 Fig). To confirm that the high IAA stem levels of rms2 is due to an impaired auxin response, rather than a lack of SL-mediated feedback suppression of IAA levels, IAA levels were quantified in the rms1-1 rms2-2 double mutant (Parvus background) and compared to WT and single mutants. For both basal and upper internodes, IAA levels were higher in rms1-1 rms2-2 than in WT and rms1-1 and rms2-2 single mutants (S6 Fig). This result indicates an additive effect of SL deficiency and a lack of auxin response on IAA content in the stem, similar to results reported with Arabidopsis axr1 and max1 mutants [79]. In Arabidopsis, a significant proportion of the auxin in stems is derived from active apices, including branches [79]. Accordingly, a strong gradient in IAA concentration along the Arabidopsis inflorescence stem is observed, with higher levels towards the stem base [32,80,81]. In contrast, in all our experiments on pea where different nodes along the stem were collected, the higher IAA levels were observed in the upper node below the apex indicating that the reduced IAA levels observed after SL application was not the result of reduced IAA export from active apices of axillary branches after SL application, at least in this upper node.
To investigate the possible mechanism(s) of SL-mediated regulation of IAA levels, and the kinetics of the response, expression levels of key auxin biosynthetic genes were analyzed. In Arabidopsis, the TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS1 (TAA1/TAR) enzymes convert tryptophan to indole-3-pyruvic acid (IPyA), which is then converted to IAA by the YUCCA (YUC) proteins [82,83]. The Pea RNA-Seq gene atlas (http://bios.dijon.inra.fr/FATAL/cgi/pscam.cgi; [68]) was used to select genes from these two families that were expressed in young shoots before flowering. Among the three TAR genes identified in pea, TAR2 (JN990989 = PsCam045859) was selected for analysis as it is widely expressed whereas TAR1 (JN990988 = PsCam038427) is specifically expressed in seeds and pods and TAR3 (JN990990 = PsCam017219) is expressed in roots, nodules and seeds [84]. Several pea YUC genes are expressed in young shoots [85]. In a preliminary experiment, transcript levels of TAR2, YUC1 and YUC2 were analyzed in the upper internode (node 6-node 7) of WTTorsdag and rms1-2 (Torsdag background) mutant plants grown hydroponically with and without SLs ((±)-3'-Me-GR24., 3 μM) for 7 days. The SL biosynthesis gene RMS5 (PsCCD7) was analyzed to confirm the effectiveness of the SL treatment. Transcript levels of the four genes followed the same pattern with high levels in rms1 compared to WT and a significant decrease in rms1 when grown with SLs (S7 Fig). TAR2 and YUC1 were chosen for further analysis together with RMS5.
Transcript levels of the selected genes were quantified after applying SLs for 0.5 h, 1 h, 2 h, 3 h, 4 h and 6 h to the SL-deficient mutant rms1-2 (Torsdag background). As expected, transcript levels of RMS5 were higher in rms1 controls than in the WT and decreased significantly 2 h after SL treatment (Fig 6A). TAR2 and YUC1 followed the same pattern as RMS5 with a significant decrease in TAR2 and YUC1 transcript levels after 30 min and 1 h, respectively (Fig 6B and 6C). Thus, SL treatment can significantly reduce TAR2 transcript levels in upper internodes as soon as 30 min after treatment, before any significant decreases in RMS5 transcript levels (after 2 h) and IAA levels (between 6 and 24 h) (Fig 5A). The reduction was not observed when using the rms4 SL-response mutant (S8 Fig). Together these results suggest that SLs repress IAA levels in the stem at least in part by a rapid down-regulation of transcript levels of IAA biosynthesis genes. A more detailed tissue- and/or cell-specific IAA quantification could detect whether there is a significant decrease in IAA levels at earlier time points [86].
In Arabidopsis, max mutants display increased auxin transport in the primary inflorescence stem and increased PIN1 protein accumulation at the basal plasma membrane of xylem parenchyma cells, which can be rapidly reduced by GR24 treatment [40,41]. It is well known that IAA levels and polar auxin transport (PAT) are highly interconnected. To investigate whether the SL-mediated decrease in IAA levels could be mediated by the effect of SL on PAT, we tested if SL could still elicit a reduction in IAA levels in NPA-treated plants where PAT is severely compromised.
15-d-old rms1 plants were treated with a lanolin ring around the stem of the oldest expanding internode with or without 0.1% NPA. Two days later, (±)-3'-Me-GR24 was supplied hydroponically via the roots for 24 h, after which time the internodes above and below the NPA treatment site were harvested for IAA quantification (Fig 7A). The 3 day NPA treatment induced a strong decrease in IAA levels in internodes below (54% reduction) but also above the NPA lanolin ring (30% reduction) possibly due to reduced auxin export into the stem from leaves and to the systemic effect of NPA [87]. Below the site of lanolin/NPA treatment, the effect of SL on NPA treated plants was similar to that on lanolin treated control plants (45% and 46% reduction, respectively), although the absolute reduction was lower. Above the site of lanolin/NPA treatment, the SL effect on NPA treated plants was still significant but relatively smaller (23% reduction) compared to the effect of SL in lanolin treated control plants (41% reduction). These results suggested that SLs can decrease IAA levels in stems independently of their effects on polar auxin transport.
In several species, SL application down-regulates transcript levels of SL biosynthesis genes in the shoot, indicative of negative feedback control on SL biosynthesis [5,13,14]. In pea, a major contributor to this feedback is the RMS2-dependent shoot-to-root signal. Our findings suggest RMS2 encodes the unique pea F-box protein of the AFB4/5 clade and yeast two- hybrid assays support the hypothesis that RMS2 functions as an auxin receptor (Figs 1 and 4). Together, this suggests the RMS2-dependent feedback signal is very likely auxin. Since all pea SL-defective mutants display high transcript levels of RMS1 and low X-CK, it was proposed that a non-response to SL activated this shoot-to-root feedback signal [44,53,55]. Our demonstration that SL represses IAA levels in stems strengthens the hypothesis that auxin is an intermediate for the feedback control of SL biosynthesis, with SL repressing auxin biosynthesis and very likely auxin export from source tissues, and auxin(s) stimulating SL biosynthesis (Fig 8).
Higher levels of stem IAA have been frequently observed in SL-defective mutants ([38,48,80,81,88], this work). Here we demonstrate that SLs decrease IAA levels in pea internodes with a significant reduction observed 24 h after SL feeding through the roots (Fig 5). When analyzing transcript levels of IAA biosynthesis genes in the internode below the apex, a significant reduction was observed within 30 min for TAR2 and 1 hour for YUC1 after SL application, whereas a significant decrease in the transcript levels of the SL biosynthesis gene RMS5 was observed after 2 hours (Fig 6). These results indicate that the synthetic SL analog used in our experiments, or some derivative, was transported from root to shoot within 30 min (at a distance of at least 40 cm). The only SL transporter that has been identified to date is the PLEIOTROPIC DRUG RESISTANCE1 from Petunia axillaris (PDR1), a SL cellular exporter, expressed in root cortex and shoot axils [89]. To explain the rapid root-to-shoot long distance SL transport observed here, xylem transport seems more likely. Grafting experiments have clearly demonstrated that a wild-type rootstock can inhibit the shoot branching of an SL-deficient scion and that SL can only move in a root-to-shoot direction [46]. Recent work with SL root applications and SL analysis in shoot tissues has shown that root-to-shoot SL transport is a highly structure- and stereospecific process [90]. Currently it is unclear which SLs (non-canonical SL vs canonical SLs) or SL-derived metabolites are transported from the root to inhibit shoot branching. Nevertheless, some SL-related molecule can rapidly move to the shoot and decrease the expression of IAA biosynthesis genes in the stem. However, we cannot rule out additional effects of SL on other parts of the auxin synthesis and breakdown pathway to regulate the pool of IAA in the stem. SL-mediated repression of IAA biosynthesis has previously been proposed as a mechanism by which SL attenuates shoot gravitropism and tiller/branch angle in rice and in Arabidopsis [91].
In Arabidopsis, the synthetic SL analogue GR24 was shown to reduce PAT in the main stem by removing PIN1 from the plasma membrane by a rapid (<30 min), cycloheximide-independent, clathrin-dependent mechanism [40]. As IAA and PAT influence each other [92], we tested whether the decrease in IAA levels observed after SL application was due to the effect of SL on PAT. SL could still elicit reductions in IAA levels in NPA treated plants, suggesting that SLs also have an effect on auxin biosynthesis in stems that is mostly independent of PAT (Fig 7). The SL-triggered changes in stem IAA levels are therefore likely due to a combination of changes in auxin synthesis and changes in IAA export from young expanding leaves, which are a major source of stem auxin.
Using different mutants, we demonstrated that SLs reduce IAA levels via RMS3 and RMS4 but this action does not require RMS2 (Fig 5). The TCP transcription factor BRC1 is unlikely to be involved in this SL-mediated repression of auxin biosynthesis and RMS1 expression [33]. Indeed, the high tillering rice fc1 / Osbrc1 mutant has normal D10/OsCCD8 transcript levels [18] and normal IAA levels in shoot apices [17] in comparison to all SL-defective mutants. Similarly, the highly branched pea Psbrc1 mutant has WT or lower RMS1 transcript levels and in Psbrc1 the profiles and absolute amounts of X-CKs are not significantly different from WT, whereas X-CK levels are very low in SL biosynthesis and response mutants [33,49,53]. In addition, Arabidopsis brc1 mutants have normal stem auxin transport, further separating the activity of SL in modulating auxin homeostasis and BRC1 expression [35,37].
In Arabidopsis, the SMXL6-8 proteins activate shoot branching, with SLs stimulating their ubiquitination and subsequent degradation via the proteasome [30,31]. In Arabidopsis seedlings, MAX4 transcript abundance is higher than WT in max3/Atccd7 and max2 but lower than WT in smxl6/7/8 triple mutants, max3smxl6/7/8, and max2smxl6/7/8 quadruple mutant plants [30]. Therefore, when SMXL6-8 proteins are non-functional, the feedback signal is suppressed even in SL-defective mutant backgrounds where it is usually triggered. Furthermore, in 35S:SMXL6D-GFP Arabidopsis plants bearing a dominant mutation conferring SL resistance equivalent to the rice d53 mutant, MAX4 transcript levels are higher in comparison to WT and in 35S:SMXL6-GFP transgenic plants [28,29]. Together, these results suggest the SMXL6-8 proteins are necessary for activating the feedback signal and the down-stream stimulation of SL-biosynthesis gene transcript levels. SMXL6-8 are also known to mediate the effects of SL on auxin transport [32]. Further studies with IAA quantifications in the recessive and dominant mutants for the SMXL6-8 proteins are necessary to confirm the involvement of these proteins in the regulation of IAA levels by SLs.
As RMS1 transcript levels are high in all SL-defective mutants and low in rms2 epicotyls compared to WT, it was proposed that the RMS2 gene is involved in the feedback control of SL biosynthesis [44,48,49,53]. However, double rms mutants with rms2 have intermediate levels of RMS1 transcripts and X-CK root export [20,43–45,49,53], suggesting rms2 does not completely prevent feedback upregulation of RMS1 expression and repression of X-CK [44,53]. A simple explanation for this is that other TIR1/AFB auxin receptors also play a role in mediating auxin-dependent feedback. In our experiments using Arabidopsis afb mutants, afb4 afb5 double mutants have increased branching compared to WT. Yet, triple and quadruple mutants of the TIR1/AFB2 clade display a clear highly branched and dwarf phenotype in Arabidopsis [93] and downregulation of OsTIR1 and OsAFB2 results in higher tiller numbers in rice [94]. Higher shoot branching and low MAX3/MAX4 expression is also observed in auxin-related mutants such as axr1 or the semi-dominant bodenlos (bdl) mutant containing a mutation that stabilizes the IAA12 protein [15]. Together these results suggest that members of the AFB4/AFB5 clade are unlikely to be the only auxin receptors specifically involved in mediating the auxin-dependent negative feedback in shoot branching regulation. It may also explain why the rms2 mutant responds to IAA in some assays, e.g. RMS1 transcript levels are increased after IAA application but do not attain the high levels observed in the SL-response rms3 and rms4 mutants [44].
The elevated IAA levels observed in SL defective mutants may also result from auxin exported from the branches and entering the main stem and this auxin likely also participates in feedback up-regulation of SL biosynthesis genes [32,55,80]. Indeed, a strong IAA concentration gradient which increases towards the stem base is observed in Arabidopsis, particularly in highly branched mutants [32,80,81]. This gradient is thought to be due to the increased number of active apices exporting IAA into the basal stem combined with their increased auxin transport activity [32,80]. Interestingly, an opposite IAA gradient was found in pea, with higher IAA concentrations in the upper internode; IAA levels in this internode are even higher than those in the apical part of the shoot [77]. This pattern is also observed in highly branched mutants ([48,77]; this work). The origin of these opposing IAA gradients is not clear and requires further investigation. They may be due in part to different experimental systems and/or developmental stages, as shoot branching is generally analyzed before floral transition in pea, versus after floral transition in Arabidopsis [95].
The higher IAA level in the upper internode of rms mutants suggests that, at least in pea, it is the non-response to SL (possibly mediated by high levels of SMXL proteins), rather than the high shoot branching per se, which activates the biosynthesis of auxin and feedback up-regulation of SL biosynthesis gene transcript levels. This hypothesis was already tested using the suppressed axillary meristem1 (sax1) mutation which inhibits the formation of axillary meristem at most nodes [53,96]. Grafting experiments showed that X-CK was similarly reduced when WT rootstocks were grafted to either rms4 single mutant or rms4 sax1 double mutant scions, despite having strongly reduced branching in rms4 sax1 shoots due to the absence of most axillary meristems [53]. Another example where high IAA levels are not related to high shoot branching is the maintenance of high IAA levels in rms2 shoots even when branching is suppressed by grafting to WT rootstocks [66]. This observation supports the hypothesis that the high IAA content of rms2 is due to its non-response to auxin. A shoot-derived signal, very likely IAA, has also been proposed in pea, as a transient increase in RMS1 transcript levels occurs in rms2 epicotyls only when a strong basal branch is growing above it [55]. These data suggest that rms2 can respond to this lateral branch-derived signal. Interestingly, the rms4 mutant showed RMS1 transcript levels similar to WT in the internode below the apex, where IAA levels are also comparable to WT (Fig 5B), whereas constitutive elevated RMS1 transcript levels in rms4 epicotyls are observed [55]. These data demonstrate the dynamic spatio-temporal control of auxin-SL feedback which necessitates careful analysis of phytohormone distribution [86].
Low SL levels in shoots have hampered efforts to perform direct quantifications in aerial tissues. The low RMS1 expression in rms2 epicotyls suggests that the mutant is highly branched in part because of reduced SL production [55]. However, quantifications of the three main SLs found in pea (orobanchol, fabacyl acetate, orobanchyl acetate) in root exudates and root tissues of rms4 and rms2 under low phosphate conditions are similar to those in WT [97]. Nevertheless, branching in rms1 shoots can be completely rescued by WT rootstocks, but only partially by rms2 rootstocks [48]. This suggests that rms2 mutants are highly branched at least in part due to reduced SL production. The strong additive branching phenotype of the rms1 rms2 double mutant compared to the single mutants [48] indicates that another component, possibly the high CK levels measured in both rms2 shoot tissues and X-CK, also contributes to the increased shoot branching of rms2. In Petunia, comparison of DAD1/PhCCD8 and DAD3/PhCCD7 transcript levels in dad roots and stems shows elevated PhCCD7/8 gene expression in stems relative to WT, but not in roots [16,52]. Thus, there is likely to be complex and precise regulation of SL biosynthetic gene expression and SL levels in different tissues. Another layer of regulatory complexity is added by the homeostastic regulatory mechanisms controlling auxin levels, as demonstrated by the recent discovery of the DIOXYGENASE FOR AUXIN OXIDATION1 (DAO1) which catalyzes the oxidation of IAA into 2-oxindole-3-acetic acid (oxIAA) [98–100].
Here we demonstrated that RMS2 encodes a member of the TIR1/AFB family of auxin receptors of the AFB4/5 subclade. We confirmed the role of auxin in mediating the negative feedback of SL and propose a model where auxin and SLs regulate each other’s metabolism, and the distribution of auxin is dynamically controlled by growing branches and polar transport. We cannot rule out that other mechanisms are involved in the negative feedback of SL. A direct SL-feedback mechanism may be identified when the function and targets of the SMXL proteins are better understood. Moreover, the selectivity of AFB4/AFB5 in picolinate auxin perception, conserved between Arabidopsis and pea, and the maintenance during evolution of the AFB4/5 sub-clade, are quite intriguing; a specific endogenous ligand for AFB4/5 may yet be discovered [101]. Better clarity of the SL feedback mechanism(s) at play may be gained through further analysis of such a ligand and the affinities of particular IAAs to AFB4/5.
The pea (Pisum sativum) branching rms1-1 (WL5237), and rms2-2 (WL5951) mutants were derived from the tall line Parvus. The rms1-1 rms2-2 double mutant was kindly given by Christine Beveridge (University of Queensland, AU). The rms2-1 mutant (K524), the rms3-2 mutant (K564), and the rms4-1 mutant (K164) were obtained in the tall line Torsdag. The rms1-10 (M3T-884) and rms4-3 (M3T-946) were obtained from the dwarf cv Térèse. All these mutants were described in [66,71,102]. The rms1-2 (Torsdag) mutant line was obtained by backcrossing the rms1-2 allele from the line WL5147 derived from Weitor; [48] into the WT line Torsdag three times. The rms2-1 mutant in Térèse background was obtained by backcrossing the rms2-1 allele (from the K524 line in Torsdag background) into the WT line Térèse seven times. The mapping population was an F2 of 528 individuals derived from the cross (K524 x JI281). The parental line JI281 has been used previously to generate a molecular map of pea [67] and was obtained from JIC Norwich-UK (http://www.jic.bbsrc.ac.uk/germplas/pisum/index.htm).
All Arabidopsis (Arabidopsis thaliana) lines used were in the Col-0 background and mutant lines for max2-1 [103], max4-1 [19], tir1-1 [104], afb2-3 [105], afb4-8, and afb5-5 [73] were previously published.
Pea plants were grown in glasshouse (23°C day/ 15°C night) under a 16-h photoperiod (the natural daylength was extended or supplemented during the day when necessary using sodium lamps) in pots filled with clay pellets, peat, and soil (1:1:1) supplied regularly with nutrient solution (140 mg/l N, 76 mg/l P2O5, 231 mg/l K2O, 146 mg/l CuO, 15 mg/l MgO; 0.82 mg/l K/(Ca + Mg); 3.64 mg/l NO3-/NH4+, diluted 200 times in water). Nodes were numbered acropetally from the first scale leaf as node 1. Arabidopsis plants were cultivated in a glasshouse or growth chamber (20°C day and night) under a 16-h photoperiod and humidity 70%.
The F2 population of 528 individuals were phenotyped for RMS2 and genotyped for molecular markers designed on the basis of the conservation of synteny between the pea and Medicago truncatula genomes (http://jcvi.org/medicago/). Putative pea orthologues of M. truncatula genes located on chromosome 5 in the vicinity of genes involved in hormone signaling were identified in silico in the transcriptome databases of NCBI (expressed sequence tags, transcriptome shotgun assembly). In order to identify polymorphisms, the corresponding genomic sequences were PCR amplified and sequenced in both parents of the mapping population. The mapping population was then genotyped using CAPS (cleaved amplified polymorphic sequences) assays or sequencing. Primers and polymorphisms used for these markers are given in Table 1. For sequencing PsAFB4/5 in WT and mutant lines, the primers used are given in Table 2.
Pea hydroponic culture was done as described in [78] using 33L-polyvinyl chloride opaque pots. The hydroponic culture solution was continuously aerated by an aquarium pump and was replaced weekly. Acetone or (±)-3′-Me-GR24 (dissolved in acetone) (kindly provided by F.D. Boyer, Institut de Chimie des Substances Naturelles) were added to the hydroponic culture solution to give a final concentration of 0 or 3 μM of (±)-3′-Me-GR24 and 0.01% acetone. (±)-3′-Me-GR24 corresponds to compound 23 in [78].
After 2 weeks, plants were sprayed with control solution (water, 4% ethanol) or with a solution of picloram (4-amino-3,5,6-trichloro-2-pyridinecarboxylic acid) at 0.83 mM (200 g/ha) or 2.07 mM (500 g/ha). Chlorophyll content was estimated with a SPAD (Soil Plant Analysis Development) chlorophyll meter (Minolta, SPAD-502 model, Tokyo, Japan) after 8 days on the stipule at node 6 (2 repetitions per stipule on 8–12 plants).
The in vitro IAA treatment of isolated internodes was adapted from [106]. Internodes 4–5 was harvested from 16-d-old plants and incubated in buffer or buffer supplemented with IAA (10 μM). After 3 h, the internodes were collected for RNA extraction (3 biological repeats of 6 internodes).
RNA extraction and cDNA synthesis were adapted from [33]. Total RNA was isolated from 8 to 10 pea internodes or 10 to 15 Arabidopsis basal stem using TRIZOL reagent (Invitrogen) following the manufacturer’s protocol. DNase treatment was performed to remove DNA using the Qiagen RNase-Free DNase Set (79254) and the RNeasy Mini Kit (74904) and eluted in 50 mL of RNase-free water. RNA was quantified using NanoDrop 1000 and migrated on gels to check RNA non-degradation. The absence of contamination with genomic DNA was checked using 35 cycles of PCR with RMS1 primers (5′-GGA ATG GTC CGG GCATGT G-3′ and 5′-TGA GAC TCG ATC TGC CGG TGA-3′). Total cDNA was synthesized from 2 μg of total RNA using 50 units of RevertAid H Moloney murine leukemia virus reverse transcriptase in 30 μL following the manufacturer’s instructions with poly(T)18 primer. cDNA was diluted 10 times before subsequent analysis.
Quantitative reverse transcription-PCR analyses were adapted from [33]. They were performed using SsoAdvanced Universal SYBR Green SuperMix (Biorad). Cycling conditions for amplification were 95°C for 10 min, 50 cycles of 95°C for 5 s, 62°C for 5 s, and 72°C for 15 s, followed by 0.1°C s–1 ramping up to 95°C for fusion curve characterization. Three biological repeats were analyzed in duplicate. To calculate relative transcript levels, the comparative cycle method based on non-equal efficiencies was used [107]. Transcript levels for the different genes were expressed relative to the expression of the PsACTIN gene for pea and of the AtAPT gene for Arabidopsis.
For qPCR in Arabidopsis, the following oligonucleotides were used: AtAPT, 5′-CGG GGA TTT TAA GTG GAA CA-3′ and 5′-GAG ACA TTT TGC GTG GGA TT-3′; AtMAX3, 5′-TCG TTG GTG AGC CCA TGT TTG TC-3′ and 5′-TCT CCA CCG AAA CCG CAT ACT C-3′ [108]. For qPCR in pea, the following oligonucleotides were used: PsACTIN, 5′-GTG TCT GGA TTG GAG GAT-3′ and 5′-GGC CAC GCT CAT CAT ATT-3′; PsRMS5, 5′-TGA CCG ACG GTT GTG ATT TGG-3′ and 5′-GCG GCA TCT TAA AGA CTC CGT AC-3′; PsYUC1, 5′-TTG CTA CCG GTG AAA ATG CTG A-3′ and 5′-CAT GAA AAT GTT CCA TAC CAT GAA TC-3′; PsYUC2, 5′- AGA GAA TGC CGA GGC TGT TGT G-3′ and 5′-AAG TTC CAT TCC AGA ATT TCC ACA TCC AA-3′ [85]; PsTAR2, 5′- TGG TGA ACC GTG GTG CAT TG-3′ and 5′- GCT GGT TGA GGT TCC AAC ACC TG-3′ [84].
IAA was extracted from 100 mg of fresh powder per sample with 0.8 mL of acetone/water/acetic acid (80/19/1 v:v:v). Indole-3-acetic acid stable labelled isotopes were prepared and used as internal standards (2 ng/sample) as described in [109]. The extract was vigorously shaken for 1 min, sonicated for 1 min at 25 Hz, shaken for 10 minutes at 4°C in a Thermomixer (Eppendorf®, and then centrifuged (8,000g, 4°C, 10 min.). The supernatants were collected, and the pellets were re-extracted twice with 0.4 mL of the same extraction solution, then vigorously shaken (1 min) and sonicated (1 min; 25 Hz). After the centrifugations, the three supernatants were pooled and dried (Final Volume 1.6 mL).
Each dry extract was dissolved in 100 μL of acetonitrile/water (50/50 v/v), filtered, and analyzed using a Waters Acquity ultra performance liquid chromatograph coupled to a Waters Xevo Triple quadrupole mass spectrometer TQS (UPLC-ESI-MS/MS). The compounds were separated on a reverse-phase column (Uptisphere C18 UP3HDO, 100*2.1 mm*3μm particle size; Interchim, France) using a flow rate of 0.4 mL min-1 and a binary gradient: (A) acetic acid 0.1% in water (v/v) and (B) acetonitrile with 0.1% acetic acid. The column temperature was 40°C. We used the following binary gradient (time, % A): (0 min., 98%), (3 min., 70%), (7.5 min., 50%), (8.5 min., 5%), (9.6 min., 0%), (13.2 min., 98%), (15.7 min., 98%). Mass spectrometry was conducted in electrospray and Multiple Reaction Monitoring scanning mode (MRM mode), in positive ion mode. Relevant instrumental parameters were set as follows: capillary 1.5 kV (negative mode), source block and desolvation gas temperatures 130°C and 500°C, respectively. Nitrogen was used to assist the cone and desolvation (150 L h-1 and 800 L h-1, respectively), argon was used as the collision gas at a flow of 0.18 mL min-1.
Y2H assays were carried out as in [75,110]. The plasmids pGILDA-TIR1, pB42AD-IAA7, pB42AD-IAA3, and pB42AD-ASK1 were described previously [75,110]. The mutant iaa7 construct with three substituted residues in the degron was produced by site directed mutagenesis of the pB42AD-IAA7 plasmid with the primers 5′-GCT AAA GCA CAA GTG GTG AGA TGG TCA TCT GTG AGG AAC TAC AGG A-3′ and 5′-TCC TGT AGT TCC TCA CAG ATG ACC ATC TCA CCA CTT GTG CTT TAG C-3′. WT and mutant RMS2 cDNA sequences were amplified using primers 5′-GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT CAT GAG AGA AAA CCA TCC TCC-3′ (start codon in bold and attB1 site underlined) and 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC TCA CTA CTG CAG AAT GGT AAC AT-3′ (STOP codon in bold and attB2 site underlined) and recombined into pDONR207 using BP Clonase (Invitrogen) then recombined into a Gateway-compatible version of pGILDA using LR Clonase II (Invitrogen). Site-directed mutagenesis of pDONR207 containing RMS2 coding sequence was performed using the QuickChange II XL Site Directed Mutagenesis kit (Stratagene) and primers 5′-CCT AAT TTG CAG AAA CTT AAA ATC AGG GAC AGT CCC TTC GGG G-3′ and 5′-ACA TCA AGT CGG TTA CCG TCA AGA GAA AAC CTA GGT TTG CGG ATT-3′ to obtain the rms2-1 and rms2-2 mutant sequences, respectively. AtAFB5 was amplified using the primers 5′-CAC CAT GAC ACA AGA TCG CTC AGA AAT-3′ and 5′-TAA AAT CGT GAC GAA CTT TGG TG-5′ and cloned into pENTR D/TOPO (Invitrogen), and a Myc tag added by ligating a double-stranded oligo (5′-CGC GAA CAG AAA CTG ATC TCT GAA GAA GAT CTG TAG-3′ plus 5′-CGC GCT ACA GAT CTT CTT CAG AGA TCA GTT TCT GTT-3′) into the AscI restriction site. The resulting entry clone and a Myc tag only control were recombined into the same pGILDA-derived destination vector to produce the pGILDA-AtAFB5-Myc and pGILDA-Myc (control) plasmids, respectively. Yeast strain EGY48 [p8Op:lacZ] was co-transformed with one pGILDA plasmid and one pB42AD plasmid and transformants selected on a medium lacking uracil, histidine, and tryptophan. Independent transformants were cultured and dilutions spotted on SD/Gal/Raf/X-Gal plates with or without IAA. The lexA-RMS2, lexA-rms2-1 and lexA-rms2-2 proteins were expressed to similar levels based on detection on a Western blot using an anti-lexA antibody (Millipore, 06–719) (S4 Fig).
Statistics were performed with the software R and MEGA7 was used for phylogenetic trees [111].
|
10.1371/journal.ppat.1000718 | Immature Dengue Virus: A Veiled Pathogen? | Cells infected with dengue virus release a high proportion of immature prM-containing virions. In accordance, substantial levels of prM antibodies are found in sera of infected humans. Furthermore, it has been recently described that the rates of prM antibody responses are significantly higher in patients with secondary infection compared to those with primary infection. This suggests that immature dengue virus may play a role in disease pathogenesis. Interestingly, however, numerous functional studies have revealed that immature particles lack the ability to infect cells. In this report, we show that fully immature dengue particles become highly infectious upon interaction with prM antibodies. We demonstrate that prM antibodies facilitate efficient binding and cell entry of immature particles into Fc-receptor-expressing cells. In addition, enzymatic activity of furin is critical to render the internalized immature virus infectious. Together, these data suggest that during a secondary infection or primary infection of infants born to dengue-immune mothers, immature particles have the potential to be highly infectious and hence may contribute to the development of severe disease.
| Dengue virus represents a major emerging arboviral pathogen circulating in the (sub)tropical regions of the world, putting 2.5 billion people at risk of infection. Each of the four circulating serotypes can cause disease ranging from febrile illness to devastating manifestations including dengue hemorrhagic fever and dengue shock syndrome. Severe illness is observed in individuals experiencing a re-infection with a heterologous dengue virus serotype and in infants born to dengue-immune mothers, presumably due to antibody-dependent enhancement of infection. Interestingly, it has been recently reported that patients experiencing a secondary infection have elevated levels of antibodies directed against the prM protein of immature dengue virus particles. Although it is known that cells infected with dengue virus release substantial amounts of prM-containing virions, numerous functional studies have demonstrated that immature particles lack the ability to infect cells. Herein, we show that essentially non-infectious fully immature dengue virions become virtually as infectious as wild type virus particles in the presence of prM antibodies. Anti-prM antibodies facilitate efficient binding and entry of immature dengue virus into cells carrying Fc-receptors. Furthermore, furin activity in target cells is critical for triggering infectivity of immature virus. These data indicate that immature dengue virus has the potential to be highly infectious and hence may contribute to disease pathogenesis.
| Dengue virus (DENV) represents a major emerging arthropod-borne pathogen. There are four distinct serotypes of DENV which, according to WHO estimates, infect about 50-100 million individuals annually, mostly in the (sub)tropical regions of the world. While most DENV infections are asymptomatic or result in self-limited dengue fever (DF), an increasing number of patients present more severe, potentially fatal clinical manifestations, such as dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). It is well established that a major risk factor for the development of DHF/DSS is secondary infection with a heterotypic virus serotype [1]–[3]. Also primary infection of infants born to dengue-immune mothers may lead to severe disease [1],[4],[5]. These observations have led to the hypothesis of antibody-dependent enhancement (ADE) of infection [3],[6],[7]. Increased disease severity appears to correlate with high circulating virus titers [3], [8]–[11], suggesting that antibodies directly influence the infectious properties of the virus. The molecular mechanisms by which antibodies enhance DENV infection however remain elusive.
DENV, as well as other major human pathogens like West Nile virus (WNV), yellow fever virus, and tick-borne encephalitis (TBEV) belong to the Flavivirus genus within the family Flaviviridae. Flaviviruses enter cells via clathrin-mediated endocytosis and fuse from within acidic endosomes, through which the viral genome gains access to the target cell cytoplasm[12]. Following RNA replication and protein translation, immature virions, which contain heterodimers of the transmembrane proteins E and a precursor form of M (prM), are assembled within the ER. Subsequently, the particles mature by passing through the Golgi and trans-Golgi network (TGN) [13]. In the acidic environment of the TGN, the virion undergoes a conformational change and the cellular endoprotease furin cleaves prM into M and a peptide (“pr”) that remains associated with the virion [14]. Upon release, the pr peptide dissociates from the virion, resulting in the formation of mature progeny virions.
Cells infected with DENV secrete high levels (∼30%) of prM-containing immature particles [15],[16] suggesting that cleavage of prM to M is not efficient. These DENV particles are released from infected cells as fully immature prM-containing particles and partially immature particles containing both prM and M proteins in the viral membrane [17]. Extensive functional analyses have revealed that fully immature flaviviruses lack the ability to infect cells, as the presence of uncleaved prM in the virion blocks the E glycoprotein from undergoing the pH-induced conformational changes that are required for membrane fusion [16], [18]–[22]. Although immature particles are therefore generally considered as irrelevant by-products of infected cells, the rates of prM antibody responses are significantly higher in patients with secondary infection compared to those with primary infection [23]. Furthermore, previous reports show that prM antibodies can enhance DENV infection. Enhancement of infection was observed for wild-type virus [24],[25], presumably due to the presence of uncleaved prM in these preparations, and with DENV particles containing high levels of prM generated from cells treated with chloroquine [26]. It is thus quite puzzling if indeed the presence of prM obstructs DENV infectivity, how immature particles contribute to disease pathogenesis and what role do anti-prM antibodies play in the enhancement of infection? The present study addresses these questions.
First, we investigated if prM antibodies are able to render fully immature DENV infectious. To this end, immature DENV-2 strain 16681 particles were produced in furin-deficient LoVo cells. We have used this procedure before and showed that LoVo-derived particles have an average content of 94%±9% prM [16]. Furthermore, we demonstrated that the specific infectivity of LoVo-derived fully immature DENV is at least 10,000-fold reduced compared to that of wild-type virus on cells highly permissive to infection [16]. The infectious properties of fully immature DENV virions were determined in Fc-receptor-expressing K562 cells in the absence or presence of increasing concentrations of the 70–21 antibody. This is an IgG2a antibody that has been isolated from DENV-infected mice and is mapped to amino acids 53-67 of prM [25]. Antibodies recognizing this epitope are abundantly present in sera of DHF/DSS patients [16],[27]. K562 cells were infected with DENV at a multiplicity of 100 genome-containing particles per cell (MOG 100). The number of genome-containing particles (GCP) was determined by quantitative PCR analysis of reverse-transcribed viral RNA [15]. At 24–48 hours post-infection (hpi), cells were fixed and prepared for flow-cytometric analysis to determine the number of infected cells, measured on the basis of dengue E protein expression. We observed that 43 hpi is optimal for read-out as it represents a single round of infection together with a high mean fluorescence intensity per infected cell (Fig. S1).
In agreement with our previous study [15],[16], we observed that fully immature DENV particles are essentially non-infectious as the number of E-positive cells did not exceed the limit of detection (Fig. 1A). Remarkably however, substantial numbers of E-positive cells were observed upon infection of cells with fully immature particles opsonized with the anti-prM antibody (Fig. 1A). Subsequent titration of the cell supernatants at 43 hpi revealed that opsonization of immature DENV with anti-prM antibody dramatically enhanced (up to 30,000-fold) virus particle production (Fig. 1B). The results show that prM antibodies render essentially non-infectious immature DENV nearly as infectious as wild-type virus (Fig. 1B). Enhancement of immature DENV infectivity was seen in a broad antibody concentration range, even at conditions of high antibody excess.
To ensure that the observed high level of enhancement is not restricted to a single antibody we performed additional experiments with the murine IgG2a prM antibody 2H2 [28]. The results show that 2H2 stimulated the infectious properties of fully immature particles up to 1,000 fold (Fig. S2A). Although, this antibody has previously been shown to enhance DENV infectivity [26], the power of enhancement observed here is striking and demonstrates that prM antibodies render essentially non-infectious fully immature DENV highly infectious.
Subsequently, we investigated the enhancing properties of both prM antibodies in Fc-receptor-bearing human monocytic U937 cells and observed that the antibodies again significantly stimulate the infectivity of immature particles (Fig. 1C, S2B). Thereafter, we studied the infectious properties of immature DENV particles in primary human PBMCs, cells which are known to be involved in dengue pathogenesis. The results show that, also under these conditions, prM antibodies render fully immature particles infectious (Fig. 1D, S2C).
To better understand the mechanism by which prM antibodies trigger infectivity of immature DENV, we analyzed the distinct steps in the cell entry pathway of the virus. First, the binding of immature virions to K562 cells was determined by quantitative-PCR. In order to determine the number of bound GCP per cell, the amount of virus added per cell was increased 10-fold compared to the concentration used in the infectivity experiments. The results show that antibody-opsonized immature DENV binds approximately 30-fold more efficiently to cells than immature particles in the absence of antibody (Fig. 2A). Indeed, immature particles opsonized with anti-prM bound almost as efficiently to cells as wild-type DENV in the absence of antibody. Moreover, immature DENV particles failed to interact efficiently with baby hamster kidney cells (BHK-15), cells which are highly permissive for dengue infection (data not shown) suggesting that the observed lack of infectivity is partially related to the poor binding efficiency of immature particles to cells. It is likely that binding of virus-antibody complexes is mediated by direct interaction of the antibody with the Fc-receptor expressed on the cell surface. Indeed, treatment of cells with an anti-CD32 antibody to block FcγII-receptor interaction, or opsonization of particles with mAb70-21 F(ab')2 fragments severely reduced virus particle production upon infection of K562 cells with opsonized immature virions, whereas it had no effect on infection with wild-type virus (Fig. 2B). Although this antibody has been previously described to enhance the infectious properties of wild-type DENV in cells with or without Fc-receptors [25],[27], clearly in the case of immature particles interaction with the Fc-receptor is important for infectivity. Taken together, these data indicate that prM antibodies facilitate efficient interaction and cell entry of virus-immune complexes via the FcγII-receptor.
Efficient FcγIIR-mediated cell entry does not however clarify what is the trigger for immature virions to become infectious, since the presence of prM has been shown to obstruct membrane fusion activity of the virus [14],[16],[18]. One could speculate that anti-prM antibody bound to immature virions induces a conformational change that would enable the E protein to trigger membrane fusion irrespective of the presence of prM. Another scenario might be that prM-containing virions mature upon cell entry since furin, although predominantly present in the TGN, also shuttles between early endosomes and the cell surface. To verify the potential involvement of furin during virus cell entry, we investigated the infectious properties of antibody-opsonized immature DENV in cells treated with furin inhibitor, decanoyl-L-arginyl-L-valyl-L-lysyl-L-arginyl-chloromethylketone (decRRVKR-CMK). In aqueous solution, decRRVKR-CMK has a half-life of 4–8 h [29] and therefore it is not expected to interfere with the maturation process of newly assembled virions within the infected cell. The results show that inhibition of furin activity completely abrogated virus particle production in cells infected with antibody-opsonized immature virions, whereas infection of cells with wild-type virus remained unaffected under these conditions (Fig. 3A).
To further substantiate the role of furin in triggering viral infectivity, we generated a furin cleavage-deficient virus (pDENprMΔ90) by deletion of the lysine on the position 90 (87-R-R-E-K-R-91) within the furin recognition sequence. Subsequently, DENVprMΔ90 virus and wild-type DENV-2 16681 (generated from pD2/IC-30P) virus were produced by transfection of RNA transcripts derived from the cDNA plasmids into BHK-15 cells. Virus production was measured by determining the number of physical particles based on GCP and the number of infectious units as measured by plaque assay. The presence of physical particles was further evaluated in three-layer ELISA experiments, by coating plates with a similar number of genome-containing DENVprMΔ90 particles and LoVo-derived immature particles. Similar OD values were measured for DENVprMΔ90 and LoVo-derived viruses (data not shown), which confirms the presence of physical particles and suggests that the number of genome-containing particles is accurately determined. Subsequent titration studies revealed that the specific infectivity of DENVprMΔ90 mutant virus is reduced by a factor of 12.000 compared to that of wild-type virus (generated from pD2/IC-30P) and is comparable to LoVo-derived immature virus. Next, K562 cells were infected with DENVprMΔ90 mutant virus opsonized with increasing concentrations of prM antibody 70–21. Figure 3B shows that disruption of the furin-recognition motif within the prM protein of the virus abrogates the enhancing activity of the anti-prM-antibody, demonstrating that enzymatic cleavage of prM to M by furin is critical to render immature DENV infectious.
To address the question as to whether prM to M cleavage can occur upon interaction of immature DENV particles with antibodies, we incubated 35S-methionine-labeled immature particles in the absence and presence of antibodies with exogenous furin for 16 h at pH 6.0 [16]. Protein visualization was done by SDS-PAGE analysis. In agreement with previous studies, we observed that exogenous furin treatment induces efficient cleavage of prM to M (Fig. 3C). Importantly, we found that the presence of prM antibodies does not affect DENV maturation, as virtually complete cleavage of prM to M was observed (Fig. 3C).
It has been postulated that antibody-mediated entry of DENV leads to a higher production of virus particles per infected cell, a phenomenon often referred to as intrinsic ADE [30]. In this part of the study, we investigated whether prM-mediated entry of immature DENV supports intrinsic ADE. Since immature particles are essentially non-infectious in the absence of antibodies, we compared the production of prM-opsonized immature DENV particles with wild-type virus in K562 cells. For accurate comparison, we first searched for a condition that gives a similar percentage of infected cells. Infection of K562 cells with prM-opsonized immature DENV at a MOG of 100 leads to 0.53%+/−0.14 infected cells (Fig. 1, 4A). Comparable numbers of infected cells were detected for wild-type DENV at MOG 10 (Fig. 4A). Under these experimental conditions, no differences were observed in E protein expression and production of virus particles (Fig. 4B–C), which indicates that the presence of prM antibodies, while evidently stimulating the infectious properties of immature virions, has no enhancing effect on the number of progeny virions produced per cell.
Given the high number of prM-containing particles in wild-type DENV preparations it is possible that prM antibodies also enhance the infectious properties of wild-type DENV. Indeed, in agreement with previous studies, opsonization of wild-type virus with prM antibodies results in a significant increase of viral infectivity (Fig. 5A-C, Fig. S3) [24],[25]. The level of enhancement is dependent on the cell type used and comparable to what has been described before for E antibodies [31],[32]. Enhancement of wild-type DENV infection was observed at higher antibody concentrations compared to that of immature particles. Although we do not completely understand these differences, we think that this may be related to the presence of structurally distinct immature virus particles (individual variations in prM/M content) in wild-type preparations [17]. Importantly, no enhancement of infection was observed in cells treated with furin inhibitor (Fig. 5D), demonstrating that furin activity in the target cells plays a vital role in triggering the infectious properties of antibody-opsonized immature particles in wild-type DENV preparations. Collectively, these results illustrate that prM antibodies enhance the infectious properties of prM-containing particles in wild-type DENV preparations and therefore may be important in disease pathogenesis.
As a first step towards elucidation of the implications of our findings in disease pathogenesis we evaluated the enhancing properties of 7 convalescent serum samples from patients infected with DENV-2. The infectious properties of immature particles opsonized with various dilutions of polyclonal sera were determined in U937 cells, since this cell line expresses both Fc receptors CD32 and CD64 on the cell surface. At 43 hr post-infection, the medium was harvested and the production of virus particles was measured by plaque assay. No plaques were found in the absence of sera and in the presence of DENV-naïve serum (Fig. 6). Convalescent sera from two distinct DENV-2 infected patients significantly enhanced the infectious properties of immature DENV particles at a 10,000 dilution (Fig. 6). Sera from two other patients enhanced the infectivity of immature particles to a minor extent as only a low number of plaques (average of 1.5 plaques) was observed. The three remaining patient sera did not show any effect on viral infectivity of immature particles as no viral plaques were observed. As expected, nearly all of the analyzed patient sera enhanced the infectious properties of wild-type virus particles (Fig. S4).
Multiple studies have shown that immature particles are non-infectious, the presence of prM obstructing the low-pH-induced conformational changes in the viral E glycoprotein required for membrane fusion of the virus [14],[16],[18],[21],[22],[33]. On the other hand, prM antibodies have been shown to enhance DENV infection [24],[25]. In this report, we show that the lack of infectivity of fully immature particles in the absence of antibodies is primarily related to inefficient binding of immature virions to the cell surface. If binding is facilitated through anti-prM antibodies, immature DENV particles become highly infectious presumably due to efficient intracellular processing of prM to M by the endoprotease furin.
Maturation upon entry has been previously reported for other enveloped viruses. Zhang and co-workers [34] showed that the infectivity of immature particles of Semliki Forest virus, an alphavirus, can be triggered by furin during viral endocytosis. It is likely that DENV maturation also occurs within acidic endosomes, since previous in vitro experiments have revealed that cleavage of immature particles by furin is dependent on the exposure of the virus to low pH [16]. We propose that the acidic conditions of the endosome, similar to those in the acidic TGN during processing of newly assembled virions, triggers an initial conformational change in the virion such that furin is able to cleave prM to M and the “pr” peptide. Interestingly, a recent report has shown that upon cleavage of prM a large fraction of pr peptide remains associated with the virion and that back-neutralization to pH 8.0 is required to release the pr peptide from the virion [14]. The authors interpreted this as a mechanism preventing newly assembled cleaved virions from undergoing membrane fusion in the acidic TGN. However, this notion is difficult to reconcile with our present observations, since virions that have matured within acidic endosomes of target cells do not return to neutral-pH conditions before initiating infection. One may speculate that the pr peptide stabilizes the E protein to such an extent that it survives the mildly acidic lumen of the TGN (∼pH 6.0), but is released at the more acidic pH of endosomes (∼pH 5.0) such that the E proteins have the capacity to rearrange to their fusion-active conformation. Another possibility is that upon cleavage of prM the pr peptide associates with the prM antibody instead of the E protein, thereby enabling the E proteins to adopt the fusion-active conformation.
The observed infectious potential of immature DENV virions in the presence of anti-prM antibodies may have important implications for our understanding of the processes involved in dengue pathogenesis. We speculate that in the early stages of a primary infection, before the appearance of virus-specific antibodies, immature virions would fail to penetrate host cells and therefore are of minor significance in disease development. On the other hand, during a secondary infection or primary infection of infants born to dengue-immune mothers, immature particles may become highly infectious due to the presence of anti-prM antibodies and hence may contribute to an increased dengue-infected cell mass and a high circulating virus titer, one of the preludes for the development of severe disease symptoms [3], [8]–[11]. Importantly, anti-prM antibodies may activate the infectious properties of a large population of virus particles, since we recently observed that a typical DENV-2 preparation of the prototype strain 16681 contains as much as 30% prM [16]. Taken together, our results suggest that immature DENV particles act as a veiled pathogen and can, like mature DENV contribute to the disease pathogenesis.
Variable levels of enhancement were seen with DENV-immune sera. As expected, virtually all of analyzed DENV-immune sera stimulated the infectivity of wild-type DENV. Interestingly, sera from 2 out of 7 patients significantly enhanced the infectious properties of immature particles. This suggests that individual patients develop different responses to prM. On the basis of these results, we believe that it is important to further investigate the antibody responses in DENV-infected patients and to unravel if patients with prM antibodies are more susceptible to develop severe disease. In this respect, it is interesting to note that the rates of prM antibody responses are significantly higher in patients experiencing a secondary infection compared to a primary infection [23]. Clearly, future clinical studies are required to obtain further evidence for the role of immature particles and prM antibodies in disease development.
Aedes albopictus C6/36 cells were maintained in minimal essential medium (Life Technologies) supplemented with 10% fetal bovine serum (FBS), 25 mM HEPES, 7.5% sodium bicarbonate, penicillin (100 U/ml), streptomycin (100 µg/ml), 200 mM glutamine and 100 µM nonessential amino acids at 28°C, 5% CO2. Baby Hamster Kidney-21 clone 15 cells (BHK-15) cells were cultured in DMEM (Life Technologies) containing 10% FBS, penicillin (100 U/ml), streptomycin (100 µg/ml), 10 mM HEPES, and 200 mM glutamine. Human adenocarcinoma LoVo cells were cultured in Ham's medium (Life Technologies) supplemented with 20% FBS at 37°C, 5% CO2. Human erythroleukemic K562 cells were maintained in DMEM supplemented with 10% FBS, penicillin (100 U/ml), and streptomycin (100 µg/ml) at 37°C, 5% CO2. Human leukemic monocyte lymphoma U937 cells were maintained in Iscove's modified Dulbecco's medium (GIBCO) supplemented with 10% FBS, 4 mM L-glutamine, penicillin (100 U/ml), and streptomycin (100 µg/ml) and adjusted to contain 1.5 g/l sodium bicarbonate, 10 mM HEPES and 1.0 mM sodium pyruvate (GIBCO). Cells were incubated at 37°C at 5% CO2. Human peripheral blood mononuclear cells (PBMCs) were maintained in RPMI 1640 medium supplemented with 10% FBS, penicillin (100 U/ml), and streptomycin (100 µg/ml). PBMCs were isolated from heparinized blood samples collected from healthy persons using standard density centrifugation procedures with Lymphoprep™ (AXIS-SHIELD). The PBMCs were used immediately after isolation or cryopreserved at −150°C. On the day of infection, the percentage of CD14+, CD19- population within isolated PBMCs was determined (5%–10% depending on the blood donor) using cell surface markers CD-14 -FITC and CD19-R-PE purchased from commercial source (IQ Products).
DENV-2 strain 16681, kindly provided by dr. Claire Huang (Center for Disease Control and Prevention, USA), was propagated on C6/36 cells as described before [16]. Briefly, monolayer of C6/36 cells was infected at multiplicity of infection (MOI of 0.1). At 96 hpi, the medium was harvested, cleared from cellular debris by low-speed centrifugation, aliquoted, and stored at −80°C. Immature DENV particles were produced on LoVo cells as described previously [16]. Briefly, LoVo cells were infected at MOI 10. Virus inoculum was removed after 1.5 h and fresh medium was added after washing the cells twice with PBS. At 72 hpi, the medium containing the virus particles was harvested, cleared from cellular debris by low-speed centrifugation, aliquoted, and stored at −80°C. [35S] methionine-labeled immature virus was prepared, as described previously [16]. Briefly, cells were infected at a MOI of 10. At 2 hpi, 400 µCi of [35S]methionine (Amersham Biosciences) was added to 20 ml of medium and incubation was continued overnight. At 23 hpi, the medium was supplemented with an additional 200 µCi of radioactive label. At 72 hpi, the supernatant containing the viral particles was cleared from cell debris by low-speed centrifugation and the virions were pelletted and further purified on a discontinuous (20 and 55% w/v) Optiprep™ gradient (Axis-Shield) by ultracentrifugation. Virus was harvested from the gradient interface, aliquoted and stored at −80°C. Virus preparations were analyzed with respect to the infectious titer and the number of genome-containing particles, as described previously [15],[16].
The furin-cleavage mutant (pDENprMΔ90) was generated by deletion of the lysine codon within the furin-recognition site at position 90 of prM. The mutation was introduced in the DENV-2 16681 infectious cDNA clone (pD2/IC-30P) [35]. Briefly, two PCR fragments were generated using the following primers: forward primer A (5′-CTC AAC GAC AGG AGC ACG ATC AT- 3′) and reverse primer A (5′- GAG TGC CAC TGA TCT TTC TCT TC-3′) and forward primer B (5′- GAA GAG AAA Δ GAT CAG TGG CAC TCG TT-3′) and reverse B (5′-GTG TCA TTT CCG ACT GCA TGC TCT-3′). The PCR fragments were ligated, cut with SacI and Sph1, and ligated into pD2/IC-30P. The introduced deletion was confirmed by DNA sequence analysis using an automated capillary sequencing system (ABI). RNA transcripts were generated from pDENprMΔ90 and pD2/IC-30P using T7 RNA polymerase and transfected into BHK-15 cells cells by electroporation (Bio-Rad Gene Pulser apparatus; two pulses at 1.5 kV, 25 µF, and 200 Ω). At 12 hours post transfection (hpt) cells were washed extensively to remove remaining RNA copies. Virus preparations were harvested at 72 hpt, cleared from cellular debris by low-speed centrifugation, aliquoted, and stored at −80°C. Virus preparations were analyzed with respect to the infectious titer and the number of genome-containing particles, as described previously. The antigenic reactivity of DENVprMΔ90 was compared to LoVo-derived virus by standard three-layer ELISA. Briefly, microtiter ELISA plates (Greiner bio-one) were coated with 4×106 GCP of different virus preparations per well in 100 µl coating buffer, overnight. After blocking with 2% milk in coating buffer for 45 min, 100 µl of two-fold serial dilutions of anti-DENV mAbs were applied to the wells and incubated for 1.5 h, in triplicate. Subsequently, 100 µl of horseradish peroxidase-conjugated goat anti-mouse IgG-isotype antibody (Southern Biotech) was applied for 1 h. All incubations were performed at 37°C. Staining was performed using o-phenylene-diamine (OPD) (Eastman Kodak Company) and absorbance was read at 492 nm (A492) with an ELISA reader (Bio-tek Instruments, Inc.).
Convalescent sera from DENV-2 immune, hospitalized patients were kindly provided by dr. G. Comach (Biomed-UC, Lardidev, Maracay, Venezuela) and dr. T. Kochel (U.S. Naval Medical Research Center Detachment, Lima, Peru). All sera samples analyzed were obtained between 20–28 days following DENV-infection.
Virus or virus-antibody complexes were added to 2×105 K562 cells, at a multiplicity of 100 genome-containing particles (MOG) per cell. After 1.5 h incubation at 37°C, the inoculum was removed and fresh medium was added to the cells. At 24–48 hpi, the medium was harvested and virus production was analyzed by plaque assay on BHK-15 cells, as described previously [36]. To measure the number of infected cells, cells were fixed at 24–48 hpi, stained with 3H5-conjugated Alexa647, and analyzed using a FACS Calibur cytometer. For virus-antibody complex formation, virus particles were incubated for 1 h at 37°C with various dilutions of monoclonal prM antibody 70-21 and 2H2 in cell culture medium containing 2% FBS prior to the addition to cells. To investigate the involvement of the Fc receptor, mAb 70–21 F(ab')2 fragments produced by use of the immobilized pepsin (Pierce) were used. Alternatively, K562 cells were pretreated with 25 µg/ml of anti-FcγRII antibody (MCA1075PE, Serotec) for 1 h at 37°C, after which access antibody was removed by extensive washing. In furin blockage experiments, cells were treated with 25 µM of furin-specific inhibitor, decanoyl-L-arginyl-L-valyl-L-lysyl-L-arginyl-chloromethylketone (decRRVKR-CMK) (Calbiochem) prior and during virus infection. In control sample for the decRRVKR-CMK activity, additional 25 µM of the inhibitor was added to ensure blockage of the progeny virus maturation.
To determine the number of bound genome-containing particles per cell, virus or virus-antibody complexes were incubated with 2×105 K526 cells at MOG 1000 for 1 h at 4°C. Subsequently, cells were washed three times with ice-cold PBS containing MgCl2 and CaCl2 (Life Technologies) to remove unbound virus-antibody complexes. Then, viral RNA was extracted from the cells by use of the QIAamp Viral RNA mini Kit (QIAGEN). Thereafter, cDNA was synthesized from the viral RNA with reverse transcription-PCR (RT-PCR), copies of which were quantified using quantitative PCR [15].
[35S]methionine-labeled immature particles or viral immune complexes were incubated with furin [New England BioLabs] for 16 h at pH 6.0, as described previously [16]. Following furin treatment samples were subjected to sodium dodecyl sulphate-polycrylamide gel electrophoresis (SDS-PAGE) analysis to visualize the protein composition.
|
10.1371/journal.pgen.1006649 | Overexpression of the WOX gene STENOFOLIA improves biomass yield and sugar release in transgenic grasses and display altered cytokinin homeostasis | Lignocellulosic biomass can be a significant source of renewable clean energy with continued improvement in biomass yield and bioconversion strategies. In higher plants, the leaf blade is the central energy convertor where solar energy and CO2 are assimilated to make the building blocks for biomass production. Here we report that introducing the leaf blade development regulator STENOFOLIA (STF), a WOX family transcription factor, into the biofuel crop switchgrass, significantly improves both biomass yield and sugar release. We found that STF overexpressing switchgrass plants produced approximately 2-fold more dry biomass and release approximately 1.8-fold more solubilized sugars without pretreatment compared to controls. The biomass increase was attributed mainly to increased leaf width and stem thickness, which was also consistent in STF transgenic rice and Brachypodium, and appeared to be caused by enhanced cell proliferation. STF directly binds to multiple regions in the promoters of some cytokinin oxidase/dehydrogenase (CKX) genes and represses their expression in all three transgenic grasses. This repression was accompanied by a significant increase in active cytokinin content in transgenic rice leaves, suggesting that the increase in biomass productivity and sugar release could at least in part be associated with improved cytokinin levels caused by repression of cytokinin degrading enzymes. Our study provides a new tool for improving biomass feedstock yield in bioenergy crops, and uncovers a novel mechanistic insight in the function of STF, which may also apply to other repressive WOX genes that are master regulators of several key plant developmental programs.
| The leaf blade of higher plants serves as a solar panel in which it captures solar energy and carbon dioxide to produce chemical energy in the process of photosynthesis. Thus, the ultimate source of food and feed for most heterotrophic organisms, including humans, comes from the photosynthetic activity of leaves. Plant biomass also promises to be a significant source of transportation fuels which could alleviate some of the stigma of environmental pollution, scarcity and finite resources associated with gasoline. Wider leaf blade may be expected to increase biomass yield and overall plant growth due to larger photosynthetic surface area. Here we introduced a leaf blade outgrowth regulatory factor, STF, from eudicot species into three grasses, switchgrass, Brachypodium and rice, and found that the transgenic plants formed much wider leaves compared to controls. Consequently, transgenic switchgrass plants produced approximately two-fold more total biomass and solubilized sugars without acid pretreatment, demonstrating a novel approach for improving biomass feedstock properties. We also show that the transgenic rice seedlings accumulate the phytohormone cytokinin at higher levels, uncovering a novel mechanism that links STF activity to cytokinin homeostasis. Our work will significantly advance understanding of the mechanistic function of WOX genes in plant development.
| Plant biomass is an abundant source of renewable energy and biomaterials, and sustainable lignocellulosic fuel ethanol production from biomass feedstocks has a great potential to be exploited as an alternative energy source to meet increasing energy demands worldwide [1]. The United States, for example, has projected to meet approximately 30% of its energy demands by 2030 from such renewable sources [2]. However, apart from the logistics of biomass transportation and processing, significant challenges still persist in biomass feedstock yield and saccharification efficiency. Plant cell wall, the most abundant plant biomass, is composed of cellulose and hemicellulose matrix polysaccharides copolymerized with a phenolic polymer lignin forming a complex crosslink [3–5]. This makes the polysaccharides recalcitrant to enzymatic digestion to soluble sugars (saccharification) for microbial conversion to biofuels [6]. Current biomass conversion technologies utilize acid pretreatment at high temperatures to break apart the lignin polymer and expose the polysaccharides. Such a pretreatment, in addition to cost and environmental pollution, negatively impacts downstream microbial fermentation, reducing the market competitiveness of biofuels. Accordingly, enhancing biomass yield and saccharification efficiency has become a major research focus for the genetic improvement of bioenergy crops. Switchgrass is one of the dedicated bioenergy crops in the USA [7] and research has been intensified in the last few years to increase yield and reduce lignin content in an attempt to improve its feedstock properties [8–10].
The leaf blade is the energy powerhouse of plants where solar energy and CO2 are assimilated to produce the chemical energy used in food, feed and biofuels. Since the leaf blade essentially serves as a solar panel in capturing sunlight, its size and design should have a significant bearing on biomass productivity through increasing photosynthetic efficiency [11–13]. Redesigning the leaf blade is, therefore, potentially a major target for improving biomass feedstock yield. Blade outgrowth is regulated by several antagonistically acting polarity factors that are exclusively expressed either on the upper (adaxial) or lower (abaxial) side of the leaf at least in eudicots. These factors include AS1, AS2, HD ZIP Ⅲ genes and tasiR-ARFs on the adaxial side and KAN, FIL, YAB, miRNA165/6 and ARF3/4 on the abaxial side in Arabidopsis and are required for polarity specification and cell differentiation in their respective domains [14–18]. Extensive studies in Arabidopsis over the past two decades revealed that the combined action of polarity factors and multiple phytohormones is required for the establishment and growth of a determinate bilaterally symmetrical leaf blade from undifferentiated pluripotent cells of the shoot apical meristem (SAM). The leaf marginal meristem (blastozone) has long been recognized as the site of cell proliferation for lateral expansion of the leaf blade after recruitment of leaf founder cells from the SAM and establishment of the leaf primordium [19–21]. However, leaf growth in the proximal-distal (length) direction appears to be to some extent independent from growth in the medial-lateral (width) direction as demonstrated by several genetic mutants affected only in leaf width [22] including the bladeless lam1 mutant of Nicotiana sylvestris and the stenofolia (stf) mutant of Medicago truncatula.
In monocots, Wavy auricle in blade 1 (Wab1) and Liguleless narrow-R (Lgn-R) mutants in maize [23–25] and narrow leaf 1 (nal1) in rice [26] display narrow leaf blades but defects in these mutants appear to include proximal-distal growth as well. On the other hand, the maize, narrowsheath1 (ns1) and narrowsheath2 (ns2) double mutant has a very narrow leaf blade affected in medial-lateral growth [27] without significant defect in leaf length. ns1 and ns2 are duplicate WUSCHEL-related homeobox (WOX) transcription factors homologous to Arabidopsis WOX3/PRS [28]. Mutations in homologous genes, nal2 and nal3 double, also cause narrow leaves in rice [29, 30]. The nal2/3 double mutant displays a pleiotropic phenotype including narrow-curly leaves, more tillers, fewer lateral roots, open spikelets and narrow-thin grains [30], indicating a widespread effect on overall plant development. Auxin transport related genes are found to be altered in expression in the nal2/3 double mutant [30], and the OsWOX3A protein, encoded by NAL2/3, is shown to be involved in negative feedback regulation of GA biosynthesis [31]. Transcriptome analysis in a laser dissected ns1/2 mutant shoot apex in maize also identified changes in hormonal signaling pathways including auxin and jasmonate [32]. However, the actual molecular mechanisms for the function of these WOX genes in blade lateral outgrowth remains unclear.
We cloned the stf and lam1 mutants previously and shown that they are caused by mutations in the same gene that encodes for a putative WUSCHEL-related homeobox (WOX) transcription factor [33] similar to petunia MAW and Arabidopsis WOX1 [34]. STF is expressed at the adaxial-abaxial juxtaposition of the growing leaf primordium that includes the leaf margins and the middle mesophyll, and critically regulates blade outgrowth by activating cell proliferation [33], which was confirmed by Arabidopsis WOX1 [35], suggesting a WUS like function in leaf margins. STF promotes cell proliferation through a transcriptional repression activity [36, 37] that involves the corepressor TOPLESS (MtTPL) [38]. Transcriptional repression activity is a requirement for STF’s blade outgrowth function, and multiple phytohormones including auxin and cytokinin have been proposed to be involved in STF function [33, 39], but the connection between transcriptional repression and activity of hormones has not been firmly established.
Here we report that ectopic expression of STF in three monocot species, switchgrass, Brachypodium and rice leads to improvement in biomass yield. We show that STF directly binds to several regions in the promoters of cytokinin oxidase/dehydrogenase genes and represses their transcription allowing accumulation of active cytokinin pools, highlighting a novel mechanism for WOX-mediated cell proliferation via transcriptional repression.
The Medicago WOX gene STF is a master regulator of plant growth and development required for leaf blade outgrowth, leaf vascular patterning, stem thickness, inflorescence fusion, petal expansion, ovule development and female fertility [33]. Although STF-like sequencers are conserved in eudicots and the early diverging angiosperm Amborella trichopoda [40], obvious STF homologues have not been identified in monocotyledonous plants [33, 38]. We wondered whether STF could be used to modify leaf size and thereby increase vegetative biomass in grasses. To test the hypothesis that the STF gene could be used to manipulate leaf size and improve biomass yield in grasses, we introduced the full-length STF CDS into the bioenergy crop switchgrass (Panicum virgatum L.) as well as two grass models Brachypodium and rice under the control of the maize UBIQUITIN (UBI) promoter by using Agrobacterium-mediated transformation. Meanwhile, we generated UBI::GUS and UBI::GFP transformants as controls. We observed that STF overexpressing transgenic lines in Brachypodium, rice and switchgrass showed significant morphological changes in leaf blade expansion compared with control plants (Fig 1). Each of the STF transformants showed wider leaf blade and thicker stem than controls but also displayed plant height phenotypes depending on STF expression levels. In general, increasing STF expression levels were correlated with increasing phenotypic severity. While modest level of STF expression promoted leaf expansion and stem thickness in all the three species, high level of expression drastically reduced plant height, and caused leaf curling and scattered deformation on the leaf vasculature although the blade still remained wider than controls (Fig 1 and S1 and S2 Figs).
In agreement with the morphological changes observed above, histological analysis showed that the cell number as well as the number of vascular bundles were significantly increased in STF overexpressing leaves and culms in all the three monocot plants (Fig 2A–2D and 2F). Cross section through the leaf blade or stem indicated that the number of veins was significantly increased (p < 0.01) in the leaf (Fig 2A–2D) and stems were significantly thicker (Fig 2F) in all the transgenic lines compared with their respective controls. Control plants were transformed with UBI::GUS in Brachypodium and switchgrass, and with UBI::GFP in rice using the same vector pMDC32. Measurement of leaf width in transgenic switchgrass was consistent with the number of veins in quantifying blade lateral expansion (S1 Table). However, examination of leaf epidermal cells in UBI::STF transformants in Brachypodium and switchgrass showed that the cell size was not obviously changed (Fig 2E), suggesting that the wider leaf blade and thicker stem phenotypes were mainly caused by enhanced cell proliferation. Consistent with this, quantitative real time PCR (qRT-PCR) analysis showed that the transcript level of the cell division marker Histone H4 was significantly increased in STF overexpression lines in all the three species (Fig 2G). These results together indicate that ectopic expression of STF in switchgrass, rice and Brachypodium leads to significant increase in plant size primarily through promoting cell proliferation.
To evaluate the impact of STF on biomass production, we measured agricultural traits in the bioenergy crop switchgrass including leaf blade length and width, plant height, internode diameter, tiller number and flowering time (S1 Table), which divided the transgenic switchgrass lines into three categories: Group Ⅰ, Group Ⅱ and Group Ⅲ. Representative lines for each group were shown in Fig 3A. Among these, the transgenic lines with high STF transcript levels (Group Ⅲ) displayed severe morphological alteration including twisted and curled leaf blade, reduced internode and plant height and delayed flowering. The low and moderate STF expressing transgenic lines, Group Ⅰ and Group Ⅱ, respectively, on the other hand, exhibited normal or even slightly enhanced plant height resulting in an overall improved plant stature (Fig 3B). STF transgenic rice and Brachypodium lines also showed a similar dosage-dependent effect on overall growth and development (S1 and S2 Figs).
To quantitatively determine these improvements, we evaluated total above ground dry biomass yield after maturity. We evaluated three independent transgenic lines in Group Ⅰ (STF-16, 17, and 21) and Group Ⅱ (STF-2, 4, and 10), and two independent lines in Group Ⅲ (STF-9 and 12) using three independent UBI::GUS transformed controls. The average dry weight of Group Ⅰ and Group Ⅱ transgenic switchgrass had a 1.68 and 1.95 fold increase, respectively, in total biomass compared with the controls, whereas strong STF expression in the group Ⅲ transgenic switchgrass led to reduced biomass production lower than controls due to stunted growth (Fig 3C). Further enzymatic hydrolysis of the dry biomass without pretreatment indicated that, excepting the Group Ⅲ transgenic lines, the total amount of solubilized sugar yield of Group Ⅰ and Group Ⅱ had increased by over 1.53 and 1.75-fold per plant, respectively (Fig 3D). This is a significant improvement in the feedstock properties of switchgrass for cellulosic ethanol production because more solubilized sugar implies more ethanol without pretreatment.
Lignin negatively impacts biomass recalcitrance and reduces bioconversion to ethanol. Composition analysis of dried STF transgenic switchgrass was carried out to further assess lignin content and its amenability to biofuel production. Measurement of acid detergent lignin content (ADL) and Phloroglucinol-HCl staining showed no obvious difference in lignin deposition pattern or ADL content in Group Ⅰ and Group Ⅱ transgenic switchgrass compared with the controls, whereas the lignin content in Group Ⅲ transgenic plants was slightly decreased (Fig 3E and S2 Table). Cellulose and hemicellulose analysis also indicated no difference in Group Ⅰ and Group Ⅱ, but slightly decreased in Group Ⅲ (S2 Table). These biochemical analyses indicated that overexpression of STF could significantly increase biomass yield and sugar release without altering relative cell wall components, demonstrating a great potential of STF to developing superior switchgrass varieties for biofuel production. Moreover, we observed that the Group Ⅰ and Group Ⅱ transgenic plants displayed better regenerative capacity in forming new leaves after harvest cutting (Fig 3F), suggesting that weak and moderate level of STF expression is conducive to facilitated switchgrass growth even at early developmental stages.
To gain insight into the mechanistic effects of STF on cell proliferation and overall plant development, we performed transcript profiling analysis using the switchgrass Affymetrix GeneChip and compared gene expression between three independent STF overexpressing Group Ⅱ plants and three independent controls transformed with GUS. Profiling analysis was performed in 6–8 cm newly generated tillers approximately 3 weeks after cutting. A total of 886 probes were significantly altered with a 2-fold or more difference compared to controls, in which 665 probes were downregulated (S3 Table) and 221 probes were upregulated (S4 Table), consistent with the primarily transcriptional repression function of STF in its native host. Gene Ontology assignments indicated that a wide range of functional groups were represented in both the upregulated and downregulated genes (S3 Fig). In agreement with the histological observation, this analysis identified genes that are known to be involved in cell proliferation such as Histone H4 and Expansin as upregulated in STF transgenic plants (Fig 4), which were confirmed by qRT-PCR (S4 Fig). The microarray data analysis also revealed that putative cytokinin oxidase/dehydrogenase (CKX) genes were downregulated in STF transformants (Fig 4, Table 1). Cytokinin oxidases/dehydrogenases catalyze the irreversible degradation of cytokinins and play important roles in maintaining cytokinin homeostasis. The phytohormone cytokinin (CK) affects many aspects of plant developmental programs, including a prominent role in the regulation of cell proliferation, plant growth and determination of organ size [41]. It is plausible that increasing the CK levels by reducing expression of CKXs could result in enhanced biomass and grain production as well as increased plant stature [42, 43]. However, several genes involved in auxin signaling/response were also altered in expression (Table 1). The phenotypes of STF transgenic lines in the three grass species and the microarray data prompted us to think that the effect of STF may, at least in part, be connected to cytokinin levels.
To confirm that overexpression of STF in switchgrass, Brachypodium and rice reduced CKX gene expression, we isolated 9 CKX family members from switchgrass and 11 members from each of Brachypodium and rice and examined their transcript levels by qRT-PCR. We found that the expression levels of 5 out of 9 PvCKX family members were considerably reduced in STF overexpressing switchgrass (Fig 5A). Similarly, the transcript levels of three BdCKX family members in Brachypodium and four OsCKX family members in rice were also found to be significantly reduced in STF transgenic lines compared to the controls (Fig 5B and 5C), indicating that the downregulation of CKXs in transgenic switchgrass microarray was not an isolated event. The CKXs that are repressed by STF across the three species are phylogenetically close to each other except PvCKX6, which appears to have no partners acting in the same way in rice or Brachypodium (S5 Fig). Since CKX enzymes are responsible for cytokinin degradation, we directly measured active and transiently inactive cytokinin levels in the STF overexpressing rice plants, for which optimized methods have been established [44]. We found that the amount of 5 of the 6 CK species measured, iP, isopentenyladenine; iPR, iP riboside; iP9G, iP 9-glucoside; tZ, zeatin; tZ9G, zeatin 9-glucoside, were significantly increased in STF overexpressing rice than the control plants (Fig 5D, S5 Table), suggesting that the repression of CKXs has a biological significance in promoting cytokinin levels. These data are consistent with the cell proliferation promotion activity of STF and suggest that the enhanced cell proliferation in STF transformants could in part be caused by elevated CK levels through repressing the expression of CKX family CK degrading enzymes, providing a novel mechanistic insight for STF molecular function.
Owing to the fact that STF acts primarily as a transcriptional repressor [36, 38] and specific CKXs are repressed in STF overexpressing switchgrass, Brachypodium and rice transgenic lines, we hypothesized that STF may directly repress the CKX family genes to promote CK activity. To test this hypothesis, we performed three complimentary experiments. First, we tested direct DNA binding of STF to CKX promoters in vitro using DNA binding assay. WOX family proteins have been reported to bind the G-box, the known TAAT motif as well as consensus sequences like CAAT and TTAA [31, 45]. Sequence analysis revealed that these binding sequences were present in the proximal promoter regions of the downregulated CKX family members like OsCKX9 and OsCKX11 in rice, BdCKX11 in Brachypodium and PvCKX4b in switchgrass (Fig 6A). We tested whether the STF protein can directly bind to DNA fragments containing these sequences in vitro. The STF homeodomain region was fused to a Trigger Factor (TF), a molecular chaperone protein that improves protein solubility [46] for expression in E.coli and purified using Profinity IMAC Ni-Charged Resin (BIO-RAD). Our results showed that the recombinant TF fused STF homeodomain protein was able to bind the fragments containing the conserved G-box, TAAT motif and CAAT and TTAA sequences, while binding was not detected by the control TF protein alone (Fig 6B), suggesting that the STF repression of CKXs is mediated by direct binding to their specific promoter sequences.
Second, using a well-established transient dual luciferase assay system in rice leaf protoplasts, we found that coexpression of the STF effector protein and the luciferase reporter constructs driven by the 1.8–2.8 kb promoters of OsCKX9 and OsCKX11 (Fig 6C), both downregulated in STF overexpressing rice, resulted in a significant reduction of the luminescence intensity compared with the GFP control effector protein (Fig 6D). Similarly, coexpression of the STF effector with the switchgrass pPvCKX4b and Brachypodium pBdCKX11 reporter constructs significantly reduced luminescence in the rice protoplast luciferase assay (Fig 6D), indicating that STF recognizes these CKX promoters in protoplasts and represses their activity.
Third, we tested direct in vivo binding of STF protein to OsCKX promoters by Chromatin Immunoprecipitation (ChIP) assay. Using UBI::GFP-STF transgenic rice and anti-GFP antibody, we found that the promoter regions of OsCKX9 and OsCKX11 were enriched in the GFP-STF chromatin (Fig 6E and 6F and S6 Fig). We tested five potential binding sites (P1-P5) within 2.8 kb region upstream of the translational start site, three non-specific regions (P6-P8) further upstream, one in the coding region (P9) and three non-specific regions downstream of the translation stop site (P10-P12) of OsCKX11 (Fig 6F). We found that the P1-P5 putative binding regions were significantly enriched compared to the non-specific regions (P6-P12) or the background signal in the OsActin control (Fig 6F), indicating that STF directly binds to multiple regions on CKX promoters, consistent with the DNA binding and dual luciferase assay results. These results together highlight a mechanism by which STF may act to modulate cytokinin levels in leaf development. Our data also demonstrate that ectopically expressing the Medicago WOX gene STF in switchgrass promotes leaf blade outgrowth and biomass accumulation with increased solubilized sugars, suggesting a potential in biomass feedstock improvement.
The plant-specific WOX family transcription factors are master regulators of plant growth and development including shoot and root apical meristems, lateral organs, organ size and vasculature [28, 47–53]. The Medicago WOX gene STF and its orthologues are key regulators of leaf blade outgrowth in the medial-lateral direction and floral organ fusion in several eudicot species [33–35, 54]. Here we reported that ectopic expression of STF in switchgrass, Brachypodium and rice universally promotes cell proliferation in vegetative organs leading to wider leaf blades, thicker stems and overall significant increase in total biomass yield. Lignocellulosic biomass has a potential to displace a significant portion of gasoline as a renewable energy source [1, 6, 55], and switchgrass is one of the dedicated bioenergy crops in the United States for lignocellulosic biofuel production [7, 56]. Although a 1.3 billion ton biomass production capacity per annum was projected in the US alone, this estimation takes into account that current biomass production efficiency per unit of land would be at least doubled including in marginal lands that are not in current crop production systems [2], indicating that biomass yield is still a major challenge for sustainable biofuel production. Another major hurdle is bioconversion efficiency imposed by the cell wall lignin polymer. Lignin, a phenolic polymer, is a major component of most plant cell walls which is recalcitrant to saccharification through enzymatic digestion [4, 5]. Lignin forms complex cross-links with two other major cell wall polymers, cellulose and hemicellulose, making these components inaccessible to saccharification without pretreatment with strong acids. This indicates that deconstruction of plant cell walls is a significant challenge and lignin content is an important biomass feedstock trait determining bioconversion efficiency. In fact, reducing lignin content by genetic engineering for feedstock quality improvement is currently a major undertaking in several laboratories. Here we demonstrate that it is possible to contribute to both biomass yield and sugar release in switchgrass and other grasses using the leaf development master regulator STF. Switchgrass Alamo plants overexpressing STF showed approximately a 2-fold increase in above ground total dry weight production and approximately 1.8-fold increase in the release of solubilized sugars (Fig 3), improving biomass feedstock properties without necessarily altering the relative composition of cell wall polymers. Similar successes have been reported in switchgrass using maize Corngrass1 miRNA [10] and rice miRNA156 [9] resulting in improved saccharification efficiency by altering developmental phase changes. Overexpression of a switchgrass ERF gene PvERF001 was also recently reported to increase biomass yield [8], but to our knowledge, this is the first report showing significant improvement in biomass yield and sugar release without pretreatment using a WOX transcription factor in switchgrass, providing a new tool for genetic modification of grasses.
STF and its orthologues are unique among the leaf blade regulators in the sense that they are expressed at the adaxial-abaxial juxtaposition [33–35, 54] unlike the well-studied polarity factors that are axial-specific. In this middle region at the leaf margin and middle mesophyll, STF and Arabidopsis WOX1 promote lateral expansion of the leaf blade by activating cell proliferation [33, 35] analogous to WUS in the SAM [50] and WOX5 in the RAM [48, 57]. STF is clearly shown to affect free auxin and ABA levels by direct measurements but has also been proposed to act as a master switch affecting several developmental programs probably through regulating multiple hormone homeostasis including auxin, cytokinin, and metabolic sugars based on transcriptomic, transgenic and metabolomics data analyses [33, 39]. But the detailed mechanism is unknown. STF physically interacts with the transcriptional co-repressor MtTPL and primarily acts as a transcriptional repressor for its cell proliferation activity in leaf blade outgrowth [36–38]. Since Arabidopsis TPL is known to modulate auxin signaling via interaction with repressive auxin response factors (ARFs) [58], it could be assumed that at least the STF-TPL complex may recruit ARFs to explain the observed effects of STF on auxin levels. However, ARF gene expression was also altered in the Medicago stf mutant microarray data [33], in the current STF transgenic switchgrass microarray, as well as in Arabidopsis wox1/prs double and petunia maw mutants quantified by qRT-PCR [34], suggesting a potential for STF direct effects on auxin signaling/homeostasis, although such direct effects are yet to be shown mechanistically.
Here we demonstrate that STF can directly bind to multiple regions on the promoter of cytokinin oxidases/dehydrogenases (CKXs) in vitro and in vivo, and represses their transcription in transgenic switchgrass, Brachypodium and rice (Figs 5 and 6), providing a novel mechanism to control local cytokinin homeostasis. CKXs are induced by cytokinins in plant tissues and catalyze the degradation of active cytokinins in a feedback inhibition to maintain cytokinin homeostasis [59]. This suggests that STF improves active cytokinin pools by preventing cytokinin degradation through repressing CKX genes (Fig 7). Indeed, direct measurement of active cytokinins and directly convertible cytokinin conjugates confirmed that 5 of the 6 cytokinin species analyzed were significantly increased in STF overexpressing transgenic rice lines. Since cytokinins are major regulators of cell proliferation [60], this finding is consistent with the STF’s role in the promotion of cell proliferation by transcriptional repression activity in leaf blade outgrowth and total biomass accumulation in the native Medicago and transgenic grasses. Activation of cytokinin signaling by WOX genes has been established for WUS in Arabidopsis SAM maintenance through repression of A-type ARRs including ARR5, ARR6, ARR7 and ARR15, which are negative regulators of cytokinin signaling [49, 61]. WOX9/STIP is also reported to act downstream of cytokinin sensing in the Arabidopsis SAM and proposed to activate the A-type ARRs [62]. Rice OsWOX4, that performs an equivalent function to Arabidopsis WUS in shoot meristem maintenance in rice, has also been reported to mimic cytokinin application in inducing calli in transgenic rice [63] although the mechanism is not known. Activation of cytokinin activity by repressing CKXs may thus be yet another mechanism to control cytokinin response, which may also have important implications in SAM maintenance. We checked the effect of STF on CKXs in Medicago truncatula, where STF is native. In the stf mutant microarray, one of only two genes included in the chip (Medtr4g044110) was modestly upregulated at 1.32 and 1.47 times, for 2 different probe sets, in the mutant compared to wild type, the other gene was unchanged at 1.08 [33]. We were able to identify seven definitive CKX-like genes in the annotated version 3.5 of the M. truncatula genome. We tested the expression of all of them by semi quantitative PCR in the unexpanded young leaves of wild type R108 and two stf mutant alleles, stf-1 and stf-2. Our preliminary results showed that one of them (Medtr3g036100) was significantly induced in the mutants, and two others, Medtr4g044110 and Medtr2g039410, were slightly induced, while the other four were basically unchanged (S7 Fig). The Medtr4g044110 weak induction in the mutants is consistent with the stf microarray data. This preliminary observation suggests that the effect on cytokinin activation may be part of the STF function in eudicots. In fact, the first true leaf-like structure with apparent petiole and blade outline was obtained in Nicotiana sylvestris lam1 mutants only after application of auxin and cytokinin together to the shoot apex [33], suggesting that both auxin and cytokinin signaling pathways and/or their crosstalk could be components of the STF function in leaf blade outgrowth.
Nevertheless, STF homologues are not found in monocots [33, 34, 38, 64, 65] and the biological significance of the cytokinin mechanistic model is yet to be tested by the endogenous grass WOX gene(s). In monocots, the WOX3 family is proposed to perform an equivalent function [39]. At least in maize and rice, where information is available, the function of STF/WOX1 appears to be met by the WOX3 family members NS1 and NS2 in maize [27], and NAL2 and NAL3 in rice [29, 30]. In eudicots, one WOX3/PRS and at least one STF/WOX1 genes are present but monocots have at least two WOX3/NS genes instead. Arabidopsis PRS is required for lateral axis-dependent development of flowers, lateral sepals, lateral stamens and leaf stipules [27, 28, 66], but prs mutants do not display narrow leaf blades although PRS is found to redundantly function with WOX1 in leaf blade outgrowth [34, 35]. In M. truncatula and probably other eudicots, the WOX3/LFL role is restricted to flower development [52]. It is likely that WOX1 and WOX3 have separate roles in most eudicots, other than Arabidopsis, and leaf blade development in the medial-lateral axis is governed by the STF/WOX1 family, while redundant WOX3 family genes fulfill this role in monocots. Thus, it is not surprising if some of the mechanisms are conserved between STF/WOX1 and WOX3/NS functions. The actual mechanism of NS and NAL function is unclear but the involvement of hormone signaling, especially auxin, has been predicted from transcriptome analysis [29, 32] and OsWOX3A has been reported to act in GA homeostasis [31]. It would be interesting to see if the ns1/2 and nal2/3 mutants have altered cytokinin activity or CKXs expression levels, and also complement these mutants with STF/WOX1 to understand the extent of mechanistic conservation between monocots and eudicots in lateral outgrowth of the leaf blade.
Our data suggest that promotion of cytokinin activity by direct repression of cytokinin degradation may be a general mechanism for the action of repressive WOX genes that are involved in cell proliferation during meristem maintenance and lateral organ development including WUS, NAL and NS genes. Further experiments are needed to confirm the validity of this hypothesis. Taken together, our results provide a powerful tool for genetic modification of biomass yield and sugar release in perennial and annual grasses, and uncover a novel mechanistic insight directly connecting cytokinin activity and cell proliferation by repressive WOX genes that have far reaching consequences in regulating plant vegetative and reproductive developmental programs.
Brachypodium distachyon inbred line Bd21-3, Oryza sativa varieties Nipponbare and the lowland-type switchgrass cultivar Alamo were used in this study. Brachypodium and switchgrass plants were grown in the greenhouse under a 26°C/16-h (day) and 23°C/8-h (night) photoperiods with lighting supplied by parabolic aluminized reflector lamps (average 390 μE⁄m2⁄S1). The rice (Oryza sativa L.) plants were cultivated in experiment field and green house in Beijing.
The Maize Ubiquitin promoter from pTCK303 [67] was cloned into pMDC32 Gateway vector substituting 2×35S promoter to generate pMDC32-pUBI destination vector by Hind Ⅲ and Kpn Ⅰ, and STF, GFP and GUS were cloned into pMDC32-pUBI destination vector by using the Gateway system (Invitrogen). To generate the UBI::GFP-STF vector, GFP and STF were cloned separately, with 18 bp overlapping sequence between 3’GFP and 5’STF to acquire the GFP-STF sequence cloned into pMDC32-pUBI destination vector by using the Gateway system (Invitrogen). Constructs were introduced into Agrobacterium tumefaciens by electroporation or the freezing transformation method. A. tumefaciences strain AGL1 was used for Brachypodium, switchgrass and rice transformation as previously described [9, 68, 69].
Total RNA was extracted from 6–8 cm newly generated tillers approximately three weeks after cutting (for switchgrass Microarray and qRT-PCR analysis), 2 weeks old seedlings (for rice qRT-PCR analysis) and 1 month seedling shoots (for Brachypodium qRT-PCR analysis) of UBI::STF, UBI::GUS and UBI::GFP transgenic plants by using TRIzol Reagent (Invitrogen). cDNA was generated by reverse transcription with SuperScript Ⅲ (Invitrogen). Quantitative RT-PCR was performed as previously described [36], with at least 2 biological and 3 technical replicates for both samples and controls. The CKX gene ID and primers used were listed in S6 and S7 Tables.
Tissue fixation and embedding were performed as described [69]. The tissues were sliced into 10 μm sections with a Leica RM2145 microtome, affixed to microscope slides, and stained with toluidine blue. Cross sections of the flag leaves and internode Ⅱ (for rice and Brachpodium) and internode Ⅲ (for switchgrass) of the controls and STF transformants were stained with phloroglucinol-HCl reagent as previously described [70]. Images were taken under an Olympus BX-51 compound microscope.
6–8 cm newly generated tillers, approximately 3 weeks after cutting, of three independent Group Ⅱ UBI::STF transgenic lines (STF-2, 4, and 10) and three independent UBI::GUS lines were used for total RNA extraction and microarray experiment. The Microarray analysis of transgenic switchgrass plants were performed as previously described [9]. Data analysis of differentially expressed probe sets on the chip was performed by associative analysis as described [71].
Extraction and determination of CKs from the top two leaves of three UBI::STF transgenic rice lines and wild type at vegetative stage (two months old after planting) were performed by using a polymer monolith microextraction/hydrophilic interaction chromatography/electrospray ionization tandem mass spectrometry method as described [44].
Total above-ground shoot tissues at flowering stage were harvested for the dry biomass yield analysis and further biofuel property evaluation. Lignin analysis and enzymatic saccharification were performed as previously described [9].
The STF cDNA corresponding to the N-terminal and homeodomain (HD) regions with amino acids 1 to 226 was cloned into the E. coli expression vector pCOLD-TF with His tag (Takara, TF indicates Trigger Factor, a molecular chaperone protein of original nuclear pullulan increasing the protein solubility) using EcoR Ⅰ and BamH Ⅰ restriction enzymes. Expression of pCOLD-TF, and pCOLD-TF-STF(HD) in BL21 cells was induced with 0.2 mM isopropyl-1-thio-D-galactopyranoside at 18°C for 16 h. Fusion protein was purified using Profinity IMAC Ni-Charged Resin (BIO-RAD) according to the manufacturer’s protocol and quantified by the Bio-Rad protein assay reagent. DNA binding assay was performed as previous described [72]. The putative STF binding fragments were incubated with purified His-TF alone and with the His-TF-STF(HD) fusion protein, and the DNA binding activity (protein bound DNA) was determined by qRT-PCR after washing and elution. Primers used were listed in S7 Table.
The coding sequences of GFP-STF and GFP were cloned into p2GW7 using the Gateway system (Invitrogen) to yield effector plasmids. For the reporter plasmid, a mini 35S promoter [73] with BamH Ⅰ was inserted into the pGreen Ⅱ-0800-Luc vector by exonuclease Ⅲ, to generate the destination vector pGreen Ⅱ-0800-p35S mini-Luc, and the promoter of OsCKX9, OsCKX11, BdCKX11 and PvCKX4b were cloned into pGreen Ⅱ-0800-p35S mini-Luc by restriction digestion to generate the reporter plasmid. Primers used were listed in S7 Table. Transient expression assays were performed in rice protoplasts as previously described [36, 74]. For each transformation, 5 μg of reporter plasmid and 5 μg of effector plasmid were used. Luciferase activities were detected by Dual-Luciferase Reporter Assay System (Promega) as previously described [75].
The UBI::GFP and UBI::GFP-STF transgenic rice lines were used for ChIP assay according to the method described previously [76] with some modifications. Briefly, 1 g tissue of 10-day-old T2 seedlings per sample was harvested from plants grown in greenhouse. Samples were cross-linked with 1% (v/v) formaldehyde under vacuum for 10 min, quenched with Gly (0.2 M) for 5 min, and then ground to powder in liquid nitrogen. The chromatin complexes were isolated, sonicated and then incubated with polyclonal anti-GFP antibodies (Abcam, AB290). The precipitated DNA was recovered and used as a template for qRT-PCR analysis. The input DNA and no antibody–precipitated DNA were used as positive and negative controls, respectively. The primers used for the ChIP assays were described in S7 Table.
|
10.1371/journal.pbio.0060250 | Rapid Interhemispheric Switching during Vocal Production in a Songbird | To generate complex bilateral motor patterns such as those underlying birdsong, neural activity must be highly coordinated across the two cerebral hemispheres. However, it remains largely elusive how this coordination is achieved given that interhemispheric communication between song-control areas in the avian cerebrum is restricted to projections received from bilaterally connecting areas in the mid- and hindbrain. By electrically stimulating cerebral premotor areas in zebra finches, we find that behavioral effectiveness of stimulation rapidly switches between hemispheres. In time intervals in which stimulation in one hemisphere tends to distort songs, stimulation in the other hemisphere is mostly ineffective, revealing an idiosyncratic form of motor dominance that bounces back and forth between hemispheres like a virtual ping-pong ball. The intervals of lateralized effectiveness are broadly distributed and are unrelated to simple spectral and temporal song features. Such interhemispheric switching could be an important dynamical aspect of neural coordination that may have evolved from simpler pattern generator circuits.
| As for all vertebrates, the songbird cerebrum has two halves (or hemispheres), each of which controls mainly the muscles in one half of the body. Many motor behaviors such as singing rely on high coordination of activity in both hemispheres, yet little is known about the neural mechanisms of this coordination. By using electrical stimuli to briefly perturb the activity of neurons in the motor pathway during song production, we study their involvement in generating the different elements of a song in zebra finches. We find mostly disjoint time intervals in which stimulation of either the right or left hemisphere is effective in distorting a song. This interhemispheric switching of stimulation effectiveness is evidence of a novel form of ping-pong–like motor coordination. Because left–right alternation is the basis of many motor patterns such as swimming and walking, we speculate that interhemispheric switching in songbirds has its evolutionary roots in older circuit principles invented for locomotion.
| Owing to its complexity and high precision, birdsong has provided an important animal model for studies of motor control. Adult zebra finch songs are formed by repetitions of a highly stereotyped motif that is composed of two to eight syllables and is acquired from a tutor during a critical sensorimotor period [1]. Because the stereotypy of birdsong is sustained after removal of auditory feedback, birdsong has been thought to be organized by a “central motor program” [2–4].
The main cerebral brain areas for vocal production are the robust nucleus of the arcopallium (RA), HVC (used as a proper name), and the lateral magnocellular nucleus of the anterior nidopallium (LMAN), the latter of which forms the output of an avian basal-ganglia pathway [5]. Song-related neural activity in premotor brain areas is precisely coordinated across hemispheres, because both hemispheres contribute to the production of one unique and highly stereotyped song. This precise coordination is illustrated by the strong synchronization of multiunit activity in left and right HVC during singing [6].
A useful method to probe the functional roles of premotor brain areas is electrical stimulation. In general, electrical stimulation during motor production leads to specific behavioral distortions that depend on the location of stimulation electrodes [7,8] as well as on the stimulation time (or phase) within ongoing motor patterns [9–12]. For example, in LMAN, which is involved in modulating birdsong by social context [13,14], unilateral electrical stimulation induces small transient effects on sound amplitude or sound pitch, depending on the precise stimulation time within the ongoing song motif [10]. In HVC, which generates adult song by means of ultrasparsely firing “clockwork” neurons [15,16], unilateral electrical stimulation also leads to transient song degradations such as syllable distortions and syllable truncations [17]. More importantly, both LMAN and HVC stimulation sometimes induce nontransient effects such as song stoppings or early song restarts [4]. During such restarting events caused by HVC stimulation, ongoing premotor activity in the contralateral HVC is reset within a few tens of milliseconds [18]. Given that there are no direct interhemispheric connections between cerebral song-control areas, interhemispheric synchronization and resetting must rely on common inputs to the song-control system from interhemispherically connected mid- and hindbrain areas [19–23] (Figure 1).
To explore the mechanisms of interhemispheric coordination and the dependence of song distortions on stimulation time, we chronically implanted HVC in adult male zebra finches with bipolar stimulation electrodes. We trained an artificial neural network to reliably detect the earliest possible note in a song motif in real time and stimulated either right or left HVC with a brief 0.4-ms biphasic (0.2 ms/phase) current pulse at random time lags after detection. We frequently interleaved stimulation trials by catch trials in which no stimulation was delivered. We also explored temporally modulated effectiveness of LMAN stimulation by using suitable multipulse current trains delivered to bipolar stimulation electrodes implanted in LMAN [10].
In line with earlier work, we found that unilateral HVC and LMAN stimulation distorted songs at the levels of song syllables and song motifs (Figure 2A) [4,10,17,18]. By definition, syllable-level effects were restricted to the stimulated or the subsequent syllable and consisted of either syllable distortions or syllable truncations. On the other hand, motif-level effects were manifest in longer time windows after stimulation and consisted of sudden song stopping or early motif restarts (see Materials and Methods for exact definition of effects). The prevalence of syllable- and motif-level distortions caused by HVC and LMAN stimulation is reported for all birds in Table S1.
When hundreds of stimulated motifs were reordered by stimulation time, a temporal contiguity of stimulation effects became apparent in which nearby stimulation times led to qualitatively similar song distortions (Figure 2B). Hence, song distortions were not random, but were often deterministically linked with stimulation time, possibly caused by strong perturbation of stereotyped premotor activity.
A more detailed analysis revealed that song distortions most frequently occurred on the syllable level within several tens of milliseconds after stimulation. The probability of sound-amplitude distortions sharply increased 20 ms after stimulation, peaked roughly 50 ms after stimulation, and decayed thereafter (Figure 2C). This sharp rise agrees with measurements of air sac pressure deviations, the average onset of which lags HVC stimulation by 15–20 ms [17], whereas the late decay suggests that some perturbations of neural activity were transient and affected only a subpopulation of neurons.
Interestingly, on a fine time scale, not all distortions were locked to stimulation time. We occasionally observed syllables that were truncated, not with a fixed delay to stimulation, but during a fixed time point with respect to the unperturbed motif (Figure 2D and 2E). In these cases, stimulation needed to occur within some time interval before a particular note in order to truncate that note, revealing that the motor program exhibits time points of high perturbation sensitivity. And, more interestingly, sometimes stimulation effects such as early motif restarts occurred neither after a fixed latency to stimulation nor at a fixed time point of the unperturbed motif, but at some intermediate time (Figure 2D), further demonstrating nonlinear timing aspects of the song motor program.
We automated the inspection of song distortions by analysis of sound amplitudes. We were mostly interested in motif-level effects because these seemed to arise from wide-spread and irreversible perturbation of premotor activity. For each stimulation time, we computed a late-effect (LE) value, defined as the fraction of 3.9-ms time bins in a 78–312-ms window after stimulation in which sound amplitudes were significantly different from amplitudes recorded during catch trials (see Materials and Methods). LE curves as a function of stimulation time had many sharp peaks that corresponded to different motif-level effects, separated by troughs in which stimulation was rather ineffective (on the motif level). When we increased the stimulation currents, the set of effective stimulation times grew, as revealed by LE peaks that grew in height and width (Figure 2F). At the extreme of very high currents on the order of 0.5–1 mA, birds always stopped singing, and significant LEs were seen for all stimulation times (n = 3 birds, unpublished data). In this study, our experimental strategy was to rapidly tune stimulation currents in order to observe highly modulated LE curves with coexistence of very large and close to zero values, a task that typically was achieved within 2 d. At the current intensities chosen, LE curves displayed diverse peaks (the mean peak width at the effectiveness threshold was 20 ms, median 8 ms, range 4 to 160 ms, n = 20 HVC stimulation sites in 10 birds). This wide range of peak widths in LE curves indicates that HVC stimulation perturbed neural activity on multiple time scales. The strong modulation of LE curves suggests rapid waxing and waning of the ipsilateral HVC drive, raising the question about modulation in the contralateral hemisphere.
To probe evidence of lateralized stimulation effectiveness, we implanted birds with stimulation electrodes in both left and right HVC, and performed unilateral stimulation in randomly chosen hemispheres and at random time lags after note detection. After sorting all trials recorded over 1–3 d by hemisphere and stimulation time, a remarkable complementarity became apparent: For most stimulation times, stimulation effects were seen either for right- or left-side stimulation, but not for both (Figure 3A; see Figure S1 for all birds used in our study). LE curves associated with left and right HVC stimulation were strongly modulated, but in an alternating fashion. We quantified the interhemispheric complementarity of stimulation effectiveness by the correlation coefficient (CC; see Materials and Methods) between right and left LE curves, and found that negative correlations prevailed (average CC −0.36, range −0.68 to −0.01, n = 10 birds). To assess the significance of these anticorrelations, in three birds we implanted two pairs of stimulation electrodes in right HVC (in a cross arrangement). By running the same experimental protocol on the two ipsilateral stimulation sites in HVC, we found that CCs between corresponding LE curves were positive (average CC 0.36, range 0.25 to 0.46, n = 3 birds), illustrating that the dependence of stimulation effects on electrode position within HVC is weak and demonstrating that the anticorrelation of stimulation effectiveness in bilateral stimulation experiments was highly significant. Moreover, in two birds, we implanted stimulation electrodes in right HVC and right LMAN, and also found positive CCs between corresponding LE curves (0.65 and 0.51, Figure 3B). The CCs in all birds are depicted in Figure 3C (see Table S2 for additional characterizations of the complementarity of right and left LE curves). We interpret this complementarity as evidence that interhemispheric motor coordination involves temporally alternating neural mechanisms.
We were interested in determining whether the events at which stimulation effectiveness switched from one hemisphere to the other were locked to salient song features and whether the resulting switching intervals obeyed any regularity. Visually, the effectiveness of electrical stimuli appeared to switch several times from one hemisphere to the other within a song motif, but often there was no obvious relationship between the discrete switching events and the sound spectrum produced at these times (inset of Figure 3A). When we assessed the events at which the effectiveness of electrical stimuli switched from one hemisphere to the other in terms of onsets of contralateral effectiveness (LE values larger than baseline), the mean switching interval was 35 ms (median 28 ms, range 4 to 150 ms). By contrast, when the switching events were defined by joint occurrence of ipsilateral ineffectiveness and onsets of contralateral effectiveness, the mean interval was 64 ms (median 44 ms, range 4 to 240 ms). Hence, on average, stimulus effectiveness switched back and forth between hemispheres within a few tens of milliseconds. However, our estimates of lateralized effectiveness and switching intervals must be interpreted with caution because of the aforementioned dependence of LE peak widths on stimulus current, implying that switching intervals depend (nontrivially) on stimulus current. Nevertheless, because we found broadly distributed switching intervals both across all birds and within single birds, there is little evidence of periodicity in this interhemispheric switching process.
We further explored whether effective stimuli and their lateralization were related to specific sound features. Zebra finches mostly expire during syllables and inspire during syllable gaps [24]. Both expiratory and inspiratory nuclei in the brainstem project bilaterally and therefore may be involved in controlling effectiveness switching. Because we did not measure bronchial air flow, here we inferred respiratory patterns from sound pitch curves using the simplifying assumption that zero pitch during syllable gaps corresponded to inspiration and nonzero pitch to expiration. We defined a rhythm curve as being equal to one during expiration and zero during inspiration. There was no significant coherence between this rhythm curve and either the right or left LE curves (see Materials and Methods). These results were unchanged when we defined expiratory patterns in terms of pitch values in the limited range 20–5,000 Hz (thereby assuming that some high-pitched notes are generated during inspiration). Similarly, there was no significant coherence between right/left LE curves and each of the following: sound-amplitude curves, pitch curves (see Materials and Methods), syllable onset curves, and syllable offset curves (the latter were binary curves in which a pulse of variable width was set at the transitions between inspiration and expiration as assessed by the rhythm curve). Thus, the evidence for a consistent relationship between stimulation effectiveness and simple sound features is rather weak. Notice though that all our conclusions were reached from just a few seconds of effect-curve data (15 birds) and that it would be worthwhile to reinvestigate the relation between stimulation effectiveness and song features in the future provided a larger body of interhemispheric stimulation data will be available.
We have demonstrated an interhemispheric switching process for vocal production. In this process, the motor program exhibits perturbation sensitivity that rapidly alternates between hemispheres. Such alternation is surprising given that HVC activity is highly synchronized across hemispheres during singing, and suggests that motor dominance rapidly switches back and forth between hemispheres. Possibly, the apparent alternation of dominance is related to birds' ability to independently control the two halves of their vocal organ [25,26]. However, alternation is not synonymous with independent control as it represents a restriction on independence. It is difficult to ascertain which hemisphere is dominating at any time in this switching process, because we were not able to find a simple relationship in zebra finches between stimulation effectiveness and either song features or song rhythm. On the one hand, one could argue that stimulation should be more effective in a dominant hemisphere because this hemisphere is being perturbed while generating a song in both syringeal halves. On the other hand, one could argue that stimulation should be less effective in a dominant hemisphere because the perturbation is not strong enough to overrule the ongoing activity there. In the following, we discuss the evidence for these two interpretations, as well as for interpretations on whether stimulations perturb activity in local or in distributed networks.
From existing data, we cannot infer whether or not the motor apparatus necessitates continuous and simultaneous drive from both cerebral hemispheres: adult birds do not sing normally after unilateral RA lesions [19], but these data do not exclude the possibility that at any time, the effective motor program resides in just a single hemisphere and bounces back and forth between hemispheres during singing. For example, if singing at all times is based on activity in just a single hemisphere and the drive provided by premotor activity in the other hemisphere is temporarily gated off, then we would conclude that the dominant hemisphere is the one in which low-intensity stimulation is effective. In this view, stimulation of the nondominant hemisphere above a given current threshold would also be able to distort songs, because strong perturbations might ultimately find their way to the dominant side (past the gate) where they could interrupt the ongoing motor program. However, if normal song production at all times requires simultaneous contributions from both hemispheres, then high stimulation effectiveness might be an attribute of the nondominant hemisphere, because this hemisphere can be perturbed at lower stimulation currents. On the dominant side then, low-intensity stimulation would be corrected by redundant neural mechanisms that were not sufficiently perturbed by the stimulation.
Not only the dominance question is difficult to address, but it is similarly difficult to tell whether song disruptions were entirely due to perturbation of local ongoing HVC activity or of a larger distributed network. For example, the number of spiking RA-projecting HVC neurons might drift randomly up and down during the song motif (with some inertia). Such random drifts could be associated with a compensatory increase in the number of spiking neurons in the contralateral HVC and thus to alternation of dominance. A compensatory process could be regulated during song development (e.g., by neurogenesis [27] and programmed cell death), and therefore alternating dominance would not have to rely on real-time interhemispheric communication. According to this interpretation, LMAN and ipsilateral HVC stimulation lead to similar song distortions because LMAN stimulation perturbs RA-projecting HVC neurons, for example, via RA [28]. Although at this stage we cannot rule out this scenario, it is unclear why compensatory mechanisms would act across hemispheres, but not within the same hemisphere. Furthermore, it is difficult to reconcile this scenario with observations of interhemispheric synchronization of HVC activity and with some stimulation effects such as early song restarts.
The more likely scenario within which our observations can be explained is that LMAN and HVC stimulation induce similar song distortions because of widespread perturbation of subpallial structures via RA. Because we observed a wide range of switching intervals, we found little support for the idea that switching times are determined by fixed signal propagation times (for example as reverberating activity in closed synaptic loops) or by the fixed period of a simple pattern generator circuit. Rather, some switching events may arise from detection of specific premotor patterns in one hemisphere that are subsequently relayed to the contralateral hemisphere.
Interhemispheric switching processes in relation to motor production have been reported also in mammals, for example during the preparation of vocal production in humans, in which effectiveness of transcranial magnetic stimulation (TMS) of motor cortex alternates between hemispheres [12]. Interhemispheric switching has also recently been shown to exist during perceptual rivalry, as evidenced by the hemispheric dependence of magnetic and calorimetric stimulation [29]. Interhemispheric switching may thus be a fundamental mechanism by which sensory and motor-related activity is coordinated across hemispheres. In mammals, interhemispheric coordination seems to be mainly mediated by corticocortical projections [30,31]. However, during saccadic eye movements of split-brain monkeys, activity in the two hemispheres has been shown to remain coordinated despite the lack of cerebral commissures, suggesting that subcortical pathways can subserve coordination also in the mammalian brain [32], and suggesting that similarities may exist between interhemispheric coordination in avian and mammalian brains.
Based on networks models, switching has been proposed to depend on competitive interactions [33] mediated by inhibition [34]. Evidence for interhemispheric inhibition has been found in TMS studies of human motor cortex [35,36]. We speculate that interhemispheric switching in songbirds could also rely on inhibitory mechanisms. A possible function of such inhibition could be to suppress mirror-symmetrical movements, which are thought to represent one of the default operation modes of bilateral motor systems [37]. In this sense, interhemispheric inhibition would coexist with more cooperative (excitatory) interactions between hemispheres. Inhibitory gating mechanisms could be mediated, for example, via tonically spiking Uva projection neurons (see also Figure 1) [38], and excitatory mechanisms could be relayed by respiratory nuclei, known to generate mirror-symmetrical respiratory patterns [39].
The reported interhemispheric switching process is reminiscent of one of the most prominent motor programs with left–right alternating dynamics, which is locomotion. In vertebrates, locomotion is subserved by central pattern generators in the spinal cord, which can display sustained rhythmic activity with left–right alterations even in in vitro preparations [40]. Because locomotion is much older than birdsong on an evolutionary time scale, phase-alternating neural circuits must have existed long before birds started to sing. Possibly, principles of limb coordination in locomotor circuits have been replicated by evolution for the more recent advent of birdsong. Some support for this idea comes from the conservation of bilateral projection patterns in brainstem nuclei of songbirds and non-songbirds [41], suggesting that old brain circuits have evolved to support new functions.
Adult (>90 d old) male zebra finches (Taeniopygia guttata) were used for experiments. Birds were selected on the basis of singing frequency and song complexity, and were isolated in a sound-attenuating chamber. To maximize singing frequency, birds had visual contact to one or more female zebra finches through the glass door of the chamber. A total of 15 birds were used; data in one bird were discarded because HVC stimulation did not reliably produce motif-level effects. At the end of experiments, electrolytic lesions were performed at the stimulation sites by DC current injections (15 μA for 20 s), birds were killed by overdose of Nembutal, and stimulation sites were verified in histological brain sections. All procedures were approved by the Veterinary Office of the Kanton of Zurich.
We delivered electrical stimuli with uniformly distributed probability over the time span of song motifs using custom written Labview software (National Instruments Corporation). With probability 0.35, detection triggered microstimulation at site A, with probability 0.35 at site B; and with probability 0.3, no stimulation was delivered (catch trials).
Electrodes were made of 50-μm stainless steel wire. Electrical stimuli in HVC consisted of a single 0.4-ms biphasic (0.2 ms/phase) current pulse of amplitude between 100 μA and 1 mA. In LMAN, electrical stimuli consisted of trains of ten biphasic current pulses at 400 Hz (0.4 ms/phase; train duration 23.3 ms) and amplitudes in the range 10–100 μA. The current threshold at which single-pulse stimulation in LMAN induced motif-level effects (song suspensions) was high (typically >1 mA). For this reason and to adhere to previous stimulation studies [4,10], we chose a multipulse paradigm in LMAN in which we stimulated for ten pulses at low currents (10 ∼ 100 μA per pulse).
We distinguished among different syllable and motif-level effects as follows:
Syllable truncations. First, we measured baseline distributions of syllable lengths from data of selected catch trials (only complete motifs). Stimulated syllables were then classified as truncated if their duration was within the lowest percentile of the baseline distribution. We searched for truncations only in a time window up to 156 ms (corresponding to 40 time bins of 3.9 ms or 128 sound samples each) after stimulation.
Syllable distortions. In each time bin after note detection, we calculated the baseline distribution of sound amplitudes during selected catch trials (no spontaneous song stopping). We then counted the number of 3.9-ms time bins up to 78 ms post stimulation time in which the stimulation-related sound amplitudes were significantly different from baseline (percentile p < 0.025 or p > 0.975). If this number was large enough (binomial test, alpha = 0.05), then we classified this stimulation effect as a syllable distortion. Distortions and truncations were not mutually exclusive.
Motif stoppings and restarts. For each bird, we chose a sound-amplitude threshold slightly above cage-noise level (we found that a threshold of 20% into the 1–99th percentile interval worked well for all birds). For all stimulation trials, under visual supervision, we then used this threshold to mark the offset time of every prematurely stopped motif and the successive restart time of the following note (independently of whether this note come from a song syllable, an introductory note, or a call). If the offset time fell into a window from 0 to 156 ms after stimulation and there was no restart until 312 ms, we then classified the stimulation effect as a stopping event. If, on the other hand, there was a restart after a premature offset within 312 ms after stimulation, then the stimulation effect was a restart. Hence, restarts and stoppings were mutually exclusive (however, song stoppings and syllable truncations were not).
All songs (stimulation and catch trials) were aligned by detection time. For each stimulation site, we sorted the trials by stimulation time and grouped them into 9.75-ms sets with centers separated by 3.9 ms from each other. With a mean stimulation range of approximately 500 ms and typical detection of 800–2,000 song motifs per day, we obtained roughly three to eight stimulation trials per set per day. Typically, we collected a mean of 10–20 trials per set and then tested for each set whether the sound amplitudes in 3.9-ms bins after stimulation were different from amplitudes in matched time bins during catch trials using the Kolmogorov-Smirnov (KS) test (p < 0.01). For each set, we quantified the stimulation effect by the fraction of time bins in which significant differences were detected. LE curves were based on bins ranging from 78 to 312 ms after stimulation (bins 21 to 80). Early-effect (EE) curves were based on bins ranging from 0 to 78 ms after stimulation (bins 1 to 20). To assess the time scales of song perturbations, we computed the peak widths in LE curves at the effectiveness threshold, defined by the baseline LE value during catch trials (binomial test, p < 0.01).
Our results did not depend critically on the EE and LE time windows in which syllable-level and motif-level effects were assessed. We chose the offset of the LE window (312 ms) as a compromise between being large enough to yield high sensitivity and small enough to not extend too far beyond the motif end where songs became highly variable. We set the onset of the LE window (or offset of the EE window, 78 ms) so as to exceed the peak time of stimulation effectiveness (Figure 2D), which was within 70 ms of stimulation (in agreement with previous reports [4]). Small changes in the LE window onsets (from 58.5 to 117 ms) and LE window offsets (from 234 to 390 ms) did not affect our findings of interhemispheric switching in any way.
By experimental design, our results were robust to variability in sound amplitudes caused by movements of the bird's head relative to the microphone. That is, head-position variability must have had identical influences on sound amplitudes recorded during catch trials and during stimulation trials because we randomly chose all stimulation parameters right after each detection event (i.e., whether and where to stimulate, and the stimulation time). Hence, by design there were no correlations between head position and stimulation parameters.
We assessed the similarity between effect curves x and y associated with different stimulation sites by the (Pearson) CC:
, where Cov(x,y) is the covariance between x and y. Because effect curves were nonnegative, stimulation times for which both x and y were ineffective (compared to sound-amplitude variability before stimulation, binomial test at 99% significance level) imposed a bias toward positive correlations. To avoid this bias, we ignored bilaterally ineffective stimulation times when calculating the CC (for LE curves, these were 32% of all stimulation times). Note that our conclusions were unchanged when CCs were calculated over the full set of stimulation times (thereby imposing a positive bias): the difference between average CCs in unilateral and bilateral stimulation experiments was highly significant in either case (p < 0.001, Wilcoxon rank sum test).
The relationship between right/left LE curves y and the rhythm curve z was investigated by the coherence
, where p(yz) is the cross-spectral density, and p(yy) and p(zz) are the power spectral densities of LE and rhythm curves, respectively. We chose the coherence function because its phase insensitivity allowed us to detect significant correlations irrespective of their time lag. We assessed the significance of coherence peaks by testing whether these exceeded two jackknife estimates of standard deviation (corresponding to 95% confidence). The ten jackknifes were defined by leaving out each of the ten birds from the analysis.
By visual inspection, stimulation effectiveness at the syllable level showed weaker interhemispheric complementarity than effectiveness at the motif-level. Yet, EE curves associated with bilateral stimulation (average CC −0.13, range −0.68 to 0.43, n = 10) showed significantly lower correlations (p = 0.019, Wilcoxon rank sum test) than EE curves associated with unilateral stimulation (average CC 0.35, range 0.04 to 0.80, n = 5 birds). As before, to compute these CCs, we only considered stimulation times that were associated with effectiveness in at least one hemisphere, thereby omitting 25% of stimulation times (compared to omission of 32% for LE curves). In conclusion, alternating effectiveness was seen most clearly for stimuli that disrupted normal singing, but also for stimuli that induced immediate amplitude distortions.
We investigated the possibility that pitch differences exist between times at which right and left HVC stimulation is effective. The coherence between the sound pitch curve and either right or left LE curve was not significant, neither when we considered the full pitch curve nor when we clamped the pitch curve to zero below either 2 or 5 kHz. Similarly, the median pitch during right-effective stimulation was not statistically different from the median pitch during left-effective stimulation (Wilcoxon rank sum test, p = 0.4). We also tested whether pitch differences were seen at a particular time lag after effective stimulation times. We found that the median pitch 40 ms after left-effective stimulation was significantly higher than 40 ms after right-effective stimulation (p = 0.031, n = 10 birds). However, when we excluded any one of two particular birds from the analysis, then significance broke down (p > 0.1). Significance also broke down when assessed using a shuffle predictor of pitch differences in songs of randomly shuffled syllables and gaps from different birds (Monte Carlo simulations, p > 0.05).
|
10.1371/journal.ppat.1003479 | Reprogramming of Murine Macrophages through TLR2 Confers Viral Resistance via TRAF3-Mediated, Enhanced Interferon Production | The cell surface/endosomal Toll-like Receptors (TLRs) are instrumental in initiating immune responses to both bacteria and viruses. With the exception of TLR2, all TLRs and cytosolic RIG-I-like receptors (RLRs) with known virus-derived ligands induce type I interferons (IFNs) in macrophages or dendritic cells. Herein, we report that prior ligation of TLR2, an event previously shown to induce “homo” or “hetero” tolerance, strongly “primes” macrophages for increased Type I IFN production in response to subsequent TLR/RLR signaling. This occurs by increasing activation of the transcription factor, IFN Regulatory Factor-3 (IRF-3) that, in turn, leads to enhanced induction of IFN-β, while expression of other pro-inflammatory genes are suppressed (tolerized). In vitro or in vivo “priming” of murine macrophages with TLR2 ligands increase virus-mediated IFN induction and resistance to infection. This priming effect of TLR2 is mediated by the selective upregulation of the K63 ubiquitin ligase, TRAF3. Thus, we provide a mechanistic explanation for the observed antiviral actions of MyD88-dependent TLR2 and further define the role of TRAF3 in viral innate immunity.
| In response to viral infection, cells of the innate immune system synthesize and release members of the type I interferon protein family. The interferons form an essential line of defense, both by slowing viral growth and by expanding the cellular immune response. The synthesis of interferon is initiated by recognition of viral constituents by one or more innate receptors. Among these receptors, Toll like receptor 2 (TLR2) has been shown to be critical for the immune response to a number of viruses, yet TLR2 only directly initiates Type I interferon production in a very small set of innate immune cells. We have discovered that TLR 2 can contribute to the antiviral interferon response much more broadly by indirectly governing the production of interferon induced by other Toll like receptors as wells as downstream of the cytosolic Rig-I like receptors. This happens through the TLR2-dependent up-regulation of a critical signaling element, TRAF3. We also demonstrate that this TLR2 dependent regulation of interferon may be important in biological scenarios involving co-infection of virus and Gram positive bacteria, but not Gram negative bacteria.
| The last few years have seen an explosion in the characterization of mechanisms for the recognition of microbial pathogens by the innate immune system. In particular, sensors that recognize molecular signatures of viral infection have been the subject of many exciting discoveries. Among the currently known innate immune antiviral sensors are the cytosolic RNA receptors, Retinoic acid-inducible gene 1 (RIG-I), and Melanoma differentiation-associated protein 5 (MDA5) [1]–[3], as well as, DDX21 and DHX36 (DDX/TRIF) [4]. A cytosolic DNA sensing multi-protein complex has been identified that responds to DNA virus infections, although the apical sensors for this pathway have not been fully elucidated [5], [6]. In addition, the nucleic acid sensing endosomal Toll-like receptors (TLRs), i.e., TLR3, TLR7/8, and TLR9, as well as cell surface expressed TLR4, have known virus-derived ligands [7], [8]. A common and critical feature of each of these innate viral surveillance systems is the ability to induce type I interferons (IFNs). IFNs are a family of pleotropic cytokines that are secreted and act in a paracrine or autocrine manner to induce an incredibly diverse array of genes with direct antiviral properties [9], [10]. Additionally, type I IFNs serve as a point of connection between innate and adaptive responses, in that they educate and enhance the adaptive response and lead to viral clearance [11].
Among the TLRs, TLRs 2 and 5 are unique in that they utilize MyD88-dependent signaling exclusively and, thus, do not induce type I IFNs in macrophages and dendritic cells [12]. As a result, these receptors are generally described as being antibacterial. While there are no known virus-derived ligands for TLR5, there are a number of reports in the literature describing TLR2 activation by viral ligands, including a recent report that has shown that viral, but not bacterial, TLR2 ligands may induce IFN in a subset of monocytes by unknown mechanisms [13]. In addition, mouse infection models with several different viruses have shown deficiencies in viral clearance in mice with a targeted deletion in TLR2 [14]–[16]. Presently, the mechanism by which TLR2 contributes to an antiviral state in the absence of direct IFN induction is not clear.
Along with the well documented role that individual TLR and RLR sensors play in responding to infection, it is becoming increasingly clear that individual innate immune sensing systems do not operate in isolation, but that significant cross-talk between innate systems occurs [17]. Indeed, the biological relevance of such cross-talk has already been demonstrated in cases of polymicrobial infection [18], [19].
Herein, we describe a TLR2-dependent mechanism for the governance of subsequent type I IFN production via both the TLR and RLR systems. This pathway shapes antiviral immunity in vitro in murine primary macrophages, and in vivo in mouse models of viral infection. In response to prior stimulation or “priming” with TLR2 ligands, subsequent type I IFN induction via all known IFN-β-inducing innate immune pathways is strongly potentiated. The underlying mechanism for this potentiation was identified as being largely due to the up-regulation of the E3 ubiquitin ligase, TRAF3. These findings not only explain how bacterial or viral TLR2 ligands may selectively augment a subsequent TLR-mediated IFN response to virus, but also reveal a new degree of mechanistic cooperativity between TLRs and the cytosolic RLRs in the host response to virus infection.
To characterize further the effects of TLR cross-talk on the induction of important inflammatory genes, primary mouse peritoneal macrophages were treated with media alone, or media supplemented with ligands for TLR 2 (P3C) or TLR4 (LPS). After overnight stimulation, the primary stimulus was removed and the cells washed extensively and allowed to rest for 60 minutes. The macrophage cultures were next re-stimulated with the TLR4 ligand, E. coli LPS, for 2 or 4 hrs and examined for gene induction by qRT-PCR. LPS induction of both the classical pro-inflammatory genes IL-6 and IL-12 p40 was strongly inhibited by prolonged TLR pre-stimulation (Figure 1, A and B). This is the predicted pattern described previously and known as “homotolerance” or “heterotolerance,” respectively [20]. Unexpectedly, however, when we examined the effect of TLR pre-stimulation on the LPS-mediated induction of type I interferon (IFN-β), we found the nature of the effect to be critically dependent on whether the initial stimulation had come through TLR2 or TLR4. Pretreatment with LPS (TLR4) sharply inhibited subsequent IFN-β induction in response to LPS (Figure 1C). However, pretreatment with P3C (TLR2) markedly potentiated IFN-β by nearly 10-fold at the mRNA level when compared to media-pretreated macrophages (Figure 1C). These TLR2-dependent changes in LPS-mediated gene induction were also observed at the level of protein (Figure 1 E). This pattern of enhanced IFN-β induction was also seen when induced by virally relevant ligands (forthcoming figures; to be presented later). Enhancement of subsequent IFN-β induction by pre-stimulation was independent of IFN signaling as the IFNAR−/− macrophages exhibited a similar pattern of induction (Figure S2)). Additionally, pre-treating macrophages with a panel of other ligands relevant to innate immunity did not reproduce the priming effect seen with TLR2 ligands (Figure S1).
As a role for TLR2 in potentiating Type I interferon has not been described we sought to distinguish the molecular determinants of this effect. TLR2 functions as an obligate heterodimer with either TLR1 or TLR6, depending on the nature of the ligand [21]. We used ligands for TLR2/6 (P2C) or TLR2/1 (P3C) heterodimers in pretreatment experiments, and both potentiated subsequent IFN-β production in response to LPS via TLR4 (Figure 1D). The effect of TLR2 pre-stimulation on IFN-β was long-lived, lasting for at least 72 hrs following the removal of the pre-treatment TLR2 stimulus (Figure 1F). We also observed a requirement for a prolonged initial exposure (at least 8 h) to TLR2 ligands to elicit a measurable increase in IFN-β induction (Figure S3).
The unexpected result that “priming” peritoneal macrophages with TLR2 ligands enhanced TLR4-dependent IFN-β mRNA expression led us to speculate that components of the signal transduction apparatus downstream of TLR4 were being augmented by TLR2 stimulation. To test this hypothesis, we initially profiled MAPK signaling in response to LPS in media-pretreated, LPS-pretreated, or P3C-pretreated macrophages. Stimulation of naïve macrophages with LPS resulted in robust activation of the MAPKs p38, ERK p42/p44, and JNK as evidenced by increased phosphorylation (Figure 2). Pretreating macrophages with LPS overnight completely ablated LPS-driven MAPK activation (Figure 2), a result consistent with previous reports on endotoxin tolerance [20], [22]. Pretreatment of macrophages with the TLR2 ligand P3C also reduced the activation of p38 and ERK, and almost completely inhibited activation of JNK by LPS (Figure 2). These data confirm prior reports that TLR2-mediated “heterotolerance” is less efficient than TLR4 “homotolerance” with respect to MAPK activation [20]. Based on these data, augmentation of MAPK signaling induced by TLR2-mediated “priming” cannot account for TLR2-mediated upregulation of IFN-β mRNA.
Two discrete signaling arms are initiated upon TLR4 signaling and contribute to IFN-β activity [20], [23], [24]. Classical NF-κB activation through the canonical kinase, IKKβ, results in phosphorylation of the transcription factor, p65, that subsequently contributes to induction of not only most proinflammatory cytokines and chemokines, but also IFN-β. Activation of the non-canonical kinase, TBK-1, via the TRIF/TRAM arm of the TLR4 signaling pathway leads to activation of transcription factor, IRF-3, that, together with NF-κB and other transcription factors, activates transcription of IFN-β and other similarly regulated genes. We examined IKKβ activation by phospho-specific Western analysis of medium-pretreated and P3C-primed macrophages. LPS induced strong IKKβ activation in naïve macrophages by 20 min that declined significantly by 60 min (Figure 3A). In marked contrast, LPS induced extremely weak IKKβ activation in the TLR2 (P3C)-pretreated cells that was detected transiently at 40 min (Figure 3A). LPS-mediated TBK-1 phosphorylation in TLR2-primed cells activity was reduced when compared to that achieved in naïve cells, but was not inhibited to the degree as IKKβ (Figure 3B). As neither IKKβ nor TBK-1 activity could account for the increase in IFN-β induction in TLR2-primed macrophages, we sought to examine the effect of TLR2 priming directly on key transcription factors known to be involved in regulating the IFN-β promoter [25]. We observed that phosphorylation of the NF-κB constituent p65 on residue serine 536 was dramatically enhanced in the P3C-pretreated cells when compared to the naïve macrophages (Figure 3C). Unexpectedly, we also observed a highly significant increase in the steady-state levels of total p65 in TLR2-primed cells that likely contributes to the increase in phosphorylated p65 (Figure 3C). Phosphorylation of the transcription factor IRF3 (on serine 396) was increased in P3C-primed macrophages; however, activation was not accompanied by increase in the levels of total IRF3 and, therefore, increased total IRF3 cannot explain the observed increase in phosphorylation (Figure 3D).
The transcription factor c-Jun is a constituent of the heterodimeric transcription factor, AP-1, that has also been shown to be essential for activation of IFN-β transcription [26]. In the case of TLR4, c-jun phosphorylation is also known to be downstream of MAPK signaling. In contrast to p65 and IRF3, LPS-driven phosphorylation of c-jun was significantly reduced in P3C-primed peritoneal macrophages (Figure 3E). This result is in agreement with the previous MAPK results showing a loss of JNK activation in P3C-pretreated macrophages (Figure 2). Both p65 and IRF3 are phosphorylated by their proximal kinases, IKKβ and TBK-1, respectively, while in the cytosol. Following phosphorylation, both p65 and IRF3 enter the nucleus where they are competent to bind to the IFN-β promoter. To examine whether the observed increased phosphorylation of p65 and IRF3 resulted in increased nuclear import, we performed nuclear fractionation of both medium-pretreated and P3C-primed macrophages after stimulation with medium only or with LPS for 60 min, followed by Western analysis for p65 and IRF3 (Figure 3F). We observed significantly augmented nuclear accumulation of both p65 and IRF3 in the P3C pre-treated cells when compared to the medium-pretreated, naïve macrophages (Figure 3F).
Since nuclear accumulation of p65 may not correlate directly with DNA binding, we performed an EMSA using our nuclear fractions from naïve, LPS-pretreated, or P3C-primed macrophages that were re-stimulated with media or LPS for 60 min (Figure 3G). Nuclear proteins from naïve (media-pretreated) macrophages exhibited strong binding to the NF-κB probe when stimulated with LPS (Figure 3G). The binding complex could be super-shifted with antibodies directed against either p65 or p50 (Figure 3G). LPS-tolerized macrophages re-stimulated with LPS showed diminished NF-κB activation, and the mobility of this complex could also be super-shifted by addition of either anti-p65 or anti-p50 antibodies. P3C-pretreated macrophages displayed NF-κB binding activity that was similar in intensity to that seen in LPS-stimulated, medium-pretreated macrophages.
A key question that arises from these observations is the following: why, as in the case of the P3C-primed macrophages, where NF-κB activity is not strongly diminished, do we not see an increase in the activity of all known NF-κB-responsive genes, e.g., IL-6 or IL-12 p40 (Figures 1A and 1B)? One answer may lie in the works of Medzhitov et al. and McCall et al. who showed that in LPS homotolerance, some normally LPS-responsive genes are transcriptionally silenced through chromatin remodeling such that these promoters are no longer able to interact with transcription factors, regardless of the levels of activated transcription factors [27], [28]. Therefore, we hypothesized that if such a state also occurs in TLR2-induced heterotolerance, it may be possible to abrogate the inactive/tolerant phenotype by preventing remodeling during the initial (“tolerizing”) stimulus. To test this hypothesis, primary macrophages were treated with media alone, Pam3Cys, or Pam3Cys in the presence of Trichostatin A (TSA), a known inhibitor of histone deacetylase activity. Each group of cells was then re-stimulated with LPS as before and assayed for the induction of IL-12 p40 by qRT-PCR. As expected (Figure 1B), TLR2 priming of macrophages eliminated induction of IL-12 p40 mRNA by LPS (Figure 4A). Concurrent treatment of macrophages with TSA and TLR2 ligand significantly preserved the LPS inducibility of IL-12 p40 (Figure 4A). P3C-primed macrophages exhibited a significance increase in IFN-β mRNA induced by LPS that was reversed by the presence of TSA. Thus, manipulation of histone deacetylase activity reverses “heterotolerance.”
While the treatment with TSA strongly argues for a role for chromatin remodeling in TLR2 heterotolerance leading to diminished IL-12 p40 mRNA expression, treatment with chemical inhibitors does not directly assay for transcription factor binding. We therefore also performed chromatin immunoprecipitation (ChIP) analysis on medium- vs. P3C-pretreated macrophages (Figure 4, C and D). Following 60 min of LPS stimulation in naïve or P3C-primed cells, cross-linked chromatin was immunoprecipitated with antibody against either IRF3 or p65/RelA. PCR was used to amplify bound chromatin fragments corresponding to the IFN-β enhancer region. Consistent with our results with the EMSA, p65 binding activity was maintained in P3C-primed cells at levels seen in medium-pretreated macrophages. However, IRF3 binding to the IFN-β promoter was significantly enhanced in the P3C-primed cells Figure 4C and D). As an additional control, we assayed for the recruitment of RNA pol II to the IL-12p40 promoter following 60 minutes of LPS stimulation in naïve and TLR2 primed macrophages (Figure S4). TLR2 priming strongly inhibited RNA pol II recruitment to IL-12 p40, arguing for negative chromatin remodeling in the regulation of this promoter.
Having narrowed the likely mechanism for TLR2 priming of Type I interferon to increased transcriptional activity, we interrogated the larger biological significance of this phenomena. Given that the primary function of IFN-β is in antiviral defense, we next examined the effect of TLR2 priming on the course of a model virus infection. To this end, WT primary peritoneal macrophages were infected in vitro with Vesicular Stomatitis Virus (VSV Indiana Strain) at increasing MOI, without or with prior P3C priming. Additionally, we primed IFN-β−/− macrophages with TLR2 ligand and these cells were similarly infected with VSV. Significant inhibition of VSV-induced cytopathic effect (CPE) was observed in cells pretreated with P3C (Figure 5A). The protective capacity of TLR2 priming was dependent on having an intact IFN-β gene (Figure 5A) as evidenced by the failure of P3C to protect IFN-β−/− macrophages. TLR2 priming inhibited viral growth, rather than simply inducing cell death, as plaque titration of VSV showed a >2 log inhibition (Figure 5B, left panel). Quantitation of CPE by measuring the crystal violet stain eluted from cells remaining bound to the plate strongly paralleled quantification of viral replication obtained by plaque assay (Figure 5B, right panel). We next sought to determine whether VSV infection of P3C-primed cells elicited greater production of IFN-β mRNA than seen in naïve cells. To this end, naïve or P3C-primed macrophages were infected with VSV, and IFN-β mRNA was measured by qRT-PCR 6 hours post-infection. As predicted, VSV infection induced greater levels of IFN-β mRNA after pre-stimulation (Figure 5C). Interestingly, TLR2 priming resulted in diminished IL-6 mRNA following infection (Figure 5C). To extend our virus studies further, naïve and primed macrophages were infected with a second RNA virus, influenza. The mouse adapted influenza strain, PR8, also elicited greater production of IFN-β in P3C-primed cells as well as diminished production of IL-6 mRNA (Figure 5D). As both VSV and influenza are relatively small RNA viruses, we also utilized Vaccinia Virus (WR strain) as a model DNA virus. We observed enhanced IFN-β induction by Vaccinia Virus at 8 hours post-infection, although we were not able to detect IL-6 at this time point (Figure 5E).
The significant protection afforded by P3C priming in vitro encouraged us to extend our virus studies in vivo. Female BALB/c mice were administered 500 µl of PBS or PBS containing 100 µg Pam3Cys i.p. Twenty-four hrs later, mice were infected intranasally (i.n.) with 1×106 pfu VSV. Mice were monitored for mortality and morbidity for up to 10 days post-infection. We observed complete protection from a VSV LD50 in mice that received a single priming dose of P3C (Figure 5F; p<0.05). Similar experiments were carried out in which medium- or P3C-pretreated mice were challenged with ∼5000 TCID50 influenza PR8 (∼LD100). P3C-primed mice exhibited a delayed mean time to death upon PR8 challenge compared to unprimed, infected mice (data not shown). To examine the contribution of TLR2 to viral defense in the absence of prophylactic P3C administration, WT or TLR2−/− mice were infected i.n. with a lethal dose of VSV. Loss of TLR rendered animals more susceptible to fatal VSV infection (Figure 5G).
While it is clear that TLR2 can contribute to the regulation of IFN-β and innate antiviral immunity, our previous experiments utilized a pure, synthetic ligand, Pam3Cys, to establish TLR2 priming. Whether such priming could be accomplished by biological organisms in the context of infection has not been shown. To address this issue, peritoneal macrophages were incubated with heat-killed Gram positive bacteria Staphylococcus aureus overnight at an MOI of 1. Macrophages pre-incubated with S. aureus were subsequently stimulated with E. coli LPS or transfected with poly I:C to simulate a viral infection. Macrophages exposed to S. aureus displayed strongly ablated induction of IL-12 p40 and enhanced induction of IFN-β in response to LPS (Figure 6A). Similarly, transfected poly I:C induced far greater IFN-β in bacterially primed cells (Figure 6B). We confirmed these observations with a second clinically relevant Gram positive bacteria, S. pneumoniae (Figure 6B). Our initial experiments demonstrating divergent effects for TLR2 and TLR4 on subsequent type I IFN induction (Figure 1C) led us to speculate that a priming effect for IFN-β might be a common effect of Gram positive, but not Gram negative bacteria. To test this hypothesis, peritoneal macrophages were pre-incubated with media alone or heat-killed E. coli and then stimulated either with LPS or by transfected poly I:C. In marked contrast to Gram positive bacteria, pre-treatment with E. coli strongly suppressed IFN-β mRNA induced by both LPS and transfected poly I:C (Figure 6C). In addition to bacteria, some viruses are known to signal through TLR2, including Vaccinia Virus [13]. We assayed the ability of UV-inactivated Vaccinia Virus to regulate IFN-β production by LPS and transfected poly I:C. Vaccinia Virus was capable of effective priming for both TLR4 and the RLRs (Figure 6D). To demonstrate that in vivo priming by one organism might influence an infection by a second, we injected BALB/cJ mice i.p. with PBS or heat-killed S. aureus 24 hours prior to i.n. infection with VSV. We observed a significant protective effect induced by S. aureus against VSV infection (Figure 6E).
While it is clear that both VSV and influenza infections can involve TLR4 [29], [30], the cytosolic nucleic acid innate immune receptors play a large role in detecting RNA viruses and initiating an IFN response via the TBK-1/IRF3 axis. It was therefore important to determine if the antiviral effects of TLR2 priming are limited to the TLRs, and specifically TLR4, or whether IFN-β induction is potentiated by TLR2 priming through other families of cytosolic pattern recognition receptors. To this end, we examined the expression levels of the known cytosolic RNA sensors MDA5, RIG-I, and MAVS in medium- and P3C-treated macrophages. We observed a striking up-regulation in the expression of MDA5 protein following 24 hrs of P3C stimulation (Figure 7A). We also observed a much more modest, but reproducible, increase in RIG-I protein levels. As seen previously [31], MAVS appears as several bands of differing electrophoretic mobilities, of which only one appears to be increased in P3C-treated macrophages. The increase in abundance of these cytosolic RNA sensors led us to speculate that there may be a functional increase in sensitivity of this system in P3C-treated macrophages. Therefore, naïve and P3C-treated primary macrophages were transfected with High Molecular Weight (HMW) and Low Molecular Weight (LMW) poly I:C to stimulate MDA5 and RIG-I, respectively. As a control, we additionally treated control and P3C-pretreated macrophages with “free” poly I:C to stimulate TLR3. We examined a time course of IRF3 activation by Western analysis in each case (Figure 7, B–D) and IFN-β mRNA induction by qRT-PCR (Figures 7 E–G). HMW and LMW poly I:C, as well as soluble poly I:C, elicited dramatically greater activation of IRF3 in P3C-primed cells, correlating in each case with a strong increase in induction of IFN-β at the mRNA level. Interestingly, in the cases of transfected poly I:C, we did not observe an inhibition of IL-6 induction in P3C-primed cells.
In seeking a common signaling element that might account for the potentiation in IFN-β induction by both TLRs and RLRs, we postulated that the ubiquitin ligase TRAF3 would be a strong candidate to mediate this effect. TRAF3 is immediately upstream of IRF3 activation in both the TLR and RLR pathways and TRAF3-null immune cells are defective in IFN induction via TLRs [32]–[34]. Initially, we profiled TRAF3 mRNA over a time course of Pam3Cys treatment and found a late increase in TRAF3 consistent with the time of IFN-β potentiation (Figure 8A). We examined the levels of TRAF3 and TRAF6 in the naïve and P3C-pretreated macrophages by Western analysis. Dramatically, we found that protein levels of TRAF3, but not TRAF6, were significantly higher in P3C-primed cells (Figure 8B). As a control, we tested additional innate immune ligands, and found that while TLR3 and TLR4 were capable of inducing TRAF3 mRNA to limited extents, the effect was significantly weaker than for TLR2 ligands (Figure S5). To ascertain whether elevated levels TRAF3 might lead to enhanced signaling complex formation, we performed co-immunoprecipitations from macrophages following 45 min of LPS treatment utilizing an antibody against IRF3 and blotting for co-precipitating TRAF3. We observed significantly greater amounts of TRAF3 in complex with IRF3 in TLR2-primed cells (Figure 8C). While TRAF3 is an essential component of TRIF signaling to IFN-β, it is not known whether an increase in TRAF3 by itself can potentiate TLR4-dependent IFN-β induction and IRF-3 activation. To test this hypothesis, we over-expressed TRAF3 in MAT4 cells, HeLa cells that constitutively express TLR4, stimulated them with LPS, and examined TBK-1 and IRF3 activation. TRAF3 over-expression significantly increased both signaling leading to P-IRF3 (Figure 8D, left panel) and IFN-β gene induction (Figure 8E, left panel). Over-expression of TRAF3 also increased IRF3 activation and IFN-β gene induction induced by transfection of cells with HMW poly I:C (Figures 8D and 8E, left panels respectively). Overexpression of TRAF3 did not significantly affect LPS-driven p65 phosphorylation (data not shown). To complement our over-expression studies, we performed TLR2 priming and LPS re-stimulation studies using littermate control and Lys-Cre TRAF3flox/flox (TRAF3−/−) peritoneal macrophages. Loss of TRAF3 severely compromised LPS-driven IFN-β induction irrespective of prior TLR2 priming (Figure 8F, left panel). The overall enhancement of LPS-dependent IFN-β induction due to priming in TRAF3+/+ macrophages was 4.2-fold compared to unprimed cells. In contrast, TLR2 priming only enhanced IFN-β induction by an average of 1.7-fold in TRAF3−/− macrophages. The loss of TRAF3 had little effect on the induction of IL-12 p40 (Figure 8F, right panel), thus illustrating the specificity of TRAF3 for the interferon inducing pathway.
In conclusion, we have identified a novel mechanism that accounts for the ability of TLR2 to increase antiviral immunity despite an inability to increase IFN levels directly. TLR2 priming of macrophages results in increased TRAF3 levels that lead to the selective priming of IFN-β production by multiple innate immune surveillance pathways and increased viral resistance.
The distinct subcellular localizations and the overlapping ligand chemistries among the receptors in the TLR and RLR families underscores the complementarity between these systems in providing a comprehensive surveillance network against viral infection. While often regarded as a receptor for bacterial lipoproteins, a critical role for TLR2 in antiviral defense has been established in several systems. The human viral pathogens Cytomeglovirus (CMV), Respiratory Syncytial Virus (RSV) and Varicella Zoster all have been shown to either directly encode TLR2 ligands among their structural proteins, or to require functional TLR2 for immune clearance in vivo [14], [16], [35]. Of particular interest for our present study, splenic levels of IFN-β are diminished in TLR2−/− mice infected with mouse CMV [15]. A recent report has extended the connection between TLR2 and the innate response to virus [13]. In this work, using Vaccinia Virus as a model, an as yet unknown ligand on UV-inactivated Vaccinia was shown to be a potent TLR2 agonist. Unexpectedly, while Vaccinia could trigger TLR2-dependent cytokines from all TLR2-expressing cell types tested, it also selectively induced type I IFN in a TLR2-dependent manner, from Ly6Chi “inflammatory monocytes.” However, the molecular mechanism underlying this phenomenon has yet to be fully elucidated. Our current work also provides evidence that UV-inactivated Vaccinia is a potent TLR2 agonist, and we propose that TLR2-based “priming” may be an additional mechanism by which TLR2 contributes to type I IFN production in tissues where innate immune cells other than inflammatory monocytes predominate. Such priming might work systemically during virus infection as free viral proteins capable of acting as TLR2 ligands are shed into the circulation from an infected tissue and travel to uninfected sites and encounter TLR2-expressing innate immune cells. However, it is important to note that a TLR2 priming-based mechanism need not only be of significance during infection with a single viral pathogen. We have demonstrated that TLR- and RLR-mediated IFN-β production is potently enhanced by TLR2 ligands derived from both Gram positive bacteria as well as virus. This could play a role in the pathogenesis of polymicrobial co-infections involving both a bacterial and a viral component. Indeed, synergistic IFN production has already been demonstrated in vivo in a model involving co-infection of Streptococcus pneumonia and influenza A in the upper airways of mice [36]. In this model, Gram positive bacterial infection enhances viral induced IFN-β, that in turn, enhances bacterial growth [36].
Recent literature also suggests that TLR-RLR cross-talk with respect to IFN production may not only occur under dynamic conditions of infection and inflammation. Steady-state priming of innate IFN production in macrophages by unknown components of the intestinal microbiota has been definitively demonstrated [37]. Our data also support the notion that prior ligation of TLR2 leads to a reduction in the subsequent induction of classical pro-inflammatory cytokines such as IL-12 and IL-6, while simultaneously increasing IFN-β responsiveness. In fact, a dynamic balance between IL-12 and IFN-β was recently suggested and shown to depend mechanistically on the levels of IRF transcription factor activation [19]. Whether such a mechanism is operative in our experimental systems remains to be determined.
Our data strongly supports the hypothesis that this potentiation likely occurs, in part, via a TLR-dependent amplification of the levels of the K63 ubiquitin ligase TRAF3. TRAF3 is uniquely positioned at a common node in the IFN-inducing pathways downstream of both TLRs and RLRs, and our data clearly show that TLR2 pre-stimulation greatly increases the amount of IRF3 activation for a given dose of challenge ligand. Precisely how, on an inter-molecular level, this priming affect occurs remains to be elucidated. This question is made all the more interesting, given that the positive effects of TRAF3 in our system take place in a context in which we see diminished TBK-1 activation downstream of TLR4 (Fig. 3B). The elegant studies by Oganesyan et al. [33] and Hacker et al. [32] support a model in which in response to TLR4 ligation, TRAF3 is part of a multi-protein complex that contains both TBK-1 and IRF3 and in which TRAF3 is clearly required for IRF3 activation. Since over-expression of TRAF3 alone is not sufficient to initiate signaling to the IFN-β promoter ([33] and Figure 8D), we propose that the principal effect of elevated TRAF3 levels is to dramatically enhance the kinetics by which TBK-1 can interact with and phosphorylate IRF3. Such a schema would be logical given the enzymatic activity of TRAF3 as a K63 ubiquitin ligase capable of assisting in the assembly of multi-protein complexes. This scenario is the first of which we are aware, in which TRAF3 levels are used as a rheostatic mechanism by which the innate immune response is tuned. The potential cell- and tissue-type specificity of the effects of TLR2 priming, and by extension, TRAF3, in reshaping the transcriptional responses to innate immune receptor ligation remains an area for future investigation. This is particularly relevant given that in contrast to macrophages and dendritic cells, TRAF3−/− B cells exhibit increased responses to TLR ligands [38], [39]. Additionally, TRAF3 has a well-established role as a negative regulator of the non-cannonical NF-κB pathway in B cells [40] that may provide greater complexity to the crosstalk between TLR2 and other pathways such as those downstream of CD40. It should be noted that the presence of residual priming effects of TLR2 ligands in the TRAF3−/− macrophages (Figure 8 F) indicates that additional TLR2-responsive factors beyond TRAF3 contribute to this phenomena and need to be further investigated. For example, one possible additional factor that might contribute to enhanced IFN-β induction could be the cytosolic receptor MDA5, as steady-state levels of MDA5 protein are dramatically up-regulated following TLR2 priming (Figure 7A).
In addition to TRAF3, we found that levels of the NF-κB family member, p65, were greatly enhanced following TLR2 priming. Such TLR-dependent regulation of p65 has not been previously reported, but is analogous to the accumulation of RelB reported in endotoxin tolerance [41]. The molecular mechanism of this p65 regulation remains to be elucidated, but may represent an exciting new area of governance for NF-κB responses. The enhanced accumulation of p65 protein levels may be important in maintaining NF-κB activation downstream of TLR4 in situations of reduced IKKβ kinase complex activity such as is found in TLR2-primed cells (Figure 3A).
It should also be noted that the present work has several important implications with respect to our understanding of macrophage TLR tolerance. Tolerance in macrophages as a result of prolonged exposure to ligands for TLRs has long been seen as a means to limit excessive acute inflammation in cases of disseminated infection. At the same time, not all TLR-responsive genes are “tolerizable” [42], although the mechanism for the discrimination between tolerized and non-tolerized genes is far from clear. Our work suggests that a macrophage made “tolerant” by exposure to TLR2 ligands may, in fact, selectively prime some classes of promoters, such as those responsive to IRF3, by up-regulating certain downstream signaling nodes (e.g., TRAF3) and in this way render the host hyper-vigilant to viral infection.
Animal work performed for this study complied with all applicable provisions of the Animal Welfare Act, the U.S. Government Principles for the Utilization and Care of Vertebrate Animals Used in Testing, Research, and Training, the Public Health Services (PHS) Policy on the Humane Care and Use of Laboratory Animals and the Guide for the Care and Use of Laboratory Animals (8th Ed.). The IACUC protocol governing this work was reviewed by the IACUC committee of the University of Maryland Baltimore School of Medicine. IACUC protocol A3200-01. This committee specifically approved the protocol for this work.
Primary murine peritoneal macrophages were prepared as described previously [20]. Briefly, 3 ml of 3% sterile fluid thioglycollate (Remel) was injected i.p. into 6–8 wk old, wild-type (WT) C57BL/6J mice or BALB/cJ (Jackson Laboratories, Bar Harbor, ME). Four days later, macrophages were harvested by peritoneal lavage with sterile saline. IFN-β-null mice (IFN-β−/−) and MyD88-null mice (MyD88−/−), backcrossed onto a C57BL/6J background (N≥8), were bred in-house as described previously [43]. IFNAR−/− mice and control littermates were a kind gift of Dr. Matthew Frieman (University of Maryland Baltimore). TLR2 null mice (TLR2−/−) were a kind gift of Dr. Rose Viscardi (University of Maryland Baltimore). TRAF3−/− and littermate control MEFs were a kind gift of Dr. Genhong Cheng (UCLA Medical School) and were maintained in DMEM (BioWhittaker) supplemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin, and 100 mg/ml streptomycin. Lys-Cre TRAF3flox/flox macrophages were generated by Dr. Ping Xie (Rutgers University). HEK293T (ATCC, Manassas, VA) cells were cultured in DMEM (BioWhittaker) supplemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin, and 100 mg/ml streptomycin. The RAW 264.7 macrophage-like cell line (ATCC) was cultured in RPMI 1640 (BioWhittaker) supplemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin, and 100 mg/ml streptomycin. MAT4 cells (a kind gift of Dr. Liwu Li, Virginia Tech University, Blacksburg, VA) were maintained in DMEM supplemented as with 293T cells.
Antibodies against phospho-IRF-3 (Serine 396), total IRF-3, phospho-TBK-1 (Serine 172), total TBK-1, phospho-IKKβ (Serine 176/180), total IKKβ, phospho-ERK1/2 (Thr202/Tyr 204), total ERK1/2, phospho-p38 (Thr 180/Tyr 182), total p38, phospho-JNK (Thr 183/Tyr 185), total JNK, phospho-c-jun (Serine 63), total c-jun and phospho-p65 (Serine 536), MDA5, MAVS, RIG-I, and STING were obtained from Cell Signaling (Danvers, MA). Antibody against p50 was purchased from Millipore (Bellirica, MA). Polyclonal antibodies against total p65, C-rel, as well as monoclonal anti TRAF3 were from Santa Cruz Biotechnology (Santa Cruz, CA). Polyclonal anti-TRAF3 antisera was obtained from IMGENEX. Anti DHX58/LGP2 was obtained from Abnova (Taipei City, Taiwan). Protein-free, phenol/water-extracted Escherichia coli K235 LPS was prepared as described elsewhere [44]. S-[2,3-Bis(palmitoyloxy)-(2-RS)-propyl]-N-palmitoyl-(R)-Cys-Ser-Lys4-OH (P3C), R848, CpG, poly dA:T, recombinant Salmonella typhimurium flagellin, and transfection reagent-conjugated high molecular weight (HMW) and low molecular weight (LMW) polyinosinic:polycytidylic acid (poly I:C) were purchased from Invivogen (San Diego, CA). Trichostatin A (TSA) and MG132 were obtained from CalBiochem (Carlsbad, CA). Heat killed S. aureus and S. pneumonia were purchased from Invivogen (San Diego, CA). Strain DH5α E.coli was obtained from Invitrogen (Grand Island, NY) and were heat killed by incubation at 60°C for one hour prior to use.
MAT4 cells were plated at a density of 5×105 cells per well of a six well dish and transfected 24 hrs later with 1 µg of empty vector, or vector that expresses human TRAF3 using Superfect (Qiagen) according to manufacturer's recommended protocol. Twenty-four hr after transfection, cells were stimulated with media alone or media containing 100 ng/ml LPS or transfected with poly I:C. Cells were then washed 3× in PBS and whole cell lysates harvested.
Whole cell lysates from primary murine macrophages or MAT4 HeLa cells were obtained by the addition of lysis buffer (20 mM HEPES, 1.0% Triton X-100, 0.1% SDS, 150 mM NaCl, 10 mM NaF, 1 mM PMSF) and subsequent incubation at 4°C. Cell lysates were separated by electrophoresis in a denaturing SDS-PAGE gel, and subsequent transfer to PVDF membrane. Blots were incubated overnight in relevant primary antibodies at 4°C, washed 3X with PBS, and then incubated with appropriate HRP-conjugated, secondary antibody (Jackson Immunochemicals, West Grove, PA). Blots were developed following incubation in ECL Plus Western Blotting Detection Reagent (Amersham Bioscience, Piscataway, NJ).
Vesicular Stomatitis Virus (Indiana Strain; ATCC) was grown and titered as previously described [45]. For infection of primary macrophages, macrophages were washed 1X with PBS and infected at the indicated multiplicity of infection (MOI) in serum-free RPMI for 1 hr at 37°C with occasional rocking. Infection media was removed and cells were cultured for an additional 24–48 hrs in RPMI containing 2% FBS. For in vivo infection, BALB/c mice were first anaesthetized with isofluorane and subsequently infected intranasally (i.n.) with virus resuspended in PBS as described [43].
Influenza strain A/PR/8/34 (“PR8”) was obtained from the ATCC and propagated as described previously [46] and was the kind gift of Dr. Donna Farber (Columbia University, NY). In vitro infections were conducted as described above for VSV. Vaccinia Virus (Western Reserve) was obtained from the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH: Vaccinia WR from Dr Bernard Moss. In vitro Vaccinia infections were performed as descried for VSV. UV inactivation of vaccinia was carried out using a StrataLinker (Stratagene corp.) at a dose of 5 J/cm2.
Nuclear extracts were prepared using a nuclear extraction kit (Active Motif, Carlsbad, CA) according to the manufacturer's instructions. The NF-κB consensus oligonucleotide, 5′-agttgaggggactttcccaggc-3′, from the murine IgκB light chain gene enhancer was synthesized by the Biopolymer and Genomics Core Laboratory (University of Maryland, Baltimore, MD). DNA probes were 32P end-labeled with T4 polynucleotide kinase (Invitrogen Life Technologies), as recommended by the manufacturer. EMSA was conducted as described previously [47].
Total mRNA was isolated from peritoneal macrophages using TRIZOL (Invitrogen Carlsbad, CA) reagent according to manufacturer's instructions. A total of 1 µg of RNA was utilized in oligo(dT) cDNA synthesis (Promega RT system A3500). qRT-PCR was carried out using an ABI Prism 7900HT Sequence Detection System (Applied Biosystems) utilizing SYBR Green Reagent (Applied Biosystems) and transcript-specific primers. mRNA expression profiles were normalized to levels of the housekeeping gene hypoxanthine-guanine phosphoribosyltransferase (HPRT) in each sample and the fold change in expression was calculated by the 2−ΔΔCt method [48]
ChIP assays were carried out using the Active Motif (Carlsbad CA) ChIP-IT Express Enzymatic kit according to manufacturers instructions.
|
10.1371/journal.ppat.1003787 | Serotonergic Chemosensory Neurons Modify the C. elegans Immune Response by Regulating G-Protein Signaling in Epithelial Cells | The nervous and immune systems influence each other, allowing animals to rapidly protect themselves from changes in their internal and external environment. However, the complex nature of these systems in mammals makes it difficult to determine how neuronal signaling influences the immune response. Here we show that serotonin, synthesized in Caenorhabditis elegans chemosensory neurons, modulates the immune response. Serotonin released from these cells acts, directly or indirectly, to regulate G-protein signaling in epithelial cells. Signaling in these cells is required for the immune response to infection by the natural pathogen Microbacterium nematophilum. Here we show that serotonin signaling suppresses the innate immune response and limits the rate of pathogen clearance. We show that C. elegans uses classical neurotransmitters to alter the immune response. Serotonin released from sensory neurons may function to modify the immune system in response to changes in the animal's external environment such as the availability, or quality, of food.
| The nervous and immune systems respond quickly and precisely to changes in the environment. Communication between these systems may underlie neurological disorders such as depression, and explain why environmental factors, such as psychological stress, increase infection susceptibility. What are the molecular mechanisms that link the nervous and immune systems? C. elegans has a well described nervous system and can be infected by several pathogens, making it an appropriate model organism with which to address this question. We show that synthesis of the neurotransmitter serotonin, in sensory neurons exposed to the environment, alters susceptibility to infection with Microbacterium nematophilum. Unlike infection with Pseudomonas aeruginosa, where serotonin signaling is required for behavioral avoidance, here serotonin signaling suppresses the immune response by regulating a G-protein signaling pathway in epithelial cells. Thus, we show that altering levels of serotonin can trigger different outcomes depending on the environment. We identify a single neuron capable of modifying the immune response and demonstrate that C. elegans uses classical neurotransmitters to modify its immunity. Our work demonstrates that C. elegans can be used to study reciprocal cross talk between neurotransmitters and immune responses that may be important for the pathology of disorders including depression.
| The nervous and immune systems respond quickly and precisely to the presence of pathogenic microbes in an animal's environment. Whilst the immune system activates cellular defenses to recognize and eliminate pathogens, changes in neuronal signaling alter animal behavior to avoid these microbes. Data from mammalian models suggests that bidirectional communication between these two systems can modify responses to infection [1], and this relationship may explain why psychological stress increases susceptibility to infections [2]. Because of the complicated nature of the mammalian brain and immune system the molecular mechanisms that underlie neuronal regulation of the immune response remain unclear.
The free-living soil nematode Caenorhabditis elegans utilizes conserved signaling pathways to trigger behavioral and innate immune responses to infection by several natural and clinically-relevant pathogens provided as a food source [3]. This, together with its simple and well-described nervous system, has resulted in several studies identifying the neuronal signals that influence C. elegans behavioral and immune responses to infection [4]–[9]. In the presence of pathogens such as Serratia marcescens [8], Microbacterium nematophilum [9] and Pseudomonas aeruginosa [5], [6] C. elegans uses chemosensory neurons to recognize the pathogen triggering changes in neuronal signaling that cause it to alter its behavior and avoid these potential harmful bacteria. Interestingly, neuronal signals can also directly modify C. elegans immune responses. Release of the insulin like neuropeptide INS-7 from neuronal dense core vesicles suppresses the C. elegans intestinal immune response triggered by infection with P. aeruginosa PA14 [4] and this pathway is utilized by the pathogen to suppress host immune defenses [10]. The neuronal cytokine DBL-1(TGFβ) promotes expression of the caenacin family antimicrobial peptide cnc-2 in the epidermis during infection with the fungal pathogen Drechmeria conispora [7]. Several studies have implicated mammalian neuropeptides and peptide hormones in neuronal regulation of immunity (reviewed in [11]) suggesting that these relationships are conserved, and studies in C. elegans have identified neurons as important modifiers of the immune response. However, although an octopamine receptor has been shown to regulate the C. elegans immune response [12], the function of neurotransmitters in C. elegans immunity remains unexplored.
In mammals a number of neurotransmitters act on the immune system to modify its function [13]. One of these is the classical monoamine neurotransmitter serotonin (5-hydroxytryptamine) [14], [15]. Dysregulation of mammalian serotonin signaling is associated with mood disorders including depression, and depressed patients show decreased natural killer (NK) cell activity [16]. These cells are important components of the innate immune system linking serotonin signaling to immune regulation. Furthermore NK cell activity can be enhanced by treatment with anti-depressants, such as Prozac, that act as selective serotonin reuptake inhibitors (SSRI's) [17]. NK cells are not the only immune cells affected by serotonin. Several other cells of the immune system express serotonin receptors including dendritic cells [18], macrophages [19] and mast cells [20]. Serotonin affects both innate and adaptive immunity enhancing the proliferation of B [21] and NK cells [22], promoting stimulation of T cells by macrophages [23] and acting as a chemotactic agent for mast cells [20] and eosinophils [24]. However there is still much unknown about how serotonin functions in the immune response. Genetic approaches will be key to understanding the role of serotonin in immune function. Mice lacking the serotonin biosynthetic enzyme, tryptophan hydroxylase TPH1, have some defects in their immune function [25] however TPH1 acts redundantly with TPH2 [26] and TPH1 knockouts still retain the ability to synthesis serotonin in serotonergic regions of the brain [26]. C. elegans only has one tryptophan hydroxylase ortholog, tph-1, and animals carrying the putative null allele tph-1(mg280) are deficient for serotonin production [27] allowing the immune function of neuronal serotonin to be studied in vivo.
In C. elegans serotonin signaling allows animals to respond to changes in their environment by modulating locomotion [28], feeding [29], defecation [30] and egg laying [1], [31], [32] behaviors. Serotonin signals the presence of food causing starved animals to stop moving when they encounter a bacterial lawn. Animals lacking tph-1 behave as if they are starved, decreasing their feeding and egg laying rates [2], [27]. Interestingly serotonin signaling is also required for C. elegans to respond to infection by P. aeruginosa PA14 [3], [5], [6]. Animals that lack tph-1 are more susceptible to PA14 than wild type animals. However serotonin signaling is not required for the C. elegans immune response in this context, and changes in susceptibility of tph-1 mutant animals are due exclusively to behavioral pathogen avoidance [4]–[9]. Exposure to PA14 increases serotonin levels in chemosensory neurons and promotes aversive learning so that animals that have been previously exposed to PA14 alter their olfactory preferences to avoid these toxic bacteria [6], [8].
Using the natural C. elegans pathogen Microbacterium nematophilum we have identified a role of serotonin signaling in suppressing the immune response. Wild type C. elegans tends to avoid bacterial lawns contaminated with M. nematophilum [9] and following infection an immune response is triggered that includes swelling around the rectal opening and upregulation of host defense genes [5], [6], [33], [34]. Unlike during P. aeruginosa infection, serotonin signaling was not require for avoidance of M. nematophilum but instead suppressed the immune response by activating signaling via the G-protein GOA-1(Gαo) in rectal epithelial cells. This suppression required serotonin synthesis in the chemosensory neuron, ADF, which contacts the animal's environment via ciliated sensory endings, and the serotonin receptors SER-1 and SER-7. Our data demonstrates that C. elegans uses the classical neurotransmitter serotonin to modify its immune response. These signals may function to modify the immune system in response to changes in the animal's external environment, such as the availability of food.
Infection of C. elegans with the naturally-occurring pathogen, Microbacterium nematophilum triggers an immune response that includes transcription of host defense genes and swelling around the rectal opening, known as the Deformed anal region (Dar) phenotype [4], [33], [34]. To determine whether serotonin was able to modulate the C. elegans immune response we exposed adult wild type animals to M. nematophilum on plates containing exogenous serotonin and scored the Dar phenotype in their progeny. Treatment with serotonin caused a decrease in the number of Dar animals when compared to untreated controls (Figure 1A and B), however serotonin treatment did not alter the ability of the pathogen to attach to the cuticle. We observed similar levels of SYTO13-labeled M. nematophilum adhering to the rectum of serotonin-treated animals that were Dar-defective and control animals (Figure 1C, D and E).
In these experiments animals were exposed to exogenous serotonin throughout development raising the possibility that serotonin treatment alters development to indirectly affect the Dar phenotype. The Dar phenotype requires signaling in the rectal epithelium [35], [36], therefore we first checked that the rectal epithelial marker, LIN-48 (OVO-like transcription factor) [37], was correctly expressed following treatment with exogenous serotonin. Using transgenic animals expressing lin48p::FP we did not observe any changes in the expression of this rectal epithelial marker following treatment with exogenous serotonin (Figure S1). To confirm that post-developmental treatment with exogenous serotonin still suppressed the Dar phenotype we also infected animals on plates containing exogenous serotonin at different developmental stages and scored the Dar phenotype when they reached adulthood. Although fewer control animals became Dar when they were infected at the L3/L4 stage (as we have previously observed [10], [36]), we were still able to suppress this Dar phenotype by infecting animals in the presence of exogenous serotonin 10–18 hours prior to adulthood (L3/L4 stage), indicating that exogenous serotonin does not indirectly affect the Dar phenotype by altering development (Figure 1B).
Treatment of C. elegans with exogenous serotonin causes dramatic behavioral changes including inhibition of locomotion, stimulation of egg laying and increased pharyngeal pumping [7], [28], [29], [31], [32]. To confirm that defects in these behaviors did not alter the Dar phenotype we infected a number of mutants that phenocopy the effect of exogenous serotonin but do not act in the serotonin signaling pathway. We did not observe any significant differences in the Dar phenotype, when compared to untreated controls, (Table S1) indicating that the effects of serotonin on the immune response are specific and not a secondary consequence of these physiological changes.
Serotonin signaling modulates chemosensory avoidance responses in C. elegans [11], [38]. To ensure that the differences we observed reflected a role for serotonin in the immune response, rather than increased behavioral avoidance of the pathogen in the presence of serotonin, we modified our infection assay by spreading M. nematophilum to the edges of the plate. In this ‘big-lawn’ assay animals were unable to avoid the pathogen. Animals raised on “big lawns” still displayed a decrease in the Dar phenotype (Figure 1A), indicating that serotonin inhibits the immune response directly rather than modifying C. elegans exposure to the pathogen.
Exogenous serotonin treatment inhibits the immune response, but does endogenous serotonin signaling suppress the wild type immune response? The C. elegans tryptophan hydroxlase gene tph-1 is required for endogenous serotonin biosynthesis and animals carrying the putative null allele tph-1(mg280) are deficient for serotonin production [12], [27]. Because increasing serotonin signaling suppressed the immune response we predicted that blocking serotonin signaling, using tph-1 mutants, would enhance the immune response. However, approximately 90% of wild type animals are able to mount a Dar immune response to contamination of a bacterial lawn with 10% M. nematophilum (Figure 1A and 2A), making it difficult to observe treatments that enhance this phenotype. To determine whether loss of serotonin synthesis enhanced the Dar phenotype we modified our infection assay so that bacterial lawns were contaminated with 0.05% M. nematophilum. Under these conditions 60.3% of wild type progeny became Dar (Figure 2A) although 94.65% remained infected as assessed by SYTO13 staining. These conditions allowed us to identify mutations and treatments that enhance the Dar phenotype.
When tph-1(mg280), or another putative null allele tph-1(n4622), animals were grown on lawns contaminated with 0.05% M. nematophilum we observed an increase in the percentage of progeny with the Dar phenotype, when compared to wild type controls (Figure 2A). On lawns contaminated with 10% M. nematophilum wild type controls, tph-1(mg280) and tph-1(n4622) progeny were all over 90% Dar (Figure 2A).
To determine whether the enhanced Dar phenotype of tph-1(mg280) and tph-1(n4622) was due to a decrease in serotonin synthesis we grew animals lacking tph-1 on 0.05% M. nematophilum infection plates supplemented with exogenous serotonin. Treatment with exogenous serotonin was able to rescue the enhanced Dar response observed in tph-1(mg280) and tph-1(n4622) animals (Figure 2A), confirming that wild type levels of serotonin, synthesised by TPH-1, are required to suppress the wild type Dar response.
C. elegans uses behavioral avoidance strategies, as well as immune responses, to promote its survival in the presence of pathogens [6], [8], [13], [39]. tph-1(mg280) and tph-1(n4622) animals exhibit defects in avoidance of Pseudomonas aeruginosa [6], [14], [15] To analyse the role of serotonin synthesis in M. nematophilum avoidance behavior, a phenotype that is dependent on locomotion, we used the tph-1(n4622) allele, since the tph-1(mg280) allele has a background mutation that affects locomotion [40]. Over 70% of wild type animals were found outside bacterial lawns contaminated with M. nematophilum and this distribution was not altered in tph-1(n4622) animals (Figure S2A). Furthermore, both wild type and tph-1(n4622) animals showed a strong preference for E. Coli vs M. nematophilum in food choice assays (Figure S2B and C). Together our data indicates that serotonin signaling is not required for avoidance of M. nematophilum.
TPH-1 is expressed in the serotonergic neurons ADF, NSM and HSN and occasionally in AIM and RIH [16], [27]. To determine the site of action for TPH-1 in regulating the Dar phenotype we performed rescue experiments using a TPH-1 cDNA expressed from either the ADF-specific srh-142 promoter [17], [41] or the ceh-2 promoter that is expressed in NSM neurons [18], [42]. Expression of TPH-1 cDNA in ADF chemosensory neurons, but not in the neurosecretory motor neuron NSM, restored the number of Dar animals to wild type levels in tph-1(mg280) animals grown on lawns contaminated with 0.05% M. nematophilum (Figure 2B). TPH-1 cDNA expressed in both the ADF and NSM neurons of tph-1(mg280) animals did not enhance this rescue (Figure 2B). These results suggest that serotonin synthesis by TPH-1 in the ADF chemosensory neurons is able to inhibit the Dar phenotype in wild type animals.
We next asked how the Dar response is advantageous to infected C. elegans and whether this advantage could be suppressed by serotonin's effects on the Dar phenotype. M. nematophilum is found associated with the cuticle around the C. elegans rectal opening [19], [33] and the Dar phenotype may protect animals from severe infection by distorting the animal's anal region, allowing more rapid clearance of the pathogen from the rectum. Using the vital dye Syto13 to label M. nematophilum attached to the rectal opening, we monitored clearance of the pathogen from the rectum following infection. Wild type animals were able to clear more than 50% of the Syto13 labeled M. nematophilum within 90 minutes of transfer to plates without any bacteria (Figure 1F). Animals that had cleared the SYTO13 labeled M. nematophilum after 90 minutes remained Dar (Figure S3A–C), indicating that loss of pathogen from the rectal opening is not sufficient to reverse the Dar phenotype.
Many dar-defective mutants fail to show any SYTO13 staining in the presence of M. nematophilum, indicating that these genes are required for pathogen recognition and binding [3], [20]. However, a second class of dar-defective mutants remain SYTO13 positive [3], [21], [36]. This second class of genes act downstream of pathogen binding to trigger the Dar response to infection. Using this class of Dar-defective mutants (mpk-1(ku1), unc-73(ce362) and those described in this paper) we repeated our clearance assays. When Dar and Dar-defective animals were scored indifferently for the presence of SYTO13 labeled M. nematophilum we observed a significant decrease in the ability of these mutants to clear labeled pathogen when compared to wild type controls (R. McMullan and A. Anderson, data not shown), demonstrating that the Dar phenotype protects C. elegans during infection, at least in part, by increasing the rate of pathogen clearance.
Using this clearance assay we first asked whether exogenous serotonin treatment of wild type animals; which decreases the percentage of Dar animals, was able to alter the rate of pathogen clearance. Wild type animals treated with serotonin cleared the pathogen infection significantly more slowly than control animals (Figure 1F), indicating that exogenous serotonin is able to inhibit the immune response. The Dar-defective phenotype was observed in approximately 43% of wild type progeny when animals are infected in the presence of exogenous serotonin. To confirm that this Dar-defective phenotype was associated with slower pathogen clearance rates we repeated our clearance assay separating Dar and Dar-defective animals prior to SYTO13 labeling. Almost all animals were SYTO13 positive immediately following labeling and the rate of pathogen clearance in Dar animals was similar to that observed in untreated wild type animals (Figure S3D). However pathogen clearance rates were significantly slower in Dar-defective animals (Figure S3D) confirming the ability of exogenous serotonin to decrease the Dar phenotype results in inhibition of the immune response and decreased pathogen clearance.
To determine whether endogenous serotonin was also able to alter pathogen clearance rates we performed clearance assays using tph-1(mg280) and tph-1(n4622) animals lacking endogenous serotonin synthesis. Interestingly, although these animals show a wild type Dar response when infected with 10% M. nematophilum (Figure 2A) both tph-1(mg280) and tph-1(n4622) animals cleared SYTO13 labeled M. nematophilum significantly faster than wild type animals (Figure 2C). This is consistent with our observation that tph-1(mg280) and tph-1(n4622) animals grown on lawns contaminated with 0.05% M. nematophilum have a larger percentage of Dar animals than wild type controls and suggests that TPH-1 activity in wild type animals inhibits the immune response triggered by infection by M. nematophilum.
To determine whether the enhanced immune response of tph-1(mg280) and tph-1(n4622) was due to a decrease in serotonin synthesis we grew animals lacking tph-1 on infection plates supplemented with serotonin. Treatment with exogenous serotonin was able to rescue the increased clearance rate observed in tph-1(mg280) and tph-1(n4622) animals (Figure 2C) confirming that endogenous levels of serotonin synthesis by TPH-1 are required to suppress the wild type immune response.
To determine where TPH-1 activity was required to reduce pathogen clearance rates we again expressed TPH-1 cDNA in either ADF or NSM neurons, or both. Expression of a TPH-1 cDNA in ADF chemosensory neurons, but not in the neurosecretory motor neuron NSM, was sufficient to rescue the rate of M. nematophilum clearance to wild type rates in tph-1(mg280) animals and this rescue was not enhanced when TPH-1 was expressed in both ADF and NSM (Figure 2D).
Serotonin signaling in ADF chemosensory neurons suppresses the epithelial immune response to infection with M. nematophilum, raising the possibility that these neurons are able to sense and respond to the presence of pathogen by modifying their serotonin signaling. Serotonin levels in ADF can be altered by regulating transcription of tph-1, or via post-translational mechanisms that alter TPH-1 activity. Transcription of tph-1 in ADF neurons is increased by exposure to pathogenic Pseudomonas aeruginosa PA14 [5], [6], [22], neuronal activity [23], [41] and heat stress [20], [43] and these changes can be monitored in vivo using strains expressing a fluorescent transgene under the control of the tph-1 promoter. To determine whether infection with M. nematophilum altered transcription of tph-1 we propagated animals stably expressing a tph-1p::DSRED transgene on small bacterial lawns contaminated with virulent, or avirulent, forms of M. nematophilum. Infection with virulent, or avirulent, forms of M. nematophilum did not alter the expression pattern of tph-1p::DSRED (data not shown) however infection increased expression levels of tph-1p::DSRED in ADF and NSM neurons relative to animals propagated on E. Coli alone (Figure 3 and Figure S4).
Wild type animals tend to avoid bacterial lawns contaminated with virulent M. nematophilum ([9], [24] and Figure S3) therefore it is possible that increased tph-1p::DSRED expression in the presence of M. nematophilum is caused by reduced contact with the bacterial lawn. To test this we first expressed the tph-1p::DSRED transgene in egl-30(ad805) animals that fail to avoid M. nematophilum contaminated lawns [36]. We did not observe any significant increases in tph-1p::DSRED expression in these egl-30(ad805) animals (Figure 3 and Figure S4). We also repeated our tph-1p::DSRED measurements using animals infected in a modified “big lawn” assay where they were unable to leave the food. Again we were unable to observed any significant increases in tph-1p::DSRED expression when animals were unable to avoid M. nematophilum contaminated lawns (Figure 3 and Figure S4). Together these results indicate that the changes in TPH-1 expression levels we observed were largely due to reduced contact with the bacterial lawn when it was contaminated with virulent M. nematophilum and not due to infection.
Several receptors have been identified that bind serotonin in C. elegans, including a serotonin gated chloride channel (MOD-1) and several G-protein coupled receptors (GPCRs) (SER-1, SER-4 and SER-7) [25], [44]–[46]. Infection of mod-1(ok103), ser-1(ok345), ser-7(tm1325), ser-7(ok1944), ser-4(ok512) mutants and ser-1(ok345);ser-7(tm13325) double mutants indicates that at least two GPCRs (SER-1 and SER-7) are required for serotonin effects on the immune response (Figure S5).
GPCRs activate intracellular signaling via specific G-proteins. In C. elegans the effects of serotonin on neuronal activity at the neuromuscular junction are mediated by the G-protein GOA-1(Gαo) [26], [47]. Therefore we asked whether GOA-1(Gαo) signaling was also required for serotonin-mediated inhibition of the immune response. We first infected several available goa-1(Gαo) mutants (goa-1(sa734), goa-1(n1134) and goa-1(n363)) with M. nematophilum as adults or larvae to determine whether loss of GOA-1(Gαo) signaling altered the Dar phenotype. For reasons that we were unable to determine, the majority of these animals arrested at the L1/L2 larval stage and failed to reach adulthood. Therefore we were unable to score the Dar phenotype of these animals (data not shown).
Genetic and biochemical experiments have shown that the conserved Regulator or G-protein signaling (RGS) protein, EGL-10, is a specific inhibitor of GOA-1(Gαo) activity [26], [48]. Therefore we used animals carrying null mutations in egl-10 to increase goa-1(Gαo) signaling indirectly, mimicking the effects of too much serotonin signaling. When egl-10(md176) and egl-10(n692) adults were infected with M. nematophilum we observed a significant decrease in the percentage of progeny exhibiting the Dar phenotype (Figure 4A). However similar levels of SYTO13-labeled M. nematophilum were still observed adhering to the rectum of these animals indicating that increased GOA-1(Gαo) signaling did not alter the ability of the pathogen to attach to the cuticle (Figure 4B and C). Furthermore, this decrease in the Dar phenotype was not caused by increased behavioral avoidance of the pathogen because the Dar phenotype was still decreased when egl-10(n692) were infected using a modified ‘big-lawn’ assay where they were unable to avoid the pathogen (Figure 4A). Consistent with this decrease in the Dar phenotype we observed that egl-10(n692) mutants cleared pathogen infections from their rectal opening more slowly than wild type (Figure 4D), indicating that activating GOA-1(Gαo) is able to suppress the immune response.
The Dar phenotype requires activation of multiple signaling pathways in the C. elegans rectal epithelium [27], [35], [36]. To determine the site of action for EGL-10 (and therefore GOA-1(Gαo)) in the immune response we performed rescue experiments using EGL-10 cDNA expressed in the rectal epithelial cells, using a 1.3 Kb egl-5 promoter fragment [28], [49]. Expression of EGL-10 cDNA in the rectal epithelium was sufficient to rescue the defective Dar phenotype in egl-10(n692) animals (Figure 4A). In addition rectal epithelial expression of EGL-10 rescued the slow clearance of pathogen in egl-10(n692) animals. Indeed these animals cleared labeled pathogen faster than wild type animals (Figure 4D) suggesting that overexpression of EGL-10 cDNA in these transgenic animals was able to decrease GOA-1(Gαo) activity and indicating that GOA-1(Gαo) signaling acts in the rectal epithelial cells of wild type animals to inhibit the immune response. To confirm that EGL-10 was required for the Dar response to infection rather than development of the rectal epithelium we also performed rescue experiments in egl-10(n692) mutants using a heat shock inducible EGL-10 cDNA transgene. We were able to partially rescue the Dar phenotype in egl-10(n692) mutants by expressing EGL-10 cDNA 10–18 hours prior to adulthood (L3/L4 larval stage), indicating that GOA-1(Gαo) signaling in adult animals is required for this response (Figure 4A).
Animals lacking GOA-1(Gαo) are resistant to the effects of exogenous serotonin treatment on egg laying and locomotion [29], [30], [47], [50]. To determine whether serotonin inhibits the immune response via a GOA-1(Gαo) signaling pathway we first treated adult animals with decreased rectal epithelial GOA-1(Gαo) signaling, (by overexpressing the RGS, EGL-10 in the rectal epithelial cells of wild type animals), with exogenous serotonin and scored the Dar phenotype of their progeny. Addition of exogenous serotonin during M. nematophilum infection suppressed the Dar phenotype of wild type animals but not those overexpressing EGL-10 (Figure 5A). Furthermore exogenous serotonin was no longer able to reduce the rate of pathogen clearance when EGL-10 was overexpressed in rectal epithelial cells (Figure 5B). Together these data indicate that wild type levels of GOA-1(Gαo) signaling in the rectal epithelium are required for serotonin's inhibitory effects on the immune response.
We also asked whether increasing GOA-1(Gαo) signaling, using egl-10(n692) mutants, was able to rescue the higher pathogen clearance rate we observed in tph-1 mutant animals unable to synthesis serotonin. Pathogen clearance rates in tph-1(mg280);egl-10(n692) and tph-1(n4622);egl-10(n692) double mutants were indistinguishable from egl-10(n692) animals (Figure 5C). We also observed a significant decrease in the Dar phenotype in these double mutants that was similar to the decrease observed in egl-10(n692) (Figure 5A) further indicating that GOA-1(Gαo) acts downstream of serotonin to inhibit the immune response.
Taken together our data indicates that serotonin synthesized in ADF chemosensory neurons acts via GOA-1(Gαo) signaling in the rectal epithelium to suppress the immune response (Figure 6).
The Dar phenotype requires activation of multiple signaling pathways in the C. elegans rectal epithelium [30], [35], [36]. We have previously shown that signaling via the G-protein EGL-30(Gαq) is required for the Dar response to infection with M. nematophilum [36]. To determine whether serotonin signaling acts in this EGL-30(Gαq) pathway, or a distinct pathway, we used transgenic animals overexpressing a constitutively active form of EGL-30(Q205L) (Gαq) in the rectal epithelium (RE::EGL-30*). When this transgene is expressed in wild type animals approximately 49% of them develop the Dar phenotype in the absence of infection (Table 1 and [36]) although all animals expressing this transgene are still able to trigger the Dar phenotype when infected with M. nematophilum (data not shown).
We asked what effect blocking serotonin signaling, in two different ways, had on the Dar phenotype caused by RE::EGL-30*. Firstly, we blocked serotonin signaling by treating uninfected RE::EGL-30* animals with the serotonin receptor antagonist methiothepin. Treatment of adult RE::EGL-30* animals with methiothepin increased the percentage of Dar progeny from 49.5% to 76.5% (Table 1). Secondly, we expressed RE::EGL-30* in tph-1(mg280), or tph-1(n4622), animals in the absence of infection. We observed an increase the percentage of Dar animals from 49.5% in wild type animals to 75.3% in tph-1(mg80) and 76.5% in tph-1(n4622) animals (Table 1). This increase could be rescued by expression of the TPH-1 cDNA in ADF, but not NSM, neurons of tph-1(mg280) (Table 1). These results support our previous observations that blocking serotonin signaling increases the Dar phenotype when animals are infected with 0.05% M. nematophilum. We also asked what effect blocking serotonin signaling had on the Dar phenotype of infected egl-30(ad805) animals. These animals are strongly Dar-defective [36] and we did not observe any significant differences in the Dar phenotype of egl-30(ad805) animals and the tph-1(n4622);egl-30(ad805) double mutant (Table 1). Taken together these results indicate that serotonin is acting either downstream of, or in parallel to, the EGL-30(Gαq) signaling pathway.
We next asked what effect activating serotonin signaling would have on the Dar phenotype triggered by RE::EGL-30*. To do this we grew uninfected RE::EGL-30* animals in the presence of exogenous serotonin for at least one generation. Serotonin treatment did not significantly alter the number of Dar animals in this transgenic strain (Table 1). EGL-30(Gαq) activates a Rho GEF TRIO – RHOA - RAF signaling pathway to trigger the Dar phenotype in response to infection and activation of downstream components of this pathway in the rectal epithelium; using cell-specific overexpression of constitutively active forms of RHO-1(G14V) (RhoA) (RE::RHO-1*) or LIN-45(S312A,S453A) (Raf) (RE::LIN-45*), also results in the Dar phenotype in the absence of infection. We observed no significant decrease in the number of Dar animals when we grew uninfected RE::RHO-1* or RE::LIN-45* animals in the presence of exogenous serotonin for at least one generation (Table 1). This data most strongly supports the hypothesis that serotonin acts upstream of the EGL-30(Gαq) signaling pathway during this response.
During neurotransmission the EGL-30(Gαq) pathway acts antagonistically to GOA-1(Gαo) [47], [51] with GOA-1(Gαo) reported to act upstream of, or in parallel to EGL-30(Gαq) [51], [52]. Our data suggests that this antagonism also exists in the C. elegans rectal epithelium where it plays an important role in modulating the immune response. Therefore we asked whether GOA-1(Gαo) and EGL-30(Gαq) act in the same, or parallel, pathways by expressing RE::EGL-30* in egl-10(n692) animals. Activation of GOA-1(Gαo) signaling in egl-10(n692) did not significantly alter the number of RE::EGL-30*-induced Dar animals relative to wild type animals (Table 1). Furthermore, when two other transgenes that cause the Dar phenotype (RE::RHO-1* and RE::LIN-45*) were expressed in uninfected egl-10(n692) animals the percentage of Dar animals was not significantly different from that observed in wild type animals expressing these transgenes (Table 1). This data supports the model that the serotonin - GOA-1(Gαo) pathway acts upstream of, or in parallel to, the EGL-30(Gαq) pathway to regulate the immune response (Figure 6).
Insulin-like neuropeptides and cytokines, produced in C. elegans neurons, can influence its ability to mount an immune response to pathogen infection [4], [7]. In mammals a number of neurotransmitters can also influence the immune response [13] however the regulation of C. elegans immunity by neurotransmitters remains largely unexplored. Although Sun et. al. [12] have demonstrated that signaling downstream of the neurotransmitter octopamine suppresses the immune response to P. aeruginosa infection, the function of neurotransmitters themselves remains undetermined. Here we show that the classical neurotransmitter, serotonin, suppresses the immune response to infection with the pathogen M. nematophilum. Serotonin synthesized in chemosensory neurons acts, via at least two serotonin GPCR's, to regulate G-protein signaling in rectal epithelial cells and suppress the Dar phenotype. This leads to a reduction in the animal's ability to clear the pathogen infection.
This is the first demonstration of a role for serotonin signaling in regulation of a C. elegans immune response, however serotonin regulates both innate and adaptive mammalian immune responses [14], [15] suggesting further parallels between C. elegans and mammalian immunity.
Interestingly, although serotonin does not appear to modify immune responses triggered by infection with other pathogens, it does alter C. elegans behavioral response to these pathogens. When animals are exposed to Pseudomonas aeruginosa PA14 they modify their olfactory preferences so that they learn to avoid these pathogenic bacteria [6]. This learnt avoidance results in increased survival following infection [5] and requires serotonin signaling in the same chemosensory neuron required to suppress the immune response to M. nematophilum [6], Although C. elegans avoids lawns contaminated with M. nematophilum we did not observed any role for serotonin signaling in this behavioral response. Thus serotonin signaling in the same neuron can trigger distinct responses, requiring different target cells, depending on the environment C. elegans encounters. It is possible that different levels of serotonin are required to trigger these different responses. Alternatively signaling in other neurons may be regulated by infection. Changes in the activity of these neurons may modify how downstream target cells respond to the presence of serotonin.
Although C. elegans neurons have been shown to produce insulin-like neuropeptides and cytokines that influence the immune response [4], [7] the identity of these neurons has remained largely elusive. Here we identify a single neuron that influences C. elegans immunity.
In the C. elegans adult hermaphrodite only ADF, NSM, HSN, AIM, RIH and VC4/5 neurons have been shown to contain serotonin [28], [53], [54] and the tryptophan hydroxylase TPH-1, required to synthesis serotonin, is only expressed in ADF, NSM and HSN [27], suggesting that AIM, RIH and VC4/5 take up serotonin synthesized by other neurons. Using cell specific rescue experiments we have shown that expression of TPH-1 (and therefore serotonin synthesis) in ADF is required to suppress the immune response to infection by M. nematophilum.
ADF neurons are a set of two bilaterally symmetrical chemosensory neurons that contact the external environment via sensory cilia in the amphid [55], raising the possibility that chemical cues present in the environment can alter ADF activity and influence the immune response. What are the environmental signals that trigger ADF activation leading to serotonin synthesis and suppression of the immune response following M. nematophilum infection?
One possibility is that chemical cues produced by the pathogen itself are detected by ADF. Pseudomonas aeruginosa infection increases TPH-1 expression [5], [6] and stimulates ADF neuronal activity [56] to promote behavioral avoidance of the pathogen [6]. In contrast M. nematophilum does not appear to regulate serotonin synthesis in ADF, because we only observed changes in TPH-1 expression under conditions where there was decreased contact with the M. nematophilum contaminated bacterial lawn. However, we cannot exclude the possibility that other aspects of ADF activity, including regulation of TPH-1 activity or serotonin release, are regulated by the pathogen. Regardless of how infection with M. nematophilum regulates serotonin signaling, suppression of this pathway alone is not sufficient to trigger the Dar phenotype. We did not observe the Dar phenotype in uninfected tph-1 mutants or in wild type animals treated with methiothepin (R. McMullan and A. Anderson, unpublished observation) indicating that infection with M. nematophilum must regulate additional signaling pathways (including the EGL-30(Gαq) pathway) to trigger the immune response.
Another environmental cue that may alter serotonin synthesis to suppress the immune response is the presence of food. In C. elegans serotonin signals the presence of food [28], and because ADF neurons directly contact the environment it has been suggested that they may couple environmental food signals with serotonergic neurotransmission [54]. In support of this we observed that tph-1 mutant animals (that behave as if they were starved in the presence of food), were able to respond to lower levels of M. nematophilum infection than wild type animals. We have also obtained anecdotal evidence that infection causes starved animals to become Dar more easily than well-fed animals (R. McMullan and A. Anderson, unpublished observation) suggesting that the presence (or absence) of food is able to influence the immune response. Furthermore serotonin signaling is able to suppress the Dar response when it is triggered in the absence of infection (using RE::EGL-30*) because decreasing serotonin synthesis (using tph-1 mutants), or signaling (using methiothepin), both increased the percentage of Dar positive RE::EGL-30* animals. One possible explanation for these observations is that in the presence of food, wild type levels of serotonin inhibit the Dar phenotype. It is possible that the presence of food is sensed by ADF, increasing serotonin synthesis and suppressing the immune response. In this way C. elegans would be able to integrate several environmental signals including the availability of food and the presence of pathogenic microbes and respond accordingly.
We have shown that serotonin synthesized and released from the amphid chemosensory neuron ADF in the animal's head acts on rectal epithelial cells located in the animal's tail. How does serotonin released from these cells influence signaling in distant target cells in the tail?
ADF forms synapses with 17 other interneurons and sensory neurons and gap junctions with an additional two sensory neurons (www.wormweb.org). The rectal epithelial cells, where GOA-1(Gαo) signaling is required to suppress the immune response, have not been reported as postsynaptic targets of ADF suggesting that serotonin is not released directly onto these cells. In principle it is possible that infection of C. elegans with M. nematophilum alters the connectivity of ADF such that the rectal epithelium become a postsynaptic target however we did not observe any gross changes in the expression pattern of a tph-1p::DSRED reporter, that is expressed in ADF neurons, following M. nematophilum infection (data not shown) and it seems unlikely that infection would cause such dramatic reorganization of the nervous system. Alternatively, serotonin may diffuse from its release sites to act on distant cells. This is consistent with previous C. elegans work showing that serotonin released from ADF travels extrasynaptically to RIH and AIM interneurons where it accumulates using the serotonin transporter mod-5 [54]. Extrasynaptic serotonin signaling is conserved in the vertebrate brain [57] and may also contribute to serotonin regulation of the mammalian immune response as B cells take up serotonin released from the noradrenergic neurons that innervate lymphatic tissue [58].
Another possibility is that serotonin does not act directly on the rectal epithelial cells. Several studies have placed GOA-1(Gαo) signaling downstream of serotonin in the regulation of locomotion [47], [52]. These studies have defined the site of action for GOA-1(Gαo) as the cholinergic motor neurons located in the ventral nerve cord [47]. Until recently, it remained unclear whether serotonin acts directly on serotonin receptors expressed on these cells to influence GOA-1(Gαo) signaling or whether serotonin simulates interneurons to release signals that activate GOA-1(Gαo) – coupled GPCRs on these motor neurons. Work by Gürel et. al. [59] identifies the serotonin receptors required for control of locomotion as MOD-1 and SER-4. SER-4 and MOD-1 expression were detected in a non-overlapping subset of head and tail interneurons while MOD-1 expression was also found in GABAergic motor neurons located in the ventral nerve cord indicating that serotonin must act indirectly on cholinergic motor neurons to regulate GOA-1(Gαo) signaling and locomotion [59]. Could this also be the case for serotonin regulation of the immune response? At least two serotonin GPCRs; SER-1 and SER-7, mediate the effect of exogenous serotonin on the immune response however expression of these receptors has not been reported in the rectal epithelium. Perhaps M. nematophilum infection alters the expression of these receptors, and in the future determining the expression pattern of these receptors during infection will address serotonin's mechanism of action.
In C. elegans cholinergic motor neurons a network of G-proteins regulates acetylcholine release to alter locomotion [60]. EGL-30(Gαq) and GOA-1(Gαo) act antagonistically to regulate acetylcholine release and control locomotion [51]. egl-30 mutants have decreased acetylcholine release resulting in slower locomotion rates [51] while in goa-1 mutants acetylcholine release and locomotion rates are increased [47]. Animals lacking goa-1 are resistant to the effects of serotonin on locomotion suggesting that slowing responses, triggered by serotonin in the presence of food, may be mediated by GOA-1(Gαo) signaling in cholinergic neurons [47]. We have previously shown that EGL-30(Gαq) is required in the rectal epithelium for the immune response to infection by M. nematophilum [36] and here we show that GOA-1(Gαo) acts antagonistically to EGL-30(Gαq) in these cells to suppress the immune response. Our data demonstrate that the same G-protein network is activated in different tissues (neurons and epithelial cells), to elicit different responses (locomotion and immunity).
In neurons, genetic data places GOA-1(Gαo) and EGL-30(Gαq) in parallel pathways [51] however GOA-1(Gαo) may regulate the activity of the RGS EAT-16, which inactivates EGL-30(Gαq), placing GOA-1(Gαo) signaling upstream of EGL-30(Gαq) [61]. Our genetic data suggests that GOA-1(Gαo) acts either upstream of, or in parallel to, EGL-30(Gαq) in the immune response. The Dar response provides an opportunity to study the interactions between these pathways in a new context and it will be interesting to determine whether the immune response is altered in eat-16 mutants.
Although serotonin acts upstream of G proteins in both neurons and epithelial cells, G-protein signaling activates different downstream signaling pathways in each of these cell types. In neurons EGL-30(Gαq) and GOA-1(Gαo) act antagonistically to control levels of the second messenger diacylglycerol (DAG) [47], [51] however DAG is not required to alter rectal epithelial cell shape and size and trigger the Dar phenotype (R. McMullan, unpublished observation). Conversely the ERK MAP Kinase pathway is required downstream of EGL-30(Gαq) in epithelial cells to trigger that Dar phenotype but does not appear to affect acetylcholine release [36].
Alterations in serotonin signaling have been implicated in multiple neurological disorders including anxiety, depression, autism and Alzheimer's disease [62]–[64]. There is growing evidence of altered immune function in these disorders that may contribute to their pathology (reviewed in [15]). Indeed in 1991 Smith proposed the macrophage theory of depression suggesting that excessive secretion of cytokines caused depression by altering serotonin levels in the brain [65]. Although it is not possible to ascertain the psychological state of model organisms such as C. elegans they have proved extremely useful in investigating the conserved molecular mechanisms that underlie changes in serotonin signaling in response to the environment. We have yet to determine whether the C. elegans immune response is able to reciprocally regulate serotonin signaling however analysis of serotonin-regulated behaviors following infection should begin to address this question. Our work demonstrates that C. elegans can also be used as a model to study the reciprocal cross talk between neurotransmitters and the immune response that may be important for the pathology of disorders such as depression.
C. elegans strains used in this study are detailed in Table S1. Gene ID's for genes used are detailed in Table 2. All strains were cultivated at 20°C on nematode-growth media (NGM) plates seeded with E. Coli OP50, unless otherwise stated, and maintained as described previously [66]. Where indicated Methiothepin (50 µg/ml) or Serotonin creatine sulfate (3.8 mg/ml) was added to NGM before pouring and plates were used within 5 days. Unless indicated adult animals were transferred to drug plates and their progeny were scored.
Plasmids (listed as pRJM) were constructed using standard techniques, and verified by sequencing. Transgenic strains (listed as impEx) were isolated by microinjection of the plasmid together with acr-2::gfp (a gift of J. Kaplan, Massachusetts General Hospital) or unc-17::gfp (a gift of S. Nurrish, University College London, UK) at 50 ng/µl as a marker. In all experiments matched animals not expressing the injection marker were assayed in parallel as a control. Data was only included if the phenotype of non-transgenic animals was comparable to that of the parental strain.
The wild type EGL-10 cDNA was isolated from wild type N2 RNA using standard techniques and verified by sequencing. This cDNA was subcloned into either the pPD49_78 heat shock vector (a gift of A. Fire Stanford University CA) (pRJM174) or a vector driving expression from a 1.3 Kb egl-5 promoter fragment that drives GFP expression in B, K, F, U, P12.pa and three body wall muscles in the posterior [49] (pRJM176). These plasmids were injected at 20 ng/µl into egl-10(n692). impEx031 and impEx020 contain extrachromosomal versions of pRJM174 and pRJM176 respectively. impEx031;egl-10(n692) were backcrossed to remove the egl-10(n692) mutation in order to obtain animals overexpressing EGL-10 in the rectal epithelium.
Assays were performed essentially as described in McMullan et. al. [36] with the following changes. OP50 or CBX102 bacterial cultures were grown to the same optical density in LB and 40 µl was placed on opposite sides of a 60 mm NGM plate. One-day old adult animals were transferred to NGM plates lacking food for 30 minutes and then washed in M9 and allowed to settle before aspiration. A suspension of animals in a drop of M9 was placed equidistant from each bacterial lawn, numbers of animals varied from 25 to 100. Choice index = (number of animals on lawn A- number of animals on lawn B)/number of animals on lawn A+B. In all experiments lawn A was OP50 and B was CBX102 M. nematophilum. Experiments were performed in triplicate and repeated at least three times.
Infection with M. nematophilum was performed as described previously [36]. NGM plates were seeded with either 10% or 0.05% M. nematophilum diluted in OP50 E. Coli. Unless otherwise stated adult animals were transferred from OP50 plates to infection plates and maintained at 20°C. F1 progeny were scored for the presence or absence of the Dar phenotype once they reached L4 or adult stages. In the case of hs::EGL-10;egl-10(n692) animals (Figure 4A) and wild type animals in Figure 1B synchronized populations of L1 animals were obtained by bleaching. These animals were transferred to infection plates as L1's or grown on standard E. Coli OP50 plates for 24 (L2/L3 stage), or 48 hours (L3/L4 stage), before transferring to infection plates. This generation was assayed for the presence of the Dar phenotype when animals reached L4 or adult stages. Between 30 and 50 animals were scored per plate. Experiments were performed in triplicate and repeated at least three times.
For experiments using exogenous serotonin plates were prepared as described above and were seeded with 10% or 0.05% M. nematophilum diluted in OP50 E. Coli. Plates were used within 5 days. Animals were not pretreated with serotonin prior to M. nematophilum infection and serotonin was present throughout infection with M. nematophilum.
For imaging experiments the avirulent M. nematophilum strain UV336 [67] was used and plates were prepared identically to plates seeded with virulent, CBX102, M. nematophilum.
SYTO13 staining was performed as described previously [35]. Following incubation with SYTO13 animals were either transferred to unseeded plates, for clearance assays as described below, or mounted for imaging.
Animals were infected with M. nematophilum as described above and SYTO13 labeling was performed as described previously [35] except that after a 60 minute incubation with SYTO13 10–20 µl of settled, stained worms were transferred to NGM plates lacking food. After drying the number of animals colonized by SYTO13 positive M. nematophilum was scored using a Nikon SMZ1500 microscope with GFP filter. A ring of 150 mM Copper Sulphate, 2% SDS was use to prevent animals escaping from the plates. Unless indicated both Dar and Dar-defective animals were scored indifferently for the presence of SYTO13 labeling. The percentage of Dar animals was scored prior to SYTO13 labeling. To confirm that loss of SYTO13 labeling reflected a loss of M. nematophilum attachment wild type animals were washed from clearance assay plates after 90 minutes and restained with SYTO13. Approximately 10% of animals were SYTO13 positive indicating that M. nematophilum was no longer attached to the rectal opening of the majority these animals. Between 20 and 50 animals were scored per plate. Experiments were performed in triplicate and repeated at least three times. We observed some variability in the rate of pathogen clearance between experiments therefore assays were performed by two different people and data was only included if results were comparable. Furthermore, each graph only contains data from experiments where all genotypes presented were assayed in parallel.
10–20 µl of settled, SYTO13 labeled animals were added to an equal volume of 600 mg/ml 2,3-Butanedione monoxime in M9 and mounted on 2% agarose pads. Adult animals expressing tph-1p::DSRED (vsIs97) were infected with virulent or avirulent forms of M. nematophilum and the first generation progeny were fixed using 4% paraformaldehyde and imaged by mounting on 2% agarose pads.
Animals were viewed on a Nikon Eclipse Ti inverted microscope using a Nikon ×40 objective (for SYTO13 labeled animals) or ×60 objective (for animals expressing tph-1p::DSRED). Images were obtained using Nikon NIS elements BR software. When tph-1p::DSRED expressing animals were used two images were acquired for each animal to allow the fluorescence intensity of ADF and NSM neuronal cell bodies to be measured.
Images were thresholded to highlight SYTO13 staining, or the ADF/NSM cell body, using Nikon NIS elements BR software and the average fluorescence intensity was the average pixel value within this thresholded area. Controls (wild type animals for SYTO13 experiments and uninfected animals for tph-1p::DSRED) were imaged in parallel and experiments were performed on at least three separate occasions. At least 30 animals were imaged per condition.
Expression from the heat shock promoter was achieved using two rounds of heat shock for 60 min separated by 30 min at 20°C. Heat shock was performed at 0, 24 and 48 hours after transfer of L1's to M. nematophilum plates when animals were at approximately L1, L2/3 and L3/4 stage respectively. A heat shock temperature of 33°C was used. Animals were allowed to recover at 20°C before scoring for the Dar phenotype when the animals reached adulthood.
In all cases statistical analysis was performed using Prism 6 (GraphPad Software). The percentage of Dar animals was compared using an unpaired two-tailed t-test or one-way ANOVA followed by Tukey HSP Post hoc multiple comparison test. Food choice index data was compared using a one-way ANOVA followed by Tukey HSP Post hoc multiple comparison test. Clearance assay data was compared using a two-way ANOVA followed by Tukey HSP Post hoc multiple comparison test. * P≤0.05, ** P≤0.01, *** P≤0.001, **** P≤0.0001, n.s. P>0.05
|
10.1371/journal.ppat.1004000 | Cytomegalovirus m154 Hinders CD48 Cell-Surface Expression and Promotes Viral Escape from Host Natural Killer Cell Control | Receptors of the signalling lymphocyte-activation molecules (SLAM) family are involved in the functional regulation of a variety of immune cells upon engagement through homotypic or heterotypic interactions amongst them. Here we show that murine cytomegalovirus (MCMV) dampens the surface expression of several SLAM receptors during the course of the infection of macrophages. By screening a panel of MCMV deletion mutants, we identified m154 as an immunoevasin that effectively reduces the cell-surface expression of the SLAM family member CD48, a high-affinity ligand for natural killer (NK) and cytotoxic T cell receptor CD244. m154 is a mucin-like protein, expressed with early kinetics, which can be found at the cell surface of the infected cell. During infection, m154 leads to proteolytic degradation of CD48. This viral protein interferes with the NK cell cytotoxicity triggered by MCMV-infected macrophages. In addition, we demonstrate that an MCMV mutant virus lacking m154 expression results in an attenuated phenotype in vivo, which can be substantially restored after NK cell depletion in mice. This is the first description of a viral gene capable of downregulating CD48. Our novel findings define m154 as an important player in MCMV innate immune regulation.
| Cytomegalovirus (CMV) has developed diverse tactics to elude the host immune response and guarantee its survival. The signalling lymphocyte-activation molecules (SLAM) family of receptors encompasses a number of adhesion molecules expressed on the surface of leukocytes that play critical roles in both innate and adaptive immunity. In this study, we report that murine CMV drastically reduces the expression of several SLAM family receptors at the cell surface of infected macrophages, most likely as part of its immunoevasion mechanisms. We have identified a murine CMV gene product (m154) that downregulates CD48, a SLAM family member that functions as a ligand of CD244, a molecule involved in the regulation of natural killer (NK) and cytotoxic T cell functions. We show that during infection, m154 targets CD48 for degradation. Moreover, this viral protein contributes to increased MCMV growth during acute infection in the mouse by protecting against NK cell mediated surveillance. These findings are important for better understanding CMV pathogenesis, and provide a novel example of host innate immune subversion by CMV.
| Pathogens have recourse to innumerable tactics for evading host immune surveillance. Viruses, and in particular large DNA viruses such as herpesviruses, are endowed with the capacity to encode multiple products committed to altering, during all stages of their life cycle, several functions of the innate and adaptive immune system. The homeostatic equilibrium achieved between host immune responses and viral immune escape mechanisms empowers these viruses to successfully establish their characteristic lifelong infections. Human cytomegalovirus (CMV), the prototype β-herpesvirus, usually leads to asymptomatic infections in healthy individuals where it remains in a latent state for life, going through sporadic reactivation and leading to severe diseases in immunocompromised patients [1], [2]. The generation of an efficient host-elicited immune response against CMV includes the induction of natural killer (NK) cells, antibody and T-cell mediated responses [3]. As a consequence, CMV has evolved diverse countermeasures to avoid recognition by T cells, allowing it to interfere with the surface expression of major histocompatibility complex class I (MHC class I) and class II and costimulatory molecules, compromising antigen presentation [3]–[6]. Likewise, the virus counteracts NK cell triggering, primarily by suppressing the expression of ligands for activating receptors while preserving engaged inhibitory receptors [7]–[9]. In addition, CMV alters the function of cytokines and their receptors, and interacts with complement factors. While great strides have been made in recent years in identifying CMV inhibitors of immune response mechanisms, current consensus is that among the vast amount of genetic CMV material still requiring a functional assignment, the virus harbours as yet uncovered immunoevasins directed against already known or new immunological targets. Due to the species-specific nature of human CMV (HCMV) replication, infection of mice with murine CMV (MCMV) has proven to be an invaluable model for studying aspects of the biology underlying CMV infection. In this regard, the MCMV system has been widely used to unveil new immunomodulatory molecules and to explore their roles in infection and viral pathogenesis [10].
The signalling lymphocyte-activation molecules (SLAM) family of cell-surface receptors is a distinct structural subgroup of the immunoglobulin (Ig) superfamily differentially expressed on hematopoietic cells and found to play pivotal roles in both innate and adaptive immunity [11]–[13]. Among other activities, SLAM immunomodulatory receptors regulate cell adhesion, cytokine production, and cytotoxicity of NK and CD8+ T cells. The SLAM family currently consists of nine members, CD48, CD84, CD150 (SLAM), CD229, CD244 (2B4), CD319 (CRACC), CD352 (SLAMF6, NTB-A; Ly108), CD353 (BLAME) and SLAMF9. One of the hallmarks of this class of receptors is that they interact with members of the same family via their amino-terminal Ig-V domains. While most of them are typically self-ligands, participating in homophilic interactions, CD48 is a heterophilic receptor for CD244 [14]. The cytoplasmic domain of most SLAM family members carry one or more copies of a distinctive immunoreceptor intracellular tyrosine-based switch motif (ITSM) [13], [15]. Upon receptor engagement, these motifs undergo phosphorylation and recruit with high affinity and specificity adaptor molecules like the SLAM-associated protein (SAP) [11], [16].
In particular, CD48 is a GPI-anchored glycoprotein with expression in a broad range of cells of the hematopoietic lineage, especially on antigen-presenting cells [17]. CD244, the high-affinity counter receptor of CD48 both in humans and mice, is a transmembrane surface glycoprotein with an intracellular tail containing four ITSMs. It is highly expressed on NK cells, and to a lesser extent on other cytotoxic cells such as CD8+ T cells, basophils, and eosinophils. CD244 is an important activating receptor for the regulation of CD8+ T and mature NK cells, promoting cell-mediated cytotoxicity and cytokine release [18]–[20]. Engagement of CD244 by its ligand leads to the polarization and release of cytolytic granules into the contact zone between NK and target cells [21].
SLAM family receptors have been shown to play specific roles in viral pathogenesis. Various morbilliviruses, including the highly contagious measles virus, employ CD150 as the principal receptor to enter into a subset of immune cells, facilitating their spread and contributing to viral-induced immunosuppression [22], [23]. In response to Epstein-Barr virus infection, CD48 is strongly induced on the surface of B lymphocytes and may aid viral trafficking [24]. In addition, we have recently shown that HCMV encodes UL7, a CD229 structural homologue capable of interfering with proinflammatory responses [25]. The role of SLAM receptors in antiviral immunity has been clearly documented in the X-linked lymphoproliferative syndrome, a rare immunodeficiency human disease in which impaired signalling functions of the SLAM receptors, stemming from mutations in the SAP-encoding gene, is associated with an extreme sensitivity to infection with Epstein-Barr virus [26]. Therefore, since SLAM receptors are active components of host immunity, viruses might have evolved immune evasion manoeuvres to specifically ablate triggering of such receptors. Indeed, this is the case for HIV-1, which utilizes Vpu to elude NK cell recognition through the downregulation of NTB-A expression on the surface of infected CD4+ T cells [27]. Whether this is a more generalized phenomenon and what consequences modulating SLAM receptors may cause in the infected host remain unknown.
In this study we show that MCMV infection efficiently decreases the expression of several SLAM family receptors at the cell surface of macrophages, and we pinpoint m154 as the viral downregulator of CD48. We found that m154 helps to debilitate the effectiveness of anti-MCMV triggered NK cell responses, thereby meliorating viral growth in vivo. Thus, we present here a novel strategy evolved by CMV to subvert detection by NK cells during acute infection, based on the modulation of a SLAM family member.
SLAM family members are differentially expressed among hematopoietic cells. As macrophages play a key role in MCMV infection with regard to viral replication, dissemination, and the establishment of latency [28]–[30], and constitute one of the principal effectors of innate immunity, we selected this particular cell type to explore potential SLAM perturbations upon MCMV infection. Using flow cytometry, we first assessed whether CD48, CD84, CD150, CD229, CD244 and Ly108 were present on the surface of thioglycollate-elicited peritoneal macrophages. The lack of a commercially available antibody against CD319, CD353, and SLAMF9 prevented the study of these receptors in this cell type. As shown in Figure 1A, the SLAM receptors CD48, CD84, CD229, and Ly108 were expressed on the macrophage surface, whereas CD150 and CD244 could not be detected. We found however, that CD150 was present at the surface of LPS-treated mouse peritoneal macrophages (data not shown), consistent with earlier studies [31]. Thus, peritoneal macrophages represent an MCMV permissive cell type expressing a number of SLAM receptors, allowing us to examine whether members of this family could be targets for modulation during MCMV infection.
We then infected peritoneal macrophages with MCMV-GFP at a multiplicity of infection (moi) of 2. The use of MCMV-GFP, based on the bacterial artificial chromosome (BAC)-cloned MCMV genome pSM3fr-GFP, which contains a GFP gene inserted within the ie2 locus [32], allowed us to track and selectively analyze infected cells in the cultures. Under these conditions, infection rates reached approximately 50%. At different times (24 h, 48 h, and 72 h) after infection, cells were stained for the surface expression of CD48, CD84, CD229, and Ly108. Notably, MCMV infection resulted in the significant progressive downregulation of all the four receptors analyzed over the course of the infection, when compared to both non-infected cells (GFP negative) from the same culture (Figure 1B) or with mock-infected macrophages (data not shown). Surface reductions in CD84 and Ly108 were already observed at 24 h post-infection (hpi), and at 48 hpi for CD48 and CD229, becoming for all four receptors more pronounced at 72 hpi. Thus, by 72 hpi macrophages demonstrated a dramatic loss in expression of the four SLAM receptors analyzed. As expected [6], a significant surface decrease in MHC class I molecules was also detected in infected cells. Similar results were obtained when experiments were performed with wild-type (wt) MCMV not expressing GFP (data not shown).
We further analyzed the effect of the viral dose on the alteration of SLAM surface expression by infecting peritoneal macrophages at different mois, ranging from 0.5 (∼5% infected macrophages) to 5 (∼70% infected macrophages), with MCMV-GFP. As depicted in Figure 2A, there was a strong dependency on the viral dose for cell-surface reduction of SLAM receptor expression concomitant with the downmodulation of MHC class I, which in turn correlated with the extent of infected peritoneal macrophages.
To determine whether viral gene expression was required for SLAM downregulation, macrophages were treated with UV-inactivated MCMV. The results showed no decrease in CD48, CD84, CD229, or Ly108 surface expression after infection of macrophages for 72 h with the UV-inactivated virus (Figure 2B), indicating that SLAM downregulation could be attributed to specific MCMV genome-encoded products. Moreover, for Ly108, cell-membrane expression levels after infection with UV-inactivated MCMV were even higher than those of uninfected cells, most likely due to the viral-dependent macrophage activation (data not shown).
Altogether these results show that MCMV encodes gene products that efficiently diminish the cell-surface levels of SLAM family members.
Since CD244, the high affinity receptor for CD48, is expressed in NK and cytotoxic CD8+ T cells known to play a prominent role in the clearance of MCMV infection, we decided to further explore the consequences of the cell-surface depletion of CD48, and sought to identify the viral product(s) causing it.
The potential modulators of SLAM receptors would most likely be genes dispensable for viral replication in vitro. Thus, to identify the MCMV gene product(s) that might mediate the downregulation of CD48, we systematically screened the viral genome utilizing a panel of mutant viruses bearing deletions of approximately 10–15 kbp each in non-essential regions. These mutant viruses were based on the BAC-cloned MCMV genome containing the GFP. Peritoneal mouse macrophages were infected with wild-type and mutant MCMVs, and at 72 hpi were tested for surface expression of CD48. After infection with the deletion mutant MCMV-GFPΔm144-m158 (Figure 3A) missing genes extending from m144 to m158, cell-surface CD48 was restored, reaching levels comparable to that of non-infected cells (Figure 3B). As expected, due to the lack of the m147.5 gene this deletion mutant was also capable to revert the cell-surface expression of CD86 [33], whereas it did not significantly affect the downregulation of other SLAM receptors, such as Ly108. At this point, three additional viral mutants, MCMV-GFPΔm144-m148, MCMV-GFPΔm149-m153, and MCMV-GFPΔm154-m157 all containing smaller specific deletions within the m144-m157 region (from m144 to m148, from m149 to m153, and from m154 to m157, respectively) (Figure 3A) were assessed for their capability to interfere with CD48. As shown in Figure 3B, only the MCMV mutant in which the genetic region encompassing m154 to m157 was removed, efficiently relieved CD48 downregulation, while levels of CD86 remained similar to those present in wt MCMV-infected macrophages. CD86, however, was not reduced from the macrophage surface after infection with either MCMV-GFPΔm144-m148 or MCMV-GFPΔm149-m153, mutants that do lack the m147.5 gene. To further narrow down the possible viral CD48 downregulators, we examined two additional viral mutants containing deletions within the m153-m157 genomic region, MCMV-GFPΔm153-m154 and MCMV-GFPΔm155-m157 (Figure 3A and data not shown). Notably, the MCMV mutant lacking m153 and m154 genes, but not the viral mutant missing genes m155 to m157, reverted CD48 downregulation (Figure 3B, and data not shown). As a role of the m153 gene in CD48 cell-surface alteration had been excluded after analyzing MCMV-GFPΔm149-m153, we deduced that the m154 gene product was the one leading to reduced macrophage-surface expression of CD48 during MCMV infection. This observation was confirmed with a viral mutant bearing a deletion in m154, MCMVΔm154 (Figure 3A), which was able to ablate downregulation of CD48 to an extent comparable to that of mock-infected cells, whereas it maintained the downregulation of Ly108 and CD84 (Figure 3C). As Tang and co-workers [34] in a reassessment of global MCMV ORFs using DNA microarray analysis reported two additional small ORFs, m154.3 and m154.4, potentially expressed in infected NIH 3T3 cells, that partially overlapped with ORF m154 and which therefore were interrupted in the deletion mutant MCMVΔm154, a new recombinant MCMV carrying a smaller internal deletion in m154 that preserved intact both m154.3 and m154.4 (MCMVΔm154Int) was generated. In a manner similar to MCMVΔm154, MCMVΔm154Int did not significantly alter CD48 surface levels (Figure 3C). These data further confirmed that the observed rescue of CD48 surface density in infected macrophages was the result of deleting the m154 gene. Thus, we concluded that m154 abrogates the surface expression of CD48.
The m154 gene belongs to the m145 gene family [35], comprised of eleven members, some of which encode molecules that adopt an MHC class I fold [36] and which are known to be involved in the modulation of immune responses. In contrast to other members of this family, the m154 gene has no homology with MHC class I genes. It encodes a 368-aa type I transmembrane protein with a 23-aa putative N-terminal signal peptide, a 300-aa ectodomain, a 23-aa transmembrane domain, and a 22-aa C-terminal cytoplasmic tail (Figure 4A). The ectodomain is a mucin-like domain displaying a striking number of serine (29) and threonine (84) residues that are potential O-linked glycosylation sites, and contains one putative N-glycosylation site (at position 161). A search of the available sequence databases using the m154 deduced amino acid sequence revealed no significant degree of sequence identity between m154 and other known viral or host proteins.
In order to examine m154 expression during the viral infection, we raised a specific monoclonal antibody (mAb; m154.4.113) against the protein, using a peptide corresponding to its cytoplasmic tail as an immunogen. Peritoneal macrophages, either mock-infected or infected for 72 h with wt MCMV, were analyzed by Western blot with this mAb. A single protein band with an apparent molecular mass of ∼60 kDa was detected only in the infected cell (Figure 4B). The migration of the detected protein differed from the predicted size of the mature m154, which is 38 kDa, being highly suggestive of an extensive glycosylation occurring via its copious serine and threonine residues. To identify the expression kinetic class of m154, we infected macrophages in the presence of either the viral DNA synthesis inhibitor phosphonoacetic acid (PPA), which prevents late viral gene expression, or the protein translation inhibitor cycloheximide (CHX), which selectively limits viral gene expression to immediate early genes. As shown in Figure 4B, m154 was not recovered after release from the CHX block in the presence of actinomycin D, whereas under these conditions, the major immediate early MCMV protein IE1 was abundantly found, as expected. m154, however, was readily detected after PPA treatment, indicating that this viral protein is expressed with early kinetics.
Infected macrophages were also examined by indirect immunofluorescence to determine the subcellular localization of m154. The protein was strongly expressed on the cell membrane and to a lesser extent in the cytoplasm (Figure 4C). Biotin-labelling of proteins on the surface of wt MCMV-infected macrophages, followed by immunoprecipitation with the anti-m154 mAb, SDS-PAGE, and subsequent Western blot probed with labelled streptavidin, confirmed the presence of m154 at the cell surface (Figure 4D). Localization of m154 on the cell surface was also observed after ectopic expression of m154. Thus, when 300.19 cells stably transfected with an HA-m154 fusion protein containing the influenza hemagglutinin (HA) epitope tag inserted at the N-terminal end of m154 were analyzed by flow cytometry using an anti-HA antibody, a cell-surface pattern of HA staining was observed (Figure 4E). It must be noted, however, that when expressed in isolation, m154 exerted no overt effects on the surface levels of CD48, which was constitutively expressed in 300.19 cells (Figure 4F). This result suggests the need of additional MCMV encoded proteins or virally induced cellular molecules for m154 to operate appropriately.
To asses the ability of m154 to downregulate CD48 when ectopically expressed in the context of an MCMV infection, we generated a viral mutant (MCMVm154Ectop, Figure 5A) in which we inserted the m154 ORF plus 210 nt of its corresponding putative promoter and 60 nt including its putative polyA signal, into the genome of MCMVΔm154 behind the ie2 ORF. As shown in Figure 5B, MCMVm154Ectop largely reduced CD48 macrophage-surface levels when examined at 72 hpi, time at which m154 could be clearly detected by indirect immunofluorescence in the MCMVm154Ectop-infected cell, where it displayed a distribution comparable to that observed during wt MCMV infection (compare Figure 5C and Figure 4C).
Based on all these findings, we concluded that m154, the MCMV downregulator of CD48, is an early-phase mucin-like protein with a predominant cell-surface localization in the infected cell.
Although MCMV-encoded products with immunomodulatory properties are not believed to play a role in the viral replication cycle, we analyzed whether m154 affected MCMV growth in tissue culture. To this end, single-step growth curves of MCMVΔm154 and wt viruses were determined in mouse embryo fibroblasts (MEFs) and peritoneal macrophages after infection at a low moi. MCMVΔm154 displayed plaque morphologies in MEFs and growth kinetics in both cell types that were indistinguishable from those of wt virus, confirming the lack of involvement of m154 in the viral replication cycle (Figure 6A and data not shown).
In addressing the mechanism by which m154 downregulates CD48, we first considered the possibility that this viral protein may affect CD48 transcription. However, when we compared CD48 RNA levels in wt MCMV-infected macrophages with mock-infected cells by reverse transcriptase (RT)-PCR, we found no substantial change in CD48 mRNA content (Figure 6B). This observation suggested that CD48 expression was being altered through post-transcriptional mechanisms. Therefore, we examined CD48 protein in cell lysates of wt MCMV-infected cells at different time points by using Western blot. As depicted in Figure 6C, CD48 (∼40 kDa band) levels were drastically lower in total cell lysates of infected macrophages, especially after 48 h of infection. This decrease in CD48 occurred concomitantly with the appearance of m154, which was readily detected 48 h after infection, reaching a maximum at 72 hpi and then continuing to accumulate in the infected cell. Thus, the data pointed towards proteolytic degradation of CD48 during MCMV infection. To assess which protein degradation pathway was involved, we used proteasomal or lysosomal proteolysis inhibitors. MG-132 is considered a specific 26S proteasome inhibitor, while leupeptin is a reversible and competitive intralysosomal proteolysis inhibitor that specifically blocks serine and some cysteine proteases. As revealed by Western blotting, treatment with MG-132 was able to significantly restore CD48 expression (Figure 6D). We also observed however, that the presence of leupeptin partially abrogated CD48 degradation. Consistent with these findings, immunofluorescence microscopy assays evidenced enhanced CD48 signals in the wt MCMV-infected macrophages exposed to the two different proteolysis inhibitors (Figure 6E, panels j and n) as compared to that of the untreated-infected cells (Figure 6E, panel f). Moreover, co-localization of m154 and CD48 could be clearly visualized in both MG-132 and leupeptin treated-infected macrophages (see panels l and p in Figure 6E). Altogether, the data indicate that MCMV targets CD48 for degradation, likely using both the proteasome- and the lysosome-mediated mechanisms.
As indicated, the natural ligand for CD48 is CD244, a molecule that is expressed on all NK cells, and to a lesser extent, on other cytotoxic leukocytes. We first sought to explore whether infection of macrophages with MCMV resulted in a reduced recognition by CD244 due to the loss of CD48 on the cell surface. For this purpose, we generated a soluble murine CD244-Fc fusion protein containing the ectodomain of CD244 fused to the Fc portion of the human IgG. As shown in Figure 7A, binding of CD244-Fc fusion protein to macrophages was significantly decreased upon wt MCMV infection. On the other hand, the fusion protein interacted with MCMVΔm154- and MCMVΔm154Int-infected cells in a similar manner to non-infected cells (Figure 7A).
The interaction of CD48 with CD244 increases NK cell activation, triggering cytotoxicity. Thus, by suppressing CD48-surface expression, m154 could help MCMV elude NK cell-mediated immune responses. To ascertain whether this was the case, we compared the degranulation capacity of NK cells after exposure to macrophages infected with wt MCMV or MCMVΔm154. For this purpose, we used a flow cytometric-based assay to measure NK cell-surface expression of LAMP-1 (CD107a). NK cells purified from mouse spleens were incubated with mock-, wt MCMV- or MCMVΔm154-infected cells at a macrophage/NK ratio of 1∶1. As expected, the percentage of CD107a+ NK cells specifically augmented in response to the viral infection as compared to non-infected cultures. No substantial differences could be detected, however, in any of the experiments performed, when we compared the percentage of CD107a+ NK cells incubated with wt MCMV and those incubated with mutant MCMV-infected macrophages (e.g. mock: 5.5%±1.7; wt MCMV: 31.6%±2.9; MCMVΔm154: 26.3%±1.3). In contrast, we observed markedly increased CD107a externalization in the NK cell population responding to the m154 defective MCMV, as indicated by a 2-fold increase in the CD107a mean fluorescence intensity (MFI) on the CD107a+ NK cells exposed to MCMVΔm154-infected cultures as compared to wt MCMV (Figure 7B and 7C). Thus, the mean number of granules discharged by individual degranulating NK cells during stimulation by MCMV-infected macrophages was lower when m154 was being expressed. To evaluate whether the effects on NK-cell responses observed were caused by m154 acting on the CD48/CD244 axis, we performed degranulation assays on co-cultures of NK cells and MCMVΔm154-infected cells pre-incubated with the CD244-Fc fusion protein. As shown in Figure 7D, the CD244-Fc fusion protein partially blocked CD107a surface expression on NK cells exposed to the MCMVΔm154-infected cells, while an irrelevant control Fc fusion protein did not have a significant impact. Together, the results indicate that m154 contributes to confer protection to MCMV-infected macrophages against NK cell attack, and that these effects are mediated, at least in part, through m154 downregulation of CD48.
We reasoned that reduction of CD48 surface expression on antigen-presenting cells may contribute to the host's impaired ability to control viral growth. We therefore sought to explore the impact that m154 plays in the context of an acute viral infection by inoculating BALB/c mice with MCMVΔm154 or wt MCMV. By day 2 after infection, we could observe that mice intraperitoneally (i.p.) inoculated with 2×106 plaque forming units (PFU) of MCMVΔm154 had gained a larger percentage of body weight than mice infected with the same dose of wt MCMV (Figure 8A). Moreover, while wt MCMV-infected animals lost a substantial percentage of body weight by days 4, 6 and 8 after infection, animals infected with MCMVΔm154 did not experienced any weight loss during the course of the assay. Thus, at day 8 after infection the average body weight of mice infected with wt MCMV was 14.4 g±4.0, whereas mice infected with MCMVΔm154 had an average body weight of 18.3 g±4.3 (data not shown). In agreement with the loss of body weight, wt MCMV-infected mice also developed more exacerbated clinical signs of disease, such as ruffled hair, hunched posture and lethargy (data not shown). When we analyzed the frequency of infected peritoneal macrophages, we did not find significant differences between wt MCMV- and MCMVΔm154-infected mice (wt MCMV: 3.0%±0.6; MCMVΔm154: 2.9%±0.2). Neither, the nature of the cellular influx to the peritoneal cavity, as determined by the levels of neutrophils (CD11b+ Gr-1+), macrophages (CD11b+ Gr-1−), T lymphocytes (CD3+) or B lymphocytes (IgM+) appeared to be distinct amongst the two groups of infected animals (Figure S1). We subsequently determined the replication levels of the viral deletion mutant in several target organs of the animals at different days post-infection. As depicted in Figure 8B, while at day 2 post-infection, comparable viral titers were observed in the spleens of wt MCMV- and MCMVΔm154-infected mice, at day 4 after infection, viral titers in MCMVΔm154-infected animals were around 32-, 6-, 9-, and 4- fold lower in spleen, liver, kidney, and lung, respectively, than those found in the same organs of wt MCMV-infected mice. Likewise, at day 8 post-infection, viral loads of MCMVΔm154 were considerably lower in the organs analyzed (kidney, heart, lung, and salivary glands) compared to those of wt MCMV. Indeed, at this time point, viral titers were below the assay's detection limit in a number of the MCMVΔm154-infected animals (Figure 8B). Comparable results were obtained when mice infected with MCMVΔm154Int were analyzed at day 4 post-infection (Figure 8C). Thus, we can conclude that MCMVs lacking the m154 gene are attenuated in all major organs targeted during MCMV infection.
These results, showing the in vivo effects of m154 as early as 4 days post-infection, together with the in vitro data pointing to a contribution of m154 impairing NK degranulation against the infected macrophages, were highly indicative of an MCMV evasion mechanism involving NK cell immune surveillance. Therefore, we decided to examine whether the reduced attenuation of MCMVΔm154 was a consequence of its enhanced susceptibility to NK cells during the in vivo infection. Mice were specifically depleted of NK cells by treatment with rabbit antiserum to asialo GM1. Four days after infection with 8×105 PFU of wt MCMV or MCMVΔm154, mice were sacrificed and assayed for viral loads in spleen and liver, the two predominant organs in which NK cells have been reported to intervene in the control of MCMV. As expected, all mice treated with anti-asialo GM1 antibody had significantly higher viral titers in their spleens and livers as compared to the corresponding untreated control mice (Figure 8C). However, the extent to which these viral titers were elevated following NK ablation was considerably superior in the MCMVΔm154-infected animals (in particular in the spleen, 101-fold) than in animals infected with wt MCMV (9-fold in spleen). Thus, this substantial restoration of MCMVΔm154 replication demonstrates that m154 promotes MCMV growth in vivo by subverting NK cell responses.
For an effective immune response against many viral infections, antigen-presenting cells such as dendritic cells and macrophages must expose a concerted repertoire of receptors that alert T and NK cells for their efficient activation. In this context, distortion of the surface receptor content is a maneuver widely adopted by numerous viruses to elude the immune system and secure an optimal milieu for their replication and dissemination. In this study we show that several cell-surface molecules of the SLAM family, which operate as co-signalling molecules triggering distinct signal-transduction networks in T, NK and antigen-presenting cells, are targeted by MCMV. Notably, CD48, CD84, CD229 and Ly108 get differentially restricted from the cell surface within the window of time it takes for the virus to complete its life cycle and produce productive progeny. Hence, the fact that CMV might have an active interest in interrupting SLAM interactions through the downregulation of the specific receptors/ligands in the infected cell indicates that, at least for the four SLAM members analyzed in our study, engagement of the corresponding receptors/counter receptors should exert prevailing activating signals in key immune cells during infection.
In this study, we decided to further explore in more detail the loss of CD48-surface expression after MCMV infection. CD48, a GPI-anchored molecule with broad expression in hematopoietic cells, is a SLAM receptor not involved in the homophilic interactions distinctive of this family, its natural ligand being CD244 [17]. Accordingly, reduction of CD48 from the surface of MCMV-infected macrophages leads to a drastic decrease in CD244 binding compared to that observed in mock-infected cells. By screening a battery of MCMV deletion mutants, we identified m154 as the viral downregulator of CD48. Thus, deletion of both the complete m154 sequence or of an internal part of this viral gene from the MCMV genome is sufficient for restoring the surface levels of CD48 back to those found in non-infected cells. Moreover, m154 ectopically expressed within the MCMV genome leads to a significant decrease of CD48 on the surface of the infected macrophage. The m154 ORF encodes a type I transmembrane protein containing a remarkable mucin-type extracellular region. By generating a specific mAb (m154.4.113) against the cytoplasmic tail of this protein, we found that m154 expression is initiated in the early phase of infection and continues throughout the infection cycle, a time frame that is concomitant with the progressive downregulation of CD48 in the infected macrophage. In addition, we show that this viral protein preferentially localizes on the surface of the infected cell.
m154 belongs to the m145 family of glycoproteins [35], despite not presenting the MHC class I protein fold characteristic of some family members. Notably, several of the ten members (m17, m145 to m158) that comprise this family have been reported to perform immunoevasive activities [37]. In particular, the m145, m152, and m155 proteins each downregulate one or more ligands of the activating NK cell receptor NKG2D (H60, RAE1, or MULT-1; [38]–[42]). Additionally, m152 causes intracellular retention of MHC class I molecules [43], while m155 reduces cell-surface expression of the costimulatory molecule CD40 [44]. Finally, the m157 protein interacts with Ly49 NK cell receptors and engages both NK activating (Ly49H) and inhibitory receptors (Ly49I) [36], [45], [46]. Thus, m154 can be now added to the group of molecules within the m145 family that operates as an immunoevasin. While it is important to point out that m154 is able to selectively downregulate CD48, since surface molecules like CD86, Ly108, or CD84, which are used as specificity controls, are not affected by m154, we can not exclude, however, that this viral protein also has a multifunctional nature and targets additional immune receptors.
In terms of delineating the mechanism that leads to the loss of surface CD48, we determined that it does not occur at transcriptional level. Instead, our data show that m154 appears to be majorly causing CD48 degradation. The viral protein leads to a major reduction in the total cellular amount of CD48. Through Western blot analysis and immunofluorescence microcopy, we found that treatment of infected cells with either the serine and cysteine protease inhibitor leupeptin or the proteasome inhibitor MG-132 stabilizes CD48 expression in a certain degree, suggesting that both the lysosomal and proteasomal degradation pathways play a role in the downregulation of CD48. In addition, using these proteolysis inhibitors, co-localization of m154 and CD48 could be appreciated in the wt MCMV-infected macrophages. Interestingly, the m154 cytoplasmic tail displays a motif that has been implicated in lysosomal targeting, and two overlapping recognition sites for the adaptor protein AP-2, which is involved in clathrin-dependent endocytosis. While the detailed mechanism by which m154 operates remains to be elucidated, it does not seem to involve the overall trafficking of GPI-linked receptors, as cell-surface expression of other GPI-anchored membrane molecules, such as CD55, are not affected by this viral protein (data not shown).
m154 does not have a counterpart in any other of the CMV species whose genome has been sequenced so far. However, akin to MCMV, we have previously reported that CD48 is also downregulated in HCMV-infected macrophages [47]. Therefore, each CMV might have evolved its own CD48-specific inhibitor, as yet to be identified for HCMV, emphasizing the importance of targeting this molecule to evade NK cell recognition during infection. In addition, and similarly to MCMV, we found that other SLAM receptors are markedly reduced from the cell surface of macrophages upon HCMV infection (Angulo, unpublished observations). Whether the overall downmodulation of SLAM receptors in the infected cell is an inherent and unique property of CMVs, reflecting selection pressures faced in their specific niches, or whether this might be used by other viruses as an immune evasion mechanism, remains to be explored. Notably, CD48 and NTB-A have been also reported to be negatively regulated by HIV, leading to impaired NK cell recognition and lysis of the infected CD4+ T cells, and being the viral accessory protein Vpu identified as the NTB-A downregulator [27], [48].
Increasing evidence indicates that CD244 contributes to the regulation of both NK cell antiviral activity and virus-specific CD8+ T cell functionality in humans and mice [49]–[51]. Engagement of CD244 by CD48 in the NK cell results in the recruitment and clustering of the receptor into lipid rafts, the phosphorylation of the ITSMs within its intracellular tail, and the subsequent association with the adapter molecule SAP [18], [52]–[54]. This triggers a signalling cascade, leading to the formation of the NK cell synapse, which is characterized by the polarized release of cytolytic granules containing perforin and granzymes. The NK cell synapse is most likely critical for activated NK cells to interact in a productive manner with MHC class I-negative target cells and induce potent cell cytotoxicity [55]. On the other hand, CD244 can inhibit NK cell activation in the absence of functional SAP, such as occurs in cells from patients with X-linked lymphoproliferative syndrome [56]. Taken together, these observations indicate that CD244 and SAP modulate the activity of normal NK cells. Here, we specifically show that disruption of the m154 gene in MCMV leads to an enhanced antiviral NK cell response in vitro. In particular, this viral protein limits NK degranulation capacity against MCMV-infected cultured macrophages. Moreover, we present that the NK cell response to cells infected with the MCMV lacking the m154 gene can be partially inhibited by preincubation of the infected macrophages with the CD244-Fc fusion protein. Hence, we infer from these results that by downregulating CD48, m154 may help protect MCMV-infected cells from NK killing. We cannot discard the possibility, however, that m154 might be also be capable of exerting other functions that contribute to these effects.
As expected, the m154 gene is not required for replication in vitro and an MCMV lacking m154 has not an altered growth phenotype in cultured MEFs or macrophages. It is of note that parental and mutant viruses used throughout the study do all derive from MCMV-BAC pSM3fr, containing an mck-2 frameshift mutation associated with reduced ability to infect macrophages and a diminished capacity to attract leukocytes [57], [58]. However, the fact that all of these recombinant MCMVs have the same pSM3fr background makes them comparable at the level of the MCK-2 phenotype. In contrast to the in vitro observations, the severity of the infection of viruses that do not express m154 was significantly impaired in vivo, where they exhibited a substantial restricted growth in all organs analyzed. By day 4 post-infection, the differences in splenic and liver growth between wt MCMV and MCMVΔm154 were around 30- and 10-fold, respectively, consistent with m154 counteracting NK cell responses, which are crucial to the early control of MCMV replication. Accordingly, we show that mice depleted of NK cells with an antibody to asialo GM1, a glycosphingolipid present at high concentrations in this cell population, selectively improved the in vivo replication of the deletion mutant, confirming that the mechanism by which m154 exerts its protective role is NK cell dependent.
Because CD244 expression is not restricted to NK cells, the impact of m154 might have implications that extend beyond the regulation of NK cell function. This receptor is also present at lower levels on other cytotoxic cells, such as CD8+ T cells, γδ T cells, basophils, and eosinophils. In particular, upon interaction with CD48, CD244 helps initiate signalling and cellular cytotoxicity in CD8+ T cells [49]. Hence, one could speculate that other non-NK cell-related mechanisms might be also contributing to the net protective effects of m154 in vivo. However, the fact that NK cell depletion results in the near complete rescue of the m154-deleted MCMV growth phenotype in vivo indicates that, at least under the conditions of early acute infection analyzed here, the potential of this viral protein to influence processes mediated by other immune cell subsets might be relatively minor. It remains to be determined whether additional effects of m154 could be of relevance later during the infection or in other scenarios the virus might encounter.
In summary, our study presents the SLAM family of immunoreceptors as a novel target of manipulation by CMV, adding to the diversity of molecular strategies incorporated by this pathogen to escape immune detection. We have identified for the first time a herpesviral gene implicated in the downregulation of the SLAM member CD48, and documented its protective role in vivo by counteracting NK cell responses. The knowledge gained from the findings reported in this manuscript will contribute to a better understanding of the complex host-CMV interactions and provide additional insights into the functioning of the SLAM receptors in viral immunity. Finally, the identification of novel players that increase the CMV burden early on during infection could prove helpful for the future development of antiviral reagents.
All procedures involving animals and their care were approved (protocol number CEEA 308/12) by the Ethics Committee of the University of Barcelona (Spain) and were conducted in compliance with institutional guidelines as well as with national (Generalitat de Catalunya decree 214/1997, DOGC 2450) and international (Guide for the Care and Use of Laboratory Animals, National Institutes of Health, 85-23, 1985) laws and policies.
The cell lines NS-1 (mouse myeloma) and 300.19 (mouse pre-B) were obtained from the American Type Culture Collection. Cells were grown in RPMI 1640 medium (GIBCO-BRL, Paisley, UK) supplemented with 10% fetal bovine serum (Sigma Aldrich, St. Louis, MO), 100 U/ml penicillin, 100 U/ml streptomycin, 1 mM sodium pyruvate, and 2 mM L-glutamine (GIBCO-BRL). 300.19 cells were maintained in media supplemented with 0.05 mM 2-mercaptoethanol (GIBCO-BRL). Primary mouse embryonic fibroblasts (MEFs) were cultured in Dulbecco's modified Eagle's medium (DMEM; GIBCO-BRL) supplemented as indicated above. Primary macrophages were elicited from peritoneal exudate cells (PECs) following i.p. injection of 1 ml of 3% thioglycollate (Sigma Aldrich) into BALB/c mice. PECs were removed by peritoneal lavage. Cells were plated out at 2×105 cells/ml in supplemented RPMI 1640 medium, and incubated for 2 h at 37°C, 5% CO2, after which nonadherent cells were washed away with phosphate buffered saline (PBS). Macrophage preparations were confirmed by flow cytometry using the markers F4/80 and CD11b (about 95% were F4/80+CD11b+). NK cells were obtained from mouse spleen using the mouse NK cell isolation kit II (Miltenyi Biotec, Bergisch Gladbach, Germany) on an AutoMACS (Miltenyi Biotec).
The BAC-derived MCMV, MW97.01, based on the MCMV Smith strain (ATCC VR-1399) and referred to here as wt MCMV [59], and the MCMV-GFP recombinant virus, a derivative of MW97.01 carrying the GFP gene [32] were used as parental viruses throughout the study. Recombinant strains MCMV-GFPΔ6 lacking genes from m144 to m158 (referred to here as MCMV-GFPΔm144-m158), MCMV-GFPΔ6S1 lacking genes from m144 to m148 (referred to here as MCMV-GFPΔm144-m148), MCMV-GFPΔ6S2 lacking genes from m149 to m153 (referred to here as MCMV-GFPΔm149-m153), MCMV-GFPΔ6S3 lacking genes from m154 to m157 (referred to here as MCMV-GFPΔm154-m157), MCMV-GFPΔm153-m154 lacking genes m153 and m154 (referred to here as MCMV-GFPΔm153-m154), and MCMV-Dm155Dm157FRT lacking genes from m155 to m157 (referred to here as MCMV-GFPΔm155-m157) have been described previously [33], [40]. For the generation of recombinant MCMV lacking m154 ORF (referred to here as MCMVΔm154) a kanamycin resistance (KanR) cassette was amplified from the plasmid pGP704 with primers Dm154Fw (5′- CCC GCC AAT CAC ATT CAC GAG GGG GTG CTC CGA GAT ACG GTC TCG ACC ACA GGA CGA CGA CGA CAA GTA -3′) and Dm154Rv (5′- CAC ATA AGA CTC GTC ATA ACC TTC CCC GAG TGC CAC CTC CCC ACC CTT ATC GTC TCA GGA ACA CTT AAC -3′) (underlined letters are homologous to the MCMV genome). For the generation of recombinant MCMV lacking only ORF m154 and without affecting ORFs m154.3 and m154.4 described in Tang et al. [34], referred to here as MCMVΔm154Int, a KanR cassette was amplified from the pGP704 with following primers: Dm154bFw (5′- CCG CTG CGG ACG CGA TCT CTT CGG CAA CCC CTA GTG CAG GTG CCG TTA GGA CGA CGA CGA CAA GTA A -3′) and Dm154Rv. PCR fragments were inserted into the m154 ORF of the MCMV BAC MW97.01 by red-α, -β, -γ-mediated recombineering [60]. Subsequently, the KanR cassette was excised by FLP recombinase. For MCMVm154Ectop, the m154 ORF plus sequences containing the putative promoter and the polyadenylation signal were PCR amplified using primers m154-ek.for (5′- CGC GTT AAC CCC GTA TAA ACA CCG CAC CAG A -3′) and m154-ek.rev (5′- CGC AGA TCT ATG TCC TGA CAG ATT ATC GTG GT -3′) with MW97.01 DNA as a template. The PCR product was cloned into pOri6K-F5 [61] and then the m154 sequences together with the adjacent KanR cassette (flanked by mutant [F5] FRT sites) was amplified using primers m154-ins.for (5′- AAC CAC GGG TTC TTT CTC TTG ACC AGA GAC CTG GTG ACC GTC AGG AAG AAG ATT CAG TGA CAG GAA CAC TTA ACG GCT GA -3′) and m154-ins.rev (5′- GTC CGA TGA ATA AAA CCT CTT TAT TTA TTG ATT AAA AAC CAT GAC ATA CCT CGT GTC CTC CCC GTA TAA ACA CCG CAC CA -3′). The m154 sequences and the KanR cassette were inserted downstream of the ie2 (m128) ORF of the MCMVΔm154 BAC by red-α, -α, -γ-mediated recombineering followed by excision of KanR as described above. The integrity of the MCMVΔm154, and MCMVΔm154Int, and MCMVm154Ectop genomes was verified by restriction analysis and sequencing. Viral stocks were prepared by infecting MEFs at low moi. Cell supernatants were recovered when maximum cytopathic effect was reached, and cleared of cellular debris by centrifugation at 1.750× g for 10 min. Viral titers were determined by standard plaque assays on MEFs. Peritoneal macrophages were infected with parental MCMV, MCMV-GFP or derived deletion mutants at an moi ranging from 0.2 to 10. Adsorption was for 2 h at 37°C, 5% CO2, including a centrifugal enhancement of infectivity step [62]. Cells were then washed in PBS before fresh medium was added. The percentage of macrophages infected by recombinant viruses not expressing GFP was estimated by indirect immunofluorescence 24 h after infection using the anti-MCMV IE1 mAb Croma 101 followed by goat anti-mouse IgG Alexa Fluor-488. UV-inactivation of virus was performed using a UV crosslinker (HL 2000 hybrilinker, UVP [254 nm UV], Upland, CA) for 3 min at 360 mJ/cm2.
The anti-m154 mAb (clone m154.4.113, IgG2a) was generated by fusing an NS-1 myeloma cell line with spleen cells from a BALB/c mouse immunized three times with the synthetic peptide corresponding to the intracellular tail of m154 (HRWEDDKGGEVALGEGYDESYV) conjugated to KLH (Proteogenix, Oberhusbergen, France). The hybridoma was subcloned at least three times. The following mAbs were obtained from Biolegend (San Diego, CA): anti-mouse CD48-PE, anti-mouse CD48-Alexa Fluor 488, CD84-PE, Ly108-PE, H2-PE, CD150-PE, CD86-PE, F4/80-pacific blue, CD49b-PE/Cy7, CD107a-Alexa Fluor 488, CD3-Alexa Fluor 647, and streptavidin-PE. Biotin anti-mouse 2B4, recognizing mouse CD244, anti-mouse CD229-APC, CD11b-PE, and Gr-1-APC were purchased from Becton Dickinson Bioscience (San Diego, CA), anti-mouse CD48 (HM48-1) from Santa Cruz Biotechnology (Santa Cruz, CA), and IgM-FITC from Southern Biotech (Birmingham, AL). Anti-mouse IgG Alexa fluor 555 and anti-mouse IgG-Alexa Fluor 488 were purchased from Invitrogen (Carlsbad, CA). Anti-rabbit IgG and anti-Armenian hamster IgG (H+L) labelled with horseradish peroxidase (HRP) were obtained from Jackson Immuno Research Laboratories (West Grove, PA), and anti-mouse IgG (H+L) labelled with HRP from Promega (Heidelberg, Germany). Biotin-conjugated anti-HA and anti-mouse β-actin were purchased from Sigma Aldrich. Isotype control IgGs directly conjugated with the corresponding fluorochromes were obtained from Immunotools (Friesoythe, Germany). The MCMV IE1 specific mAb Croma 101 has been previously described [63].
A murine CD244-Fc fusion protein containing the CD33 leader peptide and the Fc region of human IgG1 was obtained by inserting sequences corresponding to the CD244 ectodomain into the mammalian expression vector signal pIg-Tail (R&D Systems, Wiesbaden, Germany) as previously described [64]. The construct corresponding to the Ig fusion protein was subcloned into the expression vector pCI-neo. An NS-1 stable transfectant secreting the CD244-Fc fusion protein was obtained by electroporation and selection using 1.2 mg/ml geneticin (G418) (GIBCO-BRL). Eight million cells were electroporated (280V, 950 µF) with 8 µg of linearized DNA using the Gene Pulser II Apparatus (Bio-Rad, Hercules, CA). The transfected cells were plated in flat-bottom, 96-well tissue culture plates (Costar, Corning, NY) by limiting dilution and the clone producing the highest amounts of fusion protein was cultured in INTEGRA CL 1000 flasks (Integra Biosciences AG, Chur, Switzerland). The supernatant containing the fusion protein was purified using the Affi-Gel Protein-A MAPS II kit (Bio-Rad). An m154 N-terminal HA fusion protein (HA-m154) was constructed using a PCR product of m154 obtained with primers m154BglII (5′- CCA AGA TCT TTG GGT CGT TTA GAG CTT -3′) containing a BglII restriction site, and m154PstI (5′- CTC CTG CAG TCA CAC ATA AGA CTC GTC -3′) containing a PstI restriction site, and DNA extracted from MCMV virions as a template. The PCR product was inserted into the pGEMT vector and a BglII-PstI fragment corresponding to the m154 gene without the signal peptide was then excised and inserted in frame with the HA at the N-terminal end of the m154 protein into the mammalian expression vector pDisplay (Invitrogen) treated with BglII and PstI. 300.19 stable transfectants were obtained using the same protocol as indicated for NS-1 cells, except that plasmids expressing HA-m154 or the corresponding empty pDisplay vector were transfected using the Amaxa Cell Line Nucleofector Kit V (Amaxa AG, Koeln, Germany) according to the manufacturer's protocol, and selection performed with 0.8 mg/ml of G418.
Flow cytometry was performed using standard procedures. Fc fusion protein stainings were performed using 2 µg of biotinylated Ig fusion protein followed by incubation with streptavidin-PE. Samples were analyzed using a FACSCanto II (BD Biosciences) flow cytometer and processed with the accompanying FlowJo software (Tree star Inc, Ashland, OR). Ten thousand cells were counted for each sample. Cell viability was measured using the LIVE/DEAD Fixable Violet Dead Cell Stain Kit (Invitrogen) according to the manufacturer's instructions. The FACSCalibur (BD Biosciences) was used for analysis of the cellular influx to the peritoneal cavity.
Total RNA was isolated from peritoneal macrophages either uninfected or MCMV-infected for 72 h by the TRIzol method (Invitrogen). RT-PCR was then carried out using the SuperScript III First-strand Synthesis System for RT-PCR (Invitrogen) according to the manufacturer's protocol. Briefly, RNA samples were treated with RNase-free DNase I (Promega) for 30 min at 37°C, and the DNase was inactivated at 65°C for 10 min. The RNA was reverse transcribed using oligo(dT) primers at 50°C for 50 min, and reactions were terminated by heating at 85°C for 5 min. The reverse-transcribed products were treated with RNase H for 20 min at 37°C and amplified using specific primers. Primers m154For (5′- CTT GGA TCC ATG CGG GCG ATG TTA CGG -3′) and m154Rev (5′- CTC GGA TCC CAC ATA AGA CTC GTC ATA -3′) were used to amplify a 1116-bp fragment within the MCMV m154 gene, primers mCD48For (5′- ATG TGC TTC ATA AAA CAG GG -3′) and mCD48Rev (5′- TTG TCA GGT TAA CAG GAT CCT GTG -3′) were used to amplify a 726-bp fragment within the murine CD48 gene, and primers β-actinFor (5′- TAT CCT GAC CCT GAA GTA CC -3′) and β-actinRev (5′- TCA TCT TTT CAC GGT TGG CC -3′) were used to amplify a 170-bp fragment within the murine β-actin gene. PCRs were performed under the following conditions: 1 cycle at 94°C for 5 min; 30 cycles of 1 min at 94°C, 1 min at 58°C, and 1 min at 72°C; and 1 cycle at 72°C for 10 min. Control reactions carried out in the absence of RT were used to assess the specific detection of RNA. Amplified products were separated on a 1% agarose gel and visualized by ethidium bromide staining.
For Western blot analysis, peritoneal murine macrophages, either mock-infected or infected with MCMV at an moi of 10 were used. For selective expression of viral immediate-early proteins, cells were incubated from 30 min prior to infection to 4 h post-infection in the presence of CHX (100 µg/ml; Sigma Aldrich), followed by incubation in the presence of actinomycin D (10 µg/ml; Sigma Aldrich) for another 12 h. Selective expression of early genes was carried out by treatment of the cells with PPA (250 µg/ml; Sigma Aldrich) for 72 h. For proteolysis inhibition experiments, cells were treated with 75 µM MG-132 (Sigma Aldrich) or 250 µM leupeptin (Sigma Aldrich) for 6 h and 24 h, respectively, before harvesting. Under the conditions used, MG-132 and leupeptin did not generally affect cell viability as assessed by trypan blue cell staining. At the indicated times after infection for each specific case, samples were lysed in protein sample buffer and boiled for 5 min. Cell lysates were subjected to SDS-PAGE in 10% acrylamide gels and subsequently transferred to nitrocellulose membranes (Protran, Whatman Schleicher & Schuell, Germany). Equal quantities of total protein were analyzed per lane. Membranes were probed using the mAb anti-m154 (clone m154.4.113), mAb anti-IE1 Croma 101, and an anti-mouse IgG (H+L) HRP as a secondary antibody, and mAb anti-mouse CD48 (HM48-1) followed by a HRP-conjugated goat anti-Armenian hamster IgG (H+L) antibody. β-actin was detected using the mAb anti-β-actin and an HRP-conjugated goat anti-rabbit IgG as a secondary antibody. Blots were developed using a SuperSignal West Pico Chemiluminescent Substrate (Pierce, Rockford, IL) according to the manufacturer's protocol.
Peritoneal murine macrophages, mock-infected or MCMV-infected at different mois, were cultured on glass coverslips in 24-well tissue culture plates. When indicated, cultures were exposed to proteolysis inhibitors as indicated for the Western blot analysis. At specific time points after infection, the cells were washed in PBS and fixed and permeabilized using ice-cold methanol and 0.3% Triton X-100 (for IE1 detection), or fixed in ice-cold acetone (for m154 and CD48 detection), and subsequently blocked with 1% bovine serum albumin (Sigma Aldrich; for IE1 detection) or with 20% rabbit serum (Linus) and 6% fetal bovine serum in PBS (for m154 and CD48 detection). The cells were stained with anti-m154 mAb (clone m154.4.113), or with MCMV IE1 mAb Croma 101, using as secondary antibodies a goat anti-mouse IgG (H+L) Alexa Fluor 555 or Alexa Fluor 488, or directly with anti-mouse CD48-Alexa Fluor 488. Nuclei were counterstained with DAPI reagent (Invitrogen). The samples were mounted in ProLong Gold antifade reagent (Invitrogen). Fluorescence images were obtained using a Nikon Eclipse E600 microscope (Nikon, Tokyo, Japan) or an inverted Leica DMI6000B microscope and the LAS AF software from Leica Microsystems (Wetzlar, Germany).
Peritoneal macrophages were surface-labeled with biotin (Sigma Aldrich) and lysed in protein sample buffer. Cell lysates were precleared 3 times for 30 min using protein G Sepharose (GE Healthcare) and incubated overnight with anti-m154 mAb and protein G Sepharose. Immunoprecipitates were washed, eluted, subjected to SDS-PAGE in 10% acrylamide gels and transferred to nitrocellulose membranes. Membranes were probed with streptavidin-POD conjugate (Roche Diagnostics GmbH, Mannheim, Germany) and blots developed as for the Western blot analysis.
Multi-step growth in vitro was analyzed by infecting MEFs or peritoneal macrophages in 24-well plates with wt MCMV or MCMVΔm154 at an moi of 0.025 and 0.2, respectively. After a 2 h adsorption period, cells were washed with PBS and incubated in the corresponding medium supplemented with 3% fetal bovine serum. At specific time points after infection, the amount of extracellular (MEFs) or cell-associated (macrophages) infectious virus present in the cultures was determined as previously described [65] by a standard plaque assay on MEFs.
NK cell degranulation was evaluated using the CD107a mobilization assay. Cultures of peritoneal macrophages, either mock-infected or infected with 10 PFU/cell of MCMV or MCMVΔm154 for 72 h, were incubated for 5 h at 37°C with purified NK cells in the presence of monensin (BD Biosciences) and anti-CD107a-Alexa Fluor 488, at an effector-to-target cell ratio (E/T) of 1∶1. NK cells treated with 0.5 µg/ml ionomycin (Sigma Aldrich) and 50 ng/ml PMA (Sigma Aldrich) were used as a positive control for degranulation. Cells were then washed in PBS supplemented with 2 mM EDTA, stained for 30 minutes at 4°C with anti-CD49b-PE/Cy7, recognizing DX5, and analyzed by flow cytometry. When stated, MCMVΔm154-infected macrophages were pre-incubated with 10 µg/ml of the indicated Fc fusion protein for 30 min at 37°C, cultures washed, and subjected to the CD107a mobilization assay using an E/T ratio of 0.5∶1.
Seven-week-old BALB/c.ByJ female mice were obtained from Harlan (Netherlands) and housed in the vivarium (University of Barcelona) under specific-pathogen-free conditions. Mice were i.p. inoculated with 5×105 or 2×106 PFU of tissue culture-propagated wt MCMV, MCMVΔm154, or MCMVΔm154Int recombinants. When specified, NK cells were depleted by i.p. injection of rabbit antiserum to asialo GM1 (Wako Pure Chemical Industries, Osaka, Japan) at a concentration of 25 µg per mouse, one day before infection and on day 2 after infection. The efficacy of depletion was assessed by cytofluorometric analyses of spleen cells using an anti-mouse pan-NK cell mAb CD49b-PE/Cy7. At designated times after infection, mice were sacrificed, and specific organs were removed and harvested as a 10% (weight/volume) tissue homogenate. Tissue homogenates were sonicated and centrifuged, and viral titers from the supernatants were determined by standard plaque assays, including centrifugal enhancement of infectivity [62] on MEFs. In experiments evaluating the cellular influx to the peritoneal cavity, mice were sacrificed 2 days after infection and cells present in the peritoneal cavity harvested with 5 ml of PBS. Total number of cells were determined with a cell counter, and stained with a combination of mAbs CD11b-PE and Gr-1-APC or mAbs CD3-Alexa Fluor 647 and IgM-FITC to distinguish the different cellular subsets (macrophages [CD11b+ Gr-1−], neutrophils [CD11b+ Gr-1+], T lymphocytes [CD3+] or B lymphocytes [IgM+]). The number of peritoneal macrophages infected in vivo was assessed by IE1 staining of peritoneal lavage-derived macrophages of mice infected for 16 h with 2×106 PFU.
The signal peptide cleavage site and the transmembrane region were predicted by using the SignalP 4.1 (www.cbs.dtu.dk/services/SignalP/) and the TMHMM 2.0 (www.cbs.dtu.dk/services/TMHMM-2.0/) servers, respectively. The N-glycosylation and O-glycosylation motifs of the protein were identified by using the NetNGlyc 1.0 Server, (www.cbs.dtu.dk/services/NetNGlyc), and the NetOGlyc 4.0 Server (www.cbs.dtu.dk/services/NetOGlyc/), respectively.
Analyses were performed with GraphPad Prism software (version 3.03, GraphPad Software, San Diego, CA). Statistical significance of viral titers between experimental groups was determined with the Mann-Whitney test (two-tailed). P-values less or equal to 0.05 (*), 0.01 (**), and 0.001 (***) were considered statistically significant.
|
10.1371/journal.pntd.0002572 | Characteristics of the Human Host Have Little Influence on Which Local Schistosoma mansoni Populations Are Acquired | Brazil remains the country in the Americas with the highest prevalence of schistosomiasis. A combination of control efforts and development, however, has sharply reduced its intensity and distribution. The acquisition of specific schistosome populations may be dependent on host characteristics such as sex, age, geography, work, habits and culture. How these and other host characteristics align with parasite subpopulations may guide approaches to improve control.
A cohort of more than 90% of the residents in two rural communities in Brazil participated in an epidemiologic survey of demographic, socio-economic and behavioral characteristics. The variables sex, age, intensity of infection, socio-economic index, % lifetime spent on site, previous infection, and trips outside the district were used to group parasites infecting individuals. Schistosoma mansoni infection status was determined by examination of stools submitted on 3 different days. The aggregate of eggs collected from the whole stool was used to determine degree of population differentiation from allele frequencies for 15 microsatellites.
Infection prevalence was 41% for these communities, and the epidemiologic characteristics were similar to many of the endemic areas of Brazil and the world. Parasite population structuring was observed between the two communities (Jost's D 0.046, CI95% 0.042–0.051), although separated by only 8 km and connected by a highway. No structuring was observed when infected individuals were stratified by host's biologic, demographic or epidemiologic characteristics. Those most heavily infected best reflected the communities' overall parasite diversity. The lack of differentiation within villages suggests that individuals are likely to get infected at the same sites or that the same parasite multilocus genotypes can be found at most sites. The geographic structuring between villages and the lack of structuring by age of the host further supports the impression of a population little affected by migration or drift.
| Schistosomiasis is one of the world's most important parasitic infections. Its elimination has proved difficult even in countries such as Brazil where access to treatment is readily available. Infection is the result of human contact with surface water where there are infected snails, so that human biology and habits may bring different individuals in contact with different groups of parasites. Identification of schistosome subpopulations may assist understanding transmission patterns and guide control efforts. We compared microsatellite allele frequencies from all of the infections in 2 small villages and determined that the movement of parasites between them was limited. Individual infections were distinct composites of parasites, but if infected humans were grouped by demographic and epidemiologic characteristics, there was no evidence that specific parasite subpopulations were being selected in these types of hosts. Infections were also not differentiated when stratified by host's age indicating that the populations were stable over time. Since the infection cycle requires human fecal contamination of water, local human behavior can to some degree be inferred from the patterns of schistosome subpopulation distribution.
| The transmission of schistosomiasis is influenced by human culture, occupations and demographics among other factors. Also, our group and others have demonstrated that each individual host carries only a portion of the total available parasite genetic variability [1], [2], [3], [4], [5], [6], [7], and thus host-to-host structuring may exist due to each individual's personal characteristics, such as age, sex, social status or residence. These are factors that may bring them into contact with genetically distinct populations of parasites or even influence their susceptibility. While these epidemiologic relationships are usually explored by associating human demographics with infection prevalence or intensity, by using genetic markers we can also determine if these host characteristics are associated with acquiring different parasite subpopulations.
An immediate problem for any such analysis is how to sample the parasite population. Due to the biology and local distribution of the parasite Schistosoma mansoni, sampling for genetic analysis is not straightforward. The snail host, where asexual reproduction takes place, lives an average of 3 months, and cercariae collected at one point in time do not represent the whole genetic diversity found in humans [7]. In addition to differences in behavioral factors and biological susceptibility of the human host, the intermittent presence in the snail host increases the potential for differential acquisition of parasite genotypes.
Sexual reproduction takes place in the human host where the adult worms are inaccessibly located in mesenteric veins. A portion of the hundreds of eggs produced daily by worm pairs remains trapped in tissues and will not contribute to the succeeding generation, whereas the majority of these progeny is shed in stool. New individuals enter the host only by infection, a form of migration. Since the adult parasites are long-lived, humans can accumulate a variety of individuals over time. Our approach to the population genetics of S. mansoni has been to analyze allele frequencies obtained by extracting DNA from the aggregate of eggs isolated from single stools. In this way the reproducing population of schistosomes from many individuals (e.g., most of the residents of small communities) can be analyzed with a large sample size and a minimum of selection bias.
An important problem for all genetic studies is determining appropriate sample size and avoiding selection bias. The population structure of most organisms is studied by collecting a sample of discrete genotypes and then aggregating or pooling these into allele frequencies for the whole population. This approach is dependent on the quality of the sampling performed. Depending on the organism and the specific population, sample sizes of 30 [8] or hundreds [9], [10] have been deemed necessary to provide an adequate sample. Parasite populations add unique challenges to the problem of sampling since they are not simply structured as discrete organisms scattered or clustered across a landscape. They exist as populations within individual hosts (infrapopulations) as well as the collection of parasites within one host species (component populations) [11]. The latter represents the full genetic potential of which the infrapopulations are each a small sample. For the individual human infection with S. mansoni, a typical 200 g stool with a light infection of 40 eggs/g will have a total of 8,000 eggs. The miracidial stage can be hatched from eggs and collected for study. Samples of 10, 20, 30 individual miracidia may be small when diversity is high, and there may be bias for which eggs will hatch into miracidia and which can be collected. Further, the process of hatching and collecting individual parasites limits the number of infected people that can be examined. How to sample, what to sample and how much to sample has never been defined for schistosomes. Our approach to the population genetics of S. mansoni has been to analyze allele frequencies obtained by extracting DNA from an aggregate of eggs isolated from the whole stool of infected individuals. In this way the transmitted population of schistosomes from many or even all individuals (in the case of a small community) can be analyzed with a large sample size from many infrapopulations and a minimum of selection bias. Sampling larger numbers of infrapopulations also allows for stratifying hosts for comparisons. Finally, using this approach we have shown that the stool egg population has a similar genetic composition to the adult worm population [12], [13].
In this paper we assessed risk factors for infection and differentiation of parasite infrapopulations by genotyping the aggregate of eggs obtained from infected individuals in two small rural villages. We divided parasites into “component” populations based on host geography as well as host biology, demography and epidemiology. We then estimated differentiation between these groups from their infrapopulation or component population allele frequencies. Although we previously observed structure based on geographic distance between these two nearby communities [3], we found little population structuring within the villages or between hosts. Finally, we explore the implications of these findings for the nature of schistosome populations in rural communities.
The Committee on Ethics in Research of the Oswaldo Cruz Foundation of Salvador, Bahia, the Brazilian National Committee on Ethics in Research and the Institutional Review Board for Human Investigation of University Hospitals Case Medical Center, Cleveland, Ohio approved the study design. All subjects provided written informed consent or in the case of minors, consent was obtained from their guardians. All aspects of the study have been conducted according to the principles expressed in the Declaration of Helsinki.
Two rural Brazilian communities – Jenipapo (population 482) and Volta do Rio (population 367) – were studied because of their high prevalence of schistosomiasis, their size and their relative isolation. They are administered by the municipality of Ubaíra (roughly equivalent to a county in the USA) and are located in the Jiquiriçá River valley in the state of Bahia. By road they are 270 km SE of the State capitol and principal city, Salvador. Each was at least 12 km from a major town and 8 km distant from each other. Volta do Rio is also divided geographically into an upper and lower section with a 40 m difference in height above the river (Figure 1). The major sources of livelihood are planting cacao, bananas cassava, cattle raising and other animal production. There is a Federal Family Health Program clinic in Jenipapo with a permanent staff consisting of a nurse, dentist and part-time physician. Volta do Rio has a simpler health post that employs only a group of nurses. Jenipapo also has primary and secondary schools attended by all of the nearby small communities, including Volta do Rio.
As previously described [3], an epidemiologic and parasitologic survey was conducted for all inhabitants ≥1 year old who agreed to participate. Questions concerning housing, sanitary habits, socio-economic conditions and water contact were asked as part of the epidemiologic survey. For water contact, individuals or guardians for minors <10 years of age were asked if they frequently used any of the 8–9 previously identified major water contact sites and what activities they tended to perform there. The socio-economic evaluation was based on the Criteria for Economic Classification of Brazil (http://www.abep.org/novo/Content.aspx?ContentID=139). These criteria with revisions have been used nationally for more than a decade to characterize the purchasing power of the Brazilian population using possessions (color TV, radio, bathroom, car, washing machine, videocassette/DVD, refrigerator, freezer), services (maid/housekeeper) and degree of education of the head of household. The index places households within 8 categories ranging from minimum monthly wage to 13X minimum monthly wage. The interpretation of these categories is weighted for metropolitan regions of the country including Salvador, Bahia.
Three stool samples each on different days were requested from each resident over a period of 1 week for quantitative examination by the Kato-Katz method. Individuals who tested positive for S. mansoni infection were treated with a single oral dose of praziquantel according to Brazilian Ministry of Health guidelines [14]. Those found to have intestinal nematodes were treated with mebendazole.
All stools were weighed to the nearest 0.01 g on a digital balance upon arrival in the laboratory. Whole stools from single individuals that were positive for S. mansoni were homogenized in a blender containing 200 ml of 2% saline followed by selective sieving [15] through two mesh nylon filter bags (FSI, Michigan City, Indiana, USA) with 300 and 55 µm pore sizes, respectively. The retained material was then sedimented in 2% saline. Since eggs were among the densest elements in the stool [16], the bottom 5 ml of sediment was collected and kept frozen at −20°C until used for DNA isolation.
The 5 ml frozen stool sediment was mixed with 5 ml 2X extraction buffer (50 mM NaCl, 100 mM Tris–HCl, pH 7.5, 10 mM EDTA, 1.0% SDS) and 10 ml H2O-saturated and Tris-buffered phenol, pH 7.5. This was followed by two chloroform/Isoamyl extractions [3]. The DNA was then ethanol precipitated and suspended in 10 mM Tris, pH 7.5, 1 mM EDTA. Finally, the sample was treated with cetyl trimethylammonium bromide (CTAB) to remove PCR inhibitors [17].
To genotype S. mansoni eggs, 15 microsatellite markers were used as described previously [2], [3]. For each marker a duplicate PCR reaction using 2 µL of extracted DNA from stool was performed, totaling 30 reactions per sample. PCR products from each sample were combined into groups of three or four markers and processed on an Applied Biosystems 3730xl DNA Analyzer. PeakScanner software (Applied Biosystems, Carlsbad, CA) was used to determine peak heights from which allele frequencies were calculated. Successful PCR reactions were defined as those in which there was at least one peak >500 pixels in the size range expected for a given marker. All peaks less than 100 pixels were excluded. We attempted to genotype all samples, and if multiple samples from the same individual amplified, their mean allele frequency was used. Subsequent population analyses were limited to those samples where a minimum of 12 out of 15 markers genotyped successfully.
Information collected during the study was double-entered into the program Epi Info version 3.5.3 [18]. Pearson's chi-square and Student's t-test were used to compare categorical and continuous data, respectively, and a p-value of 0.05 was used as the criterion for statistical significance. Multivariable analyses were carried out using logistic or linear regression in SPSS (Version 17). Individuals with missing data were dropped for the analysis of that variable.
For population genetic analyses, allele counts for each sample were calculated by multiplying the allele frequencies at a microsatellite locus by the total egg counts found on the Kato-Katz assay. Infrapopulations were stratified by the host's residence, sex, age, intensity of infection, household, travel history, number and location of water contacts and socio-economic condition. Genetic differentiation between populations was expressed as the index Jost's D [19] calculated using the program SPADE (http://chao.stat.nthu.edu.tw). D is a true differentiation index and does not rely on assumptions of Hardy-Weinberg equilibrium, which do not apply to infrapopulations. After grouping, each pair of infrapopulations can be compared within a group or the combined allele numbers and allele frequencies can be used to form a component population. We make the following differentiation and diversity comparisons:
Egg counts were recorded as eggs per gram of stool (epg) and log-transformed to approximate a normal distribution for analyses. Arithmetic means were calculated for group Di, Dic and AE. For the Di and the Dic group means were compared by bootstrapped Student's t-test with 1000 resamples, since the distribution of these measures is unknown. There is no standard for effect size for these new types of comparison. For the Dc, we follow the convention used for interpreting FST values [21]. Dc values from 0–0.05 indicate little differentiation; from 0.05–0.15, moderate differentiation; and above 0.15, great differentiation [22]. Changes in D rather than the absolute value below the 0.05 range, however, may still indicate a significant obstacle to gene flow.
The study group consisted of 814 of the 849 (96%) inhabitants residing in the 243 households of the two villages. The mean age was 31.5 years (±22.2), and slightly more women than men were enrolled (53.7%). Most subjects were born in their current municipality (83.7%), and the average percent of lifetime spent in the municipality of Ubaíra was 93.5%. Considering the history of travel outside of the district, 25.3% reported any travel, and a minority (19.5%) of those who traveled reported contact with surface water. There were some differences for the two geographically distinct areas of Volta do Rio (VdR). The percent of those traveling outside of the district was greater for individuals from lower VdR than upper VdR (34.2 vs. 22.2%, p = 0.02), but they remained outside of the area for similar lengths of time (61.92 vs. 57.82 days, p = 0.51). There were significantly more individuals in upper VdR who had at least one family member infected (36.9 vs. 25.2%, p = 0.02).
Most demographic and epidemiologic characteristics were similar for both villages (Table 1), with the exception of the socio-economic index and sanitation. Jenipapo had a somewhat greater purchasing power for (12.0 vs. 10.7, p = 0.017). The mean socio-economic index for the two localities was 11.4±4.3, which corresponds to the second lowest of the 8 income categories used nationwide. A socio-economic index of 11 points translated to a family income of approximately $330/month in 2009. Nearly all homes in both villages have piped water and indoor flush toilets. The 2 most common destinations for these toilets was either a septic tank or the river. Despite a lower socio-economic index, the disposal of human waste was more adequate in VdR than Jenipapo, and upper VdR had better waste disposal than lower VdR. In VdR the Jiquiriçá River is shallow, sluggish and seasonal, while at Jenipapo the Jiquiriçá is joined by a major stream that maintains flow in the river throughout the year. This may explain the different approaches to sanitation. Drinking water in both communities comes from sources several km away from the river.
The prevalence of S. mansoni infection was higher in Jenipapo (45.8%) than VdR (35.1%), but the mean intensity of infection was similar (Table 1). The lower limit of detection was 8 epg and the highest mean intensity observed was 3,792 epg. Some 31.3% of residents knew someone with current or past infection with S. mansoni, and 34.2% had one or more relatives with schistosomiasis. Two hundred and ninety seven individuals (37%) reported past infection with S. mansoni, and 93.6% of those reporting infection also reported being treated, most often with oxamniquine (64.0%). None had been treated with praziquantel, which was newly approved in Brazil for treatment of schistosomiasis at the time of the study. No variables or contact points were correlated with intensity of infection.
Characteristics that were associated with a higher risk for S. mansoni infection were living in Jenipapo, age (2nd, 3rd and 4th decades compared to 1st, Figure 2) and male sex (Table 2). Traveling outside of the municipality of Ubaíra in the past year was not associated with an increased risk for infection, but water contact while traveling was (OR of 2.3, p = 0.012) compared to those reporting no contact. A self-reported history of past infection overall had no correlation with risk, but reporting past treatment for S. mansoni did (OR 3.07, p = 0.02).
Eight water contact points in Jenipapo and nine in VdR were identified as those most commonly visited by villagers. The number of visits and nature of activities at each site were asked during the epidemiologic survey. The risk of being infected with S. mansoni increased substantially as the individual had contact with an increasing number of sites (Table 2). After adjusting for age and sex, people who reported contact with one point in Jenipapo and two in VdR were significantly more likely to be infected (Table 3). All of these points were common crossings to reach from one side of the river to the highway. A log was used as a temporary bridge at one point each in the two villages. However, at contact point 5 in Jenipapo (Figure 1C) the activity most associated with infection was fishing (OR = 2.96, p = 0.012), which is usually performed while wading in the river. In VdR, at the point not used for crossing the river (P3, Figure 1B) formed a pool, and bathing here was most associated with infection (OR = 3.55, p<0.001). Working and walking at this site were protective (OR = 0.1, p = 0.012 and OR = 0.043, p = 0.036, respectively), while fishing and playing in the water were also associated with a risk for infection (OR = 4.18, p = 0.048 and OR = 4.93, p = 0.026, respectively). The only significant activity associated with those who used the site and were uninfected was collecting water (OR = 5.1, p<0.048).
While individuals younger than 15 years old did not report more water contact than those older (p = 0.540), the type of contact may have involved more or longer exposure. Water contact for children tended to involved leisure activities such as walking (p = 0.02), swimming (p<0.001) and playing (p = 0.002) compared to older individuals who contacted water primarily through activities associated with labor, such as working (p = 0.02) and obtaining water (p = 0.05). Fishing was equally frequent between both age groups. Males did report visiting 1.5 times as many water contact points as females (p = 0.001).
Our previous study [3] used only samples that were positive by Kato-Katz in all 3 stools (n = 116). For the analysis here we included all samples regardless of the number of stools positive for S. mansoni, thus, genotypes from 226 of the 335 infected individuals (67.5%) were included for analysis. Of those genotyped, 51.8% were genotyped for 3 samples, 14.6% for 2 samples and 33.6% for only 1 sample. The differentiation between the two geographic component populations of the two villages (D = 0.046, CI95% 0.042–0.051) was similar to that previously reported [3]. To determine whether related parasites clustered with host characteristics, component populations were formed by grouping infrapopulations based on host epidemiologic characteristics of sex, age, household, economic status, place of birth, frequency of travel, previous infection and number and location of water contacts. Differentiation between these component populations was analyzed for the Di, Dc, Dic and effective allele number.
Di was significantly different for the individual villages and both villages combined when infections were grouped by sex, age, infection intensity and certain water contact sites (Table 4). We also tested similarity of infrapopulations within households. Only 11% and 5% of households in Jenipapo and VdR, respectively, had more than one member infected. The mean Di for household members in Jenipapo was 0.065±0.040 and 0.086±0.047 in VdR compared to 0.095±0.033 and 0.123±0.066 for all those infected in Jenipapo and VdR, respectively. The bootstrapped t-tests for the mean Di of household clusters versus all individuals in the village were significantly smaller (p = 0.004, p = 0.030; Jenipapo and VdR, respectively).
The Dc indicates that the composition of the populations based on host characteristics differ little in their genetic composition. The Dic was significant for age overall, but this was mainly due to a difference in VdR where children ≤15 acquired parasites that were more genetically differentiated from the whole community of parasites than those infecting adults. Overall and in both communities, infrapopulations from heavy infections were less differentiated from the community's component population than lighter ones.
The AE was only significantly different for intensity of infection for the two villages combined as well as separately. This is a measure of diversity, and higher intensity infections averaged higher effective allele numbers. AE was also associated with the socio-economic index in Jenipapo.
Since different age groups may be exposed to different subpopulations of S. mansoni, we further stratified age into 4 groups: 0–7, 8–15, 16–40 and >40. We found that the youngest age group gave the highest Dc in pairwise comparisons and the highest mean Dic of any group, but this age group was also the smallest (n = 12), had the lowest prevalence and the lowest intensity of infection. When 12 individuals with similar intensities of infection (sample sizes) were compared from each age group, these differences resolved.
The communities of Jenipapo and Volta do Rio are typical of the region in their level of development and access to sanitation. They also are similar to many other areas endemic for schistosomiasis in their age-specific prevalence and intensity of infection [23]. Differences in the prevalence of infection between these otherwise similar communities may be due to differences in how human waste is handled. In part these choices may be the result of the presence of constant flow in the river in Jenipapo and seasonal flow in VdR. Consistent with this, upper VdR which is much further from the river had higher use of septic tanks and fewer homes reporting using the river. In Brazil, economic development and control efforts using education and the drug oxamniquine (used prior to praziquantel) have greatly reduced the amount of hepatosplenic disease, but the infection prevalence in many areas has not changed. In these two villages, the current prevalence when based on a single stool examination is no different from the 15–20% prevalence observed for the state of Bahia in the 1950's [24], [25] and at the start of control programs in 1976 [24], [26], [27]. When multiple stool samples are examined, the true prevalence of infection is even two to threefold higher.
Some common risk factors found in other communities can be identified in this study. Age between 10 and 20 was associated with the highest prevalence and intensity. In Brazil, male sex is associated with increased risk [28], but in other parts of the world infection can be more prevalent in females [29]. This difference is likely due to differing sexual roles in work, play and ultimately water contact. Water contact is an essential step in transmission of schistosomiasis. While this variable would seem to be a strong risk factor with high correlation with infection, it has been difficult to measure and then associate with intensity [30]. Even when water contact is directly observed [29], the frequency of contacts is not always predictive. Questionnaires have been the simplest and least expensive way to assess risk factors. In Brazil, questionnaires have been shown to produce reliable responses that correlate with risk of infection [31], [32], but even here there can be significant place-to-place variation [33] requiring questions tailored to the specific location. Water contact has been solicited in multiple ways in terms of location, type of activity, time of contact and percent body exposure. We asked only which sites were visited and what activities were commonly performed there. The questionnaire was administered prior to all stool examinations and thus not biased by knowledge of the infection status of respondents. We found that simply counting the number of sites visited was most associated with prevalence of infection. Further, while travel away from the area was not associated with infection, self-reported travel combined with surface water contact was associated. These associations tend to validate the responses given by the residents.
In addition to risk for prevalence and intensity of infection, we sought to identify risk factors for acquiring specific parasite populations. The moderate differentiation between infrapopulations indicates that each individual collects a limited portion of the total genetic variability from the component population. This non-homogeneous distribution together with differences in water contact, occupation, habits, sex, years of exposure, etc are all reasons for structuring of the parasite population within different demographic categories of the human host. No population structuring, however, was observed. In another human population with a different intensity of transmission or different economic, cultural or geographic organization the distribution of parasites might be different.
Geographic structuring showed that over a short distance schistosome gene flow is limited in this region (Dc Jenipapo/VdR = 0.046). By contrast, within the two villages, when we assessed differentiation based on individual water contact sites, we found no difference in Dc among the sites or for number of sites visited. In VdR in particular, where there is a significant geographic difference in the height of the two parts of the city relative to the Jiquiriçá River and a highway between them, we were unable to demonstrate geographic differentiation. This indicates that, within the resolution of our methodology, local gene flow is high within the villages, but not between them. Individuals tend to be infected at the same sites or the same parasite multilocus genotypes can be found at most sites. They do not tend to contaminate the waters in nearby villages, and the school sanitation system (serving children from both villages) is unlikely to contribute to the local parasite population.
The marked difference in age-specific prevalence, intensity and perhaps increased exposure during water contact suggest that children are likely to be more exposed than adults to the current component population present in the resident snails. However, the lack of differentiation by age suggests that current and past populations are largely undifferentiated, and that over at least the last 5 years (the 95% CI for parasite life-span is 5.7–10.5 years [34]) there has not been a large degree of migration or selection, also supported by the geographic structuring between the villages.
The amount of differentiation within the groups of infrapopulations (Di) defined by host characteristics was often significantly different for multiple host factors, but we have no basis for comparison to say if this is biologically meaningful. By contrast the index Dic was significantly different for only intensity of infection in both villages. The socio-economic index and age are variably associated, but these may be secondarily related to intensity. The Dic is a measure of how differentiated an individual infrapopulation is from the whole adult worm/egg parasite community. It serves as a useful measure of how effectively individual hosts within a group sample the component population. The higher the intensity of infection, the more samples are present, which results in a better representation of the component population. This will reduce D for the infrapopulation relative to the component population. For the mean effective allele number, a measure of diversity, only intensity of infection (<400 or >400 epg) showed significant difference for both villages. This is consistent with expectation. In this area, the sampling of the most heavily infected, who are usually between 7 and 15 years of age, might be the best way of estimating the composition of the component population without sampling everyone. This limited sample would still lack precision, and age alone was not significantly associated with the Dc or Dic for Jenipapo.
An important issue for these conclusions is the sensitivity of the methods employed for differentiating parasite subpopulations. We do know that our approach is sensitive enough to differentiate the component population for the two villages, and that in the laboratory, different laboratory-maintained S. mansoni populations from the same laboratory and different lots of parasites from the same life cycle can be distinguished [12]. We were able to genotype at least 1 sample from 67% of those infected. Since we have shown infrapopulation allele frequencies are stable over at least the span of a week, obtaining a single stool is unlikely to be a source of error. Most of those we were unable to genotype had low egg counts and low DNA concentrations [3]. Most of the cryptic infections we failed to detect are also likely to have been of low intensity. They were, therefore, less likely to contribute significantly to the genetic composition of their component populations. The relative relationship of the genetic composition of eggs to adult worms is unknown in natural infections, but in laboratory infections in mice, allele frequencies between these two stages were very similar [12].
There is no one approach that address all problems in population genetics, but the approach taken here is well suited to measure differentiation, since it allows for many large samples. Certain population genetic indices, such as the FIS, cannot be well estimated from aggregated data, however. In addition, we are unable to identify null alleles. This should not affect estimates of differentiation for populations in which the rate of null alleles is likely to be similar.
Attempts to control S. mansoni infection in Brazil were successful in decreasing intensity of infection, and therefore, morbidity and mortality of the disease, but the infection has far from disappeared. An understanding of the dynamics of transmission and the distribution of the parasite at the population level can contribute to planning control measures. We show that there is little population sub-structure by host characteristics to influence how praziquantel therapy should be distributed. There are no special reservoirs of distinct parasite populations within the community, and much of transmission is local with good evidence for a barrier to gene flow with a nearby community. Future studies will examine how applicable the patterns seen in these communities are to others in Brazil and elsewhere. Until elimination has been achieved, surveillance and treatment will need to be continued and improvements in sanitation advanced.
|
10.1371/journal.ppat.1007509 | Identification of a short, highly conserved, motif required for picornavirus capsid precursor processing at distal sites | Many picornaviruses cause important diseases in humans and other animals including poliovirus, rhinoviruses (causing the common cold) and foot-and-mouth disease virus (FMDV). These small, non-enveloped viruses comprise a positive-stranded RNA genome (ca. 7–9 kb) enclosed within a protein shell composed of 60 copies of three or four different capsid proteins. For the aphthoviruses (e.g. FMDV) and cardioviruses, the capsid precursor, P1-2A, is cleaved by the 3C protease (3Cpro) to generate VP0, VP3 and VP1 plus 2A. For enteroviruses, e.g. poliovirus, the capsid precursor is P1 alone, which is cleaved by the 3CD protease to generate just VP0, VP3 and VP1. The sequences required for correct processing of the FMDV capsid protein precursor in mammalian cells were analyzed. Truncation of the P1-2A precursor from its C-terminus showed that loss of the 2A peptide (18 residues long) and 27 residues from the C-terminus of VP1 (211 residues long) resulted in a precursor that cannot be processed by 3Cpro although it still contained two unmodified internal cleavage sites (VP0/VP3 and VP3/VP1 junctions). Furthermore, introduction of small deletions within P1-2A identified residues 185–190 within VP1 as being required for 3Cpro-mediated processing and for optimal accumulation of the precursor. Within this C-terminal region of VP1, five of these residues (YCPRP), are very highly conserved in all FMDVs and are also conserved amongst other picornaviruses. Mutant FMDV P1-2A precursors with single amino acid substitutions within this motif were highly resistant to cleavage at internal junctions. Such substitutions also abrogated virus infectivity. These results can explain earlier observations that loss of the C-terminus (including the conserved motif) from the poliovirus capsid precursor conferred resistance to processing. Thus, this motif seems essential for maintaining the correct structure of picornavirus capsid precursors prior to processing and subsequent capsid assembly; it may represent a site that interacts with cellular chaperones.
| The picornavirus family includes clinically important human and animal pathogens, for example: poliovirus, rhinovirus (causing the common cold) and foot-and-mouth disease virus (FMDV) that infects cloven-hoofed animals. Picornaviruses contain a positive-sense RNA genome surrounded by a protein shell, also called a capsid. The capsid proteins are made from a precursor and correct processing and assembly of these capsid proteins is necessary in the virus life cycle to create new infectious virus particles. In this study, we have identified a short motif (just 5 amino acids long) within the capsid precursor, which is highly conserved among picornaviruses. Deletion of this motif inhibited processing of the junctions between the mature structural proteins within this precursor, with one junction being more than 400 amino acids away from this region. This motif also seems to be required for the optimal accumulation of the capsid precursor in cells. We hypothesize that the motif may be involved in binding to a cellular protein, such as a chaperone, to stabilize the capsid precursor and promote its correct folding to allow it to be processed by the viral protease prior to capsid assembly.
| Picornaviruses comprise a large family of non-enveloped RNA viruses that includes important human and animal pathogens. Examples include poliovirus (PV) (genus: Enterovirus), hepatitis A virus (Hepatovirus), encephalomyocarditis virus (Cardiovirus) and foot-and-mouth disease virus (FMDV) (Aphthovirus).
In picornavirus particles, the RNA genome (ca. 7,100–8,900 nt) is surrounded by a protein shell (capsid) consisting of the four structural proteins VP1, VP2, VP3 and VP4 [1], with the exception of parechoviruses and kobuviruses in which the VP0 (the precursor of VP2 and VP4) remains uncleaved (reviewed by [2]). The capsid is composed of 60 copies of each of these structural proteins; VP1, VP2 and VP3 are exposed on the surface of the particle while VP4 is entirely internal [3–5].
Translation of the positive-sense RNA genome is dependent on the internal ribosomal entry site (IRES) within the 5′ untranslated region (UTR) that directs cap-independent translation initiation [6]. During and after translation of the single open reading frame, processing of the newly synthesized polyprotein occurs (reviewed in [2]). Usually three or four primary products are formed, namely the Leader (in many picornaviruses), the capsid precursor P1 or P1-2A (depending on the genus) and the precursors of the non-structural proteins, namely P2 and P3. Many of these viruses, e.g. members of the Cardiovirus, Hepatovirus and Aphthovirus genera, have a Leader protein at the N-terminus of the polyprotein, i.e. upstream of the capsid precursor. In the Aphthoviruses, the Leader protein is a protease (Lpro), which cleaves itself from the N-terminus of the P1-2A precursor, see Fig 1. Cleavage of the junction between the structural and non-structural proteins, at either the VP1/2A or the 2A/2B junction, is usually mediated by the 2A protein, but the function of the 2A protein varies between the genera [7], see Fig 1A. In the cardio- and aphthoviruses cleavage at the 2A/2B junction (at the C-terminus of 2A) is protease independent and happens during translation by a process termed “ribosomal skipping” [8] or “StopGo” [9]. In this case, the 2A protein remains attached to the precursor of the structural proteins (as P1-2A) until it is removed by the 3C protease (3Cpro), see Fig 1B. In the enteroviruses, the cleavage at the VP1/2A junction (i.e. at the N-terminus of 2A), to release P1, is mediated by the 2A protein that is a chymotrypsin-like protease [10,11].
The P2-P3 junction and the other protein junctions within these precursors are cleaved by 3Cpro to produce the mature non-structural proteins. However, the P1 capsid precursor of enteroviruses requires the 3CD protease (3CDpro) for its processing [12,13] whereas for the cardio- and aphthoviruses the 3Cpro is sufficient to cleave the P1-2A precursor into three structural proteins (VP0, VP3 and VP1) plus 2A [1,14], see Fig 1B. During capsid assembly, VP0 is cleaved (in most picornaviruses) to generate VP2 and VP4 by a process that is currently not understood.
There are seven different serotypes of FMDV: O, A, C, SAT 1, SAT 2, SAT 3 and Asia 1. There is a high level of sequence variation between the surface exposed structural proteins of these different serotypes. The internal VP4 protein is the most conserved of the capsid proteins with 81% of the residues being invariant [15]. In contrast, only 26% of the VP1 protein residues are invariant and furthermore it ranges in size (209–213 aa) between serotypes [16]. VP1 is the most surface exposed capsid protein [3] and has been one of the most studied FMDV proteins due to its antigenic importance and role in virus attachment [17].
One of the antigenic sites in VP1 is located on the G-H loop (including residues 141–160), which contains an arginine-glycine-aspartate (RGD) motif that is involved in the attachment of the virus to cellular integrin receptors [18,19]. Surprisingly, previous work has demonstrated that a cell-culture adapted FMDV, lacking part of this G-H loop (aa 142–154), is still able to replicate and grow normally in cell culture through the use of heparan sulfate proteoglycans (HSPG) as receptor [20].
Viruses have only a very limited coding capacity within their genomes and thus they rely on cellular factors and pathways to complete their life cycle. Several studies have suggested that cellular chaperones, including various different heat shock proteins (Hsps), are required to facilitate virus entry, genome replication, protein expression and protein assembly for a variety of viruses, including picornaviruses. Viral proteins, like cellular proteins, are dependent on such chaperones for their correct folding and assembly [21–24]. Studies on the role of Hsp90, using specific inhibitors, have shown that these agents reduce the replication of diverse viruses in vitro. The Hsp90 appears to be involved in the regulation of viral polymerase function in the case of herpesvirus [25] and hepatitis B virus [21], whereas this chaperone seems to be required for capsid processing and assembly in different picornaviruses [23,26]. Hsp90 and Hsp70 have been reported to interact with the PV capsid precursor, P1 [23,27]. The interaction between PV P1 and Hsp90 (possibly together with Hsp70), and likely in conjunction with its co-chaperone p23, is believed to protect the P1 from degradation by proteasomes (which remove misfolded proteins) and is also involved in the folding of P1 allowing it to be correctly processed by the 3CDpro [23].
Recently, we have shown that impeding the processing of one of the cleavage sites within the FMDV P1-2A, at either the VP0-VP3 or the VP3-VP1 junctions, did not block processing of the other cleavage sites, indicating that processing of these junctions is mutually independent [28]. However, in an earlier study, it was shown that truncation of VP1 (removing the C-terminal 42 amino acids of VP1) completely blocked processing of the residual capsid precursor at both the VP0-VP3 and the VP3-VP1 junctions by 3Cpro in a cell-free system [29]. Similarly, truncating the PV P1 precursor, by removing 50 aa from the C-terminus of VP1 (302 residues in length), blocked cleavage of the 2 junctions within the P1 precursor in vitro [30]. The basis for these effects has not been explained. However, taken together, these results suggest that the C-terminus of VP1 is important in relation to the processing of the entire capsid precursor of picornaviruses.
In this study, we have now identified a short region within the C-terminus of VP1 that is critical for the processing of the FMDV capsid precursor. This region contains a stretch of five amino acids that are very highly conserved amongst all FMDVs. Furthermore, this region is also strongly conserved between most other picornaviruses, including PV, suggesting a shared role for this motif for capsid processing and assembly within the picornavirus family.
Previous studies have shown that truncation of the FMDV P1-2A, by removal of the 2A peptide and the C-terminal 42 residues of VP1, completely abrogated processing by 3Cpro in vitro [29] even though the cleavage sites between VP0 and VP3 and between VP3 and VP1 were unmodified. To confirm these observations, within cells, stop codons were introduced at different positions within the P1-2A coding sequence. Transient expression assays were used to express the FMDV A22 Iraq P1-2A capsid precursor and its derivatives, within BHK cells, both in the absence and presence of the FMDV 3Cpro. The plasmids encoding both the P1-2A (wt) and the P1 alone (truncated to the first amino acid of the 2A peptide) served as positive controls. Both of these controls yielded the expected products corresponding to the P1-2A precursor and the P1 precursor (approximately 85 kDa), respectively in the absence of 3Cpro (Fig 2, lanes 1 and 3). When these plasmids were co-transfected with a plasmid that expresses the 3Cpro, both of these products were efficiently processed as indicated by the production of VP0 (approximately 37 kDa) (Fig 2, lanes 2 and 4). Thus, the absence of the 2A peptide did not affect processing of the capsid precursor by 3Cpro (as observed previously [14,26]).
Plasmids encoding mutant precursors, truncated to residue 205 in VP1 and 199 in VP1 (VP1 being 211 aa in length in FMDV A22 Iraq (wt)), generated products of approximately 85 kDa in the absence of 3Cpro (Fig 2, lanes 5 and 7), and these were efficiently processed in the presence of 3Cpro (Fig 2, lanes 6 and 8). The four additional mutants, P1 (VP1 Y185Stop), P1 (VP1 L158Stop), P1 (VP1 A107Stop) and P1 (VP1 L53Stop) all yielded products corresponding to their expected size in the absence of 3Cpro (Fig 2, lanes 9, 11, 13 and 15), however it is noteworthy that these truncated products accumulated to a lower level in the cell lysates. Strikingly, no processing of these truncated precursors was detected for any of these four mutants in the presence of 3Cpro (Fig 2, lanes 10, 12, 14 and 16) although each of these products contained the unmodified VP0/VP3 and VP3/VP1 junctions. As expected, no products were detected in the negative control (no DNA) (Fig 2, lane 17).
In order to map the determinants of capsid processing more precisely, plasmids were constructed to express mutant forms of the P1-2A precursor with fairly small internal deletions within the C-terminal portion of VP1. To serve as positive controls, both the P1-2A (wt) and a mutant form with a deletion within VP1, designated P1-2A (VP1 Δ142–154), were included. The latter deletion is tolerated by the infectious virus [20] and thus it was expected that 3Cpro should be able to fully process all of the junctions in this deletion mutant. As expected, expression of both the P1-2A (wt) and the P1-2A (VP1 Δ142–154) led to the synthesis of products corresponding to the P1-2A precursor (approximately 85 kDa) (Fig 3, lanes 1 and 13). Furthermore, both the P1-2A (wt) and the P1-2A (VP1 Δ142–154) products were efficiently processed in the presence of 3Cpro (Fig 3, lanes 2 and 14). Notice that the VP1 product derived from the P1-2A (VP1 Δ142–154) mutant migrated faster than the VP1 produced from the P1-2A (wt) ((approximately 28 kDa) due to the internal deletion (note that these antibodies do not recognize VP3 [31], but presumably this was also made).
Five different short deletions were introduced into the region of VP1 spanning residues 185–199 (the region found to be critical by the truncation analysis), namely P1-2A (VP1 Δ185–199), P1-2A (VP1 Δ185–189), P1-2A (VP1 Δ188–192), P1-2A (VP1 Δ191–195) and P1-2A (VP1 Δ194–199). Each of these constructs generated products that were very similar in size as the wt P1-2A in the absence of 3Cpro (Fig 3, lanes 3, 5, 7, 9 and 11). However, in the presence of 3Cpro the mutant having the largest deletion, P1-2A (VP1 Δ185–199) could not be processed (Fig 3, lane 4). The same product, corresponding to the P1-2A precursor, was observed both in the absence and presence of 3Cpro. Similarly, the mutants P1-2A (VP1 Δ185–189) and P1-2A (VP1 Δ188–192) were also not processed in the presence of 3Cpro (Fig 3, lanes 6 and 8). It is again noteworthy that the mutant P1-2A products that could not be processed accumulated to a lower level in the cell lysates than the P1-2A precursors that could be processed (c.f. lanes 3, 5, 7 and 1, 9, 11, 13). In contrast, co-expression of 3Cpro with the P1-2A (VP1 Δ191–195) and P1-2A (VP1 Δ194–199) led to production of VP0 indicating that processing of these mutant precursors had occurred (Fig 3, lanes 10 and 12). However, it is noteworthy that no product corresponding to VP1 was detected, when P1-2A (VP1 Δ191–195) was co-expressed with 3Cpro (Fig 4A, lane 4). Furthermore, unexpectedly, when the P1-2A (VP1 Δ194–199) was co-expressed with 3Cpro a major product corresponding to the intermediate VP3-VP1 (approximately 49 kDa) was detected (Fig 4A, lane 6). Only a weak signal corresponding to the mature VP1 was detected indicating severe inhibition of processing at the VP3/VP1 junction in this mutant (Fig 4A, lane 6), n.b. this cleavage site is located over 190 residues away in the linear sequence. No products were detected in the negative control lane. (Fig 4A, lane 9).
Due to inefficient detection of VP1 from some of the mutant precursors, an extra modification that blocks processing of the VP1/2A junction (2A L2P) [32] was introduced into the plasmids that express P1-2A (VP1 Δ191–195), P1-2A (VP1 Δ194–199) and the positive controls; P1-2A (wt) and P1-2A (VP1 Δ142–154). The additional modification (2A L2P) ensured that the 2A peptide remained fused to the VP1 (as VP1-2A). Each of these constructs generated products corresponding to the P1-2A precursor in the absence of 3Cpro (Fig 4B, lanes 1, 3, 5 and 7). The 2A L2P substitution increased the sensitivity of VP1 detection when using the anti-FMDV A-Iraq antibody. This showed that the P1-2A (wt + 2A L2P) and the P1-2A (VP1 Δ142–154 + 2A L2P, positive control) precursors were fully processed to yield VP0 and VP1-2A in the presence of 3Cpro as expected (Fig 4B, lanes 2 and 8). It also verified that cleavage at the VP3-VP1 junction in the P1-2A (VP1 Δ194–199 +2A L2P) occurred at a slower rate compared to wt, since the VP3-VP1-2A intermediate was far more abundant for the P1-2A (VP1 Δ194–199 + 2A L2P) than for the P1-2A (wt +2A L2P) in the presence of 3Cpro (compare lanes 6 and 2 in Fig 4B). It should be noted that some mature VP1-2A could be detected from the P1-2A (VP1 Δ194–199 + 2A L2P) and thus cleavage of the VP3/VP1 junction was not completely blocked (Fig 4B, lane 6). Furthermore, the P1-2A (VP1 Δ191–195 + 2A L2P) could be processed to generate VP0 and VP1-2A (Fig 4B, lane 4). However, the VP3-VP1 intermediate produced from the P1-2A (VP1 Δ191–195 +2A L2P) mutant was also more abundant than the intermediate seen with the P1-2A (wt + 2A L2P) indicating that this mutant also had a slower processing at the VP3/VP1 junction (Fig 4B, lane 4).
The cleavage of the unmodified VP1/2A junction in the P1-2A precursors with different internal deletions, was investigated using an anti-2A antibody. As expected, both the P1-2A (wt) and the positive control P1-2A (VP1 Δ142–154) generated products of approximately 85 kDa corresponding to the P1-2A precursor in the absence of 3Cpro (see supplementary material S1 Fig, lanes 1 and 13). In the presence of 3Cpro, no products were detected by the anti-2A antibodies from either the P1-2A (wt) or the positive control P1-2A (VP1 Δ142–154) indicating that the VP1/2A junction had been processed (S1 Fig, lanes 2 and 14); note the 2A peptide itself is only 18 residues long and is not detected by immunoblotting. The two mutants, P1-2A (VP1 Δ191–195) and P1-2A (VP1 Δ194–199) that showed slower processing of the VP3-VP1 junction also generated products corresponding to the P1-2A precursor in the absence of 3Cpro (S1 Fig, lanes 9 and 11). However, in the presence of 3Cpro, no products were detected by the anti-2A antibodies (S1 Fig, lanes 10 and 12), indicating that these two deletions in VP1 did not affect processing of the VP1/2A junction. Surprisingly, the non-processable precursors, i.e. P1-2A (VP1 Δ185–199), P1-2A (VP1 Δ185–189) and P1-2A (VP1 Δ188–192), could not be detected using the anti-2A antibody, either in the absence or presence of 3Cpro, and thus we cannot conclude whether cleavage of this junction was affected by the deletions (S1 Fig, lanes 3–8). No products were detected in the negative control (No DNA, S1 Fig, lane 15).
Alanine-scanning mutagenesis was employed to identify individual residues within the C-terminal region of VP1 (between residues 185 and 199 of VP1) that are required for 3Cpro processing of the P1-2A precursor. The wt and mutant precursors were expressed alone and also in the presence of the FMDV 3Cpro as above. As expected, the P1-2A (wt) and all 15 of the single amino acid substitution mutants each generated products corresponding to the P1-2A precursor in the absence of 3Cpro (see Figs 5, 6 and S2 (supplementary material), odd numbered lanes). The wt and some 13 different mutant P1-2A precursors, excluding the mutants P1-2A (VP1 Y185A) and P1-2A (VP1 R188A), were processed by 3Cpro to yield VP0 and VP1 (Figs 5, 6 and S2 (supplementary material) even numbered lanes). In contrast, the P1-2A (VP1 Y185A) and P1-2A (VP1 R188A) mutants were highly resistant to cleavage by the 3Cpro (Fig 5, lanes 4 and 10). Furthermore, it was again apparent that the accumulation of these mutant P1-2A products in the cell lysates was lower than for the wt precursor and for the other mutants that could be processed (Fig 5, lanes 3 and 9). Thus, the single amino acid substitutions VP1 Y185A and VP1 R188A were individually able to severely inhibit processing at both the VP0/VP3 and the VP3/VP1 junctions within the P1-2A precursor and had a deleterious effect on the level of the unprocessed product generated within cells.
Surprisingly, none of the single alanine substitutions in the VP1 194–199 region had any effect on the processing of the junctions within the P1-2A precursor (S2 Fig, lanes 4, 6, 8, 10, 12 and 14). None of these produced the severe block on cleavage of the VP3-VP1 junction that was detected with the P1-2A (VP1 Δ194–199) mutant (Fig 4, lane 6). However, interestingly, the P1-2A (VP1 V193A) was processed more slowly at the VP3-VP1 junction compared to the P1-2A (wt) and the other alanine mutants (Fig 6, lane 12).
The cleavage of the VP1/2A junction of the P1-2A precursors with different alanine substitutions, was also investigated using the anti-2A antibody. As expected, the P1-2A (wt) generated a product of approximately 85 kDa corresponding to the P1-2A precursor in the absence of 3Cpro (see supplementary material S3 Fig, lane 1). However, in the presence of 3Cpro, no product (containing 2A) was observed from the P1-2A (wt) showing that VP1/2A junction had been processed (S3 Fig, lane 2). The two mutants, P1-2A (VP1 C186A) and P1-2A (VP1 P187A) that were correctly processed at the VP0/VP3 and the VP3/VP1 junction also generated products corresponding to the P1-2A precursor in the absence of 3Cpro (S3 Fig, lanes 5 and 7). However, as with the wt protein, in the presence of 3Cpro no products including 2A could be detected, indicating that these two substitutions individually did not prevent processing at the VP1/2A junction. Neither of these mutant precursors, with single amino acid substitutions, which were highly resistant to cleavage at the VP0/VP3 and the VP3/VP1 junctions, i.e. P1-2A (VP1 Y185A) and P1-2A (VP1 R188A), could be detected by the anti-2A antibody, either in the absence or presence of 3Cpro. Thus, we cannot conclude whether this junction was affected by these substitutions (S3 Fig, lanes 3, 4, 9 and 10). These results are consistent with the inability to detect the mutant capsid precursors P1-2A (VP1 Δ185–199), P1-2A (VP1 Δ185–189) and P1-2A (VP1 Δ188–192), with the anti-2A antibody, as shown in S1 Fig (see above).
To confirm the importance of the YCPRP motif in the context of the virus itself, specific mutations have been introduced into the full-length FMDV cDNA, that encode single amino acid substitutions (to Ala) within the YCPRP motif. In addition, a deletion of the sequence encoding residues VP1 185–190 from the full-length FMDV cDNA was also made. RNA transcripts were prepared in vitro from each of the mutant plasmids and introduced into BHK cells. The initial harvests, prepared after 24h, were passaged onto fresh BHK cells and the appearance of cytopathic effect (CPE) observed. Clear CPE was observed with the wt transcript and from the mutants encoding the VP1 C186A and P189A substitutions. In contrast, no CPE was apparent for the mutants encoding the VP1 Y185A, P187A and R188A substitutions or with the mutant lacking residues VP1 185–190 (see Table 1). Sequencing of the P1-2A coding region from the rescued viruses (FMDV VP1 C186A and FMDV VP1 P189A revealed that the introduced mutations were retained and that no secondary mutations had occurred. These results verified the critical importance of residues Y185 and R188 in VP1 for P1-2A processing (Fig 5) and for virus infectivity. It is noteworthy that the P187A mutant was also non-infectious (Table 1) although the capsid precursor processing could be observed in the transient expression assay (see Fig 5, lane 8).
The FMDV 3Cpro is able to cleave a variety of different junction sequences in the virus polyprotein [33]. We have shown previously that blocking cleavage of one junction in the FMDV P1-2A did not affect processing of the other junctions [28]. In the current studies, it has been shown that modifications that modify or delete a short motif in the C-terminus of VP1, can prevent processing of the FMDV capsid precursor P1-2A at each of the usual cleavage sites, which are far separated, in the linear sequence, from the site of the modifications. The VP0/VP3 cleavage site is more than 400 amino acids away from the modified motif in the linear sequence while the VP3/VP1 junction is almost 200 amino acids away. It seems very likely that this reflects a major change in protein conformation for these mutant proteins. Viral proteins, like cellular proteins, are dependent on cellular chaperones for correct folding, assembly and function [24]. The viral capsid precursor must fold to a conformation that is soluble and recognizable by the viral protease to be processed. After the cleavage of the precursor, the mature capsid proteins assemble around the viral genome to form the protein shell, which contains 60 copies of each of the subunits. These structures must be stable both within, but also outside, the host cells to permit virus spread. Moreover, the virus particle must also be able to disassemble upon entry into cells to deliver the viral genome to initiate a new infection. Thus, the core structure of the capsid proteins (as distinct from the antigenic loops) is probably tightly constrained.
Within the picornavirus family, the general structure of the capsid proteins are very similar [2]. Several chaperones are known to facilitate folding of picornavirus capsid proteins [23,26]. The mature picornavirus capsid proteins are generated by cleavage of the P1, P1-2A or L-P1-2A precursors. Both Hsp90 and p23, a co-chaperone of Hsp90, have been reported to be required for processing of the PV P1 precursor into the mature structural proteins [23]. Similarly, inhibitors of Hsp90 have been shown to impede processing of the wt FMDV capsid precursor in cell-free assays [26]. However, interestingly, hepatitis A virus (HAV) is not sensitive to the inhibition of Hsp90 function [34]. This indicates that HAVs might employ other strategies for correct folding of the capsid precursor. However, it is noteworthy that HAV also has several unique characteristics that distinguish it from most other members of the picornavirus family, e.g. slow growth rate, lack of capsid protein myristoylation and use of only a single viral protease (3Cpro) for polyprotein processing [35–38].
An earlier study showed that Hsp90 mediates PV P1 folding in cells. Inhibition of this chaperone lead to misfolding of P1, which resulted in the targeting of the PV P1 for degradation by the cellular quality-control system (proteasome pathway), and thus the level of the PV P1 was strongly reduced [23]. These observations are consistent with the results presented here on the FMDV P1-2A. All of the FMDV P1-2A precursors that cannot be processed by 3Cpro accumulated to a lower level than the P1-2A (wt). This was apparent for the truncated precursors (VP1 Y185Stop, VP1 L158Stop, VP1 A107Stop, VP1 L53Stop), precursors with small internal deletions (VP1 Δ185–199, VP1 Δ185–189, VP1 Δ182–192) and two precursors with single amino acid substitutions (VP1 Y185A and VP1 R188A). Thus, it may be that the mutant precursors, which cannot be processed, are misfolded and therefore targeted for degradation, hence the reduced level of these products within cells.
Interestingly, Geller et al., [23] showed that inhibition of the Hsp90 chaperone in a cell-free system (rabbit reticulocyte lysate), where the proteasomal degradation system is inhibited by free hemin, did not reduce the yield of P1 [23]. However, even in the absence of proteasomal degradation, the Hsp90 was still required for P1 to fold into a processing-competent conformation, since the PV P1 precursor, in the absence of Hsp90, adopted a misfolded conformation that could not be recognized by the 3CDpro and thus could not be processed into the mature capsid proteins [23]. The clear resistance to processing of certain mutant FMDV P1-2A proteins (in which the YCPRP motif is modified or deleted) and their reduced accumulation within cells is entirely consistent with these results (see Fig 2B, Fig 3, Fig 4 and Fig 5).
As indicated above, a critical region that is required for the correct processing of the FMDV capsid precursor has now been identified. This motif (YCPRP) is very highly conserved among FMDVs. Indeed, the YCPR sequence was found to be completely conserved in over 100 FMDV strains, with representatives from all 7 serotypes [15]); only variation to YCPRA has been observed (Fig 7). However, previously, no function for this conserved sequence had been identified. The YCPRP motif is also highly conserved among other picornaviruses as well, e.g. it exists as FCPRP in cardioviruses and WCPRP in enteroviruses, see Fig 8. Indeed, both Y to F and Y to W are very conservative amino acid substitutions, since all three amino acids have similar properties with non-polar, aromatic side chains. This high conservation likely reflects its importance for correct folding of the capsid precursor. The high resistance to cleavage of the junctions between the structural proteins following substitution of residues VP1 Y185 and VP1 R188 individually indicates that correct cleavage may be dependent on the interaction with several amino acids in this region and thus the whole motif seems to be of high importance for correct folding and subsequent processing of the capsid precursor. Furthermore, these results are consistent with the observations that the substitutions VP1 Y185A and VP1 R188A, that each prevent P1-2A processing by 3Cpro in cells (Fig 5), also block FMDV infectivity (Table 1). It is interesting to note that the VP1 P187A mutant was also non-infectious (Table 1) even though processing of the P1-2A could still be observed (Fig 5, lane 8). The high conservation of this motif clearly reflects its sensitivity to modification.
An earlier study has shown that removing 50 residues from the C-terminus of the PV VP1 prevented cleavage of the two junctions, VP0/VP3 and VP3/VP1, within the capsid precursor in vitro [30]. Significantly, these 50 amino acids include the highly conserved motif (WCPRP) identified here, and thus indicates the importance of this motif, not only for FMDV, but also more widely within the picornavirus family. Similarly, as indicated above, removal of 42 residues (including the YCPRP) from the C-terminus of the FMDV VP1 protein completely prevented cleavage of the capsid precursor by the 3Cpro in a cell-free system [29]. Recently, we have shown that blocking cleavage of one of the junctions within the FMDV P1-2A precursor did not block the cleavage of the other junction within the capsid precursor [28]. Thus, the severe inhibition of cleavage of both junctions likely reflects a changed overall structure of the capsid precursor, thereby preventing cleavage of both junctions.
It is interesting to note that in HAV, the equivalent region of VP1 has the sequence YFPRA, perhaps the two substitutions together account for the lack of sensitivity of HAV assembly to Hsp90 inhibitors [34]. It can be proposed that the conserved motif serves as a binding site for an important chaperone, e.g. Hsp90 (or its partners), that is necessary for correct protein folding. A proposed model for this interaction is shown in Fig 9. A co-chaperone of Hsp90, called p23, also seems to be involved in the correct folding of the PV P1. It has been reported that treatment with geldanamycin (GA) did not affect the PV P1-Hsp90 interaction, but abolished the P1-p23 interaction and thereby affected P1 maturation [23], thus indicating different possibilities for chaperone interaction at this specific site.
Picornaviruses are able to adapt very rapidly since they have an RNA dependent RNA polymerase with a high error rate and no error correction mechanism. However, Geller et al., [23] showed that PV was unable to adapt to an Hsp90-independent P1 folding pathway during several passages in cells in culture or in PV-infected mice when the function of Hsp90 was inhibited by the presence of GA. Thus it seems that for the virus to adapt to a folding pathway without the involvement of Hsp90 requires extensive change [23].
It is interesting that the deletion VP1 Δ194–199 strongly inhibited cleavage at the VP3-VP1 junction, without affecting the cleavage of the VP0-VP3 junction (Fig 4, lane 6). Surprisingly, the alanine scanning substitutions through this specific region did not identify any individual residue that affected cleavage of any of the junctions (S2 Fig, lanes 4, 6, 8, 10, 12 and 14). However, interestingly the P1-2A (VP1 V193A) mutant, modified at a residue adjacent to the deletion, also displayed a slower processing rate of this VP0/VP3 junction compared to the wt and the other alanine mutants (Fig 6, lane 12). However, this VP1 V193A mutant does not seem to affect the processing of the VP3-VP1 junction to the same extent as the VP1 Δ194–199 mutant. In addition, the VP1 Δ191–195 mutant also showed a lower processing of this VP3/VP1 junction (Fig 4B, lane 4) as judged by the elevated level of the VP3-VP1 product. These results indicate that the cleavage of the VP3-VP1 junction, may be dependent on the interaction with several amino acids and that residues within the VP1 aa 193–199 region are important for optimal processing of the VP3-VP1 junction, more than 190 aa away from the site.
We have noted previously that the K210E change in VP1 that severely limited processing at the VP1/2A junction also enhanced the yield of VP3-VP1-2A [14]. These studies also identified a genetic link between the processing of the VP1/2A junction and the substitution E83K in VP1. Furthermore, Escarmis et al., [39] showed that the substitution M54I within the VP1 of serotype C FMDV resulted in less efficient processing at the VP3/VP1 junction. Thus, there are multiple, complex, interactions, some of which operate “at a distance”, that govern picornavirus capsid protein processing and assembly.
The plasmid pO1K/A22 contains a T7 promoter upstream of a full-length FMDV cDNA with the A22 Iraq capsid coding sequence within an FMDV O1K backbone as previously described [28,40,41]. To investigate the effect of different modifications within the P1-2A, the FMDV cDNA was digested with ApaI and then religated to remove most of the sequence encoding the non-structural proteins (including the 3Cpro) downstream of the 2A-peptide, as described previously [28]. These constructs contained a modified form (W52A substitution) of the Lpro to overcome the negative effect of the L protease on protein expression in cells. The modified Lpro with the W52A substitution retains the L/P1 cleavage activity but is defective at inducing cleavage of the translation initiation factor eIF4G [42]; the primer sequences used to make this modification are listed in S1 Table. This parental plasmid is referred to throughout as P1-2A (wt) and all modifications were made in this background.
Variants of the plasmid, with in-frame stop codons introduced to truncate the capsid precursor at different sites either at the start of the 2A sequence or within the VP1 coding sequence, were generated using site-directed mutagenesis [43]. Briefly, fragments were amplified in PCRs using Phusion High-Fidelity DNA Polymerase (Thermo Fisher Scientific), according to the manufacturer’s instructions, to create mega-primers, using the P1-2A (wt) plasmid as template and reverse primers specifying the introduction of STOP codons, together with the forward primer; 14TPN9_F, see S1 Table. The PCR products (between 350 and 900 bp in length depending on where the modification was made) were gel purified using the GeneJET Gel purification kit (Thermo Fisher Scientific). These PCR products were used as megaprimers for a second round of PCR (500 ng megaprimer, and 100 ng template), using the P1-2A (wt) as template to produce the modified plasmids. After the PCR and subsequent DpnI digestion of the template plasmid, the products were transformed into chemically competent Escherichia coli (E. coli) cells. Plasmids were amplified from individual colonies, purified using the GeneJet Plasmid Miniprep Kit (Thermo Fisher Scientific) and screened by Sanger Sequencing using the BigDye Terminator v.3.1 Cycle Sequencing kit and a 3500 Genetic Analyzer (Applied Biosystems). Plasmids encoding the desired modifications were amplified and purified using the QIAGEN Plasmid Midi Kit (Qiagen). All of these constructs encoded P1 that was truncated at different sites by introducing two STOP codons. The plasmids (listed in S1 Table) were labelled to denote the location of the Stop codons as follows; P1-2A (2A L1Stop), P1-2A (VP1 I205Stop), P1-2A (VP1 D199Stop), P1-2A (VP1 Y185Stop), P1-2A (VP1 L158Stop), P1-2A (VP1 A107Stop) and P1-2A (VP1 L53Stop); for primer sequences, see S1 Table.
Variants of the P1-2A plasmid that express mutant proteins with various different deletions in the C-terminal region of VP1 between residues VP1 185 and VP1 199 were created using site-directed mutagenesis, essentially as described above; for primer sequences see S1 Table. The plasmids were labelled as follows: P1-2A (VP1 Δ185–199), P1-2A (VP1 Δ185–189), P1-2A (VP1 Δ188–192), P1-2A (VP1 Δ191–195) and P1-2A (VP1 Δ194–199). Furthermore, an additional positive control was included in which 13 amino acids within VP1 were deleted, this construct was called P1-2A (VP1 Δ142–154). An earlier study had shown that FMDV with this deletion is able to replicate [20,44].
Alanine substitutions were introduced at the codons for each residue individually between VP1 185 and VP1 199, with the exception of VP1 A192, where the original alanine codon was substituted by one encoding a serine. The mutations were produced using site-directed mutagenesis, as described above; the primer sequences are listed in S2 Table. These modifications resulted in 15 different plasmids: P1-2A (VP1 Y185A), P1-2A (VP1 C186A), P1-2A (VP1 P187A), P1-2A (VP1 R188A), P1-2A (VP1 P189A), P1-2A (VP1 L190A), P1-2A (VP1 L191A), P1-2A (VP1 A192S), P1-2A (VP1 V193A), P1-2A (VP1 E194A), P1-2A (VP1 V195A), P1-2A (VP1 S196A), P1-2A (VP1 S197A), P1-2A (VP1 Q198A) and P1-2A (VP1 D199A).
Baby hamster kidney (BHK) cells (originally obtained from the ATCC (CCL-10)) were grown in 35-mm wells to about 90% confluence, when they were infected with the recombinant vaccinia virus, termed vTF7-3 [45] that expresses the T7 RNA polymerase. All the various P1-2A plasmids and the 3C plasmid (pSKRH3C [46]) express the FMDV cDNA under the control of a T7 promotor. After one hour incubation at 37°C, the vaccinia virus was removed and the cells were transfected with the specified plasmid DNA using FuGENE 6 (Promega), as described previously [47]. To obtain the highest levels of processed capsid protein expression, 1000 ng of the P1-2A plasmid alone or with 10 ng of the 3Cpro plasmid were used for each transfection of cells [28]. The cells were incubated in a CO2 incubator, at 37°C overnight and then, after removal of the medium, lysed with 500 μl Buffer C (20mM Tris-HCl (pH 8.0), 125 mM NaCl and 0.5% NP-40); the cell extracts were clarified by centrifugation at 18,000 x g for 10 min at 4°C.
Immunoblotting was performed using clarified cell lysates mixed with 2 x Laemmli sample buffer (Bio-Rad) (containing 25 mM DTT). The proteins were separated by SDS-PAGE using a 12% Bis-Tris gels (Bio-Rad) and transferred to PVDF membranes (Milipore), by wet blotting, at 200 mA for 1.5 hours. PBS containing bovine serum albumin (BSA) (5%) and Tween20 (0.1%) was used as blocking buffer (1 hour at room temperature (RT)) and dilution buffer for the guinea pig anti-FMDV O-Manisa (Man) antisera (prepared “in house”, as used previously [28]) (overnight at 4°C) and their corresponding secondary antibodies (2 hours at RT). Guinea pig anti-FMDV O-Man antisera was used for detection, since it is very efficient in detecting the denatured capsid proteins from various FMDV serotypes. PBS containing skimmed milk powder (5%) and Tween20 (0.1%) was used as blocking buffer (1 hour at RT) and dilution buffer for the primary guinea pig anti-FMDV A-Iraq antisera (prepared “in house”, as used previously [28]) and the anti-2A-peptide antibody (overnight at 4°C) and their corresponding secondary antibody (2 hours at RT). The proteins were detected using the following primary antibodies: guinea pig anti-FMDV O-Man antisera (1:1000), guinea pig anti-FMDV A-Iraq antisera (1:500) or FMDV anti-2A-peptide antibody (1:1000) (Rabbit, ABS31 Merck Millipore). Appropriate HRP-conjugated secondary antibodies (Dako) and a chemiluminescence detection kit (Pierce ECL Western Blotting Substrate, Thermo Fisher Scientific) were used to detect the proteins bound by the primary antibodies. Images were captured using a Chem-Doc XRS system (Bio-Rad).
The plasmid pO1K/A22 contains a full-length cDNA corresponding to a chimeric FMDV genome as previously described [28]. It includes the capsid coding sequence from FMDV A22 Iraq and the rest of the genome from FMDV O1K. Briefly, fragments were amplified in PCRs, using the pO1K/A22 plasmid as template, together with primers specifying the desired mutations, see S1 and S2 Tables. The PCR products were gel purified and used as megaprimers for a second round of PCR, again using the wt pO1K/A22 as template to make full-length plasmids of approximately 11,000 bp. After the PCR and subsequent DpnI digestion of the template DNA, the products were introduced into E. coli cells. Plasmids were amplified from individual colonies and sequenced. The plasmids, containing the full-length wt or mutant FMDV cDNAs, were linearized by digestion with HpaI and then transcribed in vitro using the MEGAscript T7 Transcription Kit (Thermo Fisher Scientific). An aliquot (1 μL) of each RNA sample was visualized following agarose gel electrophoresis to check yield and integrity and the rest (19 μL) was introduced into BHK cells by electroporation as described previously [28]. The cells were transferred to Falcon flasks and Eagle’s medium containing 5% calf serum was added. The cells were incubated overnight at 37°C and then harvested. Aliquots (1ml) of the harvest were inoculated onto fresh BHK cells and the appearance of cytopathic effect monitored at 1 and 2 days post-inoculation. For samples displaying CPE, viral RNA was isolated using the RNeasy Mini Kit (Qiagen) and reverse transcribed using Ready-To-Go You-Prime First-Strand Beads (GE Healthcare Life Sciences) together with random primers. The cDNA corresponding to the P1-2A region was amplified as four overlapping fragments of around 1000 bp by AmpliTaq Gold DNA Polymerase (Thermo Fisher Scientific) as described previously [28] and then sequenced. For each RNA, a negative control, lacking the reverse transcriptase, was included in the RT-PCRs to verify that the PCR products were obtained from viral RNA and not from residual plasmid template.
|
10.1371/journal.pgen.1007682 | Single copy/knock-in models of ALS SOD1 in C. elegans suggest loss and gain of function have different contributions to cholinergic and glutamatergic neurodegeneration | Mutations in Cu/Zn superoxide dismutase 1 (SOD1) lead to Amyotrophic Lateral Sclerosis (ALS), a neurodegenerative disease that disproportionately affects glutamatergic and cholinergic motor neurons. Previous work with SOD1 overexpression models supports a role for SOD1 toxic gain of function in ALS pathogenesis. However, the impact of SOD1 loss of function in ALS cannot be directly examined in overexpression models. In addition, overexpression may obscure the contribution of SOD1 loss of function in the degeneration of different neuronal populations. Here, we report the first single-copy, ALS knock-in models in C. elegans generated by transposon- or CRISPR/Cas9- mediated genome editing of the endogenous sod-1 gene. Introduction of ALS patient amino acid changes A4V, H71Y, L84V, G85R or G93A into the C. elegans sod-1 gene yielded single-copy/knock-in ALS SOD1 models. These differ from previously reported overexpression models in multiple assays. In single-copy/knock-in models, we observed differential impact of sod-1 ALS alleles on glutamatergic and cholinergic neurodegeneration. A4V, H71Y, G85R, and G93A animals showed increased SOD1 protein accumulation and oxidative stress induced degeneration, consistent with a toxic gain of function in cholinergic motor neurons. By contrast, H71Y, L84V, and G85R lead to glutamatergic neuron degeneration due to sod-1 loss of function after oxidative stress. However, dopaminergic and serotonergic neuronal populations were spared in single-copy ALS models, suggesting a neuronal-subtype specificity previously not reported in invertebrate ALS SOD1 models. Combined, these results suggest that knock-in models may reproduce the neurotransmitter-type specificity of ALS and that both SOD1 loss and gain of toxic function differentially contribute to ALS pathogenesis in different neuronal populations.
| In all SOD1 ALS patients, cholinergic spinal motor neurons degenerate, but degeneration of cortical glutamatergic neurons is less common. Despite decades of work, it remains unclear why some disease alleles (e.g. A4V) primarily affect cholinergic spinal neurons, while other alleles affect both cholinergic and glutamatergic neurons. New genome editing techniques allowed us to create the first C. elegans knock-in/single-copy models for SOD1 ALS by directly editing the C. elegans sod-1 gene to recreate SOD1 amino acid changes that cause ALS in patients. These new models are complementary to previously described overexpression models, which revealed mutant SOD1 toxic gain of function properties. By contrast, in the new C. elegans knock-in models, we find that both loss and gain of sod-1 function contribute to neurodegeneration. C. elegans cholinergic motor neuron loss is primarily driven by toxic gain of function, but glutamatergic neuron loss is primarily driven by loss of function. Only cholinergic and glutamatergic neurons degenerate in C. elegans knock-in models; dopaminergic, serotoninergic and GABAergic neurons do not. This pattern of neuronal loss is reminiscent of the pattern of neuronal loss seen in SOD1 ALS patients. Strikingly, in the C. elegans A4V model, only cholinergic neurons are lost. Our results suggest that an underlying premise of the ALS field–that identical pathological mechanisms lead to degeneration of cholinergic and glutamatergic neurons–should be reconsidered. Mechanisms that predominantly drive glutamatergic and cholinergic neuron degeneration in ALS may not be identical.
| Amyotrophic lateral sclerosis (ALS) is an adult-onset fatal neurodegenerative disorder marked by the progressive loss of glutamatergic and cholinergic motor neurons. It is the most common motor neuron disorder affecting adults. The first ALS-linked mutations were discovered in a gene encoding the antioxidant enzyme Cu/Zn superoxide dismutase 1 (SOD1) [1]. Roughly 1% of all cases are caused by mutations of SOD1.
In patients, misfolded SOD1 is a major constituent of proteinaceous inclusions in the affected neurons [2–4] and pathogenic SOD1 variants inevitably lead to cholinergic motor neuron degeneration. However, ALS is inherently heterogeneous: relative involvement of the glutamatergic corticospinal tract [5] and glutamatergic sensory neurons [6,7] differs greatly among patients, and clinical presentation of ALS, including age of disease onset, progression, severity and duration, varies [8,9]. Consequently, how mutant SOD1 mediates its toxic function in different neuronal populations remains largely unknown.
Published studies to date have primarily relied on human mutant SOD1 protein overexpression models to examine neuronal dysfunction [10–15]. Overexpression models can recapitulate several key aspects of ALS pathogenesis, including motor neuron degeneration, protein aggregation and motor dysfunction. Furthermore, most ALS SOD1 alleles have a dominant pattern of inheritance [16]. These findings support a role for gain of toxic SOD1 function in disease pathogenesis.
However, SOD1 overexpression models may not permit the study of disease mechanisms in entirety for several reasons. First, overexpression of wild type SOD1 protein has deleterious effects in model organisms [17,18], making it difficult to dissociate the impact of ALS mutations from protein overexpression. Second, overexpression models make it challenging to discover SOD1 loss of function contributions to ALS pathogenesis. Recently, single-copy SOD1 knock-in models in flies, fish and mice have been reported [19–21]. Defects observed in these knock-in models differ dramatically from those observed in overexpression models and are often less severe. Furthermore, knock-in mice and fly models suggest that both SOD1 loss and gain of function may contribute to disease pathogenesis for the limited number of SOD1 alleles tested. A single-copy ALS SOD1 knock-in mouse model shows peripheral neuropathy, reminiscent of SOD1 null mice [20]. And, eclosion defects in single-copy ALS SOD1 knock-in models in flies were rescued by the introduction of wild type SOD1, consistent with a loss of function defect [19]. Still, why SOD1 patient alleles differ in presentation and progression remains unclear, and overexpression models likely complicate this analysis.
Previously reported ALS SOD1 overexpression models in C. elegans demonstrate neuronal and muscular dysfunction [13,22–24]. To compliment these overexpression models, we generated single-copy ALS SOD1 knock-in models in C. elegans. We edited the C. elegans sod-1 gene to create A4V, H71Y, L84V, G85R and G93A missense mutations. In the resulting single-copy ALS model animals, we observed oxidative stress induced glutamatergic and cholinergic neuron degeneration. While H71Y and G85R affected both cholinergic and glutamatergic neurons, A4V and G93A models affected cholinergic neurons only. However, other neuron classes were relatively spared. Overall, we found that cholinergic and glutamatergic neurons are differentially sensitive to SOD-1 loss and gain of toxic function. Our results suggest that both loss and gain of toxic SOD-1 function may be involved in disease pathogenesis.
To design single-copy C. elegans ALS SOD1 models, we compared human SOD1 and C. elegans SOD-1 protein sequences. Alignment of the two proteins revealed 71% similarity and 56% identity between species (Fig 1A; P00441 and C15F1.7b.1, BLAST). To create A4V, H71Y, L84V, G85R and G93A models, we mutated conserved amino acid residues in the C. elegans sod-1 gene, and generated single-copy ALS models using two different strategies (Fig 1B and 1C). To avoid confusion, C. elegans models generated herein were named based on human amino acid numbering. Formal allele designations with C. elegans amino acid changes can be found in S3 Table.
Using Mos1-mediated single copy insertion (MosSCI [25]), we recreated ALS SOD1 mutations for A4V, H71Y, and G85R. MosSCI relies on excision of a known transposon to facilitate insertion of transgenic DNA fragments into a previously defined genomic location. ALS-associated sod-1 alleles for A4V, H71Y, and G85R were individually inserted into the MosSCI cxTi10882 site on chromosome IV with the sod-1 promoter, exons, introns and 3’ sequences (Fig 1B). To control for the sod-1 gene relocalization, we also inserted the entire wild type sod-1 gene in the same location on chromosome IV (Fig 1B). To facilitate the selection of transgenic animals, these transgenes were introduced alongside an unc-119(+) rescue construct. Therefore, an additional “empty” negative control carrying the unc-119(+) construct alone was generated; these lack any of the sod-1 sequences inserted in the remainder of the MosSCI models. All alleles generated by MosSCI were named with an “M” superscript (sod-1WTM, sod-1A4VM, sod-1H71YM, sod-1G85RM and emptyM), to distinguish them from the endogenous C. elegans sod-1 gene on chromosome II. Subsequently, each transgene was crossed into the sod-1(tm776) loss of function background, referred to hereafter as sod-1(-). The resulting strains were homozygous for control or single-copy ALS sod-1 transgenes. All the comparisons between ALS sod-1 models and controls reported here were made in sod-1(-) background, unless indicated otherwise.
Additionally, using CRISPR/Cas9-mediated homologous recombination (HR), we directly edited the endogenous C. elegans sod-1 gene to recreate ALS SOD1 mutations for L84V, G85R and G93A in the endogenous sod-1 gene on chromosome II (Fig 1C). CRISPR/Cas9-mediated genome editing requires introduction of silent codon changes into endogenous sod-1. Consequently, a wild type control was generated containing identical silent mutations (Fig 1C). Models generated by CRISPR/Cas9 were named with a “C” superscript (sod-1WTC, sod-1L84VC, sod-1G85RC and sod-1G93AC). To demonstrate reproducibility across strains generated by MosSCI and CRISPR/Cas9, we created the G85R allele twice, once using each technique (Fig 1B and 1C).
Single-copy models may differ from overexpression models in severity and type of defects observed [19,20]. Thus, we compared the new single-copy ALS sod-1 models to previously published neuronal overexpression models provided by the Horwich lab [13]. These animals overexpress human SOD1G85R protein in neurons and have severe locomotion defects (Fig 1D), while animals overexpressing wild type human SOD1 have relatively normal locomotion. We found that none of the single-copy ALS sod-1 model animals carrying patient amino acid changes had overt locomotion defects (Fig 1D). Furthermore, lifespan was only slightly decreased in single-copy ALS sod-1 animals, with the exception of G85R models, which had normal lifespan (S1A and S1B Fig).
ALS SOD1 mutations lead to formation of SOD1-rich proteinaceous inclusions in motor neurons [2–4]. Consistent with this observation, overexpression of human ALS SOD1 in C. elegans leads to formation of SOD1 inclusions in C. elegans neurons and muscles [13,23]. We found that expression of human wild type SOD1 tagged with YFP (hSOD1WT-YFP) resulted in small cytosolic inclusions in the ventral nerve cord motor neurons of a small fraction of wild type control animals (Fig 2A–2C). Next, we examined the impact of single-copy ALS sod-1 models on neuronal inclusions formed by human SOD1WT-YFP in the same motor neurons. With the exception of sod-1L84VC, single-copy ALS SOD1 animals were more likely to have hSOD1WT-YFP inclusions compared to wild type control animals (Fig 2A–2C; P < 0.05, chi-square test). While single-copy ALS sod-1 models increased the formation of hSOD1WT-YFP protein inclusions, loss of sod-1 did not alter formation of hSOD1WT-YFP inclusions in these neurons (Fig 2B and 2C; P = 0.61 for emptyM vs sod-1WTM and P = 0.12 for sod-1(-) vs sod-1(+); sod-1(+) refers to the standard N2 wild type unedited allele). We conclude that loss of sod-1 function does not alter propensity to form neuronal SOD1 inclusions in C. elegans motor neurons, but that most C. elegans single-copy ALS sod-1 models show increased hSOD1WT-YFP inclusion formation.
Oxidative damage may induce SOD1 misfolding, resulting in aberrant protein accumulation [23,26]. Consequently, loss of sod-1 function, coupled with an increase in oxidative damage, could exacerbate defects in ALS SOD1 models. To determine the impact of oxidative stress in single-copy ALS sod-1 models, we exposed animals to paraquat, an oxidative stress inducing herbicide [27]. Loss of sod-1 function decreased survival under paraquat-induced oxidative stress (S2A and S2B Fig, P < 0.001 for emptyM vs sod-1WTM and sod-1(-) vs sod-1(+), log-rank test), replicating a previously published finding [28]. Similarly, ALS knock-in alleles dramatically decreased survival under paraquat-induced oxidative stress, with the exception of sod-1A4VM (S2A and S2B Fig; P < 0.01, log-rank test). By contrast, without external stress, survival was unaffected or slightly decreased in these genotypes (S1A and S1B Fig). We conclude that paraquat-induced oxidative stress may exacerbate or reveal survival defects in single-copy/knock-in models.
We also examined the impact of loss of sod-1 function on formation of SOD1 inclusions after exposure to oxidative stress. Formation of hSOD1-YFP inclusions in ALS sod-1 model animals and in animals lacking sod-1 was measured after a brief period of paraquat exposure (3 hours). A significantly larger fraction of sod-1H71YM and sod-1G85RM animals had more neuronal inclusions per animal compared to wild type controls (Fig 2D, P < 0.05, chi-square test). Moreover, sod-1A4VM, sod-1H71YM and sod-1G85RM animals had larger neuronal inclusions compared to wild type controls (Fig 2E). Loss of sod-1 function in emptyM controls lead to formation of bigger, but not more, neuronal inclusions (Fig 2D and 2E). In summary, ALS sod-1 models lead to increased or accelerated formation of hSOD1WT-YFP protein inclusions, with the exception of sod-1L84VC. Loss of sod-1 function did not initiate formation of hSOD1WT-YFP inclusions in C. elegans neurons, suggesting sod-1 gain of function likely drives formation of hSOD1WT-YFP inclusions in ALS models.
Because hSOD1WT-YFP inclusions were not increased in the C. elegans sod-1L84VC model, the utility of this model remains unclear. To be comprehensive in our analysis, we include sod-1L84VC in all studies presented below.
Distinct neuronal populations are at greater risk for degeneration in ALS patients. Cholinergic and glutamatergic motor neurons are often disproportionately affected [29]. We examined the impact of ALS-associated mutations on cholinergic motor neuron survival in single-copy and overexpression models for ALS SOD1 in C. elegans. Survival of cholinergic motor neurons was assessed based on retention/loss of GFP or mCherry in animals carrying the unc-17p::GFP or cho-1p::mCherry transgenes, respectively, which express fluorescent proteins specifically in cholinergic neurons (Fig 3A). In the absence of oxidative stress, cholinergic motor neurons were intact in all examined strains (more than 30 animals per genotype scored with no neurons missing; [13]). However, overnight exposure to paraquat-induced cholinergic motor neuron degeneration in sod-1A4VM, sod-1H71YM, sod-1G85RM, sod-1G85RC and sod-1G93AC animals, as well as in animals overexpressing the human SOD1G85R-YFP protein (hSOD1G85R-YFPOE), compared to appropriate wild type controls (Fig 3B, Part II and III; P < 0.05, chi-square test). Conversely, degeneration was not observed in control animals lacking sod-1 or in sod-1L84VC animals after paraquat treatment (Fig 3B, Part I and II). Between 2 to 4 motor neurons were lost in affected animals out of the 20 neurons scored in the posterior ventral nerve cord. These results suggest that loss of sod-1 function is not sufficient to induce cholinergic motor neuron loss after exposure to oxidative stress. By contrast, most ALS mutations sensitize animals to oxidative stress and lead to cholinergic motor neuron loss.
The majority of ALS SOD1 patients carry a wild type SOD1 allele in addition to the mutated SOD1 allele. Consequently, we tested the impact of wild type endogenous sod-1 on oxidative stress induced motor neuron degeneration. Crossing the endogenous unedited wild type sod-1(+) allele on chromosome II into sod-1A4VM, sod-1H71YM or sod-1G85RM animals did not rescue paraquat-induced cholinergic motor neuron degeneration (Fig 3B, Part IV), consistent with a gain of toxic function mechanism in ALS sod-1 alleles. Similarly, crossing sod-1(+) into the hSOD1G85R-YFPOE animals failed to rescue stress induced degeneration (Fig 3B, Part V). Thus, in both single-copy and overexpression ALS SOD1 models, SOD1 gain of toxic function and oxidative stress results in cholinergic neurodegeneration.
Neuromuscular junction (NMJ) dysfunction is an early defect in ALS patients [30]. Heterologous overexpression of human SOD1G85R in C. elegans neurons also leads to NMJ dysfunction [13], based on resistance to aldicarb, an inhibitor of acetylcholinesterase. Exposure to aldicarb leads to acetylcholine buildup at the NMJ, with the consequent hyperexcitation of postsynaptic muscles and paralysis over a characteristic time course (Fig 4A). Either hypersensitivity or resistance to aldicarb indicates defective neuromuscular signaling and suggests increased or decreased overall NMJ cholinergic signaling.
We confirmed that neuronal overexpression of the human mutant SOD1G85R causes aldicarb resistance, compared to animals overexpressing the human wild type SOD1 protein (Fig 4B, left panel; P < 0.05, log-rank test and [13]). However, we found that both transgenic strains were more resistant to paralysis by aldicarb than non-transgenic wild type C. elegans (Fig 4B, left panel). Thus, overexpression of wild type human SOD1 is deleterious and causes defects in C. elegans NMJ function.
Next, we examined the impact of single-copy ALS models on NMJ function. The response of wild type control animals was indistinguishable from non-transgenic C. elegans (Fig 4C). And, in single-copy sod-1A4VM, sod-1L84VC, sod-1G85RM and sod-1G93AC animals, response to aldicarb was normal (Fig 4C; sod-1A4VM vs sod-1WTM, P = 0.20; sod-1L84VC vs sod-1WTC, P = 0.76; sod-1G85RM vs sod-1WTM, P = 0.071; sod-1G93AC vs sod-1WTC, P = 0.44, log-rank test). However, sod-1G85RC and sod-1H71YM animals were hypersensitive to aldicarb compared to wild type controls (Fig 4C; P < 0.05, log-rank test). Thus, even though single-copy models are not uniform in their response to aldicarb, none of these strains were resistant to aldicarb-induced paralysis. Overall, we observed dramatically different consequences for C. elegans NMJ function in overexpression versus single-copy ALS model animals.
The impact of sod-1 loss of function on C. elegans NMJ function has not been reported previously. We found that loss of endogenous sod-1 function lead to aldicarb hypersensitivity, compared to standard N2 strain sod-1(+) controls (Fig 4C, left panel; P < 0.001 for sod-1(-) vs sod-1(+)). To confirm this change was due to decreased sod-1 activity, and to establish conservation of SOD1 function across species, we undertook phenotypic rescue studies. Introduction of C. elegans sod-1WTM to sod-1(-) fully restored normal response to aldicarb (Fig 4C, right panel; sod-1WTM vs emptyM), and introduction of neuronal human SOD1WT-YFP partially restored aldicarb response (Fig 4B, right panel; sod-1(-); hSOD1WT-YFPOE vs sod-1(-)). These results suggest that NMJ functional defects in ALS sod-1 model animals may be, in part, driven by loss of sod-1 function.
To confirm that sod-1 loss of function contributes to NMJ defects, we also examined the consequences of altering sod-1 dosage. First, we confirmed that restoring sod-1 function would restore normal NMJ function in sod-1H71YM animals. We crossed the unedited endogenous sod-1(+) allele into sod-1H71YM allele background. These animals had normal response to aldicarb (Fig 4D, left panel; sod-1(+); sod-1H71YM vs sod-1(+); sod-1WTM, P = 0.05).
To confirm that aldicarb hypersensitivity in ALS sod-1 animals is driven primarily by sod-1 loss of function, we examined aldicarb response in animals heterozygous for ALS sod-1 transgenes. Homozygous ALS sod-1 animals were crossed to homozygous sod-1WTM or emptyM males carrying a GFP-expressing transgene. Then, the heterozygous GFP-positive cross-progeny were examined for aldicarb resistance after paraquat treatment. As control, we confirmed that emptyM/emptyM cross-progeny were hypersensitive to aldicarb compared to sod-1WTM/sod-1WTM cross-progeny (Fig 4D, right panel).
Introduction of one functional copy of sod-1 yielded sod-1WTM/emptyM animals, which were normal in response and indistinguishable from sod-1WTM/WTM animals (Fig 4D, right panel; sod-1WTM/WTM vs sod-1WTM/emptyM, P = 0.16). Next, we determined if the sod-1H71YM aldicarb response defect is recessive. We found that heterozygous sod-1WTM/H71YM animals had normal aldicarb response (Fig 4D, sod-1WTM/H71YM vs sod-1WTM/WTM, P = 0.23). Taken together, these findings suggest that loss of sod-1 function contributes to the NMJ functional defects in both sod-1(-) and sod-1H71YM animals.
NMJ functional defects are often associated with defective localization of NMJ presynaptic proteins. A previous study has reported defective presynaptic synaptobrevin/SNB-1 localization at the NMJ of the human SOD1G85R overexpression animals [13]. To assess the impact of ALS sod-1 models on NMJ proteins, we determined the localization and intensity of fluorescently-labelled pre-synaptic proteins in live animals using previously described transgenic strains and protocols in adult sod-1H71YM and sod-1G85RM animals. Size, distribution and intensity of presynaptic synaptobrevin/SNB-1 and intersectin-1/ITSN-1 fluorescent punctae were normal in sod-1H71YM and sod-1G85RM animals (S3A and S3B Fig). Unlike overexpression model animals, presynaptic protein levels and localization were not different in single-copy ALS sod-1 animals. While this approach may not reveal subtle defects in NMJ structure, we conclude that sod-1H71YM and sod-1G85RM have no major impact on the NMJ structure or number in C. elegans.
Considering that oxidative stress exacerbated defects in single-copy ALS sod-1 animals, we examined the impact of this stress on locomotion. Adult sod-1G85RC animals were treated overnight with paraquat and tested for swimming locomotion (Fig 4E) using computer vision software CeleST [31]. Increased locomotion activity was observed in sod-1G85RC and sod-1(-) animals compared to appropriate wild type controls (Fig 4F–4H, P < 0.05, Kruskal-Wallis test); Activity Index, Travel Speed, and Wave Initiation Rate metrics were all increased, consistent with increased activity. Combined, NMJ defects and locomotion assays results suggest that loss of sod-1 function may increase NMJ signaling, without dramatic perturbation in synapse structure and number. Cholinergic motor neuron defects might be expected to change locomotion in sod-1G85RM model animals; we found that oxidative stress also increased their locomotion activity, compared to controls. At first glance, this is counterintuitive as cholinergic motor neuron degeneration should impair locomotion. However, modest loss of cholinergic motor neurons does not dramatically impair C. elegans locomotion based on laser ablation studies [32,33]. Additionally, paraquat is a strongly aversive stimulus and C. elegans can respond to noxious environments with a coordinated escape response by decreasing spontaneous reversals and increasing forward locomotion speed [34]. Consistent with this, we observed that placing sod-1G85RM, sod-1H71YM sod-1G85RC, or sod-1 loss of function animals overnight on culture dishes containing paraquat resulted in high rates of escape; after 24 hours, roughly 40% of sod-1H71YM, 47% of emptyM, and 38% of sod-1G85RM animals left the 2.5mM paraquat culture dishes, compared to only 2% of sod-1WTM control animals. Similarly, 23% of sod-1G85RC animals left after overnight exposure to paraquat, compared to only 1% of sod-1WTC animals (3 trials, >80 animals per genotype total). Combined, our results suggest that the increased locomotion activity observed in sod-1G85RC animals is likely an escape response and that single-copy ALS sod-1 models differ significantly from previously described human SOD1 overexpression models in their impact on NMJ function.
Neuron loss in ALS patients is generally limited to specific subsets of cholinergic and glutamatergic neurons [29]. We examined the specificity of neurodegeneration in single-copy/knock-in ALS model animals. Neurotransmitter-type specific GFP reporter constructs for glutamatergic, dopaminergic, serotonergic, and GABAergic neurons were used to facilitate scoring of neuron loss, i.e. dat-1p::GFP, tph-1::GFP, unc-47p::GFP, and osm-11p::GFP. As the single-copy sod-1G85R models show robust cholinergic motor neuron loss after paraquat stress, this genetic model was used to assess loss of other GFP-labelled neurons after paraquat-induced oxidative stress. For dopaminergic neurons, all four CEPs and both ADE neurons were scored, as well as ADE neuron sensory processes. For serotonergic neurons, both NSM and both ADF neurons were scored. For GABAergic neurons, nineteen GABAergic motor neurons in the ventral cord were scored. For glutamatergic neurons, both ASH neurons were scored.
A previous study in C. elegans found significant dopaminergic neuron degeneration in younger animals after a more extended exposure period to paraquat [35]. However, we found no dopaminergic or serotonergic neuron loss in sod-1G85RC, sod-1(-), or control animals after paraquat-induced oxidative stress (Figs 5A, 5B, S4A and S4B), and dopaminergic ADE neuron sensory processes did not degenerate after paraquat treatment (S4B Fig). We conclude that neither sod-1G85RC nor sod-1 loss leads to loss of dopaminergic or serotonergic neurons in single-copy ALS animals.
C. elegans have both cholinergic and GABAergic motor neurons that directly synapse onto muscles; these neurons coordinately and reciprocally regulate muscle excitability [36]. After paraquat treatment, roughly 25% of sod-1G85RM animals had GABAergic motor neuron loss, while sod-1WTM or sod-1(-) animals had less than 10% affected animals (S4D Fig). No GABAergic motor neurons were lost in more than 30 sod-1G85RM animals tested without paraquat stress. Extending this analysis, we examined GABAergic motor neuron survival in sod-1A4VM and sod-1H71YM animals. Less than 20% and 10% of animals lost GABAergic motor neurons after paraquat treatment in these genotypes, respectively, which was not significant (S4D Fig). We conclude that GABAergic motor neurons are affected only in the single-copy sod-1G85RM model.
A subset of ALS patients have glutamatergic cortical motor neuron degeneration and/or sensory neuropathy [5–7,29]. After paraquat treatment, approximately 20% of the glutamatergic ASH sensory neurons were lost in sod-1G85RC animals (S4C Fig). Similarly, about 30% of the glutamatergic ASH sensory were lost in sod-1(-) animals after paraquat treatment, while virtually all neurons were intact in sod-1WTC animals (S4C Fig). No neurons were lost without paraquat stress in any genotype. These results suggested that glutamatergic neurons are also lost in the single-copy sod-1G85RC ALS model animals. Combined with the results above, we find that neuron loss occurs in sod-1G85RC animals after oxidative stress, and primarily affects motor neurons and glutamatergic neurons, with relative sparing of serotonergic and dopaminergic neurons.
Degeneration of neuronal processes has been previously reported in a single-copy ALS SOD1 mouse model and in other C. elegans models of neurodegenerative disease [20,37]. In the head and tail, C. elegans have glutamatergic sensory neuron processes that are exposed to the environment, which allows uptake of lipophilic fluorescent dyes, such as DiD (C67H103ClN2O3S). Degenerative sensory process loss or cell death completely stops dye uptake by individual neurons, as shown in C. elegans models of Huntington’s Disease polyglutamine toxicity [37]. Without paraquat-induced oxidative stress, dye uptake was normal in all C. elegans ALS sod-1 model animals and in animals lacking sod-1 function (more than 30 animals scored per genotype, more than two trials). ASH, PHA and PHB neurons failed to take up DiD in sod-1(-) animals after paraquat-induced oxidative stress (Fig 5D and 5G), but DiD uptake was normal in five other classes of glutamatergic neurons. Single-copy/knock-in sod-1H71YM, sod-1G85RM, sod-1G85RC and sod-1L84VC animals had defective dye uptake in the same neurons after exposure to oxidative stress, although the penetrance of this defect varied between genotypes with little degeneration in sod-1L84VC (Fig 5C, 5D and 5G). Again, five other classes of glutamatergic neurons that take up DiD were unaffected in all of these genotypes. No dye uptake defects were observed in sod-1A4VM and sod-1G93AC animals, even after paraquat treatment (Fig 5D and 5G).
The results above suggest that decreased sod-1 function renders glutamatergic neurons hypersensitive to oxidative stress. To test this hypothesis, we undertook phenotypic rescue experiments. As SOD-1 function is conserved across species, we crossed the previously described transgene expressing human wild type SOD1-YFP in C. elegans neurons into sod-1(-) animals. DiD uptake defects were dramatically reduced in the resulting sod-1(-); hSOD1WT-YFPOE animals (Fig 5G, right), consistent with conserved dismutase activity and loss of sod-1 function contributing to this defect. The G85R missense allele is reported to dramatically decrease human SOD1 dismutase activity [38]. We found that crossing the hSOD1G85R-YFPOE transgene into sod-1(-) animals did not ameliorate the dye uptake defects (Fig 5G, right). We conclude that decreased sod-1 function can lead to oxidative stress induced neurodegeneration of glutamatergic neurons in C. elegans.
To confirm that glutamatergic neurodegeneration in ALS sod-1 models is driven primarily by sod-1 loss of function, we examined glutamatergic neurodegeneration in animals heterozygous for ALS sod-1 transgenes. Homozygous ALS sod-1 animals were crossed to homozygous sod-1WTM or emptyM males carrying a GFP-expressing transgene. Then, the heterozygous GFP-positive cross-progeny were examined for glutamatergic neuron degeneration after paraquat treatment. As control, we confirmed that emptyM/emptyM cross-progeny had dye-uptake defects, compared to sod-1WTM/sod-1WTM cross-progeny. (PHA and PHB in Fig 5H, ASH in S4E Fig). Animals carrying a single sod-1WTM allele (sod-1WTM/emptyM) had no degeneration, confirming that loss of sod-1 function is not dominant. When we examined animals hemizygous for sod-1H71YM or sod-1G85RM (sod-1H71YM/emptyM and sod-1G85RM/emptyM), they had ASH, PHA and PHB dye-uptake defects after paraquat treatment, suggesting that these alleles cause sod-1 loss of function (Figs 5H and S4E). Consistent with this, dye-uptake defects were not seen in sod-1WTM/sod-1H71YM or sod-1WTM/sod-1G85RM animals, (sod-1WTM/ALS allele vs emptyM/ALS allele, Figs 5H and S4E). Combined, these studies suggest that H71Y and G85R decrease sod-1 function, leading to stress induced degeneration in a subset of glutamatergic neurons.
Both SOD1 loss and gain of function in ALS may contribute to degeneration and their relative contributions may shift in different neuronal subtypes. Overexpression models are ideally suited to examine gain of function consequence of disease alleles. We assessed the consequences of overexpressing wild type or mutant human SOD1 in the glutamatergic neurons of otherwise normal C. elegans. Degeneration was not observed in hSOD1WT-YFPOE animals when the unedited endogenous sod-1(+) allele was present, based on dye-filling after paraquat treatment. By contrast, a modest level of degeneration was observed in sod-1(+); hSOD1G85R-YFPOE animals (Fig 5H, right). This result is consistent with a deleterious impact of overexpressed human SOD1 G85R, which may antagonize endogenous sod-1 or have a novel, toxic gain of function.
Neurodegenerative diseases lead to behavioral changes. C. elegans ASH neurons are critical for eliciting a mechanosensory response when the nose contacts a physical barrier during forward locomotion. Loss of both ASH neurons eliminates ~60% of this behavioral response, but response is intact if one ASH neuron is present [39]. We examined mechanosensory response after exposure to paraquat and found defective nose touch avoidance response in sod-1G85RM and emptyM animals (Fig 5E and 5F). This defective response is consistent with dye uptake defects and neuronal loss observed in these genotypes. We also undertook a behavioral rescue experiment. Animals expressing neuronal hSOD1WT-YFP had normal nose touch avoidance response compared to non-transgenic control animals (Fig 5F, bottom). These results suggest that nose-touch avoidance defects in ALS sod-1 model animals may be, in part, driven by loss of sod-1 function.
Herein, we report the first single-copy/knock-in models for ALS SOD1 in C. elegans generated using targeted genome editing. A4V, H71Y, L84V, G85R and G93A patient amino acid changes were introduced into the C. elegans sod-1 gene, using two different strategies. We found that C. elegans single-copy/knock-in models for A4V, H71Y, G85R and G93A accelerated accumulation of the YFP-tagged wild type human SOD1 protein in C. elegans motor neurons. We also found that SOD1 mutations differentially impact glutamatergic and cholinergic neurons in C. elegans. A4V, G85R, H71Y and G93A lead to oxidative stress induced loss of cholinergic motor neurons, while L84V, H71Y and G85R lead to oxidative stress induced degeneration of glutamatergic neurons. Other neuronal populations (dopaminergic and serotonergic) were relatively spared, suggesting that single-copy ALS SOD1 knock-in models in C. elegans may recapitulate the selective sensitivity of cholinergic and glutamatergic neurons in ALS caused by SOD1 mutations. Furthermore, we found that sod-1 loss of function is a major contributor to glutamatergic neuron degeneration after oxidative stress. However, sod-1 gain of function likely drives cholinergic motor neuron degeneration. Combined, these results suggest that C. elegans knock-in models, at a minimum, complement overexpression models and provide unique insights into why different SOD1 mutations lead to degeneration in different types of neurons.
Cytoplasmic SOD1 aggregates are found in ALS SOD1 patients and in most ALS SOD1 overexpression models [15,23,40–42]. C. elegans single-copy/knock-in models for A4V, H71Y, G85R and G93A increased accumulation of the YFP-tagged human wild type SOD1 protein in small cytosolic inclusions within C. elegans motor neurons, even under optimal growth conditions. These hSOD1WT-YFP inclusions were possibly seeded by misfolded mutant C. elegans SOD-1 protein in single-copy/knock-in models. Are these inclusions caused by loss or gain of sod-1 function? Because inclusions were not increased in C. elegans with decreased sod-1 function, this is likely a gain of function defect. It seems likely that introduction of the ALS patient amino acid changes into the C. elegans SOD-1 protein conferred a neomorphic/novel gain of function that increased inclusion propensity. In contrast with these four alleles, we found that hSOD1WT-YFP inclusions were not increased in the motor neurons of sod-1L84VC animals. Previous studies have shown that driving human L84V SOD1 expression in cells or transgenic mice results in SOD1 aggregation [40,43]. The failure of sod-1L84VC to increase inclusions in C. elegans neurons may arise from differences between human and C. elegans SOD1 proteins. Consequently, the usefulness of the C. elegans sod-1L84VC model remains unclear.
ALS is characterized by degeneration of lower cholinergic motor neurons. However, a significant fraction of patients present with cortical glutamatergic neuron degeneration [44]. Given that cholinergic motor neurons always degenerate and the technical challenges of studying cortical neurons, most work in the ALS field focuses on cholinergic motor neurons. In overexpression ALS SOD1 mice and in other models, high level expression of ALS SOD1 patient alleles leads to degeneration that has been ascribed to a SOD1 neomorphic/novel toxic gain of function [15,41]. This is consistent with the observation that most ALS SOD1 alleles are dominant in patients. Under standard culture conditions, C. elegans cholinergic motor neurons do not die in young adult animals, in either single-copy/knock-in models described herein or in human SOD1 ALS models overexpression models described previously [13]. We found that cholinergic motor neurons in all single-copy/knock-in ALS sod-1 models were hypersensitive to oxidative stress, except sod-1L84VC. Oxidative stress lead to motor neuron loss in A4V, H71Y, G85R and G93A single-copy/knock-in animals, and in hSOD1G85R-YFPOE overexpression model animals [13], which has not been reported previously. Motor neurons were not lost in animals lacking sod-1 function and introduction of wild type sod-1 did not rescue cholinergic motor neuron degeneration in A4V, H71Y or G85R single-copy/knock-in models. Oxidative stress exposure also increased the number and/or size of neuronal SOD1 inclusions in these ALS SOD1 models. These results are consistent with previous work showing misregulation of cellular stress response pathways in ALS SOD1 models [45] and increased SOD1 aggregation after oxidative stress [23,46]. Combined, these results suggest that A4V, H71Y, G85R and G93A SOD1 ALS alleles confer a toxic gain of function that is deleterious to C. elegans cholinergic neurons.
Defects in glutamatergic corticospinal tracts can be detected in many ALS patients [5,29] and glutamatergic sensory neurons are also lost in patients [6,7]. In C. elegans single-copy/knock-in G85R and H71Y SOD1 ALS model animals, glutamatergic neurons were also hypersensitive to oxidative stress; paraquat treatment led to degeneration of neurons in these animals, but not in control animals. Similarly, loss of sod-1 function resulted in glutamatergic neuron degeneration after oxidative stress. The defects in H71Y and G85R single-copy/knock-in animals are recessive and introduction of hSOD1WT-YFP ameliorated defects in sod-1(-) animals. Combined, these results suggest that oxidative stress hypersensitivity in glutamatergic neurons is a C. elegans sod-1 loss of function defect. This is consistent with studies in mice; SOD1 loss of function results in degenerative changes [38,47]. The neurodegeneration observed here in C. elegans glutamatergic and cholinergic neurons stands in contrast to the lack of neurodegeneration observed in dopaminergic and serotonergic neurons under the same oxidative stress conditions. Additionally, even after oxidative stress, C. elegans GABAergic motor neurons are relatively spared in single-copy/knock-in models. This level of specificity is unusual and suggests that the mechanisms underlying ALS specificity for specific neuronal classes/neurotransmitter subtypes may be recapitulated in C. elegans knock-in models.
Work with overexpression models has been critical for the field and has definitively shown that ALS SOD1 proteins have toxic gain of function properties. However, use of overexpression models is subject to two caveats. First, overexpression models compare the toxic effects of the mutant SOD1 protein to the effects of overexpressing the wild type protein. We and others find that overexpressing wild type SOD1 has deleterious consequences [17,41,48]. Increased levels of wild type human SOD1 alters NMJ function in C. elegans; hSOD1WT-YFPOE animals paralyze more slowly than normal animals in the presence of aldicarb (Fig 4B). And, increased levels of human wild type SOD1 protein in C. elegans neurons decreased survival under paraquat-induced oxidative stress (S2C Fig). The deleterious consequences of wild type SOD1 protein overexpression likely complicate analysis of ALS using overexpression models. The second caveat is that overexpression studies are not designed to determine if ALS SOD1 alleles also decrease SOD1 function in vivo and thereby contribute to ALS-associated defects. Although ALS SOD1 overexpression models will continue to play a critical role in understanding gain of function mechanisms underlying ALS, we suggest that relying exclusively on these models is not optimal.
There is mounting evidence that SOD1 loss of function contributes to ALS-associated pathology. Studies in Drosophila and mouse suggest that SOD1 loss of function contributes to defects in SOD1 ALS models [19,30]. Our results with C. elegans models support this; both SOD1 gain and loss of function contribute to defects. Clearly C. elegans sod-1 loss of function renders glutamatergic neurons hypersensitive to oxidative stress, leading to degeneration and death. By contrast, SOD1 gain of function is the predominant driver of cholinergic motor neuron loss after oxidative stress; neurons are lost in both overexpression and single-copy/knock-in model animals. However, oxidative stress is still required for cholinergic neuron loss in these models, suggesting decreased sod-1 activity may also contribute. It is unclear why oxidative stress is required for motor neuron loss in C. elegans models. There are at least two possibilities: oxidative stress may induce premature aging [49] or mutant ALS SOD1 may impair oxidative stress response by antagonizing normal SOD1 function [28]. In the latter scenario, ALS SOD1 alleles might cause an antimorphic gain of function, in addition to the widely appreciated neomorphic/novel toxic gain of functions. An obvious molecular mechanism for mutant ALS SOD1 to antagonize normal SOD1 function is to drive misfolding and/or sequestration of normal SOD1 into aggregates. Potential antimorphic SOD1 gain of function is supported by evidence that increasing wild type SOD1 levels can exacerbate defects caused by mutant ALS SOD1 in other model systems [19,48], that expressing human SOD1G85R in wild type C. elegans causes hypersensitivity to oxidative stress (S2C Fig), and that expression of human SOD1G85R in C. elegans glutamatergic neurons results in glutamatergic neuron degeneration after oxidative stress (Fig 5H).
Is there a direct mapping of neurotransmitter-subtype sensitivity to different sod-1 disease alleles between human ALS patients and C. elegans knock-in models? This remains unclear for two reasons: human genetic diversity and cross-species conservation. First, A4V is the most common SOD1 disease allele in the United States patient population. Because of the large A4V patient population, one can be confident that this allele usually leads to cholinergic spinal motor neuron degeneration and loss, with characteristic sparing of glutamatergic upper motor neurons. The C. elegans sod-1 A4V model reproduces this neurotransmitter-subtype sensitivity, as only cholinergic motor neurons are lost after oxidative stress, not glutamatergic neurons. By contrast, the other ALS patient alleles examined here are much less frequent in the patient population, with H71Y reported in only 1 family [50] and L84V in two patients [50–52]. Human populations are genetically diverse and we expect that this diversity will impact many aspects of ALS, including which neurons are predominantly affected in a given individual or family. Consequently, we are unable to cross-examine disease severity and progression in patients carrying different SOD1 mutations. And, ALS may be polygenic in some patients, increasing the challenge of ascribing specific defects to specific patient alleles. Indeed, previous studies on ALS SOD1 enzymatic activity failed to show a direct link between loss of enzymatic function and disease severity [53,54]. Additionally, in the patient populations, differences in lifestyle and environmental exposure may also impact ALS. By contrast, C. elegans and other model organisms are relatively isogenic and live in homogeneous environments, which may facilitate studies that directly compare disease alleles. Finally, in patients, ALS impacts neurons in mid-life suggesting that aging or accumulated damage may contribute to disease. Results presented here focus on stress induced degeneration in young adult animals; we have not yet explored how aging might influence neuronal susceptibility to degeneration in these models. The combined efforts of patients, human geneticists, clinicians, researchers using model organisms, and scientists using other approaches will be required to untangle the connections between different SOD1 ALS alleles, neuronal populations affected in the corresponding patients, and the genetic background in diverse patient populations.
Results presented herein suggest that an underlying premise of the ALS field–that identical pathological mechanisms lead to degeneration of cholinergic and glutamatergic neurons–should perhaps be reconsidered. Mechanisms contributing to glutamatergic and cholinergic neurons may not be identical. The differential susceptibility of cholinergic and glutamatergic neurons in C. elegans single-copy/knock-in sod-1 models suggests that 1) decreased sod-1 function may be more deleterious for glutamatergic neurons and 2) gain of function may be the major contributor to cholinergic neuron degeneration. To our knowledge, this hypothesis has not been previously explored and it may shed light on the connections between ALS and Frontotemporal Dementia (FTD). There is considerable genetic and pathological overlap between these diseases [55], but it remains unclear why specific genes are associated only with ALS, only with FTD, or associated with both diseases. Exploring the hypothesis that mechanisms underlying glutamatergic neurodegeneration are distinct from mechanisms underlying cholinergic neurodegeneration may be useful in delineating and dissecting the pathological pathways that underlie these devastating diseases.
Cholinergic (unc-17p::GFP or cho-1p::mCherry), GABAergic (unc-47p::GFP), glutamatergic (osm-10p::GFP), serotonergic (tph-1p::GFP) and dopaminergic (dat-1p::GFP) specific neuronal markers were individually crossed into ALS models to assess neuron loss. A full list of strains used in this study can be found in S2 Table. Day 1 adult animals were mounted on 2% (vol/vol) agar pads and immobilized with 30 mg/mL 2-3-butanedione monoxime (BDM, Sigma) in M9 buffer. Fluorescent neurons were visualized and scored at the microscope for cell death based on loss of neuronal GFP under 63x or 100x objectives (Zeiss AxioImager ApoTome and AxioVision software v4.8). For scoring cholinergic (unc-17p::GFP or cho-1p::mCherry) neurons, animals missing at least two neurons were scored as defective. For scoring unc-47p::GFP, nineteen GABAergic ventral nerve cord motor neurons were scored. For scoring osm-10p::GFP, only ASH amphid sensory neurons neurons were scored due to variable/faint GFP expression in ASI and PHB neurons. For paraquat trials, animals were exposed to 2.5 mM paraquat overnight on plates.
Day 1 adult animals were washed off plates with M9 and incubated with DiD (Fisher DilC18(5) D307) in a microfuge tube as in [58]. After 1.5 hours, animals were spun down at 10000 rpm for 1 min, and transferred to a regular NGM plate. After 1 hour, animals were mounted on 2% (vol/vol) agar pads and immobilized with 30 mg/mL 2-3-butanedione monoxime (BDM, Sigma) in M9 buffer. Fluorescent neuronal cell bodies were visualized and scored for lack of dye uptake under 63x or 100x objectives (Zeiss AxioImager ApoTome and AxioVision software v4.8). For paraquat trials, animals were exposed to 2.5 mM paraquat overnight on plates.
Animal reared under normal culture conditions at 25 oC were scored for survival on alternating days starting from the first day of adulthood. FUDR was omitted from the growth medium as it alters C. elegans lifespan for some genotypes [59]. To avoid progeny contamination and overcrowding, aging animals were transferred to a new seeded plate every other day until all animals stopped laying eggs. Animals unresponsive to touch were scored as dead. To score survival on paraquat, we prepared fresh 2.5 mM paraquat (Sigma-Aldrich 856177) plates every week. Again, animals were transferred to new paraquat plates every day until all animals stopped laying eggs. In both assays, bagged animals or animals that left the plate were censored; these animals were included in lifespan determinations until the day before censoring.
Day 1 adult animals expressing human SOD1WT-YFP were mounted on 2% (vol/vol) agar pads and immobilized with 30 mg/mL 2-3-butanedione monoxime (BDM) (Sigma) in M9 buffer. Animals were quantified for inclusions within the motor neurons along the ventral nerve cord under 63x or 100x objectives (Zeiss AxioImager ApoTome and AxioVision software v4.8). For paraquat trials, animals were exposed to 2.5 mM paraquat for 3 hours on plates, as overnight exposure to paraquat problematically increased background fluorescence.
Animals reared at 25 oC were scored for paralysis on 1 mM aldicarb (Sigma-Aldrich 33386) over the course of 7 hours. NGM plates containing 1 mM aldicarb were freshly poured the day before the assay. Aldicarb plates were seeded with 30 ul of OP50, and left to dry open-lid under the hood for 30 minutes. Day 1 adult animals were then transferred onto aldicarb plates, and scored for paralysis every hour. Aldicarb-induced paralysis was scored as inability to move/pump to sequential prodding with a metal wire twice in the tail and then twice in the head. Paralyzed animals were removed from the plate and not re-counted.
For synaptic puncta imaging, day 1 adult animals were mounted on 2% (vol/vol) agar pads and immobilized using 30 mg/mL 2-3-butanedione monoxime (BDM) (Sigma) in M9 buffer. Images were captured in z-stacks from dorsal cord posterior to vulva (100x objective, Zeiss AxioImager ApoTome and AxioVision software v4.8). Data from three independent trials (n > 20 animals in total/genotype) was analyzed. Puncta total intensity, width, and linear density were quantified using the Punctaanalyser program in Matlab (v6.5; Mathworks, Inc., Natick, MA, USA; RRID:SCR_001622) [60].
Data collection and analysis were performed by experimenters blinded to genotype and, when possible, treatment. Quantitative data was analyzed using Graph Pad Prism 6 software (La Jolla, CA). Statistical significance for the survival and aldicarb resistance assays was determined with log-rank test. Kruskal-Wallis test was used to determine statistical significance for the swimming locomotion assays. For the remainder of the assays, two-tailed t-test or chi-square test was used to determine significance. A value of P < 0.05 was used to establish statistical significance. Error bars in figures represent error of the mean (S.E.M.).
|
10.1371/journal.pcbi.1003731 | iRegulon: From a Gene List to a Gene Regulatory Network Using Large Motif and Track Collections | Identifying master regulators of biological processes and mapping their downstream gene networks are key challenges in systems biology. We developed a computational method, called iRegulon, to reverse-engineer the transcriptional regulatory network underlying a co-expressed gene set using cis-regulatory sequence analysis. iRegulon implements a genome-wide ranking-and-recovery approach to detect enriched transcription factor motifs and their optimal sets of direct targets. We increase the accuracy of network inference by using very large motif collections of up to ten thousand position weight matrices collected from various species, and linking these to candidate human TFs via a motif2TF procedure. We validate iRegulon on gene sets derived from ENCODE ChIP-seq data with increasing levels of noise, and we compare iRegulon with existing motif discovery methods. Next, we use iRegulon on more challenging types of gene lists, including microRNA target sets, protein-protein interaction networks, and genetic perturbation data. In particular, we over-activate p53 in breast cancer cells, followed by RNA-seq and ChIP-seq, and could identify an extensive up-regulated network controlled directly by p53. Similarly we map a repressive network with no indication of direct p53 regulation but rather an indirect effect via E2F and NFY. Finally, we generalize our computational framework to include regulatory tracks such as ChIP-seq data and show how motif and track discovery can be combined to map functional regulatory interactions among co-expressed genes. iRegulon is available as a Cytoscape plugin from http://iregulon.aertslab.org.
| Gene regulatory networks control developmental, homeostatic, and disease processes by governing precise levels and spatio-temporal patterns of gene expression. Determining their topology can provide mechanistic insight into these processes. Gene regulatory networks consist of interactions between transcription factors and their direct target genes. Each regulatory interaction represents the binding of the transcription factor to a specific DNA binding site near its target gene. Here we present a computational method, called iRegulon, to identify master regulators and direct target genes in a human gene signature, i.e. a set of co-expressed genes. iRegulon relies on the analysis of the regulatory sequences around each gene in the gene set to detect enriched TF motifs or ChIP-seq peaks, using databases of nearly 10.000 TF motifs and 1000 ChIP-seq data sets or “tracks”. Next, it associates enriched motifs and tracks with candidate transcription factors and determines the optimal subset of direct target genes. We validate iRegulon on ENCODE data, and use it in combination with RNA-seq and ChIP-seq data to map a p53 downstream network with new predicted co-factors and targets. iRegulon is available as a Cytoscape plugin, supporting human, mouse, and Drosophila genes, and provides access to hundreds of cancer-related TF-target subnetworks or “regulons”.
| Precise regulation of gene expression is imperative for all biological processes. Sequence-specific transcription factors (TFs) bind to their DNA recognition sites within cis-regulatory elements and thereby contribute to the control of the transcriptional initiation rate of their target genes through an interplay with other transcription factors, co-factors, chromatin modifiers, and transcription factories [1]–[3]. The human genome encodes for about 1800 sequence-specific TFs, each of which regulates hundreds of target genes [1], [4], [5]. Because TFs play key roles in gene expression, they are often considered the master regulators of cellular processes. Thus, the mapping and characterization of their regulon (all the target genes of a TF) can provide crucial insight into the biological processes they control [6], [7]. For example, in cancer, ∼40% of the driver mutations affect TFs, and many of the key oncogenes and tumor suppressors, such as p53, MYC, E2F, and NF-κB, are transcription factors [8]. Identification of the TFs that operate a perturbed gene network, and detecting their target genes, are instrumental steps in uncovering key insights into oncogenic programs, including the discovery of therapeutic targets [9]–[12]. For example, although many target genes have been described for the tumor suppressor p53 [9], [13], [14], several aspects of the gene regulatory network (GRN) downstream of p53 remain unknown. For example, it is still unclear whether p53 also directly represses target genes; whether p53 cooperatively regulates target genes with particular co-factors; and whether different target genes are regulated depending on the cancer type, or depending on the context of p53 activation. The situation is obviously worse for less studied TFs for which often none or only few target genes are known.
The targets of a known TF can be identified experimentally with relatively high accuracy through chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-Seq) [15]. However, ChIP-Seq has limitations because it is usually applied to cells in culture rather than to the actual biological sample (e.g., a tumor); and it focuses on a single TF at a time, that has to be chosen a priori. When the TF is not known in advance, or when only gene expression profiling can be performed, regulatory relationships can be uncovered by reverse-engineering a gene regulatory network starting from the expression data. One approach to solve this problem is by exploiting the fact that genes that are co-regulated by the same TF commonly share binding sites for this TF. However, detecting these short and variable TF binding sites (TFBS) within large non-coding regions represents a computational challenge when working with human or mouse genomes. Although a lot of progress has been made over the last decade and many motif discovery methods have been developed and refined (reviewed in [16]–[19]), motif discovery methods alone are not sufficient to map a gene regulatory network, nor can they be applied to noisy gene sets containing mixtures of targets of multiple TFs. This is true for both motif discovery methods relying on de novo detection and those relying on the enrichment of known position weight matrices (PWM). Additionally, many tools have a motif-oriented output, making it difficult to identify the possible upstream TF. A further limiting factor is that many methods are restricted to using human annotated PWMs (e.g. TRANSFAC [20], JASPAR [21] or UNIPROBE [22]), limiting the number of TFs that can be identified as candidate network regulators based on motif enrichment. Therefore, although cis-regulatory sequence analysis has great potential in resolving direct TF-target interactions, it has until today seen limited applications towards gene regulatory network mapping.
Finally, the recent availability of thousands of ChIP-Seq datasets, both from ENCODE [23], and other resources [24], yields new opportunities to discover master regulators from co-expressed gene sets [25], while at the same time pose challenges on how to integrate these data with motif discovery.
Here, we aim to tackle some of these challenges by increasing the performance of motif detection to yield high-confidence results, even in noisy gene sets. Motif detection is followed by the annotation of the discovered motifs with associated TFs and direct targets. To this end, we have collected more than nine thousand PWMs from various sources and from different species and link them to candidate binding TFs using a “motif2TF” procedure. This will allow the user to link hitherto anonymous motifs, and motifs of TFs from other species, to candidate human TFs. Furthermore we developed a user-friendly Cytoscape plugin [26], called iRegulon, allowing the integration of predicted cis-regulatory binding sites directly into a biological network. Finally, we extend and generalize this framework towards combined motif and track discovery on a co-expressed gene set, incorporating more than 1000 ChIP-Seq tracks. The iRegulon Cytoscape plugin is available via the Cytoscape App Store [27] and can be downloaded from http://iregulon.aertslab.org/.
The goal of iRegulon is to enable gene regulatory network mapping directly based on motif enrichment in a co-expressed gene set. As motif discovery method we have chosen to elaborate on the recent ranking-and-recovery methods [28]–[32] (Fig. 1). In the ranking step we generate whole-genome rankings of 22284 human RefSeq genes for a library of PWMs where a PWM is a matrix representation of a regulatory motif (Table 1). For each gene, a regulatory search space (500 bp, 10 kb or 20 kb around the Transcription Start Site (TSS), see Materials and Methods) is scanned for homotypic cis-regulatory modules (CRM) using a Hidden Markov Model [33] (Fig. S1). Starting from a library with N PWMs, N ranked lists of genes are generated, each with the most likely genomic targets of a particular motif at the top of the ranking [28], [29]. Next, orthologous search spaces in ten other vertebrate genomes are determined by UCSC liftover tool [34] and are subsequently scanned with the same PWMs. The rankings for different species are combined by rank aggregation [35] into one final ranking for each PWM in our library. For the PWM libraries we have collected and reformatted most of the available libraries into a “6K collection” (N = 6383 PWMs) and a “10K collection” (N = 9713 PWMs) (Table 1). These libraries contain PWMs from different species and also include candidate PWMs for unknown TFs. The results of the ranking step are N human gene rankings stored in an SQLite database. We also generated similar databases using mouse and Drosophila as reference species, in case the input gene set is derived from mouse or fruit fly.
The recovery step uses as input any set of co-expressed genes (Fig. 1B). The enrichment of these genes is determined in each of the N motif-based rankings using the Area Under the cumulative Recovery Curve (AUC), whereby the AUC is computed in the top of the ranking (default set to 3%, see Fig. S2 for validation). The AUC values are normalized into a Normalized Enrichment Score (NES) on which we set a default cutoff of 3.0, corresponding to a False Discovery Rate (FDR) between 3% and 9% (Fig. S3 and Materials and Methods). The leading edge of candidate targets is selected as the optimal subset of highly ranked genes compared to the genomic background and compared to the entire motif collection as background (Fig. 1B and Materials and Methods).
We have previously successfully applied the ranking-and-recovery method for Drosophila, namely in cisTargetX [29] and i-cisTarget [28]. These methods have been proven successful in identifying upstream regulators and direct target genes from co-expressed gene sets for Atonal [29], Shavenbaby [36], Fruitless [37], EcR [38], Dichaete [39], Glass [40], dJun/Vri [41], and Rfx [42]. Here, we apply this framework for the first time to human and mouse and we add two novelties to facilitate GRN mapping. The first is a motif2TF procedure that links an enriched motif (PWM) to a candidate binding TF (Fig. 1C and Materials and Methods). For this step we constructed a database of motif-TF direct annotations, TF-TF edges as defined by gene homology [43], [44], and motif-motif edges as defined by motif similarity (using Tomtom [45]). The database links 6031 motifs from the “10K” collection to 1191 human TFs. The advantage of this method is that it allows discovery of motif-TF links based on orthology and based on similarities between annotated and “unknown” motifs in the collection. Application of this method adds 247 more TFs to be identified than the 944 directly annotated TFs in human, and vastly increases the number of different motifs per TF (see Materials and Methods for more detailed description). The second novelty is the availability of the method as a Cytoscape [26] plugin, called iRegulon. The plugin works on any input network and returns a combination of regulators, their direct targets within the input network, and their binding motifs. A detailed description on the use of the plugin is provided in Fig. S4. This is, to our knowledge, the first method that brings cis-regulatory sequence analysis into Cytoscape. This dramatically changes the way motif discovery is performed, because instead of a list of promoter sequences used as input, now any set, network, or pathway of genes can be used as input. Instead of a list of enriched motifs, regulons, are the output, containing the candidate TFs along with their optimal direct target subsets. iRegulon results can be immediately used to map (or annotate) gene regulatory networks and be integrated with the extensive array of regulatory, expression, and annotation tools available within Cytoscape.
To evaluate the performance of iRegulon, we derived direct target gene sets for 115 sequence-specific TFs from the ENCODE ChIP-Seq data [46], and for each target set we investigate whether the ChIP'ped TF can be correctly recovered (see Materials and Methods). Out of 115 tested TFs, iRegulon correctly identifies up to 94 TFs (82.6%) with Normalized Enrichment Scores (NES) above 3 (Fig. 2A, and Materials and Methods). We found iRegulon to be robust to noisy gene sets by adding increasing levels of noise (negative genes) to each set of targets (Fig. 2B). The motif2TF step is crucial to link an enriched motif to a candidate TF; and including motifs from other species and unknown motifs allow detecting many more correct regulators compared to using only known human motifs from TRANSFAC or JASPAR (Fig. 2C). After optimizing the parameters of iRegulon and motif2TF (see Materials and Methods and Fig. S2), we compared iRegulon with eight other motif discovery methods that use a similar input (a set of co-expressed genes) and generate a similar output (candidate regulators) using a non-ambiguous subset from Factorbook [46] (Materials and Methods). iRegulon identifies the correct TF at the first position in 17/30 cases while the other tools on average detect only 5.1/30 TFs at the first position (Fig. 2D, Table S1). Interestingly, the improved performances of iRegulon are not only due to the large PWM collection and the motif2TF mapping. Indeed, iRegulon still outperforms the other methods when using only the JASPAR collection and disabling the motif2TF step (Fig. S2C) or vice versa, when manually promoting similar motifs in the other tools to the correct TF (dashed bars in Fig. 2D). As expected, the true positive target gene recovery is significantly higher when iRegulon uses a 20 kb search space around TSS compared to using only the proximal promoter (Wilcoxon rank-sum paired test, p-value = 0.004) (Fig. S2D). We conclude that the core motif discovery framework of iRegulon is better than other tools, and that the large motif collection and the motif2TF step deliver a marked step forward in TF identification performance.
In the validation and benchmark analyses above we used gene sets derived from ENCODE ChIP-Seq data as input for iRegulon. In this section, we explore more realistic types of inputs, such as co-expressed genes downstream of a TF perturbation [47]; genes involved in the same signaling pathway (e.g., KEGG [48], Reactome [49] or Gene Ontology [50]); highly connected genes in a biological network (e.g., GeneMania [51] or STRING [52]); shared targets of a common microRNA. In the first example, we applied iRegulon to a set of 171 genes that are significantly up-regulated under hypoxia [53]. iRegulon yields a top-scoring regulon that contains HIF1A as master regulator, along with 94 predicted direct target genes (Fig. S5A). The predicted HIF1A targets are likely functional targets because they overlap much more (41%) with known HIF1A targets [54] than the non predicted targets (15%). More systematically, when applied to 76 co-expressed gene sets obtained after a genetic perturbation of the TF (gene sets from MSigDB [47]), the perturbed TF is recovered in 38 cases (50%) and as the top ranked master regulator in 18 cases (24%). The lower recall to detect the correct upstream TF compared to ChIP-derived gene sets is expected because not all TF perturbation experiments successfully result in significant gene expression changes of the direct target genes.
Next, we analyzed a set of 161 genes involved in the NOTCH signaling pathway and identified the top two regulons to be controlled by HEY1/HEY2/HES1 and RBPJ, two major players involved in NOTCH signaling (Fig. S5B). We also analyzed 1198 genes involved in immune response (GO:0006955), and as expected we found the IRF and REL/NF-κB regulons, with 806 and 711 direct target genes respectively, highlighting their role as master regulators of the immune response (Fig. S5C). We also analyzed all 2233 TF-centered subnetworks within protein association networks and found enrichment of direct targets for 151 (13.2%) and 159 TFs (14.6%) for GeneMania and STRING networks, respectively, indicating that transcriptional interactions are partially represented in protein-protein interaction networks as well (Fig. S5D). Finally, we analyzed 159 sets of known microRNA targets, for which iRegulon identified significant cross-talks (feed-forward loops) between the predicted TF and microRNA regulons (Fig. S5E). While previous methods have thus far been validated and applied to co-expressed gene sets derived from gene expression profiling, here we show that motif discovery with iRegulon can quickly identify master regulons on diverse types of gene sets, as long as a small fraction of the input set is directly co-regulated by the same TF.
We now applied iRegulon to study the gene regulatory network downstream of the p53 tumor suppressor. p53 functions mainly, if not exclusively, as a TF which regulates the expression of hundreds of genes that in turn mediate its biological activities including induction of cell-cycle arrest, senescence and apoptosis [55], [56]. Although p53 is one of the most-studied transcription factor and hundreds of target genes have already been identified [14], [55], many aspects of its downstream network remain unresolved and a more comprehensive understanding of the p53 downstream signaling network is crucial given its importance in oncogenesis.
We first determined a p53-dependent gene signature in the MCF-7 human breast cancer cell line by RNA-seq upon stabilization of p53 by the non-genotoxic small molecule Nutlin-3a [57]. This treatment resulted in significant up-regulation of 801 genes and down-regulation of 790 genes. Both up- and down-regulated gene sets were subsequently analyzed with iRegulon (Fig. 3A). The top-scoring regulon in the list of up-regulated genes is confirmed as the p53 regulon, with 307 genes predicted to be direct targets (Fig. 3A and Table S2). This indicates that p53 itself is the master regulator of the downstream network and directly controls many up-regulated genes, but not all of them (at least 38%). A Gene Ontology (GO) enrichment analysis of the 307 predicted direct targets identifies p53-related processes and pathways, such as “p53 signaling pathway” (adjusted pvalue = 3.18e-21) or “Apoptosis” (adjusted p-value = 6.76e-07), while the set with the remaining 494 up-regulated genes show no significant GO term enrichment (data not shown).
In this particular experimental setup the master regulator, namely p53, was specifically perturbed and thus known a priori. Yet, even under such circumstances there are two important advantages of using a computational regulatory analysis with iRegulon. First, the explicit finding of the p53 motif as top ranked indicates that p53 directly controls a large portion of the up-regulated genes but not all, creating two clearly distinct subsets. Second, we discover potential p53 co-factors and secondary regulons downstream of p53. Particularly, among the 801 genes that are activated downstream of p53, we found three other regulons, one operated by activator protein 1 (AP-1, heterodimer composed of JUN/FOS/FOSL1/FOSL2), another by a Forkhead TF (FOX), and another by NF-Y (Fig. 3A, Table S3A). These secondary regulons show extensive overlap with the primary p53 regulon, indicating that these TFs may be important contributors in gene regulation downstream of p53 (Fig. 3B). The AP-1 regulon, sharing 136 genes (59% of its regulon) with the p53 regulon might indicate a prevalent co-factorship between the two proteins, something that has been reported before but never on such an extended scale [58], [59]. In addition, one of the shared p53-AP1 targets is GADD45A, a gene involved in DNA damage repair, that has been shown to be a bona fide target of both p53 and AP-1 [60]. Interestingly, two subcomponents of the AP-1 complex, FOS and FOSL1, are themselves up-regulated upon p53 stabilization, and are among the predicted direct p53 targets (Table S4). These results, together with the fact that the AP-1 motif was not enriched among the down-regulated genes indicate a positive, synergistic effect of the p53 and AP-1 regulons.
Nutlin-3a treatment also resulted in 790 significantly down-regulated genes. Interestingly, the analysis of this set with iRegulon does not detect the p53 motif as enriched. It does however identify E2F as master regulator with an astounding 653 (82.7%) predicted direct targets (Table S3B). Moreover, three E2F family members, namely E2F1, E2F2, and E2F8 are all strongly and significantly down-regulated upon Nutlin-3a treatment (around 10-fold down with p-value<1.0E-64), indicating the marked involvement of this protein family in the repressive mechanisms of p53. Similarly, iRegulon points towards NF-Y as an important second master regulator of a large number of down-regulated genes (493 genes). Both E2F and NF-Y have been reported as important players for p53-mediated down-regulation of genes [61], [62]. This may happen through p21 regulated cyclin dependent kinases, resulting in a lack of phosphorylation of NF-Y and Rb which ultimately renders both NF-Y and E2F (through Rb) inactive [63], [64]. Interestingly, the majority of NF-Y's predicted regulon overlaps with that of E2F, with only a very small number of genes predicted as NF-Y only targets (Fig. 3B). The enriched Gene Ontology terms of these overlapping target genes are related to cell-cycle processes, an expected result since both E2F and NF-Y have been established to regulated cell cycle-related genes, often in a cooperative manner [65]–[67]. In contrast to E2F, NF-Y itself is not down-regulated as a gene by p53 activation. However, it is possible that NF-Y is regulated at the protein level rather than at the transcriptional level in response to p53 activation. All together, these findings support the notion of an indirect rather than a direct p53 repressive process largely working through the p53-p21 axis, which affects both E2F and NF-Y [63], [68]. All together, iRegulon generates marked ideas concerning p53, which are further elaborated upon in the next section.
To test the predicted p53 regulon we determined the genome-wide chromatin occupancy by p53 in Nutlin-3a stimulated MCF-7 cells using high-coverage ChIP-Seq (∼30 Million uniquely mapped reads). Fig. 4A shows the raw ChIP-Seq data for the known p53 target CDKN1A, with a very strong peak overlapping the known p53 binding site in the promoter of CDKN1A [69]. To avoid arbitrary thresholds on peak calling we used lenient peak calling settings to rank all genes in the genome according to their likelihood of being a p53 target based on ChIP peaks only (see Materials and Methods). To assess whether this ranking yields true p53 targets on top, we curated 223 bona fide p53 targets from the literature and public databases (Table S5), and indeed found these targets to be significantly enriched in the top of this ranking (Fig. 4B, p-value = 1.40E-24). Within the same ranking, the 307 predicted p53 targets by iRegulon are nearly as significantly enriched in the top as the curated targets (p-value = 2.60E-24), while the 494 remaining up-regulated genes are not significantly correlated with the ChIP peak data (p-value = 0.096). Importantly, this result shows that iRegulon is not only able to identify the master regulator, but is also able to correctly distinguish between direct and indirect targets from a set of co-expressed genes. Only two up-regulated genes with a high ChIP peak, namely PLK3 and DDB2, were missed by iRegulon. About 100 up-regulated genes have a small ChIP peak but have not been predicted by iRegulon as target genes. These peaks are likely false positive ChIP peaks because they do not show p53 motif enrichment when analyzed separately (Fig. S6A–C). Finally, to compare how many targets are missed by iRegulon, and how many by ChIP-Seq, we again used the set of curated targets, and found comparable numbers of false negatives, namely six for iRegulon and five for ChIP-Seq (Fig. 4C). In the previous section we had also found that gene repression downstream of p53 is indirect through E2F, which has been shown recently to be mediated by p21 and RB [63], [68]. If this is true, then the down-regulated genes should not contain p53 ChIP peaks. To test this, we plotted the recovery of the 790 down-regulated genes along the p53 ChIP-peak-based gene ranking generated above (Fig. 4B). Similar to the indirect up-regulated genes, the down-regulated genes are completely depleted of p53 ChIP peaks (p-value = 1.0). On the other hand, the down-regulated genes are positively correlated with E2F1 ChIP-Seq data in MCF-7 from ENCODE (Fig. S6D). When combining all the small p53 ChIP-Seq peaks that are detected amongst the down-regulated genes, the p53 motif is not found by de novo motif discovery, while the ChIP peaks of direct up-regulated targets are strongly enriched for de novo p53 motifs (Fig. S6A–C). From the ChIP-Seq validation data, we conclude that iRegulon predicts the correct master regulators (p53 and E2F) and that predicted target genes of these TFs significantly overlap with ChIP-Seq derived targets. By combining iRegulon and ChIP-Seq data, we propose a set of 110 “top targets” of p53 in MCF-7 that are directly and positively regulated. When further comparing these predicted targets to recent reports of several p53 targetomes based on combining gene expression profiles with p53 ChIP-Seq data under different experimental conditions [58], [59], [68], we could confirm many common targets, but also uncovered 56 new direct p53 target genes with our analysis (Table S6).
To explore the relevance of the newly identified p53 targets in other tumor types, we applied iRegulon in a meta-analysis to about twenty thousand cancer gene signatures, i.e. differentially expressed genes obtained from cancer specific experiments. We reasoned that those target genes that are recurrently predicted across cancer gene signatures, might contribute to the tumor suppressor role of p53. We used gene signatures from GeneSigDB [70], MSigDB [71] and from gene modules generated across 91 large cancer microarray data sets (see Materials and Methods and Fig. 5A). Out of 23172 signatures, p53 is found as regulator in 709 signatures. We merged the direct p53 targets across all these signatures into a network and weighted the edges according to the recurrence of this p53-target interaction across all signatures. Many previously known p53 targets and many ChIP-Seq derived targets are recovered using this analysis (GSEA NES = 3.01, FDR<0.001) (Fig. S7). Of the 110 predicted p53 targets in MCF-7 cells (as defined above), 44 are also predicted as p53 target in cancer gene signatures (grey area in Fig. 5B). These genes are predicted as p53 targets by iRegulon and show a significant ChIP peak and are represented in the p53 cancer-related meta-regulon. Amongst these 44 genes, 20 were previously indicated as well established p53 targets (genes in squares in Fig. 5B). When extending the analysis and including target genes recently reported in literature [58], [59], [68], it becomes clear that most overlap coincides within this metatargetome (34/44) (Table S6). Keeping in mind that many of the p53 targets reported by others were found using different cell lines, the enriched overlap within this metatargetome can be interpreted as a sign that these genes represent a core set targeted by p53 regardless of the cell type. Interestingly, when looking at targets like RAP2B, NHLH2, SLC12A4, and ALDH3A1, they could not have been identified through motif discovery in proximal promoters only, because the p53 binding sites are located either further upstream (∼1 kb for RAP2B and ∼5 kb for ALDH3A1) or in introns (NHLH2 and SLC12A4) (Fig. 5C).
Next we confirmed experimentally whether these four targets are bona fide p53 transcriptional targets. They are all induced in a p53-dependent manner in various cellular model systems including normal diploid human fibroblasts (BJ cells) and various cancer cell lines (i.e. HCT116 and MCF-7) (Fig. 5D). Except ALDH3A1, they are also all significantly induced upon exposure to the DNA damaging agent doxorubicin, a well-established p53 inducer (adjusted p-value<0.05). Their kinetic of induction both in response to Nutlin-3a and DNA damage is comparable to the one seen with known direct p53 targets such as CDKN1A further supporting a direct role for p53 in their regulation (Fig. S8). Finally, for all except one we could confirm luciferase reporter activity of the predicted p53 enhancer region (Fig. 5E). Enhancer-reporters for ALDH3A1, NHLH2 and RAP2B show a significant induction after Nutlin-3a treatment in wild type but not in a p53 knock-down (KD) cell line (p-value<0.05). SLC12A4 does not have a significant induction in either cell-type. Note that our positive control enhancer, namely the CDKN1A promoter, is a very responsive p53 target and likely responds to low levels of p53, which could explain the induction that is still observed even under p53 KD conditions. Functionally, these validated p53 target genes have been implicated in p53-regulated processes such as the control of cell volume, growth and movement (SLC12A4 and RAP2B) and metabolism (ALDH3A1 and NHLH2).
We extended our motif discovery approach to allow the discovery of significantly enriched ChIP-Seq tracks in a set of co-expressed genes. We created a database with track-based gene rankings from a collection of 1118 ChIP-Seq experiments against 246 human sequence-specific TFs across 40 cell types and apply the same “ranking-and-recovery” enrichment calculation as employed earlier (see Materials and Methods). These and other recent resources further enlarged our motif collection to 9713 distinct PWMs (“10K collection”) (Table 1). To test whether motif and track discovery can be performed simultaneously, we combined the motif-based rankings and the track-based rankings into one enrichment analysis, although each AUC score distribution is kept separate for normalization (Fig. 6A–B). Applied to the 801 p53-dependent up-regulated gene set, the combined approach still detects p53, AP-1, NFY, and FOX in the top motifs. Both for p53 and AP-1, enriched ChIP-Seq tracks are found by the track discovery, being our in-house performed p53 ChIP-Seq in MCF-7 after Nutlin-3a (ranked first of all tracks, NES = 5.18) and the FOSL2 ChIP-Seq tracks in MCF-7 from ENCODE (NES = 3.30) (Fig. 6C–D, Table S7). In addition, we found five more candidate TFs with a putative role in the network downstream of p53 that were not detectable using the 6K motif collection only (Fig. 3). Three of these additional candidates, namely RFX5, NR2F2, and NFI have both their ChIP-seq track and motif enriched while two more candidates, namely p300 and TCF12 only show track enrichment (Fig. 6D). To our knowledge, no interaction of these TFs with p53 has been reported in the literature. Although the targetomes of the co-factors overlap to some extent (20–42%) with p53 targets, they have a considerably large set of target genes independent of p53. Hence, with these additional TFs added downstream of p53, we can once more explain an additional fraction of the up-regulated gene set, with all the ChIP-Seq track-derived interactions together regulating 542 of the 801 genes. RFX5 is of particular interest since the gene itself is strongly up-regulated by p53 and is in fact among the core set of 801 up-regulated genes (log2FC = 1.9 and adjusted p-value = 1.05E-15). RFX5 is mainly known as a regulator of MHC-II genes, and indeed, among the top predicted RFX5 target genes downstream of p53 we find HLA-F, MR1, and other genes involved in antigen and interferon-related processes. Interestingly, RFX5 has recently also been shown to act as a DNA mismatch repair stimulatory factor [72], and several p53-shared RFX5 targets, such as DDB2 and BBC3, are in fact related to DNA damage response (adjusted p-value = 6.99E-5, Wikipathway ID:WP707) (Fig. 6E). Hence, RFX5 can be considered as a new candidate co-factor to modulate certain aspects of the p53-regulated response, and may explain why MHC-II genes are up-regulated in a p53-dependent manner. This proof-of-principle of combined motif and track enrichment paves the way towards further integration of regulatory track data and enhancer prediction data to map gene regulatory networks.
We have optimized and expanded motif discovery methods and used large collections of up to 10.000 candidate motifs to facilitate translation of motif detection results into a network biology framework. By adding this network-layer on top of cis-regulatory motifs, we could generate direct insight into a biological process, rather than producing a mere list of enriched motifs from a gene set. iRegulon outperformed existing methods at detecting the correct upstream regulator. We found that using PWMs from other species than human greatly helps motif detection in human data sets. Many TFs are conserved from human to mouse, and even from human to fly or yeast, and sometimes the yeast or fly PWM is of higher quality or better captures the specificity of DNA binding. In addition, we found that using multiple PWMs for the same TF is an advantage and leads to higher performance of TF recovery compared to using non-redundant motif collections. Our motif collection also contains an important fraction of “novel” motifs for unknown TFs. These motifs are mostly derived from whole-genome computational predictions. In some cases these unknown motifs are clustered together in the output of iRegulon alongside a known motif, and can thereby lead to candidate TF predictions, while in other cases they may represent orphan motifs (with unidentified TFs). The mixture of known and unknown motifs creates a hybrid motif detection approach, combining de novo motif discovery and pattern matching approaches.
Large-scale analyses of co-expressed gene sets of different origins, including co-expression, TF binding (ChIP), protein-protein association networks and microRNA targets, suggest that by exploiting the genome sequence, together with other species' genomes and collections of consensus TF binding sites, the most relevant sub-networks that underlie observed changes in gene expression or observed genetic interactions can be reconstructed. In up to 70% of the cases, the upstream regulatory factor can be identified, along with a set of direct targets. Therefore iRegulon provides an alternative approach to probe a particular biological process when gene expression data is available but the TF is not known in advance and/or ChIP-Seq is not feasible. By combining iRegulon with RNA-Seq, the resolution of gene expression profiling and gene regulatory network mapping can be increased, allowing the characterization of any cell type, cellular response, or tumor sample, up to the single cell level.
Multiple regulons are often discovered from one co-regulated gene set. This is expected because in higher vertebrates gene regulation is combinatorial, where multiple TFs cooperate, either through binding in the same CRM (called heterotypic CRMs), or in separate CRMs of the same target gene [17]. In addition, the targets of a TF can be TFs themselves, and in turn activate or repress their own targets. For example, in the p53-dependent gene set iRegulon identified not only p53 as regulator, but also a previously known co-factor AP-1 and new regulators downstream of p53 such as RFX5. Interestingly, FOS and FOSL1, important members of the AP-1 complex, and RFX5, were all identified in this study as targets of p53. These regulators can explain a large proportion of the possible target genes of p53 as being indirect and regulated by another TF. When we extended our ranking-and-recovery framework to include more than one thousand ChIP-Seq data tracks, we also found the respective ChIP-Seq peaks for AP-1, RFX5, and several other co-factors as significantly enriched in the p53 downstream network. The joint finding of both a motif and a track for the same transcription factor strongly increases the confidence for these factors to play a role in the network as master regulator (i.e., directly controlling many target genes). Nevertheless, we envision that in most cases the motif enrichment alone, without any track enrichment, can directly lead to candidate master regulators, because ChIP-Seq data is condition-specific and is currently available for relatively few transcription factors.
The absence of a regulator in the output of iRegulon, when neither a motif nor a track is enriched, can also be informative. For instance, neither the p53 motif nor its ChIP-Seq track are found enriched among the down-regulated genes, leading to the hypothesis that p53 does not act as a direct repressor, but only as an activator. Rather, iRegulon points to E2F as the master regulator of the down-regulated genes, both by its motif and track. This finding can be explained as indirect down-regulation of E2F targets and has recently been experimentally established: p21 controls RB1-mediated repression of E2F targets, including E2F family members themselves, thereby reinforcing this signal further [63], [68].
Our experimental findings on the p53 regulon were obtained in MCF-7 breast cancer cells. Usually, one iRegulon analysis is focused on one biological process, and predicts transcriptional targets that are relevant in that particular cell type or condition under study. We show that it is also possible to apply iRegulon more systematically on multiple signatures to identify cancer-related ‘meta-regulons’. They often represent the canonical, high-confidence target genes and agree well with ENCODE ChIP-Seq data (Fig. S7). This shows that relevant TF-target interactions can be identified purely from the genome sequence, thereby creating a valuable resource for less studied TFs.
Three predefined regulatory search spaces are used in this manuscript from small to large regions: 500 bp upstream of TSS [TSS−500 bp,TSS]; 10 kb around TSS [TSS−5 kb,TSS+5 kb]; 20 kb around TSS [TSS−10 kb,TSS+10 kb]. If another gene is located within the upstream region, then the region is cut where this neighboring gene begins or ends (depending on which strand this gene is located on). Coding exons are excluded from the search space to avoid bias towards these exons through conservation. Notice that there can be multiple regions per gene (various upstream regions for alternative transcripts, and multiple introns) (see example in Fig. S1). When multiple regions are scored for a given gene, the rank of the highest ranked region is taken into account as the final rank of the gene.
Motif detection relies on an offline scoring step whereby every gene in the human genome, along with orthologous sequences in ten other vertebrate genomes, is scanned with Cluster-Buster [33] for homotypic clusters of motifs using a library of N position weight matrices (PWMs), generating a database of N ranked lists of genes, each with the most likely genomic targets of a motif at the top of the ranking.
As in the case of motif detection, TF ChIP-Seq track detection also relies on an offline scoring step whereby every gene in the human genome is scored with M sets of ChIP-Seq peaks (broad or narrow), generating a database of M ranked lists of genes, each with the most likely genomic targets of a TF at the top of the ranking.
Our motif enrichment analysis differs from standard gene set enrichment methods such as GSEA, which uses Kolmogorov-Smirnov statistics [71]. In our method, we calculate the top enrichment of a single gene set over Nmotif genomic rankings while gene set enrichment methods assess the significance of many gene sets for one genomic ranking. Enrichment is determined by the Area Under the Recovery Curve (AUC) of the cumulative recovery curve for the input set, along the whole-genome ranking. As we are mostly interested in highly ranked genes, the AUC is computed in the top of the ranking (default set to 3%, see Fig. S2B for validation) for all PWMs or tracks of the collection. A Normalized Enrichment Score (NES) for a given motif/track is computed as the AUC value of the motif/track minus the mean of all AUCs for all motifs (or tracks), and divided by the standard deviation of all AUCs. When the distribution of AUCs follows a normal distribution then the NES score is a z-score indicative of the significance. The default NES cutoff in iRegulon is 3.0, corresponding to FDR between 3% and 9% (Fig. S3).
For each enriched motif, the candidate targets are selected as the optimal subset of highly ranked genes compared to the genomic background and to the entire motif collection as background. This step is illustrated in Fig. 1B. The target gene recovery is plotted along the whole-genome ranking for a given motif (blue curve) and compared to the average recovery + (2× standard deviation) (red curve) for all motifs in the collection. Similarly to the GSEA approach [71], the leading edge corresponds to the rank where the difference between the signal (blue curve) and the background (red curve) is maximal within the top ranked genes (the latter is defined by the Rank Threshold parameter). The input genes that have a better ranking than the rank at the leading edge are predicted as target genes for the given motif or track.
Enriched motifs are linked to candidate TFs, which could potentially bind to the motif. If we use only the direct annotations, only a small fraction of motifs (20%) can be associated to human TFs (521 TFs with “6K” collection, 944 TFs with “10K” collection). We developed a database, which we term the motif2TF database, corresponding to a network of associations between motifs and TFs where motif-TF edges correspond to all motif-TF direct annotations (from different species), TF-TF edges are defined by homology (using Ensembl Gene Trees [43], [44]), and motif-motif edges are defined by motif similarity, defined by the Tomtom p-value [45]. For each motif all possible TFs are associated following different paths in the motif2TF network. In the plugin at the client side, these TFs are ranked, prioritizing directly annotated TFs, then the TF present in the input set, then the ones that are found by gene homology and finally the TFs found using motif similarity. Figure 1C illustrates the different possible paths on a motif2TF subnetwork. Motif M1 is not directly annotated to any TF (so it can be part of the unknown motif collections), but is similar to two other motifs, namely M3 and M4, both of which are directly annotated. Motif M4 is directly annotated to a human TF (TF1), while M3 is a motif annotated for a TF from another species (TF7). Three TFs in human (TF1, TF8, TF6) are possible orthologs of TF7. In this example, the link between M1 and TF1 would go via the path through M4, which is the shortest and best path (rather than via M3 and TF7). For M1, motif2TF returns TF1, TF6, and TF8 as candidate TFs, which are subsequently ranked. The second example is for motif M2 which is annotated for TF5 in another species. Three human transcription factors (TF2, TF3, TF4) are possible orthologs of TF5, which may represent for example a family of homologous TFs such as GATA factors, E2F factors, or ETS factors. In such a TF family, the consensus motif may indeed be shared by multiple family members and therefore iRegulon may return multiple or all family members as candidates. When multiple TFs are returned, we give priority to a TF when it is part of the input genes. In this example, TF2 will be preferentially associated to M2 as it belongs to the input genes (encoded by TG5 in the Figure).
ChIP-Seq data was downloaded as hg19 aligned bed files (view = peaks) from the TFBS ENCODE collections available from the following servers: http://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeSydhTfbs/ http://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeHaibTfbs/ http://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeUchicagoTfbs/. Almost one thousand files (999) were downloaded corresponding to 160 sequence-specific TFs (TFSS): 672 files for HAIB (Hudson Alpha Institute), 323 for SYDH (Stanford/Yale/USC/Harvard) and 6 files for Uchicago. Files corresponding to Input and RNA Polymerase 2 (“Pol2”/“Pol2(phosphoS2)”) were not downloaded. 115 TFs are detectable in iRegulon (i.e., at least one motif in the collection of 6383 motifs can be connected to the TF), corresponding to 786 ENCODE datasets. Each query set consists of the top 200 target genes presenting a ChIP peak in a predefined search space, i.e., for each search space tested (500 bp upstream of TSS; 10 kb around TSS; 20 kb around TSS), we define a different set of target genes, so that each target gene contains a ChIP peak within the chosen motif search space. The ChIP-Seq scoring of the genes has been done as mentioned earlier in the Track-based rankings section. Finally, note that our motif collection does not contain PWMs derived from these datasets (so we rely on other, previously curated PWMs to identify the correct TF). The Factorbook dataset collection is a subset of this ENCODE selection corresponding to 254 ChIP-Seq files (121 from HAIB, 129 from SYDH and 5 from Uchicago), inferred from the list of signatures published in the Table S1 of the FactorBook reference publication [46]. 126 out of these 254 FactorBook signatures have the canonical motif corresponding to the ChIP'ped TF. From these we randomly selected one signature per TF for which the canonical motif was predicted as top 1 by their motif discovery pipeline (inferred from Table S1A [46]). The list of the 30 used datasets is presented in Table S3. Different types of control gene sets were selected, namely: from ENCODE ChIP-Seq we used (1) genes without a ChIP-Seq peak in the corresponding search space; (2) TF neighborhoods for 1150 TFs, containing for each TF all the genes within 5 Mb flanking the TF; and (3) 1161 random signatures. Datasets are available on our laboratory website (http://www.aertslab.org). We also got similar performances using 631 uniformly reprocessed ChIP-Seq data generated in NarrowPeak format by the ENCODE Analysis Working Group downloaded from http://genome.ucsc.edu/cgi-bin/hgFileUi?db=hg19&g=wgEncodeAwgTfbsUniform (data not shown).
The classical motif discovery algorithms that originated in the late 1990s can be put in two categories: string-based or enumeration methods and matrix-based approaches. The string-based approaches rely on the detection of statistically over-represented words (oligonucleotides or spaced motifs) compared to a given background [83]–[88]. Matrix-based approaches make use of position weight matrices (PWMs) as a predictive model for TF binding sites, which can be graphically represented as a motif logo [89], and use optimization algorithms (Expectation-Maximization [90], greedy algorithm [91], [92] or Gibbs sampling [93]–[95]) to find the most common motifs to all input sequences. Most of these methods performed well on yeast or bacterial promoter sequences, but they showed limited performances when applied to mouse or human [96]. These methods could be improved by phylogenetic footprinting [97]–[103] and by applying genome-scale methods that exploit the entire gene expression data set rather than a set of co-expressed genes [104]–[106]. Current developments have on the one hand focused on the application of the early algorithms to ChIP-Seq data [107]–[111], and on the other hand on the application of motif discovery to gene sets, with the aim to increase the performance in higher eukaryotes such as fly, mouse and human, using large sequence search spaces. This category of PWM enrichment methods is represented by phylCRM/Lever [30], DIRE [80], [112], PASTAA [32], [113], PSCAN [114], Allegro [115], HOMER [116], OPOSSUM [117] and i-cisTarget [28]. They all use libraries of candidate PWMs and apply PWM enrichment statistics, often combined with other cues, such as comparative genomics and TF binding site clustering. By using libraries of PWMs for known TFs (e.g., PWMs derived from protein binding microarrays), these methods promote a TF to a candidate master regulator of the gene set when its PWM is found enriched. We used all methods in this category of PWM enrichment methods that are available online, that can work on human gene sets, and that can be practically performed on 30 sets of 200 human genes.
Thirty gene sets from FactorBook were selected for motif discovery tool comparison (Fig. 2D, Table S1). These gene sets have been selected because the motif of the ChIP'ped TF was detected as top enriched motif in the top 500 peaks in FactorBook. We extracted the top 200 genes having the highest peaks in their 20 kb region around the TSS. The comparison was performed on TF and motif recovery using the parameters indicated in Table S3. The parameters were left to default and when possible, we only adjusted the parameters to allow for larger upstream regions (when possible we choose TSS+−10 kb). iRegulon was compared to eight other publicly available motif enrichment tools, namely OPOSSUM [117], DIRE [80], [112], PASTAA [32], [113], PSCAN [114], Clover [16], AME [118], Allegro [115] and HOMER2 [116] (in the case of Homer2, de novo and known motif discovery are performed simultaneously but we consider them as different approaches and validate them separately). We selected these tools because they mostly take as input a set of human co-expressed genes, and they all return, at least to some extent, information on which TF could be regulating the input genes. For this reason, it not feasible to compare iRegulon with classical de novo motif discovery methods (e.g., MEME-like methods) because such methods are intractable on large human gene sets (e.g., 200 genes×20 kb×10 species represents a sequence set of 40 Mb), and they result in new motifs rather than candidate TFs. We also attempted to use SMART [119] but we did not succeed in running the software. For tools that require regulatory sequences as input (AME and Clover) we used the same sequences as used by iRegulon. For some tools like Clover, it is theoretically possible to use a large search space but one run on one dataset takes too long (∼17 hours), and therefore we limited the analysis to 500 bp promoter sequences. In the case of AME, we found no positive results with a large search space (data not shown), so we show the results with the default search space. For comparison, we used the number of motifs/TFs found in top 1 and within top 5 positions. The total number of detected motifs was not reported for comparison, because some tools use more stringent thresholds than others. All these tools rely on the available motif annotation to identify the candidate TF such as Jaspar (J) or Transfac (T). However, we also manually re-associated the detected motifs to candidate TFs (mainly by comparison of the detected motif with the FactorBook motif) (see column “USING SIMILARITY” in the Table S3). For Homer2, 14 motifs that are derived from ENCODE ChIP-Seq data matching the actual Factorbook ChIP-Seq data were discarded from their in-house PWM collection to avoid over-fitting (indeed, iRegulon does not include FactorBook PWMs either, nor do any of the other tools). Note that for the other large-scale analysis (e.g. full ENCODE analysis), we use a command-line version of iRegulon.
At the client side, iRegulon is implemented in JAVA as a Cytoscape plugin, which can be downloaded from http://iregulon.aertslab.org. The iRegulon plugin is connected to the server-side daemon over the Internet. The iRegulon server-side daemon is implemented in Python and uses MySQL to store and query the PWM-based whole-genome rankings (see below). After submitting a gene set or network to the service, the results are returned to the client, and this happens on-the-fly, and takes about one minute. The user can browse through the motif discovery results, select a TF among the prioritized list of TFs, and add upstream regulators and direct regulator-target ‘edges’ to the input gene set or network under study. A detailed description on the use of the plugin is provided in Figure S4. In addition, the plugin allows querying cisTargetDB to obtain the meta-regulon for a given TF, i.e. targets found recurrently predicted for this TF by iRegulon across thousands of signatures/gene sets. iRegulon results were obtained by running the Cytoscape plugin v0.97 on Cytoscape 2.8.1. The current version of iRegulon (1.2) supports the “10K” motif collection and the track discovery. The source code of the iRegulon plugin is also available from the iRegulon website (http://iregulon.aertslab.org).
iRegulon was applied in batch (i.e., using the GMT file format as input for the command line version of iRegulon) to 3447 signatures in GeneSigDB (version 4), 6753 signatures from MSigDB (version v3 collection 2) and 12972 bi-clusters we obtained in-house. Bi-clustering was performed with Ganesh clustering algorithm [120], [121] using default settings to 91 microarray datasets. The 91 datasets were retrieved as normalized (fRMA) microarray data from InsilicoDB [122]. iRegulon results on all these gene sets is stored in a MySQL database, from which all summaries per motif and subsequently per TF are computed, resulting in a meta-regulon per TF. In a meta-regulon, each target gene is annotated with a number that represents the number of gene sets where the TF is found enriched and the gene is among the optimal subset of direct targets.
GO enrichment analysis was performed using DAVID [123], [124] or BINGO [125]. GSEA analysis on ChIP-Seq data was performed to avoid arbitrary peak score cutoffs. The genome was ranked according to the MACS ChIP-peak score (score range between 0 and 1517.33 for p53) within an area of 20 kb around the TSS of 22284 RefSeq genes. Functional categories found enriched for co-factors of p53 were calculated by DAVID and WebGestalt [126] based on Gene Ontology and KEGG pathways.
Cells were kept in culture at 37°C, with 5% CO2 and in RPMI medium (+ L-glutamate, Gibco) supplemented with 10% fetal bovine serum (Invitrogen), 0.4 mM sodium pyruvate (Gibco), 100 µm/ml penicillin/streptomycin (Invitrogen), 1× non-essential aminoacids (Gibco) and 10 µg/ml Insulin (Sigma).
p53-Wild-Type MCF-7 cells were plated onto 24-well plates (60000 cells/well). The next day, cells were either stimulated with 5 µM Nutlin-3a or left untreated. After 24 h, cells were washed in PBS (Gibco) and prepared for RNA extraction according to the RNeasy protocol (Qiagen), yielding around 2 µg of total RNA per sample. The quality of the RNA samples were checked using a Bioanalyzer 1000 DNA chip (Agilent) after which libraries were constructed according to the Illumina TruSeqTM RNA Sample preparation guide. Final libraries were pooled and sequenced on the HISeq 2000 (Illumina), generating approximately 30 million reads of 50 bp length. After removing adapter sequences reads were mapped to the human reference genome (hg19) using TopHat v1.3.3 [127] with default settings. Reads were aggregated with HT-Seq (–str = no parameter, version 0.5.3p3) using the human RefSeq annotation, release 42. DESeq [128] was used to normalize and to calculate differential expression between Nutlin-3a stimulated and non-stimulated samples. A final list of differentially expressed genes was obtained using adjusted p-value<0.05 and |log2FC|>1. The threshold of 2-fold up-regulation was supported by the observation that the strongest enrichment of the targets from the KEGG p53 signaling pathway is observed among the top 648 up-regulated genes (GSEA leading edge corresponds to log2FC = 1.182).
p53 wild-type MCF-7 cells were seeded at a density of 5 million cells per 15 cm dish and grown ON at 37°C to 80–90% confluency. Cells were then stimulated with 5 µM Nutlin-3a for 24 h. ChIP samples were prepared following the Magna ChIP-SeqTM preparation kit using the p53 antibody (DO-1, SCBT). Per sample, 5–10 ng of precipitated DNA was used to perform library preparation according to the Illumina TruSeqTM DNA Sample preparation guide. In brief, the immunoprecipitated DNA was end-repaired, A-tailed, and ligated to diluted sequencing adapters (dilution of 1/100). After PCR amplification with 15–18 cycles and gel size selection of 200–300 bp fragments, the libraries were sequenced using the HiSeq 2000 (Illumina). Cleaned reads were mapped to the human reference genome hg19 (UCSC) using bowtie (v2.0.0-beta3) with the addition of parameter –local, allowing for further soft clipping of the reads. Reads with a mapping quality below 4 were removed. Peak calling was performed using MACS (version 1.4.2) [129] either with the default p-value threshold (3634 peaks) of 1.0E-5 or using p-value<0.05 (lenient setting to generate the whole-genome ranking).
MCF-7, HCT116 (human colon carcinoma cell line) and BJ cells were treated continuously with 10 µM Nutlin-3a or a pulse of 5 µM Doxorubicin and total RNA was harvested at different time points. Reverse transcription was performed using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Real Time quantitative PCR reactions were run on LightCycler480 (Roche) in 384-well format, using SYBR-Green Fast Universal PCR Master Mix (Applied Biosystems). Multiple primer pairs were tested for each target, and melting curve analysis confirmed amplification of a single product. Normalization was done with the most stable reference genes, assessed by GeNorm analysis [130]. The normalized relative fold changes were log-transformed before performing two-sided t-test to determine significance levels. The p-values were further corrected for multiple testing by very stringent Bonferroni correction. RT-qPCR primer sequences: NHLH2-fw-CACTGTGGGAGGATCTGAGC; NHLH2-rev-ATAAAGGGGCACTTCGCCTG; ALDH3A1-fw-CTGCAGGGAACTCAGTGGTC; ALDH3A1-rev-GGTACAGATCCTTGTCCAGGT; SLC12A4-fw-GGGAACAACATTCGCAGCAG; SLC12A4-rev-AGTGGCATTCGACGTGTCAT; RAP2B-fw-GCGCACAAAAGCCAAACGC; RAP2B-rev-AGACACCCTGGCCAATGCAA.
MCF-7 cells (WT or p53-KD) were seeded in a 24 well plate at a density of 50 000 cells per well. After 24 h, cells were transfected using Fugene HD (Promega) in a 1∶3 ratio. 400 ng of luciferase reporter plasmid containing one of the enhancers of interest (CDKN1A, RAP2B, ALDH3A1, SLC12A4 and NHLH2) was mixed with a β-gal plasmid in a 1∶10 ratio to correct for transfection efficiency. The next day, cells were stimulated with 5 µM Nutlin-3a. After 24 h, the transfected cells were harvested and luciferase and β-galactosidase activities were measured following the manufacturer's instructions (Applied Biosystems). The p-values were calculated using a t-test.
The RNA-Seq and ChIP-Seq data are available from the NCBI GEO database (GSE47043).
|
10.1371/journal.pgen.1000703 | Dissection of the Complex Phenotype in Cuticular Mutants of Arabidopsis Reveals a Role of SERRATE as a Mediator | Mutations in LACERATA (LCR), FIDDLEHEAD (FDH), and BODYGUARD (BDG) cause a complex developmental syndrome that is consistent with an important role for these Arabidopsis genes in cuticle biogenesis. The genesis of their pleiotropic phenotypes is, however, poorly understood. We provide evidence that neither distorted depositions of cutin, nor deficiencies in the chemical composition of cuticular lipids, account for these features, instead suggesting that the mutants alleviate the functional disorder of the cuticle by reinforcing their defenses. To better understand how plants adapt to these mutations, we performed a genome-wide gene expression analysis. We found that apparent compensatory transcriptional responses in these mutants involve the induction of wax, cutin, cell wall, and defense genes. To gain greater insight into the mechanism by which cuticular mutations trigger this response in the plants, we performed an overlap meta-analysis, which is termed MASTA (MicroArray overlap Search Tool and Analysis), of differentially expressed genes. This suggested that different cell integrity pathways are recruited in cesA cellulose synthase and cuticular mutants. Using MASTA for an in silico suppressor/enhancer screen, we identified SERRATE (SE), which encodes a protein of RNA–processing multi-protein complexes, as a likely enhancer. In confirmation of this notion, the se lcr and se bdg double mutants eradicate severe leaf deformations as well as the organ fusions that are typical of lcr and bdg and other cuticular mutants. Also, lcr does not confer resistance to Botrytis cinerea in a se mutant background. We propose that there is a role for SERRATE-mediated RNA signaling in the cuticle integrity pathway.
| As the skin of a plant, the epidermis mediates a broad set of protective functions which includes defense against abiotic environmental stresses and pathogens. The majority of its barrier capacity is localized to the outermost cell wall, which is covered by a waxy cuticle. Several distinct cuticular mutants in the model plant Arabidopsis produce a remarkable syndrome that is characterized by ectopic cell adhesion and changes in plant morphology. We used these mutants to study the constitution of the cuticle and the activation of the molecular compensatory mechanisms that are important for adaptation. We examined whole-genome responses in these mutants and used an appropriate statistical procedure to reveal the genes which change their expression. We then applied the same approach to the analysis of hundreds of datasets in repositories. The comparison of gene expression profiles identified the gene SERRATE, which encodes a protein of RNA–processing multi-protein complexes, and further analysis revealed that the syndrome is suppressed in double mutants, as predicted. Our finding suggests that the mechanism which operates to control the integrity of the cuticle involves the regulation of small–RNA signaling.
| The ability to maintain the barrier properties of the epidermis, which covers the aerial surface of higher plants, is largely due to their outermost cell walls which are impregnated and covered with specialized lipids. The fine structure and composition of this complex layer, called the cuticle, has been the subject of numerous studies [1],[2].
The innermost periclinal layer of the cuticle is, in fact, a cutinized portion of the epidermal cell wall, in which cell wall polysaccharides are perhaps cross-linked to phenolics and aliphatic components of the cutin. The presence of phenolics, which may contribute to the barrier function, is evident in this layer through fluorescence microscopy and chemical analysis [3]. Under transmission electron microscopy, however, this cutinized layer of the cell wall is often heterogeneous in appearance and penetrated by tufts of fibrillar material, or is sometimes barely visible. This is in contrast to the opaque stripe of the continuous cuticle proper which, at a higher resolution, often appears to be finely lamellated; it is composed of polyester cutin and non-hydrolyzable polymer cutan and is, essentially, free of cell wall polysaccharides. In the early stages, in particular, it may have a pectinaceous under-layer. Wax forms the outermost structural layer, although a certain amount of it (intracuticular wax) permeates the interior of the cuticle. Individual plant species display some variations of this standard pattern.
Waxes and cutin represent two groups of cuticular lipids that are supposed to be primarily responsible for the barrier function of the plant epidermis. The structural, biochemical, biophysical and molecular genetic aspects of cuticular lipids and their role in the defense against pathogens have been reviewed elsewhere [2], [4]–[10].
Plant growth and development demand that the cuticle changes continuously and also maintains the balance between rigidity and flexibility. During the early stages of epidermal development, cells are covered with osmiophilic amorphous procuticle, but the lamellar structure of the cuticle proper and the reticulate fibrillar pattern of the cutinized portion of the cell wall may become distinguishable as the cuticle forms [2]. Chemical and structural changes in the cuticle may be beneficial, both in terms of the adaptation to fluctuating environmental conditions and in response to various stresses. However, this view of the cuticle as a dynamic structure has yet to be supported by extensive data.
Recent molecular, genetic studies of cuticular mutants have lead to the identification and characterization of a number of genes involved in various aspects of cuticle formation (reviewed in [5],[6],[9],[11]).
A group of Arabidopsis mutants, including fiddlehead (fdh), lacerata (lcr) and bodyguard (bdg), which are thought to be defective in the biosynthesis of cuticular polyesters, reveals secondary phenotypes which include drastic changes in cell differentiation, plant architecture, organ morphology, pathogen resistance and other elements. This suggests that there exists a pathway that is not only essential for cuticle formation, but may, directly or indirectly, control various cellular processes as well [9]. One of these is the adhesive interaction between the epidermal cells of different organs [12],[13]. No mechanism has yet been demonstrated which might account for the association between the series of phenotypes in the cuticular mutants.
Studies of one of these mutants, bdg, which exhibits defects that are characteristic of the loss of cuticle structure, paradoxically revealed that this cuticular mutation accumulates significantly more cutin monomers in the residual cell-wall bound lipids, and more wax [14]. It was, therefore, suggested that plants are capable of repairing cuticular perturbations and re-establishing cuticle homeostasis [14]. Fdh and lcr, like bdg, belong to a class of cuticular mutants that are characterized by secondary phenotypes which include misshapen cells and organs and epidermal fusions [14]–[16]. Their cuticular phenotypes were not, however, examined in detail.
Herein, we report that both lcr and fdh display an increase in cutin and wax constituents and provide insight into the structural aspects of their cuticle. Using a microarray-based transcriptome analysis, we demonstrate the transcriptional upregulation of wax, cutin, and cell wall and defense genes in lcr, bdg and fdh. We propose that this is a compensatory adaptive response, representing a part of a cell-wall integrity maintenance mechanism. To compare the responses induced by mutations in cuticular and cellulose synthase genes, we used the meta-analytical method MASTA (MicroArray overlap Search Tool and Analysis), which has recently been developed in our lab. When utilizing it for in silico suppressor/enhancer screens, we identified SERRATE and confirmed that it is required for organ fusions in cuticular mutants. This raises the interesting possibility that there may be a connection between the cuticle formation and morphogenesis.
As do other cuticular mutations of this kind, lcr has a pleiotropic effect on plant development, which affects leaf morphology, cell morphogenesis and differentiation, shoot branching, and senescence [16]. At the rosette stage, lcr plants are easily distinguishable from wild types, but not from bdg and fdh (Figure 1A), by severe deformations of leaves and leaf fusions. When compared to wild type, the staining of rosette leaves with the water-soluble dye toluidine blue (TB) resulted in fdh, lcr, and bdg having heterogeneous, patchy patterns (Figure 1B), showing the defects of the cuticle. However, with regard to the intensity of staining, neither the fused or unfused rosette leaves of fdh, lcr and bdg were distinguishable from each other. As direct estimation of the cuticle permeability is not feasible in Arabidopsis, to extend these results, we performed an assay which measures chlorophyll leaching into alcohol [13]. As expected, the leaves of all three mutants lost chlorophyll faster than wild type when immersed in 80% ethanol (Figure 1C), thus corroborating the results of the TB staining. However, whereas lcr and bdg appeared to be very similar to each other, fdh released chlorophyll much more quickly (Figure 1C): after 20 min of incubation, fdh lost about 60% of total chlorophyll, while lcr and bdg only lost about 20%.
To study whether there is both a correlation between cuticle permeability to chlorophyll and engagement in ectopic organ fusions, in this assay we examined a sample taken from lcr rosette leaves which were not joined in a fusion. Figure 1C shows that these leaves lose the pigment faster than the wild type control does, but this was still slower than the representative lcr sample (comprising leaves joined in a fusion and leaves not joined in a fusion). This suggests that both features of the polymorphic lcr phenotype are linked.
To investigate whether the expression of LCR is restricted to the epidermis, we fused the putative 5′ regulatory regions of the LCR gene with the green fluorescent protein (GFP) reporter gene. The expression of LCR:GFP was then studied in transgenic Arabidopsis plants by confocal scanning laser microscopy, and was found to be limited to the epidermal cells of leaves, stems, sepals, petals, style, stigma and ovules (Figure 1D–1G). The expression of LCR in organ primordia was reminiscent of that of FDH and BDG, which have previously been studied in detail by using GFP fusions [14],[17]. Because LCR belongs to the CYP86A P450 gene subfamily, which includes closely related and highly conserved gene sequences in Arabidopsis, we were not able to design an LCR specific probe which would be long enough. We attempted to concatenate LCR specific sequence motifs, but the resulting probes also failed to yield a consistent in situ hybridization signal (data not shown). However, our results with the LCR:GFP plants support the microarray hybridization data which had suggested that LCR might be the epidermis specific gene in the stem [18]. Collectively, these results back-up the contention that lcr is a typical cuticular gene.
To determine whether the lcr mutation distorts or disrupts the cuticle, we examined the epidermis of aerial organs in lcr by transmission electron microscopy (TEM). In leaves and petioles of wild type plants, cutin deposition in the epidermal cell wall forms a regular membranous structure (called the cuticle proper) on the outer side (Figure 2A, 2C, 2G, 2L). When viewed under TEM, this electron-dense layer was not only discontinuous and deformed in lcr, but was also characterized by the irregular deposition of multi-layered, electron-dense, sharply outlined material, as well as the presence of empty spaces within the deeper layers of the cell wall (Figure 2D, 2E, 2H–2K, 2M–2O). The presence of empty cavities (Figure 2H and 2N) and the over-deposition of an electron-dense material close to the cell wall surface (Figure 2D, 2H, 2K), indicate infiltration, or bursting, of the cell wall materials through the defective cuticle proper in this mutant (Figure 2O), thereby leading to cutin juxtaposition in the inner layers (Figure 2H, 2M, 2N). Supernumerary layers of cutin-like materials may lead to the conclusion that lcr does not suffer from a lack of cutin, but rather from the structural dysfunction of its cuticle, even though a cutin overlay was very thin in many instances (Figure 2D, 2K, 2N). In some cases, electron-opaque material seemed to crystallize inside the lcr cell wall, giving its cuticle a composite appearance (Figure 2E, 2K, 2O). Although many features make the lcr cuticle resemble that in bdg [14] and the CUTE plants [19], this has not yet been observed in any other mutant, and appears to be characteristic of lcr, which exhibited an extraordinarily irregular cuticle.
In various fusion zones in lcr, the cell walls often seemed to be merged, with little or no trace of the intrinsic cuticular membrane, although some inclusions of electron-dense material could be found in areas where cell walls have not been completely fused (Figure S1A, S1B). When exposed to mechanical tension, fused organs might separate, only remaining connected by a fine thread (Figure S1C).
The outer cell wall of epidermal cells in lcr was, generally, not as regular as in wild type plants, with severe deformations and some darker stripes giving it a plastic appearance (Figure 2H–2J). This leads to speculation that lcr has a cell wall phenotype.
From the results of the chlorophyll leaching measurements, it might have been expected that fdh would display a highly disorganized cuticle, resembling that in lcr. However, Lolle and co-workers reported that the cuticle could always be detected in the fusion zones separating different organs, and three lipid stains failed to detect any differences between fdh-1 and wild-type tissues [12],[13]. The fdh-1 allele has been isolated in the Landsberg erecta (Ler) genetic background. Since the lcr transposon insertion allele used in this study was found in the Columbia (Col) ecotype, we sought to examine the cuticle in an fdh allele with the same genetic background. We made the decision to characterize fdh-3940S1 [20], which has also been used in the chlorophyll leaching assays described above.
The extensive investigation under TEM did not reveal any visible ultrastructural changes in the fdh-3940S1 cuticle (hereafter referred to as fdh) in different organs when compared to wild-type. This makes fdh different to the lcr, bdg and hth mutants and the CUTE plants [19]. A typical, continuous electron-dense layer was found to be deposited in the epidermal cell wall in fdh leaves (Figure 2B, 2F) and in the fusion zones (Figure S1), showing that two knock-out alleles of fdh behave similarly in the Ler and Col genetic backgrounds. While the cuticle proper in fdh showed no major ultrastructural changes that were detectable by TEM, some images gave the impression that its surface may have a rather more diffused appearance, lacking sharp margins.
These results imply that structural defects in the cuticle proper, which are detectable under TEM, may not account for the molecular sieving properties of the cuticle that were estimated by molecular leaching assays. This also suggests that the organ fusions observed in fdh, lcr, bdg and some other cuticular mutants are not a direct consequence of major structural changes in their cuticles, thereby calling into question the conventional view that the bare cell walls of epidermal cells interact to produce a fusion.
One of the cuticle's primary roles is to act as a protective barrier against pathogen attack [10]. It has, however, been reported that some Arabidopsis cuticular mutants, such as lacs2, lcr, bdg and transgenic CUTE plants, are, in fact, more resistant to the major necrotroph Botrytis cinerea than the wild type [21]–[23]. The reasons for this paradoxical resistance remained unclear. The easier diffusion of a plant-produced toxin through the mutant cuticles was, however, considered to be one of the possible mechanisms of this resistance [21], suggesting that the highly permeable cuticle would greatly increase the resistance of fdh to this pathogen. To test this concept, we compared the wild type and three mutants by the detached-leaf assay which appeared to have the best consistency [24]. As controls, we used the B. cinerea-resistant lacs2 and the hypersusceptible phytoalexin deficient3 (pad3) which is impaired in the accumulation of the antifungal metabolite camalexin [25]. The level of susceptibility after droplet inoculation was measured as the percentage of lesions larger than the original inoculation site (percentage of outgrowing lesions) and as the average lesion area (Figure 3). Three days post-inoculation (dpi) with B. cinerea (strain 2100 in 1/4 PDB), lacs2 and bdg showed lower susceptibility (P<0.0001), with 16% and 18% of outgrowing lesions, respectively, as compared to 72% in the wild type, and 37% and 42% in lcr and fdh, respectively (Figure 3B). Consistent with the described hypersusceptibility to B. cinerea, 96% of outgrowing lesions were identified in pad3 (Figure 3B). The higher resistance phenotype of cuticular mutants was revealed by both the lower proportion of outgrowing lesions and lesion area, the latter being on average very similar in all cuticular mutants and always significantly smaller when compared to the wild type (Figure 3C). One interesting observation is that, under our experimental conditions, some leaves of bdg and lacs2 remained free of disease symptoms at 3 dpi (∼27% and ∼56% respectively), whereas all inoculated leaves of lcr and fdh showed signs of fungal infection. Comparable results were obtained with the B. cinerea strain B05.10 (data not shown), corroborating the view that FDH deficiency afforded similar protection against B. cinerea as that observed in other cuticular mutants. However fdh does not seem to be more resistant than bdg and lcr, implying that the resistance phenotype does not correlate well with the cuticular permeability when measured by the chlorophyll leaching rate and the TB stainability.
Although lcr and fdh were recognized as classical cuticular mutants revealing enhanced epidermal permeability (Figure 1B and 1C) [13],[16],[21],[26], the chemical compositions of their cuticular lipids had not been characterized in detail. LCR, which functions as a fatty acid ω-hydroxylase when expressed in yeast, was proposed as being active in the biosynthesis of the ω-hydroxy and α,ω-dicarboxy fatty acids that are major cutin monomers in Arabidopsis [16],[27]. The loss of the LCR function is presumed to reduce the accumulation of the respective cutin monomers.
Based on the amino acid sequence similarity, it has been suggested that FDH encodes a β-ketoacyl-CoA synthase (KCS) which is involved in microsomal fatty acid elongation [20],[28]. In FDH-deficient plants, cutin monomers could, therefore, potentially either decrease or comprise shorter chain monocarboxylic and dicarboxylic fatty acids.
To determine the effects of fdh and lcr mutations on the chemical composition of cuticular polyesters, we analyzed residual-bound lipids in leaves (Figure 4A). This approach gives a good approximation of the monomer composition of pure cutin, which is difficult to isolate in sufficient amounts in Arabidopsis [29]. This analysis was conducted twice on both mutants, using plant material grown under similar greenhouse conditions in different seasons (Figure 4A and Figure S2). Both experiments yielded similar results, with lcr and fdh accumulating higher levels of the C18:2 α,ω-diacid which is a major cutin component in Arabidopsis [29],[30]. The increase, when compared to wild type, was approximately three times and twice in lcr and fdh, respectively (Figure 4A). Remarkably, the levels of C18:2 ω-hydroxy acid, which is a precursor to C18:2 α,ω-diacid, also increased two-fold in both mutants.
Both two and one and half times differences were also found for the C18:3 ω-hydroxy acid in lcr and fdh, respectively. It is also worth noting that both accumulated approximately less than twice the C18:2 acid in residual-bound lipids. The partially (by ∼25–40%) reduced content of the C16:0 acyl chains in the three classes of fatty monomers (acids, ω-hydroxy acids, α,ω-diacids), might be evidence of an enhanced C16 elongation in the mutants.
Examination of the cutin composition analysis data (Figure 4A and Figure S2) revealed that no shift towards shorter carbon chains could be detected in fdh. Moreover, no consistent decrease in the levels of ω-hydroxy fatty acids and α,ω-diacids could be detected in lcr. Given the increase in the levels of C18 ω-hydroxy fatty acids and α,ω-diacids, it could be proposed that the major changes in cuticular lipids observed in lcr and fdh are not due to the deficiency caused by the respective mutations. Instead, it could be evidence of an induced response to these mutations, which leads to the incorporation of more cutin-like material in the outer epidermal cell wall of the mutants. This response could play a compensatory role which contributes to the survival of the mutant plants. It is particularly noteworthy in this context that lcr and fdh appear to possess remarkably similar cutin monomer compositions.
There is a strong line of evidence linking the over-accumulation of epi- and intra-cuticular wax to the cuticular deficiency in bdg [14]. We, therefore, analyzed leaf wax composition in lcr and fdh (Figure 4B), revealing that the total amount of wax had been increased two-fold and three-fold in lcr and fdh, respectively; on average, wild-type leaves had a wax load of 0.72±0.07 µg/cm2 compared to 1.56±0.25 µg/cm2 and 2.66±0.25 µg/cm2 in lcr and fdh, again respectively. The major constituents of wax, which are alkanes in Arabidopsis, were increased from 2.3 times (C33 in lcr) to 9 times (C27 in fdh). The levels of free fatty acids and alcohols had also increased, but to a lesser extent, with up to a 3.3-fold increase in the C32 fatty acid (Figure 4B). The marked difference between the two mutants was the extent to which wax aldehydes were produced. Whereas these had not changed in lcr when compared to wild type, the fdh leaf wax appeared to contain much greater amounts of all aldehyde species than the wild type did, with C28 exhibiting the biggest (nearly a 830-fold) increase.
We also examined the leaf epidermis in lcr, fdh, bdg and wild type by cryo-scanning electron microscopy (SEM) which preserves wax morphology. Under SEM, the leaf surface in the wild type appeared to be smooth, without appreciable sculpturing. However, in all three mutants examined, a considerable number of ripples and plate-like wax crystals gave their surfaces a more ruffled appearance (Figure S3).
We concluded from these results that lcr and fdh respond to the loss of respective gene functions by the over-accumulation of wax in leaves. This response is quite similar with respect to alkanes, but only fdh appears to activate a pathway for aldehyde biosynthesis. Our SEM results suggest that stem wax is also affected in these mutants (Figure S4), but a detailed analysis would be beyond the scope of this paper.
The cuticular phenotypes of lcr and fdh described above, and the analysis of their cuticular lipids, suggest that these mutations may induce a kind of a response which includes the genes associated with cutin and wax biosynthesis. If this response is controlled at the level of transcription or mRNA stability, coordinated changes in the abundance of transcripts which encode corresponding genes should be observed in the cuticular mutants. To test this possibility, we studied the gene expression changes by using microarray hybridization with the Arabidopsis ATH1 Genome Array (Affymetrix). We then compared gene expression in young rosette leaves from fdh, lcr and bdg mutants to that in wild type. As described in Materials and Methods, RNA-derived probes were prepared from three biologically independent samples for each mutant.
Taking into account the low replicate numbers of the microarray data, we have chosen to detect differentially expressed genes (DEGs) using the Rank Product (RP) method, as suggested by Breitling and co-workers [31]. This produces a good performance, in particular for replicate numbers below ten [32]. We recently revealed that RP outperforms Cyber-T, Local Pooled Error (LPE), two-sample Bayes T, Empirical Bayes, SAM, fold change and the ordinary t-test in terms of the validity of the DEG lists [33]. The significance of the detection is assessed in RP by a non-parametric permutation test which evaluates the percentage of false positive predictions (pfp) or the false discovery rate (FDR). In this study, we regarded genes with a pfp of less than 5% (0.05) to be significantly differentially expressed because, for them, the probability of being consistently selected by the RP method is greater than 95%. This filtering resulted in a list of 440 DEGs in fdh when compared to wild type, followed by lcr (260 DEGs) and bdg (126 DEGs). The DEGs are listed in Tables S1, S2. Figure 5A summarizes the findings, and shows Venn diagrams which represent the number of genes that were changed and up or downregulated in the three cuticular mutants. The microarray analysis suggests that the majority of DEGs were upregulated: 88% for fdh, 91% for lcr, and 93% for bdg (Figure 5A). It also reveals large overlaps between misregulated genes in the different mutants. This supports the notion that these mutants exhibited similar transcriptional changes, as suggested by similarities in their organ fusion phenotypes and the composition of their cuticular lipids. The expression of only 13 genes (10%) was specifically changed in bdg, compared to 50 specific genes (19%) in lcr and 240 specific genes (54%) in fdh (Figure 5A). Table S3 includes 89 genes which were found in the overlap between the genes that were differentially expressed in all of these mutants. Remarkably, these 89 common genes represent 71% of all of the DEGs in bdg. A simplified version of this table is shown in Table 1.
To substantiate this computational analysis, we re-calculated the microarray data as one experiment which consisted of two groups: the cuticular mutant group with nine replicates and the wild type group with three. This experimental design should minimize the inter-dependence between mutant versus wild type group differences. One would also expect statistical power to increase as the number of replicates goes from three to nine for the mutant group. By using the same parameter settings in the Rank Product method, and the same pfp cut-off value, we obtained a list of 744 upregulated DEGs. Compared to the 87 commonly upregulated genes which were identified with the first approach, many more candidates were detected by Rank Product this time, suggesting that the actual number of genes discriminating between the three cuticular mutants and wild type is higher. However, 74 (85%) of these 87 genes were found in the overlap with the top-87 gene list from the second approach. A comparison of gene ranking also reveals that there is a significant consensus between the lists obtained by the two approaches.
To further corroborate our results, we sought to first demonstrate by semi-quantitative RT–PCR that the selected genes are, in fact, up or downregulated as predicted by the microarrays. We have chosen 12 candidate genes from among those which are commonly upregulated in the three mutants (Table S3), as well as two genes which were not included in this list. The first was CER1, which is thought to be directly involved in wax biosynthesis, although its enzymic function remains unknown; the other was a RAP2.6-like gene (RAP2.6L) which encodes an AP2/EREBP domain protein (Text S1). RAP2.6L was one of the genes which appeared to be upregulated, but did not meet the criteria because one of its pfp values was slightly above the 0.05 cut-off (0.052). From the microarray hybridization analysis, we estimated that the selected genes were upregulated in the range of 3 to 172-fold in the mutants (Table S3).
The results of semi-quantitative RT–PCR assays, shown in Figure 5B, indicated that all selected genes were consistently upregulated in the mutants. Interestingly, one can also observe a correlation between the genes that are strongly upregulated in both assays (namely the microarray and the semi-quantitative RT–PCR) with the LTPs, PPT, RAP2.6 and DAISY genes which display the most distinct expression between mutants and wild type. However, we did not aim to quantitatively evaluate gene expression measurements, or compare respective fold changes which would have required the use of real-time, quantitative RT–PCR.
One of the genes which was found to be highly up-regulated in the leaves of all three mutants (in lcr 5.8-fold, in fdh 11.7-fold and in bdg 9.0-fold, as revealed by microarrays) was DAISY (Table 1). It was shown to encode a KCS which is involved in the biosynthesis of aliphatic suberin in roots. The roots of the daisy mutant accumulate significantly less C22 and C24 very-long-chain fatty acid derivatives in suberin, suggesting that it functions as a docosanoic acid synthase [34]. The RT–PCR analysis and promoter-GUS fusions revealed that it was also expressed in various aerial organs of the plant, although its expression levels in leaves were very low [34]. While DAISY was almost undetectable in unwounded rosette leaves, its expression was rapidly induced by wounding, and correlated with suberin deposition around punctures [34]. These features render DAISY a good target for confirmation by in situ hybridization. In particular, we wondered whether it would be specifically induced in the leaf epidermis of the cuticular mutants.
The in situ hybridization results from wild type leaves, as shown in Figure 6A, demonstrate that the expression of DAISY is restricted to the xylem and phloem in vascular bundles of older rosette leaves; the hybridization signal was also observed the transmitting tract and ovules (Figure 6B). In the lcr and bdg mutants, the signal was also detected in leaf primordia and young developing leaves (Figure 6C–6 F), and the intensity of labeling in these organs was similar to that in the vasculature (Figure 6E and 6F). In the mutants, DAISY was ectopically expressed in all cell types in leaves, including the epidermis, and careful examination did not reveal any cell specificity. We concluded that the enlargement of the domain of DAISY expression obtained by in situ hybridization is in agreement with our microarray data and the results of semi-quantitative RT–PCR, all of which evidence the induction of DAISY in young leaves of lcr and bdg. We also concluded that the data from the microarray hybridization are reliable and suitable for a comparative analysis.
To find out which biological processes are most affected in cuticular mutants, we first used the Classification SuperViewer [35] which analyses Gene Ontology (GO) annotations (ATH_GO_GOSLIM.20070512) in order to identify overrepresented GO terms when compared to the entire Arabidopsis genome. The most prominent functional group of the upregulated genes (Table S3) was represented by “cell wall” related proteins, with a 15.8±4.7-fold enrichment (mean±standard deviation for 100 bootstraps) when compared to the whole genome. These were followed by the “extracellular” (7.6±3.3), “response to abiotic or biotic stimulus” (3.5±1.0), “plasma membrane” (2.2±1.7), “response to stress” (2.1±0.9), “other enzyme activity” (2.1±0.5), “developmental processes” (1.7±0.6), “hydrolase activity” (1.7±0.5), “transport” (1.7±0.6), and “other membranes” (1.7±0.3) groups. Other listed terms included “electron transport or energy pathways”, “transcription factor activity”, “signal transduction” and “transcription”, but these were less conspicuous. These results strongly indicate that the cell wall undergoes extensive changes in response to cuticular mutations. Given that the cuticle is an essential part of the cell wall, this should not come as a surprise. However, it is interesting that the genes associated with responses to abiotic or biotic stimuli were only third in this list, suggesting that the response to cuticular dysfunction leads to a specific compensatory response whereby the normal homeostasis of the cell wall (and cuticle) is altered, and the viability of the mutants is increased.
Our survey also showed that genes potentially associated with the cell wall, the cuticle and defense responses are upregulated in the cuticular mutants (Text S1). Although the list of commonly misregulated genes presented in Table S3 may not be complete and may contain about 5% of falsely discovered genes, this is a further indication that bdg, lcr and fdh respond similarly to the dysfunction by remodeling their cuticles and cell walls and activating defenses.
The above findings, when taken together, suggest that the transcriptional activation of target genes is an adaptive response to cuticular mutations such as bdg, lcr and fdh. Moreover, the fact that the three mutants display a peculiar phenotype, which is comprised of the overproduction of wax, organ fusions, irregular leaf shapes and defects in cell differentiation, suggests that the underlying signaling pathways may be distinct from, but overlap with, those that are activated in response to conventional biotic and abiotic factors.
To identify related pathways, we sought to determine which transcriptional responses were the most similar to those observed in bdg, lcr and fdh. We also thought that contrasting these results to a similar analysis for cell wall deficient mutants would help further in the definition of the mechanism by which cuticular mutations induce plant defenses.
To this end, a comparative analysis of differentially regulated genes among the three mutants should be extended to include several hundred of the microarray datasets that are available for Arabidopsis. This is challenging to implement because the generally low consistency of differentially expressed gene (DEG) lists achieved with the use of t-type tests has been reported by several groups, including those participating in the MicroArray Quality Control (MAQC) project [36]. It would be beyond the scope of this paper to go further into the computational details but, using the human MAQC project [36] and the Arabidopsis datasets available from the Gene Expression Omnibus (GEO), we have conducted a comparative survey of the acceptability of several statistical approaches. This revealed that the DEG lists selected by the Rank Product method [31] outperform, in terms of consistency, those generated by seven other methods [33].
Building upon this finding, we developed guidelines for a large-scale comparative analysis of DEG lists, and re-analyzed over 600 contrasts (e.g. mutant vs. wild type or treatment vs. control comparisons) with the Rank Product method, using the raw probe intensity data from the Affymetrix CEL files that we obtained from several databases and authors. We termed this approach MASTA (MicroArray overlap Search Tool and Analysis), after the phrase FASTA that is used for sequence comparison.
Mutations in CELLULOSE SYNTHASE4 (CESA4)/IRREGULAR XYLEM5 (IRX5) result in modifications in the composition and structure of the secondary cell wall and lead to the specific activation of defense pathways, suggesting that a cell wall integrity system which is similar to that of yeast may exist in plants [37]. This makes the side-by-side meta-comparison with cuticular mutants worthwhile. We, therefore, probed the MASTA database with the Rank Product generated lists of DEGs from both this study and the cesA4/irx5 microarray experiment [37]. For ease of computation, each probe comprised equal numbers (in this case 200) of the top up and downregulated DEGs, ranked by gene pfp (FDR) score.
The original MASTA output files are too long to be included in their entirety, and only selected subsets (126 out of 1208 overlaps) are, therefore, shown as examples in Figure 7 for lcr and cesA4/irx5. For each probe, a MASTA search performs pairwise testing for overlaps between up and downregulated genes in the probe and target DEG lists, thus totaling four combinations for each comparison. Following the customary terminology used in genetic analysis, in MASTA a coupling-phase overlap refers to an overlap between up and upregulated genes, or between down and downregulated genes, whereas repulsion-phase overlap refers to an overlap between up and downregulated genes, or between down and upregulated genes in the probe and target DEG lists, respectively.
The MASTA revealed that the DEGs in cuticular mutants were in the coupling phase with those in the CUTE plants [19], with 53 genes (P<2.0×10−64) in the overlap between the upregulated genes (Figure 7). The cesA4/irx5 mutation induced a number of genes which are shared with cuticular mutants, but gene overlaps varied between 29 and 36 (P<5.3×10−27 and P<6.2×10−37, respectively) (Figure 7).
It is apparent from the comparative analysis that lcr transcriptional responses are quite similar to those induced by wounding. The overlap between upregulated genes was conspicuously like the early wounding response (15 min after wounding), but it became stronger at subsequent points in time. Although wounding-inducible genes showed statistically significant overlaps with DEGs in cesA4/irx5, the salt treatment, osmotic stress and drought appeared to misregulate most of the similar set of genes in cesA4/irx5. Large gene overlaps were noticeable for both upregulated and downregulated genes (Figure 7), with salt stress being the most similar to cesA4/irx5.
Remarkable differences between lcr and cesA4/irx5 were seen in the overlaps with transcriptional responses to growth regulators: steroids, methyl jasmonate (MeJ), naphthaleneacetic acid (NAA) and abscisic acid (ABA). While hardly any overlaps above the threshold line were detected with the genes downregulated in lcr, the strong repulsion-phase overlaps were displayed with genes downregulated in cesA4/irx5. This suggests that a number of steroid-inducible genes (from 50 to 60) are suppressed in the cesA4/irx5 plant, thereby evidencing the fact that steroid hormones play an essential role in the cesA4/irx5 phenotype. Adding to the difference between cuticular and cell wall integrity signaling pathways, was the presence of strong coupling phase overlaps between cesA4/irx5 DEGs and those induced by MeJ, NAA and ABA (Figure 7). The involvement of ABA and jasmonic acid (JA), predicted by MASTA, was in accordance with the results of Hernandez-Blanco [37] and co-workers, who used the Genevestigator Meta-Analyzer tools (www.genevestigator.ethz.ch/at/) to compare selected upregulated genes which showed fold-change values >2. Remarkably, although cesA4/irx5 was not tested, irx1 and irx3 were crossed with two ABA-insensitive mutants, abi1 and aba3, as well as the double homozygous mutants, irx1 abi1 and irx3 aba3, and appeared to be unviable two to three weeks post-germination.
Therefore, based on these findings, we propose that in contrast to cesA4/irx5, major gene expression changes in lcr are induced independently of MeJ, NAA and ABA, and brassinosteroid signaling. Taken together, these data suggest that the underlying response mechanisms in the cesA4/irx5 and lcr mutants to cell wall and cuticle defects, respectively, are sufficiently different.
The suppressor/enhancer screens involving the mutagenesis of the targeted mutant provide a solution to the problem of revealing additional genes in a given pathway. Alternatively, a set of known mutants may be crossed with the mutant of interest to make a series of double mutants, then enabling their phenotypes to be evaluated. The public availability of various mutant microarray datasets provides an opportunity to rapidly assess the potential of a large number of genes as genetic modifiers by using MASTA prior to the crosses. When resulting from probing with the mutant of interest, the coupling and repulsion phase overlaps indicate the presence of putative enhancers and suppressors, respectively.
To identify these, we surveyed the most significant gene overlaps with the mutant microarray datasets. A particularly noticeable case in Figure 7 was the coupling-phase overlaps of cesA4/irx5 with pad4 (54 upregulated genes, P<3.7×10−66) and eds1 (59 upregulated genes, P<4.1×10−75). This suggested that both could be enhancers of cesA4/irx5 in double mutants. EDS1 and PAD4 encode lipase-like proteins which can form a heterodimer and are required for the accumulation of the plant defense signal, salicylic acid [38]. Eds1, but not pad4, showed a coupling-phase overlap with lcr (38 upregulated genes, P<6.3×10−40), suggesting that it could enhance the lcr phenotype in double mutants, whereas pad4 could not.
Three cases in Figure 7, which we considered to be particularly relevant as suppressors, were the recessive mutations in SERRATE (SE), SENSITIVE TO FREEZING2 (SFR2) and SENSITIVE TO FREEZING6 (SFR6). From 23 to 45 (P<3.1×10−19 to 6.3×10−51), the genes that were repressed in the se, sfr2 and sfr6 mutants were induced in lcr, fdh and bdg, exhibiting a repulsion phase overlap.
The se-1 mutants exhibit conspicuous leaf serration, and are affected, not only in other aspects of leaf development, but also in embryogenesis, flowering time and seedling responses to the hormone, ABA. The stronger se alleles, namely se-2 and se-3, severely disrupt both meristem activity and leaf polarity [39]. SERRATE (At2g27100) is a zinc-finger protein which has been shown to participate in both RNA splicing and the processing of pri-miRNA transcripts into mature miRNAs [39]–[43]. The MASTA search suggests that SE has a different role in the cell wall stress pathway, with 19 downregulated genes in a coupling-phase overlap with cesA4/irx5 (Figure 7).
We reasoned that if SE is required for the induction of the responses in cuticular mutants, some aspects of their phenotypes should be the opposite to those of se because MASTA reveals a repulsion phase overlap for their misregulated genes. We had previously noticed that lcr and bdg do, indeed, possess smooth-edged elongated leaves [14],[16], and this case has, therefore, been selected for further testing.
From the above analysis with MASTA, one can anticipate the suppression of either phenotype in double mutants, although given the fact that SE acts in the context of a multiprotein complex which is involved in RNA processing, se is more likely to be epistatic to the cuticular genes. To corroborate this prediction, we obtained double mutants with the weak se-1 allele [41]. The Figure 8A shows that both se lcr and se bdg feature serrated leaves and, essentially, look like se plants [41]; (se fdh is not yet available because the genes are tightly linked on the long arm of chromosome 2). Remarkably, under normal growth conditions, the double mutants failed to develop the ectopic organ fusions, leaf deformations and plant architecture that were characteristic of the single cuticular mutants in this class (Figure 8A) [9],[14],[16].
The TB staining phenotype of lcr could be significantly reverted in a se mutant background (Figure 8B). After infection with B. cinerea, se lcr and se plants tended to have larger lesion areas than wild types (Figure 8C) (the lesion areas are not shown as the lesions often covered the entire leaf surface in se and se lcr). The rate of infection was higher in se lcr and se mutants than that in wild type plants (Figure 8D). Remarkably, leaves of se bdg exhibited TB staining pattern and resistance to B. cinerea similar to that in bdg (Figure 8).
These findings suggest that secondary phenotypes in distinct cuticular mutants result from the induction of a response, which requires SE. This is also evidence that MASTA provides a powerful way of identifying suppressors and enhancers in the pathway of interest.
It might be anticipated that cuticular mutants would, in general, have reduced levels of cutin monomers or wax and display conspicuous cuticle defects, however our data suggest a more complex picture when considering the cuticular mutants which display the organ fusion phenotype. We showed that the cuticle is highly disorganized in lcr, with cutin-like depositions and cavities in the inner layers of the cell wall. This closely resembles the cuticle of bdg and the CUTE plants [14],[19] as well as, to a lesser extent, ace/hth [27]. Yet, we also demonstrated that TEM did not reveal any visual aspects in the fdh cuticle which appear to be different from wild-type. However, these organ-fusion mutants are noticeably similar in the accumulation of higher levels of wax and cutin constituents in the residual bound lipids. This sets them apart from the wild-type-looking plants of cuticular mutants which exhibit a concomitant lowering of the levels of cutin components, such as lacs2 and att1 [21],[44],[45]. We also found that the chemical composition of the cutin, as revealed by the analysis of leaf residual bound lipids, does not show a decrease in ω-hydroxy and α,ω-dicarboxy fatty acids in lcr; there was also no reduction in the lengths of the fatty acid chains in fdh.
It is, therefore, becoming evident that a simple case scenario does not seem to be plausible for all cuticular mutants, meaning that other mechanisms need to be taken into account in order to understand the precise nature of their phenotypes and the role of cuticle in development and immunity. It may be proposed that some cuticular mutations induce a kind of a damage response that is similar to the activation of the cell wall integrity pathway in response to cell wall disrupting drugs and mutations in fungi and plants [37],[46].
Consistent with the previously observed induction of stress and defense-related genes in some cesA mutants [37] and the CUTE plants [23], we found by microarray gene expression profiling in lcr, fdh and bdg that a number of upregulated genes fall into this class. The GO analysis suggests that the functions of these genes are mainly associated with the cell wall.
We showed that misregulated genes in cuticular and cell wall mutants substantially overlap, but these genes are significantly more similar to each other than to those in the cesA4/irx5 mutants. This may be indicative of the existence of specific signaling routes which engage transcriptional control mechanisms in the cuticular mutants. The fact that cuticular mutations confer a highly pleiotropic phenotype, including organ deformations and fusions, the extensive branching, delayed senescence and resistance to the fungal necrotroph B. cinerea [9],[10] that were not observed in cell wall mutants is in accord with this notion.
Using biochemical analytical techniques may make it difficult or impossible to separate the overlapping loss-of-function and response phenotypes, thereby emphasizing the role of genetic approaches. To further assess the complexity of the factors responsible for the range of cuticular phenotypes, we used a meta-analytical methodology, MASTA, which has involved a reappraisal and re-analysis of several hundred Arabidopsis microarray datasets from public databases and authors [33]. MASTA is based on the proof-of-concept study showing that, a reference database of expression profiles which correspond to diverse chemical treatments and mutations in yeast can be used to functionally annotate uncharacterized genes and pharmacological perturbations in this substance [47].
Using this bioinformatics tool for the in silico suppressor screen, we identified SE as a gene which is epistatic to lcr and bdg. This prediction has been supported by three lines of evidence. First, the double mutants, se lcr and se bdg, failed to produce featured morphological changes, including organ fusions, in particular. Second, the se lcr mutant lost its characteristic TB staining on the epidermis surface. Third, se lcr lost resistance to the necrotroph, also indicating that se is an essential factor contributing to the complex phenotypes of cuticular mutants.
The ectopic organ fusion phenotype results from enhanced cell adhesion in epidermal cells and is interesting, because cell adhesion is a fundamental process underlying development. While adhesion between plant cells is generally established when cells are formed during cytokinesis, cells dynamically regulate adhesion, and may undergo separation or establish fusions de novo in a controlled manner, with pollen tube growth and carpel fusion being the best-known examples of the latter process [48],[49]. This suggests that there are multiple mechanisms by which plants can regulate adhesion between cells.
Organ fusions in the fdh mutant of Arabidopsis offered a genetic proof that the developmental program, normally limited to the gynoeceum, could be induced in the whole plant [12]. In addition to fdh, other Arabidopsis mutants have been reported to exhibit organ fusions and impaired morphogenesis [15], [16], [19], [27], [50]–[57]. However, the function of cuticle in these processes remained open to question. Given the role that the cuticle plays in the isolation of plant surfaces, it is necessary to study the corresponding genes in the cuticle context, particularly because most of the mutants seem to be the consequence of lesions in the genes which encode lipid-modifying enzymes. Remarkably, cell wall-related genes are prominent in the overlap which is comprised of the commonly upregulated genes in the cuticular mutants and downregulated genes in se (Table 1 and Table S3), Further experiments would be required to determine which of these genes, or others as yet not recognized, are involved in epidermal adhesion and leaf morphogenesis. Since the latest version of the Arabidopsis genome annotation (TAIR8) contains information about 27235 protein-coding genes, and the ATH1 array represents approximately 23750 genes (87%) [58], we note that about 13% of the downstream genes of interest may escape identification in both our and Lobbes' microarray experiments [43]. The discovery of epistatic effects could be the first step towards identifying the cell-surface molecules and understanding genetics of the putative interactional mechanisms which underlie the organ fusion phenotype.
The mechanism by which the distorted cuticle leads to a strong resistance to B. cinerea is also unclear but may involve an inhibitory action of plant-derived toxins such as camalexin and a higher permeability of cuticle to these compounds or fungal elicitors [59],[60]. Fast induction of camalexin biosynthesis genes in wounded and infected plants accounts for the strong immunity against B. cinerea [59]. MASTA revealed that DEGs in se (data not shown) were in the repulsion phase with those in cuticular mutants and wounded plants, suggesting that wounding response may be compromised by the se mutation in se lcr. However, we did not find that se bdg plants were more susceptible (and become less stainable) than bdg, indicating a greater complexity of the antifungal resistance in cuticular mutants.
The function of the nuclear-localized SE protein in the regulation of the pleiotropic phenotypes of cuticular mutants also remains to be determined. So far it is known that, like DICER-LIKE1 (DCL1) and HYPONASTIC LEAVES1 (HYL1), SE is required for miRNA biogenesis but not for sense post-transcriptional gene silencing [42],[43]. In se, 20 upregulated genes have been identified as known targets of miRNAs and/or transacting siRNAs (ta-siRNAs) [43]. However, only downregulated genes in se show an overlap with DEGs in the cuticular mutants, and none of the se downregulated genes present on the ATH1 chip are known miRNA targets [43]. Interactions between the SE, DCL1 and HYL1 proteins are essential for the efficient and precise processing of pri-miRNA [61], although MASTA does not predict the presence of epistatic interactions with dcl1 and hyl1 (data not shown). The possibility that SE may be associated with an alternative pathway in RNA signaling warrants further investigation. Nevertheless, it is an important finding that se is epistatic to the mutations in the distinct epidermis-specific cuticular genes LCR and BDG, suggesting that the impaired cuticle triggers specific cell signaling pathway. Accordingly, this study offers an intriguing and unexpected insight into how cuticle formation, cell adhesion and morphogenesis in plants may be co-regulated.
All plants used in this study were derived from Arabidopsis thaliana (L.) ecotype Columbia (Col-O).
The mutant alleles used were: fdh-3940S1 [20], lcr-3P77 [16], bdg-2 [14], se-1 [41], lacs2-3 [21] and pad3-1 [25]. Arabidopsis plants were grown in a greenhouse or controlled environment chamber at 22 to 23°C and 50 to 60% humidity under a short day photo-period (8-h light) for the first 6–7 weeks, and then under a long day photo-period (16-h light) if not, otherwise, indicated.
The putative 1.3-kb promoter region of the LCR gene was amplified by PCR with the PLCR-H (TGAACTCCAAAGCTTTACATGACTACATCG) and PLCR-Xb (CTCTAGATCTCCTCATAAACTTGGAGTGA) primers (HindIII and XbaI sites are underlined in the primer sequences), and cloned as a HindIII/XbaI fragment into the pBgreen binary vector [17]. Wild-type Arabidopsis thaliana Col-0 plants were transformed with the resulting pBLCR:GFP by vacuum-infiltration [62], and BASTA-resistant transgenic plants were selected. The analysis of GFP expression with a confocal laser scanning microscope (Leica TCS 4D) was performed in tissue sections from the pBLCR:GFP plants as described [14].
In situ hybridization was performed using the same antisense digoxigenin-labeled riboprobes that were derived from a cDNA clone of the DAISY gene, and an epidermis-specific control, as in Franke [34]. The hybridization products were revealed by an immunohistochemical reaction after incubation with an alkaline phosphatase-conjugated anti-digoxigenin antibody. Probe preparation, the hybridization procedure, and immunohistochemical detection were conducted as previously described [17].
To study the fine structure of the cuticle, plants were grown for 4–5 weeks under long day conditions. Tissue samples were embedded, and ultra-thin sections (50–70 nm) were contrasted as described in [21] (fdh) and [19] (wild type and lcr). The samples were examined with a Phillips CM12 transmission electron microscope or a Philips CM 100 BIOTWIN electron microscope. The details of the procedures can be found in our previous work [19],[21].
To examine wax coating, rosette leaves from 6–7 week-old plants grown under short day conditions, as well as stem internodes (4th and 5th) from 12 week-old plants (8 weeks under short and 4 weeks under long-day conditions) were prepared for cryo-SEM. The samples were deep-frozen and sputtered with palladium using the K1250X cryogenic preparation system (Emitech, England). Leaf surfaces were examined with a Zeiss SEM SUPRA 40VP microscope.
Toluidine blue staining was performed according to Tanaka and co-workers [26]. Chlorophyll leakage from rosette leaves into ethanol was performed according to Lolle and co-workers [13], with modifications as previously described [14].
B. cinerea strains 2100 and B05.10 were grown on the potato dextrose agar (Difco Laboratories) at 22°C for seven to nine days. Spores were harvested and washed twice in water, and filtered through Miracloth (Calbiochem). Inoculations were made as previously described [24]. Briefly, rosette leaves of four-week-old soil-grown plants were placed in square Petri dishes containing 0.8% agar with petioles in the medium. Four µl of a suspension containing 5×105 conidiospores/mL in 1/4-strength potato dextrose broth (Difco Laboratories) were deposited on the detached leaves. A single drop was applied to each leaf between the middle vein and the edge of the leaf, and the leaves were incubated under continous light at 22–24°C. Disease symptoms were scored at 3 days after challenge. High humidity was maintained by sealing the dishes with Parafilm. Macroscopic images were acquired with a digital camera (Sony DSC-W120) and the lesion area was measured in pixels using Image J (software available at http://rsbweb.nih.gov/ij/) and then converted to square millimeters. The inoculation experiments were repeated three times with detached leaves using at least 50 leaves per genotype and once in planta with similar results.
Fatty acid composition analyses of residual bound lipids and wax were performed as previously described in detail [14],[27]. Cutin and wax constituents were separated and identified by GC–MS using a gas chromatograph 6890N equipped with a quadrupole mass selective detector 5973N (Agilent Technologies, Boeblingen, Germany). The composition analysis in the lcr and fdh leaves was performed twice. In each experiment, plants were grown for 10–11 weeks under short day conditions prior to tissue harvest. For wax analysis, plants were grown for seven weeks under the same conditions.
Mutants (bdg, lcr, fdh) and WT (Col-0) were grown side by side in a growth chamber under short day conditions at 20°C/18°C. Plants were five and a half weeks old when tissue was harvested. Three independent samples, each containing typical young leaves (from 2 mm to 1 cm long) from 15 plants, were prepared per plant type. Total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) according to the manufacturer's instructions. RNA concentration and quality were assessed with agarose gel electrophoresis. The samples were sent to the Integrated Functional Genomics (IFG) platform of the Westfalian-Wilhelms-University (Muenster, Germany; http://campus.uni-muenster.de/ifg.html) for a further quality check, preparation of biotin-labeled cRNA probes, hybridization to GeneChip Arabidopsis ATH1 Genome Arrays and scanning of the slides. All of the above steps were performed according to the manufacturer's instructions. For each plant type, we had three biological and no technical replicates.
The pre-processing of raw data and the Affymetrix MAS5.0 Quality Control tests were performed using Bioconductor packages (http://www.bioconductor.org) and custom written scripts in the R programming environment (http://www.r-project.org). For the quality control tests, we used the SimpleAffy package [63]. The quality-control measures indicated that the 12 microarrays used in this study show no systematic signal distortion, similar scale factors, and adequate background levels, sufficient percentage of genes called “present” and acceptable performance in 3/5 ratio tests (Figure S5). The microarray data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE15105).
Statistical analysis was performed using custom written scripts for the Bioconductor RankProd package [64]. To estimate the false discovery rate (FDR), pfp (false positive predictions) values have been calculated from 100 permutations. The predicted differentially expressed genes (DEGs) have been ordered by increasing pfp value. For this report, a 5% (0.05) pfp cutoff has been applied to the definition of the DEGs in the mutants.
The meta-analytic software, MASTA (MicroArray overlap Search Tool and Analysis), was written to run in R (http://www.r-project.org).
A database for MASTA was created. To this end, Affymetrix raw data files (CEL files) were downloaded from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo), ArrayExpress (http://www.ebi.ac.uk/microarray-as/ae/), TAIR AtGenExpress (http://www.arabidopsis.org/index.jsp), the Integrated Microarray Database System (http://ausubellab.mgh.harvard.edu/imds), or via the NASC Affywatch subscription service (http://nasc.nott.ac.uk/). Some CEL files have been obtained from authors' websites or from the authors directly. At the time of this study, the MASTA database comprised DEGs for over 600 contrasts (mutant vs. wild type or treatment vs. control comparisons) calculated by using custom RankProd scripts in the R programming environment. Other portions of MASTA included the overlap analysis and plotting routines (details will be published elsewhere). The RankProd-selected DEGs were ordered by increasing pfp value, and the top 200 DEGs from the lists containing up and downregulated genes were used for the overlap analysis in this report. Output PDF files from MASTA were imported to Adobe Illustrator (Adobe Systems, San Jose, CA) for assembly. The statistical significance of the overlap between two DEG lists was determined by using the online program available at http://elegans.uky.edu/MA/progs/overlap_stats.html.
MASTA will be made available for viewing and downloading at http://bar.utoronto.ca/ (The Bio-Array Resource for Arabidopsis Functional Genomics).
In brief, 0.5 µg aliquots of the total RNA of each hybridization sample, treated with DNase I, were reverse-transcribed to the first-strand cDNA with a One-Step RT–PCR Kit (Qiagen). These cDNAs were used as templates for PCR under the following conditions: denaturation at 94°C (1 min); Nopt cycles of 94°C (1 min), 58°C (45 sec) and 72°C (30 sec, except for At4g30280 where it was 1 min); then 72°C final extension (15 min). Gene-specific primer pairs are listed in Table S4. The expression of ACTIN2 (At3g18780) was analyzed as an internal control. The semi-quantitative RT–PCR reactions were optimized for number of cycles Nopt to ensure product intensity within the linear phase of amplification (close to the the lower limit of the linear range) for each gene (Table S4). PCR fragments were quantified (Typhoon 8600 PhosphorImager, Amersham Biosciences) following electrophoresis in ethidium bromide-containing agarose gels.
The segregating F2 populations (about 200 plants each) derived from the lcr×se-1 and bdg×se-1 crosses respectively were tested in a blind two-stage screen for the presence of mutant and wild-type alleles in the LCR, BDG and SERRATE loci by PCR. The following PCR conditions were used: denaturation at 94°C (2 min); 36 cycles of 94°C (30 sec), 58°C (30 sec) and 72°C (50 sec); then final extension at 72°C (5 min).
At the first stage, DNA was isolated in 96-well blocks, based on the method described previously [65]. To genotype the LCR locus, amplification products were loaded on a 1.5% agarose gel. For the BDG and SE loci, half of the PCR products were first digested with MwoI or BfuCI (NEB) respectively, and then the undigested and digested PCR products were loaded onto a high-resolution agarose gel (3% GenAgarose Tiny HT, Genaxxon Bioscience).
At the second stage, candidate double mutants were re-screened by repeating DNA isolation with the DNeasy Plant Mini Kit (Qiagen) and PCR analysis. Several plants have been identified for each double mutant type. The allele-specific primers and PCR products are listed in Table S5; the sequence of the lcr-3P77 transposon insertion allele was deposited in the GenBank under accession number FJ767868.
|
10.1371/journal.pntd.0007374 | Genus-wide Leptospira core genome multilocus sequence typing for strain taxonomy and global surveillance | Leptospira is a highly heterogeneous bacterial genus that can be divided into three evolutionary lineages and >300 serovars. The causative agents of leptospirosis are responsible of an emerging zoonotic disease worldwide. To advance our understanding of the biodiversity of Leptospira strains at the global level, we evaluated the performance of whole-genome sequencing (WGS) as a genus-wide strain classification and genotyping tool. Herein we propose a set of 545 highly conserved loci as a core genome MLST (cgMLST) genotyping scheme applicable to the entire Leptospira genus, including non-pathogenic species. Evaluation of cgMLST genotyping was undertaken with 509 genomes, including 327 newly sequenced genomes, from diverse species, sources and geographical locations. Phylogenetic analysis showed that cgMLST defines species, clades, subclades, clonal groups and cgMLST sequence types (cgST), with high precision and robustness to missing data. Novel Leptospira species, including a novel subclade named S2 (saprophytes 2), were identified. We defined clonal groups (CG) optimally using a single-linkage clustering threshold of 40 allelic mismatches. While some CGs such as L. interrogans CG6 (serogroup Icterohaemorrhagiae) are globally distributed, others are geographically restricted. cgMLST was congruent with classical MLST schemes, but had greatly improved resolution and broader applicability. Single nucleotide polymorphisms within single cgST groups was limited to <30 SNPs, underlining a potential role for cgMLST in epidemiological surveillance. Finally, cgMLST allowed identification of serogroups and closely related serovars. In conclusion, the proposed cgMLST strategy allows high-resolution genotyping of Leptospira isolates across the phylogenetic breadth of the genus. The unified genomic taxonomy of Leptospira strains, available publicly at http://bigsdb.pasteur.fr/leptospira, will facilitate global harmonization of Leptospira genotyping, strain emergence follow-up and novel collaborative studies of the epidemiology and evolution of this emerging pathogen.
| Leptospirosis, caused by pathogenic Leptospira strains, is an emerging bacterial zoonotic disease mostly affecting humans in tropical countries. Despite its public health importance, little is known about the strains that are circulating worldwide due to the lack of a universal common language on strain types. In this work we describe a new strain genotyping and classification system that is highly standardized, thus facilitating global collaboration, and that can discriminate all members of the Leptospira genus at high resolution. We then examine the genetic diversity of Leptospira strains from different origins. This study provides a framework for optimizing diagnostic methods and epidemiological surveillance of leptospirosis.
| Spirochetes constitute an evolutionarily and morphologically unique group of bacteria [1]. Pathogenic members of this phylum are the causative agents of several important diseases including leptospirosis, an emerging zoonotic disease with more than one million severe cases and 60,000 deaths every year worldwide, mostly in the tropical countries [2]. Pathogenic Leptospira species can cause a wide range of diseases in human, ranging from mild flu-like symptoms to severe complications, such as Weil's disease and pulmonary hemorrhagic syndrome, in which the case fatality rate can reach 40% [3]. Leptospirosis is expected to become more prominent worldwide due to climate change and the growing urban population living in slums. In addition, infections with pathogenic species can lead to major economic losses in livestock, as animal infections include e.g., abortion and loss of milk production [4].
The high public health and economic importance of Leptospira calls for better control of the infections the bacteria cause to both humans and animals. However, the control of Leptospira transmission is challenging for several reasons. First, the life cycle of pathogenic Leptospira is complex. Pathogenic leptospires are excreted through the urine of a wide range of animals including rodents which are asymptomatic reservoirs and livestock. Transmission to susceptible hosts usually occurs through contact with water contaminated with the urine of infected animals [5]. Therefore, multiple environmental sources of exposures, linked to multiple animal species, must be considered as possibilities.
Further complicated matters, the genus Leptospira is genetically highly heterogeneous and knowledge of its biodiversity remains largely incomplete. Taxonomically, the genus is currently subdivided into 35 species [6]. These species are ordered into three major evolutionary clades named according to their virulence status: pathogens, intermediates and saprophytes [1]. The agents of leptospirosis belong to two subclades, the pathogens (13 species) and the intermediates (11 species). The pathogenic species are responsible of the most severe infections in both human and animals, yet we know little about which component of the spirochete are critical for virulence. The species of the intermediates subclade are widely distributed in the environment [6–10] and they may be responsible for mild infections in both human and animals [11–19]. Intermediates possess most of the virulence factors found in the pathogens [1, 20]. In turn, the saprophytes form a single clade containing eleven species that are regarded as non-pathogenic environmental bacteria [1]. Saprophytes are relatively fast-growing in vitro when compared to the pathogens and lack the virulence factors described in infectious strains [1]. Classification into the three main clades has been typically performed using housekeeping and 16S rRNA genes sequencing [20].
Yet another barrier against leptospirosis control is the difficulty in isolating and cultivating Leptospira, which hinders optimal diagnostics of infections as well as laboratory identification, and hampers the constitution and maintenance of strain culture collections that are needed for microbiological studies and diagnostic or vaccine development purposes.
Finally, there is a lack of efficient strain typing methods that would allow tracking Leptospira strains (i) from their environmental or animal sources to their infected hosts and (ii) as they spread across time and space. Serotyping, which relies on the use of specific monoclonal antibodies, has led to the distinction of >300 serovars based on the structural heterogeneity of the surface-exposed lipopolysaccharides (LPS). This method has demonstrated an association of serovars with some animal reservoir hosts [21], even though the mechanisms that have allowed the adaptation of pathogenic Leptospira to various hosts are still unknown. However, serovar identification is currently performed by only two reference laboratories worldwide and is fastidious and time-consuming [22]. Furthermore, correlation between serotypes and genomic background is not always accurate, as the LPS biosynthetic locus (rfb) can be horizontally transferred between Leptospira species [23–25].
Molecular typing methods include pulsed-field gel electrophoresis (PFGE) [26, 27], and multilocus variable-number tandem-repeat analysis (MLVA) [28], but both methods have important practical limitations. Thus, PFGE [26] is not widely used and laborious, and only the most common serovars are typeable. More recently, multilocus sequence typing (MLST) was developed [29–31], but unfortunately three distinct MLST schemes have been proposed and applied to distinct collections of isolates, resulting in fragmentation of Leptospira epidemiological knowledge. Further, given the heterogeneity of Leptospira, the above methods are not universally applicable to all clades and species. In particular, MLST schemes are mainly focused on pathogens. As a consequence, current knowledge on the biodiversity and epidemiology of Leptospira is limited, and there is a critical need for a consensus Leptospira genotyping method that would be inclusive for its entire biodiversity, would facilitate fine-level strain discrimination for epidemiological purposes, and would reach high standardization allowing comparison of data from laboratories globally.
Whole-genome sequencing (WGS) has emerged as a powerful tool for bacterial strain classification and epidemiological typing [32]. The core genome MLST (cgMLST) approach, which extends the MLST concepts to the core genome, was demonstrated to be a useful high-resolution typing method in other bacterial species [33–36].
Taking advantage of the unique strain collection of the Reference Center for Leptospirosis in charge of the leptospirosis surveillance in mainland France and French overseas territories, our objectives were (i) First, to define based on genomic sequencing, the phylogenetic diversity of Leptospira, and its links with ecology and geography. In particular, our purpose was to shed light on the saprophyte and intermediate clusters, which have been scarcely studied thus far, and to include potentially novel species in this analysis. (ii) Second, we aimed to devise a genomic sequence-based genotyping method that is simultaneously universally applicable across the entire Leptospira genus and highly discriminatory at the strain level, and to propose a genomic taxonomy of Leptospira strains.
We sequenced 327 genomes from the collection of the National Reference Centre for Leptospirosis (Institut Pasteur, Paris, France), which is a globally representative strain collection of isolates from environmental, animal, and human samples gathered in the last 50 years. All strains and genome sequences used here are listed in S1 Table. Leptospira strains were grown at 30°C in liquid Ellinghausen, McCullough, Johnson and Harris (EMJH) medium. Species identification and serovar typing were performed at the National Reference Centre for Leptospirosis (Institut Pasteur, Paris, France) as previously described [37–39].
Collection of the strains was conducted according to the Declaration of Helsinki. A written informed consent from patients was not required as the study was conducted as part of routine surveillance of the national reference center and no additional clinical specimens were collected for the purpose of the study. Cultures originating from human samples were anonymized. Approval for bacterial isolation from soil and water was not required as the study was conducted as part of investigations into leptospirosis outbreaks. For New Caledonia, approval for bacterial isolation from the natural environment was obtained from the South Province (reference 1689–2017) and North Province (reference 60912-2002-2017).
Bacterial genomic DNA was purified using MagNA Pure 96 Instrument (Roche). Next-generation sequencing was performed by the Mutualized Platform for Microbiology (P2M) at Institut Pasteur, using the Nextera XT DNA Library Preparation kit (Illumina), the NextSeq 500 sequencing systems (Illumina), and the CLC Genomics Workbench 9 software (Qiagen) for analysis. Draft genomes with 50x minimum coverage, a total size < 5.3 Mb, and a minimum N50 of 10,000 nt were used for subsequent analysis. All raw reads generated and/ or contig sequences were submitted to NCBI under the project number PRJEB29877 and are available under genome accession numbers ERR3047203 to ERR3047514.
We also downloaded 182 assembled genome sequences from the NCBI and PATRIC (www.patricbrc.org) databases, including reference strains of previously described species [20] and representative isolates for each clade (S1 Table).
To determine a core gene set, 103 high-quality genome sequences of Leptospira covering the whole diversity of the Leptospira genus, i.e., representative isolates from the three clusters (50% from the pathogens, 12% from the intermediates, and 38% from the saprophytes) were selected (S1 Table); 50% of the genomes were downloaded from NCBI, the others were sequenced as described above.
From this set we inferred the genus core genome using the CoreGeneBuilder pipeline [40] and L. interrogans serovar Copenhageni strain Fiocruz L1-130 (GCF_000007685) as a reference. The pipeline’s first step relies on the eCAMBer software [41], which consists of a de novo annotation of the genomes (except the reference) using Prodigal [42] and the harmonization of the positions of the stop and start codons. In the next step, the core genome is inferred with a bidirectional best hits (BBH) approach as previously described by Touchon et al. [43]. We used CoreGeneBuilder default settings except for the synteny parameters (options–R and–S) both of which were set to 1. A gene was considered as part of the core genome if found in at least 90% of our genomes. Genes were not requested to be present in all genomes, as this stringent definition of a core genome would have resulted in too few genes given the diversity of Leptospira. Instead, the set of genes defined using the relaxed requirement of 90% presence can be viewed as a “soft core genome”. This resulted in an initial core genome containing 764 genes. We then filtered out some genes based on the following criteria. (i) First, we removed potential paralogs. Indeed, the presence of paralogs inside a typing scheme can lead to ambiguities, as a candidate gene might be attributable to two different core gene loci. To detect those potential paralogs, we compared each allele of each locus against all the alleles of all the other loci using the software BLAT [44]. If a single hit was found between two different loci (more than 70% protein identity between two alleles), we removed both. (ii) Second, we also removed genes that belong to one of the 3 existing Leptospira MLST schemes [29, 45, 46] and the ribosomal genes, so that they can be analyzed independently. (iii) Third, we also removed loci whose length varies too much among alleles, which is useful in reduceing ambiguities during the genotyping process. We aligned the protein sequences and removed those for which the alignment contained more than 10% of gaps (total number of gaps compared to the total number of characters). (iv) We removed loci containing ambiguous characters. (v) Finally, to avoid redundancy in the information contained within the cgMLST scheme, we removed loci that were overlapping in the reference genome using the definition of Prodigal [42]: a minimum of 60 bp of overlap if genes are on the same strand, and of 200 bp if genes are on different strands. The analysis resulted in the selection of 545 core genes listed in S2 Table and this cgMLST scheme was then used to analyze the presence of genes and to call alleles in 509 genomes (S1 Table), including the 103 genomes used for core genome definition. The allele and profiles definitions of the Leptospira cgMLST scheme were made publicly available through an Internet-accessible genotyping platform at https://bigsdb.pasteur.fr/leptospira/.
To derive a phylogenetic tree based on cgMLST gene loci, the allelic sequences of each locus were extracted and aligned as protein sequences using MAFFT v7 [47]. The concatenation of all loci yielded to a supermatrix of characters. IQ-TREE v1.5.4 [48] was used to infer a phylogenetic tree from this supermatrix of characters with an LG+G evolutionary model. Branch supports were assessed with both bootstrap (1,000 replicates) and aLRT-SH methods [49]. All trees were drawn using the iTOL webserver [50].
To evaluate classical MLST against the newly defined cgMLST scheme, all available Leptospira STs were downloaded from the Oxford University MLST database at https://pubmlst.org/leptospira/ [51] which comprises schemes 1, 2, and 3 developped by Boonsilp et al. [45], Varni et al. [46] and Ahmed et al. [29], respectively (S1 Table). MLST alleles derived from our WGS data were compared to the MLST database to determine the ST of our genome assemblies. Simpson index of discrimination and Wallace or Rand indices of concordance among partitions were computed using the web site http://www.comparingpartitions.info [52, 53].
A total of 327 Leptospira isolates were sequenced, covering the diversity of the Leptospira genus. A complementary set of 182 genome sequences of Leptospira strains, mostly reference strains from the Leptospira Genome project [20], was downloaded from GenBank and PATRIC (S1 Table). The total set of 509 genomes contained representatives of most Leptospira species currently described. The clusters of pathogens, intermediates and saprophytes were represented by 402, 31, and 76 genomes, respectively. Geographically, the dataset was highly diverse: strains were isolated from different geographical areas (Africa: 19, East Asia: 17, Caribbean: 13, Central America: 7, Europe: 73, Indian Ocean: 123, Middle East: 4, North America: 24, Oceania: 11, Pacific Ocean: 14, South America: 101, Southeast Asia: 97). The ecological sources of the strains were also diverse: 111 were from the environment, 226 were from humans, while the remaining isolates were from various animal hosts, such as rodents, cows, dogs, and pigs (S1 Table). The strains corresponded to 42 species including 15 novel species isolated from the environment in Japan, Mayotte, France, Malaysia, Algeria, and New Caledonia [54]. There were 26 serogroups and 73 serovars in the dataset (S1 Table). The strains selected for this study are therefore highly diverse geographically, ecologically and taxonomically.
The general features of the 509 genomes are reported in S1 Table and summarized in S1 Fig. Genomic assembly sizes ranged from 3,450,639 to 5,267,227 base pairs. Pathogens had a heterogeneous genome size, which was larger on average than the genome size of intermediates, which in turn had a larger genome than saprophytes (p < 0.001 for both comparisons). The genomic assemblies of pathogens were more fragmented (average contig number, 222) than those of the two other clusters (52 and 47 for the saprophytes and intermediates, respectively), which may reflect the high number of mobile elements in the pathogens [55]. The guanine+cytosine content (G+C%) of genomes was higher in the intermediates (42.39%) than in the saprophytes (38.27%, p < 1e-7) and in the pathogens (38.83%, p < 1e-7). Saprophytes were more homogeneous in their G+C% content than the two other clusters (S1 Fig).
To define the phylogenetic diversity of the dataset, 545 selected genes (see Methods, section cgMLST definition) were translated, aligned and then concatenated (S2 Table). The resulting phylogenetic tree is shown in S2 Fig. ANI analysis [54] revealed 42 species defined using the 95% ANI cutoff [56, 57], including 15 novel species for which a formal description was proposed elsewhere [54]. The phylogenetic tree with representatives of each species (Fig 1) is consistent with previous data [1] showing two major clades, the “saprophytes” containing species isolated in the natural environment and not responsible for infections and “pathogens” containing all the species responsible for infections in both humans and animals, plus environmental species for which the virulence status is not clearly established. This latter clade is further subdivided in two subclades that we named P1 (formerly described as the pathogen group) and P2 (formerly described as the intermediate group). Note that two strains previously assigned to the saprophytes (strains 201400974 and E30 isolated from the natural environment in Algeria and Japan, respectively) were clearly distinct from the other saprophytes and represent new species, named L. ilyithenensis and L. kobayashii, of a novel subclade within the clade of saprophytes. We named this new subclade S2 for convenience, in comparison to S1 which is constituted by species formerly described as the saprophyte group [54]. The basal position of the saprophyte clade with respect to P1 and P2 subclades is concordant with previous studies [58, 59]. The mean genetic distances among the three main subclades S1, P1 and P2 (S3 Fig) ranged between 0.33 substitutions per site (pathogens P1- intermediates P2) and 0.47 substitutions per site (intermediates P2- saprophytes S1), underlining the fact that these subclades are separated by large evolutionary distances. In contrast, mean intra-subclade genetic distances were 0.13 (saprophytes S1), 0.12 (pathogens P1) and 0.17 substitutions per site (intermediates P2), reflecting the higher heterogeneity and deeper phylogenetic branching of the intermediates P2 subclade. The distance between the new subclade S2 and saprophytes S1 was 0.29, showing that it lies close the P1-P2 inter-subclade distance.
We found that all species were monophyletic (S2 and S4 Fig). Furthermore, as expected, the intra-species distances were much lower than the inter-species. For example, L. borgpetersenii isolates formed a tight cluster with a maximum genetic divergence of 0.179 substitutions per site. Similarly, L. interrogans isolates showed high genetic relatedness, with a maximum distance of 0.033. This is remarkable given that both species are distributed worldwide (Fig 2). L. mayottensis, which is confined to the islands of Mayotte and Madagascar, showed a level of diversity of 0.008.
The phylogenetic analysis (Fig 1) revealed some structuration and led us to recognize several subgroups of species within subclades. Regarding the subclade P1, species L. interrogans, L. noguchi and L. kirschneri clustered into one subgroup (P1-1), whereas L. borgpeterseni, L. alexanderi, L. weilii, L. mayottensis, and L. santarosai formed a second subgroup (P1-2). Two other subgroups are constituted by L. alstonii (P1-3) and L. kmetyi, L. barantonii from New Caledonia [60] and L. dzianensis isolated from the environment in Mayotte (P1-4). Finally, subgroup P1-5 comprised L. adleri, L. putramalaysiae from the environment in Malaysia and L. typperaryensis. These subgroups are consistent with previous studies [20, 59, 61, 62].
To improve resolution, separate trees were constructed for the saprophytes S1 and the intermediates P2 (S4 Fig), showing the high level of genetic diversity among environmental isolates. The saprophytes were grouped into two subgroups. Subgroup 1 (S1-1) comprised L. vanthielli, L. brenneri, L. wolbachii; two new species: L. perdikensis and L. congkakensis from Malaysia; L. meyeri, L. harrisiae and the new species L. mtsangambouenesis and L. bandrabouensis isolated from Mayotte. Subgroup 2 (S1-2) comprised L. biflexa and three new species, L. bouyouniensis, L. kemamanensis, and L. jelokensis, isolated from Mayotte and Malaysia; and L. levetti and the new species L. ellinghausenii isolated from soil in Japan.
Among the intermediates P2, three subgroups were recognizable: subgroup P2-1 with L. fainei, L. broomi, and L. inadai; subgroup P2-2 with L. wolffii; and subgroup P2-3 with L. venezuelensis, L. licerasiae, L. saintgironsiae and four new species, named L. dzoumogneensis, L. johnsonii, L. selangorensis, and L. sarikeiensis, isolated from soils in Malaysia, Japan, and Mayotte (Figs 1, S2 and S4). A scheme for classifying Leptospira strains is proposed in S3 Table.
The phylogenetic structuration reflects a strong contrast between inter- and intra-species distances, which makes it possible to assign isolates at the species level based on their genome sequence-derived phylogenetic position. This led us to re-identify some isolates. For example, strain GWTS assigned to pathogen L. alstonii based on the 16S rRNA and secY genes [63, 64] did not cluster with the L. alstonii reference strain and formed a distinct branch in our phylogenetic tree (S2 Fig). Based on ANI values with representative species, including new species described in this study, it represents a new pathogenic species that we named L. tipperaryensis (S1 Table) [54]. Similarly, strains of serovar Rushan were previously identified as belonging to L. noguchi [65] but were phylogenetically clustered with L. alstonii (Figs 1 and S2) and had ANI values of 99.29% compared with the type strain of L. alstonii. These strains therefore appear to be new members of L. alstonii. Interestingly, the L. alstonii reference strain, of serovar Sichuan, was isolated from a frog [66], as were the strains from serovar Rushan, suggesting a tropism of this species for frogs.
Species of the saprophytes and intermediates subclades were represented by few strains. In contrast, some species of pathogens subclade P1 were represented by multiple isolates (e.g., 160 for L. interrogans, 76 for L. borgpetersenii, 52 for L. kirschneri, 27 for L. santarosai, 27 for L. noguchi and 23 for L. mayottensis). Based on the present sample of Leptospira genomes, the geographic distribution of these species showed clear differences (Fig 2). L. interrogans, L. borgpetersenii and L. kirschneri were found in all world regions, even though L. kirschneri appeared more rarely in Asian and American samples than in Europe and Mayotte. In contrast, in our dataset, L. santarosai was only sampled from the American continent and the Caribbean islands and L. noguchi was found predominantly in the Americas and rarely in Asia. So far, L. mayottensis has been only isolated from the Indian Ocean islands (Fig 2).
We analyzed in more details the geographic distribution of the diversity of L. interrogans, the most common Leptospira species from human infections around the world, and which was the most represented in our dataset. S5 Fig presents a phylogenetic tree of the 152 L. interrogans isolates for which the geographic source was known; these were from 32 countries in all world regions. The data reveal extensive geographical spread of L. interrogans sublineages. Although some sublineages were sampled in a single world region (e.g., the sublineage containing serovars Szwajizak, Wewak, and Hawain originated in Oceania), it is clear that most sublineages are geographically widespread (S5 Fig). This is true even for genetically homogenous subgroups, which have limited phylogenetic depth and have therefore emerged recently. These data demonstrate the rapid spread of L. interrogans sublineages over large geographic distances.
To develop a standardized subtyping strategy for Leptospira, we analyzed genome sequences using a gene-by-gene approach [34], based on the 545 genes that were highly conserved across the genus (S1 and S2 Tables). We define this set of gene loci as a core genome MLST (cgMLST) scheme [33, 34] for Leptospira; note that due to occasional absence of a few genes in some genomes, strictly speaking this set of genes is a ‘soft core genome’. The majority of cgMLST genes (527 loci per isolate on average, 96.7%) were called successfully (i.e., an allele was defined), including in the saprophytes S1 and intermediates P2 (S1 Table). The number of successfully called alleles per isolate ranged from 436 to 545 depending on the gene (S1 Table). Hence, this cgMLST scheme allows genotyping of all Leptospira genomes, with only a few missing data points.
For high-resolution subtyping, we defined cgMLST sequence types (cgST) as groups of cgMLST allelic profiles that are identical at all loci except for missing data, which are ignored in pairwise comparisons of allelic profiles (S1 Table). Considering the 509 genomes, there were 463 distinct profiles (defined by their cgST identifier, S1 Table), i.e., most genomes could be identified by a unique allelic profile. The discriminatory power of cgST classification (Simpson’s index) was 99.9%, much higher than that of MLST: for genomes that were typeable by cgMLST and the three MLST schemes [29, 45, 46], the Simpson indices of discrimination were 0.999 (confidence interval: 0.998–1.000), 0.793 (0.735–0.851), 0.787 (0.730–0.845) and 0.787 (0.730–0.845) for cgST, MLST1, MLST2 and MLST3, respectively. Hence, as expected, the use of 545 genes instead of 7 cgMLST largely improves our ability to distinguish among Leptospira isolates.
To assess the reproducibility and stability of cgST subtyping, sequencing replicates were performed for three isolates: L. licerasiae strain VAR010, L. meyeri strain Veldrat, and L. interrogans strain L495. The two replicates of the same isolate shared the same cgST, indicating high reproducibility of cgST classification. We next analyzed a culture‐attenuated strain of L. interrogans serovar Lai that had accumulated mutations (insertions, deletions, and single-nucleotide variations) in 101 genes after serial in vitro passages over several years [67]. The derived strain (cgST20) was clearly distinct from the virulent parental strain (cgST23, differing by 15 loci). Nevertheless, these subcultures were grouped together in the phylogenetic tree (S2 Fig). Similarly, a virulence-attenuated isolate of L. interrogans serovar Manilae passaged 67 times was sequenced [68] and compared with the corresponding parental virulent culture. The cgMLST analysis classified the 2 cultures as cgST31 and cgST32, differing by only 2 alleles out of 545 genes. These results illustrate the high resolutive power of cgMLST, which can distinguish genomes of isolates that evolved in-vitro over several generations.
To evaluate the genetic diversity among isolates classified into the same cgST (or groups of cgSTs differing only by missing data in some isolates), we analyzed the three most numerous ones (highlighted with colors in S1 Table, column cgST) using a whole-genome single nucleotide polymorphisms (SNP) approach. First, cgST128 and its related cgST123 and cgST308 comprised eight L. borgpetersenii isolates from Mayotte. These differed by a maximum of 16 SNPs, and five isolates had only up to 2 SNPs among themselves. Second, cgST262 and related cgSTs (cgST130, cgST321 and cgST396) comprised 11 isolates, also of L. borgpetersenii from Mayotte. These isolates differed among themselves by up to 23 SNPs. Finally, cgST482 and related cgST484 comprised seven L. interrogans isolates from cows in Uruguay; all these isolates were identical (no SNP) except one, which differed by only three SNPs from the others. These results show that isolates sharing the same cgST, or cgSTs that are identical except for missing data, are very closely related also based on whole-genome SNPs, and include levels of whole-genome SNPs that are compatible with the isolates being part of recent chains of transmissions [69, 70].
To define groups of Leptospira strains based on cgMLST, we first explored the distribution of pairwise distances among all cgMLST profiles (S6 Fig). We also evaluated the quality of clustering, using the Silhouette index [71], resulting from the use of all possible threshold values (from 1 to 544) in single-linkage clustering (S7 Fig), revealing a plateau of maximal clustering quality between 40 and 300 allelic mismatches. Based on the above analyses, a threshold of 40 allelic differences was chosen as the cut-off value to define clonal groups (CG). In other words, a CG is defined as a group of cgMLST allelic profiles differing by no more than 40 allelic mismatches, out of 545 gene loci, from at least one other member of the group. This definition resulted in the identification of 237 CGs (S1 Table). To evaluate this choice as compared to alternative thresholds, we compared using the adjusted rand coefficient [72] the partitions (i.e., groups of isolates classified into the same CG) obtained using thresholds of 20, 30, 50, 60, 150, 200 and 300 mismatches (S7 Fig). Interestingly, confidence intervals overlapped with those of threshold 40 within a wide range of possible cutoff values (20 to 150). Hence, a choice of alternative thresholds in that range would have a limited impact on the resulting clusters. Finally, the effect of missing data (uncalled cgMLST alleles) on the clustering results was evaluated in-silico by introducing increasing amounts of missing data and assessing the resulting clusters of isolates as compared to their initial cluster (S8 Fig). This simulation showed that cluster assignment is robust to even high amounts of missing data (affecting up to 400 loci out of 545).
The clusters created at the 40-mismatch level represent a potentially useful genome-based taxonomy of Leptospira strains. To evaluate this classification system in comparison with previous Leptospira strain classifications, we first compared them to the 6- or 7-gene MLST classifications currently in use [29, 45, 46]. The three MLST classifications (S1 Table) were mapped onto the phylogenetic tree and their concordance with cgMLST was analyzed (Fig 3). A total of 260, 106, and 143 Leptospira STs are currently defined for MLST schemes 1, 2, and 3, respectively (April 2018; https://pubmlst.org/leptospira/). These MLST schemes were developed for strain typing of the main pathogenic Leptospira species but not for the saprophytes and intermediates [29, 30, 45, 46, 73]. As expected, saprophytes and most intermediates were not typeable by the three classical MLST schemes, whereas by design, all strains were typeable by cgMLST (Fig 3). Therefore, the typeability of the proposed cgMLST scheme appears greatly enhanced as compared with classical MLST.
We also assessed the concordance among assignments produced by the three MLST schemes and the cgMLST clustering into CGs, using Sankey diagrams (S9 Fig) and adjusted Rand and Wallace coefficients [72]. The adjusted Rand index of concordance of MLST with cgMLST was 0.86, 0.89 and 0.89 for MLST1 [45], MLST2 [46] and MLST3 [29], respectively. Wallace indices are not symmetrical, and thus produce two values: one for the comparison of MLST versus cgMLST clustering (i.e., how well MLST identity predicts CG identity), and one for the reciprocal comparison. The results were 0.86 and 0.86 for MLST1, 0.82 and 0.97 for MLST2, and 0.83 and 0.96 for MLST3. Hence, the CG accurately predicts with high accuracy the STs of MLST2 and MLST3. Only 4, 1 and 2 cgMLST clusters matched more than one MLST ST for scheme 1, 2 and 3, respectively (S9 Fig). Reciprocally, 26, 9 and 13 STs for MLST1, MLST2 and MLST3, respectively, were subdivided into more than one CG. In other words, despite accepting 40 mismatches within members of the groups, CG classification is still more discriminatory than each of the classical MLST systems. Note that although the low- and high-passage strains (see above) of L. interrogans serovar Lai and L. interrogans serovar Manilae were distinguishable at the level of cgST subtypes, they were classified into the same CG (CG16 and CG23, respectively), consistent with their recent evolutionary link.
To provide access to the cgMLST allele and profiles nomenclature, allowing for comparison and sharing of typing results among laboratories worldwide, a database was set-up and was made publicly accessible online (https://bigsdb.pasteur.fr/leptospira/). This database is based on the software framework Bacterial Isolate Genome Sequence Database (BIGSdb) [33, 34, 74].
The distribution of serovars and serogroups along the phylogeny showed that most serogroups had a polyphyletic distribution. The fact that phylogenies can be in disagreement with serotyping was previously reported, and some serovars or related serovars from a same serogroup may belong to different species [21]. Thus, isolates from the same serogroup can be distributed in different species or sublineages within species. For example, L. interrogans strains of serogroup Australis or of serogroup Pyrogenes did not all cluster together in the phylogenetic tree (S2 Fig).
We investigated the correspondence of cgMLST groups with serovars. Serogroups (sg) were usually sub-divided into several CGs (S1 Table). For example, the 29 isolates of sg Australis were subdivided into 14 CGs, the 42 isolates of sg Grippotyphosa fell into 16 CGs, and the 20 isolates of sg Pyrogenes were grouped into 12 CGs. At the serovar level, highly related strains belonged to the same clonal group (S1 Table). This was the case for the 19 isolates from serovars Copenhageni and Icterohaemorrhagiae, which were clustered together in CG6, and for serovars Ratnupura and Vanderhoedeni (CG185, L. kirschneri sg Grippotyphosa) and Bajan and Barbudensis (CG179, L. noguchii sg Australis). However, some serovars were genetically more heterogeneous and were themselves sub-divided into different cgMLST clonal groups (e.g. L. kirschneri and L. interrogans sv Grippotyphosa: 6 CGs; L. interrogans sv Lai, 3 CGs; L. interrogans sv Pyrogenes: 8 CGs) (S1 Table). Therefore, cgMLST groups represent a useful classification system that is genome sequence-based and is complementary to serogroup and serovar classification, which are based on surface antigens.
To explore the links between cgMLST classifications and the epidemiology of Leptospira strains, we first analyzed the correspondence of cgMLST groups with hosts. It is well established that serovars are usually associated with a specific animal reservoir; for example, rats usually carry serovars of the Icterohaemorrhagiae serogroup; and serovar Canicola is associated with dogs [21]. Here, the most frequent cgMLST clonal groups of subclade P1 contained isolates obtained from both human and animals (except in Mayotte where few isolates have been isolated from animals). Thus, isolates of L. interrogans sg Pyrogenes (CG23), L. borgpetersenii sg Ballum (CG15) and L. borgpetersenii sg Javanica (CG25), associated with human leptospirosis, were clustered by cgMLST with rodent isolates, suggesting that these serogroups are maintained in rodents and that these animals represent reservoirs of human infections (S1 Table). Similarly, CG19 corresponding to serovar Sejroe comprised human and cattle isolates (S1 Table). Some CGs were found in an even larger range of hosts. For example, the 37 isolates belonging to CG5 (serovar Pomona) were obtained from humans, dogs and cows from seven countries. Likewise, CG28 contained isolates from dogs, rodents, pigs, and humans, indicating that some CGs or serotypes are not always restricted to specific hosts and may have a more generalist ecology. The environmental strains from our study were usually not grouped with animal or human isolates, as they formed distincts CGs.
We next analyzed 90 clinical isolates collected in the island of Mayotte (Indian Ocean) over a period of 10 years (2007–2017). cgMLST separated them into 10 CGs, which were highly congruent with their serotypes and species (S1 Table). Serogroup Mini was predominant (60%) and subdivided into five CGs, which agreed with their species assignments (CG63, CG83 and CG84 for L. kirschneri, CG78 for L. borgpetersenii, and CG79 for L. mayottensis). The most frequent CG was CG78, corresponding to 39 isolates, which were distributed into 25 cgSTs and were isolated over the 10-year period. Isolates belonging to L. mayottensis were sub-divided into two CGs, CG79 (n = 7, 5 cgSTs) and CG82 (n = 16, 14 cgSTs) (S1 Table). These two groups were previously recognized by PFGE, MLST and serotyping [37, 75]. Isolates from the island of Mayotte belonged to cgMLST groups that were not found in other world regions, consistent with the unique epidemiology of leptospirosis in this insular ecosystem [37, 39]. In contrast, multiple CGs were observed in different geographical locations around the world (S1 Table). The wide geographic distribution of CGs indicates that geographic spread of Leptospira strains is faster than their genetic evolution into distinct CGs.
We next analyzed the geographic distribution of the high-resolution cgMLST types (cgST). One of the most represented cgSTs (cgST482) in our dataset is constituted by L. interrogans serovar Pomona strains (n = 6) isolated from cattle in Uruguay [76] (S1 Table). Although five out of six of these strains have been isolated from the same farm and were undistinguishable by SNP analysis, one isolate from another region of the country differed from the group of five isolates by 3 SNPs. This shows that cgST classification could possibly inform on the epidemiological links among Leptospira isolates.
Until now, a consensus approach to characterize and compare Leptospira isolates has been lacking, limiting our understanding of the biology and epidemiology of strains within this important genus and impeding progress in establishing appropriate control and prevention measures. Advanced knowledge on the diversity and distribution of Leptospira strains is also essential for the design and evaluation of the efficacy of new vaccines and diagnostic tools.
This study lays a foundation for a comprehensive understanding of the biodiversity of Leptospira and for the epidemiological surveillance of medically important Leptospira pathogens. The availability of high-throughput sequencing technologies and the reduction of their costs makes genome sequencing a viable option as the new gold standard for Leptospira genotyping and taxonomy. Recently, 14 new species were identified based on genomic comparisons and a high degree of biodiversity of Leptospira species in soils and water was recently uncovered [6, 17, 75]. Besides, there is growing evidence that “intermediate” species are responsible for mild infections in humans [6, 8, 11–17, 19, 77, 78]. Novel genotyping methods should therefore encompass the entire genus, including both potentially pathogenic and non-pathogenic strains, in order to provide universal Leptospira strain characterization systems.
The classical MLST schemes were developped using six or seven genes with a focus on pathogenic Leptospira species [29, 45, 73]. More recently, a new MLST scheme was proposed and applied to a wider collection of strains, including a few intermediate species [46, 62]. However, none of these MLST methods enables the inclusion of all major Leptospira lineages, including saprophytic strains. Here we sought to develop a cgMLST strategy, which is an extension of conventional MLST at genome scale [34]. Our comparative genome analyses resulted in the identification of 764 genus-wide core genes, including 545 that were deemed suitable for use in cgMLST genotyping. This is in accordance with previous estimates of 700 to 1,000 Leptopira core genes [6, 20, 59]. Importantly, our cgMLST scheme was developed using genomes representing the entire breadth of the phylogenetic diversity of the genus and was validated using Leptospira strains from diverse sources and geographical locations.
The cgMLST scheme was used to construct amino-acid sequence-based phylogenetic trees that were consistent with previous work and current species designations. In addition, this work revealed the existence of novel Leptospira species isolated from soils and water across a wide geographic range (Algeria, Mayotte, Japan, New Caledonia and Malaysia), including species from the new subclade S2 that is phylogenetically related to the previously known saprophytes S1. This work confirms the high diversity of Leptospira species in the natural environment [6, 60], and the novel taxa were described more formally elsewhere [54]. Further, cgMLST-based phylogenetic analysis provides high-level resolution, allowing discrimination among closely related species and strains.
Much like classical MLST data, cgMLST data can be used to devise a classification of isolates using the single linkage algorithm [79]. Here we defined clonal groups based on cgMLST with a 40 allelic mismatches cut-off value. In order to optimize discrimination among groups, this threshold was chosen as the smallest threshold within the range of thresholds that maximized the quality of clustering. We demonstrated the robustness of CG classification to missing data and to threshold choice, and therefore propose that CG identifiers will become a practical and highly stable genomic taxonomy system for Leptospira strains. However, it must be underlined that clonal groups are broad classification categories that are of limited use for transmission studies, as illustrated by the wide geographical and temporal distribution of isolates from single clonal groups. Isolates belonging to the same clonal group always belonged to the same serogroup. Conversely, strains of a given serogroup can fall into phylogenetically unrelated clonal groups, suggesting that some Leptospira serogroups are derived from multiple independent ancestors. Further, strains belonging to the same serovar were not always clustered together by cgMLST, indicating that serovars can also be polyphyletic. In contrast, genetically related serovars were sometimes conflated by cgMLST clustering. These observations underline the complementarity of cgMLST clonal groups with previous classifications based on serotyping. cgMLST allows assigning Leptospira isolates both at the species and serogroup levels, and in most cases at the serovar level as well. With the increasing description of novel species and the continuous recording of strain diversity within species by surveillance networks and microbiology laboratories, a precise understanding of the biodiversity of Leptospira strains is needed. cgMLST might represent a useful standard for classification and nomenclature, and would advantageously replace the current classical MLST nomenclatures, which are incomplete, and the serotyping nomenclature, which is complex and does not always reflects phylogenetic relationships, as is the case for other pathogens [80].
Although many CGs were found in distinct geographic regions, the island of Mayotte was a notable exception in that its CGs were endemic. The lack of dissemination of CGs from Mayotte, or of colonization of Mayotte by cosmopolitan CGs such as those of Icterohaemorrhagiae, illustrates the unique ecosystem of this island [81]. However, whether the distribution of species or CGs reported here reflects strong endemicity, or is due to currently limited sampling, will be subject of future studies. As an example of our sampling limitations, L. santarosai is not only found in America as shown in Fig 2 but also in Taiwan where this species is the most frequently encountered species in patients [82]. Isolation of additional strains from both humans and animals will also be required to evaluate whether or not environmental strains belonging to subclades P1 and P2 have the ability to cause infections.
We propose a high-resolution classification of Leptospira strains into cgSTs, which correspond to groups of isolates with total sequence identity at the 545 cgMLST genes, with a tolerance of missing data. We showed that this level of discrimination is able to distinguish among in-vitro evolved cultures. Due to the occurrence of missing data, the cgMLST profiles of some isolates can match several distinct cgSTs. Isolates with identical cgST or belonging to groups of related cgSTs (defined as matching single isolates’ profiles) were shown to differ at the whole-genome scale by less than ~30 SNPs. This level of divergence is indicative that they share a very recent common ancestor and might be part of an ongoing transmission chain [69, 70], even though genomic epidemiology applications to Leptospira remain to be evaluated taking into account its specific mutation rate and transmission dynamics.
L. interrogans and L. borgpetersenii are ubiquitous pathogenic species. This is probably due to the fact that rodents are major reservoir hosts for these species [45]. Thus, L. interrogans strains belonging to serovars Copenhageni and Icterohaemorrhagiae share the same CG regardless of their geographic origin. This limited genetic diversity and broad geographic distribution (S5 Fig) is consistent with recent evolution/expansion following extensive migration of rodents, the main reservoir of serovars Copenhageni and Icterohaemorrhagiae, and multiple introductions due to modern global transport, in particular long-range, ship-based travel and trade. Due to this rapid geographic diffusion, little phylogeographic signal was present in the dataset, rendering challenging the reconstruction of the geographic origins of L. interrogans and its sublineages with confidence. By contrast, species such as L. noguchi, L. kirschneri, and L. mayottensis are not associated with rats and are largely confined in specific geographical areas. The pathogen L. mayottensis may have been introduced into Mayotte from Madagascar via the tenrec, a small terrestrial mammal [83].
This work provides a framework for the definition of Leptospira clades, subclades, subgroups, species, as well as strains at two levels of resolution (S3 Table). The possibility for laboratories around the world to identify the same strains using a unified nomenclature and a centralised genotyping database will facilitate the sharing and dissemination of knowledge on circulating Leptospira genotypes, worldwide. The cgMLST scheme will also enable early detection of new genotypes being introduced into locations where they are not usually found. The links between genotypes and their pathogenic potential and virulence will be an important subject for future studies. For yet unknown reasons, a limited number of Leptospira serovars are much more likely to cause severe disease than others [84–87]. The role of phages, plasmids, and horizontal transfer in the acquisition of virulence factors also remains to be determined. The molecular basis of host specificity is also largely unknown. Future dedicated studies will be needed to characterize the gene content of subclades, species and strains, and their association with the clinical presentation and outcome of Leptospira infections.
|
10.1371/journal.pgen.1007306 | One for all and all for One: Improving replication of genetic studies through network diffusion | Improving accuracy in genetic studies would greatly accelerate understanding the genetic basis of complex diseases. One approach to achieve such an improvement for risk variants identified by the genome wide association study (GWAS) approach is to incorporate previously known biology when screening variants across the genome. We developed a simple approach for improving the prioritization of candidate disease genes that incorporates a network diffusion of scores from known disease genes using a protein network and a novel integration with GWAS risk scores, and tested this approach on a large Alzheimer disease (AD) GWAS dataset. Using a statistical bootstrap approach, we cross-validated the method and for the first time showed that a network approach improves the expected replication rates in GWAS studies. Several novel AD genes were predicted including CR2, SHARPIN, and PTPN2. Our re-prioritized results are enriched for established known AD-associated biological pathways including inflammation, immune response, and metabolism, whereas standard non-prioritized results were not. Our findings support a strategy of considering network information when investigating genetic risk factors.
| Integrating multiple types of -omics data is a rapidly growing research area due in part to the increasing amount of diverse and publicly accessible data. In this study, we demonstrated that integration of genetic association and protein interaction data using a network diffusion approach measurably improves reproducibility of top candidate genes. Application of this approach to Alzheimer disease (AD) using a large dataset assembled by the Alzheimer’s Disease Genetics Consortium identified several novel candidate AD genes that are supported by pre-existing knowledge of AD pathobiology. Our findings support a strategy of considering network information when investigating genetic risk factors. Finally, we developed a transparent and easy-to-use R package that can facilitate the extension of our methodology to other phenotypes for which genetic data are available.
| The discovery of disease-associated genomic variation has numerous clinical and scientific applications, including earlier disease prognosis, improved understanding of disease pathophysiology, and development of personalized treatment therapies [1]. A commonly used technique for identifying these mutations is the genome wide association study (GWAS) approach [2]. Typically, a large sample of affected and unaffected individuals are genotyped for many single nucleotide polymorphisms (SNPs) using a high-density microarray chip and then test statistically if the allele frequency of each variant is associated with disease status [2]. Significant associations in this first step (“discovery phase”) are deemed to be robust if they replicate in an independent cohort (“replication phase”). In this study, we focused on improving the replicability of GWAS results for Alzheimer disease (AD), although our methodology is applicable to genetic data for other diseases and traits. AD is a neurodegenerative disease resulting in irreversible dementia and memory loss with elevated prevalence in older populations [3]. Recent estimates suggest that approximately 5.4 million Americans have AD, and the number of cases of AD is expected to increase dramatically in future years if medical advances continue to improve life expectancy, thereby allowing more individuals to reach ages where AD is on the rise [3].
Genetic studies of AD have led to identifying numerous AD associated genes such as APP [4], PSEN1 [5], and PSEN2 [6] for early onset AD (EOAD), as well as APOE [7, 8] and SORL1 [8, 9] for late onset AD (LOAD). Common variants in more than 20 other genes have been robustly associated with AD risk [8]. However, not all AD associated genes will reach genome wide significance in current datasets of sample sizes below 100,000 individuals. It is well recognized that incorporating other forms of biological data improves confidence in genetic findings [10–12].
Our computational framework is based on the following biological hypothesis. If a known AD variant is associated with a gene that is involved in a particular biological process (BP) (e.g. inflammation), we assume as a probabilistic prior that other AD variants might be associated with proteins involved in this BP or proteins that physically interact with this BP. This hypothesis can be tested computationally using a protein interaction network [13–15] by extending the “guilt by association” principle via propagation of probabilistic evidence in a network [16, 17]. This general idea has similarity to the Google ranking algorithm of web pages, in which a web page that has a short link distance to many “important” pages will itself be considered “important”.
In the case of protein interactions, guilt by association-based inference is typically performed by inspecting the function of direct neighbors of a predicted disease gene in a protein-interaction network. This approach has been incorporated in multiple interpretation systems as well as commercially such as Ingenuity Pathway Analysis (IPA). However, it has been shown that network propagation, diffusion or other related methods that go beyond simple neighbor-based analysis can carry functional or disease associations further in the network with improved predictive accuracies [10, 11]. This idea extends to predicting both gene function and disease phenotypes associated with genes [11, 18–22].
We hypothesize that this general framework, and network diffusion in particular, can be extended to aid prioritization of AD genes. Although the underlying biology of AD may be far more diverse than a single function, there are several biological pathways that are aberrantly activated in AD brains, and not surprisingly, most of the genes identified by AD GWAS contribute to these pathways [23]. For example, a primary indicator of AD is the accumulation of amyloid beta plaques in the brain, resulting from mis-processing of APP protein [23].
We developed a novel re-prioritization approach that can be integrated easily into the current genetic analysis design (Fig 1). First, we curated the AD literature to produce a set of approximately 60 robust AD (RAD) genes that includes those that have been associated with AD at the genome-wide significance level or that contain variants shown to affect AD-related processes directly (Table 1). We then constructed a network of protein-protein interactions and applied network diffusion to score and rank genes based on their proximity to the RAD genes. Network diffusion allows modeling of indirect interactions, modules and protein complexes that are not modeled if only the direct interactions of proteins are considered. Next, we combined our genetic association results with the network diffusion scores to produce a newly re-prioritized ranking of genes. Finally, we validated our methodology using a novel approach involving bootstrap aggregation on one of the largest assembled genetic datasets of AD. Network-augmented genetic results have measurably improved replication rates in this validation approach. We also show that our main results and key predictions were essentially unchanged after restricting the RAD set to 19 genes which have had been functionally validated as well as replicated in independent datasets.
We assembled a PPI network using interactions pooled from multiple PPI databases (ConsensusPathDB [13], iRefIndex [14], and Human Interactome Y2H [15]) inspired by recent work [21]. Pooling interactions from these three databases resulted in a connected network that includes a large percentage of the genes in our GWAS dataset. We then determined if the RAD genes are proximal within this network. The first proximity measure tested was the average shortest path (ASP) distance [50]. The ASP distance between RAD genes, determined by cross-validation (See Methods), is much smaller than would be expected by random chance (Table 2). One problem is that ASP distance between RAD genes and genes with many interactions (the number of interactions a gene has corresponds to its “degree” and high degree genes are considered to be hubs) tends to be small (Table 3). In this situation, all hub genes will be falsely predicted to be AD-related. Thus, we incorporated instead the Regularized Laplacian diffusion kernel [51] which penalizes paths going through hubs. The diffusion distance between RAD genes is smaller than would be expected by chance (p = 0.00054) (Table 2). Simultaneously, the problematic hub genes in the network have discounted scores as demonstrated by the notable drop in ranking of the 10 genes with the highest number of overall interactions (Table 3).
We next tested if genes with high diffusion scores replicate more frequently in order to demonstrate that diffusion scores are informative when used in conjunction with genetic data. Bootstrap aggregation [52] was applied to our genetic dataset to produce a large number of pairs of discovery and replication datasets (See Methods). In each discovery + replication pair, we conducted a standard genetic workflow, beginning with a screen in the discovery dataset followed by validating top findings in the replication dataset. For each pair, a replication rate was calculated by determining the percentage of genes that surpass a given significance threshold also replicated. To test if network diffusion scores improved replication, we altered the standard discover + replication approach. We ranked genes by their network diffusion score and then iteratively dropped genes that had ranking diffusion scores below a given stringency threshold. At first we retained only genes in the 50th percentile of network scores, then gradually increased the threshold to only include genes in the 60th, 70th, 80th, and 90th percentiles. For each threshold, we computed the replication rate and compared to the baseline. As shown in Fig 2, filtering based upon network score percentile noticeably increased replication rate. Genes with a–log(p-value) of > 6 replicated at a rate of approximately 16% in simulations (farthest right purple point), while additional strict network filtering improved the replication rate to nearly 34% (farthest right red point).
Since filtering on network diffusion score improved replication rate, we next sought to integrate the network diffusion scores and genetic results into a single score. First, we converted the p-value of each gene from genetic analysis into a Z-score (“GWAS Z-Scores”) and then converted the network diffusion percentile of each gene into a Z-score (“Network Z-scores”). Linear regression analysis showed that the Network and GWAS Z-scores are independent (Fig 3A). Next, we assigned each gene a replication rate based upon how frequently the gene replicated in our bootstrapped validation datasets (See Methods). We observed that replication rates were higher for genes with higher network Z-scores compared to genes with lower network Z-scores (Fig 3B).
To combine the Network and GWAS Z-scores, we developed an approach that uses a linear support vector machine (SVM) [53] to determine how heavily each type of score should be weighted in order to maximize replication rate (See Methods). These weights were then used in conjunction with the meta-analysis method for combining summary results implemented in METAL [54]. The weights predicted by the SVM (Fig 4) were 0.703 (GWAS) and 0.297 (Network). As further confirmation, we conducted binomial (logit family) logistic regression using network and GWAS Z-scores as predictors and the replication class (high/low) as the outcome. Both network and GWAS score were significant, (GWAS: coefficient = -0.659, p <2.0×10−16) (Network: coefficient = -0.229, p = 0.0016). The coefficients derived from logistic regression are very similar to the SVM-derived weights (GWAS weight = 0.742, Network weight = 0.258).
Next, we applied our combined approach genome-wide, excluding the RAD genes and genes containing significantly associated variants (p <1.0x10-7) to focus on novel candidates. Among the genes with largest combined Z-scores (Table 4, S1 Table), several have important roles in inflammation. CR2 (p = 5.95×10−7) is a receptor protein involved in immune response (genecards.com [55]). SHARPIN (p = 1.43×10−5) is a component of the LUBAC complex that plays a regulatory role in inflammation [55]. PTPN2 (p = 3.21×10−5) is a phosphatase that also serves an important role in regulation of inflammation and glucose homeostasis [55]. The Bonferroni-corrected significance threshold when considering only genes in the 75th percentile of network scores is p = 1.46 x 10−5, although this is likely to be overly strict since proximally located genes are not inherited independently.
We performed pathway analysis using Gene Set Enrichment Analysis (GSEA) [56] to determine if AD-related pathways are more enriched when genes are ranked by their combined Z-scores versus GWAS-only Z-scores (See Methods). Notably, ranking genes based upon combined Z-scores resulted in several significantly enriched AD-related pathways including immune response, FOX03 targeting (indicates enrichment for aging), and hippocampal development (Table 5). By comparison, ranking genes based only upon their GWAS Z-scores resulted in virtually no significant pathways entirely (Table 6).
GWAS of AD and AD-related endophenotypes have discovered and replicated associations with more than 60 genes (Table 1), many of which have roles in AD-related pathways (amyloid β aggregation, inflammation, cholesterol transport, immune response, etc.). To identify additional AD-related genes, we hypothesized that genes having suggestive evidence for association from a genome-wide screen and protein-level interactions (both direct and indirect) are more likely to replicate. This idea has been referred to as functional linkage [57]. To test this hypothesis, we developed a novel approach for improving the prioritization of candidate disease genes that incorporates a network diffusion of scores from known disease genes using a protein network and integration with GWAS risk scores. We tested this approach on a large AD GWAS dataset and validated the performance of the methodology using bootstrap aggregation. Several novel AD genes were predicted including CR2, SHARPIN, and PTPN2.
Part of the motivation for our approach was to identify genes that are more obviously biologically relevant to AD. This is exemplified by SHARPIN, whose principal known function is to form the LUBAC complex and prevent inflammation, a major process through which amyloid aggregation and AD are thought to develop [23]. Similarly, CR2, a homolog of CR1 which is a well-established AD gene [8], is involved in immune response. Many immune response genes are differentially expressed between healthy and AD brains, and investigations into the connection between expression in cell types and the presence of AD has led to growing interest in the role microglial cells (a first responder in the immune response pathway) [58]. Finally, PTPN2 is involved in multiple AD-related pathways; it has roles in negatively regulating inflammation and de-phosphorylation of key glucose metabolism kinases including INSR and EGFR [59]. The AD-related roles of each of our novel AD gene predictions, in combination with their strong network and genetic scores, make them highly promising candidates.
One biological form of functional linkage that does not require direct physical interaction is membership in the same signaling pathway or protein complex. For example, our study identified interaction between FOXO and INSR that is consistent with evidence of a multi-link signaling pathway comprised of direct physical interactions in the insulin-signaling pathway [60]. By comparison, neighborhood enrichment approaches (i.e., testing a gene’s direct interactions) cannot detect indirect interactions. Furthermore, neighborhood enrichment approaches are unreasonable for AD because some RAD genes are network hubs (e.g., APP has more than 2000 interactions) which would result in an unreasonably high number of genes having AD-enriched neighborhoods.
Some distance metrics capture indirect interactions by calculating the proximity between a pair of genes based upon short paths between them in the network. However, after testing a simple distance metric known as average shortest path (ASP), we observed that hub genes were still the top-ranked predicted genes. Since hub genes have many interactions, they tend to have short overall paths to any genes in a network, although their functions are highly generic and unlikely tied to a particular disease. Ubiquitin C (UBC), for example, has nearly 9,000 interactions; however, this is simply because protein degradation is essential for regulating the vast majority of proteins. Therefore, a more nuanced network propagation approach can aid in making disease specific inferences.
Network diffusion is a widely used class of spectral graph clustering methods that have been applied to many computational disciplines [51]. We used this approach to propagate evidence in the form of AD scores throughout the network. A protein in the network that has a short “diffusion distance” to one or more well-established AD genes will receive a high network risk score. Notably, we observed that network diffusion down-weights hubs while simultaneously outperforming ASP distance when applying leave-one-out cross-validation to the RAD genes. Many diffusion kernels have been proposed in graph theory, however the Regularized Laplacian [51] approach used in this study has the highly desirable properties of requiring very little parameterization (in fact, only a single parameter is required to be set) and also more computationally efficient than other diffusion kernels. Network diffusion methods have been applied in other genetics research contexts such as labeling somatic network mutations in cancer [61], characterizing gene sets [62], and predicting risk genes for amyotrophic lateral sclerosis [21].
We also observed that genes with high diffusion scores tended to replicate more frequently in our 125 pairs of bootstrapped discovery and replication datasets. However, network Z-scores and GWAS Z-scores in the full dataset were not strongly correlated. Taken together, these observations indicate the importance of considering jointly protein interaction data and genetic results even though they are independent because the integration of both types of information will likely yield noticeable improvement in replicability of findings. Since our bootstrapping procedure required splitting the original dataset, the simulations were conducted using datasets that contained only one-half of the total sample. This suggests that our network scores aided in determining which genetic associations were real in datasets with reduced power. We note that our bootstrapping approach was performed on the same data from which we derived the GWAS Z-scores used to train the SVM. Therefore, the selection of combination weights may have been biased in favor of GWAS Z-scores. Furthermore, it is unclear whether the weight combination used in this study (0.297/0.703) would be appropriate for combining genetic and network data for other disorders or traits.
The GWAS approach has a very limited capability to identify the entire set of genes which contribute to the risk of a complex disease like AD, even in datasets containing up to 100,000 individuals, because some genes do not contain variants that are sufficiently frequent and/or exert a large enough effect to yield a statistically significant association. To overcome this limitation, we developed a novel SVM approach to integrate the genetic and network scores by propagating GWAS Z-scores in a PPI network. In the AD example presented here, we initialized the RAD genes to have an identical high score in the network, thereby allowing re-prioritization of genes in any AD dataset regardless of the internal Z-scores of the RAD genes.
We acknowledge that our initial choice to treat each RAD gene equally may be controversial. Arguably, we could have seeded our analyses with GWAS Z-scores for each RAD gene from the original studies. However, our approach permits unbiased exploration of interactions of all plausible AD genes and does not require adjustment to these Z-scores for sample size or allele frequencies. Moreover, results derived from weighted RAD genes would be dominated by interactions with APOE for which the significance level exceeded a–log(p-value) of more than 100 in several datasets (compared to < 10 for most other RAD genes in the total group of datasets). Also, several key AD-related genes (e.g., APP, PSEN1 and PSEN2) which show little evidence for association with individual SNP or gene-based tests for AD would be undervalued in analyses using weighted Z-scores. In order to make our software maximally flexible and support weights derived from confidence in the seed genes, we implemented an option for users to specify unequal weights on the seed genes at their own discretion.
A potential concern about our results is the strategy for selecting RAD genes because many significant GWAS findings include variants located in intergenic regions. The most parsimonious explanation is that the variant responsible for the association peak influences the nearest gene, but there is abundant evidence suggesting this assumption is often incorrect. To address this issue, we repeated our analyses using a more restricted set of RAD genes that included only those supported by genome-wide significant evidence of association with AD risk and replication in independent datasets or by other genetic evidence plus experiments linking them to AD-related pathophysiology. Our leave-one-out cross validation approach demonstrated that the genes in the restricted RAD set had closer network proximity to each other than would be expected by chance (p = 5.93x10-5, S2 Table). The statistical support for the novel genes CR2 (p = 4.09x10-7), SHARPIN (p = 1.10x10-5), and PTPN2 (p = 2.41x10-5) remained the same (S3 Table). Finally, combined Z-scores that were derived using diffusion from the more conservative RAD gene set yielded similar AD-related pathways such as Fx03 targets (FWER p = 0.064), antigen processing (FWER p = 0.02), and hippocampal development (FWER p = 0.065) (S4 Table). These results confirm that the genes with a clear functional role in AD produce network diffusion-based predictions that are consistent with the results presented here. Curiously, the inclusion or exclusion of the portion of RAD genes that have an ambiguous or non- validated functional role in AD did not affect our results.
We also acknowledge that several of the novel putative AD genes may have been erroneously prioritized because they are in the same locus with RAD genes. This concern is unlikely noting that there are several instances where a genetic association peak includes multiple genes that may have a possible functional role in AD (e.g., the MS4A gene cluster [8]). Although one of our novel AD genes, CR2, is located close to CR1, which is an unambiguous RAD gene given its robust replication in GWAS and effect on deposition of neuritic amyloid plaque [63], CR2 is also an intriguing AD candidate gene because it has been shown to regulate hippocampal neurogenesis [63]. Thus, our findings suggest that our approach will aid in predicting truly multiple AD-related genes at a locus, however additional biological evidence may be required in some instances to make this distinction.
Previous AD studies have implicated inflammation and immune response genes, but we did not observe enrichment for these pathways when incorporating only GWAS scores in the analysis. However, these and other recognized AD-related pathways emerged after applying our network re-prioritization method (Table 6) suggesting that incorporation of network data can help minimize discrepancies in predictions across different genetic datasets. On the other hand, other well-established AD-related pathways, including cholesterol metabolism and endocytosis, were not detected by our approach. Further inspection of the results revealed, for example, that enrichment for the cholesterol homeostasis pathway is not significant when applying GSEA to the genetic data only (FWER p = 1). This pathway as defined in the Molecular Signatures Database (MSigDB) is very broad and contains many genes that are weakly associated with AD which consequently diminish the enrichment of the set. The evidence for this pathway is greater in the analysis using only network scores (FWER p = 0.18), which indicates our method still improves the detection of cholesterol homeostasis. Even pathways such as HDL-mediated lipid transport that were enriched in analyses considering only genetic data (largely due to the strong signal from APOE) were not ranked highly by our network diffusion algorithm because RAD genes such as APOE are ignored to minimize bias.
Although merging of multiple databases to obtain a very highly connected network is a requirement for the diffusion algorithm to work properly, our approach offers several advantages in comparison to other network-based approaches including biological transparency, ease of integration with a variety of GWAS methods, and the ability to balance data-driven statistics and biological prior probabilities. The extensive simulations we conducted provide a general basis for further establishing the practicality of genetic and network-based integration. Our network methodology was developed with the goal of accommodating known complications of genetic analysis.
The software developed for this study is open source, accessible to most users (incorporated in an R package), and applicable to any set of variant- or gene-level disease association results. Importantly, it requires only a set of GWAS results and a list of previously known disease genes and, therefore, does not necessitate changes to previously established genetic analysis pipelines. Although we used an SVM procedure to determine the weights for the score combination, a user can specify any weights or simply use our defaults that are based on the 0.297/0.703 ratio determined by SVM. Our package is accessible through GitHub (https://github.com/lancour/ignition).
A set of genes ascribed to AD with a high degree of certainty was assembled through curation of published findings ascertained through PubMed searches that emerged from studies using a variety of approaches including GWAS of AD risk and AD-related endophenotypes, family-based linkage analysis, positional cloning, whole exome sequencing (WES), and candidate gene testing (CGS) (Table 1). Criteria for inclusion in this set included (1) genome-wide significance for GWAS and WES studies (p < 5x10-8) and LOD score > 3 for linkage studies and (2) replication of association signals in independent datasets; or (3) biological evidence that demonstrate functional relevance to AD of associated variants or the encoded protein.
A set of interacting gene-gene pairs (in HGNC symbol format) is required as input for this software. To compile this set, three databases (RefIndex v14 [14], ConsensusPathDB v31 [13], and Human Interactome Y2H DB vHI-II-14 [15]) were selected based on their demonstrated utility in recent work [21]. iREFINDEX and ConsensusPathDB interactions were filtered to remove self and complex (more than two proteins) interactions. The ConsensusPathDB interactions are given in uniProt ID format, which were converted to HGNC symbols using the official website (http://www.genenames.org). iREFINDEX provides a HGNC symbol for each interactor of an interaction when possible, and so only interactions which had a HGNC for both interactors were kept. The Human Interactome DB already provides a set of binary gene-gene interactions in HGNC format, so no processing was required. The union of the processed sets from each database was used as the final interaction set. The unified set contains 19,972 unique gene symbols and 236,642 interactions. These databases are curated collections of experimentally determined interactions (typically binding or affinity) reported in the literature, such as from co-immunoprecipitation, as well as predicted interactions in a small number of databases.
Network diffusion is a very well-studied spectral approach to graph clustering and annotation [17, 51, 64, 65]. It attempts to mimic node-to-node distance in the graph that in turn aims to capture functional relevance. The first step of the diffusion method is to model the protein interactions as a network. A network is comprised of a set of nodes, V, and a set of edges between nodes, E. For this work, nodes represent genes, and edges represent an interaction present in the unified set. Although we use unweighted edges in this work, our network methods and software are able to receive weighted input as well, such as protein interactions with confidence measures taken from STRING [66]. The construction of diffusion kernels using weighted edges has been well studied and is equally valid [51]. n is the number of nodes in the network, which is 19,972 (yielding 236,642 edges). All network methods were implemented in R. The regularized Laplacian kernel [51] is constructed by:
K=(I+αL)−1
(1)
where K is the resulting kernel, I is the identity matrix, L is the graph Laplacian, and alpha is a constant (see S1 Text and [51] for additional details). For this study, an alpha value of 0.1 was used, consistent with other work in this field [17]. Next, a network diffusion score was computed for each gene. To do this, the diffusion score vector, y, was initialized to be a length n vector that contains 1’s in the indices of the RAD genes, and 0’s otherwise. Risk scores for all genes in the graph were then derived by multiplication of K by the diffusion score vector y: ỹ = Ky.
To test if RAD genes had closer than random diffusion proximity to other RAD genes in a network, leave-one-out cross validation [67] was applied to the RAD gene set. First, a single RAD gene from the RAD set was set to 0 in the initial diffusion score vector, y. Then, diffusion scores were computed based upon this new initialization of y. The diffusion scores were sorted and the sorted rank of the removed RAD gene’s diffusion score was determined in comparison to all other non-RAD genes. This process was repeated for each gene in the RAD set, resulting in a list of ranks. If diffusion proximity is informative and potentially predictive, the average rank of the RAD genes should be significantly lower than the average rank of all genes, (n+1) / 2, which was verified using a one-tailed t-test.
The Alzheimer’s Disease Genetics Consortium (ADGC) is an NIA-funded project whose goal is to identify genes associated with an increased risk of developing late-onset Alzheimer disease (LOAD) by assembling and analyzing genetic and phenotypic data from large cohorts containing rigorously evaluated AD cases and cognitively normal controls of various ethnic ancestries. Details of ascertainment, collection, quality control (QC), and analysis of genotype and phenotype data in the individual datasets of the ADGC are provided elsewhere [8, 68]. Here we examined genotype data that were generated using high-density SNP microarrays from 32 prospective, case-control, and family-based studies of LOAD comprising 16,175 case and 17,176 controls of European ancestry. After QC steps to filter low-quality SNPs and individuals with low genotype call rates, principal components (PCs) of ancestry were computed within each dataset using EIGENSTRAT [69] and a set of 21,109 SNPs common to all genotyping platforms and datasets in order to account for population substructure in genetic association analysis. Samples with outlier PC values >six standard deviations from the mean were excluded from subsequent analyses. Genotypes for a much larger set of SNPs were imputed using the Haplotype Reference Consortium panel release 1.1 [70, 71], which includes 64,976 haplotypes derived from 39,235,157 SNPs, and the Michigan Imputation Server (https://imputationserver.sph.umich.edu/) running MiniMac3 [72, 73].
Association of AD with the imputed dosage of the minor allele for each SNP (a quantitative estimate between 0 and 2) genome-wide was conducted using logistic regression models implemented in PLINK [74] that included covariates for age-at-onset/age-at-exam, sex, the first three PCs, and an indicator variable for each dataset. Joint analysis was chosen in favor of meta-analysis to avoid problems that could be introduced if bootstrap aggregation under-sampled small cohorts, resulting in unreliable association estimates for those cohorts. To account for relatedness in family datasets, subsets of maximally-unrelated affected and unaffected individuals were sampled from each pedigree. Each variant was annotated to a gene region according to RefSeq release 69 [75] using the program ANNOVAR [76]. Then, each gene was assigned the minimum p-value of all variants annotated to it, after applying the following formula:
PgGene′=1−(1−PgBestSNP)N+12
(2)
where N is the number of variants analyzed that were annotated to the gene. Previously, this correction [77] has been shown to perform comparably to more complex adjustments based upon gene length, recombination hotspots, and similar gene features [78].
Since the availability of large AD genetic datasets is limited, bootstrap aggregation [52] was used to generate a high number of datasets for method validation. First, the full ADGC dataset was equally separated into discovery and replication halves. Then, 25 iterations of bootstrap aggregation were applied to the discovery half and then the replication half. The resultant 25 discovery and 25 replication datasets were then matched (D1 and R1, D2 and R2….D25 and R25). To further ensure robustness, the splitting procedure was repeated a total of 5 times, with 25 iterations of bootstrap aggregation applied each time, resulting in 125 total pairings (D1 and R1, D2 and R2. …D125 and R125). Each pairing represents a discovery dataset as well as an independent replication dataset.
For each pairing, the previously described genetic analysis was conducted on the discovery half. Then all genes that passed a designated significance threshold (the number of passing genes is denoted as r) were selected to be tested again in the replication half using a significance threshold of (0.05 / r). The replication rate was computed by determining the percentage of passing genes in the discovery half that also passed in the replication half. A replication rate was estimated for each pairing, and the mean replication rate was then determined. Next, the replication rate was re-determined for each pairing, with the added criterion that selected genes must also have a top percentile network diffusion score (top 10th, 20th, 30th, 40th, and 50th were tested). The average replication rate for each filtering threshold was compared to the average replication rate without filtering.
The p-values from genetic analysis of the ADGC dataset were converted to Z-scores using the qnorm function in R. Then, the network diffusion scores were converted into percentiles. The percentiles are transformed into Z-scores using the qnorm function, with the additional specification of lower.tail = F. The weighting scheme from METAL was applied to combine the GWAS and network Z-scores:
Zcombined=w1*Zgwas+w2*Znetworkw12+w22
(3)
Although any weight selection can be used, the weights were “learned” using an SVM [53] due to the observation that the GWAS and network scores did not contribute equally to predicting replication rate. First, a replication rate was determined for each gene. If a gene had a p-value of <0.05 in d discovery datasets and a replication p-value of <0.05 in r of the paired replication datasets, it was assigned a replication rate of r/d. To reduce model overfitting, create sufficient separation between the classes, and achieve a balance of high and low replicating genes, only high replication genes (≥0.7, n = 676) and low replication genes (<0.1, n = 475) representing approximately 8.4% of the total genes with both a network and GWAS scores were extracted. By comparison, using a threshold of 0.8 or 0.9 would result in an imbalanced training set with very few high replication genes because highly replicating genes are uncommon. A linear SVM [53] was trained using the network Z-scores and the genetic association Z-scores as features, and “high” and “low” as the classes. The resulting slope of decision boundary was then used to determine appropriate weights (w1 = 0.703, w2 = 0.297).
Pathway enrichment was performed using the Gene Set Enrichment Analysis (GSEA) software [56]. GSEA’s pre-ranked analysis tool requires that the user provide a numeric measure for ordering genes. To establish a baseline, enrichment was done using our internal GWAS Z-scores to order genes. Then, enrichment was done using the alternative ordering genes based upon their combined Z-scores (see above for combination method). The gene sets tested for enrichment were the GSEA C2 pathways in MSigDb, which are the “curated gene sets” compiled from multiple sources including KEGG [60], Reactome [79], and domain experts. The significance threshold was set at FDR < 0.25, as suggested previously for this hypothesis generating approach [56].
The use of de-identified human subject information for this study was approved by the Boston University Institutional Review Board.
|
10.1371/journal.pbio.1000228 | Recognition of Lyso-Phospholipids by Human Natural Killer T Lymphocytes | Natural killer T (NKT) cells are a subset of T lymphocytes with potent immunoregulatory properties. Recognition of self-antigens presented by CD1d molecules is an important route of NKT cell activation; however, the molecular identity of specific autoantigens that stimulate human NKT cells remains unclear. Here, we have analyzed human NKT cell recognition of CD1d cellular ligands. The most clearly antigenic species was lyso-phosphatidylcholine (LPC). Diacylated phosphatidylcholine and lyso-phosphoglycerols differing in the chemistry of the head group stimulated only weak responses from human NKT cells. However, lyso-sphingomyelin, which shares the phosphocholine head group of LPC, also activated NKT cells. Antigen-presenting cells pulsed with LPC were capable of stimulating increased cytokine responses by NKT cell clones and by freshly isolated peripheral blood lymphocytes. These results demonstrate that human NKT cells recognize cholinated lyso-phospholipids as antigens presented by CD1d. Since these lyso-phospholipids serve as lipid messengers in normal physiological processes and are present at elevated levels during inflammatory responses, these findings point to a novel link between NKT cells and cellular signaling pathways that are associated with human disease pathophysiology.
| A central tenet of immunology is that cellular responses that protect us from pathogens result from molecular recognition of foreign compounds (antigens). The role of self-antigens in immune activation is less clear. We show here that an endogenous lipid called lyso-phosphatidylcholine (LPC) is recognized as an antigen by a subpopulation of human T lymphocytes, called natural killer T (NKT) cells, and specifically by the best-studied subgroup of these cells known as invariant NKT (iNKT) cells. NKT cells have attracted the interest of immunologists because they can potently influence the outcome of diverse immune responses; for example, they can promote bacterial clearance and tumor rejection, and they can also quell autoimmune disease pathology. Previous studies indicated that NKT cells are activated by self-antigens, but the identity of the relevant compounds remained unclear. Our finding that LPC is a self-antigen for iNKT cells suggests that these lymphocytes are attuned to highly conserved lipid signaling pathways that are fundamental to normal physiological processes and are markedly up-regulated during inflammation. Thus, these results provide a new molecular basis for understanding how iNKT cells contribute to a wide variety of immune responses.
| Natural killer T (NKT) cells are a unique subpopulation of T lymphocytes that display innate-like characteristics and can potently modulate adaptive immune responses [1],[2]. They are among the first cells to respond during microbial infections and produce a wide variety of cytokines that have multiple effects on other immune cells [3],[4]. NKT cells are characterized by a restricted T cell receptor (TCR) usage in which the TCRα chain is invariant, and the TCRβ chains show more limited variability than those of classical T lymphocytes. The T cell receptors of NKT cells are specific for a nonclassical antigen-presenting molecule called CD1d that presents lipids and glycolipids. One of the most remarkable features of NKT cells is the source of the antigens they recognize. Unlike classical MHC-restricted T cells, which are selected for recognition of non–self compounds, NKT cells have been found to recognize both self and foreign molecules [2],[3]. Thus, NKT cells become activated in vivo even when there is no external challenge, and this property may underlie many of their immunoregulatory effects as well as their rapid activation during infection [2],[5].
Based on their restricted TCR usage, it has been proposed that NKT cells recognize a conserved set of antigens. Consistent with this, NKT cells have been found to share recognition of a class of microbial lipids in which a galactose sugar is attached in an α-anomeric configuration to a sphingolipid or a diacylglycerol [6]–[8]. Recognition of this type of glycolipid appears to be conferred by an evolutionarily conserved antigen recognition “hotspot” within the T cell receptors of NKT cells [9]–[11]. It remains unclear whether the part of the TCR that varies from NKT cell to NKT cell confers additional individual antigen recognition properties; however, a number of reports have documented antigen-specific responses that are confined to subsets of the NKT cell population, suggesting that this may indeed be the case [12]–[14].
The molecular identity of the self-antigens responsible for endogenously activating NKT cells, and how these antigens stimulate beneficial immune functions rather than uncontrolled autoreactive pathology, are major unresolved mysteries. A series of studies has indicated that the self-antigens recognized by murine NKT cells are loaded into CD1d molecules within intracellular endosomal vesicles and require specialized processing steps that take place at these sites. Mutated CD1d molecules that do not traffic through the endosomal vesicular system fail to stimulate CD1d-dependent autoreactive responses by murine NKT cells and are not able to positively select NKT cells in vivo [15],[16]. Additionally, murine NKT cells show reduced responses to self-antigens if normal endosomal functioning is inhibited, for example, by the addition of pH-altering drugs or when lysosome-resident enzymes are genetically deficient [17],[18]. A glycolipid called isoglobotrihexosyl ceramide (iGb3) that is generated in lysosomal compartments through glycosidic cleavage of the mature tetra-glycosylated form has been identified as a self-antigen recognized by murine NKT cells [19]. However, this glycolipid is not required for the development and function of murine NKT cells in vivo, suggesting that other as yet unidentified compounds also function as NKT cell self-antigens [20].
In contrast to the murine system, the self-antigen responses of human NKT cells do not require lysosomal processes [21],[22]. Mutated human CD1d molecules that do not traffic through the endosomal system stimulated normal autoreactive responses by human NKT cell clones, and drugs that alter lysosomal pH also had no deleterious effect [22]. Similarly, antigen-presenting cells (APCs) that are genetically deficient in lysosomal lipid transfer proteins stimulated normal self-antigen responses by human NKT cells [21]. Moreover, although the iGb3 glycolipid is antigenic for a fraction of the human NKT cell subpopulation [19],[23], it is not clear that this is a self-antigen for human NKT cells, since current data suggest that the iGb3 molecule is not produced in humans due to the lack of functional genes for galactosyl transferase enzymes that are required for its biosynthesis [24]. These data, demonstrating disparity between the human and murine systems, suggest there may be significant differences in the nature of the self-antigens that regulate human and murine NKT cell responses. This potentially clinically important point will not be clarified until there is a molecular understanding of the CD1d ligands recognized by NKT cells of each species. Here, we have analyzed the responses of human NKT cells to lipids found within the ligand pool of secreted human CD1d molecules.
To identify self-antigens recognized by human NKT cells, we tested their responses to synthetic preparations of compounds that were identified in a pool of ligands eluted from human CD1d molecules [25]. Lipids were pulsed onto plate-bound recombinant human CD1d-Fc fusion protein and tested for their ability to stimulate cytokine secretion by a panel of human NKT cell clones. We have found from previous analyses that the CD1d-Fc fusion protein, which is produced in a hamster cell line, does not stimulate significant responses from our NKT cells unless an antigenic lipid is added [23]. Hence, because there is little or no detectable reactivity to CD1d ligands that may be endogenously present in the recombinant molecules, this assay provides a means of assessing NKT cell responses to added ligands, even if they are relatively weak agonists [26].
Figure 1A shows a summary of responses by human NKT cell clones to glycerophospholipids and sphingolipids found within a pool of lipid ligands eluted from human CD1d molecules [25]. Ligand species were selected so as to include representative diacylated phospholipids (phosphatidylcholine, PC; phosphatidylethanolamine, PE; phosphatidylinositol, PI; and phosphatidylglycerol, PG), a tetra-acylated cardiolipin (CL) species, monoacylated lyso-phospholipids (lyso-phosphatidylcholine, LPC; lyso-phosphatidylethanolamine, LPE; lyso-phosphatidylglycerol, LPG; and lyso-phosphatidic acid, LPA), and the two most abundant sphingolipids (sphingomyelin, SM; and the ganglioside GM3). As a positive control, the CD1d-Fc molecules were pulsed with a form of the prototypical NKT cell antigen α-galactosylceramide (α-GalCer) that contains a 20-carbon fatty acyl chain with two unsaturations (C20:2) and is known to load particularly well into recombinant CD1d molecules in solution [27]. Since natural ligands generally have been found to stimulate weaker responses from NKT cells than α-GalCer, as another control, we also assessed NKT cell responses to a truncated form of α-GalCer called “OCH” that has been shown to be a weaker agonist for human NKT cells [26],[28]. Of the species tested from the CD1d ligand pool, LPC elicited the strongest NKT cell responses (Figure 1A). The NKT cell responses to LPC were generally 10- to 100-fold less than their responses to the C20:2 analog of α-GalCer and appeared similar to those induced by OCH (Figure 1A), suggesting that LPC is a weak to moderate agonist.
Variation in the strength of the responses to LPC appeared to be largely due to reactivity differences among the NKT clones. Individual NKT cell clones were quite reproducible in their responses to LPC; some clones consistently showed strong responses, some regularly showed moderate or weak responses, and some repeatedly showed little or no detectable response (Figure 1B). The strength of individual NKT clone responses to LPC did not correlate with their responses to the C20:2 analog of α-GalCer (Figure 1B), suggesting that the LPC reactivity differences were not simply due to differing activation thresholds. Titrating the concentration of LPC used to prepulse the CD1d-Fc fusion protein yielded similar dose-response curves for all of the NKT cell clones. Significant responses above background were observed at LPC pulse concentrations from about 10 to 100 µM, with a peak at about 25 µM (Figure 1C). Notably, NKT cell responses were consistently diminished or absent at higher LPC pulse concentrations (Figure 1C).
To confirm that the NKT cell responses were due to recognition of LPC and not to a contaminant in the synthetic preparations of this compound, we tested LPC purified from chicken eggs. Dose-response curves to this LPC preparation, comprising a mixture of LPC species differing in their hydrocarbon chain lengths and double bonds, were similar to those for the synthetic LPC (Figure 1D). Thus, both synthesized and natural LPC preparations were recognized by the NKT cell clones, whereas synthesized preparations of related phospholipids were not.
To further investigate NKT cell recognition of LPC, we sorted a polyclonal population of NKT cells from the peripheral blood of a healthy volunteer donor using fluorescently labeled α-GalCer–loaded CD1d tetramers and expanded the cells in vitro for a short time (less than 1 mo) by stimulating them with PHA and IL-2 in the presence of irradiated autologous mononuclear cells. The resulting population of cells showed uniformly positive staining using an anti-CD3 antibody and α-GalCer–loaded CD1d tetramer, and contained approximately equal fractions of CD4+ and CD4− cells (Figure 1E, top panels). The expanded polyclonal NKT cells showed a detectable cytokine response to plate-bound CD1d-Fc molecules pulsed with LPC and also responded to the C20:2 analog of α-GalCer (Figure 1E, bottom panels).
Our screening of NKT cell lipid recognition showed occasional weak responses to other lipids identified within the CD1d ligand pool (Figure 1A). Therefore, we evaluated the NKT cell responses to these CD1d ligands using titrated doses of lipid. NKT cell clones that responded to LPC generally showed little or no recognition of other lyso-phospholipids, suggesting molecular specificity for LPC. For example, LPA, which is identical to LPC except for the absence of the choline head group, induced little or no NKT cell activation above background (Figure 2A). Diacylated PC sometimes stimulated very weak positive responses, but in most cases, there was no significant NKT cell activation from this lipid (Figure 2B), suggesting that the lyso- form contains antigenic features not present in the diacylated lipid.
We had previously identified a human NKT cell clone that consistently demonstrated specific responses to PI and PE, although other human NKT cell clones tested in parallel showed little or no recognition of these lipids [23]. In the current analysis, we found that PI was capable of eliciting weak responses from some NKT cell clones, but in general, this lipid failed to show stimulatory effects for the clones tested here (Figure 2C). PE only rarely elicited positive responses from the panel of NKT clones (Figure 1A, and unpublished data). We also failed to detect positive responses to plasmalogen forms of PE and PC (unpublished data). Weak but detectable NKT cell responses were sometimes observed to a purified preparation of the GM3 ganglioside, although in most cases, the results for this lipid were also negative (Figure 2D). Notably, a human NKT cell clone (J3N.4), from which we previously reproducibly observed positive responses to iGb3 [23], did not respond to the structurally related compound GM3 (Figure 2D). Sphingomyelin also generally stimulated no detectable response from the NKT cell clones (Figure 1A, and unpublished data). Thus, LPC was unique among the ligand species tested here in the strength, consistency, and dose-dependence of the NKT cell responses it elicited.
Lyso-phospholipids are known to be highly bioactive molecules that can signal through G-protein–coupled receptors; therefore, it was possible that the responses we observed might be due to direct stimulation of NKT cells, rather than via TCR-mediated antigen recognition. To address this possibility, we performed a number of controls to confirm that the observed NKT cell responses were due to recognition of LPC in the context of CD1d. NKT cells that were incubated directly with LPC in the absence of CD1d molecules showed no detectable cytokine secretion, and similarly, there was no response to plate-bound negative control antibody that was prepulsed with LPC (Figure 3A). NKT cell responses to LPC-pulsed CD1d-Fc molecules were specifically blocked by an anti-CD1d antibody (Figure 3B). Additionally, the NKT cells did not respond to CD1c-Fc molecules that were prepulsed with LPC (Figure 3C), although we found that CD1c-Fc and CD1d-Fc molecules showed similar binding of a biotinylated lyso-phospholipid (Figure 3D). Together, these results demonstrate that NKT cell responses to LPC require presentation by CD1d molecules.
In our initial screening of lyso-phospholipids found in a pool of eluted CD1d ligands [25], LPC was the only species that consistently stimulated cytokine secretion from most of the NKT cell clones (Figure 1A). Since the lipid tails of all of the lyso-phospholipids tested in this analysis were identical (i.e., C18:1), this suggests that NKT cell recognition is dependent on chemical features of the head group. To investigate this further, we tested the ability of LPC-reactive NKT cells to respond to lyso-sphingomyelin (LSM), a structurally related compound that was not found in the CD1d ligand pool. LPC and LSM can be generated by similar enzymatic cleavage of the diacylated phospholipids phosphatidylcholine (PC) and sphingomyelin (SM), resulting in removal of the fatty acyl chain and the generation of lyso- species that contain a choline head group linked by a phosphate ester to a single hydrocarbon tail (Figure 4A). NKT cell clones that recognized LPC consistently also showed responses to CD1d-Fc molecules that were prepulsed with LSM, although higher molar concentrations of LSM were required to stimulate responses (Figure 4B). In contrast, there was typically no detectable response to SM, the diacylated form (Figure 4C). Sphingosine 1-phosphate, a lyso- species lacking the choline head group, stimulated little or no response from the NKT cells (Figure 4D). Hence, NKT cell responses were specific for lyso-phospholipids containing a choline head group.
Whereas lipids that appear to be abundant cellular ligands of human CD1d, such as SM, PC, PE, PI, CL, or GM3 [25],[29],[30], showed little or no antigenicity in this analysis, it is nevertheless possible that they play an important role in the CD1d antigen-presenting system by modulating the ability of other more antigenic lipids to load into CD1d molecules. We therefore investigated whether binding of these lipids to CD1d could block the subsequent presentation of an antigenic glycolipid. Recombinant CD1d-Fc molecules were preincubated with diacylated phospholipids or sphingolipids. The recombinant CD1d-Fc was then washed and incubated with a saturating concentration of the C20:2 analog of α-GalCer and tested for the ability to stimulate cytokine secretion by NKT cell clones. Pretreatment with several of the lipids, including PA, PC, PE, CL, and GM3, consistently resulted in almost complete blocking of the response to the C20:2 antigen (Figure 5A). In contrast, pretreatment with PG, PI, or SM resulted in only partial blocking of C20:2 (Figure 5A). These results suggest that a fraction of the CD1d molecules exiting the secretory pathway (e.g., those containing PG, PI, or SM) may be receptive to binding extracellular diacylated lipids such as C20:2 at the cell surface.
We next investigated the ability of lyso-phospholipids to bind to CD1d molecules containing cellular ligands. We have found that we can readily detect specific association of biotinylated LPE with recombinant CD1d-Fc molecules (Figure 3D). However, it is not clear whether the CD1d-Fc fusion proteins used in these experiments contain endogenous lipids, and if they do, whether these modulate the binding of exogenously added lipids. Therefore, we investigated the binding of biotinylated LPE to purified CD1d-β2m heterodimers that were produced in a human lymphoblastoid cell line, and for which the bound ligands have recently been characterized as a mixture of phospholipids and sphingolipids [25]. Although the signal was much lower than that observed for the CD1d-Fc fusion protein, the purified CD1d-β2m molecules also yielded a biotin signal that was significantly above the background, indicating the presence of bound LPE (Figure 5B). These results indicate that lyso-phospholipids can bind to CD1d molecules containing a complex mixture of cellular lipids.
To further investigate, we tested whether lyso-phospholipids can cause the dissociation of diacylated lipids from CD1d molecules. Recombinant human CD1d molecules produced in insect cells have a uniform charge distribution and can be visualized as a single major band on a native isoelectric focusing (IEF) gel (Figure 5C, lane 1). When the CD1d molecules are loaded with a charged lipid such as the trisialoganglioside GT1b, the band shifts due to the acidic charge of the bound lipid (Figure 5C, lanes 2 and 3). Binding of a neutral lipid (e.g., α-GalCer) to the CD1d-GT1b complex replaces the bound GT1b and is therefore associated with loss of the acidic charge (Figure 5C, lane 4). We found that addition of a 3-fold molar excess of either LPC or LPE to the CD1d-GT1b complex resulted in dissociation of 70%–80% of the bound GT1b, as assessed by the reduced intensity of the acidic band and the increased intensity of the basic band (Figure 5C, lanes 5 and 6). Titrating the concentration of lyso-phospholipid that was added to the CD1d-GT1b complex demonstrated that even a 1∶1 molar ratio of lyso-phospholipid to CD1d was sufficient to induce dissociation of approximately 30% of the bound GT1b (Figure 5D, lane 2), with nearly complete GT1b dissociation observed at molar ratios of 5∶1 or higher (Figure 5D, lanes 3–6). These results demonstrate that lyso-phospholipid loading into CD1d molecules is not prevented by previously bound diacylated lipids.
Previous studies have indicated that endosomal trafficking of CD1d is important for efficient presentation of certain exogenous lipids, such as α-GalCer, apparently because loading of α-GalCer into CD1d molecules occurs much more efficiently in endosomal vesicles [22],[31]. We therefore investigated the role of CD1d endosomal trafficking for presentation of exogenous LPC by APCs. As observed previously [22], human lymphoblastoid cell lines transfected with cytoplasmic tail-deleted CD1d molecules that lack the amino acid motif required for reinternalization from the cell surface show reduced α-GalCer–dependent NKT cell responses compared with transfectants expressing wild-type CD1d (Figure 6A). However, wild-type and tail-deleted CD1d transfectants stimulate similar CD1d-dependent autoreactive responses by NKT cells (Figure 6A), demonstrating that endosomal recycling of CD1d molecules is not required for presentation of antigenic cellular lipids. Addition of LPC to wild-type CD1d transfectants resulted in statistically significant increases in NKT cell cytokine secretion in three out of 17 experiments (Figure 6B). In these cases, the magnitude of the enhancement was low (a mean increase of 1.48-fold±0.234). However, addition of LPC to tail-deleted CD1d transfectants produced significantly enhanced NKT cell responses in seven out of 19 experiments, and in these cases, the magnitude of the effect was greater (mean increase of 3.67-fold±1.727). These experiments indicate that it is possible for extracellular LPC to compete with endogenous ligands and load into cell surface CD1d molecules, although this pathway does not appear to be highly reproducible. Additionally, these results suggest that endosomal recycling of CD1d molecules limits the presentation of extracellular LPC.
To further investigate, we compared two species of LPC. Most LPC species that have been identified as CD1d cellular ligands contain carbon chains with one or more double bonds [25],[29]; however, the most abundant species of LPC in extracellular fluids is often the fully saturated C16:0 carbon chain form. We found that NKT cell responses to APCs pulsed with C18:1 and C16:0 LPC appeared similar (Figure 6C), suggesting that both species can load into cell surface CD1d. Importantly, the NKT cell responses to LPC-treated APCs were completely CD1d-dependent, since the CD1d-negative parental cell line that was pulsed with LPC did not stimulate NKT cell cytokine secretion (Figure 6C). Interestingly, similar to our results using recombinant CD1d-Fc molecules for presentation, the LPC-dependent responses were highly concentration dependent and consistently appeared diminished or abrogated when the APCs were pulsed with high levels of LPC (Figure 6C).
We next tested the effect of blocking phospholipase A2 enzymes on the autoreactive responses of NKT cells. Human monocytes in peripheral blood constitutively express CD1d and stimulate CD1d-dependent cytokine secretion by human NKT cells in the absence of added antigens [32]–[34]. We isolated monocytes from human peripheral blood and preincubated them for 24 h with a polyclonal preparation of chicken antibodies (IgY) directed against secreted phospholipase A2 (sPLA2), or with a negative control preparation of polyclonal IgY [35]. The monocytes were then washed and used to stimulate cytokine secretion by human NKT cell clones. Monocytes that were pretreated with the anti-sPLA2 antibody showed significantly reduced stimulation of NKT cell cytokine secretion compared to those that were treated with the negative control antibody, or to untreated monocytes (Figure 6D). Importantly, monocyte cell surface expression of CD1d was not reduced by anti-sPLA2 antibody pretreatment (unpublished data). These results point to an important role for PLA2 enzymes, key producers of LPC in vivo, in the activation of NKT cells by physiological APCs.
To further investigate the physiological role of LPC presentation by CD1d, we analyzed IFNγ responses by human peripheral blood lymphocytes (PBLs) directly ex vivo. Lymphocytes were freshly isolated from ten healthy adult donors, and tested by ELISpot analysis for cells that produced IFNγ in response to CD1d transfected or untransfected APCs. Because the APCs used for these experiments do not express MHC class II molecules on the cell surface and have reduced MHC class I expression [36], they should not stimulate marked alloreactive responses from the peripheral blood T cell populations of most donors. Consistent with this, most donors (seven out of ten) showed little or no IFNγ secretion (i.e., less than 20 spots per well) in response to the untransfected APCs (Figure 7A, left plot). However, PBL samples that were incubated with CD1d-transfected APCs consistently showed significantly increased numbers of spots (Figure 7A, left plot), suggesting that exposure to APCs expressing CD1d stimulated lymphocytes within the samples. Notably, the increased IFNγ production did not require the CD1d+ APCs to be prepulsed with antigen, suggesting that the responses are due to recognition of an endogenous antigen. PBL samples that were incubated with CD1d-transfected APCs prepulsed with C20:2 consistently showed a further increase in the number of spots (Figure 7A, middle plot), suggesting that additional T cells were activated by CD1d-mediated presentation of the α-GalCer analog. Most donors (eight out of ten) showed increased numbers of spots in response to CD1d tail-deleted APCs pulsed with LPC, compared to CD1d tail-deleted APCs treated with vehicle alone (Figure 7A, right plot). Six of the eight “responding” donors showed marked increases in the number of spots detected in response to the LPC-pulsed APCs (Figure 7B). These data suggest that CD1d-restricted T cells that respond to LPC as an antigen are present in the blood of healthy human adults.
The results presented here show that a fraction of human NKT cells specifically recognize LPC and LSM. Recognition of these lipids was observed using NKT cells that express semi-invariant T cell receptors and recognize a class of foreign antigens called α-GSLs [23],[26]. Semi-invariant NKT cells (or “iNKT” cells) such as these have been associated with beneficial immunoregulatory effects in a variety of murine models and also appear deficient in certain human autoimmune conditions [1]–[3]. It has been hypothesized that iNKT cell recognition of self-antigens allows them to perform immunoregulatory functions without foreign antigenic stimulation; however, the specific mechanisms by which this may occur have remained unclear. Our results indicate that the functions of NKT cells may be regulated by conserved lipid signaling pathways that operate during normal physiology and that have elevated activity during pathophysiological processes.
It has recently been shown that LPC can be isolated from human CD1d molecules purified from human lymphoblastoid cell lines [25],[29], providing strong evidence that lyso-phospholipids such as LPC can successfully compete with other types of self-lipids for loading into CD1d molecules. LPC is produced by the action of PLA2 enzymes, which are a functionally defined superfamily comprising at least 15 distinct types of proteins that localize to a variety of intracellular and extracellular sites [37]. Therefore, multiple sources of LPC may be available for loading into CD1d molecules. For example, stimulation of APCs by growth factors, cytokines, neurotransmitters, hormones, and other extracellular signals can lead to the activation of cytoplasmic PLA2 enzymes and release of LPC into the cytoplasm [38]. Additionally, the recent identification of a lysosome-resident PLA2 enzyme that is up-regulated in human monocytic cells upon stimulation through the retinoid X receptor suggests that LPC is produced within lysosomes after certain kinds of cellular activation [39]. Finally, several types of secreted PLA2 enzymes produce LPC by cleaving PC on the outer leaflet of the plasma membrane [40], and this LPC could load into CD1d molecules at the cell surface.
Our data indicate that secreted PLA2 enzymes are important for autoantigenic stimulation of NKT cells, since treatment of monocytes with an IgY preparation that was raised against purified sPLA2 protein specifically blocked their subsequent activation of NKT cells. This finding is consistent with the possibility that the cell surface is an important site of LPC production for loading into CD1d molecules. However, it is not clear from our results that high concentrations of extracellular LPC facilitate the activation of iNKT cells, since we have consistently found that NKT cells show little response to CD1d-mediated presentation of LPC when the lipid is added in concentrations above about 50 µM. The reason for this is unknown; our binding studies suggest that LPC does bind to CD1d molecules at these lipid concentrations. Nevertheless, this failure of high concentrations of LPC to activate NKT cells may be physiologically significant, since it occurs when either plate-bound recombinant CD1d-Fc molecules or CD1d-transfected APCs are used for LPC presentation. We also find that transfectants expressing wild-type CD1d molecules (which continuously recycle from the cell surface through endosomal compartments and back to the cell surface) show only a limited ability to present exogenously added LPC, whereas transfectants expressing tail-deleted CD1d molecules that are deficient in internalization from the cell surface appear more efficiently able to present exogenous LPC. This observation suggests that the normal recycling of CD1d molecules on APCs may limit the presentation of extracellular LPC. Thus, it remains to be determined whether autoreactive iNKT cell activation is most effective when LPC is produced at concentrations and cellular locations that are associated with normal physiological states or is further enhanced by elevated extracellular levels of LPC that are associated with inflammation.
We show here that lymphocytes that produce IFNγ in response to CD1d+ APCs are consistently present in the peripheral blood of healthy adult humans, and that for many donors, there is an increase in the frequency of IFNγ-producing cells when the APCs are prepulsed with LPC. It is not clear whether the LPC-reactive lymphocytes detected in this analysis are iNKT cells or whether they belong to a different subset of CD1d-restricted T cells. For example, blood samples from human multiple myeloma patients were recently reported to contain elevated frequencies of LPC-reactive CD1d-restricted T cells [41]. However, the LPC-reactive T cells from multiple myeloma patients did not utilize the characteristic T cell receptor of NKT cells and demonstrated skewed cytokine production, suggesting that they comprise a distinct CD1d-restricted T cell population [41]. Since LPC accumulates to greatly increased concentrations in blood and other bodily fluids in chronic inflammatory conditions such as multiple myeloma, it is possible that the T cell populations detected in blood of multiple myeloma patients were specifically expanded as a result of the disease state. It is not clear whether these LPC-reactive T cells play a pathogenic or a regulatory role in multiple myeloma.
Unfortunately, it has been difficult for us to gauge peripheral blood frequencies of LPC-reactive T cells in healthy donors because we have not obtained reproducible staining using LPC-loaded CD1d tetramers. Thus, it is not clear what fraction of the total iNKT cell population normally recognizes LPC, or what fraction of the total LPC-reactive T cell population is normally comprised of iNKT cells. However, our results do clearly demonstrate that not all iNKT cells recognize LPC. Approximately 75% of the NKT cell clones tested (eight out of 12) showed responses to LPC, whereas the remainder did not respond to this antigen but did respond well to the α-GSL used as a control. The ability of individual NKT cell clones to respond to LPC was generally very reproducible, and therefore, the most likely explanation for the clonal variation is that the TCR β-chain sequences of some clones permit recognition of this antigen, whereas other TCR β-chain sequences do not. Since the NKT cell clones that failed to respond to LPC nevertheless demonstrate detectable autoreactive responses to CD1d molecules expressed on APCs ([23], and unpublished data), these results suggest that some iNKT cells may recognize another, as yet unidentified, endogenous ligand. Alternatively, our results are also consistent with the possibility that additional autoreactive responses by iNKT cells result from recognition of very weak agonists that are abundant constituents of the ligand pools of human CD1d molecules, such as diacylated glycerophospholipids (e.g., PC, PI, and PE) and glycosphingolipids such as GM3 [25],[29].
It has recently been demonstrated that an autoreactive subset of noninvariant CD1d-restricted T cells found in mice can recognize sulfatide, a glycolipid derived from myelin, and that a lyso- form of sulfatide is more potently antigenic than the diacylated form [42],[43]. Hence, it may be a common finding that lyso-lipid species are more antigenic for CD1d-restricted T cells than their diacylated counterparts. Thus, perhaps autoreactive CD1d-restricted T cells monitor endogenous levels of cleaved lipids. In this case, oxidizing agents and lipases that generate these compounds may play a key role in the activation of these natural T cell populations. This possibility adds a new dimension to observations that sPLA2 enzymes play important roles, not only in inflammatory conditions, but also in host defense during microbial infections [40], since part of the immunological effects of these enzymes may result from their production of antigens that stimulate CD1d-restricted T cells. Similarly, the observation that lyso-phospholipids such as LPC can serve as potent immune adjuvants that enhance antigen-specific antibody production and cytotoxic T cell activation raises the possibility that these effects of LPC may involve the specific activation of NKT cells [44], since NKT cells are known to potently enhance memory responses by antigen-specific B cells and T cells [45],[46]. Thus, understanding the role of self-antigens such as LPC in regulating the responses of human NKT cells and other CD1d-restricted T cell populations may provide critical new insights into beneficial immune activation as well as disease pathology.
The glycosphingolipid α-GalCer and its OCH and C20:2 structural analogs were synthesized as described previously [27],[47]. Synthetic preparations of the following lipids were obtained commercially (Matreya or Avanti Polar Lipids): C18:1/C18:1 phosphatidic acid, C18:1/C18:1 phosphatidylcholine, C18:1/C18:1 phosphatidylethanolamine, C18:1/C18:1 phosphatidylglycerol, C18:1/C18:1 phosphatidylinositol, C18:1/C18:1/C18:1/C18:1 cardiolipin, sphingomyelin (containing a C18:1 acyl chain), C18:1 lyso-phosphatidic acid, C18:1 and C16:0 lyso-phosphatidylcholine, C18:1 lyso-phosphatidylethanolamine, C18:1 lyso-phosphatidylglycerol, lyso-sphingomyelin, and sphingosine-1-phosphate. Purified preparations of the ganglioside GM3 (from bovine buttermilk) and lyso-phosphatidylcholine (from chicken eggs) were purchased from Avanti Polar Lipids. Diacylated lipids were dissolved in DMSO at a concentration of 100 µg/ml and stored frozen at −20°C. Lyso-phospholipids were dissolved in 50% DMSO/dH2O at a concentration of 400 µg/ml and stored frozen at −20°C. Lipids were warmed to room temperature, then sonicated at 60°C in a heated water bath for 20 min before use.
Human NKT cell clones were established as described previously [23], and maintained at 37°C with 5% CO2 in the following culture medium: RPMI 1640; 2 mM l-glutamine; 100 µg/ml penicillin and streptomycin; 10% fetal bovine serum (Hyclone); 5% bovine calf serum (Hyclone); 3% human AB serum (Atlanta Biologicals); supplemented with 400 U/ml recombinant human IL-2 (Chiron). The NKT cell clones were periodically restimulated by incubating them with irradiated allogeneic peripheral blood mononuclear cells (PBMCs) and 30 ng/ml anti-CD3 monoclonal antibody (mAb) (clone SPVT-3b). Polyclonal NKT cells were expanded from freshly isolated PBMCs from a healthy adult donor as follows: monocytes and B lymphocytes were removed by magnetic depletion using anti-CD14 and anti-CD19 microbeads (Miltenyi Biotec), and the remaining cells were incubated with human CD1d tetramer loaded with the C20:2 analog of α-GalCer, then the labeled cells were separated using goat anti-mouse IgG magnetic microbeads (Miltenyi Biotec). The positively selected cells were stimulated to proliferate by exposure to irradiated autologous PBMCs, in medium containing 250 ng/ml PHA-p. Recombinant human IL-2 (Chiron) was added after 2 d at a concentration of 40 U/ml, and titrated up to 400 U/ml over a period of 10 d. Experiments were performed on the polyclonal NKT cells within 3–4 wk of the initial sorting from fresh blood.
APCs expressing wild-type or tail-deleted CD1d molecules were generated using the human lymphoblastoid 3023 cell line, as described previously [22]. The untransfected 3023 parental cell line was maintained in the following culture medium: RPMI 1640; 2 mM l-glutamine; 100 µg/ml penicillin and streptomycin; 5% bovine calf serum (Hyclone); 5% fetal bovine serum (Hyclone); 0.5 mg/ml G418 (Mediatech). For the transfected cell lines, this culture medium was supplemented with 0.5 µg/ml puromycin (Sigma-Aldrich).
Recombinant human CD1d-Fc fusion protein was prepared as previously described [23]. CD1d-Fc fusion protein and anti-CD11a antibody (clone HI111, BioLegend) were coated onto high protein binding 96-well microtiter plates at 0.5 µg and 0.05 µg per well, respectively. Where indicated, the CD1d-Fc was replaced by human CD1c-Fc fusion protein or an isotype-matched negative control antibody (clone UPC-10). The wells were then incubated for 16–20 h at 37°C with C20:2, OCH, or test lipids diluted in 25% DMSO/dH2O. The wells were washed with sterile PBS, then RPMI, then RPMI containing 10% fetal bovine serum, and NKT cell clones (5×104/well) were added in a final volume of 200 µl/well in culture medium (RPMI 1640; 2 mM l-glutamine; 100 µg/ml penicillin and streptomycin; 10% fetal bovine serum (Hyclone); 1 mM sodium pyruvate; 55 µM 2-mercaptoethanol; and nonessential amino acids). Where indicated, anti-CD1d antibodies (clone CD1d42.1) or an isotype-matched negative control antibody (clone P3) were added to the wells at a final concentration of 10 µg/ml, prior to the addition of NKT cells. Supernatants were collected after 18–24 h, analyzed for granulocyte macrophage colony-stimulating factor (GM-CSF) by ELISA (BioLegend), and quantified by comparison to recombinant human GM-CSF standards (PeproTech). Using this protocol, the means ± standard deviations of the background GM-CSF secretion were as follows: NKT cells exposed to plate-bound anti-CD11a without CD1d-Fc molecules, 47.9±88.9 (n = 44); NKT cells exposed to plate-bound anti-CD11a and untreated CD1d-Fc molecules, 44.2±67.9 (n = 69); NKT cells exposed to plate-bound anti-CD11a and vehicle-pulsed CD1d-Fc molecules, 45.2±80.2 (n = 210).
Lyso-phosphatidylethanolamine (LPE) was biotinylated using Sulfo-NHS-biotin (Pierce), according to the manufacturer's protocol. The biotinylated lipid was dissolved in DMSO at a concentration of 100 µg/ml and sonicated at 60°C in a heated water bath for 20 min. Biotinylated LPE in PBS supplemented with 1 mg/ml BSA was incubated at the indicated concentrations for 2 h at 37°C with recombinant CD1c-Fc or CD1d-Fc fusion proteins, or with secreted native CD1d molecules produced in a human lymphoblastoid cell line as described [25]. The lipid-treated CD1 molecules were then incubated in microtiter plates coated with anti-CD1c mAb (clone F10/21A3), anti-CD1d mAb (clone CD1d42), or an isotype-matched negative control mAb (clone P3), to allow assessment of the CD1-dependent binding compared to the background, and biotinylated-LPE was detected using streptavidin-alkaline phosphatase (Zymed).
Lyso-phospholipid association with CD1d was also tested using an assay that measures displacement of a charged lipid ligand that is prebound to the CD1d [48]. A 6-His–tagged construct of the human CD1d ectodomain was coexpressed with human β2-microglobulin using a baculovirus insect expression system. CD1d protein was purified using Ni-NTA resin, followed by size-exclusion chromatography over a Superdex 200 column (GE Healthcare). The CD1d was loaded with purified trisialoganglioside GT1b (Matreya), as described previously [48]. Untreated or GT1b-loaded CD1d preparations were incubated for 2 h at 37°C at a protein concentration of 40 µM in HBS, in the presence of the indicated concentrations of α-GalCer, C18:1 LPC, C18:1 LPE, or GT1b as a control. The species were then separated according to charge on a native isoelectric focusing gel (IEF PhastGel, GE Healthcare), and protein bands were visualized by Coomassie stain.
To assess inhibitory effects on antigen presentation, microtiter plate wells coated with recombinant CD1d-Fc fusion protein and anti-LFA-1 mAb were incubated for 24 h at 37°C with vehicle (25% DMSO in dH2O), or with vehicle containing the indicated lipids at a final concentration of 75 µM. The wells were then washed with PBS, and a solution of 0.6 µM C20:2 (dissolved in PBS supplemented with 1 mg/ml BSA) was added and incubated for 24 h at 37°C. The wells were washed again, and NKT cell clones (5×104/well) were added and incubated for 18–20 h at 37°C with 5% CO2. Supernatants were collected and analyzed by standardized ELISA for GM-CSF concentration. Percent inhibition was calculated by the following formula: 1−(GM-CSF produced in response to lipid pretreated CD1d/GM-CSF produced in response to vehicle pretreated CD1d)×100.
Wild-type or tail-deleted CD1d transfected or untransfected 3023 human B lymphoblastoid cells were pulsed for 4 h at 37°C with α-GalCer, or lyso-phosphatidylcholine (LPC), or vehicle (DMSO) alone. The cells were washed with culture medium, then co-incubated at a 1∶1 ratio (5×104/well each) with NKT cells, in a final volume of 200 µl. Supernatants were collected after 18–24 h, and analyzed for NKT cell production of GM-CSF using a standardized ELISA.
Polyclonal anti-sPLA2 IgY antibodies were prepared by immunizing Single-Comb White Leghorn laying hens with sPLA2 IB enzyme purified from porcine pancreas (Novozyme) in complete Freund's adjuvant (CFA). Negative control IgY antibodies were prepared by immunizing the hens with CFA alone [35]. The two IgY antibody preparations were purified from egg yolks by extraction with polyethylene glycol, followed by dialysis using a 50 kDa molecular weight (MW) cutoff membrane. The presence of IgY specific for sPLA2 in the immunized antibody preparation was confirmed by ELISA, whereas the negative control antibody preparation showed no detectable anti-sPLA2 antibody signal (Figure S1A). The anti-PLA2 IgY preparation was capable of reducing the conversion of PC to LPC by a secreted PLA2 enzyme in vitro, whereas the negative control IgY did not have this effect (Figure S1B). Additionally, we observed specific binding of the anti-PLA2 IgY to the cell surface of freshly isolated human monocytes (Figure S1C), suggesting that antibodies within the preparation recognize human PLA2 molecules. To assess the effect of anti-sPLA2 antibody treatment on NKT cell activation, monocytes were isolated from human PBMC samples by magnetic sorting using CD14 microbeads (Miltenyi Biotec). The monocytes were incubated for 18–24 h at 37°C and 5% CO2 in culture medium containing 20 µg/ml anti-sPLA2 or negative control IgY, or in culture medium with no added antibodies. The monocytes were washed with fresh medium and then combined at a 1∶1 ratio with NKT cells (5×104/well of each). Supernatants were collected after 24 h and analyzed by ELISA for the production of GM-CSF and IL-13 (BioLegend).
Human PBMCs were purified from fresh blood obtained from healthy adult donors using Ficoll-Paque density gradient centrifugation (GE Health Sciences), and B cells, monocytic cells, and plasmacytoid DCs were removed by magnetic sorting using beads specific for CD19, CD14, and BDCA-4 (Miltenyi). CD1d transfected or untransfected 3023 cells were incubated for 2 h at 37°C in culture medium containing LPC (10 µM), or the C20:2 analog of α-GalCer (260 nM), or vehicle (DMSO) alone, then washed and resuspended in fresh medium. PBMCs and APCs were added in a 1∶1 ratio (100,000 cells per well total) in serum-free medium (CELLect medium, MP Biomedicals) to triplicate wells of 96-well PVDF membrane plates (Whatman) coated with anti-human IFNγ mAb (clone NIB42 from BioLegend). The cells were incubated for 48 h at 37°C and 5% CO2. Secreted IFNγ was detected using biotinylated anti-human IFNγ mAb (clone M701B from Thermo Scientific), and revealed by development with streptavidin-alkaline phosphatase and BCIP/NBT chromogenic substrate. Spots were quantitated using AID 5.0 software. Background signal from analysis of PBMCs without added APCs was typically less than 20 spots per well.
|
10.1371/journal.pgen.1001110 | Incremental Genetic Perturbations to MCM2-7 Expression and Subcellular Distribution Reveal Exquisite Sensitivity of Mice to DNA Replication Stress | Mutations causing replication stress can lead to genomic instability (GIN). In vitro studies have shown that drastic depletion of the MCM2-7 DNA replication licensing factors, which form the replicative helicase, can cause GIN and cell proliferation defects that are exacerbated under conditions of replication stress. To explore the effects of incrementally attenuated replication licensing in whole animals, we generated and analyzed the phenotypes of mice that were hemizygous for Mcm2, 3, 4, 6, and 7 null alleles, combinations thereof, and also in conjunction with the hypomorphic Mcm4Chaos3 cancer susceptibility allele. Mcm4Chaos3/Chaos3 embryonic fibroblasts have ∼40% reduction in all MCM proteins, coincident with reduced Mcm2-7 mRNA. Further genetic reductions of Mcm2, 6, or 7 in this background caused various phenotypes including synthetic lethality, growth retardation, decreased cellular proliferation, GIN, and early onset cancer. Remarkably, heterozygosity for Mcm3 rescued many of these defects. Consistent with a role in MCM nuclear export possessed by the yeast Mcm3 ortholog, the phenotypic rescues correlated with increased chromatin-bound MCMs, and also higher levels of nuclear MCM2 during S phase. The genetic, molecular and phenotypic data demonstrate that relatively minor quantitative alterations of MCM expression, homeostasis or subcellular distribution can have diverse and serious consequences upon development and confer cancer susceptibility. The results support the notion that the normally high levels of MCMs in cells are needed not only for activating the basal set of replication origins, but also “backup” origins that are recruited in times of replication stress to ensure complete replication of the genome.
| Proper replication of the genome is essential for maintenance of the genetic material and normal cell proliferation. DNA replication can be compromised by exogenous factors and genetic disruptions. Such compromise can lead to disease such as cancer, which is characterized by genomic instability (an elevated mutation rate). Because the DNA replication apparatus is essential, relatively little is known about how genetic variants impact the health of whole animals. In this report, we studied mice bearing combinatorial mutations in a component of the replication apparatus, the MCM2-7 helicase. MCM2-7 is a complex of 6 proteins that are essential for initiating DNA replication along chromosomes, and to unwind the DNA during DNA replication. We find that although cells have excess amounts of MCM2-7 to support proliferation under normal circumstances, that incremental MCM depletions have multiple drastic effects upon the whole animal, including embryonic lethality, stem cells defects, and severe cancer susceptibility. Additionally, we report that mouse cells regulate and coordinate the levels of usable MCM proteins, both at the level of synthesis and also by regulating access to chromatin. The implication is that genetic variants that impact MCM levels, even to a minor degree, can translate into disease.
| In late mitosis to early G1 phase of the cell cycle, DNA replication origins are selected and bound by the hexameric origin recognition complex (ORC; [1]). ORC then recruits the initiation factors CDC6 and CDT1, which are required for loading MCM2-7, thereby forming the “pre-replicative complex” (pre-RC). The formation of pre-RCs is termed origin “licensing” and this gives origins competency to initiate a single round of DNA synthesis before entering S phase. MCM2-7 is a hexamer of six distinct but structurally-related minichromosome maintenance (MCM) proteins (reviewed in [2]–[5]). In vivo and in vitro evidence indicates that the MCM2-7 complex is the replicative helicase [6]–[8].
MCM2-7 proteins are abundant in proliferating cells [9], and are bound to chromatin in amounts exceeding that which is present at active replication origins or required for complete DNA replication [10]–[14]. Although these and other studies showed that drastic decreases in MCMs are tolerated by dividing cells, there are certain deleterious consequences. In Xenopus extracts and mammalian cells, excess chromatin-bound MCM2-7 complexes occupy dormant or “backup” origins that are activated under conditions of replication stress, compensating for stalled or disrupted primary replication forks [11], [15]–[16]. The depletion of these backup licensed origins was associated with elevated chromosomal instability and susceptibility to replication stress, factors that might predispose to cancer.
In previous work, Shima et al found that a hypomorphic allele of mouse Mcm4 (Mcm4Chaos3) caused high levels of GIN and extreme mammary cancer susceptibility in the C3HeB/FeJ background [17]. This provided the first concrete evidence that endogenous mutations in replication licensing machinery may have a causative role in cancer development. The ethylnitrosourea (ENU)-induced Mcm4Chaos3 point mutation changed PHE to ILE at residue 345 (Phe345Ile). This amino acid is conserved across diverse eukaryotes and is important for interaction with other MCMs [18]. Budding yeast engineered to bear the orthologous mutation exhibit DNA replication defects and GIN [17], [19]. Surprisingly, MEFs from Mcm4Chaos3 mice not only had reduced levels of MCM4, but also MCM7 [17], suggesting that the point mutation might destabilize the MCM2-7 complex. Subsequently, it was reported that mice containing 1/3 the normal level of MCM2 succumbed to lymphomas at a very young age, and had diverse stem cell proliferation defects [20]. These mice also had 27% reduced levels of MCM7 protein, and their cells exhibited decreased replication origin usage when under replication stress (treatment with hydroxyurea) conditions [21]. These studies imply that relatively modest decreases in any of the MCMs may be sufficient to cause cancer susceptibility, developmental defects, and GIN [20].
Here, we report that genetically-induced reductions of MCM levels in mice, achieved by breeding combinations of MCM2-7 alleles, caused several health-related defects including increased embryonic lethality, GIN, cancer susceptibility, growth retardation, defective cell proliferation, and hematopoiesis defects. Remarkably, genetic reduction of MCM3, which mediates nuclear export of excess MCM2-7 complexes in yeast [22], rescued many of these defects, presumably attributable to observed increases in chromatin-bound MCM levels. These data suggest that relatively minor misregulation or destabilization of MCM homeostasis can have serious consequences for health, viability and cancer susceptibility of animals.
To extend previous findings that Mcm4Chaos3Chaos3 cells exhibited decreases in MCM4 and MCM7 protein, and to determine if the decreased levels were differentially compartmentalized in the cell, we quantified soluble and chromatin-bound MCM2-7 levels in mouse embryonic fibroblasts (MEFs) by Western blot analysis. As shown in Figure 1A, all MCMs were decreased in both compartments by at least 40% compared to WT cells. Because Mcm4Chaos3/Chaos3 MEF cultures have slightly decreased proliferation and G2/M delay (Figure 1A and [17]), it is possible that the lower MCM levels in mutant MEFs are entirely attributable to growth defects. To test this, we assessed the levels of nuclear MCM2 in S-phase cells by flow cytometry (Figure 1B). Although MCM2 levels in WT and Mcm4Chaos3/Chaos3 G1 nuclei were essentially the same (P = .65; t-test), mutant cells transitioned from G1 to S with 40% less nuclear MCM2 content than in WT (P<.02; t-test). The levels of nuclear MCM2 in WT decreased through S phase more sharply than in mutants, which transitioned to G2 with only ∼23% less than controls (Figure 1B). This differential decline is apparent in the flow plots, where WT cells exhibit a greater downward slope in the S compartment (Figure 1B). The decreases in MCM2 from early to late S were 51% in WT and 38% in mutants. The MCM2 intra-S modulation phenomenon is also addressed in subsequent experiments. The marked differences in nuclear MCM2 concentration between actively proliferating (S-phase) WT and mutant cells indicates that a biochemical or regulatory basis, rather than a population skewing, underlies the differences in protein levels.
Another possible explanation for the coordinated decrease in MCMs is that the mutant MCM4Chaos3 protein destabilizes the MCM2-7 hexamer and causes subsequent degradation of uncomplexed MCMs. Other groups reported that knockdown of Mcm2, Mcm3, or Mcm5 in human cells decreased the amount of other chromatin-bound MCMs [15]–[16], leading to a similar proposition that the cause was MCM2-7 hexamer destabilization [16]. If true, then we would expect mRNA levels to be unchanged in mutant cells. To test this, we performed quantitative RT-PCR (qRT-PCR) analysis of Mcm2-7, and several control housekeeping genes in Mcm4Chaos3/Chaos3 MEFs. Analysis of 5 littermate pairs of primary MEF cultures revealed that transcript levels for each of these genes in mutant cells was 51–65% of WT, similar to the protein decreases (Figure 1C). Levels of mRNA in the 7 housekeeping genes analyzed were not altered significantly (Figure 1C, right panel). This data suggest that either reduced MCM4 levels per se, or defects resulting from the Mcm4Chaos3 allele, cause a decrease in the levels of all Mcm mRNAs. Interestingly, the mRNA reduction appears to occur post-transcriptionally, a phenomenon that is currently under investigation (Chuang and Schimenti, unpublished observations).
The Mcm4Chaos3 allele was identified in a forward genetic screen for mutations causing elevated micronuclei (MN) in red blood cells, an indicator of GIN [17]. While the altered MCM4Chaos3 protein may cause DNA replication errors as does a yeast allele engineered to contain the same amino acid change [19], it is also possible that the decrease in overall MCM levels in Mcm4Chaos3 mutants contributes to, or is primarily responsible for, elevated S-phase DNA damage and GIN as is seen in various cell culture models (see Introduction). To test this possibility, we generated mice from ES cells bearing gene trap insertions in Mcm2, Mcm3, Mcm6, and Mcm7 (Figure 2A; alleles are designated as Mcm#Gt). These gene traps are designed to disrupt gene expression by fusing the 5′ end of the endogenous mRNA (via use of a splice acceptor) to a vector-encoded reporter, resulting in a fusion protein lacking the C-terminal portion of the endogenous (MCM) protein. As with a previously-reported Mcm4 gene trap [17], each of these alleles proved to be recessive embryonic lethal (Figure S1). Furthermore, each allele appeared to be a null, since mRNA levels in heterozygous MEF cultures were ∼50% lower than WT controls (Figure 2B). To determine if heterozygosity for various Mcms caused pan-decreases in Mcm mRNA levels as does homozygosity for Mcm4Chaos3, mRNA levels for each of the Mcm2-7 genes were also quantified. Whereas Mcm2Gt/+ cells did show ∼20% decreases in the other Mcms, the Mcm3, Mcm4, Mcm6 and Mcm7 gene trap alleles did not (Figure 2B). Thus, it appears that the marked Mcm pan-decreases in Mcm4Chaos3/Chaos3 cells are not due to decreased Mcm4 RNA per se, but rather a response to replication defects cause by the mutant protein. Notably, the pan Mcm2-7 downregulation in Mcm2Gt/+ cells is consistent with the observation that MCM7 is decreased in Mcm2IRES-CreERT2/IRES-CreERT2 mice, although mRNA levels were not evaluated in that study [20].
After breeding the gene trap alleles into the C3HeB/FeJ genetic background for at least 2 generations (Mcm4Chaos3/Chaos3 females get mammary tumors in this background), blood MN levels were measured. Heterozygosity for each allele caused an increase in the fraction of cells with MN (Figure 2C). Compound heterozygosity further increased MN on average, as did heterozygosity for 3 or more gene traps (Figure 2C), indicating that genetically-based decreases in any of the MCMs precipitate GIN.
As outlined above, previous studies showed that reductions of particular MCMs in cells or mice reduces the levels of other MCMs, causing GIN, cancer, and developmental defects. However, the reduction in MCM levels required to precipitate these consequences, and whether there is a threshold effect, is unclear. To explore the consequences of incremental MCM reductions on viability and cancer in mice, we crossed the Mcm4Chaos3 and gene trap alleles into the same genome. In the case of Mcm2, there was a striking and highly significant shortfall of Mcm4Chaos3/Chaos3 Mcm2Gt/+ offspring at birth (Figure 3A; Figure S2). Heterozygosity for Mcm2Gt itself was not haploinsufficient, as indicated by Mendelian transmission of Mcm2Gt in crosses of heterozygotes to WT (119/250; χ2 = 0.448). These results demonstrate that there is a synthetic lethal interaction between Mcm4Chaos3 and Mcm2Gt that is related to MCM2 levels. Additionally, the surviving Mcm4Chaos3/Chaos3 Mcm2Gt/+ offspring were severely growth retarded; males weighed ∼50% less than Mcm4Chaos3/Chaos3 siblings (Figure 3B; this genotype causes disproportionate female lethality). Another indication of a quantitative MCM threshold effect is that C3H-Mcm4Chaos3/Chaos3 mice are developmentally normal, but Mcm4Chaos3/Gt animals die in utero or neonatally (Figure 3A) [23].
The synthetic interaction between Mcm4Chaos3 and Mcm2Gt might be specific, or it may reflect a general consequence of reduced replication licensing (and consequent elevated replication stress). We therefore tested whether hemizygosity for Mcm3, Mcm6 or Mcm7 would also cause synthetic phenotypes in the Mcm4Chaos3/Chaos3 background. The Mcm4Chaos3/Chaos3 Mcm6Gt/+ genotype caused highly penetrant embryonic lethality; only 10% of the expected number of such animals survived to birth (Figure 3A; Figure S2). The Mcm4Chaos3/Chaos3 Mcm7Gt/+ genotype caused both embryonic and postnatal lethality. The number of liveborns was ∼50% of the expected value, and only 8% of those (5/62) survived to weaning (Figure 3A; Figure S2). Additionally, as with Mcm2, hemizygosity for Mcm6Gt and Mcm7Gt in the Mcm4Chaos3/Chaos3 background caused growth retardation (Figure 3B). The decrease in male weight was ∼20% and ∼80% respectively, compared to Mcm4Chaos3/Chaos3 siblings at the oldest age measured (Mcm4Chaos3/Chaos3 Mcm7Gt/+ animals died before wean, so the oldest weights were taken at 10 dpp). In contrast to the synthetic phenotypes with Mcm2, 4, 6 and 7, there was no significant decrease in viability (Figure 3A) or weight (not shown) in Mcm4Chaos3/Chaos3 Mcm3Gt/+ mice. This seeming inconsistency is addressed in the following section.
As mentioned earlier, mice with ∼35% of WT MCM2 protein, but not 62%, showed early latency (10–12 week) lymphoma susceptibility [20]. To identify if there is a critical MCM threshold for cancer susceptibility, we aged a cohort of Mcm2Gt/+ mice, representing approximately intermediate MCM2 levels. As shown in Figure 4A, these animals did not show a dramatic cancer-related mortality in the first 12 months of life. However, we did find that ∼3/4 of these animals had tumors at death or necropsy by 18 months of age (data not shown). These combined data are suggestive of a potential gradient of susceptibility, but that there is a critical minimum threshold of MCM levels, between ∼35 and 50% in the case of MCM2, required to avoid early cancer and other developmental defects.
To further resolve this phenomenon, surviving Mcm4Chaos3/Chaos3 Mcm2Gt/+ mice were aged and monitored. They began dying at 2 months of age, and all were dead (or sacrificed when they appeared moribund) by 7 months (Figure 4A). Gross necropsy and histopathological analyses revealed or suggested lymphomas/leukemias in 20 of these animals (summarized in Table S1 with histological examples in Figure S3; detailed histopathology analysis of a T cell leukemic lymphoma is presented in Figure 4B). Six of these had chest tumors that were likely thymic lymphomas. The cause of death for the remaining 7 animals was undetermined. Consistent with previous studies [17], most Mcm4Chaos3/Chaos3 mice hadn't yet succumbed from tumors or other causes by 12 months of age. Additional animals of these genotypes are incorporated in Figure 6, but histopathological analyses weren't conducted. These data show clearly that removing a half dose of MCM2 from Mcm4Chaos3/Chaos3 cells is sufficient to produce greatly elevated cancer predisposition to the already-underrepresented survivors at wean. Mcm4Chaos3/Chaos3 Mcm2Gt/+ MEFs had 45% the amount of Mcm2 mRNA as Mcm4Chaos3/Chaos3 cells (Figure 7C), which already had a 38% reduction compared to WT (Figure 1). Thus, Mcm2 RNA was reduced to ∼17% of WT. To determine if elevated GIN might be responsible for the cancer susceptibility phenotype, we measured erythrocyte MN. Whereas the percentage of micronucleated RBCs in Mcm4Chaos3/Chaos3 mice was 4.18±0.26 (mean±SEM, N = 12), Mcm4Chaos3/Chaos3 Mcm2Gt/+ mice averaged 5.85±0.47 (N = 16), indicating a synergistic increase (P<0.01). Overall, the data support the notion that in whole animals, reduction of MCMs to under 50% of WT causes severe developmental and physiological problems.
The data reported here and elsewhere [17], [20] support a model where phenotypic severity is proportionally related to MCM concentrations. However, our genetic experiments uncovered one notable exception: hemizygosity for Mcm3 did not cause any severe haploinsufficiency phenotypes (increased lethality and decreased weight) as did Mcm2/6/7 in the Mcm4Chaos3/Chaos3 background, or Mcm4Gt in trans to Mcm4Chaos3 (Figure 3A; Figure S2). Since extreme reductions of MCM3 in cultured human cells caused GIN and cell cycle arrest [16], the absence of synthetic effects with McmChaos3 led us to hypothesize that either mice are more tolerant to lower levels of this particular MCM, or that MCM3 is present in a stoichiometric excess compared to the other MCMs, at least in a subset of cell types. To explore these issues we performed additional phenotype analyses, and also sought to uncover potential effects of MCM3 reduction by reducing other MCMs simultaneously.
Strikingly, rather than exacerbating the synthetic lethality in Mcm4Chaos3/Chaos3 Mcm2Gt/+ mice, Mcm3Gt heterozygosity significantly rescued their viability to 72.5% from 29.7% (Figure 5A and Fig S3). Not only was viability rescued, but also growth (weight) of Mcm4Chaos3/Chaos3 Mcm2Gt/+ Mcm3Gt/+ survivors compared to Mcm4Chaos3/Chaos3 Mcm2Gt/+ animals produced from the same matings (Figure 5B). Mcm3 hemizygosity also significantly rescued the near 100% lethality of Mcm4Chaos3/Gt animals (nearly 6 fold increased viability), and doubled the viability of Mcm4Chaos3/Chaos3 Mcm6Gt/+ mice (Figure 5A; Figure S3). Rescue of Mcm4Chaos3/Chaos3 Mcm7Gt/+ was not observed (not shown).
The rescue of the reduced growth phenotype by Mcm3 hemizygosity led us to evaluate the proliferation of compound mutant cells. Whereas Mcm4Chaos3/Chaos3 and Mcm4Chaos3/Chaos3 Mcm3Gt/+ primary MEFs proliferated at identical rates, Mcm4Chaos3/Chaos3 Mcm2Gt/+ MEFs showed a severe growth defect beginning ∼5 days in culture (Figure 5C). As with whole animals, MEF growth was partially but significantly rescued by Mcm3 hemizygosity.
Since the Mcm4Chaos3 and Mcm2Gt alleles causes elevated GIN (micronuclei in RBCs), we considered the possibility that the Mcm3 rescue effect might be related to an attentuation of GIN. Accordingly, we measured MN levels in Mcm4Chaos3/Chaos3 mice with different combinations of other Mcm mutations. As shown in Figure 5D, hemizygosity for Mcm2 and Mcm7 caused a significant elevation in MN levels, unlike Mcm3. However, the increased MN in Mcm4Chaos3/Chaos3 Mcm2Gt/+ was not rescued by Mcm3 hemizygosity. This suggests that the synthetic lethality and mouse/cell growth defects are not related to GIN per se. However, in the course of measuring MN in enucleated peripheral blood cells, we noticed that the ratio of CD71+ cells was significantly higher in both Mcm4Chaos3/Chaos3 Mcm2Gt/+ and Mcm4Chaos3/Chaos3 Mcm7Gt/+ mice (3.3 and 6.2 fold, respectively; Figure 5E). This increase in the ratio of reticulocytes (erythrocyte precursors; immature RBCs) to total RBCs is characteristic of anemia. Hemizygosity for Mcm3, which alone had no effect on CD71 ratios of Chaos3 mice, corrected completely this abnormal phenotype in Mcm4Chaos3/Chaos3 Mcm2Gt/+animals (Figure 5E).
Because MCM2-depleted mice were reported to have stem cell defects [20], and Mcm4Chaos3/Chaos3 Mcm#Gt/+ mice had clear developmental abnormalities, we examined the efficiency of reprogramming mutant MEFs into induced pluripotent stem cells (iPS). The efficiency was quantified using either : 1) iPS-like colony formation, or 2) cells counts of SSEA1 and LIN28 positive cells by flow cytometry. Both gave similar results. Mcm4Chaos3/Chaos3 Mcm2Gt/+ cells were severely compromised in the ability to form iPS cells compared to Mcm4Chaos3/Chaos3 (∼200 fold less efficient; Figure 5F). However, additionally reducing Mcm3 by 50% increased iPS formation from both Mcm4Chaos3/Chaos3and Mcm4Chaos3/Chaos3 Mcm2Gt/+ MEFs by ∼2.5 and 10 fold, respectively.
Finally, we found that reduced MCM3 levels could rescue the cancer susceptibility of two different Chaos3 models. As shown earlier (Figure 4), Mcm4Chaos3/Chaos3 Mcm2Gt/+ mice were highly cancer-prone with an average latency of <4 months. When a dose of Mcm3 was removed from mice of this genotype, lifespan was extended dramatically in both sexes as a consequence of delayed cancer onset, and the cancer spectrum shifted from lymphoma/thymoma towards mammary tumors (Figure 6A). Additionally, hemizygosity of Mcm3 delayed (or eliminated) the onset of mammary tumorigenesis in Mcm4Chaos3/Chaos3 females by ∼4 or more months (Figure 6B). However, although Mcm3 hemizygosity rescued viability of Mcm4Chaos3/Gt mice (Figure 5A), these animals were cancer prone with a shorter latency (by ∼6 months) and different spectrum (primarily lymphomas) than Mcm4Chaos3 homozygotes.
We considered two possibilities to explain the surprising phenotypic rescues of reduced MCM genotypes (Mcm4Chaos3/Chaos3 ; Mcm4Chaos3/Chaos3 Mcm2/6Gt/+ ; Mcm4Chaos3/Gt) by additional MCM3 reduction (Mcm3Gt/+). One is that the phenotypes are related to altered stoichiometry of MCM monomers, and that disproportionally high amounts of MCM3 relative to MCM4 and MCM2/6/7 have a dominant negative effect. However, as demonstrated above, levels of MCM3 are proportionally reduced in Mcm4Chaos3/Chaos3 cells (Figure 1). The second possibility is that decreased levels of MCM3 leads to a favorable change in the amounts or subcellular localization of MCMs. Various experiments have indicated that MCM2-7 hexamers or subcomplexes must be assembled in the cytoplasm before nuclear import in yeast [4], and in mice, nuclear import appears to require MCM2 and MCM3 [24]. MCMs shuttle between the nucleus and cytoplasm during the cell cycle in S. cerevisiae. Although in most other organisms MCMs are reported to be predominantly and constitutively nuclear localized throughout the cell cycle, dynamic redistribution between the nucleus and cytoplasm has been observed in hormonally-treated mouse uterine cells [25]. In budding yeast, nuclear export is dependent upon Mcm3, which has a nuclear export signal (NES) that is recognized by Cdc28 to promote export of MCM2-7 [22]. Analysis of mouse and human MCM3 using NES prediction software (www.cbs.dtu.dk/services/NetNES/) [26] revealed the presence of homologously-positioned, leucine-rich potential NESs (Figure 7A). Therefore, we hypothesized that the rescue of phenotypes by Mcm3 hemizygosity is due to decreased MCM protein export from the nucleus, or alternatively, increased nuclear import or stabilization that allows greater access of all MCMs for licensing chromatin.
To explore this hypothesis, we performed Western blot analysis of MCM levels in Mcm4Chaos3/Chaos3 MEFs with or without the Mcm3Gt and/or Mcm2Gt alleles, and examined the effects of Mcm3 dosage on the levels of nuclear and chromatin-bound MCM2 and MCM4. The results are presented in Figure 7B. In all cases, the genetic reductions of Mcm2 and Mcm3 led to corresponding decreases in the cognate mRNA levels (Figure 7C), with only minor additional decreases of other MCM mRNAs (beyond that already caused by homozygosity for Mcm4Chaos3) occuring in the context of Mcm2 hemizygosity (similar to Mcm2Gt/+ MEFs in Figure 2B). The overall levels of total, nuclear, and chromatin-bound MCM2 and MCM4 were unaffected by hemizygosity of Mcm3 in Mcm4Chaos3/Chaos3 cells (Figure 7B). When Mcm2 levels were genetically reduced by half, a condition causing the severe phenotypic effects described earlier, this caused a marked decrease in the level of chromatin-bound MCM3 and MCM4 (in addition to MCM2 itself), although total and nuclear MCM3/4 levels were affected to a lower degree or not at all. Strikingly, the decreased levels of chromatin-bound MCM2/3/4 in Mcm4Chaos3/Chaos3 Mcm2Gt/+ MEFs were reversed by Mcm3 heterozygosity, but levels of total MCM2 and MCM4 were not restored. The increase of chromatin-bound MCMs occured despite the presence of less MCM3, suggesting that MCM3 is present at levels in excess of that needed to bind chromatin, presumably for pre-RC formation in the context of the MCM2-7 hexamer. In conclusion, a 50% reduction in total MCM3 increases MCM2/4 loading onto chromatin when MCM2 is otherwise limiting, and this rescue is associated with amelioration of several phenotypes.
We found that elevation of nuclear MCMs in the Mcm3Gt/+ MEFs was often (as shown in Figure 7B), but not consistently elevated across samples by Western analysis (not shown). Therefore, we quantified MCM2 during the cell cycle by flow cytometric analysis of nuclei from 7 replicate MEF cultures. Similar to WT MEFs (examples in Figure 1B), NIH3T3 cells showed a decrease of nuclear MCM2 during S phase progression (Figure 7D, left panel). However, all genotypes with in the Mcm4Chaos3/Chaos3 background had a reduced decline. Thus, for comparative quantitation across genotypes, we compared the levels of MCM2 levels at the beginning of G1 vs. that in S phase (regions used for these calculations are indicated in the left panel), using the calculation described in the Figure 7 legend. The data are graphed in the right panel. The data revealed that regardless of genotype, the difference in average amounts of nuclear MCM2 at the beginning and end of G1 (ΔG1) did not vary. Compared to Mcm4Chaos3/Chaos3, cells lacking 1 dose of Mcm2 had relatively lower levels of S phase MCM2 (ΔS) compared to early G1. Additional removal of an Mcm3 dose partially rescued the ΔS value, indicating that these cells had ∼16% more nuclear MCM2 in S phase compared to Mcm4Chaos3/Chaos3 cells hemizygous for Mcm2 alone, despite overall reduced MCM2 levels in the cell (Figure 7B, left panel).
MCM2-7 proteins exist abundantly in proliferating cells and are bound to chromatin in amounts exceeding that required to license all replication origins that initiate DNA synthesis [9]–[12], [14]. The role of excess chromatin-bound MCM2-7 has been a mystery referred to as the “MCM paradox” [27], perpetuated by observations that drastic MCM reductions in certain systems can be compatible with normal DNA replication or cell proliferation [13], [28]–[30]. However, these circumstances are not universal, and reductions are not entirely without consequences. Early studies showed that a reduction in MCMs resulted in decreased usage of certain ARSs [12] and conferred genome instability [31] in yeasts. In cell culture systems, depletion of certain MCMs have been found to cause cell cycle defects, checkpoint abberations and GIN [13], [16]–[17], [29], [32].
Recent work has shed light on aspects of the MCM paradox. Using Xenopus egg extracts attenuated for licensing by addition of geminin (an inhibitor of CDT1, which is required for MCM loading onto origins), one study proposed that excess chromatin-bound MCM2-7 complexes license “dormant” origins that can be activated to rescue stalled or damaged replication forks, a situation that can become important under conditions of replication stress [11]. Similar results were subsequently reported for human cells depleted of MCMs by siRNA [15]–[16], and for replication stressed MCM2-deficient MEFs [21]. Our finding that nuclear MCM2 levels decrease as S-phase progresses, and moreso in WT than in Mcm4Chaos3/Chaos3 MEFs, is consistent with the dormant origin hypothesis. The decrease may reflect displacement of dormant hexamers by active replisomes, followed by subsequent degradation or nuclear export. If WT nuclei have more dormant licensed origins than Chaos3 mutants, then WT cells would be expected show a greater loss of MCMs.
The isolation of Mcm4Chaos3 provided the first demonstration that mutant alleles of essential replication licensing proteins can cause GIN and cancer [17]. Diploid budding yeast containing the same amino acid change in scMcm4 as the mouse Mcm4Chaos3 exhibited Rad9-dependent G2/M delay (Rad9 is a DNA damage checkpoint protein), elevated mitotic recombination, chromosome rearrangements, and intralocus mutations [19] (Li, X. and Tye, B., personal communication). One explanation for these outcomes is that the Chaos3 mutation impairs MCM4 biochemically in a manner leading to elevated replication fork defects, and that these defects lead to the GIN and cancer phenotypes. Alternatively, and/or in addition, the observed associated pan-reductions of MCMs in mouse cells [17] raised the possibility that decreased replication licensing might be the primary or ancillary cause for the mouse phenotypes.
The subsequent finding that mice (Mcm2IRES-CreERT) containing ∼1/3 the normal level of MCM2 had GIN and and cancer lent support for the idea that reductions in MCMs contribute to the Chaos3 phenotypes [20]. Although amounts of all MCMs were not investigated in Mcm2IRES-CreERT/IRES-CreERT mice, 65% reduction of MCM2 caused a reduction of dormant replication origins in MEFs that were replication stressed by hydroxyurea [21]. In Mcm4Chaos3/Chaos3 mice, we hypothesize that in the context of Mcm2, 6 or 7 heterozygosity, which further reduces overall and chromatin-bound MCM levels below that already caused by Mcm4Chaos3 (measured to be <20% of WT mRNA levels for Mcm2), MCMs are reduced to a degree that compromises cell proliferation. This then translates into the various developmental defects and increased cancer susceptibility we observed. Whatever the exact mechanistic cause of these phenotypes, it is clear that the phenotypes are related to reduction of one or more MCMs below a threshold level that is <50%. The severe developmental consequences of MCM depletion in mice suggests that certain cell types in the developing embryo are highly sensitive to the effects of replicative stress, and/or that relatively minor cell growth perturbations of such cells are not well-tolerated in the context of complex, rapidly-occuring developmental events. The molecular basis for these phenotypes does not appear to be directly related to GIN, because whereas Mcm3 hemizygosity rescued several phenotypes, and delayed cancer latency in Mcm4Chaos3/Chaos3 mice, it did not concommitantly decrease MN. This suggests that phenotypes such as decreased proliferation and embryonic death are caused by genetically-induced replication stress, moreso (or in addition to) than GIN alone.
Our genetic studies indicate that there is a quantitative MCM threshold required for embryonic viability, as demonstrated by the synthetic lethalities we observed when combining homozygosity of Mcm4Chaos3 with Mcm2Gt, Mcm6Gt or Mcm7Gt heterozygosity, but not in the heterozygous single mutants. Additionally, the Mcm4Chaos3/Gt genotype, which reduced MCM levels below 50%, caused embryonic and neonatal lethality [17]. Underscoring the exquisite sensitivity of whole animals to subtle perturbations in the DNA replication machinery were the remarkable phenotypic rescues (viability, growth, iPS efficiency, etc.) by Mcm3 hemizygosity. The decreased MCM dosage led to increases in S phase nuclear MCMs and chromatin-bound MCMs, presumably reflecting increased replication origin formation. The various single and compound mutants described here and elsewhere [20], which show that 50% reductions of any one MCM is well-tolerated but decreases of ∼2/3 are not, supports the idea of a threshold effect, and suggests that the threshold lies somewhere between 1/3 and 1/2 of normal MCM levels (at least in the cases of MCM2, MCM6 and MCM7).
These results also emphasize the importance of relevant physiological models, both in general and with respect to the MCMs. RNAi knockdown of MCM3 in human cells to ∼3% normal levels was still compatible with normal short-term proliferation, although the cells had GIN and high sensitivity to replication stress [16]. It is doubtful such a drastic situation would be recapitulated in vivo (it would likely result in embryonic lethality as in Mcm3Gt/Gt mice). Nevertheless, it is noteworthy in that study that MCM3 depletion was better tolerated than knockdowns of any other member of the replicative helicase.
The finding that reductions in MCM3 rescued MCM2/4/6 depletion phenotypes lends insight into dynamics and regulation of mammalian DNA replication. In budding yeast, MCMs shuttle between the nucleus and cytoplasm during the cell cycle. MCM2-7 multimers must be assembled in the cytoplasm before being imported into the nucleus during G1 phase [4]. The MCM2-7 importation is dependent upon synergistic nuclear localization signals (NLS) on Mcm2 and Mcm3 [22]. In order to prevent over-replication of the genome, MCMs are exported from the nucleus during S, G2 and M [4]. This export is dependent upon Mcm3, which has a nuclear export signal (NES) that is recognized by Cdc28 to promote MCM2-7 export in a Crm1-dependent manner [22].
In contrast to budding yeast, MCMs that have been studied (MCM2/3/7) are primarily nuclear-localized throughout the cell cycle in metazoans and in fission yeast [4]. Upon dissociation from chromatin during S phase, MCM2-7 complexes are reported to remain in the nucleus but are sequestered via attachement to the nuclear envelope or other nuclear structures [24], [33]–[35]. Interestingly, mcm mutations in fission yeast that disrupt intact MCM2-7 heterohexamers triggers active redistribution of MCMs to the cytoplasm [36]. Additionally, re-distribution of MCMs between the cytoplasmic and nuclear compartments has been observed in hormonally-treated mouse uterine cells [25].
Our observations support the idea that intracellular re-distribution of MCMs also occurs in mammals, and that it is an important regulatory process. Staining of MCM2 in intact nuclei of normal NIH 3T3 fibroblasts and MEFs show a steady decline (but not elimination) as S phase progresses. Furthermore, it appears that the process of nuclear MCM2 elimination during S phase is regulated, since in situations of decreased MCMs (as in the Mcm4Chaos3/Chaos3 mutant), there is decreased loss of nuclear MCM2 during S phase.
Three lines of experimentation implicate MCM3 as playing a key role in regulating intracellular MCM localization: 1) Rescue of reduced-MCM phenotypes by genetic reduction of MCM3; 2) Increased S-phase nuclear MCM2 by Mcm3 hemizygosity in MCM-depleted cells (Figure 7D); and increased chromatin-bound MCM2/4 by Mcm3 hemizygosity in MCM-depleted cells. Our data suggests that MCM3 acts as a negative regulator that prevents re-assembly or reloading of MCM complexes as they dissociate from DNA during replication. As described earlier, mouse and human MCM3 have predicted NESs in similar positions of their primary amino acid sequences as do the yeast genes. Thus, one explanation for these phenomena is that decreased MCM3 suppresses MCM2-7 nuclear export, which occurs normally and which may be accentuated by the Chaos3 mutation in a fashion analogous to mcm mutant fission yeast discussed above [36]. This would effectively increase the amounts of MCMs available for replication licensing. More work is required to determine if the rescue mechanism is indeed related to a decrease in MCMs export, as opposed to direct or indirect involvement in other events such as increased nuclear import or enhanced chromatin loading.
With respect to the early lymphoma susceptibility phenotype in Mcm4Chaos3/Chaos3 Mcm2Gt/+ mice, it is unclear whether the type of tumor is dictated primarily by the particular Mcm depletion (in this case MCM2, thus resembling Mcm2IRES-CreERT2/IRES-CreERT2 animals), the genetic background, or the age of particular cancer onset (if animals die of thymic lymphoma at an early age, they will be unable to manifest later-arising mammary tumors). The compound mutant mice used for the aging aspects of this study were bred to at least the N3 generation in strain C3H. Mcm4Chaos3/Chaos3 mice congenic in this background are predisposed exclusively to mammary tumors, whereas lymphomas were observed in mutants of mixed background [17]. Presently, we favor the idea that genetic background and age of tumor type onset are primary determinants of the cancers that arise in the mice we have studied thus far. Genetic background has also been reported to influence tumor latency in MCM2-deficient mice [21].
The MCM2-7 pan-reduction in Chaos3 cells is consistent with other studies involving mutation or knockdown of a single MCM in mammalian cells [16], [20], [29], [37]. In these examples of parallel MCM decreases, the general assumption is that there is hexamer destabilization or impaired MCM chromatin loading followed by degradation of monomers. However, we found that the protein decreases are related to decreased mRNA levels. These large (∼40%) decreases do not appear to be attributable to transcriptional alterations from cell cycle disruptions (these cells have a small elevation in the G2/M population), but rather occur at the post-transcriptional level (unpublished observations). Since we also found that MEFs carrying only 1 functional Mcm2 allele caused ∼20% decreases of Mcm3-7 mRNAs, it is possible that mRNA downregulation drove MCM reductions in these other model systems. However, the mechanism for coordinated mRNA regulation, and what triggers it, is a mystery that we are currently investigating.
Our data contribute to a growing body of data that replication stress, which can occur via perturbations of the DNA replication machinery, plays a significant role in driving cancer [38]–[41]. While the Mcm4Chaos3 mutation is an unique case, the deleterious consequences of MCM reductions suggest that genetically-based variability in DNA replication factors can have physiological consequences. Such variability in functions or levels may be caused by Mendelian mutations or multigenic allele interactions. Mutations affecting transcriptional activity of one or more Mcms, which might occur in non-coding cis-linked sequences or unlinked transcription factors, could have such effects. This has implications for cancer genome resequencing projects, whereby such mutations would not be obviously associated with MCM expression. The allelic collection we generated, when used alone or in combination with each other or Mcm4Chaos3/Chaos3 mice, allow the generation of mouse models with a graded range of MCM levels. These should be valuable for investigations into the impact of replication stress on animal development, cancer formation, and cellular homeostasis.
MEFs from 12.5- to 14.5-dpc embryos were cultured in DMEM+10% FBS, 2 mM GlutaMAX, and penicillin-streptomycin (100 units/ml). Assays were conducted on cells at early passages (up to P3). For cell proliferation assays, 5×104 cells were seeded per well of a 6 well plate. They were then cultured and harvested at the indicated time points to perform cell counts.
Doxycycline inducible lentiviral vectors [42] were prepared by co-transecting viral packaging plasmids psPAX2 and and pMD2.G along with vectors encoding rtTA, Oct4, Sox2, Klf4, or c-Myc (plasmids were obtained from Addgene.org, serial numbers 12259, 12260, 20323, 20322, 20324, and 20326) into 293T cells using TransIT-Lt1 transfection reagent (Mirus). Viral supernatants were collected at 48 and 72 hours, and concentrated using a 30kd NMWL centrifugal concentrator. MEFs from 13.5d embryos, up to P3, were seeded to gelatin coated tissue plates at a density of 6.75×103 cells/cm2 and allowed to attach in standard MEF media for 24 hours before infection with lentiviral vectors. After 24 hours incubation the culture media was changed to KO-DMEM supplemented with 15% KO serum replacement (Gibco), recombinant LIF, 2 µg/mL doxycycline (Sigma), 100 µm MEM non-essential amino acids solution, 2mM GlutaMax, 100 units/mL penicillin and 100 µg/mL streptomycin (Gibco). The induction media was refreshed daily for 13 days until the cells were passaged to 100 mm plates prepared with irradiated feeders. Cells were cultured for an additional 10 days in the induction media in the absence of doxycyline before iPS colony counting, cell counts, and flow cytometry.
For flow cytometric quantification of iPS cells derived from reprogramming of MEFs, ∼1×106 cells were trypsinized for 10 minutes, then washed twice with cold PBS. They were gently but completely resuspended in 1ml of 4% paraformaldehyde in PBS at room temperature for 30 minutes. The fixed cells were pelleted by centrifugation at 500×G for 2 minutes and washed twice with 10 ml TBS-TX (0.1% Triton X-100) buffer. For antibody staining, the cells were blocked with 1ml TBS-TX buffer with 1% BSA for 15 min at room temperature, then stained with primary “stemness” antibodies (monoclonal anti-SSEA1, Millipore; rabbit polyclonal anti-LIN28, Abcam) for 60 min, washed twice, then secondary antibody was applied for 60 minutes. Immunolabeled cells were analyzed by flow cytometry using a 488nm laser. Secondary antibodies were goat anti-mouse IgG-FITC (South Biotech) and goat anti-rabbit IgG-594 (Molecular Probes). Cells were considered to be iPS cells if they were LIN28/SSEA1 positive. Calibration of the flow cytometer and gates were set using untransfected MEFs as negative controls, and v6.4 ES cells as positive controls.
For quantification by colony formation, plates containing the passaged reprogrammed cells were examined microscopically at 20×, and 4 fields were scored and averaged. Colonies were considered as iPS clones based on morphological criteria: well defined border, three-dimensionality, and tight packing of cells.
Micronucleus assays, which include CD71 staining, were performed essentially as described [43].
MEFs were plated at 4×106 cells/150 mm culture dish for 60 hr, trypsinized, then resuspended in 1ml PBS. To the suspension was added TX-NE (320 mM sucrose, 7.5 mM MgCl2, 10 mM HEPES, 1% Triton X-100, and a protease inhibitor cocktail). The cells were gently vortexed for 10 seconds and incubated on ice for 30 min. Dounce homogenization was unnecessary. Nuclei were then pelleted by centrifugation at 500×G for 2 min and washed twice with 10 ml TX-NE, then resuspended in 1ml TX-NE. Nuclei yield and integrity was monitored microscopically with trypan blue staining. The nuclei were fixed by adding 15ml cold methanol for 60 min on ice. The fixed nuclei were pelleted by centrifugation at 500×G for 2 min, then washed twice with 10 ml TBS-TX (0.1% TX-100). 1×106 nuclei were placed into 1.5ml tubes in 1ml TBS-TX buffer+1% BSA for 15 min at room temperature. The primary antibody (Rabbit anti-mouse MCM2) was added for 60 min, then secondary antibody (FITC goat anti-rabbit) was added for 60 min. Finally, the nuclei were stained with propidium iodide (PI), and RNAse treated (batches optimized empirically) for 30 mins. Immunolabeled nuclei were analyzed by flow cytometry (using a BD FACSCalibur cytometer with CellQuest software), exciting the PI and FITC with a 488nm laser.
ES cell lines containing gene trap insertions in Mcm genes were obtained from Bay Genomics [Mcm3 (RRR002), Mcm6 (YHD248), Mcm7 (YTA285)] or the Sanger Institute [Mcm2 (ABO178)]. The Mcm4 line was previously reported [17]. Allele names are abbreviated as, for example, Mcm3Gt instead of the full name Mcm3Gt(RRR002)Byg. All of the original ES cells were of strain 129 origin, and the alleles were backcrossed into C3HeB/FeJ for ≥4 generations.
To identify the exact insertion sites of the gene trap vectors, a “primer walking” procedure was used. This involved priming PCR reactions with :1) a fixed vector primer, and 2) one of a series of primers series corresponding to the intron in which the vector presumably integrated. PCR products were then sequenced. Genotyping of gene-trap-bearing mice was performed either by PCR amplification of the neomycin resistance gene within the vector, or by using insertion-specific assays (Table S2).
Cytosolic and chromatin-bound protein was extracted as described [44]. Antibody binding was detected with a Pierce ECL kit. Band were quantified using NIH Image J software. Antibodies- aMCM2: ab31159 (Abcam); aMCM3: 4012 (Cell Signaling); aMCM4: ab4459 (Abcam); aMCM5: NB100-78261 (Novus); aMCM6: NB100-78262 (Novus); aMCM7: ab2360 (Abcam); aBeta-actin: A1978 (Sigma); aTBP: NB500-700 (Novus).
Total RNA from P1 MEFs was DNAse I treated, then cDNA was synthesized from 1 µg of total RNA using the Invitrogen SuperScript III ReverseTranscriptase kit with the supplied Olige-dT or random-hexamer primers. qPCR reactions were performed in triplicate on 1 ng or 10 ng of cDNA by using the SYBR power green RT-PCR Master kit (Applied Biosystems; 40 cycles at 95°C for 10 s and at 60°C for 1 min), and real-time detection was performed on an ABI PRISM 7300 and analyzed with Geneamp 5700 software. The specificity of the PCR amplification procedures was checked with a heat-dissociation step (from 60°C to 95°C) at the end of the run and by gel electrophoresis. Results were standardized to β-actin. The PCR primers are listed in Table S1.
|
10.1371/journal.pbio.0050274 | Hypoinsulinemia Regulates Amphetamine-Induced Reverse Transport of Dopamine | The behavioral effects of psychomotor stimulants such as amphetamine (AMPH) arise from their ability to elicit increases in extracellular dopamine (DA). These AMPH-induced increases are achieved by DA transporter (DAT)-mediated transmitter efflux. Recently, we have shown that AMPH self-administration is reduced in rats that have been depleted of insulin with the diabetogenic agent streptozotocin (STZ). In vitro studies suggest that hypoinsulinemia may regulate the actions of AMPH by inhibiting the insulin downstream effectors phosphotidylinositol 3-kinase (PI3K) and protein kinase B (PKB, or Akt), which we have previously shown are able to fine-tune DAT cell-surface expression. Here, we demonstrate that striatal Akt function, as well as DAT cell-surface expression, are significantly reduced by STZ. In addition, our data show that the release of DA, determined by high-speed chronoamperometry (HSCA) in the striatum, in response to AMPH, is severely impaired in these insulin-deficient rats. Importantly, selective inhibition of PI3K with LY294002 within the striatum results in a profound reduction in the subsequent potential for AMPH to evoke DA efflux. Consistent with our biochemical and in vivo electrochemical data, findings from functional magnetic resonance imaging experiments reveal that the ability of AMPH to elicit positive blood oxygen level–dependent signal changes in the striatum is significantly blunted in STZ-treated rats. Finally, local infusion of insulin into the striatum of STZ-treated animals significantly recovers the ability of AMPH to stimulate DA release as measured by high-speed chronoamperometry. The present studies establish that PI3K signaling regulates the neurochemical actions of AMPH-like psychomotor stimulants. These data suggest that insulin signaling pathways may represent a novel mechanism for regulating DA transmission, one which may be targeted for the treatment of AMPH abuse and potentially other dopaminergic disorders.
| Abuse of psychostimulants such as amphetamine remains a serious public health concern. Amphetamines mediate their behavioral effects by stimulating dopaminergic signaling throughout reward circuits of the brain. This property of amphetamine relies on its actions at the dopamine transporter (DAT), a presynaptic plasma membrane protein that is responsible for the reuptake of extracellular dopamine. Recently, we and others have revealed the novel ability of insulin signaling pathways in the brain to regulate DAT function as well as the cellular and behavioral actions of amphetamine. Here we used a model of Type I diabetes in rats to uncover how insulin signaling regulates DAT-mediated amphetamine effects. We show that by depleting insulin, or through selective inhibition of insulin signaling, we can severely attenuate amphetamine-induced dopamine release and impair DAT function. Our findings demonstrate in vivo the novel ability of insulin signaling to dynamically influence the neuronal effects of amphetamine-like psychostimulants. Therefore, the insulin signaling pathway, through its unique regulation of brain dopamine, may be targeted for the treatment of amphetamine abuse.
| Virtually all major classes of abused drugs share an ability to enhance dopamine (DA) transmission throughout midbrain reward centers [1,2]. Once DA is released into the synapse, the DA transporter (DAT) is the primary mechanism for clearing the transmitter from the extracellular space, particularly within the striatum [3–5]. DAT is a target of multiple psychomotor stimulants including cocaine, methamphetamine and amphetamine (AMPH) [6]. Dysregulation of DAT function has been implicated in a wide variety of neuropsychiatric pathologies, including attention-deficit hyperactivity disorder, depression and bipolar disorder [1,7].
DA clearance is dynamically modulated by several signaling pathways [8–10]. Importantly, recent studies suggest a unique role for insulin and insulin-like growth factors (e.g., IGF1 and IGF2) in this modulation [11–14]. Insulin receptors (IRs) and receptors for IGF1–2 are found on DAT-expressing midbrain DA neurons [15–18]. Insulin and IGF1–2 receptors function as receptor tyrosine kinases (RTKs), which have been shown to regulate the activity of a variety of neurotransmitter transporters [19–22]. Additionally, RTKs are known to stimulate phosphotidylinositol 3-kinase (PI3K) signaling, which in turn activates protein kinase B (PKB), also known as Akt [23,24]. Akt is a central player in insulin and growth factor signaling and a regulator of several cellular functions including cell growth and apoptosis [25]. Recently, the PI3K/Akt signaling pathway has been shown to regulate DA clearance [11] and has been implicated in cocaine sensitization [26], alcohol tolerance [27] and opioid dependence [28]. The mechanism underlying the regulation of DA clearance by PI3K seems to rely on DAT trafficking, as Garcia et al. [13] and Wei et al. [29] recently demonstrated that Akt activity is critical for sustaining human DAT (hDAT) membrane expression and function.
In vivo evidence supporting insulin and PI3K signaling pathways in the control of DA clearance comes from Patterson et al. [30], who demonstrated that in rats, hypoinsulinemia induced by food deprivation decreases the maximum velocity [Vmax] for DA uptake (with no significant change in the affinity constant [Km] for DA), as determined by rotating disk voltammetry on striatal suspensions. Consistently, the uptake of DA, as determined ex vivo by using striatal synaptosomes and in vivo by high-speed chronoamperometry (HSCA), is severely reduced in rats previously depleted of insulin with the diabetogenic agent streptozotocin (STZ) [14].
AMPH-like stimulants are actively transported by catecholamine carriers such as DAT [6]. As substrates, AMPHs not only competitively inhibit DA reuptake and thereby increase synaptic DA, but also promote reversal of transport, resulting in efflux of DA via the DAT [6]. This efflux results in an increase in extracellular DA and is believed to be of major importance for the psychomotor stimulant properties of AMPHs [6]. Because insulin and PI3K signaling have been shown to fine-tune DAT cell surface expression [13,29], it is possible that inhibition of PI3K signaling in vivo, by reducing DAT cell surface expression, inhibits AMPH-induced DA efflux and, hence, its behavioral effects. The ablation of pancreatic β cells by STZ in rats is a model of insulin depletion, and as such, we hypothesized that PI3K signaling in the brain, as well as DAT cell surface expression and possibly DAT-mediated behavioral effects of AMPH, would be reduced following STZ pretreatment. In support of our hypothesis are studies showing that, insulin-depleted, diabetic rodents have significantly reduced basal locomotor activity [14,31,32] and are resistant to the motor stimulant properties of AMPH and other related psychomotor stimulants [31,33,34]. Likewise, the reinforcing potential of AMPH, as determined by the daily maintenance of intravenous AMPH self-administration, is significantly blunted in the STZ model of hypoinsulinemia [12]. Therefore, in light of these data, it is possible that the regulation of AMPH-induced DA efflux—promoted by insulin and PI3K signaling—is mediated by changes in DAT cell surface expression.
Here we show that pharmacological manipulation of the PI3K signaling pathway caused by hypoinsulinemic conditions or selective pharmacological inhibition/activation of PI3K dramatically regulates the ability of AMPH to evoke DAT-mediated DA release in the striatum, as determined by HSCA. Consistently, in hypoinsulinemic rats we observed a blunting of AMPH-evoked striatal activation measured by functional magnetic resonance imaging (fMRI). We couple these findings with biochemical data showing that PI3K/Akt signaling is reduced under hypoinsulinemic conditions, as is the cell surface distribution of the DAT within striatum. Collectively, these data support the novel concept that insulin signaling—possibly through PI3K and/or Akt—plays a critical role in DA homeostasis by regulating DA clearance and the increases in extracellular DA induced by AMPH-like psychomotor stimulants.
PI3K signaling, which is stimulated by activation of IRs and other RTKs [23], plays a critical role in the maintenance of DA clearance and DAT cell surface expression [11,13,21,35]. Therefore, it is conceivable that PI3K signaling and ultimately Akt, by fine-tuning DAT expression at the plasma membrane [13,29], regulate the ability of AMPH to cause DAT-mediated DA efflux.
To test this hypothesis, we first altered PI3K signaling in vivo by depleting circulating plasma levels of insulin, a potent hormonal activator of the PI3K/Akt pathway [23,24], using the antibiotic STZ [36]. We administered a single dose of STZ (65 mg/kg) by tail vein injection at least 7 d prior to experiments. This regimen led to a significant increase in blood glucose levels: 532 ± 39 mg/dl (STZ-treated rats) versus 108 ± 21 mg/dl (untreated controls) (p < 0.001, Students t-test; n = 11–12 rats). Radioimmunoassay data from our laboratory suggest that STZ reduces striatal levels of insulin by at least 50% (M. Shiota, Vanderbilt Diabetes Center, unpublished data).
Importantly, in striatum—a region that contains abundant DATs [37–39] and IRs [15,17,18] and that participates in the reward pathway [1,2]—STZ treatment inhibited Akt activity. To assess Akt activity in these studies, we measured its ability to phosphorylate in vitro GSK3α [29]. As seen in Figure 1, STZ treatment reduces basal Akt activity, reflected by a decreased phosphorylation of GSK3α with respect to untreated controls. In three independent experiments, STZ treatment in rats led to a 44 ± 16% decrease in Akt activity measured from striatal synaptosomes (Figure 1B), suggesting that the STZ treatment significantly downregulates basal PI3K signaling in striatum.
To verify whether inhibition of PI3K signaling induced by STZ treatment correlates with changes in AMPH-induced DA efflux, we used HSCA to measure the release and clearance kinetics of striatal DA in vivo [14,40]. One week after STZ treatment (blood glucose: 495 ± 31 mg/dl [STZ-treated rats] versus 115 ± 5 mg/dl [saline-treated controls], p < 0.001, Students t-test; n = 6), HSCA recordings were carried out. In saline-treated rats, ejection of AMPH (400 μM/125 nl) caused a robust release of DA that was rapidly cleared from the extracellular space (Figure 2A). In contrast, AMPH elicited significantly less DA release in STZ-treated rats, and the released DA was cleared more slowly in these animals (Figure 2A). The slope of the rising portion of the DA signal indicates the rate of DAT-mediated DA efflux, which is primarily dependent on the affinity and turnover rate of DA and is independent of DA content [14,40]. Analysis of the rising phase of the trace revealed that DA efflux rates in STZ-treated rats were severely attenuated compared with those of saline-treated control rats (Figure 2B). STZ-treated rats also had a significantly lower amount of released DA (Figure 2C). Furthermore, STZ-treated animals also displayed significant deficits in DAT-mediated DA clearance, indicated by the reduced slope of the descending phase of the DA signal compared to saline-treated rats (Figure 2D). These data suggest that under hypoinsulinemic conditions, in which PI3K signaling is diminished, the ability of AMPH to cause DA efflux is impaired, possibly by decreasing DAT function.
It is possible that factors other than PI3K signaling (e.g., altered blood glucose levels) might contribute to the blunted AMPH-induced DA release caused by STZ treatment. To address this concern, we selectively inhibited PI3K activity within the striatum of naive rats using LY294002 and then recorded AMPH-induced DA efflux in this region using HSCA. Figure 3 shows the effect of LY294002 pretreatment on AMPH-induced DA release. LY294002 (1 mM/125 nl) or vehicle (artificial cerebrospinal fluid [aCSF] in DMSO) were infused into the striatum by way of a calibrated micropipette positioned adjacent to the recording electrode. AMPH (400 μM/125 nl) was infused 0, 45 and 90 min later. The inhibition of PI3K led to a significant reduction in the ability of AMPH infusions to induce DA efflux 45 and 90 min after treatment (Figure 3).
The precise concentration of LY294002 or AMPH that reaches the recording site is unknown, because it depends on diffusion through the extracellular matrix [41]. However, it has been estimated that there is at least a 10-fold dilution in drug concentration when ejected from a micropipette at a distance of 300 μm from the recording electrode [42], which is the separation distance that was used in the current studies. Additional studies from our laboratory suggest that a 10- to 200-fold dilution in drug concentration occurs by the time it diffuses to the recording electrode [43]. Thus, a barrel concentration of 400 μM AMPH or 1 mM LY294002 when pressure-ejected into brain would yield concentrations at the recording electrode of approximately 2–40 μM or 5–100 μM, respectively. Previous studies have shown that these concentrations of AMPH are consistent with those reported in brain after systemic administration of a behaviorally effective dose of AMPH and its derivatives [44,45]. Furthermore, the concentrations of LY294002 are similar to those that are able to regulate cocaine sensitization [26].
DAT is dynamically regulated at the plasma membrane by a number of intracellular signals [9,10,46,47], and recent data have also shown that transporter levels can be regulated by DA [48], pseudosubstrate stimulants such as AMPH [48,49], and inhibitors of DAT function such as cocaine [50]. To evaluate whether the reduction in AMPH-induced DA efflux caused by hypoinsulinemic conditions is promoted by a decrease in DAT cell surface expression, we evaluated DAT levels at the plasma membrane in striatal synaptosomes from rats made hypoinsulinemic with STZ [13]. As shown in Figure 4, chronic depletion of insulin results in a significant (>40%) decrease in the level of biotinylated, membrane-associated DAT within synaptosomes, indicating that DAT cell surface expression was significantly reduced in hypoinsulinemic rats. These findings, together with our electrochemical data (Figures 2 and 3), support the hypothesis that the reduction in AMPH-induced DA efflux caused by STZ treatment is a consequence of a reduction in DAT levels on the plasma membrane and are consistent with the previously reported blunted behavioral properties of AMPH under hypoinsulinemic conditions [12,33,34,51].
To further explore noninvasively the effect of STZ, hypoinsulinemia and downregulation of the PI3K signaling on AMPH-induced DA efflux, we used blood oxygenation level–dependent (BOLD) fMRI, which is sensitive to fluctuations in blood/hemoglobin oxygenation that closely reflects changes in neuronal activity [52,53]. In recent years, fMRI has proven useful in the study of the neural and pharmacological properties of psychostimulants within small laboratory animals [54–60]. Notably, when examined in rodents, BOLD responses to AMPH are linearly correlated with AMPH-induced changes in extracellular DA levels within the striatum [54,55]. In the present study, the BOLD responsiveness of the DAT- and IR-rich striatum to AMPH stimulation in normal and hypoinsulinemic rats was measured at 9.4 T using T2*-weighted multi-slice gradient echo imaging. Figure 5 shows that STZ-pretreated rats displayed a marked reduction in striatal activation in response to an acute exposure to AMPH (3 mg/kg, intraperitoneal [i.p.]). Figure 5A depicts representative BOLD activation maps from untreated control versus STZ-treated animals, each co-registered to high-resolution anatomic templates acquired in the same animals. Compared to untreated control rats given acute saline, those receiving AMPH exhibited significant BOLD activation in the dorsolateral striatum. However, this response was absent in rats rendered hypoinsulinemic by STZ treatment.
To quantify the effects of hypoinsulinemia on the striatal BOLD signal we performed region-of-interest (ROI) analysis of dorsolateral portions of this structure, which is predominantly innervated by the substantia nigra compacta and where t-maps indicated strong AMPH-evoked BOLD activation that was sensitive to insulin depletion. Figure 5B–5D summarizes the results of this analysis across groups of subjects (n = 5–6 per treatment group). When compared to treatment with saline, AMPH-treated animals exhibited a strong BOLD signal increase above baseline in the dorsolateral striatum. In contrast, there was no significant AMPH-induced BOLD signal change from baseline in STZ-pretreated, insulin-depleted rats (Figure 5B). In Figure 5C, post-injection traces from animals within each of the four treatment conditions described in Figure 5A and 5B were integrated and compared using one-way analysis of variance (ANOVA): F3,31 = 3.30, p < 0.05. Multiple comparisons between group pairs were conducted post hoc using the Newman-Keuls test: p < 0.05 compared to *Baseline, +Saline and #Untreated Control. ROI analysis of the ventral striatum (nucleus accumbens, NAc), which is innervated by the ventral tegmental area (VTA), revealed that the BOLD response to AMPH challenge did not significantly differ between STZ-treated and control animals (unpublished data). Likewise, prelimbic and cingulate cortices, also innervated by the VTA and where the norepinephrine transporter is the predominant carrier supporting DA inactivation [61], failed to show significant differential AMPH-induced BOLD responses after STZ (unpublished data).
To further elucidate the links between the PI3K signaling pathway, DAT function and AMPH action, we activated the PI3K pathway pharmacologically within the striatum of STZ-treated, hypoinsulinemic animals by locally infusing insulin just before the delivery of a brief AMPH pulse in this region. One week after depleting insulin with STZ treatment, local (striatal) application of exogenous insulin (10μM/100 nl) 2 min before AMPH infusion (400 μM/125 nl) almost fully restored to control levels the rate and the amount of AMPH-evoked DA release (Figure 6A and 6B), as well as the rate of DAT-mediated clearance (Figure 6C). These data further support our hypothesis that PI3K signaling is crucial for AMPH to stimulate DA efflux.
In recent years, the PI3K/Akt signaling pathway has been heavily implicated in the development, progression and maintenance of drug dependence [26–28]. Regulation of DAT plasma membrane expression (and subsequently of extracellular DA) by PI3K signaling is emerging as an important mechanism linking neurotransmitter transporter function to psychomotor stimulant abuse [11,13]. Profound adaptations within the neuronal dopaminergic system occur in experimentally induced diabetic mice [51]. Compared to controls, STZ-treated hypoinsulinemic rats display a marked reduction in striatal DA clearance [14,30] and are resistant to the behavioral effects of AMPH [12,33,34,51]. In experimentally induced diabetic rats (i.e., alloxan-treated), AMPH administered acutely is less effective at producing anorexia and stereotyped behavior and at increasing locomotor activity; subsequent administration of insulin reverses this attenuated sensitivity to AMPH [33]. Importantly, Galici et al. [12] showed that there is a selective decrease in AMPH self-administration in diabetic rats, consistent with data showing that dopamine uptake is decreased in hypoinsulinemic rats [14,30]. Considering that the striatum is highly enriched in insulin [17,62] and IRs [15,17,18], as well as in DAT [37–39], these studies strongly support a role for the neuronal PI3K pathway in regulating DAT activity and extracellular DA levels, as well as in the actions of AMPH.
The link between the PI3K pathway and the actions of AMPH is further fortified by recent studies from our laboratory as well as others, showing that prolonged exposure to AMPH ex vivo and in vivo inhibits PI3K signaling, as measured by Akt activity in striatum [29,63]. Akt is a protein kinase that is immediately downstream of PI3K, and Akt activity has been shown to be essential for insulin modulation of transporter function in striatal synaptosomes and human DAT-expressing cells [11,13]. Indeed, insulin signaling increases DA uptake capacity and cell surface expression [11,13]. In contrast, in vitro inhibition of either PI3K or Akt causes a decrease in DA uptake capacity and a redistribution of DAT away from the plasma membrane [11,13].
Here we demonstrate in vivo that hypoinsulinemia and pharmacological inhibition of PI3K signaling reduces the ability of AMPH to evoke DA efflux in the striatum. The reduction in DA efflux determined by HSCA in the current studies may result either from a decreased DAT plasma membrane expression, as suggested by in vitro [11,13] and ex vivo [14,30] studies, from a diminished DA content [64], or from both. Our data suggest that it is unlikely that the reduced DA efflux is a consequence of changes in tissue DA content. This is because the analysis of the rising phase of the HSCA traces revealed that the rate of AMPH-induced DA efflux in STZ-treated rats was severely attenuated compared with that of saline-treated control rats (Figure 2B). In fact, the slope of the rising portion of the DA signal represents the rate of DAT-mediated DA efflux, which is primarily dependent on the affinity and turnover rate of DA and is independent of DA content [14,40]. In addition, DA clearance as measured by HSCA that is independent of DA content but dependent on DAT number at the plasma membrane is reduced in STZ-treated animals [14]. STZ treatment does not significantly influence total DAT number or DA affinity [12,14]. Thus, the current findings suggest that a reduction in insulin signaling leads to a decrease in DAT function, a notion supported by the previous study from Owens et al. [14] showing that AMPH-naive, STZ-treated rats exhibit significantly less DA uptake in striatum as determined in vivo with HSCA and ex vivo in synaptosomes. Collectively these data support the hypothesis that hypoinsulinemia, by downregulation of PI3K signaling (see Figure 3), significantly reduces AMPH-induced DA efflux because of reduced DAT plasma membrane expression.
In support of the current in vivo electrochemical and ex vivo biochemical findings, STZ treatment was also found to inhibit the ability of AMPH to induce a BOLD response in the dorsal striatum. The current study did not reveal significant differences in insulin-dependent, AMPH-induced BOLD signal fluctuations in the NAc (unpublished data). Possibly, this was due to the limited radio frequency penetration of the surface coil used in this study into more ventral brain areas such as the NAc. Importantly, others have reported decreases in AMPH-induced DA release in NAc dialysates collected in freely moving rats [65]. Because hyperglycemia has been shown to not significantly influence BOLD signals [66], our data suggest that blunting of the AMPH-induced BOLD response in the DAT-dense striatum of insulin-depleted rats (Figure 5) is not due to STZ-mediated metabolic abnormalities. Importantly, in the striatum, the AMPH-induced BOLD response has been shown to correlate with striatal extracellular DA levels [54,55] and, consequently, with DAT-mediated reverse transport of DA. These data further support our hypothesis that STZ treatment, by decreasing PI3K signaling in striatum, downregulates AMPH-induced DA efflux measured by HSCA (Figures 2, 3 and 6) and fMRI (Figure 5). Our results are consistent with in vitro studies demonstrating that the blockade of insulin signaling decreases the number of active DATs on the plasma membrane [13] as we currently demonstrate in DAT cell surface biotinylation studies from striatal preparations (Figure 4). These data support the hypothesis that the attenuated rate of AMPH-induced striatal DA efflux in hypoinsulinemic rats results from a DAT trafficking phenomenon [14]. Conceivably, in STZ-treated animals, insulin stimulation of PI3K signaling should restore DA clearance and AMPH-induced DA efflux. Figure 6 shows that local application of insulin in STZ-treated rats almost completely restores AMPH-stimulated DA efflux.
We demonstrate here that PI3K signaling regulates the pharmacological actions of drugs (e.g., AMPH) that act on dopaminergic systems. Importantly, our data show that hypoinsulinemia reduces basal PI3K signaling and impairs the ability of AMPH to increase extracellular DA levels. Therefore, PI3K signaling may provide a new cellular target for the development of novel treatments of AMPH abuse and regulation of dopaminergic tone.
All procedures were approved by the Vanderbilt University Medical Center and the University of Texas Health Science Center at San Antonio Institutional Animal Care and Use Committees and were conducted according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals. For all experiments, male Sprague–Dawley (Harlan, Indianapolis, Indiana, United States) rats (275–350 g) served as experimental subjects. STZ is an antibiotic that destroys the insulin-secreting β cells of the pancreas [36] and has previously been used to induce chronic hypoinsulinemia in rats by our laboratories [12,14]. STZ (Sigma-Aldrich; http://www.sigmaaldrich.com) was freshly dissolved in ice-cold 100 mM citrate saline (pH 4.5) for all studies. Rats received STZ (50 mg/kg, i.p. for HSCA studies; 65 mg/kg into the tail vein for fMRI studies) and were returned to their home cages for at least 7 d. Blood glucose was measured with a glucometer (Advantage Accu-Chek, Roche Diagnostics; http://www.roche.com) before STZ and just before an experiment. Animals were considered hypoinsulinemic when their glucose levels exceeded 300 mg/dl.
Preparation of synaptosomes was performed as described previously [11,12,14]. Rats were killed by decapitation, their brains were removed and their striata were rapidly dissected on a plastic dish placed on ice. Tissue was homogenized in ice-cold Krebs-Ringer buffer (125 mM NaCl, 1.2 mM KCl, 1.2 mM MgSO4, 1.2 mM CaCl2, 22 mM NaHCO3, 1 mM NaH2PO4, 10 mM glucose, pH 7.4) containing 0.32 M sucrose using a glass-Teflon homogenizer. Homogenates were centrifuged at 1,000g for 10 min at 4 °C, and the resulting supernatants were centrifuged at 16,000g for 25 min at 4 °C. P2 pellets were then placed on ice and resuspended immediately prior to experiments.
Akt activity assays were performed as described previously [29]. Striatal synaptosomes were lysed for 45 min at 4°C in a buffer containing 20mM Tris (pH 7.5), 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM β-glycerolphosphate, 1 mM Na3VO4, 1 μg/ml leupeptin and 1 mM PMSF. Lysed proteins (∼400 μg; BioRad DC Protein Assay Kit; http://www.biorad.com) underwent immunoprecipitation with an Akt-specific monoclonal antibody as part of a commercially available Akt activity assay kit (BioVision; http://www.biovision.com). Activity of the immunoprecipitated Akt was determined in vitro with the addition of recombinant GSK3α as the kinase substrate; the resulting phosphorylated GSK3α (pGSK3α) was determined by immunoblotting (see below) using phosphospecific antibodies to GSK3α (Ser 21, diluted 1:1000), provided in the Akt activity assay kit.
Biotinylation studies were performed as described previously [13,48,49] with modification. Striatal synaptosomes were washed twice with warm Krebs–Ringer bicarbonate (KRB) buffer (containing 145 mM NaCl, 2.7 mM KCl, 1.2 mM KH2PO4, 1 .2 mM CaCl2, 1.0 mM MgCl2, 10 mM glucose, 0.255 mM ascorbic acid, and 24.9 mM NaHCO3) and then incubated in the same buffer for 1 h at 37 °C. The reaction was stopped in ice and the samples were washed with phosphate-buffered saline ( PBS) containing 0.1 mM CaCl2 and 1 mM MgCl2 (PBS-Ca-Mg) and incubated with EzLink Sulfo-NHS-SS-Biotin (2.0 mg/ml in PBS-Ca-Mg; Pierce Chemical; http://www.piercenet.com) on ice for 30 min. The reaction was quenched by washing twice with 4 °C PBS-Ca-Mg containing 100 mM glycine (PBS-Ca-Mg-glycine) followed by an incubation with PBS-Ca-Mg-glycine for 15 min on ice. Synaptosomes were then washed twice with cold PBS-Ca-Mg before lysis with 1 ml of radioimmunoprecipitation assay (RIPAE) buffer (10 mM Tris pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.1% SDS, 1% sodium deoxycholate and 1% Triton X-100) containing protease inhibitors (0.5 mM phenylmethylsulfonyl fluoride, 5 μg/ml leupeptin and 5 μg/ml pepstatin) for 1 h 30 min on ice with agitation. Lysates were then centrifuged at 14,000g for 30 min at 4 °C. After isolation of supernatants, biotinylated proteins were separated by incubation with 90 μl ImmunoPure immobilized streptavidin beads (Pierce) for 1 h at room temperature with agitation. Beads were washed three times with RIPAE buffer; biotinylated proteins were then eluted in 50 μl of 2X SDS-PAGE sample loading buffer at room temperature. Total cell lysates (∼300 μg protein) and the biotinylated (cell surface) fraction (∼10% of total; 30 μg protein) underwent immunodetection for DAT as described below.
Determination of biotinylated DAT immunoreactivity was conducted with some modification according to previously described methods [13,29]. Briefly, synaptosomal lysates were separated by SDS-PAGE, and resolved proteins were transferred to polyvinylidene difluoride (PVDF) membranes (BioRad), which were incubated for 1–2 h in blocking buffer (5% dry milk and 0.1% Tween 20 in Tris-buffered saline). To quantify biotinylated (surface) DAT, immunoblots were incubated with mouse monoclonal primary antibodies to the N terminus of the rat DAT (antibody 16, 1:1000, [67]), generously provided by Roxanne Vaughan (University of North Dakota School of Medicine and Health Sciences, Grand Forks, North Dakota, United States). All proteins were detected using HRP-conjugated goat anti-mouse secondary antibodies (1:5000; Santa Cruz Biotechnology; http://www.scbt.com). After chemiluminescent visualization (ECL-Plus; Amersham; http://www.amersham.com) on Hyperfilm ECL film (Amersham), protein band densities were quantified (Scion Image; http://www.scioncorp.com) and normalized to the appropriate total protein amount. Immunoblotting experiments were performed in triplicate, analyzed (GraphPad v4.0; http://www.graphpad.com) and reported as mean ± standard error of the mean.
HSCA was conducted using the FAST-12 system (Quanteon; http://www.quanteon.cc) as previously described with some modification [14]. Recording electrode/micropipette assemblies were constructed using a single carbon-fiber (30 μm diameter; Specialty Materials; http://www.specmaterials.com), which was sealed inside fused silica tubing (Schott, North America; http://www.schott.com). The exposed tip of the carbon fiber (150 μm in length) was coated with 5% Nafion (Aldrich Chemical Co.; http://www.sigmaaldrich.com; 3–4 coats baked at 200 °C for 5 min per coat) to provide a 1000-fold selectivity of DA over its metabolite dihydroxyphenylacetic acid (DOPAC). Under these conditions, microelectrodes displayed linear amperometric responses to 0.5–10 μM DA during in vitro calibration in 100 mM phosphate-buffered saline (pH 7.4).
Animals were anesthetized with injections of urethane (850 mg/kg, i.p.) and α-chloralose (85 mg/kg, i.p.), fitted with an endotracheal tube to facilitate breathing, and placed into a stereotaxic frame (David Kopf Instruments; http://www.kopfinstruments.com). To locally deliver test compounds (see below) close to the recording site, a glass single or multi-barrel micropipette (FHC; http://www.fh-co.com) was positioned adjacent to the microelectrode using sticky wax (Moyco; http://www.moycotech.com). The center-to-center distance between the microelectrode and the micropipette ejector was 300 μm. For experiments in Figure 2, the micropipette was filled with AMPH (400 μM; Sigma) or its vehicle (PBS). The study in Figure 3 used a multibarrel configuration in which barrels contained AMPH (400 μM) or vehicle (aCSF) and additional barrels contained LY294002 (1 mM; Sigma) or its vehicle (aCSF in DMSO). For experiments in Figure 6, one barrel contained AMPH (400 μM) and an adjacent barrel contained insulin (10 μM; Sigma); a third barrel contained aCSF, the vehicle for both AMPH and insulin. The electrode/micropipette assembly was lowered into the striatum at the following coordinates (in mm from bregma [68]): A/P, +1.5; M/L, ±2.2; D/V, −3.5 to −5.5. The application of drug solutions was accomplished using a Picospritzer II (General Valve Corporation; http://www.parker.com) in an ejection volume of 100–150 nl (5–25 psi for 0.25–3 s). After ejection of test agents, there is an estimated 10–200-fold dilution caused by diffusion through the extracellular matrix to reach a concentration of 2–40 μM (AMPH), 5–100 μM (LY294002) or 0.05–1 μM (insulin) at the recording electrode [43]. To record the efflux and clearance of DA at the active electrode, oxidation potentials—consisting of 100-ms pulses of 550 mV, each separated by a 1-s interval during which the resting potential was maintained at 0 mV—were applied with respect to an Ag/AgCl reference electrode implanted into the contralateral superficial cortex. Oxidation and reduction currents were digitally integrated during the last 80 ms of each 100-ms voltage pulse. For each recording session, DA was identified by its reduction/oxidation current ratio: 0.55–0.80.
At the conclusion of each experiment, an electrolytic lesion was made to mark the placement of the recording electrode tip. Rats were then decapitated while still anesthetized, and their brains were removed, frozen on dry ice, and stored at −80°C until sectioned (20 μm) for histological verification of electrode location within the striatum. HSCA data were analyzed with GraphPad Prism using three signal parameters (see Figure 2A for exemplary trace): (i) the DA efflux rate (in nM/s), which is the change in DA oxidation current evoked by AMPH application as a function of time; (ii) the maximal signal amplitude of the released DA (in μM); and (iii) the DA clearance rate (in nM/s), defined as the slope of the linear portion of the current decay curve, i.e., from 20−60% of maximal signal amplitude.
Under isoflurane anesthesia, rats were implanted with femoral artery and i.p. catheters, tracheotomized and artificially ventilated with a 30:70% O2:N2O mixture. Rats were paralyzed with a bolus infusion of pancuronium bromide (2 mg/kg; Sigma) dissolved in isotonic saline (1 ml/kg, i.p.), and the concentration of isoflurane was reduced to 0.88%. Ventilation parameters were adjusted (respiratory rate = 48–52 breaths/min; inspiration volume 14–18 cm H2O) to maintain stable blood gases, which were sampled from the arterial catheter immediately before and after the completion of MRI scans. Mean arterial gas values obtained from all 23 rats used in fMRI studies were: pH = 7.36 ± 0.06, pCO2 = 37.7 ± 6.8 mm Hg, pO2 = 140.5 ± 20.6 mm Hg. A respiration pillow sensor (SA Instruments; http://www.i4sa.com) was positioned underneath the animal's abdomen. Core body temperature and heart rate were monitored during imaging studies using a rectal probe and subdermal electrocardiograph (ECG) electrodes implanted into the forepaws. Temperature, ECG and respiratory data were collected and analyzed using an MR-compatible monitoring system (SAM-PC, SA Instruments).
To minimize motion artifacts, rats were positioned within a custom-built plexiglass stereotaxic platform and fixed in place with Teflon ear bars and an adjustable incisor bar. Attached to the platform was a socket holding a 20-mm dual transmit-receive radio frequency surface coil (Varian Instruments; http://www.varianinc.com) lowered to 1 mm above the scalp. The platform was then placed inside a 9.4 T, 21-cm bore Varian Inova superconducting magnet equipped with actively shielded gradients of 40 G/cm and peak rise times of 135 μs. The MRI system was controlled by a Varian console interfaced with a Sun Microsystems computer running VnmrJ 6.1D software (Varian). Nine contiguous coronal slices, serving as within-subject high-resolution anatomic templates, were acquired using conventional gradient echo multi-slice (GEMS) imaging. Seventy-two functional image volumes, spatially aligned with the anatomic templates, were then continuously acquired for 30 min using the following GEMS parameters: TR/TE = 220/12 ms; flip angle = 20°; NEX = 2; slice thickness = 1 mm; in-plane voxel resolution = 0.47 × 0.47 mm; matrix = 64 × 64; FOV = 30 × 30 mm; acquisition time = 25.6 s per excitation. After a 15-min baseline period, AMPH dissolved in isotonic saline was administered as a bolus i.p. infusion (3 mg/ml/kg; 20–30 s); image acquisition continued uninterrupted for 15 min after the infusion.
Analysis of fMRI data was conducted in MATLAB (v7.0.4; The MathWorks; http://www.mathworks.com) as described previously, with some modification [69]. Data were first analyzed to generate statistical parametric activation maps based on the Student's t-test. t values were computed for each image voxel by comparing the baseline signal to the post-injection signal. For each subject, colorized t values from each image voxel were registered to a high-resolution anatomic template obtained in the same subject. ROI analyses were conducted over the dorsolateral striatum, the ventral striatum (NAc) and prelimbic/cingulate cortices based on 1-mm-thick coronal slices spanning 1–2 mm anterior to bregma [68]. For each ROI, fMRI time series data underwent baseline drift correction with a linear detrending function; high-frequency noise was suppressed with a low-pass Butterworth filter. Pixel intensities from each image voxel in the ROIs were converted to percent signal changes from baseline (% ΔS/So), averaged across left and right hemispheres, integrated and analyzed by one-way ANOVA followed by a Newman-Keuls post test.
|
10.1371/journal.pntd.0002807 | Diagnostic Performance of Schistosoma Real-Time PCR in Urine Samples from Kenyan Children Infected with Schistosoma haematobium: Day-to-day Variation and Follow-up after Praziquantel Treatment | In an effort to enhance accuracy of diagnosis of Schistosoma haematobium, this study explores day-to-day variability and diagnostic performance of real-time PCR for detection and quantification of Schistosoma DNA compared to other diagnostic tools in an endemic area before and after treatment.
Previously collected urine samples (N = 390) from 114 preselected proven parasitological and/or clinical S. haematobium positive Kenyan schoolchildren were analyzed by a Schistosoma internal transcribed spacer-based real-time PCR after 14 years of storage. Pre-treatment day-to-day fluctuations of PCR and microscopy over three consecutive days were measured for 24 children using intra-class correlation coefficient. A combined ‘gold standard’ (PCR and/or microscopy positive) was used to measure sensitivity and negative predictive value (NPV) of several diagnostic tools at baseline, two and 18 months post-treatment with praziquantel.
All 24 repeatedly tested children were PCR-positive over three days with little daily variation in median Ct-values, while 83.3% were found to be egg-positive for S. haematobium at day 1 and 75.0% at day 2 and 3 pre-treatment, signifying daily fluctuations in microscopy diagnosis. Of all 114 preselected schoolchildren, repeated microscopic measurements were required to detect 96.5% versus 100% of positive pre-treatment cases by single PCR. At two months post-treatment, microscopy and PCR detected 22.8% versus 69.3% positive children, respectively. Based on the ‘gold standard’, PCR showed high sensitivity (>92%) as compared to >31% sensitivity for microscopy, both pre- and post-treatment.
Detection and quantification of Schistosoma DNA in urine by real-time PCR was shown to be a powerful and specific diagnostic tool for detection of S. haematobium infections, with less day-to-day variation and higher sensitivity compared to microscopy. The superior performance of PCR before, and two and 18 months post-treatment provides a compelling argument for PCR as an accurate and reproducible tool for monitoring treatment efficacy.
| Schistosoma haematobium is a blood fluke that causes severe urogenital pathology and affects millions of people, mainly in sub-Sahara Africa. Current diagnosis is based on microscopic examination of urine samples, but this method is not only observer dependent, but also known for its low sensitivity and high day-to-day variability. Accurate diagnosis is important to assess community levels of infections for consideration of deworming campaigns, and to monitor treatment efficacy. We evaluated a real-time polymerase chain reaction (PCR) assay for specific detection and quantification of Schistosoma DNA in urine samples from 114 preselected S. haematobium infected schoolchildren of endemic coastal Kenya and compared the outcome to several other diagnostic methods. Three urine samples collected over three subsequent days from 24 participants were used for Analyzing day-to-day fluctuations in egg counts and Schistosoma DNA levels. Urine was also tested two and 18 months after praziquantel treatment. Compared to microscopy, we observed less day-to-day fluctuations and higher sensitivity with real-time PCR, in particular when tested two months after therapy. Real-time PCR is therefore useful for more accurate identification of S. haematobium, especially in monitoring control interventions.
| An estimated 119 million people are infected with Schistosoma haematobium worldwide [1] with sub-Saharan Africa as the major geographical area at risk [2]. In up to 75% of infected individuals both urinary and genital tracts are affected, hence the recent renaming of “urinary” to “urogenital” schistosomiasis. Following effective mass drug administration campaigns, the 65th World Health Assembly held in 2012 realized that intensified control measures were needed to further reduce transmission of schistosomiasis in Africa. The implementation of specific elimination programs were even encouraged [3]. The availability of highly accurate diagnostics is vital for these control programs, particularly in post-treatment situations [4]. In areas with high transmission and subsequently high prevalence and infection intensities, where mass treatment with praziquantel (PZQ) is provided, limited diagnostic sensitivity may be justified as PZQ causes no serious side-effects and generally, the intervention will be cost-effective. However, in moderate and low transmission areas, or in areas with waning post-treatment prevalence, detecting eggs by microscopy becomes more questionable while at the same time cost-effective targeting of mass chemotherapy becomes more essential to avoid the potential emergence of drug resistance [5], [6].
For the diagnosis of urinary tract infections, the ‘gold standard’ remains microscopic detection of ova in urine samples [6]. Nonetheless, parasitological diagnosis has its limitations as the sensitivity is low and may be affected by day-to-day variability in egg excretion [7]. Furthermore, when subjects harbor low worm loads, light infections are often missed by microscopy [8]. Collection of multiple urine samples per individual on consecutive days may increase parasitological sensitivity of microscopy, but is more expensive and also time-consuming [8]. Haematuria represents a good predictor of S. haematobium as it correlates well with heavy infection levels [9]. However, blood in urine is a nonspecific symptom of schistosomiasis in areas of low endemicity and can be over- or underestimated depending on the infection prevalence in an area. Haematuria can therefore not be used as direct indicator of infection, but only as a screening method for S. haematobium morbidity at the individual level [9].
Alternatively, schistosome parasites release antigenic material into the circulation of their hosts during different stages of their life cycle, thus providing an opportunity to identify these antigens for diagnostic purposes. For example, a circulating soluble egg antigen enzyme-linked immunosorbent assays (cSEA-ELISA) was developed separately for S. mansoni [10] and S. haematobium [11] for detection of egg antigens. While the S. mansoni-specific ELISA could be used as an indicator of tissue egg load [12], urine cSEA levels were found to be primarily a marker of S. haematobium-related bladder pathology [11], [13], [14]. In addition, antigens released by the adult worm stage, i.e. circulating cathodic antigens (CCA) and circulating anodic antigens (CAA), can be used to differentiate between active and past infection as they are quickly cleared from the circulation [15], [16]. While a CCA-urine point-of-care (POC) is now commercially available and widely evaluated [17], the CAA assays have recently been much improved [18] with a view towards future POC applications.
In search of diagnostic tools with high overall accuracy when applied to a single sample, the detection and amplification of Schistosoma DNA has been evaluated in recent years [19], [20]. Real-time PCR may become an alternative to microscopy and antigen detection for diagnosing S. haematobium due to several advantages [19], [20], [21], yet its efficacy is rarely evaluated through population-based surveys in communities originating from schistosomiasis endemic regions. Recent studies based on epidemiological and diagnostic evaluations of real-time PCR using urine samples from communities in Ghana have showed promising results [19], [21]. Nonetheless, there is need to further evaluate the accuracy of real-time PCR for detection of Schistosoma DNA in urine samples pre- and post-treatment as limited data are currently available on the performance of real-time PCR as a monitoring tool after chemotherapy [22]. The current study explores pre-treatment day-to-day variability, and diagnostic performance of real-time PCR for detection and quantification of Schistosoma DNA compared to microscopy, microhaematuria and cSEA-ELISA in an endemic area before and after treatment.
This study was performed retrospectively on urine samples collected by one of the authors (AK) and his team at a larger field survey on human S. haematobium in Kenya that was initiated in 1996. All samples were anonymized before analysis. The study was approved by the Ministry of Health's ethical committee (Ministry of Health Kenya) and the Danish Central Medical Ethics Committee. Informed consent was given by the pupils' parents, the education office and the local administration.
Geographical, demographical and epidemiological details of the study area have been published elsewhere [23]. At collection of data and samples the area was known for high S. haematobium prevalence and infection intensity, while S. mansoni infections are hardly seen on coastal Kenya [24]. Urines from a total of 114 schoolchildren aged between six and 15 years from two schools (Kibaokiche and Tsunguni) in coastal Kenya were analyzed. The 114 children were part of a larger group selected for post-treatment follow-up studies based on positive S. haematobium egg excretion and/or presence of microhaematuria (grade 1) at baseline examination before treatment (attachment S1). The cohort examined for the current study consisted of 114 schoolchildren, who had been present at all follow-up time points and where an urine sample was available for real-time PCR analysis. All infected children in the two schools received supervised treatment with a single regimen of PZQ (40 mg/kg) after baseline investigations and at completion of the study [23]. Aliquots of urine (5–10 ml) were taken before filtration. All samples were stored at −20°C and sent at frozen condition to the Department of Parasitology, Leiden University Medical Centre (LUMC), The Netherlands for quantification of cSEA by ELISA. In 2010, DNA was extracted of 390 urine samples and tested with real-time PCR at the LUMC: i) day-to-day variation was assessed in urine samples collected randomly among 24 schoolchildren prior to treatment on three consecutive days (day = 1, 2, 3); ii) urine samples from 114 participants were analyzed with the aim of determining the correlation with and performance of PCR as compared to other diagnostic tests at baseline/pre-treatment, two, and 18 months after treatment.
Parasitological diagnosis was performed in 1996 using standard microscopy on mid-morning urine specimens for day-to-day consistency [11], [14], [23]. Eggs were quantified using a slightly modified Nucleopore syringe urine filtration method, filtering a 10 ml duplicate aliquot from each urine sample [23]. Results were expressed as the number of eggs per 10 ml of urine (eggs/10 ml). Each slide was read by two trained microscopists and discrepancies resolved by consensus before data recording. Eggs exceeding 1000 per 10 ml urine were not counted due to clogging of filters, but recorded as >1000 eggs/10 ml. Egg counts were categorized into no infection (no eggs/10 ml), low (1-<50 eggs/10 ml), and high (≥50 eggs/10 ml) infection intensities. Microhaematuria was assessed using urine-reagent strips Hemastix® (Bayer, UK), according to the manufacturer's instructions. Microhaematuria was categorized into no infection (score 0), “trace” (score 1), low (score 2), moderate (score 3), and high (score 4) infection intensities.
The urinary levels of cSEA data were assessed by Kahama et al. in 1998, and published previously [11], [13], [14]. In short, a monoclonal antibody-based sandwich ELISA was used for quantification of cSEA in pre-treated urine specimens and concentrations were read against a standard curve. The cut-off for positivity of cSEA samples was determined at 32 ng/ml.
Urine DNA isolation, amplification and detection were performed as described elsewhere [19], [21]. In short, Schistosoma genus-specific primers (Ssp48F and Ssp124R) and the double-labeled probe Ssp78T were used to amplify a 77-bp fragment of the internal transcribed spacer-2 (ITS2) sub-unit. This PCR has been extensively validated on its specificity and appropriate positive and negative controls were included at each PCR run [19], [21]. In addition an internal control (phocin herpes virus 1 (PhHV-1)) was added to each sample reaction for detection of potential inhibition of amplification. A CFX96 real-time PCR detection system (Bio-Rad) was used for amplification and amplicon detection, and CFX Manager version 1.6.514 (Bio-Rad) for related data analysis. The PCR output is expressed as Cyclic-threshold (Ct-) values. Ct-values indicate the number of amplification cycles at which the level of fluorescent signal exceeds the background fluorescence, hence demonstrating parasite-specific DNA loads (infection intensity) in urine samples. The PCR thermocycler was set at 50 cycles in which Schistosoma DNA could be amplified. Ct-values were categorized into no infection (Ct = 50), low (35≤Ct≤50), moderate (30≤Ct<35), and high (Ct<30) intensity infections [21].
Analysis were performed using SPSS version 20·0 (Statistical Package for the Social Sciences, Chicago, IL, IBM). Main outcome variables were Ct-values and egg counts (continuous). Analysis included: 1) day-to-day variability; 2) correlation between Ct-values and the value of each diagnostic assay; 3) diagnostic performance pre- and post-treatment using sensitivity and negative predictive value (NPV). In case of missing data, outcome measures were calculated based upon available data. Due to limited missing data, this scarcely influenced the results.
The schoolchildren who provided urine samples for the current study, consisted of 49.1% boys, and ages ranged from six to fifteen years (median: 9.6 years). Kibaokiche school (56.1%) counted 14 more children than Tsunguni school. Prior to treatment, the two schools had similar infection levels. Although schoolgirls had higher egg intensities compared to schoolboys with a median positive value of 128.5 eggs/10 ml (IQR: 21.3, 341.9) for girls and 65.0 (IQR: 15.0, 374.0) eggs/10 ml for boys, boys seemed to be more prone to reinfection (73.2%) and high intensity infection with S. haematobium (42.0 eggs/10 ml; IQR: 13.5, 140.0) at 18 months post-treatment than girls (53.4%; 17 eggs/10 ml; IQR: 2.0, 164.0) (χ2: 4.8; p = 0.029). For a more detailed epidemiological description of the results, we refer to the papers by Kahama et. al. [13], [14], [23].
The variation in the different diagnostic parameters over three subsequent days prior to treatment of 24 subjects is shown in table 1. Despite that egg and/or microhaematuria positivity was a criterion for study enrolment, four subjects (16.7%) were found egg-negative by microscopy at day 1, and six (25%) on each subsequent day. Median egg counts varied across days, with the lowest median egg count (73 eggs/10 ml) on day 3. Two subjects (8.3%) remained negative by microscopy in all three repeated pre-treatment samples. No new cases were diagnosed on the third day urine sample. Daily variability of microhaematuria was substantial regarding both the proportion of positive and proportion with heavy intensity infections (score 4), the latter ranging from 54.2% to 70.8%. Similarly, daily fluctuations of cSEA-ELISA were considerable between the three subsequent days. All PCR-tested subjects were positive in all three days (100%) while median values and IQRs were comparable.
All correlation coefficients of the various diagnostic tools using the median values of the three different sampling points pre-treatment (N = 24) were highly significant (p<−0.001). Highest correlation was found between microscopy (eggs/10 ml) and microhaematuria (ρ: 0.81). Correlation with real-time PCR (Ct-values) was similar and best with both cSEA-ELISA (ng/ml) and microhaematuria (ρ: −0.71) as compared to microscopy (ρ: −0.62). Values of real-time PCR are negative as low PCR Ct-values reflect high parasite-specific DNA loads and vice versa. Day-to-day fluctuation of the parameters as determined by ICC was much less for DNA concentrations (R = 0.67; 95%CI: 0.35, 0.85; p = 0.001) than for egg counts (R = 0.29; 95%CI: −0.44, 0.67; p = 0.17). Major day-to-day fluctuations were also found for microhaematuria (R = 0.27; p = 0.19) and cSEA-ELISA (ng/ml) (R = 0.23; p = 0.23).
The proportions of S. haematobium-positives based on a single sample by the various diagnostic assays are shown in table 2. Only 83.3% of the 114 children were microscopy positive at the first urine samples at baseline. This percentage increased to 96.5% after microscopy of minimum three (up to five) consecutive urine samples, while 100% of the children were PCR-positive based on testing a single urine sample. Treatment with PZQ resulted in significantly reduced intensity of infection after two months, with almost one-fourth (22.8%) of the schoolchildren remaining egg-positive in a single urine specimen and 34.2% in multiple samples. After 18 months, infection levels increased to 63.2% by microscopy of a single urine sample and examination of minimum three (up to five) repeated samples increased positivity to 68.1%. Both microhaematuria and cSEA showed higher proportions of positives at two months post-treatment as compared to microscopy, but both tests missed a substantial number of cases at 18 months post-treatment. Real-time PCR had the highest proportion of S. haematobium-positives throughout all time points based on a single sample. A high percentage of positives were ascertained by PCR two months after treatment (69.3%), although DNA levels were relatively low compared to pre-treatment levels (median Ct-value: 35.9). At 18 months post-treatment, positivity increased to 78.9% with a median intensity of 29.9 Ct-values. The course of positive infection and infection intensity at different examination time points is shown graphically in figure 1.
Significant correlations were found between urine DNA levels and egg counts, microhaematuria and cSEA levels, both before treatment as well as 18 months post-treatment (table 3). There was no significant correlation between Ct-values and the other parameters at two months post-treatment.
The performance of each diagnostic tool using a combined ‘gold standard’ (infection positivity by either microscopy or PCR) is shown in table 4. According to the defined ‘gold standard’, 83 samples were positive for S. haematobium at two months post-treatment. Microscopy had a sensitivity of 31.3% with a NPV of 35.2%. Kappa agreement between microscopy and the ‘gold standard’ was present (κ: 0.20; p<0.0001). Both microhaematuria and cSEA had similar sensitivities and NPV, and poor Kappa agreements with the ‘gold standard’ (microhaematuria: κ: 0.13; p = 0.036; and, cSEA: κ: 0.03; p = 0.69). Real-time PCR displayed the highest sensitivity and NPV of 95.2% and 88.6%, respectively, at two months post-treatment, as well as high Kappa agreement (κ: 0.92; p<0.0001). PCR missed four out of 83 positive cases.
At 18 months post-treatment, the sensitivity of microscopy increased to 73.5% but the NPV remained below 40%, while the Kappa-statistic was 0.44 (p<0.0001). Sensitivities of microhaematuria and cSEA also increased at 18 months post-treatment, but to a minor extent compared to microscopy, and both NPVs declined. A total of 98 infections were found using the ‘gold standard’ at 18 months post-treatment, of which 90 cases were detected by real-time PCR thus showing a sensitivity of 91.8%. This was comparable to the sensitivity at two months post-treatment, although the NPV decreased to 67%. The Kappa agreement between PCR and the ‘gold standard’ was 0.76 (p<0.0001).
Analyzing multiple pre-treatment urine samples, we found the proportion of positive cases with eggs, microhaematuria, and/or cSEA to fluctuate significantly on a day-to-day basis, as determined by microscopy, urine-reagent strips, and cSEA-ELISA, respectively. Similarly, daily variations were demonstrated in intensity of infection as determined by microscopy and cSEA-ELISA, and proportion of microhaematuria with score 4. Multiple sample testing can improve the sensitivity of microscopy [27] but also increases workload and costs [8], which is cumbersome in endemic field settings. Previous studies found relatively stable microhaematuria levels over time compared to egg counts [28], [29], but the proportion of microhaematuria-positives and the proportion with score 4 varied considerable in this study. Nevertheless, blood in urine remains a useful marker for preliminary screening of communities to identify those at risk of morbidity [9], [30], [31], as the current study found a strong correlation of microhaematuria with infection as detected by all three diagnostic methods. Furthermore, microhaematuria mirrored egg counts following treatment. Both microhaematuria and cSEA may be useful as infection and pathology markers for S. haematobium, as confirmed by relatively good pre-treatment diagnostic sensitivities of both techniques. Alternatively, real-time PCR performed excellently with 100% sensitivity even when using a single urine specimen, with significant correlation and stability over three successive days. The results of the current study are, however, based on a very small sample size, which may have affected the results.
The current study showed that real-time PCR may be a good indicator of infection intensity, as measured by a strong correlation between median Ct-values and egg counts, both pre- and 18 months post-treatment. The correlation between Ct-values and egg counts might be even stronger if egg counts above 1000 eggs/10 ml urine had not been truncated. The correlation with PCR improved when microscopic assessment of infection was based on three to five samples, irrespective of the examination time point. This suggests that real-time PCR could be used as a diagnostic tool providing information about prevalence as well as intensity of infection. Similar significant correlation was found between Ct-values and egg counts by another study, using controls from a non-endemic area, and urine samples known to contain S. haematobium eggs as cases [19]. Obeng et al. found that the median Ct-values for cases with low-intensity S. haematobium infections (≤50 eggs/10 ml urine) was higher than that of cases with intense infections (>50 eggs/10 ml). Discrepancies between these values of microscopy and PCR were likely owing to the lightly infected cases who were excreting very low numbers of eggs, whereby the amount of schistosome genetic material may not reach levels detectable by PCR [19]. There was a poor correlation between egg counts and Ct-values at two months post-treatment. However, it is possible that the linear relation between egg counts and DNA loads may have been disturbed by the temporary effect of treatment [32].
In order to estimate the true-positives of a diagnostic assay in the absence of an accurate and reliable “gold standard”, latent class analysis (LCA) can be employed [33], [34]. However, LCA models are not error-free. The model maintains several assumptions (i.e. conditional independence) that are often violated in practice. Results from at least four different diagnostic assays should be used to account for conditional dependence (i.e. when two assays are based on a similar biological phenomenon), although sufficient results from so many diagnostic assays are not always readily available. In the case of inappropriate use of LCA models, estimations of sensitivity, specificity, and prevalence would only lead to bias [35]. Therefore, analogous to Midzi et al. [26] and others [25], [36], this study used the combined results of microscopy and PCR as reference for calculation of diagnostic accuracy based on both methods being 100% specific. Real-time PCR achieved 100% sensitivity after a single test pre-treatment, while microscopy could not detect all positive cases even if more than three independent urine samples were analyzed. Our study shows that the number of positives as determined by PCR at two months post-treatment is considerably higher than revealed by microscopy. Microscopy found a relatively low prevalence after treatment (22.8%) as opposed to 69.3% with PCR. This discrepancy may be explained by the low sensitivity of microscopy (31.3%) in comparison with the ‘gold standard’, which might indicate that PCR-positive, egg-negative cases were false-negatives. PZQ mainly kills mature worms, while immature schistosomes may endure during chemotherapy and redevelop into maturity several weeks after treatment [37]. This temporary discontinue in egg excretion may explain the higher numbers of microscopic false-negatives at two months post-treatment, whereas PCR can still pick-up Schistosoma-specific DNA at very low infection intensity levels [38]. Nonetheless, PZQ significantly reduced infection intensity at two months post-treatment as determined by both microscopy and PCR.
Studies have shown that, even with highly effective treatment, many people in endemic regions still harbor sufficient surviving adult schistosomes to account for light but detectable post-treatment egg excretion [5], [39]. In areas of intense transmission, people are likely to have high levels of both patent and pre-patent infections at time of treatment, with PZQ having little effect on immature schistosomes [37]. Considering the lower post-treatment sensitivity of microscopy in this study, the number and intensity of infections may have been underestimated at two months post-treatment. The high number of cases with detectable Schistosoma DNA in urine after chemotherapy shows in a more pronounced way that treatment may not be a 100% effective in removing infection [5], [39]. PCR assays detect mostly DNA that originates from schistosomes eggs in urine, stool or organ biopsy samples, and a positive PCR result is therefore dependent on whether the processed sample contains eggs [40]. In chronic, continuously exposed individuals in high intensity infection areas, retention of eggs in the bladder wall may occur [41] or stunted worms may still be present, even after chemotherapy. We therefore cannot rule out with certainty the presence of decaying worms or eggs still expressing parasite DNA [42], although Coulibaly et al. found up to 38% higher prevalences with CCA as compared to Kato-Katz in individuals infected with S. mansoni post-treatment [36]. As CCA is regurgitated into the bloodstream by actively feeding worms and successive cleared in the host's kidneys [15], [16], Coulibaly et al.'s results on S. mansoni also indicated the presence of active infection in the host three weeks post-treatment. This suggests that adult worm clearance after chemotherapy may not be as effective as previously thought, and could indicate that real-time PCR may be a useful tool for the evaluation of treatment efficacy in different Schistosoma species [8], [43]. Currently, we are awaiting the results of the renewed CCA and CAA-POC, based on the same set of urine samples. Further studies evaluating the Schistosoma DNA clearance in relation to circulating antigens after treatment will be of great interest and may provide useful information in relation to a more thorough understanding of post-treatment clearance of worms and eggs. At 18 months post-treatment, infection prevalences returned to 63.2% and 78.9% by microscopy and PCR, respectively, likely caused by reinfection [5], [39]. Given the obvious intense transmission in this area, rapid reinfection with cercariae and hence development of worms is very likely. The consequential increase in number of individuals excreting eggs as well as the quantity of eggs 18 months post-treatment diminishes the random distribution effect of samples without eggs even with active infection well-known in low or moderate infections [7], [8], [43]. Conclusive confirmation of eggs in urine by microscopy is thereby enhanced several months after treatment due to the reoccurrence of high prevalence and infection intensity. In our study, sensitivity of PCR was 92%, and sensitivity of microscopy increased to 74% at 18 months post-treatment.
Considering the high diagnostic accuracy and stability of real-time PCR over multiple time periods, the assay could be a valuable alternative to the current microscopic ‘gold standard’ for diagnosis of schistosome infections. This may especially be true for long-term monitoring of control interventions, as the assay seems particularly useful in demonstrating the presence of low intensity infections in the target population. In addition, we were able to analyze urine samples collected in 1996 and stored at −20°C for almost 15 years with real-time PCR, making retrospective diagnostic and/or epidemiological research feasible. On the other hand, real-time PCR is a laboratory-based test; the required equipment and consumables are expensive and training of laboratory technicians is essential. For cost-effective implementation, this procedure should preferably be performed in a centralized facility, meaning that, for the time being, real-time PCR cannot replace any of the existing methods readily applicable in the field. These include urine-reagent strips to detect haematuria for initial screening of those at risk of morbidity (S. haematobium) and urine POC-CCA assays for mapping and rapid screening of at-risk areas for S. mansoni [17]. The UCP-CAA strip assay (detecting all schistosome species) is much more sensitive than the POC-CCA. UCP-CAA can be utilized in the current robust format in low-resource settings, but is not yet available as a commercial POC diagnostic assay for schistosome infections [18]. Still, real-time PCR assays can make a valuable contribution in high-tech reference laboratories in especially settings approaching elimination of schistosomiasis, both as a diagnostic and a research tool, helpful in the decision-making process for appropriate intervention delivery, such as in confirmation of successful therapy, and, in the near future, vaccination [38]. Real-time PCR has many advantages: it can be run in a high through-put set-up, results can be standardized, it is possible to quantify parasites, and there is evidently the possibility of targeting several other helminth infections simultaneously in one multiplex real-time PCR. Potentially several non-human Schistosoma species could be included in order to study the distribution of zoonotic infections as well. Further studies in different endemic field settings evaluating real-time PCR using recently collected urine samples, as well as studies with larger sample sizes that validates the assay against other diagnostic methods, are encouraged.
|
10.1371/journal.pntd.0003702 | Mosquito-Disseminated Pyriproxyfen Yields High Breeding-Site Coverage and Boosts Juvenile Mosquito Mortality at the Neighborhood Scale | Mosquito-borne pathogens pose major public health challenges worldwide. With vaccines or effective drugs still unavailable for most such pathogens, disease prevention heavily relies on vector control. To date, however, mosquito control has proven difficult, with low breeding-site coverage during control campaigns identified as a major drawback. A novel tactic exploits the egg-laying behavior of mosquitoes to have them disseminate tiny particles of a potent larvicide, pyriproxyfen (PPF), from resting to breeding sites, thus improving coverage. This approach has yielded promising results at small spatial scales, but its wider applicability remains unclear.
We conducted a four-month trial within a 20-month study to investigate mosquito-driven dissemination of PPF dust-particles from 100 ‘dissemination stations’ (DSs) deployed in a 7-ha sub-area to surveillance dwellings and sentinel breeding sites (SBSs) distributed over an urban neighborhood of about 50 ha. We assessed the impact of the trial by measuring juvenile mosquito mortality and adult mosquito emergence in each SBS-month. Using data from 1,075 dwelling-months, 2,988 SBS-months, and 29,922 individual mosquitoes, we show that mosquito-disseminated PPF yielded high coverage of dwellings (up to 100%) and SBSs (up to 94.3%). Juvenile mosquito mortality in SBSs (about 4% at baseline) increased by over one order of magnitude during PPF dissemination (about 75%). This led to a >10-fold decrease of adult mosquito emergence from SBSs, from approximately 1,000–3,000 adults/month before to about 100 adults/month during PPF dissemination.
By expanding breeding-site coverage and boosting juvenile mosquito mortality, a strategy based on mosquito-disseminated PPF has potential to substantially enhance mosquito control. Sharp declines in adult mosquito emergence can lower vector/host ratios, reducing the risk of disease outbreaks. This approach is a very promising complement to current and novel mosquito control strategies; it will probably be especially relevant for the control of urban disease vectors, such as Aedes and Culex species, that often cause large epidemics.
| Mosquito-transmitted diseases are among the most challenging infectious threats worldwide. Mosquito control is crucial for preventing infection and disease, particularly when effective vaccines or drugs are unavailable. A major drawback of current mosquito control strategies is that mosquito breeding sites are often overlooked, and therefore left untreated, during control campaigns. One appealing alternative proposes exploiting the innate breeding-site–finding ability of female mosquitoes to have them disseminate tiny insecticide particles that poison their offspring. Thus far, however, this idea has only been tested in small-scale trials. Here we show that mosquitoes effectively transferred insecticide particles from dissemination stations to sentinel breeding sites over distances between 3 and 400 m in a tropical urban neighborhood. This yielded high breeding-site coverage, with up to 94.3% of sentinel breeding sites presenting evidence of contamination with mosquito-disseminated insecticide. We recorded a 10-fold increase of juvenile mosquito mortality and a 10-fold decrease of adult mosquito emergence during the four-month dissemination trial. In combination with other tactics, this approach has the potential to considerably enhance mosquito-borne disease prevention, particularly in urban settings.
| Mosquito-borne infectious diseases pose major public health challenges worldwide. Malaria and dengue are the most widespread, but other pathogens are also of concern, including viruses such as West Nile, chikungunya or Japanese encephalitis, and parasites such as those causing filariasis [1–5]. Urban vectors are especially problematic because they can transmit pathogens to large populations of susceptible humans, causing epidemics [1,2]. Since effective vaccines or treatments are available for only a few mosquito-borne diseases, prevention heavily relies on vector control; to date, however, mosquito control has proven difficult [1–3,6–9]. In particular with Aedes aegypti and Ae. albopictus (the vectors of dengue, chikungunya, and yellow fever), current strategies depend on the ability of mosquito control staff to detect and eliminate mosquito breeding sites in and around human residences [3,8]. Unfortunately, both Aedes species breed in small water-holding containers that can be difficult to detect, leading to low breeding-site coverage in control campaigns; this partially explains why the performance of such campaigns can be so poor [9]. In general, mosquito control tactics that rely on source reduction via larval habitat management all face the challenge of low coverage, whereby cryptic or inaccessible mosquito breeding sites remain untreated [3,8].
Proof-of-concept research has shown that the egg-laying behavior of female mosquitoes can be exploited to have them disseminate tiny particles of pyriproxyfen (PPF), a potent larvicide, from resting sites to nearby breeding sites [10,11]. This strategy relies on the innate ability of female mosquitoes to find and reach suitable breeding sites and on the ‘skip-oviposition’ behavior of some species, whose females visit several breeding sites to lay a few eggs in each [12,13]. The idea entails luring mosquitoes to ‘dissemination stations’ treated with PPF dust-particles that adhere to the insect’s body and are thus transferred to clean breeding sites subsequently visited for oviposition [10].
Although appealing, this approach has only been tested in small areas, with PPF dissemination measured at very short distances [10,11,14–16]; recently, a larger trial (ref. [17]) used an emulsifiable-PPF spray instead of dust-particle dissemination stations. Here, we investigate whether adult mosquitoes can transfer PPF particles from lure dissemination stations to sentinel breeding sites in a tropical neighborhood, and assess the impact of mosquito-disseminated PPF on juvenile mosquito mortality and adult mosquito emergence.
All field procedures were carried out with permission from dwelling owners. Sérgio LB Luz holds a permanent license (27733–1) from the Brazilian Institute for the Environment and Natural Resources (IBAMA) for sampling disease vectors.
The study took place at the Tancredo Neves neighborhood in the city of Manaus, Amazonas, Brazil (3°6’S, 60°1’W; Fig 1). Tancredo Neves is a lower middle-class residential neighborhood where most people live in single-family houses with a small yard; house/yard compounds will be referred to as ‘dwellings’ hereafter. Aedes aegypti and Ae. albopictus infest most dwellings in the study area, where dengue cases are common [9].
We set up a mosquito-surveillance network spanning about 50 ha of Tancredo Neves and comprising 55 randomly selected dwellings. The presence of egg-laying mosquitoes in each dwelling was sampled via ‘sentinel breeding sites’ (SBSs). SBSs were 580-ml dark-brown plastic cups (Fig 1) baited with 200 ml of hay infusion (approximately 5g dry Zoysia sp./liter of tap water, fermented for six days in a closed plastic container) diluted in 250ml of tap water. As for standard ovitraps, SBSs offer artificial breeding habitats that can be promptly set/checked and from which larvae can be removed for further analysis; we used screw-capped cups to avoid any spillover of SBS contents (hence possible cross-contamination) during SBS retrieval and transportation to the laboratory. Mosquito surveillance was run monthly from January 2011 to September 2012 (see study timeline in Fig 1) by simultaneously setting three SBSs in each dwelling during six days per month. We coded each SBS individually to ensure that each was set always at the same location; SBSs were thoroughly washed with water and soap between monthly sampling rounds. Any missing SBS was replaced by a new one with a different code in the next sampling round. Overall, we analyzed dwelling-level data from 1,075 dwelling-months (excluding dwellings that were unavailable for sampling at certain months) and breeding site-level data from 2,988 SBS-months (excluding SBSs that did not produce data in a given month because they were overturned, went missing, or corresponded to dwellings that were unavailable for sampling). Dwelling-level data allowed us to investigate whether mosquitoes would effectively disseminate PPF over the whole study area, while SBS-level data provided insight on (i) breeding-site coverage (as measured by the fraction of SBSs that became contaminated with mosquito-disseminated PPF) and (ii) the effects of the trial on juvenile mosquito mortality and adult mosquito emergence (see below).
SBSs were retrieved after six days of operation to avoid emergence of adult mosquitoes in the study dwellings. Once in the laboratory, the contents of each SBS were transferred to a white plastic cup to ease the observation of mosquito juveniles; each cup received the same code as the corresponding SBS and was capped with gauze and kept for 8–16 days to monitor mosquito development. A pinch of TetraMin fish food (Tetra, Melle, Germany) was added every other day to each cup. Mosquito larvae found in each individual SBS-month were identified as Ae. aegypti, Ae. albopictus or Culex spp. [18] (we ignore a few, rarer taxa in the present analyses), and were checked every two days to score juvenile mosquito death or adult mosquito emergence. For each SBS-month, juvenile mortality was estimated as the percent of individuals that died as larvae or pupae, and adult mosquito emergence as the sum of all individuals that emerged as adults.
‘Dissemination stations’ (DSs; Fig 1) were two-liter, black plastic cups with 400 ml of tap water and the inner wall lined with black, velvet-like cloth dusted with 5 g/m2 of PPF (Sumilarv 0.5 g granules, Sumitomo, London, UK) ground to fine powder in a metal mortar. PPF is an insect juvenile-hormone analog that kills immature mosquitoes, especially pupae, at extremely low doses; it also reduces fertility in adult mosquitoes, but has no lethal or repellent effects on them [10,16]. PPF is recommended by the World Health Organization as a safe mosquito control means even in drinking water [3], and is currently endorsed by the Brazilian Ministry of Health [8]. One hundred DSs were deployed in a sub-area of about 7 ha nested within the study area; DSs were 3 to 397 m from the nearest SBS (Fig 1). DS deployment (‘the trial’ hereafter) took place from December 2011 (month 11) to March 2012 (month 14), coinciding with the rainy season [9]; all DSs were removed from the field at the end of month 14 (Fig 1). DSs were placed in sheltered locations and checked fortnightly throughout the four months of the trial to refill water, re-dust cloth with PPF, and replace lost cups.
In the laboratory, individual SBSs were scored each month by one of the investigators (Elvira Zamora-Perea) as contaminated or not contaminated with PPF. Contamination was inferred when mosquito juveniles in the SBS developed the abnormal morphology and coloration that characterizes PPF poisoning (large bodies, blackish color; see S1 Fig). In addition to inducing these marked morphological abnormalities, PPF increases larval development time from the typical 8–9 days until adult emergence to 14–16 days until (usually) pupal death. We pre-tested the effectiveness of our PPF in a double-blind, randomized, controlled laboratory trial using 30 independent cohorts of 20 Ae. aegypti larvae each. All juveniles in the 15 cohorts treated with PPF (0.05 ppm a.i.) died over three weeks of monitoring, whereas only one larva died in the 15 control cohorts (see S1 Text). Juveniles in treated cohorts developed the morphological abnormalities typical of PPF poisoning (see S1 Fig).
After exploratory/descriptive analyses, we used generalized linear models (GLMs; binomial family, logit link) to analyze binary outcome data (i) at the dwelling level and (ii) at the breeding-site level. At the dwelling level, the binary outcome was 1 for dwellings with at least one SBS presenting evidence of contamination with mosquito-disseminated PPF and 0 otherwise. At the breeding-site level, the binary outcome was 1 for SBSs with evidence of contamination and 0 for those without. We investigated the effects of two key predictors: (a) time-period, comparing 10 months ‘before’, 4 months ‘during’, 3 months ‘early after’, and 3 months ‘late after’ the trial; and (b) log10-distance (in meters) between each dwelling and the nearest DS, which was also used to approximate the distance between SBSs set in each dwelling and the nearest DS (see Fig 1 for timeline and spatial arrangement of DSs and dwellings). Because no PPF was present in the environment before the trial, we used Firth’s correction [19] to estimate period effects with ‘before the trial’ as reference level. ‘Distance*period’ interactions were also tested. Dwelling-level models adjusted for the number of SBSs that were operational in each dwelling and month, which was specified as a three-level categorical covariate (1, 2, or 3 operational SBSs, with ‘1 SBS’ as the reference level). Relative model performance was assessed using second-order Akaike information criterion (AICc) scores and related metrics (ref. [20] and S1 Text); likelihood-ratio tests were used to evaluate covariate contribution to model fit. Categorical variables were analyzed with Pearson χ2 tests or conditional maximum-likelihood odds ratios [21]. Crude juvenile mortality rates were compared with nonparametric Kruskal-Wallis rank-sum and post hoc Tukey tests. Contour plots were built to spatially visualize GLM predictions and juvenile mosquito mortality data. We used linear regression to illustrate the effect of distance to the nearest DS on juvenile mosquito mortality. We analyzed the data using JMP 9.0 (SAS Institute, Cary, NC).
All surveillance dwellings presented evidence of contamination with mosquito-disseminated PPF in ≥1 SBS at some time-point during DS deployment. There was evidence of contamination in 75.5%, 80%, 100%, and 94.4% of surveillance dwellings in months 11, 12, 13, and 14, respectively. Afterwards, dwelling-level PPF coverage fell from 79.2–81.5% (months 15–16) to 1.9% (month 20) of dwellings. Table 1 summarizes dwelling-level data over the four study periods.
GLMs revealed strong period effects and a negative effect of distance; ‘distance*period’ interactions were not significant. According to the main-effects GLM (Table 2), the odds that a dwelling had evidence of contamination were 96.9 times higher during than before the trial (likelihood-ratio test, χ2 = 692.8, d.f. = 1, P<0.0001; Table 2, Fig 2). A substantial decline in contamination odds was detected only 4–6 months after DSs were removed (Table 2, Fig 2). The odds of contamination decreased at an average rate of 54.5% for each 10-fold increase in distance between dwellings and DSs (Table 2). A model with only period effects provided a poor fit to the data (ΔAICc = 11.87), but a distance-only model performed much worse and similarly to the intercept-only model (see S1 Text); thus, while both covariates substantially helped explain the data, period effects were more important than distance effects.
Most of the SBSs that contained larvae (i.e., were visited by ≥1 egg-laying mosquito) consistently became contaminated during the trial: from 67.9% in month 11 to 94.3% in month 13. Afterwards, contamination fell back to 65.7% in month 15 and 1.7% in month 20 (Fig 3). Overall, we found evidence of PPF contamination in >85% of SBSs that were visited by egg-laying mosquitoes during the four-month trial period, with a steep decline afterwards (Table 3).
GLMs revealed period and distance effects similar to those seen at the dwelling level (see Tables 2 and 4). ‘Distance*period’ interactions were, again, not significant; the main-effects model is presented in Table 4, and its spatially-plotted predictions are shown in Fig 4. AICc clearly favored this model over simpler versions (S1 Text).
Due to the abundance and oviposition behavior of Ae. aegypti, we hypothesized that, during the trial, SBSs with larvae of this species would have higher odds of presenting evidence of PPF contamination than SBSs without. Such odds were 329% higher in SBSs with Ae. aegypti larvae (91.2% contaminated) than in those with only Ae. albopictus and/or Culex spp. larvae (70.5% contaminated; odds ratio 4.29, 95%CI 2.47–7.54). A similar effect was recorded when comparing SBSs with Ae. aegypti larvae only (i.e., probably visited only by Ae. aegypti) vs. those without (odds ratio 3.42, 95%CI 1.89–6.41); see details in S1 Table.
Juvenile mortality was assessed based on 29,922 mosquito larvae/pupae present in 2,287 SBS-months; overall, 9.2% of those mosquitoes (95%CI 8.9–9.5%) died as juveniles. Before PPF dissemination, overall larval/pupal mortalities in SBSs were approximately 2.0/0.1% (Ae. aegypti), 1.5/0.2% (Ae. albopictus), and 6.9/0% (Culex spp.). During the trial, these figures reached peak values of 27.9/80.7%, 43.6/70%, and 16.7/54.8%, respectively; pre-trial values were restored by months 15–16 (see S1 Dataset).
Before the trial, species-pooled mean juvenile mortality across SBSs was 4.2% (SE = 0.5); 0% mortality was recorded in 87.8% of SBSs with ≥1 larva (data from 1,124 SBSs and 11,970 mosquitoes). Mean juvenile mortality across SBSs rose to 75.1% (SE = 1.8) during the four months of PPF dissemination, with 100% mortality recorded in 61.6% of SBSs with ≥1 larva (data from 427 SBSs and 2,392 mosquitoes). Mean juvenile mortality progressively declined afterwards to 15.8% early after (n = 365 SBSs) and to just 0.6% late after the trial (n = 371 SBSs).
Mean monthly mortality of Ae. aegypti juveniles in SBSs rose from a 0–10% range before the trial (median and inter-quartile range [IQR] all 0%) to 62–94% during the trial (median and IQR all 100%), and fell back to 0.3–1.4% (median and IQR all 0%) in the final 3-month period. Ae. albopictus mean monthly juvenile mortality was <2% (range across months 0–4.6%) before and about 64% (range 29.5–84.2%) during dissemination, and quickly fell back to baseline values after the end of the trial. Juvenile mosquito mortalities were significantly different across the four study periods: Kruskal-Wallis test of species-pooled mortality, χ2 = 1,140.5 (d.f. = 3, P<0.0001). Tukey tests suggested, however, that Ae. aegypti mortality was comparable before and late after the trial, with a marginally significant difference when considering all species (see details in S2 and S3 Tables).
Juvenile mosquito mortality was over 20 times higher in SBSs with evidence of PPF contamination (mean across months 67.3%; SE = 1.6) than in SBSs without such evidence (2.8%; SE = 0.3); 100% mortality was recorded in 287 of 564 contaminated SBSs (50.9%; 95%CI 46.8–55%) and in 17 of 1,723 non-contaminated SBSs (1%; 95%CI 0.6–1.5%). During the four months of PPF dissemination, mean mortality in SBSs with evidence of PPF contamination reached 87.9% (n = 364 SBSs), vs. 0.8% in SBSs without such evidence (n = 63).
Fig 5A shows monthly mean juvenile mortality (all species pooled) in SBSs that contained ≥1 mosquito larva. Fig 5B shows results for Ae. aegypti (n = 1,224 SBS-months): juvenile mortality reached 94.3% (95%CI 90.1–98.4%) in month 13, when 430 individuals in 124 SBSs were scored for mortality or emergence. Again, these large differences among periods were highly significant (Kruskal-Wallis P<0.0001).
Finally, juvenile mosquito mortality decreased with increasing distance between DSs and SBSs during and, especially, early after the trial; on the contrary, no significant distance effects were evident before or late after DS deployment (Fig 6). Although larger and more persistent effects were apparent in and near the DS-deployment sub-area, the rise of mortality was evident throughout the study area, particularly for Ae. aegypti (see S2 Fig).
The median number of mosquitoes that completed development in SBSs each month before PPF dissemination was 1,177 (IQR 851–1,427), as compared to just 107 (IQR 71.8–201.5) during the trial (see S1 Dataset). Adult mosquito emergence rose back to 1,408 (IQR 972–1,910) early after the trial and, somewhat surprisingly, peaked late after DS removal to 3,435 (IQR 3,270–4,000) (Kruskal-Wallis χ2 = 840.2, d.f. = 3, P<0.0001; see S1 Dataset). For a more general comparison, the median monthly number of immature Aedes spp. collected in SBSs in the same area and dwellings over the 28 months preceding the present study was 2,481 (IQR 1,556–2,811) (see S1 Text); at a 4% typical baseline rate of juvenile mortality, monthly adult emergence can be estimated as about 2,400 (IQR 1,500–2,700). Overall, then, monthly adult mosquito emergence from SBSs was reduced by over one order of magnitude during the trial.
This paper shows that urban mosquitoes can be very effective at transferring pyriproxyfen dust-particles from simple dissemination stations to sentinel breeding sites at the neighborhood scale. All surveillance dwellings and most SBSs had evidence of contamination with mosquito-disseminated PPF at some time-point during the trial. This dramatically increased juvenile mosquito mortality in SBSs, leading to a >10-fold decrease of adult mosquito emergence. These findings confirm previous encouraging results from laboratory assays and small-scale field trials [10,11,14–16] and demonstrate that mosquito behavior can effectively be harnessed to disseminate insecticides at the neighborhood scale in real-life settings. This approach has the potential to substantially enhance mosquito control and mosquito-borne disease prevention, particularly in urban settings.
Female mosquitoes have evolved to maximize their efficiency at locating and reaching suitable breeding sites. One egg-laying strategy, displayed by Ae. aegypti and other species that breed in small containers, is to visit several distinct sites to lay a few eggs in each. This ‘skip oviposition’ behavior [12,13] may have helped increase coverage of SBS contamination with mosquito-disseminated PPF in our trial. Our results show that very simple PPF-dusted DSs can result in highly effective mosquito-driven dissemination; emulsifiable-formulation sprays seem to be less effective [17], perhaps because mosquitoes are more likely to pick PPF particles on the dusted cloth of DSs than on sprayed surfaces. Furthermore, our DSs were deployed at sheltered sites to protect them from rain or direct sunlight and were re-dusted periodically, which probably increased dissemination efficacy [17]. These findings signal potentially important operational details for PPF-based interventions. Our data also suggest that Ae. aegypti may be more efficient at disseminating PPF than Ae. albopictus or Culex spp. (S1 Table). Interventions based on mosquito-disseminated PPF might therefore be less effective when Ae. aegypti is absent (e.g., ref. [17]).
Low breeding-site coverage is a major shortcoming of current mosquito control strategies [9,10,22]. Thus, dengue vector control campaigns based on breeding-site detection and elimination had negligible effects on dwelling infestation by Aedes spp. in our study setting [9] and elsewhere in Brazil [23]. Together with the negatively-biased house infestation indices provided by routine breeding-site surveillance, this suggests that vector control agents overlook many Aedes breeding sites while inspecting premises [9,24]. Achieving adequate coverage is even more challenging for Ae. albopictus, which can breed in natural sites such as tree holes or epiphytic bromeliads that routine vector control does not target [25,26]. This is particularly relevant in the current context of chikungunya virus reemergence and fast spread from Africa into Asia, Oceania, Europe, and, more recently, the Americas [1,2,5,27,28; www.cdc.gov/chikungunya/]. Chikungunya transmission by Ae. albopictus can be very efficient and has been reported from several settings [28,29]. By enhancing coverage of natural breeding sites, mosquito-disseminated PPF could help control this vector species and perhaps contribute to containing the expansion of chikungunya fever. The approach, in sum, provides a powerful tool to increase breeding-site coverage, extending larvicide dissemination to sites that control agents would never detect or treat, such as sites inside closed premises or most natural breeding sites; at the same time, conspicuous breeding sites such as water tanks or catch basins could be treated directly by vector control staff following standard practice. Mosquito-disseminated PPF could therefore substantially enhance current vector control tactics, not only for dengue and chikungunya [3,6,8,10] but also for other mosquito-borne pathogens including West Nile virus (see below).
We note that our trial was conducted in an open area within a neighborhood and city where mosquitoes are widespread pests. Adult mosquitoes from adjacent, untreated areas could therefore freely migrate into the trial area and replace local recruitment that was lost due to PPF; this limited our ability to investigate the broader impact of the trial on the local mosquito population. Still, even under those circumstances, we detected a measurable decline of infestations at the dwelling level (i.e., the presence of ≥1 larva in any of a dwelling’s SBSs) by Culex spp. and, to a lesser extent, Ae. albopictus after DS deployment (see below and S3 Fig). The number of immature mosquitoes collected in SBSs also decreased during DS deployment and recovered afterwards (see S4 Fig). On the other hand, immigrating adult mosquitoes may have helped increase PPF coverage of SBSs and surveillance dwellings. Our results also suggest that Ae. aegypti females, thought to be poor fliers [30,31], can carry PPF over distances up to 400 m across a neighborhood where suitable breeding sites are readily available [32,33]. This dispersal ability has implications for understanding local dengue-spread dynamics [32–34] and suggests that a productive breeding focus in a non-treated premise can act as the source of adult mosquitoes for an area of about 50 ha.
Because we set clean SBSs each month, we did not assess the residual effects of PPF, which may have been important in breeding sites we did not monitor [35]. For example, observed dwelling-level infestation (i.e., presence of at least one larva in at least one SBS) by Culex spp. steadily fell from 30.1% at baseline to 9.3% during, 6.9% early after, and 2.5% late after the trial (see S3 Fig). We speculate that this might reflect persistent contamination of some key Culex breeding sites in the study area, hinting at the potential of mosquito-disseminated PPF for the control of West Nile virus- or filariae-transmitting Culex spp. In addition, the fact that many SBSs became contaminated after DSs were removed indicates that PPF particles persisted in the environment and were still being disseminated several months after the trial ended. Early after the trial, this might be related to mid-term PPF carriage by mosquitoes that picked dust particles at DSs but only lost them after several ovipositions. However, because of the short lifespan of non-diapausing adult mosquitoes [36], long-term PPF carriage cannot explain SBS contamination events late after DS removal. The contamination of resting sites by PPF-carrying mosquitoes, with dust particles then picked-up and secondarily disseminated by other mosquitoes, is more likely in these cases. Environmental persistence is nonetheless expected to be short, because PPF degrades quickly [37]; interestingly, over the last eight months of our study the fraction of SBSs with evidence of PPF contamination closely matched the expected fraction of PPF particles remaining active in the environment, given a 30-day half-life [37]: linear regression, R2 = 0.91, P = 0.0003 (S5 Fig).
Mortality of juvenile mosquitoes in SBSs with evidence of PPF contamination was within the range reported from small-scale [10,11,16] and direct-impact PPF trials [17]. Our results are unique because they indicate that mosquito-disseminated PPF increased juvenile mortality and reduced adult emergence at the neighborhood scale, with each of these metrics changing by over one order of magnitude. Importantly, these effects came about in spite of (i) mortality reaching 100% in just over 50% of contaminated SBSs, likely because of PPF under-dosage; (ii) substantial, yet incomplete, breeding-site coverage (as measured by SBS coverage); and (iii) the fact that the trial was conducted during the rainy-cool season, when the availability of breeding sites is at its peak and Aedes populations are unlikely to undergo local extinctions [9]. This approach can therefore be expected to perform well even under constraints and imperfections that are typical of real-life vector-control campaigns. We note, in addition, that we did not investigate the effects of PPF on adult mosquitoes (e.g., malformations, shorter lifespan, or reduced fertility), and that some dead early-stage larvae may have been scavenged by other larvae before SBSs were checked in the laboratory [10,14,16]. Therefore, our results likely underestimate the impact of the trial.
A key limitation of our study is that we lacked the technical means to measure the minute, parts-per-billion concentrations of PPF that kill mosquito juveniles; hence, and as in previous trials (e.g., [10,11]), we lack direct evidence of PPF contamination in our SBSs. Still, we think it extremely unlikely that our observations might stem from any unmeasured event. Evidence of PPF-induced mortality was assessed by a researcher with broad experience in the appraisal of juvenile mosquito development and morphology in PPF laboratory and field trials [10,16]. This evidence (S1 Fig) was only recorded after DS deployment. In our pre-trial laboratory tests, PPF-treated and PPF-untreated larvae had the expected morphology and could be unambiguously identified (pers. obs.; S1 Fig). Further, the dramatic increase of juvenile mortality at the time of DS deployment (Figs 5, 6 and S2 Fig) can hardly be explained by any alternative phenomenon. Not only there was an abrupt leap in mortality as the trial started: mortality also declined with distance from DSs during and early after the trial, and gradually fell back to baseline values when the DSs were removed from the environment (Figs 5, 6 and S2 Fig). We had monitored these local mosquito populations for years before this trial and never recorded any such sudden demographic shift (e.g., [9,38]). Finally, longitudinal contamination data in individual SBSs provided no evidence of either laboratory contamination (non-contaminated SBSs were recorded every month) or persistence of contamination from one month to the next (>70% of SBSs scored as contaminated in month m were scored as not contaminated at least once in month m+1; see S1 Text). Thus, mosquito-driven PPF dissemination is by far the best explanation available for our findings.
Our results provide evidence that urban mosquitoes can be very effective at transferring PPF dust-particles from simple dissemination stations to artificial breeding sites at the neighborhood scale. Maximum monthly coverage was 94.3% for SBSs and 100% for surveillance dwellings over 50 ha, and juvenile mosquito mortality reached 87.9% in SBSs contaminated by PPF-disseminating mosquitoes. This resulted in a >10-fold rise of juvenile mosquito mortality and a >10-fold fall of adult mosquito emergence; by lowering vector/host ratios, these strong effects can help reduce the risk of arboviral disease outbreaks [39]. We conclude that this approach is a very promising complement to current mosquito control strategies, which heavily rely on the difficult task of detecting vector breeding sites and therefore perform poorly. Mosquito-disseminated insecticides could profitably be combined both with current, standard control practices and with novel, more sophisticated tactics involving transgenic or Wolbachia-infected mosquitoes [40–43].
|
10.1371/journal.pgen.1003903 | The Histone Variant His2Av is Required for Adult Stem Cell Maintenance in the Drosophila Testis | Many tissues are sustained by adult stem cells, which replace lost cells by differentiation and maintain their own population through self-renewal. The mechanisms through which adult stem cells maintain their identity are thus important for tissue homeostasis and repair throughout life. Here, we show that a histone variant, His2Av, is required cell autonomously for maintenance of germline and cyst stem cells in the Drosophila testis. The ATP-dependent chromatin-remodeling factor Domino is also required in this tissue for adult stem cell maintenance possibly by regulating the incorporation of His2Av into chromatin. Interestingly, although expression of His2Av was ubiquitous, its function was dispensable for germline and cyst cell differentiation, suggesting a specific role for this non-canonical histone in maintaining the stem cell state in these lineages.
| Many tissues in the body are maintained by adult stem cells, which are dedicated but undifferentiated precursors that both maintain their population throughout life and produce daughter cells that differentiate to replace cells lost to turnover or damage. Here we show that the histone variant His2Av is required cell autonomously for maintenance of both germline and somatic cyst stem cells in the Drosophila testis. Although His2Av is expressed ubiquitously, under normal conditions, function of this histone variant was not required for correct differentiation of stem cell progeny in testes or for the survival of cells in the developing eye. We propose that adult stem cells maintain a plastic, bipotential state able to switch between self-renewal and differentiation and that His2Av may provide a chromatin state that helps bias transcription programs towards the stem cell fate.
| Many adult tissues with short-lived, highly differentiated cells such as blood and skin replace cells lost to turnover through the proliferation and differentiation of adult stem cells. Adult stem cells must also self-renew to maintain a source of differentiating cells in the long term. The mechanisms that control the balance between self-renewal and differentiation need to be tightly regulated to maintain homeostasis of adult tissues. Although recent work has focused on signals from the local microenvironment of the stem cell niche, responses to these signals take place in the context of cell autonomous properties of the stem cell state that influence the ability of adult stem cells to maintain their identity. Likely candidates for such cell autonomous properties include the state of chromatin at key regulatory genes that influence stem cell maintenance.
The basic unit of eukaryotic chromatin, the nucleosome, is formed by DNA wrapped around an octamer containing two copies each of histones H2A, H2B, H3, and H4. Access to DNA by transcription factors and RNA polymerase is achieved by factors that control the post-translational modifications of core histones [1] and/or remodel nucleosomes [2]. The replacement of canonical histones with histone variants has recently emerged as an additional mechanism regulating chromatin accessibility [3]. Variants of the canonical histone H2A are highly conserved across species and play roles in transcriptional control, formation of heterochromatin boundaries, lineage commitment, and DNA repair. In yeast and mammals, H2AX is involved in recruiting factors to the sites of DNA damage [4] and H2A.Z is implicated in transcriptional regulation [5], [6]. In Drosophila, His2Av, the only known variant of H2A, assumes functions of both H2AX and H2A.Z [7]. Drosophila His2Av and His2A share 55% of their amino acid sequences, with the C-terminal region of His2Av considerably longer than that of His2A [8].
Here, we show that the histone variant His2Av is required cell autonomously for the maintenance of two adult stem cell populations in the Drosophila testis. The stem cell-niche microenvironment at the apical tip of the Drosophila testis consists of the germline stem cells (GSCs), which give rise to sperm [9]; the cyst stem cells (CySCs), which give rise to the cyst cells that enclose germ cells as they differentiate [10], [11]; and the post-mitotic somatic hub cells, to which GSCs and CySCs attach [12], [13]. His2Av function is required for both GSC and CySC maintenance; however, its function was dispensable for the differentiation program in the germ and cyst cell lineages. Our results suggest that in the absence of DNA damaging agents, the transcriptional role of His2Av may be required to regulate the delicate balance between self-renewal and differentiation states in adult stem cells.
Immunostaining of wild-type adult testes revealed His2Av protein expression in many cell types in the adult testis of Drosophila. At the apical tip of the testis, His2Av localized to the nuclei of somatic cells of the hub, GSCs (Fig. 1A) and CySCs (Fig. 1C). In differentiating spermatocytes, His2Av was concentrated on the autosomal and sex bivalent chromosomes within the nucleus (Fig. 1B, inset). His2Av also localized to the nuclei of differentiating somatic cyst cells associated with spermatocyte cysts (Fig. 1D).
Clonal analysis revealed that His2Av function is required cell autonomously for stem cell maintenance in the Drosophila male germline. Negatively marked GSCs lacking His2Av function were generated in adult fly testes by mitotic recombination using the FLP/FRT system in a His2Av810/+ background [14]. GSCs at day 3 post clonal induction (PCI) (Fig. 1E) and later germline clones at day 8 PCI (Fig. S1A) homozygous mutant for His2Av810 did not exhibit His2Av staining, indicating specificity of the antibody towards His2Av protein and a sharp decline in protein levels in His2Av mutant GSCs by at least day 3 PCI. At day 2 PCI, GSCs homozygous mutant for His2Av810 were detected in 75% of the testes scored, similar to the 81.4% observed in controls (Fig. 1F). By day 8 PCI, the percentage of testes with at least one marked GSC clone dropped to 2% for the His2Av810 mutant (Fig. 1F), suggesting a defect in GSC maintenance upon loss of His2Av function, while 64.8% of control testes had at least one marked GSC.
Consistent with the loss of mutant GSCs, His2Av810 mutant spermatocytes were not maintained over time after clone induction. At day 4 PCI, His2Av810 spermatocytes were observed in 86% of testes. However, by day 12 PCI, His2Av810 mutant spermatocytes were no longer observed (Fig. 1G). A genomic transgene carrying the His2Av coding sequence under control of its endogenous promoter and fused to the mRFP coding sequence (His2Av-mRFP) [15] rescued the loss of spermatocytes, indicating that the failure to maintain GSCs and their differentiating progeny was due to loss of His2Av function (Fig. 1G, Fig. S1B, C).
Knockdown of His2Av function specifically in GSCs and early germ cells by expression of a RNAi hairpin for His2Av using the nanos-GAL4-VP16 (NGVP16) driver also indicated a cell autonomous role for His2Av in GSC maintenance. By day 3 after RNAi expression, induced by shifting flies from 18°C to 30°C, His2Av protein levels in GSCs dropped considerably compared to controls (Fig. S2A, B). At day 4 after RNAi induction, visualization of testes by phase contrast microscopy revealed the presence of spermatocytes and elongated spermatids in testes expressing His2Av RNAi and in controls (Fig. 2A, B). By day 12, however, testes expressing His2Av RNAi exhibited germ cell loss and did not contain spermatocytes or elongated spermatids (Fig. 2D), while control testes at day 15 still had both cell types (Fig. 2C). Quantitation of GSC number revealed that the loss of germ cells observed 12 days after RNAi induction was due to a failure to maintain GSCs. At day 0, His2Av RNAi expressing and control testes had an average of 7 and 8.2 GSCs, respectively (Fig. 2E, F, I). By day 12, the number of GSCs adjacent to the hub in testes expressing His2Av RNAi had dropped to 0, while control testes contained an average of 7.8 GSCs per testis hub (Fig. 2G, H, I).
In contrast to its role in GSCs, His2Av was not required cell autonomously for germ cell differentiation. Germline clones homozygous mutant for His2Av810 differentiated into spermatocytes (Fig. 3A and Fig. S1A) and round and elongating spermatids (Fig. 3B, C), as observed 8 days after clone induction. Mutant onion stage round spermatids had the normal size and 1∶1 ratio of nuclei to mitochondrial derivatives, indicating successful progression through meiotic divisions (Fig. 3B). Knockdown of His2Av in late spermatogonial cysts by RNAi expressed under the control of the bam-Gal4 driver confirmed that His2Av function is dispensable for the differentiation program of germ cells at the later stages. His2Av protein levels were greatly reduced in spermatocytes upon expression of RNAi (Fig. S2C, D), yet spermatocytes lacking His2Av protein for 8 days after RNAi induction were still able to differentiate, undergo meiosis, and give rise to elongated spermatids (Fig. 3D, E).
Although His2Av mutant GSCs were lost to differentiation, they did not appear to do so by accumulating Bam protein earlier than their heterozygous counterparts. The accumulation of Bam protein in transit-amplifying spermatogonial cells stops proliferation and initiates differentiation to spermatocytes [16]. Immunostaining for Bam protein 5 days PCI revealed that neither heterozygous His2Av810/+ nor homozygous His2Av810 mutant GSCs or gonialblasts expressed Bam protein (Fig. 3F). Bam protein did accumulate at the correct time during the differentiation program in His2Av mutant cells, at the 4-cell spermatogonial stage (Fig. 3F), similar to in wild-type spermatogonial cysts. Consistent with the correct temporal accumulation of Bam protein, germ cells lacking His2Av function underwent 4 rounds of spermatogonial divisions, producing cysts with 16 spermatocytes (Fig. 3G).
Consistent with its expression in the cyst cell lineage, His2Av function was also required cell autonomously for CySC maintenance. Although both His2Av810 mutant and control CySCs were present at comparable frequencies at day 2 PCI, by day 8 PCI almost all testes lacked His2Av810 mutant CySCs, while control CySCs were maintained (Fig. 4A). Consistent with the loss of mutant CySCs, His2Av810 mutant cyst cells expressing the differentiation marker Eya were also lost over time. His2Av810 mutant cyst cells were observed in 100% of testes at day 4 PCI, but by day 12 PCI His2Av810 mutant Eya-positive cyst cells were almost entirely absent (Fig. 4B). The His2Av-mRFP transgene rescued the loss of His2Av810 homozygous mutant CySCs, suggesting that the failure to maintain CySCs was due to loss of His2Av function. At day 2 PCI, an average of 43.2% (n = 32) of testes contained His2Av810 mutant CySCs, while under the same conditions, 67.5% (n = 23) testes from sibling males carrying the His2Av-mRFP transgene contained His2Av mutant CySCs (data not shown). The percentage of testes containing His2Av810 mutant CySCs at day 8 PCI dropped to 3% (n = 26), while 42.1% (n = 31) of testes from males carrying the His2Av-mRFP transgene contained marked CySCs.
The failure to maintain CySCs was not due to downregulation of the transcriptional repressor Zinc-finger homology-1 (Zfh-1), which is expressed in CySCs and is required for CySC maintenance [17]. At day 3 PCI, when only 20% of testes scored had homozygous mutant CySCs, Zfh-1 expression in His2Av810 homozygous mutant CySCs was comparable to that in neighboring wild-type CySCs (Fig. 4C). As in the germ line, His2Av function was not required for cyst cell differentiation. Cyst cells lacking His2Av function differentiated successfully at least to the stage at which they express the differentiation marker Eya and are associated with differentiating germ cells (Fig. 4D).
In addition to the survival and differentiation of germ cells and cyst cells lacking His2Av, the classic eye test revealed that His2Av function might be dispensable for cell survival in the eye tissue. Eyes composed exclusively of cells lacking His2Av function were generated using the EGUF/hid system [18]. When mitotic recombination was not induced, eye precursor cells expressed the GMR-hid transgene and failed to develop, resulting in adult flies with tiny eyes (Fig. 4E). In contrast, when clones were induced, cells lacking His2Av function produced eyes (Fig. 4F), although they appeared slightly smaller and rougher compared to controls (Fig. 4G), suggesting that His2Av might contribute to proper cell proliferation and/or differentiation in this tissue. Together, the results from clonal and RNAi analysis in the germline, somatic cyst, and eye cell lineages suggest that in the absence of DNA damaging agents, Drosophila His2Av function is required for adult stem cell maintenance but not for cell survival or differentiation.
Analysis of His2Av mutant GSCs revealed that His2Av function was not required to maintain three previously defined STAT-dependent characteristics of GSCs: 1) attachment to the hub through E-cadherin mediated adherens junctions, 2) oriented cell division [19], and 3) upregulation of STAT92E protein in response to Unpaired (Upd) signaling from the hub. His2Av mutant GSCs localized E-Cadherin-GFP (E-Cad-GFP), expressed in GSCs by the nanos-Gal4 driver and detected 5 days PCI, to the hub-GSC interface similar to neighboring heterozygous GSCs (Fig. 5A) and as previously shown [12]. The expression of E-Cad-GFP in GSCs did not result in an increase in His2Av mutant GSC maintenance; His2Av810 mutant GSCs in testes from sibling males either expressing or lacking the expression of E-Cad-GFP were lost at the same rate (Fig. 5B). His2Av also did not appear to be required for the stereotypical orientation of centrosomes in GSCs that sets up the mitotic spindle orientation and the subsequent asymmetric outcome of GSC division [12]. Analysis of testes 3 days after induction of His2Av810 clones revealed that in GSCs with two centrosomes, one centrosome was found adjacent to the hub-GSC interface in 86.3% of His2Av810 mutant GSCs, similar to neighboring heterozygous His2Av810/+ GSCs (86.42%) and FRT control GSCs (84.62%) (Fig. 5C–E). Loss of His2Av function did not substantially alter the accumulation of STAT92E, an indicator of JAK-STAT activity [20], in GSCs. His2Av810 mutant GSCs remaining adjacent to the hub 5 days PCI had STAT92E protein levels comparable to neighboring heterozygous GSCs (Fig. 5F), suggesting that loss of GSCs in His2Av mutants is not due to failure to express STAT92E. Conversely, GSCs homozygous mutant for either Stat92E06346 (Fig. 5G) or Stat92Ejc46 (data not shown) expressed His2Av protein at levels comparable to neighboring heterozygous GSCs, suggesting that His2Av expression in GSCs was not dependent on STAT92E function.
Loss of His2Av function did not suppress the overproliferation of CySC-like and GSC-like cells in testes with forced activation of the JAK-STAT pathway. When the Upd ligand was expressed ectopically in early germ cells under the control of the nanos-Gal4 driver, larval testes heterozygous for His2Av had an abundance of small Vasa-positive GSC-like cells with dot spectrosomes and Zfh-1 positive CySC-like cells (Fig. 5H). Under the same conditions, testes from sibling nos-Gal4/UAS-Upd; His2Av810/Df(3R) BSC524 larvae also exhibited an abundance of GSC-like and CySC-like cells (Fig. 5I), although there were subtle signs of differentiating germ cells. In the absence of His2Av function, 16 out of 37 (43.24%) nos-Gal4; UAS-Upd larval testes had a few germ cell cysts containing branched fusomes (Fig. 5I″″). In the same experiment, only 1 out of 37 (2.7%) testes from nos-Gal4/UAS-Upd; His2Av/+ larvae exhibited branched fusomes. Thus, in the absence of His2Av function, a small population of His2Av mutant germ cells appears to initiate the differentiation program even under conditions of high JAK-STAT activation.
Clonal analysis suggested that the chromatin remodeling factor Domino, the homolog of yeast Swr1 [21], which exchanges His2A variant for His2A in yeast [22], [23], is required for stem cell maintenance. In the Drosophila testis, when negatively marked clones of domk08108 were generated using the FLP/FRT system, the percentage of testes carrying marked domk08108 homozygous GSCs or CySCs was indistinguishable from the control at day 2 PCI (Fig. 6A, B). However, the percentage of testes carrying marked domk08108 homozygous GSCs steadily decreased over time after clonal induction and dropped to zero by day 8 (Fig. 6A). Similarly, the percentage of testes with domk08108 mutant CySCs dropped from 74% at day 2, to 14.5% at day 4, to 0% at day 15 (Fig. 6B). Immunostaining analysis revealed that Domino function might be required for the localization of His2Av to chromatin in GSCs, similar to the function of the corresponding Swr1 complex in yeast. At day 6 PCI, nuclei in GSCs lacking Domino function had reduced levels of His2Av protein compared to control GSCs (Fig. 6C–C′″). Quantification of His2Av immunofluorescence intensity revealed that the loss of domino function reduced His2Av protein levels in GSCs by an average of 2-fold. The average ratio of His2Av immunostaining per unit area in GSCs that were homozygous for domk08108 compared to His2Av immunostaining per unit area in GSCs heterozygous for domk08108 within a testis (n = 28 testes) was 0.56. In contrast, in FRT 42D control (n = 18 testes), the ratio of His2Av immunostaining per unit area in GFP negative to GFP positive GSCs was 1.12 (Fig. 6D). Consistent with a role for Domino in His2Av incorporation and function in GSC maintenance, the loss of His2Av810 mutant GSC clones increased in a domk08108/+ genetic background (Fig. 6E). At day 2 PCI, 65.5% of testes contained His2Av810 homozygous mutant GSC clones, while under the same conditions, only 51.1% of testes from sibling males carrying the dom allele had marked GSC clones, possibly due to reduced incorporation of His2Av into chromatin before clonal induction. Under the same conditions at day 2 PCI, an average of 92% of testes from both domk08108/+; His2Av810 and sibling His2Av810 males had spermatocyte clones, suggesting that clonal induction occurred at the same rate in both genetic backgrounds (data not shown). The percentage of testes with marked His2Av mutant GSCs was also lower at days 3 and 5 PCI in males with the domk08108/+ allele compared to sibling males without the dom allele.
In contrast to the function of Domino, the chromatin remodeling factor ISWI, which also functions in GSC and CySC maintenance [24], did not appear to be required for the localization of His2Av to chromatin in GSCs. Immunostaining for His2Av protein in testes containing ISWI2 homozygous mutant GSCs 6 days PCI revealed that the levels of His2Av protein in ISWI mutant GSCs were comparable to neighboring ISWI2/+ GSCs (Fig. 6F). The average ratio of His2Av immunostaining intensity per unit area for GSCs homozygous mutant for ISWI2 to neighboring ISWI2/+ GSCs within the same testis (n = 24 testes) was 0.96 (Fig. 6D). Similarly, ISWI protein levels in the nuclei of His2Av810 mutant GSCs were comparable to that in heterozygous GSCs (Fig. 6G), suggesting that His2Av might not be required to recruit or maintain ISWI on chromatin. ISWI did not exhibit a strong genetic interaction with His2Av to maintain GSCs in the adult testes. The percentage of testes with marked His2Av810 mutant GSC clones in ISIW2/+; His2Av810 testes was comparable to that in testes lacking the ISWI allele at days 2, 3, and 8 PCI, only falling slightly at day 5 PCI (Fig. 6H).
Loss of His2Av function did not globally alter levels of the epigenetic marks associated with transcriptional state in GSCs. Immunostaining with antibodies that recognize H3K4 tri-methyl (H3K4me3) (Fig. 7A), mostly associated with transcriptionally active/poised chromatin regions, and H3K27 tri-methyl (H3K27me3) (Fig. 7B), mostly associated with transcriptionally inactive regions of chromatin [1] on testes with His2Av810 mutant GSCs 5 days PCI revealed that the levels of these epigenetic marks were comparable in mutant and heterozygous GSCs. Likewise, the protein levels of Scrawny (Scny), a histone H2B deubiquitinase required to prevent premature expression of differentiation genes in adult stem cells [25], were also not altered in His2Av mutant GSCs (data not shown). Furthermore, His2Av protein levels scored 6 days PCI were unaltered in GSCs homozygous mutant for scny02331 (Fig. 7C) or scnye00340 (data not shown) compared to neighboring scny/+ heterozygous GSCs. Although scny mutant follicle cells in the Drosophila ovary exhibit elevated levels of H3K4me3 [25], male GSCs homozygous mutant for either scny02331 (Fig. 7D) or scnye00340 (data not shown) did not exhibit changes in H3K4me3 levels compared to neighboring heterozygous GSCs. These data suggest that loss of His2Av or Scny function was not associated with dramatic changes in transcription in the testis, at least when assayed at a global level by immunostaining for histone marks.
Our results reveal that the histone variant His2Av is required cell autonomously for maintenance of two different adult stem cell types, GSCs and CySCs, in the Drosophila male gonad, but not for the differentiation of the progeny in these two stem cell lineages. The specific requirement for His2Av for adult stem cell maintenance suggests that His2Av may play critical role(s) in the mechanisms that maintain the ability of adult stem cells to self-renew rather than differentiate. His2Av function has been implicated in both transcriptional repression and transcriptional activation. His2Av could maintain adult stem cells by either favoring repression of pro-differentiation genes and/or activation of genes necessary for stem cell identity and function. In yeast, H2A.Z occupies transcriptionally inactive genes and intergenic regions [26], while in Drosophila, His2Av is required for the establishment of heterochromatin and transcriptional repression [27]. Conversely, studies indicate that in Drosophila, yeast, and chicken, His2Av is enriched at nucleosomes downstream of the transcription start site of active or poised genes [28], [29], [30]. Nucleosomes and histone dimers containing H2A.Z appear to be less stable than nucleosomes containing the canonical histone H2A [31], [32], [33]. This lower stability may favor a more open chromatin, giving transcriptional activators or repressors better access to the DNA. Consistent with this model, a recent study showed that H2A.Z promotes self-renewal and pluripotency of murine embryonic stem cells (ESCs) by facilitating the binding of Oct4 to its target genes and the Polycomb repressive complex 2 to differentiation genes [34]. However, in ESCs, unlike in Drosophila male GSCs and CySCs, His2A.Z function was also required for the expression of differentiation genes when ESCs were grown under conditions that induce differentiation [34], [35].
We propose that the requirement of His2Av for adult stem cell maintenance, but not for differentiation, may reflect a subtle role for His2Av in maintaining expression of genes required for self-renewal versus differentiation. Adult stem cells lie at the cusp of two alternate fate choices, self-renewal and differentiation; the progeny of stem cell division are maintained in a state where they can execute either self-renewal or differentiation programs depending on local cues. The requirement for this balanced, bi-potential state may make adult stem cells more sensitive to the small alterations in the relative levels of key transcripts associated with the loss of His2Av function, tilting the balance from stem cell maintenance to onset of differentiation. Consistent with the model that His2Av may alter transcriptional levels subtly, H2A.Z was shown to be required to fine-tune the transcriptional state of hsp70 and a wide variety of other genes in response to temperature changes in Arabidopsis [36], [37].
The ATP-dependent chromatin-remodeling factor Domino is required for GSC and CySC maintenance in the male germline, as previously shown for somatic follicle stem cells in the female gonad [38]. The yeast Swr1 complex containing the homolog of Drosophila Domino exchanges His2A with Htz1 (the yeast His2A variant) [22], [23], [39] and in Drosophila, Domino- containing Tip60 chromatin remodeling complex has been shown to exchange phospho-His2Av with unmodified His2Av in in vitro assays [40]. Our studies indicate that Domino function is required in vivo in GSCs for the incorporation of His2Av into chromatin. Nuclei of domino mutant GSCs had lowered but still detectable levels of His2Av protein, possibly due to the weak domino allele used in this study. Alternatively, incorporation of His2Av in some regions of the chromatin may occur independently of Domino function, as has been reported in yeast, in which stress-responsive genes exhibit Swr1-independent incorporation of Htz in the coding region [41]. Although ISWI, like His2Av, is required for GSC and CySC maintenance in the male germline [24], these proteins may function in parallel pathways to maintain adult stem cells in the testis. The ISWI containing nucleosome-remodeling factor (NURF) was shown to maintain GSCs and CySCs in the Drosophila testis by positively regulating the JAK-STAT signaling pathway; GSCs mutant for components of the NURF complex exhibited low levels of STAT92E protein [24]. In contrast, as discussed below, His2Av may function independently of the JAK-STAT signaling pathway.
Our results indicate that His2Av may function independently of the JAK-STAT signaling pathway to provide a chromatin environment that allows for stem cell maintenance. Expression of the His2Av and STAT92E proteins in GSCs was not dependent on each other. Our studies indicate that His2Av may not be required for expression of at least one other key STAT-dependent gene in CySCs. Activation of the JAK-STAT signaling pathway in response to the Upd signal from the hub is important for CySC maintenance, possibly in part through STAT-dependent transcription of Zfh-1 [17]. However, CySCs lacking His2Av function still expressed Zfh-1. In GSCs, activation of the JAK-STAT pathway is important for maintaining hub-GSC adhesion and for centrosome orientation [19], both of which appeared unaffected in His2Av mutant GSCs. Loss of His2Av function did not strongly suppress the phenotype associated with ectopic overexpression of Upd in the testis, although a few His2Av mutant germ cells were able to initiate differentiation, possibly due to relatively lower levels of JAK-STAT activation in these cells. Even though loss of His2Av normally resulted in differentiation of GSCs and CySCs, the requirement for His2Av function can be overridden by high levels of activation of the JAK-STAT pathway, possibly maintaining somatic CySCs in a stem cell like state, which may fail to provide a microenvironment for germ cells to initiate differentiation [19], [42].
Fly stocks were raised on cornmeal/molasses medium at 25°C unless stated otherwise. Stocks are from the Bloomington Stock Center unless specified otherwise. Mutant alleles used in this study include 1) w;FRT82B, His2Av810/TM6B, Tb, carrying a 311 base pair deletion that removes the second exon of the His2Av gene [43], 2) w;FRT82B, His2Av05146/TM3, 3)w;; Df(3R)BSC524/TM6b,Tb, a deletion that encompasses the His2Av gene, 4) the Stat92E alleles, FRT82B, Stat92E06346/TM3 and FRT82B, Stat92EJ6C8/TM3 (gift from E. Matunis), 5) y, w, ey-Flp, GMR-lacZ; FRT42D, domk08108/CyO, y+, a loss of function allele (also known as dom1) with a P-element inserted at the 3′ boundary of the first exon [21] obtained from DGRC, 7) y w;FRT 42D, ISWI2, sp/SM5, Cy, sp , a null allele carrying a nonsense mutation [44], 8) the scrawny alleles: FRT 80B, scnyl(3)02331 and FRT 80B, scnye00340 [25]. w; His2Av-mRFP; FRT82B, His2Av810/TM6B, Tb flies were used to rescue GSC and CySC loss. The His2Av-mRFP construct rescues the lethality of His2Av05146 mutant [15]. The following flies 1) hs-FLP122;;FRT82B, ubi-nGFP, 2) hs-FLP122;nos-GAL4;FRT82B, tub-LacZ (gift from D. Kalderon), 3) hs-FLP122;;FRT80B, ubi-nGFP, 4) hs-FLP122;FRT42D, ubi-nGFP were used to induce marked clones in the testes. y,w;ey-GAL4, UAS-FLP;FRT82B, GMR-hid/TM2 flies were used to induce marked clones in adult eyes. FRT82B, FRT80B and FRT42D were used as wild-type controls for clone induction. Other stocks used include w,sa-GFP [45], UAS-Upd [46], UAS-DEFL #6-1 (UAS-E-Cad-GFP) [47] from DGRC, nanos-GAL4, UAS-Dicer2;; nanos-GAL4VP16 (NG4VP16) and ;; Bam-GAL4. RNAi flies against His2Av (Transformant ID #110598) were obtained from the Vienna Drosophila RNAi Center.
A heteroallelic combination of His2Av810 and Df(3R) BSC524 survives until the third instar larval stage when grown at 25°C for 2 days and then shifted to 29°C. The effects of loss of His2Av function in testes ectopically expressing Upd ligand was analysed in the third-instar larval progeny of nanos-GAL4; Df(3R) BSC524/TM6B,Tb and UAS-Upd; His2Av810/TM6B,Tb. Tb-positive larvae (heterozygous for either His2Av810 or Df(3R) BSC524) expressing UAS-Upd under the nanos-Gal4 driver were used as controls
Testes were dissected in 1× phosphate-buffered saline (PBS) and fixed in 4% formaldehyde in PBS for 20 minutes at room temperature, washed twice for 30 minutes each in PBS with 0.3% Triton X-100 and 0.6% sodium deoxycholate. Testes were incubated overnight at 4°C in primary antibodies against Armadillo (Arm, mouse 1∶10; Developmental Studies Hybridoma Bank (DSHB)) [48], Fas3 (mouse 1∶10; DSHB) [49], α-spectrin (mouse 1∶10; DSHB) [50], Eyes absent (Eya, mouse 1∶10; DSHB) [51], E-cadherin (mouse 1∶10, DSHB) [52], Green Fluorescent protein (GFP, rabbit 1∶400–1∶1000; Invitrogen and Sheep 1∶1000, Abd-Serotec), β-Galactosidase (rabbit 1∶1000; Cappel), Histone H3 lysine 4 trimethyl (H3K4me3, rabbit 1∶200: Cell Signaling), Histone H3 lysine 27 trimethyl (H3K27me3, rabbit 1∶200: Cell Signaling), His2Av (rabbit 1∶1000; gift from R. Glaser) [53], Traffic-jam (Tj, guinea pig 1∶5000; gift from D. Godt) [54], Vasa (goat 1∶50; Santa Cruz Biotechnology), ©-tubulin (mouse 1∶50; Sigma), Zfh-1 (rabbit 1∶5000; gift from R. Lehman), STAT92E (rabbit 1∶1000; gift from E.Bach) [55], Scrawny (guinea pig 1∶200; gift from M. Buszczak) [25] and ISWI (rabbit 1∶100; gift from J.Tamkun) [56]. Secondary antibodies used were from the Alexa Fluor-conjugated series (1∶500; Molecular Probes). Samples were mounted in VECTASHIELD medium containing DAPI to visualize DNA (Vector Labs H-1200). Immunofluorescence images were obtained with a Leica SP2 Confocal Laser Scanning microscope. Phase and clonal analysis images were obtained using a Zeiss Axioskop microscope and SPOT RT3 camera by Diagnostic Instruments or CoolSNAPez camera by Photometrics. Images were processed using Adobe CS4 Photoshop and Illustrator. Comparison of intensity of His2Av staining in GSCs was performed using the ImageJ program [57]. The nuclear area in GSCs was selected based on the DAPI staining and the average intensity of His2Av immunostaining within the nucleus was measured using ImageJ. An average of immunofluorescence intensity per unit area for all GSCs homozygous (identified as GFP negative) or heterozygous (identified as GFP positive) for a given genotype was calculated for each testis. The relative level of His2Av protein was calculated as a ratio of the average immunofluorescence intensity per unit area for homozygous GSC to heterozygous GSC within each testis. Similar results were obtained when anti-His2Av intensity was normalized to the intensity for DAPI staining for each GSC.
Homozygous His2Av mutant clones in a heterozygous background were generated by crossing either 1) hs-FLP122;;FRT82B, ubi-nGFP, 2) hs-FLP122; FRT42D, domk08108/CyO;FRT82B, ubi-nGFP, 3) hs-FLP122; FRT 42D, ISWI2, sp/CyO;FRT82B, ubi-nGFP, or 4) hs-FLP122;nos-GAL4;FRT82B, tub-LacZ virgin females to w;;FRT82B, w;;FRT82B, His2Av810/TM6B, Tb or w;; UAS-DEFL #6-1, FRT82B, His2Av810/TM6B, Tb [The UAS-DEFL #6-1 (UAS-E-Cad-GFP) containing chromosome was recombined to the FRT82B, His2Av810 chromosome] males. Homozygous domk08108 or ISWI2 mutant clones were obtained by crossing males of the alleles to hs-FLP122;FRT42D, ubi-nGFP virgin females, while males of scny alleles were crossed to hs-FLP122;;FRT80B, ubi-nGFP males. The progeny were raised at 25°C and heat-shocked at 37°C for two hours each on two consecutive days at the pupal stage. GSCs homozygous mutant for His2Av810 or other alleles were identified by their lack of GFP (or β-Galactosidase), presence of the germ cell marker Vasa, and contact with the hub. Homozygous clones of CySCs generated by heat shock induced mitotic recombination were identified by their lack of GFP (or β-Galactosidase) and the germ cell marker Vasa, by the presence of Tj, a marker of the cyst cell lineage, and by their proximity to the hub.
Homozygous mutant germline clones generated in His2Av05146/+ resulted in the loss of mutant GSCs (Fig. S3A) and spermatocytes (Fig. S3B) over time after clone induction. However, this loss of marked cells was not associated with a loss of anti-His2Av staining (Fig. S3C′), and the loss of homozygous mutant germ cells was not rescued by the presence of His2Av-mRFP transgene (Fig. S3B), suggesting that a mutation other than His2Av on the chromosome might be responsible for GSC loss in this line.
FLP-medicated mitotic recombination was induced in eye precursor cells by crossing y,w;ey-GAL4, UAS-FLP;FRT82B, GMR-hid/TM2 virgins to males carrying FRT 82B, His2Av810 (or FRT control). Eye precursor cells carrying one copy of the dominant cell lethal transgene GMR-hid fail to survive, thereby generating eyes composed entirely of cells homozygous for His2Av810 (or the FRT control).
RNAi knockdown experiments were carried out by crossing flies carrying His2Av RNAi hairpin under the UAS regulatory sequence to either UAS-Dicer2;;NG4VP16 females or Bam-GAL4. The progeny were raised at 18°C until eclosion and transferred to and held at 30°C.
|
10.1371/journal.pgen.1006136 | Srs2 and Mus81–Mms4 Prevent Accumulation of Toxic Inter-Homolog Recombination Intermediates | Homologous recombination is an evolutionally conserved mechanism that promotes genome stability through the faithful repair of double-strand breaks and single-strand gaps in DNA, and the recovery of stalled or collapsed replication forks. Saccharomyces cerevisiae ATP-dependent DNA helicase Srs2 (a member of the highly conserved UvrD family of helicases) has multiple roles in regulating homologous recombination. A mutation (srs2K41A) resulting in a helicase-dead mutant of Srs2 was found to be lethal in diploid, but not in haploid, cells. In diploid cells, Srs2K41A caused the accumulation of inter-homolog joint molecule intermediates, increased the levels of spontaneous Rad52 foci, and induced gross chromosomal rearrangements. Srs2K41A lethality and accumulation of joint molecules were suppressed by inactivating Rad51 or deleting the Rad51-interaction domain of Srs2, whereas phosphorylation and sumoylation of Srs2 and its interaction with sumoylated proliferating cell nuclear antigen (PCNA) were not required for lethality. The structure-specific complex of crossover junction endonucleases Mus81 and Mms4 was also required for viability of diploid, but not haploid, SRS2 deletion mutants (srs2Δ), and diploid srs2Δ mus81Δ mutants accumulated joint molecule intermediates. Our data suggest that Srs2 and Mus81–Mms4 have critical roles in preventing the formation of (or in resolving) toxic inter-homolog joint molecules, which could otherwise interfere with chromosome segregation and lead to genetic instability.
| Homologous recombination (HR) is a DNA-repair mechanism that is generally considered error free because it uses an intact sister chromatid as a template. However, in diploid cells, HR can also occur between homologous chromosomes, which can lead to genomic instability through loss of heterozygosity. This alteration is often detected in genetic disorders and cancer, suggesting that tight control of this process is required to ensure genome stability. Yeast Srs2, conserved from bacteria to humans, plays multiple roles in the regulation of HR. We show here that a helicase-dead mutant of Srs2, srs2K41A, is lethal in diploid cells but not in haploid cells. Expression of Srs2K41A in diploid cells causes inter-homolog joint molecule intermediates to accumulate, and leads to gross chromosomal rearrangements. Moreover, srs2Δ mus81Δ double mutants have a severe diploid-specific growth defect with accumulation of inter-homolog joint molecules. These data demonstrate that Srs2 and Mus81-Mms4 participate in essential pathways preventing accumulation of inter-homolog recombination intermediates, thereby reducing the risk of genome instability.
| Genomes are constantly challenged by endogenous metabolic products or exogenous physical or chemical agents that can generate DNA lesions. When they go unrepaired, these DNA lesions cause stalled replication forks and/or replication-fork collapse, leading to the accumulation of single-stranded DNA (ssDNA) gaps or DNA double-strand breaks (DSBs). Homologous recombination (HR) is a highly conserved DNA-repair mechanism that is essential for the faithful repair of DSBs and has an important role in the repair of post-replicative ssDNA gaps [1–3]. Therefore, dysregulated or incomplete repair by HR can lead to genomic instability, which is a hallmark of cancer.
Rad51 is a central factor in DSB repair by HR. Rad51 forms nucleoprotein filaments on ssDNA tracts generated by 5’ to 3’ ssDNA resection from DSBs. Rad51 filaments mediate strand invasion into homologous DNA duplexes, leading to the formation of D-loops [4,5]. HR intermediates, including D-loops, can enter one of two HR sub-pathways: the synthesis-dependent strand-annealing (SDSA) pathway, which generates non-crossover products, and the canonical DSB repair (DSBR) pathway, which generates crossover or non-crossover products [6,7]. In the SDSA pathway, a newly synthesized ssDNA strand is displaced from the D-loop to anneal to the complementary strand in the original duplex, resulting in a non-crossover outcome with no change to the template DNA [1]. The DSBR pathway involves D-loop extension and annealing of the displaced strand to a second ssDNA tail of the broken duplex, forming a DNA intermediate termed the double Holliday junction. In Saccharomyces cerevisiae, several helicases function in crossover control. Srs2 and Mph1 act independently to promote SDSA by processing the HR intermediates downstream of D-loop formation [8–11]. Sgs1, together with Top3 and Rmi1, can dissociate double Holliday junctions to generate non-crossover products, thus preventing crossovers in the DSBR pathway [8,12–14]. Alternatively, double Holliday junctions can be resolved to produce crossover or non-crossover products by structure-specific endonucleases, such as the Mus81–Mms4 complex, the Slx1–Slx4 complex, and Yen1 [15–17].
Srs2 is a member of the highly conserved UvrD family of helicases that have 3’ to 5’ helicase activity [18,19]. A mutant allele of SRS2 was first isolated as a suppressor of the radiation sensitivity associated with rad6 and rad18 mutants, which are defective in post-replication repair [20–22]. In addition, mutants of SRS2 have a hyper-recombination phenotype characterized by crossover events [8,23,24]. Srs2 interacts with a sumoylated form of the DNA replication clamp, proliferating cell nuclear antigen (PCNA), which recruits Srs2 to DNA replication forks, preventing HR [25,26]. Thus, Srs2 is an anti-recombinase that prevents inappropriate HR at the replication fork and preferentially facilitates post-replication repair. These data are consistent with the fact that Srs2 disassembles the Rad51 filament and unwinds synthetic D-loop structures in vitro [27–29]. In addition to its role as an anti-recombinase, Srs2 binding to sumoylated PCNA blocks synthesis-dependent elongation of the invading strand within a D-loop structure at a stalled replication fork, limiting the frequency of crossover events [29]. Moreover, Srs2 promotes the SDSA pathway during mitotic DSB repair by removing the Rad51 filament from the second end of the DSB, and/or by facilitating the dissociation of the invading strand from the D-loop [30–32]. Phosphorylation of Srs2 by cyclin-dependent kinase 1 (Cdk1) stimulates the SDSA pathway [33]. Taken together, these observations suggest that Srs2 has two distinct functions in HR; it prevents unscheduled recombination by inhibiting Rad51-dependent formation of joint molecules and it promotes efficient DSB repair by the SDSA pathway.
During HR in diploid cells, sister chromatids are the preferred templates for HR-mediated repair (inter-sister HR), but homologous chromosomes can also be used to restore the broken DNA (inter-homolog HR), although much less efficiently. Because sister chromatids are identical, inter-sister HR is genetically silent. By contrast, the use of homologous chromosomes as repair templates has important consequences for genetic stability, and loss of heterozygosity is a frequent outcome [34]. The frequency of loss of heterozygosity is high in cancerous and aged cells, which has raised interest in dissecting the mechanisms of HR [35]. The HR process has to be tightly controlled to protect against genetic instability, but little is known about the relative contributions of each HR pathway to the processing of the two classes of recombination intermediate, involving either sister chromatids or homologs.
Our experiments were designed to explore the role of Srs2 in haploid and diploid cells by phenotypic characterization of a number of srs2 mutants as a function of cell ploidy. The Srs2 helicase-deficient mutant (srs2K41A) caused diploid-specific lethality. This lethality was suppressed by deletion of RAD51, but was independent of the phosphorylation and sumoylation of Srs2 and of its interaction with sumoylated PCNA. Expression of Srs2K41A in diploid cells led to a specific increase in G2/M-arrested cells, more abundant inter-homolog joint molecules and increased gross chromosomal rearrangements, such as chromosome loss and translocations. srs2Δ mus81Δ double mutants also demonstrated a severe, diploid-specific growth defect, with the concomitant accumulation of joint molecules. These results suggest that the mechanisms of processing inter-sister and inter-homolog joint molecules differ significantly. We propose that Srs2 and Mus81–Mms4 have critical roles in processing inter-homolog joint molecules, which could otherwise interfere with chromosome segregation and lead to genetic instability.
A previous study showed that srs2Δ diploid cells are more sensitive to methyl methanesulfonate (MMS) than srs2Δ haploid cells [21,36]. This ploidy-specific sensitivity to MMS is thought to reflect lethal outcomes of inter-homolog HR events in the absence of wild-type Srs2. To understand the role of Srs2 in inter-homolog HR, we constructed four mutants of srs2: srs2K41A lacks helicase activity [37], srs27AV cannot undergo Cdk1-dependent phosphorylation [38,39], srs23KR cannot undergo sumoylation [40], and srs2ΔSIM lacks the protein motif that mediates interaction with sumoylated PCNA [26]. These srs2 mutants and wild-type SRS2 were expressed in yeast from low-copy centromeric (pRS415_LEU2) plasmids under control of the SRS2 promoter. The plasmids were introduced into srs2Δ haploid or diploid cells and selected on SC+Glucose medium lacking leucine (SC+Glu-Leu). In this initial screen, no diploid colonies expressing Srs2K41A were detected (Table 1), suggesting that srs2K41A could be lethal or could block growth of srs2Δ diploid cells. To test this possibility, an srs2K41A allele was integrated at the SRS2 genomic locus of haploid yeast. The integrating cassette included downstream HIS3 or LEU2 selectable markers (srs2K41A_HIS3 or srs2K41A_LEU2). The endogenous SRS2 allele in a haploid strain was also linked to HIS3 or LEU2 selectable markers as a control (SRS2_HIS3 or SRS2_LEU2). A MATα strain carrying srs2K41A_LEU2 was crossed to MATa strains bearing srs2K41A_HIS3, SRS2_HIS3, or srs2Δ::HIS3. Diploids from these crosses were selected for growth on SC+Glu medium lacking histidine and leucine. As shown in Fig 1A, the srs2K41A/srs2Δ heterozygotes and srs2K41A/srs2K41A homozygotes did not grow on the selection medium, whereas heterozygous srs2K41A/SRS2 diploids exhibited normal growth. This demonstrates that srs2K41A mutants are lethal in diploids.
To investigate why srs2K41A is lethal in diploid cells, Srs2K41A and wild-type Srs2 were expressed under the control of the inducible GAL1 promoter from a single-copy integrated allele at the chromosomal AUR1 locus of srs2Δ diploid and haploid cells (Fig 1B). Hereafter, these strains are referred to as GAL-srs2K41A and GAL-SRS2, respectively. A GAL-empty strain (essentially the same as an srs2Δ strain) was constructed in a similar manner, as an additional control. The resultant haploid and diploid strains grew normally in 2% glucose-containing medium (YPD) (Fig 1C and S1A Fig), enabling the effect of conditional expression of Srs2K41A and Srs2 to be investigated.
To determine the level of expression of Srs2 in this experimental system, GAL-SRS2 diploid cells were grown for 6 h in the presence of 2% raffinose medium (YPR) and various concentrations of galactose, and whole-cell extracts were prepared and analyzed by immunoblotting with an antibody to Srs2. The results revealed that Srs2 protein was absent in cells grown in YPD or YPR, and that the abundance of Srs2 increased with increasing galactose concentration (S1B Fig). Control experiments established that GAL-SRS2 diploid cells grew normally in the presence of 0.02% galactose, but poorly in the presence of 0.2% galactose, because of high overexpression of Srs2 (Fig 1C and S1C Fig), as previously reported [36]. In addition, expression of Srs2K41A, but not wild-type Srs2, inhibited growth (despite the presence of the chromosomal SRS2+ allele) when moderately expressed in the presence of 0.05% galactose, whereas similar growth defects were not observed in the presence of 0.02% galactose (S1D Fig). Thus, srs2K41A is essentially a dominant-negative allele, and its dominancy is dependent on the ratio of wild-type Srs2 to Srs2K41A. We conclude that expression of Srs2 from the GAL1 promoter in the presence of 0.02% galactose generates a physiologically-relevant protein level, and, for the remainder of this study, cells carrying GAL1 promoter-driven expression strains were grown in YPD or YPR to repress Srs2 expression, and in YPR medium containing 0.02% galactose to induce Srs2.
To examine whether GAL-srs2K41A diploid cells could recover from growth arrest in galactose-containing medium, cells transiently grown in the presence of 0.02% galactose were transferred back to glucose-containing medium to determine the plating efficiency. The plating efficiency of GAL-srs2K41A diploids decreased rapidly with >3 h incubation in the presence of galactose, whereas no significant effect on growth was observed for GAL-srs2K41A haploid cells, or GAL-empty and GAL-SRS2 haploid or diploid cells, even after incubation for 8 h in 0.02% galactose (Fig 1D). These data show that a physiological level of Srs2K41A reduces viability of diploid cells, but not haploid cells.
In the course of these studies, Srs2K41A isolated from haploid and diploid cells was observed as multiple slow-migrating protein species on SDS-PAGE when cells were grown in the presence of 0.02% galactose (Fig 1E). Because Srs2 is phosphorylated and sumoylated in response to DNA damage [33,38,39], we postulated that the slower-migrating forms of Srs2K41A protein are phosphorylated and/or sumoylated isoforms of the protein. To test this hypothesis, plasmids that expressed Srs2K41A, Srs2K41A,7AV, Srs2K41A,3KR, and Srs2K41A,7AV,3KR from the GAL1 promoter were introduced into srs2Δ diploid cells. Each strain was grown to early logarithmic phase in glucose medium and transferred to galactose medium, and protein extracts were prepared and analyzed by western blot with an antibody to Srs2. This analysis revealed that Srs2K41A,7AV, which lacked phosphorylation sites, existed as three sumoylated isoforms that moved slightly faster than modified isoforms of Srs2K41A on electrophoresis (Fig 1F). Srs2K41A,3KR, which lacked sumoylation sites, existed as phosphorylated isoforms (Fig 1F). As expected, srs2K41A,7AV,3KR, in which all phosphorylation and sumoylation sites had been mutated, resulted in a considerable reduction in expression of modified isoforms of Srs2 (Fig 1F). These results indicate that Srs2K41A can be sumoylated and phosphorylated in the absence of DNA damage. To determine whether these modifications of Srs2K41A affected diploid-specific lethality, yeast CEN/ARS plasmids (in which srs2K41A, srs2K41A,7AV, srs2K41A,3KR, and srs2K41A,7AV,3KR were under the control of the endogenous SRS2 promoter) were constructed and transformed into the srs2Δ diploid strain. The result showed that no srs2Δ transformants expressing Srs2K41A or its derivatives were viable (no colonies were detected) (Table 1), indicating that neither phosphorylation nor sumoylation is required for the lethal effects of Srs2K41A in diploid yeast.
To learn more about how srs2K41A kills diploid yeast cells, the cell-cycle distribution and cell morphology of GAL-srs2K41A cells were investigated in haploid and diploid cells. Cells were grown to early logarithmic phase in the presence of glucose, transferred to YPR containing 0.02% galactose, and then analyzed by flow cytometry. In GAL-empty and GAL-SRS2 haploids and diploids, cell-cycle progression was not significantly altered by galactose induction (Fig 2A). However, GAL-srs2K41A diploids, but not haploids, showed apparent cell-cycle arrest at G2/M after induction of Srs2K41A. The 4C peak appeared to broaden with prolonged incubation of cells in the presence of 0.02% galactose (Fig 2A). Similar effects have been observed after extended treatment with nocodazole, a microtubule-depolymerizing drug that causes G2/M arrest [41]. Consistent with this interpretation, approximately 80% of GAL-srs2K41A diploids assumed the characteristic morphology of G2/M arrest, which involves large-budded cells with one nucleus at the bud neck and a short spindle (Fig 2B and 2C and S2 Fig). These results suggest that, in diploids, Srs2K41A causes cell-cycle arrest after bulk DNA synthesis is complete.
The checkpoint protein kinase Rad53 is phosphorylated and activated in response to DNA damage and replication stress. As shown in Fig 2D, phosphorylated Rad53 was detected in galactose-induced GAL-srs2K41A diploid and haploid cells, but not in GAL-SRS2 cells. Previous studies showed that the protein product of srs2ΔSIM, which cannot interact with sumoylated PCNA, undergoes dramatically less sumoylation in vivo [40], and srs2ΔSIM mutation suppresses the replication defects associated with overexpression of Srs2 in haploid cells [42]. In our study, the phenotypes of GAL-srs2K41A,ΔSIM diploid and haploid cells were examined. Rad53 phosphorylation and Srs2 sumoylation (and phosphorylation) were significantly reduced at 6 h after GAL-srs2K41A,ΔSIM haploid cells were transferred to 0.02% galactose, compared with levels in GAL-srs2K41A haploid cells (Fig 2E). By contrast, substantial Rad53 phosphorylation was still observed in GAL-srs2K41A,ΔSIM diploid cells, although Srs2 phosphorylation and sumoylation were strongly reduced compared with levels in GAL-srs2K41A diploid cells (Fig 2E). In addition, GAL-srs2K41A,ΔSIM diploids, but not haploids, had severe growth defects (Fig 2F). These results indicate that the srs2K41A lethality in diploid cells is not associated with activation of the DNA damage checkpoint through its interaction with sumoylated PCNA.
A well-characterized role of Srs2 is that of anti-recombinase, and in this context Srs2 dismantles Rad51 nucleofilaments on ssDNA [27,28]. Toxic HR intermediates might, therefore, accumulate in srs2K41A diploid cells, which could explain the ploidy-specific lethality of this allele. Consistent with this hypothesis, rad51Δ srs2Δ diploid strains expressing Srs2K41A from a plasmid vector were viable (Fig 3A). Similarly, the growth inhibition of GAL-srs2K41A diploids in the presence of 0.02% galactose was suppressed by the rad51Δ mutation (Fig 3B). Moreover, srs2Δ diploid cells expressing Srs2K41A,Δ783–998, which lacks the Rad51 interaction domain in Srs2 [28], were also viable (Fig 3A). Taken together, these results indicate that the lethality of srs2K41A in diploids is associated with Rad51-dependent HR in diploids.
Rad52 nuclear focus formation is an indication of HR in vivo, and many mutants with genome-maintenance defects have increased numbers of Rad52 foci compared with their wild-type counterparts [43]. The frequency of spontaneous Rad52 foci was, therefore, quantified in GAL-srs2K41A cells and appropriate control cells expressing GFP-tagged Rad52 from the endogenous RAD52 genomic locus. Few Rad52-GFP foci were observed when cells were grown in glucose-containing medium (Fig 3C and 3D). However, after 8 h incubation in 0.02% galactose, Rad52-GFP foci were markedly increased in GAL-srs2K41A diploids compared with GAL-SRS2 diploid and GAL-srs2K41A haploid cells, and most of the foci occurred in large-budded cells with a single nucleus (Fig 3C and 3D). These findings suggest that GAL-srs2K41A diploid cells accumulate HR intermediates at a much higher frequency than GAL-srs2K41A haploid cells.
To test directly whether Srs2K41A caused joint molecules to accumulate in srs2Δ diploids, diploid cells were incubated for 8 h in YPR medium with or without 0.02% galactose, harvested and used to obtain chromosomal DNA, which was analyzed by pulsed-field gel electrophoresis (PFGE). In GAL-srs2K41A diploid cells, the DNA signal corresponding to chromosomes that entered the gel decreased after induction in galactose-containing medium, and most of the DNA failed to migrate out of the well of the gel (Fig 4A). The non-migratory DNA appeared by 4 h after induction in galactose-containing medium (S3A Fig). By contrast, non-migratory DNA was not observed when DNA from galactose-induced GAL-SRS2 and GAL-empty diploid cells or GAL-srs2K41A haploid cells was analyzed by PFGE (Fig 4A). Moreover, accumulation of non-migratory DNA in GAL-srs2K41A diploid cells was suppressed by rad51Δ (Fig 4A). These results suggest that Rad51 and Srs2K41A collaborate in diploid cells to generate DNA structures that are not able to migrate out of the well during PFGE. In this context, it should be noted that the rad51Δ mutation did not suppress Rad53 activation in GAL-srs2K41A and GAL-srs2K41A,ΔSIM diploid cells under the same conditions (S3B Fig), suggesting that joint molecules per se are not direct signals for Rad53 activation.
To characterize the chromosomal structures that accumulated in GAL-srs2K41A diploid cells, chromosomal DNA samples were digested with the rare-cutter restriction endonuclease NotI prior to PFGE. Although NotI digests yeast chromosomes into multiple large and small fragments, the intensity of the DNA signal in the wells did not change significantly after digestion with NotI (Fig 4B). This observation suggested that GAL-srs2K41A diploid cells accumulated branched DNA structures, which were enriched even after digestion with NotI. To test this possibility, NotI-digested or non-digested chromosomal DNA samples were digested with purified RuvC from Escherichia coli. RuvC is a highly specific endonuclease that resolves Holliday junctions, although it also cleaves three-stranded junctions and nicked Holliday junctions [44,45]. The results showed that the action of RuvC released NotI-digested chromosomal fragments into the PFGE gel (Fig 4B), whereas non-migratory chromosomal DNA without NotI treatment was hardly resolved by RuvC (S3C Fig). NotI digestion could conceivably facilitate the formation of catalytically competent joint molecule configurations for RuvC cleavage, since junction incision by RuvC is dependent on configuration [46,47]. These results suggest that RuvC-cleavable joint molecules accumulate in GAL-srs2K14A diploid cells.
To further examine whether the DNA structures in GAL-srs2K14A diploid cells were products of inter-homolog HR, an srs2Δ haploid strain that carried an additional copy of chromosome IV (henceforth known as the srs2Δ disome) was constructed. Notably, no colonies were obtained when Srs2K41A was expressed from a plasmid vector in srs2Δ disomes, but the growth defect was rescued by deletion of RAD51 (S4A Fig). These suggest that the additional copy of a donor sequence (homologous chromosome) is a cause of the lethality of srs2Δ disomes expressing Srs2K41A, and that the growth defect of GAL-srs2K14A disomes is the result of Rad51-dependent HR. Similar experiments were performed in an srs2Δ disome in which GAL-srs2K41A was integrated at the chromosomal AUR1 locus (henceforth known as the GAL-srs2K41A disome). The GAL-srs2K41A disome strain failed to grow in the presence of galactose, whereas the haploid control strain grew normally under same conditions (S4B Fig). In PFGE analysis, the GAL-srs2K41A disome strain, but not GAL-empty and GAL-SRS2 disome strains, showed a specific loss of signal corresponding to chromosome IV in galactose-induced cells, whereas no other chromosomes were similarly affected (Fig 4C and 4D and S4C Fig). This conclusion was confirmed by Southern blotting with a probe for chromosome IV, which showed a reduction in hybridization signal in the gel and augmentation of the hybridization signal in the well during Srs2K41A expression (Fig 4C). These results suggest that inter-homolog joint molecules accumulate in GAL-srs2K41A diploid and disome cells.
Our results led to the hypothesis that unresolved joint molecules form in srs2K41A diploid cells, leading to chromosomal instability and cell death. To test this hypothesis, the frequency of loss of a pair of chromosome V homologs marked with CAN1 on the right arm and URA3 on the left arm was calculated and compared in GAL-SRS2 and GAL-srs2K41A diploid cells (Fig 5A, left panel) [48]. The observed frequency of chromosome loss in galactose-induced GAL-srs2K41A diploid cells was 15-fold higher than in galactose-induced GAL-SRS2 diploid cells (Fig 5A, right panel), suggesting that GAL-srs2K41A diploid cells have a defect in chromosome segregation, which leads to a high rate of aneuploidy. Indeed, this result probably underestimated the chromosome-loss frequencies in galactose-induced GAL-srs2K41A diploid cells because it only detected aneuploid cells that remained viable after re-plating on glucose-containing medium. To directly investigate genomic integrity, chromosomal DNA was isolated from surviving cells and analyzed by PFGE. Chromosomal abnormalities were observed in 3% (1 of 29) of galactose-induced GAL-SRS2 and 0% (0 of 29) glucose-repressed GAL-srs2K41A diploid cells (Fig 5B). By contrast, 20 of 68 survivors (29%) obtained from galactose-induced GAL-srs2K41A diploid cells showed abnormal chromosome compositions; both aneuploidy and chromosomal translocations were detected (Fig 5B). Thus, the expression of Srs2K41A in diploids dramatically increases the rates of gross chromosomal rearrangements.
It has been reported that sensitivity to MMS is enhanced in srs2Δ diploid cells relative to their haploid counterparts [21]. To gain insight into inter-homolog HR, a genome-wide screen for diploid-specific sensitivity to MMS was conducted using a library (n ≈ 4,800) of viable haploid and diploid deletion mutants, directly testing for a ploidy-specific phenotype in the presence of MMS. The complete results of the screen will be described elsewhere. Three genes were identified that function in the processing of HR intermediates (SRS2, MUS81, and MMS4). Our investigation focused on a subset of HR genes including SGS1, MPH1, MUS81, MMS4, RAD1, RAD10, YEN1, SLX1, and SLX4, which are involved in the processing of D-loops, Holliday junctions, and similar structures [11,17]. We reconfirmed that mus81Δ and mms4Δ strains were more sensitive to MMS as diploids than as haploids, whereas other HR-deficient diploid strains had similar MMS sensitivity to their haploid counterparts (Fig 6A and S5 Fig). Mus81 interacts with Mms4 to form a structure-specific nuclease, which cleaves a variety of branched structures, including 3' flaps, D-loops, and nicked Holliday junctions [49–51]. These results suggest that Mus81–Mms4 has an important role in the resolution of inter-homolog joint molecules.
The genetic relationship between Srs2 and Mus81–Mms4 was investigated by comparing the growth and viability of haploid and diploid srs2Δ and mus81Δ single-mutant and double-mutant strains. The srs2Δ mus81Δ haploid double mutant grew just as well as either single mutant, whereas srs2Δ mus81Δ diploid cells grew very poorly (Fig 6B). A similar effect was seen in srs2Δ mms4Δ cells (S6A Fig). Poor growth of srs2Δ mus81Δ diploids was rescued by expression of plasmid-borne srs27AV, srs23KR, or srs2ΔSIM, but not by the plasmid vector alone (S6B Fig, upper panel), which indicated that Srs2 rescued the growth defect of the double-mutant strain in the absence of phosphorylation, sumoylation, or interaction with sumoylated PCNA. However, plasmid-borne srs2Δ783–998, which lacks a Rad51-interaction domain, did not complement the severe growth defect of srs2Δ mus81Δ diploid cells (S6B Fig, upper panel), and deletion of RAD51 or RAD52 rescued the growth defect (S6C Fig). These results suggest that Mus81–Mms4 and Srs2 have essential roles in inter-homolog HR.
Biochemical and two-hybrid studies have shown that, in addition to Srs2Δ783–998, Srs2Δ875–902 and Srs2L844A are deficient in Rad51 interaction [52,53]. These results suggest that the amino acid residues of Srs2 that are critical for binding to Rad51 are localized in separate regions within Srs2 residues 783–998. In our study, plasmids were constructed in which srs2L844A, srs2Δ875–902, and srs2L844A,Δ875–902 were under the control of the endogenous SRS2 promoter. Poor growth of srs2Δ mus81Δ diploids was rescued by expression of plasmid-borne SRS2, srs2L844A, srs2Δ875–902, or srs2L844A,Δ875–902, but not srs2Δ783–998 (S6B Fig, lower panel). Similarly, srs2Δ diploid cells expressing srs2K41A,L844A, srs2K41A,Δ875–902, or srs2K41A,L844A,Δ875–902 from pRS415 were not able to form viable colonies (S6D Fig). These results suggest that Srs2L844A and Srs2Δ875–902 retain some ability to interact with Rad51 in vivo, consistent with the results of a previous study that the phenotype of srs2Δ875–902 cells is similar to wild-type [52].
Most srs2Δ mus81Δ diploids arrested with 4C DNA content and were large-budded cells with a single nucleus, suggesting a significant delay of entry into mitosis (S7 Fig). The simplest explanation for the synergistic growth defect in srs2Δ mus81Δ diploids is that the double mutants had unresolved inter-homolog joint molecules, which resulted in G2/M arrest, as observed in srs2K41A diploids. PFGE analysis consistently revealed that chromosomal DNA from srs2Δ mus81Δ diploid cells, but not from their haploid counterparts, formed structures that failed to enter the gel (Fig 6C). Moreover, these DNA accumulations in srs2Δ mus81Δ diploid cells were suppressed by the rad51Δ mutation (Fig 6C). This ploidy-specific behavior is consistent with our other results, and probably reflects the accumulation of inter-homolog joint molecules.
The phenotypic similarity between srs2K41A and srs2Δ mus81Δ suggested the possible functional interaction between Srs2K41A and Mus81. To address this possibility, GAL-srs2K41A mus81Δ, GAL-SRS2 mus81Δ, and GAL-empty mus81Δ diploid strains were constructed, and the effect of expressing Srs2K41A or Srs2 in srs2Δ mus81Δ diploids in the presence of 0.02% galactose was examined. In a control experiment, expression of wild-type Srs2 complemented the growth defect of the GAL-SRS2 mus81Δ mutant (Fig 6D). Notably, whereas Srs2K41A expression aggravated the growth of the GAL-srs2K41A diploid strain, it had no effect on the growth of the GAL-srs2K41A mus81Δ diploid strain (Fig 6D). These results suggest that the srs2K41A mutant behaves similarly to the srs2Δ mus81Δ double mutant. It should be noted that srs2K41A was lethal in diploid yeasts, but srs2Δ mus81Δ diploid cells were viable, albeit with poor growth, suggesting that the growth defect in srs2K41A diploids is more toxic than in srs2Δ mus81Δ diploids.
Srs2 has a dual function in HR, preventing unscheduled recombination and promoting the SDSA pathway during DSB repair. Our results showed that srs2K41A diploids, but not haploids, had a severe defect in growth. GAL-srs2K41A diploid cells showed an elevated number of Rad52 foci and an increase in the rate of gross chromosomal rearrangements, suggesting a high rate of spontaneous HR-associated DNA damage. Indeed, these growth defects were suppressed by inactivation of Rad51 and also by deletion of the Rad51 interaction domain of Srs2K41A. These results imply that DSBs are not responsible for the toxic effects of srs2K41A in diploid yeast, since the repair of DSBs is essential for cell survival and requires functional HR. Indeed, in PFGE analysis, fragmented chromosomes were not detected, but joint molecule accumulations were observed in Srs2K41A-expressing diploid cells. To repair ssDNA gaps by HR, homologous sequences located on the same or different chromosomes can serve as templates. Especially, inter-homolog HR occurs only in diploids, whereas inter-sister HR can occur in both haploid and diploid yeasts. Our results suggest, therefore, that the diploid-specific lethality of srs2K41A is the result of a failure to resolve joint molecules formed during inter-homolog HR.
Srs2 is phosphorylated by Cdk1 and sumoylated in response to DNA damage [33,38]. Cdk1-dependent phosphorylation of Srs2 is required to promote the SDSA pathway for DSB repair. Srs2 sumoylation may have a role in the assembly and/or stabilization of protein complexes involved in DNA repair, although its biological significance remains obscure [40]. In this study, low-abundance Srs2K41A underwent both phosphorylation and sumoylation at multiple sites in haploid and diploid cells, even in the absence of DNA damage. Mutational analysis revealed that sumoylation and phosphorylation of Srs2K41A were largely independent events, which was consistent with the results of a previous study [40]. Moreover, our data demonstrated that Srs2K41A sumoylation and phosphorylation were dispensable for srs2K41A lethality in diploids. By contrast, the lethality of srs2K41A in diploids required its interaction with Rad51. These results suggest that the removal of toxic Rad51 filaments by the Srs2 translocase activity may be essential for the viability of diploid cells.
Our results showed that the DNA damage checkpoint, as monitored by Rad53 phosphorylation, was constitutively activated in haploid and diploid cells expressing Srs2K41A. A previous study showed that overexpression of wild-type Srs2 in haploid cells activates the DNA damage checkpoint in a manner that requires the Srs2 SIM domain [42]. Similar observations were made in our experiments in GAL-srs2K41A haploid cells, suggesting that activation of the DNA damage checkpoint in srs2K41A haploids depends both on DNA replication and the interaction between Srs2 and sumoylated PCNA. By contrast, checkpoint activation and growth inhibition were still observed in GAL-srs2K41A,ΔSIM diploid cells. Thus, GAL-srs2K41A,ΔSIM diploids might accumulate more (or a different type of) DNA lesions than haploid cells of the same genotype, triggering the DNA damage checkpoint. In addition, the rad51Δ mutation did not suppress diploid-specific (unrelated to sumoylated PCNA) Rad53 activation of srs2K41A,ΔSIM diploids, suggesting that this checkpoint activation is unlikely to be associated with the lethality of srs2K41A in diploid cells.
Inter-homolog recombination intermediates form infrequently during HR in mitotic yeast cells. However, if they form, efficient resolution is required to prevent interference with proper chromosome segregation. Our data suggest that Srs2K41A is recruited to inter-homolog recombination intermediates through its interaction with Rad51, and, further, that Srs2K41A inhibits processing of these intermediates, possibly because it lacks a functional helicase/translocase activity. Thus, a possible explanation for Srs2K41A lethality is that in addition to impeding Srs2-dependent HR, it actively blocks a second repair pathway that resolves inter-homolog joint molecules by other helicases or endonucleases. Srs2 has been shown to physically interact with Mus81–Mms4, and to remove Rad51 from DNA, enabling Mus81–Mms4 to access and cleave DNA [54]. In addition, Srs2 and Mus81 co-localize after DNA damage, although Mus81 is fully proficient in focus formation in the absence of Srs2 [54]. A plausible mechanism for a second repair pathway is the resolution of inter-homolog joint molecules by Mus81–Mms4 endonuclease. In this context, it is notable that each of the srs2Δ, mus81Δ, and mms4Δ mutations resulted in greater sensitivity to MMS in diploids than in haploids, which was not true of sgs1Δ mutations. Moreover, srs2Δ mus81Δ diploid cells exhibited a severe growth defect and a marked accumulation of joint molecule intermediates, which were also observed in Srs2K41A-expressing diploid cells. It remains unclear whether the non-migratory DNA complexes observed during PFGE are direct substrates for Mus81–Mms4. However, our genetic and physical studies showed that the rad51Δ mutation could suppress both lethality and joint molecule accumulation in srs2K41A and srs2Δ mus81Δ diploids. Moreover, expression of Srs2K41A aggravated the growth of srs2Δ diploid cells, whereas it did not affect growth in srs2Δ mus81Δ diploid cells. Taken together, these findings suggest that the lethality of srs2K41A and srs2Δ mus81Δ diploid cells was likely to be associated with joint molecule accumulation, and that Srs2K41A actively blocks at least in part the Mus81–Mms4 pathway.
These diploid-specific phenotypes of srs2K41A and srs2Δ mus81Δ imply that inter-homolog and inter-sister HR are somewhat mechanistically different in the processing of HR intermediates. Previous studies in haploid yeast reported that the sgs1Δ srs2Δ and sgs1Δ mus81Δ double mutants, but not srs2Δ mus81Δ, are lethal in haploid yeast [55–57]. Sgs1–Top3–Rmi1 (STR) is required to prevent mitotic crossovers by dissolving double Holliday junction structures through the formation of hemicatenanes [8,13,14]. These results indicate that Sgs1 has an important role in dissociating joint molecule intermediates that arise during HR. A possible explanation for the differential results in haploid and diploid yeasts is that some of the inter-homolog joint molecules that accumulate in srs2K41A and srs2Δ mus81Δ diploid cells cannot be resolved by the STR complex. Cohesion complexes are recruited to sites of DNA damage independently of DNA replication and have a role in suppressing inter-homolog HR by holding sister chromosomes together [58–60]. We speculate that the STR complex might have limited ability to dissociate inter-homolog joint molecules that contain sister-chromatid cohesin rings because cohesin complexes sterically block the formation of hemicatenanes by restricting the decatenation of topologically linked DNA structures between homologous chromosomes. Alternatively, inter-homolog joint molecules might include specific substrates for Mus81–Mms4, such as a single Holliday junction that cannot be resolved by the STR complex. Indeed, it has been reported that joint molecules formed in the mus81Δ mutant contain single Holliday junctions [11].
Our results demonstrate that Srs2 and Mus81–Mms4 participate in essential pathways to prevent the accumulation of toxic inter-homolog joint molecules. In this context, Srs2 may prevent formation of joint molecule structures resulting from inter-homolog HR, whereas Mus81–Mms4 might have a downstream role in promoting their resolution. Unprocessed inter-homolog joint molecules result in chromosome nondisjunction, leading to genetic instability and a high likelihood of cell death. Uncontrolled inter-homolog HR in human cells is associated with genomic instability, such as loss of heterozygosity and gross chromosomal rearrangements, which are hallmarks of cancer cells. Hence, elucidation of the mechanisms controlling inter-homolog HR in diploid yeast could provide new insights into the mechanisms of cancer and aging in humans.
All yeast strains used in this study are listed in S1 Table (see Supporting Information). All double mutants and triple mutants were constructed by standard genetic methods. The details of strains and plasmids produced for and used in this study are presented in S1 File (see Supporting Information).
Cells were grown in yeast extract–peptone–dextrose medium containing 0.01% adenine sulfate (YPD) at 30°C. Cells transformed with pRS415 derivatives were selected on Synthetic Complete (SC)+Glucose medium lacking leucine (SC+Glu-Leu). For Srs2 expression from the AUR1 locus, cells grown exponentially in YPD or YP-2% raffinose (YPR) medium were further incubated at 30°C for various times in YPR medium containing galactose. In a mating assay to produce diploid cells, the resulting diploid cells were selected on SC+Glu medium lacking both histidine and leucine. Disome cells transformed with pRS415 derivatives were selected on SC+Glu medium lacking both leucine and histidine and containing 300 μg/mL G418 (Sigma-Aldrich). Cells resistant to both canavanine and 5-fluoroorotic acid (5-FOA) were selected on SC+Glu medium lacking arginine and containing 60 μg/mL canavanine and 1 mg/mL 5-FOA. For Srs2 expression from p415GAL1 derivatives, cells grown in SC+Glu-Leu medium were washed with water, and the cells (1×107 cells/mL) were further incubated at 30°C for 6 h in SC-Leu medium containing 2% raffinose and 0.2% galactose. For Srs2 expression from the AUR1 locus of disome strains, cells grown in SC+Glu-His+G418 medium containing 0.05 μg/mL aureobasidin A were washed with water, and the cells (2×106 cells/mL) were further incubated at 30°C for 3 h in SC-His+G418 medium containing 2% raffinose in place of glucose and then transferred to 0.5% galactose-containing medium.
Protein extracts were prepared from 1×108 cells using the trichloroacetic acid (TCA) method, as described previously [61]. Proteins were separated by SDS-PAGE, transferred to PVDF membrane, and probed with anti-Srs2 or anti-Rad53 polyclonal antibodies (Santa Cruz).
Yeast chromosomes were separated with CHEF-Mapper XA (Bio-Rad) in 0.8% agarose with 0.5×TBE buffer and stained using ethidium bromide or SYBR Green I (Life Technologies). Gel images were acquired with an LAS4000 mini system (GE Healthcare). The intensity of chromosome bands was quantified using Image J software (NIH). For samples digested with NotI and RuvC, the plugs prepared for PFGE were washed twice more with H buffer (Takara) containing 0.01% Triton X-100 and then washed twice with the same buffer containing 1.3 mM phenylmethylsulfonyl fluoride (PMSF). The plugs were treated with 300 units/mL NotI at 37°C for 16 h in the same buffer. The NotI treated plugs were washed twice with a buffer containing 20 mM Tris-HCl (pH7.5), 10 mM Mg(OAc)2, and 1 mM DTT, and then washed twice with the same buffer containing 1.3 mM PMSF. The plugs were further incubated at 37°C for 16 h in the same buffer containing 8 μg/mL RuvC. After incubation, the plugs were treated with proteinase K and washed twice with 0.5×TBE for PFGE analysis. Southern blotting was performed essentially as described previously [62]. Chromosomes were transferred to a charged nylon membrane (Hybond-N+; GE Healthcare) and hybridized with alkaline phosphatase-labeled probes, which were prepared from PCR products (chromosome IV; 463,680–463,707). After hybridization, the membrane was treated with CDP-Star (GE Healthcare), and chromosomes were detected with the LAS4000 mini imaging system.
The frequency of loss of a pair of chromosome V homologs marked with CAN1 on the right arm and URA3 on the left arm was determined. Briefly, cells were grown in SC+Glu medium lacking histidine and uracil, and further incubated at 30°C for 3 h in YPR medium. After incubation, galactose (0.02%) was added to the cultures, followed by incubation at 30°C for 8 h. Cells from each culture were washed and spread onto plates at an appropriate dilution to determine the total cell number (on YPD plates) and the cell number with allelic loss of chromosome V (on SC+Glu plates containing canavanine and 5-FOA). Colonies arising on YPD and SC+Glu plates containing canavanine and 5-FOA were counted after 3 or 4 days of growth at 30°C. The chromosome-loss frequency was determined from the number of colonies with both CanR and 5-FOAR per mL divided by the number of viable cells per mL, and the average from three independent experiments was calculated. p values were calculated using a Student’s two-tailed t test.
Fluorescence-activated cell sorting (FACS) analysis, microscopy, and spot assays for MMS sensitivity were performed as described previously [63].
|
10.1371/journal.ppat.1003972 | Code-Assisted Discovery of TAL Effector Targets in Bacterial Leaf Streak of Rice Reveals Contrast with Bacterial Blight and a Novel Susceptibility Gene | Bacterial leaf streak of rice, caused by Xanthomonas oryzae pv. oryzicola (Xoc) is an increasingly important yield constraint in this staple crop. A mesophyll colonizer, Xoc differs from X. oryzae pv. oryzae (Xoo), which invades xylem to cause bacterial blight of rice. Both produce multiple distinct TAL effectors, type III-delivered proteins that transactivate effector-specific host genes. A TAL effector finds its target(s) via a partially degenerate code whereby the modular effector amino acid sequence identifies nucleotide sequences to which the protein binds. Virulence contributions of some Xoo TAL effectors have been shown, and their relevant targets, susceptibility (S) genes, identified, but the role of TAL effectors in leaf streak is uncharacterized. We used host transcript profiling to compare leaf streak to blight and to probe functions of Xoc TAL effectors. We found that Xoc and Xoo induce almost completely different host transcriptional changes. Roughly one in three genes upregulated by the pathogens is preceded by a candidate TAL effector binding element. Experimental analysis of the 44 such genes predicted to be Xoc TAL effector targets verified nearly half, and identified most others as false predictions. None of the Xoc targets is a known bacterial blight S gene. Mutational analysis revealed that Tal2g, which activates two genes, contributes to lesion expansion and bacterial exudation. Use of designer TAL effectors discriminated a sulfate transporter gene as the S gene. Across all targets, basal expression tended to be higher than genome-average, and induction moderate. Finally, machine learning applied to real vs. falsely predicted targets yielded a classifier that recalled 92% of the real targets with 88% precision, providing a tool for better target prediction in the future. Our study expands the number of known TAL effector targets, identifies a new class of S gene, and improves our ability to predict functional targeting.
| Many crop and ornamental plants suffer losses due to bacterial pathogens in the genus Xanthomonas. Pathogen manipulation of host gene expression by injected proteins called TAL effectors is important in many of these diseases. A TAL effector finds its gene target(s) by virtue of structural repeats in the protein that differ one from another at two amino acids that together identify one DNA base. The number of repeats and those amino acids thereby code for the DNA sequence the protein binds. This code allows target prediction and engineering TAL effectors for custom gene activation. By combining genome-wide analysis of gene expression with TAL effector binding site prediction and verification using designer TAL effectors, we identified 19 targets of TAL effectors in bacterial leaf streak of rice, a disease of growing importance worldwide caused by X. oryzae pv. oryzicola. Among these was a sulfate transport gene that plays a major role. Comparison of true vs. false predictions using machine learning yielded a classifier that will streamline TAL effector target identification in the future. Probing the diversity and functions of such plant genes is critical to expand our knowledge of disease and defense mechanisms, and open new avenues for effective disease control.
| Bacterial leaf streak of rice (Oryza sativa), caused by Xanthomonas oryzae pv. oryzicola (Xoc), and bacterial blight of rice, caused by the closely related Xanthomonas oryzae pv. oryzae (Xoo) are important constraints to production of this staple crop in many parts of the world. Yield losses as high as 50% for blight and 30% for leaf streak have been documented [1]. Leaf steak in particular appears to be growing in importance, as high-yielding but susceptible hybrid varieties of rice are increasingly adopted (C. Vera-Cruz and G. Laha, personal communications). Xoc enters through leaf stomata or wounds and interacts with mesophyll parenchyma cells to colonize the mesophyll apoplast, causing interveinal, watersoaked lesions that develop into necrotic streaks. Quantitative trait loci for resistance to leaf streak have been characterized [2], but native major gene resistance has yet to be identified. In contrast, Xoo typically enters through hydathodes or wounds and travels through the xylem, interacting with xylem parenchyma cells through the pit membranes, and typically resulting in wide necrotic lesions along the leaf margins or following veins down the center of the leaf. Only in later stages of disease development does Xoo colonize the mesophyll. Also in contrast to leaf streak, roughly 30 independent genes for resistance (R) to blight have been identified and seven molecularly characterized [3], [4]. The basis for the distinct tissue specificities of Xoc and Xoo and the disparity in known host resistance, despite the genetic similarity of the two pathogens, is not known.
Virulence of Xoo, and of Xanthomonas that infect citrus, cotton, or pepper, is influenced by transcription activator-like (TAL) effectors [5]–[15]. Widespread in Xanthomonas, TAL effectors are proteins delivered into the plant cell via type III secretion (T3S) that transactivate effector-specific host genes [16], [17]. If activation is important for disease, the target is considered a susceptibility (S) gene [9]. Individual Xoo strains harbor multiple, distinct TAL effector (tal) genes [8], and several bacterial blight S genes have been identified. The first of these were Os8N3 (a sugar transporter gene family member also and hereafter referred to as OsSWEET11), the bZIP transcription factor OsTFXI, and the transcription initiation factor TFIIAγI, upregulated respectively by TAL effectors PthXo1, PthXo6, and PthXo7 of Xoo strain PXO99A [9], [10]. More recently, the closely related OsSWEET11 paralog OsSWEET14 (also Os11N3) was discovered to be an S gene targeted by several distinct TAL effectors from other strains [11], [18], [19]. A third close paralog upregulated during infection by some strains, OsSWEET12, also functions as an S gene, though a TAL effector that upregulates it has not yet been reported [19], [20]. The recessive blight R genes xa13 and xa25 are promoter variant alleles of OsSWEET11 and OsSWEET12, respectively, that are not activated by the corresponding TAL effector (or presumed TAL effector in the case of OsSWEET12) [9], [20]. Some TAL effectors induce host resistance by transcriptionally activating a type of dominant R gene that triggers local cell death when expressed, for example the archetypal TAL effector AvrBs3 from the pepper pathogen X. euvesicatoria [21], which activates the pepper Bs3 gene for resistance to bacterial spot [17], and the Xoo effector AvrXa27, from strain PXO99A, which induces the rice R gene Xa27 [22]. Like Xoo, Xoc strains harbor multiple tal genes [8], [23]. However, though the T3S system through which TAL effectors travel is required for leaf streak [24], the role of Xoc TAL effectors in the disease is uncharacterized, and no leaf streak S genes have been identified.
TAL effectors find their targets via a structurally modular mechanism that allows prediction of DNA specificity and customization to target nucleotide sequences of choice [25]–[29]. The modules are tandem repeats of a 33–35 amino acid sequence, exhibiting polymorphism at residues 12 and 13, together called the repeat variable diresidue (RVD). Different RVDs were shown computationally and experimentally, and later structurally to each specify a single nucleotide through direct interaction with (or exclusion of other bases by) the residue 13 side chain, such that the string of RVDs presented by the repeats “encodes” the sequence of the so-called TAL effector binding element (EBE) on the DNA [25], [26], [30], [31]. The RVD nucleotide associations observed in nature are not strictly one to one, however [26]. Indeed, all known natural EBEs contain one or more mismatches to the corresponding TAL effector RVD sequence, a mismatch being a base different from the one most commonly associated with the RVD. Furthermore, some RVDs have dual or even entirely lax specificity. So, the TAL effector-DNA binding code is partially degenerate, rendering target prediction probabilistic [26], [32]. Finally, EBEs in nature are almost all directly preceded by a 5′ thymine (T) that has been shown, in the few studied cases, to be important for TAL effector-driven gene activation as well as full affinity DNA binding [33]–[35]. The single known exception, EBETalC in the promoter of OsSWEET14, displays a cytosine (C). Although the effect of substituting a T was not tested directly, a perfect match EBE for TalC, with a T at base 0 and corrected mismatches at two other locations, indeed showed higher activity [13]
In this study, we sought to better understand bacterial leaf streak in relation to bacterial blight, particularly with an eye toward identifying determinants of tissue specificity, and to examine the roles of Xoc TAL effectors in disease. We began by comparing transcription profiles in Xoc-, Xoo-, and mock-inoculated plants by microarray analysis. We then combined the transcriptomic data with computational identification of candidate EBEs to predict TAL effector targets, and carried out experiments to differentiate real from falsely predicted ones. Screening a TAL effector mutant library of Xoc, we next identified a TAL effector that plays a major role in virulence, and we discriminated from among its two targets the first known S gene for leaf streak, in part by using designer TAL effectors to independently activate the genes. Using our complete list of newly discovered targets as well as the previously identified Xoo targets represented in our dataset, we next examined general characteristics of TAL effector driven gene expression. Finally, in an attempt to better discriminate real targets from falsely predicted ones in the future, prior to experimentation, we used machine learning to train a classifier on primary and contextual features of EBEs in the respective groups. Our results provide new insight into bacterial leaf streak, increase the number of known natural TAL effector combinations by 20, identify a new class of S gene, and advance our understanding of and ability to predict functional targeting by TAL effectors.
We initially set out to determine whether there are differences in host genome-wide expression patterns during bacterial leaf streak vs. bacterial blight that might help to explain the different tissue specificity of Xoc and Xoo. Using a vacuum infiltration approach developed from a dipping method we showed previously to be effective for both pathovars [36], we inoculated rice (cv. Nipponbare) plants en masse with Xoc strain BLS256 (hereafter Xoc refers to this strain unless otherwise specified), Xoo strain PXO99A (likewise), or a mock inoculum, harvested leaves at 2, 4, 8, 24, and 96 hours thereafter, and quantified transcript levels in these leaves for the roughly 56,000 annotated rice genes in parallel using the Affymetrix GeneChip Rice Genome Array [37]. We focused our analysis on patterns of expression across the time course rather than expression levels at a particular time point and examined three pairwise comparisons, Xoc vs. mock, Xoo vs. mock, and Xoc vs. Xoo (see Materials and Methods).
A total of 505 genes showed significantly different expression profile patterns (q≤0.3; see Materials and Methods) in one or more of the pairwise comparisons (Figure 1). Eighty and 94 genes were differentially expressed uniquely in response to Xoc or Xoo, respectively (Figure 1; Table S1 and Table S2). Only five genes were differentially expressed both in response to Xoc and Xoo relative to mock: three similarly between Xoc- and Xoo- and two with different patterns in Xoc- vs. Xoo-inoculated plants (Figure 1; Table S3). Strikingly, all of the statistically significantly differentially expressed genes showed patterns of upregulation in response to Xoc or Xoo. Expression patterns of the ten or fewer most significantly differentially expressed genes in response to Xoc, Xoo, or both are shown in Figure 2.
Singular enrichment analysis [38] of gene ontology (GO) for all Xoc- and Xoo-upregulated genes revealed broad differences in the major categories represented (Table S4 and Table S5). Six significant GO terms were identified for Xoc-induced genes. Four of these are categorized under biological processes and include coenzyme metabolic, cofactor metabolic, sulfur metabolic and, cellular amino acid derivative metabolic processes. The other two, catalytic and oxidoreductase activities, are grouped under molecular function (Table S4). For Xoo-induced genes, the significant terms all fall within the cellular component category, including membrane-bounded vesicle, vesicle, cytoplasmic membrane-bounded vesicle, and cytoplasmic vesicle (Table S5). The most abundant ontology category for genes induced by Xoc was catalytic activity, and included several glutathione S-transferase and oxidase genes (Table S4). These were part of a large group of Xoc-induced genes, distributed among several categories, with annotations that suggest roles in reactive oxygen species detoxification and redox status control (assembled together in Table S6).
Among the complete list of Xoo-induced genes are each of the bacterial blight S genes previously reported to be induced by PXO99A TAL effectors, OsSWEET11 (Os08g42350), OsTFXI (Os09g29820), and TFIIAγI (Os01g73890) (Table S2 and Table S7). Notably, none of these three genes nor any of the OsSWEET11 paralogs reported to function as bacterial blight S genes [11], [19], [20] was activated following inoculation with Xoc.
Thus, host genome wide expression patterns during bacterial leaf streak vs. bacterial blight are almost completely different.
The TAL effector inventories in Xoc and Xoo are entirely distinct. Xoc harbors 26 unique, intact TAL effector genes and Xoo 14, with no shared predicted EBEs based on RVD sequences [23], [39]. The inventories of predicted non-TAL type III effectors in Xoc and Xoo are similar, but six effector genes present in Xoc are absent from or pseudogenized in Xoo and several minor polymorphisms exist among the shared genes [23]. As a first step to determine the extent to which differences in TAL or other type III effector content might account for the differences in rice global transcription patterns we observed, we asked whether T3S is required for induction of the top ten rice genes most significantly induced uniquely following inoculation with Xoc, the top ten induced by Xoo, and all five induced in common by both strains. We compared, by RT-PCR, transcript accumulation after inoculation with the wild-type strains or with T3S-deficient derivatives BLS256hrcC− [24] and PXO99AME7 [9]. Induction of each gene required bacterial T3S (Figure 3 and [9], [10]).
Among the top ten Xoo-induced genes are the TAL effector targets OsSWEET11 (Os08g42350) and TFIIaγ1 (Os01g73890). The patterns of induction of each of the top Xoc- or Xoo-induced genes revealed by the genome-wide expression analysis described in the previous section vary, but some are similar to that of OsSWEET11 and TFIIaγ1 (Figure 2). This similarity and the T3S-dependence of expression suggested that some of these and perhaps others in the complete lists of induced genes are targets of TAL effectors.
To identify TAL effector targets, we first used the scoring function we developed previously based on observed RVD-nucleotide association frequencies [26], [32] to scan in silico all annotated rice gene promoters (the promoterome) [32] for candidate EBEs for any of the 40 total TAL effectors present in Xoc and Xoo [23], [39]. Some of these TAL effectors have new RVDs whose specificities have not been characterized. The scoring function by default treats new RVDs as wild cards, equally likely to specify any base. However, since structural studies revealed that the second residue of each RVD makes the base-specific contacts while the first stabilizes the inter-helical loop that projects that second residue into the major groove of the DNA [30], [31], we used the specificities of common RVDs for any new RVDs that share the same second position residue. These were limited to two RVDs found in Xoc TAL effector Tal2g, ‘SN’ for which we substituted nucleotide association frequencies of ‘NN’, and ‘YG’ for which we substituted those of ‘NG’. Candidate EBEs were required to be directly preceded by a T at the 5′ end and, for each TAL effector, to score below a cutoff calculated based on the distribution of scores for that effector (see Materials and Methods). This list was then cross-referenced to the GeneChip expression data, and genes with one or more candidate EBEs in the promoter that were also induced following inoculation with the corresponding strain were retained as predicted targets (Table S7).
Thirty-five of these are genes induced by Xoc (three of the 35 are also induced by Xoo), and they collectively contain candidate EBEs for 19 out of the 26 Xoc TAL effectors. Twenty-nine are genes induced by Xoo (five are also induced by Xoc), and they together contain putative EBEs for all 14 of the unique Xoo TAL effectors (Tal7a and 7b are identical to Tal8a and 8b, respectively). The latter include each of the three previously demonstrated targets of Xoo (i.e., PXO99A) TAL effectors in Nipponbare, OsSWEET11 targeted by PthXo1, OsTFXI targeted by PthXo6, and TFIIAγI targeted by PthXo7 [9], [10], [40] (the AvrXa27-activated allele of Xa27 is not present in Nipponbare). Among the five genes induced in common by Xoc and by Xoo, two were predicted to be targeted by a TAL effector from Xoo but not by one from Xoc (Os01g58240 by Tal4 and Os01g40290 by Tal7b/8b of Xoo). In the other three, sequence distinct, candidate EBEs for one or more TAL effectors from each strain were found in the promoters (EBEs for Tal2c and Tal3b of Xoc and AvrXa27 and Tal9b of Xoo in Os03g03034, for Tal1c and Tal3a of Xoc and Tal9a of Xoo in Os07g06970, and for Tal5a and Tal11a of Xoc and Tal9e of Xoo in Os02g15290).
Of the 35 total genes induced by Xoc that harbor a candidate EBE for an Xoc TAL effector, eight harbor EBEs for more than one. Likewise, of the 29 Xoo-induced genes that match an Xoo TAL effector, four genes contain EBEs for multiple Xoo TAL effectors. These results suggest for both pathovars a partial redundancy among effectors for some targets. The Xoc-induced gene Os06g14750 and the Xoo-induced gene Os07g11510 contain overlapping candidate EBEs for three TAL effectors each from those strains, Tal2a, Tal1c, and Tal11b, and PthXo6, Tal2a, and Tal5a, respectively.
The number of predicted targets for individual TAL effectors varies. In the case of Xoc, we identified five predicted targets each for Tal3b and Tal6, and one of the predicted Tal6 targets, Os12g42970, harbors two candidate Tal6 EBEs. Five Xoc TAL effectors, Tal2c, Tal5a, Tal8, Tal9b and Tal11b, have only one predicted target each. For Xoo, we predicted seven targets for PthXo6 and one target each for PthXo1, PthXo7, Tal6a, Tal7a/8a, Tal9d, and Tal9e. AvrXa27 had five predicted targets, two of which, Os06g03080 and Os06g03120, are paralogs nearly identical in their coding sequences and both represented by a single probeset. The promoters of these genes share the same AvrXa27 EBE (one of two AvrXa27 EBEs in Os06g03120), but are otherwise distinct.
In sum, all but a few of the TAL effectors of Xoc and Xoo have candidate binding sites in a gene upregulated by that strain; a total of 61 out of 179, or roughly one-third, of the genes induced following inoculation with Xoc, Xoo, or either strain are predicted targets of those TAL effectors; and within these predictions multiple targets per TAL effector as well as multiple TAL effectors per target were observed.
The next step was to determine which predicted TAL effector targets are real targets. Because several S genes for bacterial blight of rice have been characterized and all are TAL effector targets, while no S genes have yet been identified for bacterial leaf streak and the roles of TAL effectors in this disease have not been explored, we focused on the 44 TAL effector-target pairs predicted for Xoc (Table 1, taking Tal6 and Os12g42970, with its two Tal6 EBEs, as one pair). To identify real targets, we used both TAL effector loss of function and gain of function assays to test TAL effector dependence of expression. First we generated a library of Xoc TAL effector mutant strains by marker exchange mutagenesis. By mapping the mutation in several strains, we identified loss of function derivatives for all but one (Tal2a) of the TAL effectors for which we had predicted a target. And, we cloned each of the TAL effectors into a broad host range plasmid for complementation and heterologous expression (gain of function). Then we assessed by RT-PCR whether any TAL effector mutant strain failed to activate any of the corresponding predicted targets of that TAL effector, and for any that did, whether the cloned effector specifically complemented the mutation to restore activation. In parallel, we expressed each TAL effector in strain EB08 of the soybean pathogen X. axonopodis pv. glycines (Xag) [41], which neither causes symptoms nor elicits a hypersensitive reaction when inoculated to rice (cv. Nipponbare), and we determined whether the transformants specifically activated corresponding targets.
The results verified 19 of the 44 predicted Xoc TAL-effector targets as real (Table 1 and Figure S1; the Tal2a target was verified only by the gain of function experiment). Another 20 were shown not to be activated by the corresponding TAL effector and are hereafter referred to as falsely predicted targets. The remaining five could not be tested because transcript was not detected by RT-PCR, despite induction according to the GeneChip expression data. Interestingly, multiple predicted targets were verified for some TAL effectors, however, for each of the eight genes predicted to be targeted by multiple TAL effectors, only activation by one of those TAL effectors was verified.
Having identified 19 targets of Xoc TAL effectors, the next challenge was to ascertain whether any are S genes for bacterial leaf streak. Barring redundancy, i.e., targeting of the same S gene by multiple TAL effectors, which our verification experiments excluded for each target tested, loss of a TAL effector that activates an important S gene should by definition result in a reduction of virulence. We therefore first quantified the virulence of each of several mutant strains of Xoc to identify such TAL effectors, using a lesion length assay (Figure 4). Collectively, the mutants account for all 26 Xoc TAL effectors except Tal2a, for which a mutant was not isolated. Assayed on rice cv. Nipponbare plants, only mutations that map on at least one side to the 3′ end of the tal2 cluster, i.e., involving tal2f or tal2g, or that map to the tal11 cluster, which contains tal11a and tal11b, were associated with significantly reduced virulence, 49–64% and 64–79%, respectively. Thus, most of the Xoc TAL effectors, in the context of the Nipponbare host genotype, appear not to make any non-redundant, major contributions to virulence. Interestingly, this includes the TAL effectors that activate genes induced in common by Xoc and Xoo, Tal1c, Tal2c, and Tal5a (Table 1, Table S3, and Table S7).
Of the few Xoc TAL effectors pinpointed by the mutational analysis as possible virulence factors that might lead us to one or more S genes (Tal2f, Tal2g, Tal11a, and Tal11b), we had verified targets only for Tal2g (Table 1). From the code- and GeneChip expression-based analysis, Tal2f had no predicted targets, and two of the three predicted targets of Tal11a and the sole predicted target of Tal11b were shown not to be actual targets by the loss- and gain-of-function RT-PCR experiments (Table 1). So, we focused on Tal2g. Of the three mutant strains in which the mutation endpoints map within or flanking Tal2g (Figure 4: M27, M30, and M134), we chose mutant M27 for further characterization. In M27, the marker exchange endpoints suggest a complex recombination, with a disrupted tal2f on the 5′ end and a disrupted tal2b′, a pseudogene that resides 5′ of tal2f in the native chromosome, on the 3′ end. Because the apparent complex recombination might have affected several genes in the cluster, we assayed each tal2 gene (tal2a, -c, -d, -e, -f, and -g), individually on a plasmid for the ability to complement M27. Only tal2g restored virulence to M27 in the lesion length assay, and it did so fully, confirming Tal2g as the sole virulence factor among the TAL effectors whose expression is disrupted in this mutant (Figure 5A). The marker exchange endpoints in M27 could be explained by a double crossover between tal2b′ and tal2g, concurrent with the marker exchange crossovers, that positioned tal2b′ sequences at the 3′ endpoint of the exchange, with the 5′ end in tal2f, disrupting tal2g but not affecting tal2c, tal2d, or tal2e. Consistent with this, the verified targets of Tal2c and Tal2d (Os03g03034 and Os04g49194) are induced by M27 (Figure S2).
Curiously, the total population of M27 isolated from leaf homogenates at seven days after inoculation was not significantly different from that of the wild type (Figure 5B). However, we observed less bacterial exudate on the surface of M27-inoculated leaves than on leaves inoculated with wild type (see Figure 4B). When surface bacteria were isolated and quantified (see Materials and Methods), M27 indeed showed nearly a 400-fold reduction relative to the wild type, and Tal2g on a plasmid fully restored wild-type levels of exudate (Figure 5B). Thus, Tal2g is a major virulence factor in bacterial leaf streak that functions both in lesion expansion and exudation of bacteria to the leaf surface.
The two verified targets of Tal2g, Os06g46500, encoding a predicted monocopper oxidase, and Os01g52130, encoding a predicted sulfate transporter, OsSULTR3;6 [42], are among the most significantly induced genes in the GeneChip expression dataset (Table S1). To test whether either is a biologically relevant target, i.e., an S gene, we engineered designer TAL effectors (dTALEs) to specifically activate each target individually, and we tested the ability of these dTALEs to restore virulence to M27 (Figure 6). Assayed by RT-PCR, in syringe infiltrated leaves dTALE dT434 expressed in M27 specifically induced the monocopper oxidase gene, and dTALEs dT436 or dT437 induced OsSULTR3;6, each similarly to wild type and to M27 expressing Tal2g (Figure 6B). In the lesion length assay, dT436 and dT437 each restored full virulence to M27, whereas dT434 made no significant difference (Figure 6C). When surface bacterial populations were quantified over time at the inoculation site, and spread of bacteria over time was measured by quantifying total populations in contiguous leaf segments at and extending from the inoculation site, M27 expressing dT437 and M27 expressing Tal2g behaved the same as the wild type, whereas M27 expressing dT434 showed a reduction in surface population and slowed population spread equivalent to M27 carrying the empty vector (Figure 6D and Figure 6E). Scanning the rice promoterome for candidate EBEs as in our original search for potential Xoc and Xoo TAL effector targets, we found no overlap between candidate off-targets of dT436 and dT437, or between off-targets of either with genes harboring a potential Tal2g EBE. Together, the data therefore indicate that OsSULTR3;6 is the relevant Tal2g target and a major S gene for bacterial leaf streak.
As described above, in our search for TAL effector targets, we used specificity values of ‘NN’ and ‘NG’ for the ‘SN’ and ‘YG’ RVDs that are found in Tal2g. As might be expected, the list of candidate Tal2g EBEs generated using these values differed from a second list we generated in parallel using the default, wild card values. Specifically, in the list generated using the default values for ‘SN’ and ‘YG’, hereafter referred to as the default scoring list, the verified Tal2g target Os06g46500 did not make the cutoff (Materials and Methods) to be considered a candidate (indeed no sequence from any Xoc-induced gene beside OsSULTR3;6 scored well enough in this list to be considered a candidate), indicating that substituting the RVD specificity values allowed us to capture an otherwise false negative.
To further probe the validity of substituting the values, we tested the function of two candidate EBEs from the default scoring list that each scored better (lower; see Materials and Methods) than the (default-scored) EBEs in the two verified targets, but that displayed a mismatch to one or each of the two new RVDs in Tal2g based on the presumed specificities of those RVDs (Figure 7A). Though not induced by Xoc, both of the corresponding genes, Os06g13880 and Os12g36920, are induced by Xoo (Table S2), indicating that they are euchromatic. Also, the default-scored candidate EBEs, at 139 bp and 86 bp upstream of the respective annotated transcriptional start sites, are each within the range of locations displayed by the EBEs in all the targets verified in this study (152 bp or less; Table 1), so failure to be induced by Xoc likely does not relate to suboptimal EBE localization. We also chose to test a third sequence with a mismatch to one of the new RVDs, that scored just above the cutoff in the default scoring list (Figure 7A) and was therefore not considered a candidate, but was nonetheless in the promoter of an Xoc-induced gene Os05g10650 (Table S1), and therefore a potential false negative in that list. To test the function of the three sequences, we used a transient, Agrobacterium-mediated, TAL effector-driven reporter gene expression assay in Nicotiana benthamiana [40]. None of the sequences, inserted into a 343 bp fragment of the pepper Bs3 promoter just upstream of the native EBE for the cognate TAL effector AvrBs3 [25], rendered the reporter responsive to Tal2g (Figure 7B). In contrast, the EBEs from the verified targets of Tal2g resulted in strong and specific induction of the reporter by Tal2g similar to induction of the unamended reporter by AvrBs3 (Figure 7C).
Thus, in addition to capturing the verified target Os06g46500 as a candidate, the substituted scoring correctly classifies the Os12g36920 and Os05g10650 sequences as non-candidates (scored above the cutoff). The substituted scoring scores the Os06g13880 sequence as worse than the EBEs of the two verified targets, consistent with its lack of activity, but still calls it a candidate. This incongruity might be explained by the observation that the Os06g13880 sequence displays a mismatch to the first RVD of Tal2g (Figure 7A), and mismatches at the 5′ end and especially at the first position have been shown to more strongly negatively affect activity than mismatches elsewhere [43] a phenomenon not accounted for by the scoring function. Taken together, the observations overall support the assignment of the common RVD specificities for those of the new cognate RVDs, in agreement with the inference from published structural data discussed earlier.
Returning to our list of 19 new, verified TAL effector-target pairs, we next sought to determine whether the expression patterns of the targets might reveal general characteristics of TAL effector-driven gene expression. Using the normalized (log2 transformed) GeneChip expression data, we began by comparing the average transcript levels of the targets at two hours after inoculation in mock- or Xoc-inoculated plants to expression levels of 1) the 20 falsely predicted targets, 2) all genes differentially expressed (DE) in the Xoc vs. mock comparison, and 3) all genes represented on the array (Figure 8). This average basal expression level of the targets was nearly identical in mock- and Xoc-inoculated plants, similar to that of the falsely predicted targets, slightly higher than that of all genes DE in the mock vs. Xoc comparison, and markedly higher than the average expression level for all genes under either condition at any time point (4.4). Indeed, the majority (14 of 19) of the targets showed basal levels (two hours after inoculation with Xoc) higher than that average (Table S7; for the analyses presented here and throughout this section, genes represented by two probesets in any table were assigned the average values of those probesets). The target with the highest normalized basal expression level was Os03g37840 targeted by Tal4a, at 7.6, approximately 1.7 times the genome-wide average at that time for either Xoc- or mock-inoculated plants.
We next examined expression at two hours after Xoc inoculation relative to expression at 96 hours after that treatment. The average fold induction (Table 1 and Table S7) of the targets (calculated as 2average(X-Y), where fold induction of a gene is defined as 2(X-Y) for the difference between normalized expression values X and Y; see Materials and Methods) was 3.3, compared to an average of 2.7 for the falsely predicted targets (Table 1 and Table S7) and 2.6 for all the genes DE in the Xoc vs. mock comparison (Table S1 and Table S3). Compared to the average for all 19 targets, induction of 11 of the 14 targets with higher than average basal expression levels was moderate, from the overall minimum of 1.2-fold, exhibited by the Tal2a target Os02g43760, to 3.3-fold, whereas the five targets basally expressed at or below the average for the genome were induced 1.6- to 9.1-fold. The remaining three targets, which were expressed at higher than average basal levels, varied in their induction from 6.4- to the overall high of 22.4-fold exhibited by the Tal2d target Os04g49194. This latter value was second only to the 34.2-fold induction of Os01g40290 (Table S1), a gene not predicted to be an Xoc TAL effector target. The normalized expression value for the Tal2d target Os04g49194 at 96 hours after Xoc inoculation, 9.4, was also near the maximum across the genome for that time point and treatment, 10.7 (Os11g47970, probeset Os.11573.2.A2_a_at). Right behind was the sulfate transporter S gene Os01g52130 targeted by Tal2g, exhibiting induction of 13.0-fold to an expression level of 9.1. Overall, though there was not a perfect inverse correlation between basal expression level and fold-induction, expression levels of all targets at 96 hours after Xoc inoculation varied relatively little, averaging 6.9 (standard deviation, SD,1.3), suggesting that regardless of initial target expression level, TAL effectors may generally induce genes to a similar final level.
Extending the analysis to the four known Xoo TAL effector-target pairs represented in our data (Table S7), we found that the average basal expression (i.e., two hours following Xoo infection) was 5.4 (SD 0.6), nearly identical to the average basal expression of Xoc TAL effector targets (5.2 with SD 1.3). One of the Xoo TAL effector targets (Os07g06970 targeted by Tal9a, also targeted by Tal1c of Xoc) was expressed basally at near genome-average levels. It was moderately induced, 5.0-fold, by 96 hours after Xoo inoculation. The other three, like the majority of the Xoc TAL effector targets, were each basally expressed at higher than average levels. Two of these, Os01g73890 (TFIIAγ1) and Os09g29820 (OsTFX1), targeted by PthXo7 and PthXo6, respectively, also showed relatively low fold induction (3.2- and 2.2-fold, respectively). The overall average fold induction, 4.9, was higher than that of the Xoc TAL effector targets, but this number is skewed somewhat by the large change, 17.1-fold, in expression of the third target initially expressed at higher than average levels, Os08g42350 (OsSWEET11) targeted by PthXo1. Despite the small sample size, and with the PthXo1 target as a notable exception, the pattern of expression and fold-induction of the Xoo TAL effector targets overall was similar to that observed for Xoc TAL effector targets, tending toward higher than average initial levels and relatively moderate induction.
Finally, to better understand targeting and to improve prediction, we asked whether there are features of EBEs in the real targets we identified that distinguish them from those in our falsely predicted targets. Indeed, inspection of the features listed in Table 1 revealed some that appear to be characteristic of EBEs in real targets (we included both Tal6 EBEs in Os12g42970 in this analysis, for a total of 20 EBEs in real targets). First, on average, EBEs in real targets had lower relative scores. The relative score is the ratio of the actual score for a TAL effector-target alignment to the hypothetical score of that TAL effector aligned with its perfect match target; it allows comparison across TAL effectors, which is otherwise not possible because repeat number and RVD composition affect actual score [32]. The average relative score for EBEs in real targets was 1.98 (range 1.22–2.81), while for falsely predicted targets it was 2.47 (range 1.70–3.18). Second, EBEs in real targets generally ranked more highly in the collection of scores for the TAL effector across all rice promoters than the EBEs in the falsely predicted targets did: 16 of the 20 in real targets ranked in the top 200, with an average rank of 137 across all 20, while 17 of the 20 in falsely predicted targets ranked lower than 200, with an average rank of 347 for all 20. Finally, the maximum distance of an EBE in a real target from the annotated transcriptional start site was 152 bp upstream, with an average of 47 bp upstream (based on 19 that have an annotated TXS, out of the 20 total; range, 152 bp upstream to 63 bp downstream), whereas for the falsely predicted targets, the EBEs were anywhere from 22 bp downstream to 815 bp upstream, with an average distance of 293 bp upstream (based on the 18 with an annotated TXS). Proximity to a TATA box did not appear to correlate independently: nine of the EBEs in real and six of the EBEs in falsely predicted targets are within 100 bp of a TATA box.
To test whether the apparent differences in EBE features could be used to computationally discriminate between real and falsely predicted TAL effector targets and thereby improve future prediction, we took a machine learning approach and trained several Naive Bayes and logistic regression classifiers using combinations of relative score, rank, distance to TXS, and proximity to a TATA box, as well as actual score, distance to translational start site (TLS), and distance to a Y patch, a core promoter motif commonly found in plants [44]. For this analysis, we included also the known Xoo (PXO99A) TAL effector targets in Nipponbare, each of which, as noted above, was among our predictions (Table S7). Classifiers were generated using leave-one-out cross validation, a method that determines model parameters using all but one of the EBEs as the training set and then asks whether the resulting classifier correctly calls the remaining EBE. This is repeated with each EBE in turn to optimize the model. Recall, precision, and other metrics are computed based on the number of EBEs classified correctly using this procedure. A Naive Bayes classifier trained on all features achieved the highest recall, capturing 92% of the real targets (Table 2). The precision (percent of positives called that are true positives) of the classifier was 88% (Table 2), and no other classifier had a significantly better area under the receiver operating characteristic curve (AUC; Figure S3), a measure of the tradeoff between recall and precision. Notably, a logistic regression classifier using the distance to transcriptional start site alone achieved a recall almost as high as that achieved using all features, and had a similar AUC (Table 2 and Figure S3).
In this study we integrated genome-wide expression profiling, computational prediction using the TAL effector-DNA binding code, and functional analyses, and identified a TAL effector target in rice, OsSULTR3;6, that plays a major role in susceptibility of this staple crop species to a disease of increasing global importance, bacterial leaf streak of rice. Key to identifying the S gene was targeted gene activation using designer TAL effectors. Encoding a predicted sulfate transporter, the gene represents a new class of TAL effector-induced S gene, distinct from the handful that has been identified for bacterial blight of rice. Indeed, we discovered that overall, pathogen-induced host transcriptional changes in leaf streak are almost entirely different from those that take place during blight. We found that the T3S-translocated TAL effectors of the leaf streak pathogen are responsible, at a minimum, for nearly a quarter (19/85 genes) of the differential host gene expression during infection that we detected. We identified Tal2g as the major Xoc virulence factor that upregulates OsSULTR3;6, and demonstrated that the upregulation of OsSULTR3;6 contributes specifically to lesion expansion and bacterial exudation. We learned that, on average, TAL effector targets are expressed basally at higher than genome average levels and induced to a moderate extent, though OsSULTR3;6 and the blight S gene OsSWEET11 were exceptions, as two of the most highly induced genes in our dataset. Finally, the targets we identified and predictions we verified to be false allowed us to generate a Naive Bayes classifier that can be used in the future to identify the strongest candidate TAL effector targets prior to verification experiments, and that may also help optimize targeting with dTALEs. These advances leave the key question about tissue specificity unanswered, and raise other questions, but they open promising new avenues of inquiry. Also, they highlight gaps in our understanding of gene activation by TAL effectors, and point to challenges that remain in code-assisted discovery of TAL effector targets, but they demonstrate nonetheless the power of the approach we used to rapidly dissect interactions between TAL effector-wielding pathogens and their hosts.
Regarding the basis for the tissue specificity of Xoc relative to Xoo, the markedly distinct patterns of host global gene expression during bacterial leaf streak compared to bacterial blight suggest a role for host gene manipulation by the pathogens. The results of the gene ontology enrichment analysis we carried out on the differentially expressed genes raise the intriguing possibility that Xoc is uniquely able to control redox status, preventing or dampening the defense-associated oxidative burst and or affecting redox-dependent signaling pathways in the mesophyll. In a preliminary experiment to test this possibility, we observed that Xoc-inoculated leaves do show reduced overall H2O2 content at 96 h after inoculation relative to Xoo- and mock-inoculated plants (Figure S4). The reduction could relate to reduced photosynthesis, as leaves are beginning to exhibit watersoaking by this time, but it could be the direct consequence of Xoc-dependent changes in transcript levels of the redox-modulating genes, as Xoo-infected plants also exhibit watersoaking by 96 hrs, yet are unaltered in their H202 content relative to mock inoculated plants.
In contrast, the abundance of membrane associated and vesicle associated terms unique to the Xoo-induced genes is consistent with an ability to manipulate trafficking pathways that might result in nutrient export from xylem parenchyma cells into the nutrient-poor xylem, an ability Xoc may lack. This possibility aligns with the presumed role of the blight S gene OsSWEET11 as a sucrose exporter.
The extent to which TAL effectors account for the genome-wide differences in gene expression is uncertain. We observed previously that TAL effectors in Xoc and Xoo diversified subsequent to or in concert with divergence of the two pathovars [23], so it is tempting to assume a determinative role for TAL effectors in tissue specificity. However, despite our demonstration that 19 out of the 85 genes induced by Xoc are TAL effector targets, the numbers of identified targets, particularly for Xoo, are still too few to draw any conclusions from ontology enrichment analysis of just the targets. But note that targets of TAL effectors from each pathovar include one or more distinct transcription factor or putative transcription factor (bHLH family) genes: the ontology enrichment results just discussed might reflect a pervasive and determinative role of TAL effectors, through both direct and indirect effects on global gene expression patterns.
Genetic manipulation of host cells tailored to the different conditions in the mesophyll apoplast vs. the xylem is a compelling hypothesis, but one might expect some generic manipulation important for colonization both by Xoc and Xoo as well. Curiously, neither of the two genes targeted by TAL effectors from both pathovars, the OsHen1 RNA methylase gene Os07g06970 or the VQ domain containing protein gene Os02g15290, appears to play an important role in leaf streak, based on the observation that the corresponding TAL effector mutant strains M87 (tal1c) and M38 (tal5a) were not significantly reduced in virulence (Figure 4). Possible roles of these targets in bacterial blight remain to be tested.
Despite the fact that exactly half of the Xoc TAL effectors were found to activate specific targets, most of the Xoc TAL effectors appear not to play a significant role in virulence, raising the question why the pathogen harbors all 26. We screened over 150 pSM7 integrants of Xoc to find that ones showing significantly reduced lesion lengths when inoculated to rice cv. Nipponbare mapped exclusively to tal clusters 2 or 11 (Figure 4 shows representative mutants). We narrowed this further to those that affect tal2g or the two tal11 genes. We confirmed the virulence contribution of tal2g by complementation, but we did not do the same for the tal11 mutants, leaving open the possibility even that the reduced virulence of the tal11a and tal11b mutants was due to ectopic mutation. The lack of a detectable virulence contribution for most Xoc TAL effectors is not unlike observations for Xoo, in which TAL effectors contribute to virulence to different extents, with typically only one or two out of many per strain playing a major role [8], [12]. Three possible reasons for the phenomenon come to mind, none mutually exclusive. First, the non-contributing TAL effectors may be important in host genotypes other than Nipponbare, in which promoter polymorphisms can influence targeting, or in plants at different growth stages from the one we assayed, in which the physiological context might change the gene activation requirements for the development of leaf streak. Second, having many clusters of tal genes in the genome, even if most are inconsequential to infection, might provide a selective advantage over time by increasing the likelihood of recombination for adaptation to new host genotypes [45]. Finally, the contributions might be redundant, or subtle, similar to those of non-TAL type III effectors [46]. Though predicted, we confirmed no redundant targeting by Xoc TAL effectors. Rather, the functions of distinct targets could themselves be redundant or epistatic to one another, a scenario that would have escaped detection in our study, but again would provide a pathogen advantage in the face of host genotypic variation. Regarding subtle contributions of individual TAL effectors, they might collectively cause an essential perturbation.
The importance of Tal2g and the sulfate transporter gene it upregulates for lesion expansion and bacterial exudation is reminiscent of the phenotype associated with TAL effector Avrb6 of the cotton (Gossypium hirsutum) pathogen X. campestris pv malvacearum. Strains carrying the avrb6 gene cause larger water-soaked symptoms that correlate with more bacterial release to the leaf surface [6]. It has been proposed that bacterial exit and accumulation onto the leaf surface is advantageous as a means of dissemination, particularly for pathogens like Xoc that do not cause systemic infections [14], [47], [48]. AvrBs3 causes cell hypertrophy that may achieve this by reducing the volume of the apoplast and squeezing bacteria out to the surface, by inducing the pepper cell size regulatory gene UPA20 [14], [16]. PthA of X. citri triggers developmental changes that result in canker formation and rupture, releasing bacteria to the leaf surface [49]. Its target has not been reported. We have seen no evidence of hyperplasia or hypertrophy in available micrographs of Xoc infected rice leaves, nor in electron micrographs we have generated, and sulfate transporters are not known to regulate cell growth, but this possibility should be examined more closely in a future study. We hypothesize though that, as suggested by the effect of Avrb6 (the target of which is also yet to be reported), the enhanced watersoaking conferred by Tal2g upregulation of OsSULTR3;6 facilitates bacterial egress.
The rice cv. Nipponbare genome encodes 14 sulfate transporter genes phylogenetically divided into five groups [42], [50]. OsSULTR3;6 belongs to the less well characterized group 3 that includes five additional members. None of the additional members is induced by Xoc (i.e., they are absent from Table S1). A recent report demonstrated a role for the Arabidopsis group 3 sulfate transporter AtSULTR3;1 in pH-dependent sulfate uptake by chloroplasts [51]. The chloroplast is a main site for sulfate reductive assimilation for the synthesis of cysteine, which together with glutathione maintains the antioxidant capacity of the cytosol [52]–[54]. AtSULTR3;2, AtSULTR3;3 and AtSULTR3;4 also were shown to contribute [51]. In contrast, the last member of the group, AtSULTR3;5, is plasma membrane localized and cooperates in roots with the low affinity transporter AtSULTR2;1 under sulfur deficiency to increase sulfate uptake capacity for root-to-shoot vascular transport [55]. The Tal2g target OsSULTR3;6 is most similar to AtSULTR3;5 (57% identity) yet is expressed, in the absence of Xoc infection, primarily in seeds during later stages of seed development [42]. The physiological consequence of the recruitment of high OsSULTR3;6 expression to mesophyll cells by Tal2g is therefore challenging to predict. Given the M27 phenotype, an attractive hypothesis is that it alters antioxidant capacity, impinging on redox signaling to dampen defense and allow more rapid induction of watersoaking by the pathogen. Another possibility is that it enhances watersoaking more directly, either through a redox-controlled mechanism or simply by altering osmotic equilibrium.
In L. japonica, the group 3 sulfate transporter gene sst1, which is more similar to OsSULTR3;6 (56% identity) than to any other member of the gene family in rice, is essential for normal nodule development and symbiotic nitrogen fixation [56]. Its ortholog in poplar (Populus trichocarpa), PtSultr3;5, is among most highly induced transcripts during the establishment of symbiosis with the fungus Laccaria bicolor [57]. This gene is also strongly induced during both compatible and incompatible interactions with the poplar rust pathogen Melampsora larici-populina [58]. Whether the Tal2g target and these orthologs play analogous roles in such diverse plant-microbe interactions awaits in-depth functional analysis.
That a major S gene for leaf streak is a member of a large gene family recalls the situation in blight, in which five members of the large OsSWEET family can functionally substitute for one another as S genes, three so far have been shown to play that role, and distinct TAL effectors from multiple strains have been identified as the activators of two [11], [13], [19]. Whether any of the five other group 3 paralogs, or of the other 13 total members of the sulfate transporter gene family in rice can substitute for OsSULTR3;6, and whether any are actually targeted by other strains of Xoc is an important question. A tendency for S genes to be members of functionally analogous gene families would make sense from an evolutionary perspective, both for the advantage it would afford the pathogen by providing alternate targets should cis- (e.g. xa13) or trans- (e.g., Bs3) acting types of resistance be encountered, as well as the possibility it would afford the host to adapt through promoter mutation and resist targeting while maintaining essential functions. These processes might indeed drive one another [46]. On the other hand, if OsSULTR3;6 is uniquely capable among its paralogs of serving as an S gene, the likelihood of identifying moderately stable resistance by screening for or engineering promoter variants that retain endogenous function is increased.
OsSULTR3;6 was one of the most strongly induced and highly expressed genes in Xoc inoculated plants, as OsSWEET11 was in Xoo inoculated plants. These major S genes were exceptional, with the majority of TAL effector targets being induced moderately. The blight S genes OsTFIIAγ1 and OsTFX1, which contribute only moderately, were induced relatively weakly. Whether these differences reflect an evolutionary optimization of transactivation for major S genes, or gene specific differences in induction potential or optima, or chance, is unclear. The general pattern of relatively high basal expression and moderate fold increase across all identified TAL effector targets may be dominated by so-called collateral targeting inconsequential to disease and under no selection, and it suggests that TAL effectors may act as transcriptional enhancers more readily than as activators. However, the low variation in normalized expression levels for all targets at 96 h after inoculation suggests that on average, this enhancement is close to saturating.
We generally did not observe significant expression changes at early time points (i.e. 2 h and 4 h), possibly as a result of a low signal∶noise ratio caused by variation among the replicates, but expression of TAL effector targets generally increased steadily across the later time points. Though some genes were expressed at lower levels in Xoc- or Xoo-inoculated plants than in mock-inoculated plants at 96 h, no significant patterns of downregulation across all time points were observed. We tentatively conclude from these observations that TAL effectors of Xoc and Xoo do not significantly downregulate any genes in their host, despite the potential to do so through indirect effects, or theoretically, through nonfunctional binding that interferes with endogenous expression.
The average number of candidate EBEs in the rice promoterome, per TAL effector across all Xoc and Xoo TAL effectors, was 671. After excluding candidate EBEs in genes not upregulated after inoculation, that average dropped to 1.5, with some TAL effectors having none and some as many as seven. Nearly half of the filtered EBEs that were tested further (i.e., those for Xoc TAL effectors) were real. Thus, combining candidate EBE search results with global gene expression data is a robust and effective approach to identifying TAL effector targets.
Nevertheless, the method still yielded roughly as many false positives as true targets. Though upregulated during infection, false positives might include genes with EBEs that match but are inaccessible or in the wrong context to be functional, or genes with EBEs that score below the cutoff but are not actually sufficiently high affinity sites. In an attempt to decrease the number of falsely predicted targets and improve the efficiency of target identification in the future, we applied machine learning to our set of 24 real (Xoc and Xoo) and 20 falsely predicted (Xoc) targets (Table S7) using several characteristics of their candidate EBEs. The best classifier that resulted calls 22 of the real targets and three of the falsely predicted targets as real, for a recall of 0.92 and a precision of 0.88. Thus, it eliminates 85% (17/20) of the falsely predicted targets at a cost of less than 10% (2/24) of the real ones. The training set was relatively small, so these metrics may not hold strictly when applied to larger numbers of predicted targets, and even if they are relatively stable, if the goal is to comprehensively capture real targets, the classifier clearly can not be used as a strict filter. It is also important to remember that the classifier was trained only on EBEs that passed the score cutoff and were located in up-regulated genes, so performance metrics might not hold if the classifier is used on EBEs that do not meet these requirements. Rather, the probability this classifier provides should be used to prioritize already predicted targets for experimental validation (see Supporting Information for the Weka model file for the Naive Bayes classifier trained on all EBE features). Training on a greater number of targets as they are identified will improve both precision and recall, possibly even uncovering conditional relationships among characteristics of EBEs in real targets that the classifier currently calls incorrectly. With more targets, further comparison of classifiers built on subsets of EBE features might also reveal a smaller set of the most biologically relevant features that are sufficient to effectively discriminate real targets. Even with the small training set used here, the only slightly lower recall of the classifier based just on distance to transcriptional start site strongly suggests a major role for this feature.
As demonstrated by the results of our functional characterization of Tal2g EBEs and candidate EBEs (Figure 7), an important remaining challenge to eliminating false positives is a more nuanced understanding of TAL effector DNA binding. In particular, being able eventually to replace the RVD-nucleotide association frequency-based scoring matrix with one based on biochemically defined contributions of different RVD-nucleotide pairings, weighted to account for effects of position 5′ to 3′, will improve initial candidate EBE calling. Defining specificities for as many rare RVDs as possible will also be important to eliminate false positives and capture real targets for proteins like Tal2g. In this regard, we improved our predictions by substituting values of common RVDs for two rare ones, based on inference from structural data, and supported in the case of ‘SN’ by an experimental study [59].
Better understanding of TAL effector DNA interactions will also help eliminate false negatives. Without the scoring substitutions for the rare RVDs in Tal2g, one of its targets, the monocopper oxidase gene, would have been overlooked. Another example is suggested by the lack of identified targets for either Tal11a or Tal11b despite the reduced virulence of tal11 mutants (recalling however that complementation analysis was not performed to verify a role for either effector). A very low level of induction may be sufficient for function of some targets, such as an R gene like Xa27 [22] or an S gene that encodes a transcription factor, so false negatives could derive from a failure to detect differential expression in the initial transcript profiling experiment. A false negative could also result for a TAL effector with lax specificity. Xa27 again serves as an example. AvrXa27 contains at several positions an RVD with dual or no specificity; its EBE in Xa27 ranks 5,368th in the rice promoterome, nestled above the low-scoring outlier cutoff [26]. Exclusion of sites preceded by any base other than T, as specified in our search, might also pass over a real EBE. The TalC EBE in OsSWEET14, discussed in the introduction, is a salient if rare example. Two additional, theoretical examples are worth considering. The first is a gene whose expression is activated via read-through transcription by a TAL effector that targets a neighboring gene upstream. The second is a gene for which overall transcript levels do not change detectably, but which yields a unique alternative transcript when driven by the TAL effector due to TAL effector-dependent repositioning of the transcriptional start site [41], [60]. Transcript profiling by next generation sequencing of cDNA (RNAseq) [e.g., 61], in contrast to the GeneChip expression experiment that began this study, should provide the sensitivity to detect weakly expressed or weakly induced genes as well as alternative transcripts, to reduce or eliminate false negatives that might otherwise result. Regarding TAL effectors with lax specificity, EBEs with a non-canonical preceding base, and potential read-through targeting, adjusting EBE search parameters is a simple solution, but will unavoidably increase the number of false positives.
Given the current understanding of TAL effector function and the ability to predict binding sites using the code, we considered each gene that was activated by a TAL effector and that displayed a strong candidate EBE for that effector to be directly activated. Yet even meeting these criteria, it is formally possible that such a gene might be activated indirectly, i.e., in response to expression of another gene directly activated by the TAL effector. In pepper, prior to discovery of the code, direct targets of AvrBs3 were isolated by screening for transcripts whose upregulation by AvrBs3 occurs even in the presence of the eukaryotic translation inhibitor cycloheximide [14], [16]. To address the possibility that some of the Xoc TAL effector targets we identified in rice are indirect targets, using RT-PCR we tested Xoc-triggered transcript accumulation of the targets for sensitivity to cycloheximide, measured at 8, 16, 24, and 36 hr after co-infiltration (Figure S5). At the two earlier time points transcripts of all but one target accumulated similarly in response to Xoc with or without cycloheximide, and most showed identical patterns across all four time points. However, several showed distinct patterns of up and down regulation across the time points in response to cycloheximide treatment alone. Furthermore, cycloheximide treatment strongly and persistently upregulated three pathogenesis-related genes previously observed to be induced by biotic stresses, included as controls, and transiently induced a fourth. Induction of an additional control, Os05g42150, which is the most significantly Xoc-induced gene in our dataset (Table S1) and is not predicted to be a TAL effector target, was unaffected by cycloheximide at the two earlier time points and was slightly repressed at the later ones. The results overall thus reveal differing and confounding epistatic effects of cycloheximide treatment in rice that render conclusive identification of direct and indirect targets by this method impossible.
Regarding the single target showing repression of Xoc-induced transcript accumulation in the presence of cycloheximide, the monocopper oxidase gene Os06g46500, in addition to its being upregulated during infection and harboring an appropriately positioned strong candidate EBE for Tal2g, several other lines of evidence support it being a direct target. First, in the context of the Bs3 promoter, tested in N. benthamiana, that EBE is functional (Figure 7). Second, the pattern of induction of Os06g46500 by Xoc is rapid and robust, virtually identical to that of the SO4 transporter gene targeted by Tal2g (Figure S6) and similar to the patterns displayed by the verified Xoo TAL effector targets (Figure 2). Third, no other Tal2g target that might activate Os06g46500 was predicted other than the SO4 transporter gene, and activation of the SO4 transporter gene by dTALEs was not accompanied by activation of Os06g46500 (Figure 6B). Also, activation of Os06g46500 itself with a dTALE targeting a site in the vicinity of the native EBE (Figure 6) indicates that the promoter is not generally inaccessible to binding by a TAL effector. Finally, assuming that if not all, at least most of the targets in the training set for the classifier we generated by machine learning are direct targets, the leave-one-out validation tests showed that Os06g46500 shares the characteristics of those targets (including the previously confirmed targets of Xoo TAL effectors).
For Os06g46500, and the rest of the targets we identified here, assaying activation following disruption of the endogenous EBEs by site-directed mutagenesis would provide the most direct evidence for or against direct targeting, but such experiments are beyond the scope of the present study. Absent such data, it remains possible that some of the targets we identified are indirect ones. For the reasons detailed in the above two paragraphs, we conclude that this is unlikely, but to the extent that it were true, it would affect the utility of our predictive classifier.
Many crops, including rice, wheat, cotton, citrus, tomato, cassava, soybean, and others, suffer losses due to Xanthomonas spp. that deploy TAL effectors. We demonstrated here that TAL effector activity in bacterial leaf streak of rice is directly responsible for nearly a quarter of the gene activation detected during infection. Considering the likely downstream effects, the total proportion is certain to be even greater. Our study provides new insight into bacterial leaf streak of rice in relation to bacterial blight and identifies a major new S gene, but TAL effector target identification in several pathosystems is a critically important ongoing objective. Probing the diversity and functions of TAL effector activated S and R genes will expand our knowledge of disease and defense mechanisms, and our ability to exploit those mechanisms for effective disease control. Patterns of distribution of different S genes in diverse pathosystems might yet reveal causal distinctions between pathogens that colonize the mesophyll and those that invade the xylem. New targets will also refine our understanding of functional TAL effector-DNA interactions, improving our ability to use these proteins [62]. Though improvements can be made, and challenges remain, the overall combined transcriptomic and computational approach we successfully undertook constitutes a moderately high throughput method that can be applied to TAL effector target identification in many Xanthomonas-host interactions, particularly as more, complete pathogen and host genome sequences become available.
Oryza sativa ssp. japonica cv. Nipponbare plants were grown in LC-1 soil mixture (Sungro, Bellevue, WA) in PGC15 growth chambers (Percival) in trays approximately 60 cm below a combination of fluorescent and incandescent bulbs providing approximately 1,000 µmoles/m2/s measured at 15 cm, under a cycle of 12 h light at 28°C and 12 h dark at 25°C. Fertilizer (Peters Professional, St. Louis, MO) and iron chelate micronutrient (Becker Underwood, Ames, IA) were applied with watering every two days at 0.25 and 4.5 g/l, respectively, until the day before inoculation. Nicotiana benthamiana plants were grown in LC-1 in a PGC15 growth chamber at approximately 90 cm below the lights, under a cycle of 16 h light (fluorescent lighting at 22°C, and 8 h dark at 18°C, and fertilized using a surface amendment of Osmocote granules (ScottsMiracle-Gro, Maryville, OH).
Bacterial strains and plasmids used for this study are listed in Table S8. E. coli strains were grown in LB medium at 37°C and A. tumefaciens in YEP medium (10 g/l peptone, 10 g/l yeast extract, 5 g/l NaCl, 1.5% agar) at 28°C, and transformed by standard electroporation. X. oryzae strains were cultured in GYE (20 g/l glucose, 10 g/l yeast extract) at 28°C unless otherwise specified, and were transformed by electroporation as described previously [63], except that 1 µl (5 µg) TypeOne Restriction Inhibitor (Epicentre Biotechnologies, Madison, WI) was added prior. Antibiotics were used for selection as follows: ampicillin at 100 µg/ml, gentamycin at 25 µg/ml, kanamycin at 25 µg/ml, spectinomycin at 25 µg/ml, and tetracycline at 10 µg/ml for E. coli or 2 µg/ml for X. oryzae.
GeneChip data are available at the PLEXdb gene expression resource (www.plexdb.org) [68] under accession OS3 and at NCBI-GEO under accession GSE16793.
Promoter sequences, defined as the 1,000 bases upstream of the start codon, for the approximately 56,000 rice genes annotated in the MSU Rice Genome Annotation Project Release 7 (http://rice.plantbiology.msu.edu/) were searched using our previously published TAL effector-target scoring model [32], for the best-scoring site in each promoter for each of the 40 unique Xoo and Xoc TAL effectors. Scoring takes the sum of the negative log probabilities of the RVD-nucleotide pairings at a site, so a lower score is a better score. Sites were required to be directly preceded by a 5′ T. The scoring matrix was used as published and separately with the RVDs ‘SN’ and ‘YG’ assigned nucleotide association frequencies of ‘NN’ and ‘NG’, respectively (see [Results]). The distributions of the approximately 56,000 resulting scores for each TAL effector in each case were then used to calculate a cutoff for outliers, defined as the 25th percentile minus 1.5 times the interquartile range. Promoters were then rescanned to identify all sites scoring below (better than) that cutoff for each TAL effector, and those sites were retained as candidate EBEs. Finally, the list of candidate EBEs for each TAL effector was cross-referenced to the GeneChip expression data. Candidate EBEs in promoters of genes upregulated in Xoc- or Xoo-inoculated plants relative to mock were taken as predicted targets.
A library of tal gene knockout strains of Xoc was generated by transformation with the suicide (non-replicative) plasmid pSM7 (Figure 4A; Table S8) [69], pSM7 harbors a 4.5-kb PstI fragment containing all but the first 80 bp of the ORF of tal gene aB4.5 [12] with an insertion of the EZ-Tn5 <NotI/KAN-3> transposon (Epicentre) at bp 1,769 of the gene, in repeat number 9 of 17.5 (sequence available on request). The vector is pBluescript II KS(+) (Agilent Technologies, Santa Clara, CA). The transposon provides kanamycin resistance. Selection for this marker yields strains in which the cloned, mutated tal gene has undergone homologous recombination with an endogenous tal gene. Because the tal ORF is truncated at the 5′ end, either a single or double recombination that retains the transposon results in a tal gene knockout. Double recombination can knock out clustered tal genes. The 4.5 kb PstI fragment also includes the first 85 bp of the avrXa10 tal gene downstream of ab4.5, which might increase the likelihood of complex recombination. To determine the number of insertions per strain and to map insertions, genomic DNA was extracted using the GenElute Bacterial Genomic Kit (Sigma-Aldrich, St. Louis, MO). Strains with single insertions were identified by Southern blot using EZ-Tn5 <NotI/KAN-3> as a probe. Insertion endpoints were mapped by amplifying and sequencing the distal ends of 5′ and 3′ fragments flanking the transposon and extending outside the repeat region. Primers used for amplifying 5′ flanking DNA were forward primer p369 (5′-TTCTGfCCCGGACCCCAACCGGATAG), matching a conserved 5′ sequence in Xoc tal genes, and reverse primer p395 (5′-TCCCGTTGAATATGGCTCATAACACCCC), corresponding to the transposon. For the 3′ fragment, forward primer p397 (5′-GTCCACCTACAACAAAGCTCTCATCAACC), corresponding to the transposon, and reverse primer p398 (5′-TCCTCTTCGTTGAATGCC), matching a conserved 3′ sequence of Xoc tal genes, were used. Sequencing of the distal ends of the 5′ and 3′ amplicons (furthest from the repeat region) was done using tal gene plus-strand primer p396 (5′-ACCCCAACCGGATAGG) and p398, respectively. In all but a few cases, insertion endpoints were unambiguously identified by polymorphisms among the 5′ and 3′ sequences of the 26 Xoc tal genes and two tal pseudogenes (Figure 4).
Two micrograms of genomic DNA of X. oryzae pv. oryzicola strain BLS256 were digested with BamHI and separated in 1% agarose by electrophoresis. DNA fragments from 2 to 5 kb were gel purified and ligated into pBluescript II SK- (Agilent) previously digested with BamHI and dephosphorylated with alkaline phosphatase (CIP; New England Biolabs, Ipswitch, MA). The ligation reaction was then used to transform E. coli TOP10 cells, and colonies harboring TAL effector clones were identified by colony PCR using oligonucleotides p270 (GCCAAGTCCTGCCCGCG) and p271 (CCTCCAGGGCGCGTGC), which target the conserved 5′ region of Xanthomonas oryzae TAL effector genes. The tal gene fragments in these clones were tentatively identified based on size and 5′ and 3′ sequencing. Next, the corresponding SphI fragment of each tal gene, encoding the repeat region and short flanking sequences, was cloned into the tal1c backbone (i.e., lacking the corresponding SphI fragment) in the entry vector pCS466 [63] and confirmed by 5′- and 3′- sequencing with oligos p235 (GGAGGCCTTGCTCACGGATGC) and p236 (GGCCGGTGACAGCACGATCCG), respectively. For tal2g, tal4a, and tal8, BamHI fragments were cloned into pCS466 (cut with BamHI) instead, since those genes are each missing one of the SphI sites. The reconstituted genes in pCS466 were then recombined into the broad host-range destination vector pKEB31 [27], using Gateway LR Clonase (Life Technologies, Carlsbad, CA), for expression in Xanthomonas.
Xoc strains were cultured for 3–4 days on solid media then resuspended in 10 mM MgCl2 to OD600 = 0.5 (approximately 1×108 CFU/ml) and infiltrated into the abaxial surface of fully expanded leaves of 4-week old rice plants using a needleless syringe. 10 mM MgCl2 alone was infiltrated as the mock. Infiltrated tissue was collected at 48 h and RNA prepared using the RNeasy Mini Kit (Qiagen, Valencia, CA). Before elution, RNA was subjected to in-column digestion with the RNase-Free DNase Set (Qiagen). Two µg of total RNA were used for first-strand cDNA synthesis using SuperScript III reverse transcriptase (Life Technologies) and standard oligo dT20. Reverse transcriptase reactions were diluted 5 times and 1 µl was used as a template for PCR with Phire Hot Start II DNA polymerase (Thermo Scientific, Waltham, MA) together with transcript-specific oligonucleotides for 30 sec at 98°C, followed by 23–25 cycles (depending on transcript abundance) of 10 sec at 98°C, 5 sec at 60°C, and 10 sec at 72°C. The oligonucleotides used are listed in Table S9.
Rice leaves were inoculated by syringe infiltration as described above for RT-PCR. Virulence was quantified at the specified days after inoculation as lesion expansion, in mm, from the inoculation spot (Figure 4B). To measure bacterial populations, duplicate sets of three leaves per treatment per time-point were collected. One set was used to quantify total bacterial populations and the other to quantify surface bacteria. For total bacterial counts, 10 cm leaf sections centered on the infiltration spot or leaf sections as indicated in Figure 6D were cut into small pieces and ground thoroughly in 2 ml of water using a mortar and pestle. For surface bacteria, a leaf section encompassing the watersoaked area was washed with 50 µl of water twice and the wash diluted into 1 ml of water. Samples were thereafter diluted serially in sterile water and spotted on peptone sucrose agar (10 g/l sucrose, 10 g/l peptone, 1 g/l sodium glutamate, 1.5% agar) supplemented with cephalexin at 20 µg/ml. Plates were incubated at 28°C until appearance of single colonies, and colonies at the dilution they were first distinct were counted. For each replicate sample, eight such measurements were made. Results are displayed as the mean and standard deviation of all measurements for all replicates. Experiments were repeated at least three times with consistent results.
TAL Effector Targeter [32] was used to target designer TAL effectors (dTALEs) to the promoter regions of Os01g52130 and Os06g46500. dTALEs were assembled by golden gate cloning into the entry vector pTAL1 as described [27] and subsequently transferred to the broad host range destination vector pKEB31 [27] by Gateway LR Clonase (Life Technologies). RVD sequences if the dTALEs used in this are provided in Text S1.
GUS reporter assays were conducted in Nicotiana benthamiana leaves of five-week old plants (from the date of sowing) using the substrate 5-bromo-4-chloro-3-indoyl glucuronide (X-Gluc) as described [70], using three leaf discs from different plants per treatment, collected at 48 hours after infiltration of Agrobacterium. Experiments were repeated twice. Determination of total protein in sample extracts was performed using the Bradford assay kit (Bio-Rad). The vector for T-DNA delivery of avrBs3 under the 35S promoter was pGWB5-avrBs3 [40]. The equivalent construct for tal2g, pGWB5-tal2g, was made by replacing the ∼3.3 kb BamHI fragment of an avrBs3 clone in the entry vector pENTR-D (Life Technologies; gift of T. Lahaye, University of Munich) with the ∼3.2 kb BamHI fragment of tal2g, then moving the reconstituted tal2g equivalent gene to the binary destination vector pGWB5 [71] using Gateway LR Clonase (Life Technologies). The pGWB5 derivatives were introduced into Agrobacterium tumefaciens strain GV3101 by electroporation; transformants were selected with 25 µg/ml each of kanamycin and gentamycin. The reporter constructs were made by first PCR amplifying from a longer Bs3 promoter clone (gift of T. Lahaye) the AvrBs3-responsive 343 bp sequence upstream of the Bs3 start codon, using previously reported primers [70] and inserting it into the Gateway entry vector pCR8/TOPO-TA (Life Technologies). A single base substitution was then introduced by site directed mutagenesis (Agilent) to create an NcoI site 47 bp upstream of the native EBE for AvrBs3. Candidate Tal2g EBEs flanked upstream by 5 bp and downstream by 4 bp matching their native context were synthesized as double stranded oligonucleotides with NcoI overhangs (Text S1) and cloned into the NcoI site of the modified Bs3 promoter. Finally, the modified Bs3 promoter and derivatives were transferred into the binary GUS reporter vector pGWB3 [71] using Gateway LR Clonase (Life Technologies), and the resulting plasmids introduced into A. tumefaciens GV3101 as described above.
Both Naive Bayes and logistic regression classifiers were implemented using Weka 3.6.9 [72] with default options, which select the discrimination threshold that maximizes F measure. All classifiers were trained on the candidate EBEs in Table S7 that were determined to be either in real or falsely predicted targets (“Yes” or “No” in column P, “Verified”). Classifiers were trained using various subsets of the following features: relative score, actual score, rank, distance to TXS, distance to TLS, proximity to a TATA box, and distance to a Y Patch. If a predicted EBE was located in a promoter without a TATA box or without a Y patch, or with no annotated TXS, the value for that feature was considered missing and replaced with a question mark. All classifiers were evaluated using leave-one-out cross validation. The receiver operating characteristic curve and precision recall curve in Figure S3 were plotted using the ROCR package [73].
|
10.1371/journal.ppat.1004144 | Inactivation of Fructose-1,6-Bisphosphate Aldolase Prevents Optimal Co-catabolism of Glycolytic and Gluconeogenic Carbon Substrates in Mycobacterium tuberculosis | Metabolic pathways used by Mycobacterium tuberculosis (Mtb) to establish and maintain infections are important for our understanding of pathogenesis and the development of new chemotherapies. To investigate the role of fructose-1,6-bisphosphate aldolase (FBA), we engineered an Mtb strain in which FBA levels were regulated by anhydrotetracycline. Depletion of FBA resulted in clearance of Mtb in both the acute and chronic phases of infection in vivo, and loss of viability in vitro when cultured on single carbon sources. Consistent with prior reports of Mtb's ability to co-catabolize multiple carbon sources, this in vitro essentiality could be overcome when cultured on mixtures of glycolytic and gluconeogenic carbon sources, enabling generation of an fba knockout (Δfba). In vitro studies of Δfba however revealed that lack of FBA could only be compensated for by a specific balance of glucose and butyrate in which growth and metabolism of butyrate were determined by Mtb's ability to co-catabolize glucose. These data thus not only evaluate FBA as a potential drug target in both replicating and persistent Mtb, but also expand our understanding of the multiplicity of in vitro conditions that define the essentiality of Mtb's FBA in vivo.
| The development of new chemotherapies targeting Mycobacterium tuberculosis (Mtb) will benefit from genetic evaluation of potential drug targets and a better understanding of the pathways required by Mtb to establish and maintain chronic infections. We employed a genetic approach to investigate the essentiality of fructose-1,6-bisphosphate aldolase (FBA) for growth and survival of Mtb in vitro and in mice. A conditional fba mutant revealed that Mtb requires FBA for growth in the acute phase and for survival in the chronic phase of mouse infections. In vitro essentiality of fba was strictly condition-dependent. An FBA deletion mutant (Δfba) required a balanced combination of carbon substrates entering metabolism above and below the FBA-catalyzed reaction for growth and died in media with single carbon sources and in mouse lungs. Death of Δfba in vitro was associated with the perturbation of intracellular metabolites. These studies highlight how a conditional fba mutant helped identify conditions in which FBA is dispensable for growth of Mtb, evaluate FBA as a potential target for eliminating persistent bacilli and offer insight into metabolic regulation of carbon co-catabolism in Mtb.
| Metabolism is an important aspect of all host–pathogen interactions [1]. Mycobacterium tuberculosis (Mtb) has evolved to replicate and survive for decades within diverse host granulomas, including caseating, fibrotic and cavitating lesions, each providing a different environment [2], [3]. Mtb's ability to adapt to multiple environments may in part be attributed to its metabolic flexibility and its capacity to efficiently catabolize multiple carbon sources simultaneously [4]–[6]. Fatty acids and lipids are major carbon sources available to Mtb in vivo [7]–[10], but Mtb also has access to glucose and glycolytic 3-carbon compounds, requires trehalose import for virulence and can utilize CO2 as source of carbon [6], [11]–[13]. Mtb lacks classical carbon catabolite repression and it remains to be identified how co-catabolism of multiple carbon sources is regulated to achieve optimal growth [4].
Knowledge about Mtb's metabolism benefits the understanding of tuberculosis pathogenesis and can identify potential new targets for chemotherapeutic interventions, as Mtb mutants lacking metabolic enzymes are among the most attenuated in the mouse model of tuberculosis [8], [10], [14]–[17]. Fructose bisphosphate aldolase (FBA) is central to glycolysis and gluconeogenesis (Figure S1) and has been the focus of structural, enzymatic, and drug developmental studies [18]–[22]. Evidence supporting FBA's requirement for optimal growth came from the failure to isolate mutants with transposon insertions in fba, the gene encoding FBA, in standard in vitro culture conditions [23]. Also, previous attempts to delete fba from the Mtb chromosome by homologous recombination have failed [18], [19]. A conditional fba mutant generated in the attenuated H37Ra strain revealed that growth in either glucose- or succinate-containing media was strictly dependent upon induction of FBA expression, demonstrating that FBA is an essential enzyme for both glycolysis and gluconeogenesis [19]. Moreover, FBA is expressed in Mtb during mouse and guinea pig infections, and increased amounts were identified in culture filtrates during hypoxia, a condition which persistent Mtb is thought to encounter in the host [19], [24]. FBA has recently been shown to be secreted by Mtb and to bind host plasminogen, potentially enabling FBA to play a metabolism-independent role in host-pathogen interactions [19]. Importantly, Mtb FBA differs from human FBA by its reaction mechanism. FBAs catalyze the reversible cleavage of fructose-1,6-bisphosphate (FBP) to yield dihydroxyacetone phosphate (DHAP) and glyceraldehyde 3-phosphate (G3P). However, class I FBAs such as the human enzyme produce a Schiff-base reaction intermediate whereas class II FBAs, to which the Mtb enzyme belongs, require a divalent metal cation such as Zn2+ to stabilize the enolate intermediate [25]. Although some bacteria express both class I and class II enzymes, Mtb lacks an annotated class I FBA [19], [26]. Class I FBA activity in Mtb has been reported in earlier studies, but could not be detected by others [19], [27], [28]. The difference in catalytic mechanism of FBA from humans compared to that of Mtb enabled the design of bacteria-targeting class II-specific FBA inhibitors and work is ongoing to improve their efficacy [20], [22], [29], [30].
Our goal was to investigate the importance of Mtb's FBA in acute and chronic mouse infections and to further characterize the basis for its in vitro essentiality. To achieve these goals we generated Mtb strains in which FBA expression was tightly regulated by a tunable dual-control (DUC) genetic switch that combines transcriptional silencing and controlled protein depletion [31]. We found that essentiality of FBA was condition-dependent, and could be overcome by availability of two carbon sources entering metabolism above and below the FBA-catalyzed step. However this was dependent on the specific ratio of the glycolytic and gluconeogenic carbon sources, because metabolism of butyrate by a strain lacking fba was dependent on Mtb's ability to efficiently co-catabolize glucose. Mtb relied on FBA for both growth during acute and persistence during chronic mouse infections.
To investigate the role of FBA in vitro and in vivo we generated an Mtb strain in which expression of FBA is regulated by the recently described DUC switch [31], so that anhydrotetracycline (atc) or doxycycline (doxy) trigger transcriptional repression of the fba gene and simultaneous degradation of the FBA protein. We first introduced a second copy of fba with a strong promoter (Psmyc-fba) [32] via an integrative plasmid into the Mtb chromosome and then deleted the native copy. After confirming fba deletion by Southern blot (Figure S2), we generated FBA-DUC by replacing Psmyc-fba with a DAS+4-tagged fba gene, whose transcription was controlled by a TetR-regulated promoter. The DAS+4-tag allowed for proteolytic inactivation of FBA. Immunoblot analysis confirmed FBA depletion upon addition of atc (Figure S3). Similar to previous findings with an H37Ra FBA-TetON mutant [19], growth of H37Rv FBA-DUC with single carbon sources was inhibited when FBA was depleted by the addition of atc (Figure 1A–C). Unexpectedly, however, growth of FBA-DUC was unaffected by atc in media containing glucose and a second carbon source, such as butyrate, or glycerol (Figure 1D,E).
To assess if FBA is required in vivo, we infected mice with WT and FBA-DUC and monitored growth and survival in lungs and spleens (Figure 2). Mice infected with FBA-DUC were fed doxy-containing chow starting at the day of infection, on day 10, and on day 35 post-infection to determine whether FBA is required for growth and persistence of Mtb. FBA-DUC failed to replicate in lungs of mice fed with doxy starting on the day of infection and was undetectable by CFU determinations at day 10 post-infection (Figure 2A). No pathological signs of infection were observed in lungs of these mice (not shown) and no CFU were detectable in their spleens. When doxy administration was started later, CFU declined rapidly in lungs and spleens of both acutely and chronically infected mice. Decreases in CFU were accompanied by decreases in lung pathology (Figure 2B). In mice that did not receive doxy, FBA-DUC replicated and persisted similar to WT. Together these data establish that Mtb requires FBA activity for growth during acute and persistence during chronic mouse infections.
Growth of FBA-DUC with atc in dual carbon source media (Figure 1) suggested that FBA may not be essential in all conditions and that its essentiality may be carbon source-dependent. Therefore, we next attempted to delete fba by replacing the integrative, plasmid containing Psmyc-fba with a plasmid that did not contain fba. Δfba candidates grew readily on agar plates containing a carbon source combination - glucose and glycerol - that permitted growth of FBA-DUC with atc (Figure 3A). No Δfba candidates were obtained on standard Mtb agar plates with glycerol and Middlebrook OADC, a supplement containing oleic acid and glucose as carbon sources. However, doubling the glucose concentration resulted in growth of colonies albeit with slower growth rate than in the absence of oleic acid (Figure 3A). Southern blot and immunoblot confirmed that the Δfba candidates were true knockout strains (Figure S2, S3). As observed on agar plates, growth of Δfba in liquid culture was carbon source-dependent, although Δfba replicated in oleic acid-containing medium following a twenty-day lag phase (Figure 3B). This delayed growth was not from complete fba suppressor mutants because these bacteria were not capable of growth with a single carbon source (not shown). These experiments demonstrated that essentiality of FBA is conditional and suggested that Mtb lacking FBA (Δfba) has specific carbon source requirements.
Indeed, similar to the effects of conditional FBA-depletion, complete inactivation of FBA by deletion of fba caused a failure to replicate with a single carbon source but did not drastically affect growth with two carbon sources that enter central carbon metabolism above and below the FBA catalyzed step (Figure 4A). Growth of the complemented strain (Δfba-comp) was similar to that of WT in all conditions tested, demonstrating that the observed growth phenotypes were caused by the lack of FBA. CFU analysis of Δfba in single carbon sources revealed that Δfba died rapidly, with a 5 log decrease in CFU in 7 days and no detectable CFU by day 14 in media containing glycerol, butyrate, or acetate as sole carbon sources (Figure 4B, data not shown for acetate). Death in glucose-containing medium was slower, but also substantial. In contrast, Δfba survived with only a 10-fold decrease in CFU in base media without added carbon. Thus, the presence of a single carbon source resulted in killing of Δfba and the kinetics of death were dependent on the specific carbon source suggesting that the mechanism leading to death of Δfba in glucose may be different from that induced by glycerol or butyrate feeding below the FBA catalyzed step.
Δfba failed to replicate in resting bone marrow derived mouse macrophages (BMDM) in contrast to WT and Δfba-comp (Figure 5A). Activation of macrophages with IFNγ reduced replication of WT and Δfba-comp and did not affect survival of Δfba. In mice, Δfba did not establish an infection and was cleared from mouse lungs by day 10 post infection (Figure 5B). Thus, the metabolic environment in phagosomes of macrophages ex vivo and in mouse lungs did not support growth of Mtb in the absence of FBA and resulted in killing of Δfba in mice. The lack of death of Δfba in isolated macrophages suggests that the intracellular environment in macrophages ex vivo does not exactly mimic that faced by Mtb in macrophages inside the lung.
To better understand the impact of loss of FBA on Mtb metabolism we used liquid chromatography/mass spectrometry to measure metabolite levels in WT, Δfba, and Δfba-comp cultured on filters on top of agar plates containing 0.2% glucose, 0.2% glycerol, or a combination of 0.2% glucose and 0.2% glycerol (Figure 6). Following exposure to glucose alone, Δfba accumulated high levels of hexose-phosphate (P) and depleted its triose-P and phosphoenolpyruvate (PEP) pools. These metabolic changes are consistent with a defect in glucose metabolism due to the absence of FBA activity. Pentose-P and sedoheptulose-P pools were also significantly increased in Δfba suggesting either increased flux through the pentose phosphate (PP) pathway or decreased metabolism of PP pathway intermediates. In contrast, exposure to glycerol as sole carbon source resulted in depletion of hexose-P and sedoheptulose-P pools while triose-P levels were increased. The presence of both glucose and glycerol reversed most metabolic changes in Δfba except for the triose-P pool, which remained elevated, but not to the same extent as with glycerol as sole carbon source. These experiments demonstrate that metabolism of single carbon sources by Δfba significantly altered intracellular metabolite concentrations, which was accompanied by cell death (Figure 4). The buildup of PP pathway metabolites was consistent with increased flux into the PP pathway during culture on glucose; however, the PP pathway could not act as an efficient bypass to overcome loss of FBA.
Growth of Δfba required the presence of two carbon sources entering metabolism above and below the FBA catalyzed reaction and on agar plates was dependent on the amount of glucose in media containing glycerol and oleic acid (Figure 3). We therefore sought to investigate whether FBA might regulate a specific balance of glycolytic and gluconeogenic metabolism. WT Mtb replicated with glucose as sole carbon source and even low concentrations of butyrate enhanced its growth as previously reported [4] up to 0.4% when butyrate became toxic (Figure 7A). In contrast, when provided with 0.2% glucose, Δfba required at least 0.05% butyrate for growth, which was maximal with 0.1% but did not reach that of WT suggesting that in these conditions Mtb's capacity to metabolize butyrate is limited by FBA deficiency.
With a fixed concentration of 0.1% butyrate, WT growth increased with increasing amounts of glucose (Figure 7B). Growth of Δfba, however, plateaued with 0.025% glucose and was inhibited at concentrations above 0.2%. Inefficient glucose metabolism thus limited growth of Δfba in the presence of butyrate suggesting that FBA facilitates the efficient catabolism of butyrate by driving glycolysis. Together these data imply that the efficiency of glucose metabolism determines Mtb's capacity to co-catabolize butyrate.
Given the in vivo essentiality in acute and chronic mouse infections, our final goal was to evaluate to what degree FBA has to be depleted to result in growth inhibition and whether this is dependent on the available carbon source. FBA-DUC grew slower than WT in media with 0.4% glucose even in the absence of atc (Figure 1) and in glucose FBA-DUC was more sensitive to atc-induced growth inhibition than in butyrate-containing media (Figure S4). The reduced growth in glucose-containing, atc-free media (Figure 1,8A) was likely due to the low FBA amounts expressed in FBA-DUC, which were reduced by approximately 87% compared to WT even in the absence of atc (Figure 8B). Addition of 6.3 ng/ml atc depleted FBA by approximately 97% and abolished growth of FBA-DUC in glucose containing medium. In contrast, depletion by 87% or 97% was insufficient to reduce growth of FBA-DUC in medium with 0.1% butyrate as the sole carbon source. Addition of 25 ng/ml atc, which reduced FBA expression below the level of detection, was required to inhibit growth with butyrate. Thus, Mtb was more sensitive to FBA depletion when grown with glucose as sole carbon source than when metabolizing butyrate emphasizing that vulnerability to depletion of an essential protein can be dependent on the growth condition.
The experiments reported here enhance our understanding of Mtb carbon metabolism and are relevant to tuberculosis drug development. Mtb fba has previously been shown to be required for growth on standard media [19], [33] and FBA inhibitors are under development [22], [29], [30], but to our knowledge no FBA-specific inhibitor has been effective at inhibiting growth of live Mtb. Furthermore, it was unknown whether FBA is essential for growth or survival of Mtb in vivo. Using a genetic approach we evaluated FBA as a potential drug target by assessing its essentiality for growth and persistence during mouse infections. Depletion of FBA after establishment of chronic infection led to complete clearance of viable Mtb in mouse lungs and spleens, which promotes FBA as a potential therapeutic target for killing of persistent bacilli.
Unexpectedly, FBA was not essential in all conditions in vitro, and a conditional fba mutant was critical to identify conditions in which Δfba could be isolated. This deletion mutant was able to replicate when provided with combinations of carbon substrates entering metabolism above and below the FBA-catalyzed reaction. Thus, it will be crucial to determine the efficacy of FBA inhibitors against live Mtb in carbon conditions where Mtb is vulnerable to FBA inactivation in contrast to standard mycobacterial liquid culture medium, which contains a combination of glucose and glycerol making FBA dispensable.
Earlier studies have highlighted Mtb's ability to co-catabolize multiple carbon sources in vitro and in macrophages [4]–[6]. This is in contrast to other bacteria such as E. coli, which exhibits diauxic growth in the presence of multiple carbon sources resulting from their sequential metabolism [34], [35]. Here, we provide further evidence for Mtb's co-catabolism and offer insight into its metabolic regulation. Growth of WT Mtb on glucose and on butyrate, which requires β-oxidation for conversion into acetyl-CoA mimicking the fate of long chain fatty acids, was enhanced in a dose responsive manner by addition of the respective other carbon source. Growth of Δfba only occurred with two carbon substrates, feeding into either side of the FBA reaction. Additionally, deletion of FBA rendered Mtb extraordinarily sensitive to the relative concentrations of carbon sources in the medium. In media containing glucose and butyrate, Δfba was unable to efficiently utilize glucose to enhance its growth while growth of WT scaled with glucose. Metabolism of increasing amounts of butyrate was also restricted shown by Δfba's limited growth despite the presence of 0.2% glucose. FBA thus facilitates efficient butyrate catabolism through its ability to metabolize glucose and FBA inactivation resulted in suboptimal co-catabolism of glucose and butyrate.
The mechanism of death of Δfba in carbon-unbalanced media remains to be determined but may be due to buildup of FBA substrates at higher concentrations of glucose or butyrate, as well as regulatory mechanisms allowing Mtb to sense carbon levels and respond accordingly. Indeed, the inability of Δfba to grow on standard Mtb plate medium containing glucose, glycerol and oleic acid was overcome by the addition of 0.2% additional glucose, which was not required for growth in nearly identical medium lacking oleic acid.
Inactivation of FBA in E. coli prevented growth with glucose but not with glycerol and succinate as carbon substrates [36], [37]. Accumulation of sugar phosphates including the FBA substrate fructose-1,6-bisphosphate (FBP) is thought to be the cause of this growth inhibition, which is supported by suppressor mutations that prevent FBP buildup and allow growth on glucose [38]. With our metabolite extraction protocol we did not detect FBP, however other sugar phosphates including hexose-P, sedoheptulose-P and pentose-P increased dramatically in Δfba cultured on glucose, while triose-P and PEP accumulated on glycerol. The accumulation of these metabolites was not observed when both glucose and glycerol were available to Δfba, except for the triose-P pools, which remained elevated but much less than in the glycerol condition. It is thus possible that the accumulation of phosphorylated metabolites is cause for the death of Δfba in conditions where glycolytic and gluconeogenic carbon flow is unbalanced, such as with single carbon substrates or in mixes with an abundance of a glycolytic or gluconeogenic substrate.
How potent would an FBA inhibitor need to be to affect Mtb growth? We measured the vulnerability of Mtb to FBA depletion in single glycolytic and gluconeogenic carbon sources where FBA is essential. Although depleting FBA by 97% did not affect growth with butyrate as sole carbon source, it prevented growth with glucose and 87% inhibition was sufficient to reduce growth with glucose. These data reveal a differential susceptibility of Mtb to FBA depletion, depending on the available carbon source. Antibiotic targets vary widely in how much inhibition is required to stop replication or induce death and not all effective drug targets are highly vulnerable to inhibition [39]. Specific target-drug interactions can contribute to the efficacy of a compound. We cannot predict how effectively FBA must be depleted to abolish replication in vivo, but the WT-like phenotype of FBA-DUC without doxy suggests that 13% of WT FBA levels are sufficient for normal growth and persistence in mice. The fast killing of Mtb following further FBA depletion in mouse lungs and spleens suggests that it is an effective target during acute and chronic mouse infections. During infections FBA may also have an additional, metabolism-independent function in Mtb's interaction with the host as it can be secreted and bind to human plasminogen [19].
The requirement of FBA for growth and persistence in mice suggests that in vivo Mtb either faces single carbon sources or lacks access to the growth permissive ratio of carbon sources that can compensate for the lack of FBA. This is likely due to an abundance of fatty acids and lipids which are predominant carbon sources available to Mtb in vivo [7]–[10]. It is unknown if FBA is essential for growth and persistence of Mtb in humans; experimental animal models that more closely mimic human TB pathology [40] would help address this question. Given that human granulomas contain lipid-rich foamy macrophages and build up lipids in the caseum [41], [42], it is plausible that Mtb requires FBA also during human infection.
Mouse studies were performed following National Institutes of Health guidelines for housing and care of laboratory animals and performed in accordance with institutional regulations after protocol review and approval by the Institutional Animal Care and Use Committee of Weill Cornell Medical College (protocol # 0601-441A, Conditional Expression of Mycobacterial Genes).
M. tuberculosis H37Rv strains were grown in Middlebrook 7H9 liquid media containing 0.5% bovine serum albumin fraction V, 0.2% glucose, 0.2% glycerol, 0.085% NaCl, and 0.05% tyloxapol without shaking in 5% CO2 at 37°C. For carbon-defined growth curves, Mtb was cultured in Sauton's base media modified to be carbon-limited, containing 0.05% potassium phosphate monobasic, 0.05% magnesium sulfate heptahydrate, 0.2% citric acid, 0.005% ferric ammonium citrate, 0.05% ammonium sulfate, 0.0001% zinc sulfate, and 0.05% tyloxapol at pH 7.4. For solid media, Middlebrook 7H10 media with 0.5% glycerol and 10% Middlebrook OADC supplement (final concentration of 0.5% bovine serum albumin fraction V, 0.2% glucose, 0.085% NaCl, 0.006% oleic acid, 0.0003% catalase) or self-made ADNaCl (final concentration of 0.5% bovine serum albumin fraction V, 0.2% glucose, 0.085% NaCl) was used. Carbon sources glucose, glycerol, sodium acetate and butyric acid, were added at indicated concentration (%wt/vol or %vol/vol, depending on stock). When appropriate, hygromycin B (50 µg/ml), streptomycin (10 µg/ml), kanamycin (25 µg/ml), and/or zeocin (25 µg/ml) were added. Anhydrotetracycline (atc) was added at the indicated concentrations and replenished at half the initial concentration in liquid cultures every 4–5 days for growth curves but not vulnerability assays. For survival assays, bacterial culture samples were taken from growth curve cultures at the time-points indicated and plated for CFU. For metabolomic profiling, Mtb was seeded at OD580∼1 on 0.22 µM nitrocellulose filters (1 ml per filter) and cultured on Middlebrook 7H10 agar medium containing 0.5% bovine serum albumin fraction V, 0.085% NaCl, 0.2% glucose, and 0.2% glycerol for 5 days. Filters were then transferred to similar plates with defined carbon sources: 0.2% glucose, 0.2% glycerol, or both together, each at 0.2%. Mtb was harvested 24 hours later by metabolic quenching in cold acetonitrile∶methanol∶H2O (40∶40∶20) and mechanically lysed using a bead beater as described [4], [10].
M. tuberculosis H37Rv was transformed with a plasmid expressing fba under the control of a strong promoter P1 (Psmyc [32]) that integrates into the phage attL5 site in the Mtb genome. In this fba merodiploid strain deletion of native fba was achieved by allelic exchange using specialized transducing phage phAE87 as previously described [10]. After confirming removal of native fba by Southern blot, replacement transformations of the attL5 insets were performed to generate FBA-DUC and Δfba. In FBA-DUC, fba contained a C-terminal DAS+4 tag and a plasmid in the phage tweety site that allowed inducible expression of SspB as described [31]. In Δfba the attL5 site carries a kanamycin-resistant plasmid not containing fba. All plasmids were constructed using Gateway Cloning Technology (Invitrogen) using BP and LR recombinase reactions following the manufacturers instructions. The complemented mutant is Δfba transformed with a plasmid that integrates in the attL5 site expressing fba from the P1 promoter.
M. tuberculosis metabolites were separated and detected in a Agilent Accurate Mass 6220 TOF coupled to an Agilent 1200 Liquid Chromatography system using a Cogent Diamond Hydride Type C column (Microsolve Technologies) using solvents and configuration as described [43]. Metabolites were quantified by standard curves generated with authentic chemicals spiked into homologous mycobacterial lysates. Quantified metabolite concentrations were normalized to bacterial biomass of individual samples determined by measuring residual protein content (BCA Protein Assay kit; Pierce).
Protein extracts were prepared from bacterial pellets from 30 ml cultures at indicated time points in specified media. Briefly, cultures were washed with phosphate buffered saline (PBS), 0.05% Tween 80 and resuspended in 1 ml PBS, 1× protease inhibitor cocktail (Roche). Cells were lysed by bead beating three times at 4500 rpm for 30 seconds with 0.1 mm Zirconia/Silica beads. Beads and cell walls were removed through centrifugation (11000× g/10 min/4°C) and the supernatant was filtered through a 0.2 µm SpinX column (Corning). Lysate protein concentrations were determined using a DC Protein Assay Kit (Bio-Rad). For immunoblots 20-0.02 µg protein extracts were separated by SDS-PAGE, transferred to nitrocellulose membranes and probed overnight with rabbit antisera FBA [19] (1∶2000 dilution in 1∶1 PBS/Odyssey Blocking Buffer (LI-COR Biosciences), 0.1% Tween20) or PrcB (1∶18,000 dilution in 1∶1 PBS/Odyssey Blocking Buffer (LI-COR Biosciences), 0.1% Tween20). As secondary antibody IRDye 680 Donkey anti-Rabbit IgG(H+L) (LI-COR Biosciences) was used. Proteins were detected using the Odyssey Infrared Imaging System (LI-COR Biosciences).
Female C57BL/6 mice (Jackson Laboratory) were infected by aerosol using an inhalation exposure system (Glas-Col) and early-log-phase M. tuberculosis cultures as single-cell suspensions in PBS to deliver 100 to 200 bacilli per mouse. Doxycycline containing food (2000 ppm, Research Diets) was given to mice starting at the indicated time-points. Serial dilutions of lungs and spleens homogenates were cultured on 7H10 plates containing ADNaCl to determine CFU at the indicated time points. The left lobe of each lung was fixed in 10% buffered formalin, further processed for histopathology and stained with hematoxilyn and eosin. We isolated and infected bone marrow-derived mouse macrophages as previously described [44].
|
10.1371/journal.ppat.1007491 | EspL is essential for virulence and stabilizes EspE, EspF and EspH levels in Mycobacterium tuberculosis | The ESX-1, type VII, secretion system represents the major virulence determinant of Mycobacterium tuberculosis, one of the most successful intracellular pathogens. Here, by combining genetic and high-throughput approaches, we show that EspL, a protein of 115 amino acids, is essential for mediating ESX-1-dependent virulence and for stabilization of EspE, EspF and EspH protein levels. Indeed, an espL knock-out mutant was unable to replicate intracellularly, secrete ESX-1 substrates or stimulate innate cytokine production. Moreover, proteomic studies detected greatly reduced amounts of EspE, EspF and EspH in the espL mutant as compared to the wild type strain, suggesting a role for EspL as a chaperone. The latter conclusion was further supported by discovering that EspL interacts with EspD, which was previously demonstrated to stabilize the ESX-1 substrates and effector proteins, EspA and EspC. Loss of EspL also leads to downregulation in M. tuberculosis of WhiB6, a redox-sensitive transcriptional activator of ESX-1 genes. Overall, our data highlight the importance of a so-far overlooked, though conserved, component of the ESX-1 secretion system and begin to delineate the role played by EspE, EspF and EspH in virulence and host-pathogen interaction.
| Mycobacterium tuberculosis is the etiological agent of human tuberculosis, a life-threatening disease which has seen a recrudescence in the last decades due to the spread of drug-resistant bacterial strains and to co-morbidities such as HIV and diabetes. To develop effective treatment and limit bacterial dissemination within and outside the host, it is pivotal to improve our understanding of the strategies used by the pathogen to colonize the host and subvert the immune defenses. The ESX-1 secretion system represents a key player in these processes. Here we show that the EspL protein, encoded by the ESX-1 gene cluster, is essential for bacterial virulence and for stabilizing the abundance of the EspE, EspF and EspH components of the ESX-1 system. Tubercle bacilli lacking EspL cannot multiply inside macrophages, do not secrete the major virulence factor EsxA and fail to trigger the ESX-1 dependent innate immune response. EspL is thus an important but so-far neglected contributor to ESX-1 function.
| Mycobacterium tuberculosis, the etiological agent of human tuberculosis, is arguably the world’s most successful human pathogen. It is estimated that one third of the world’s population is latently infected by the bacterium [1], which can survive in a dormant state inside specialized cellular structures in the lung parenchyma called granulomas [2,3]. As a consequence of immunodeficiency or co-morbidities, like HIV or diabetes [4,5], latent M. tuberculosis can reactivate and establish an acute infectious process which leads to the disease. Host-pathogen interaction and disease progression are mediated by various virulence factors encoded by the bacterial genome, the most important of them being the ESX-1 or type VII secretion system [6].
ESX loci are characterized by genes encoding small secreted proteins with a conserved tryptophan-x-glycine (WXG) motif and by transmembrane ATPases belonging to the FtsK-SpoIIIE-like family [7,8]. Five ESX systems, implicated in different functions, exist in M. tuberculosis [9]. The ESX-1 cluster comprises approximately twenty genes and encodes a specialized secretion apparatus, which releases effectors into the extracellular milieu. The relevance of the ESX-1 genes in mycobacterial physiology was recognized when attenuation of the vaccine strain M. bovis BCG and of the vole bacillus M. microti was associated with their partial deletion [10–14]. Since then, the role played by ESX-1 in cytosolic recognition and stimulation of innate immunity [15–17], phagosomal rupture and bacterial escape [18,19], intercellular spread and systemic disease [20,21] has been the object of numerous studies. Recently, the ESX-1 secretion system has also been considered as a potential drug target for the development of anti-virulence drugs [22].
Considerable progress has been made in understanding how the system works and is regulated. Electron microscopy-based studies showed that the M. xenopi ESX-5 core membrane complex is composed of four proteins (EccB5, the ATPase EccC5, the putative channel EccD5, and EccE5) which assemble into an oligomer with a six-fold symmetry [23]. However, it is still unknown how the secreted substrates can cross the mycobacterial external membrane, or mycomembrane, although the involvement of EspC, encoded by the distal espA-espC-espD locus, has been hypothesized in M. tuberculosis [24]. The current model for ESX-1 activity proposes heterodimeric and co-dependent complexes as the secreted substrates, i.e. EsxA/EsxB and EspA/EspC [25,26]. These are targeted to the inner membrane apparatus by a bipartite secretory signal composed of the WXG motif on the first member of the dimer and of a tyrosine-x-x-x-aspartic acid/glutamic acid (YxxxD/E) motif on the second [26,27]. ESX-1 function undergoes transcriptional regulation, exerted by EspR [28], Lsr2 [29], CRP [30], MprA [31] and mIHF [32] on the espA-espC-espD locus. Recent studies showed that WhiB6, a redox sensor protein, directly controls expression of genes associated with the ESX-1 secretion system in M. marinum, such as espA, espE and eccA1, and that this regulation is strictly dependent on its Fe-S cluster [33]. In M. tuberculosis, whiB6 is part of the PhoP regulon [34] and divergently contributes to ESX-1 gene transcription in the H37R strains compared to other isolates [35]. Additional, post-transcriptional, control of secretion activity is carried out by the serine protease MycP1 through proteolytic cleavage of another ESX-1 substrate, EspB [36].
Here, we investigate the role of a previously overlooked ESX-1 component, EspL, in M. tuberculosis. We demonstrate that it is essential for mycobacterial replication inside macrophages, for eliciting innate cytokine production and for stabilizing the protein levels of the additional ESX-1 members EspE, EspF and EspH.
In order to evaluate the role of EspL in M. tuberculosis virulence and ESX-1-dependent secretion activity, construction of an unmarked deletion mutant was planned. The transcriptional profile of the H37Rv genomic region that includes espL was carefully considered to avoid polarity on the downstream gene espK. Studies by Cortes and colleagues [37] demonstrated the presence of a polycistronic RNA that covers mycP1, eccE1, espB and espL, but not espK, which is transcribed independently (S1A Fig). Additionally, sequence inspection revealed that the GTG translational start codon of espL overlaps the stop codon of the preceding gene espB. The pJG1100-derived suicide vector [38] was then constructed according to this pre-existing information and the espL coding sequence (CDS) was deleted from the chromosome by allelic exchange, from coordinate 4,360,199 to coordinate 4,360,543, thereby leaving the espB stop codon intact (S1B Fig). The resulting strain, named ΔespL, was validated by immunoblot (Fig 1A) and whole genome sequencing. The latter technology identified one single nucleotide polymorphism (SNP) in gene rv1403 (353T>C), which caused substitution of Val118 with Ala, and one SNP in ethA (A to G transition at position 368) which resulted in replacement of His123 with Arg. No other differences were noted upon comparison with the genome sequence of the parental strain, except for the intended deletion. ΔespL was transformed with the complementing plasmid pGA44-espL, which carries the espL gene under the control of the PTR promoter, or with the empty vector pGA44 as a control [39]. Immunoblot experiments proved that expression of EspL was restored in the complemented strain ΔespL/pGA-espL, whereas no band was detected in the control ΔespL/pGA44 (Fig 1A). Consistent with these findings, qRT-PCR showed that the espL mRNA was expressed at a level similar to that of the wild type when the gene was provided in trans (Fig 1B). Transcriptional analysis was extended to include espB, espK and esxA. While espB and esxA mRNA levels were not altered significantly by the mutation introduced or by the ectopic expression of espL, the amount of espK transcript was 2.5-3-fold higher in the mutant strain (Fig 1B).
The ΔespL mutant did not show any major difference as compared to the wild type strain during in vitro growth in standard medium (S2A Fig). However, infection of THP-1 cells demonstrated severe reduction of the cytotoxicity of the mutant, which allowed cell survival to a similar extent as upon infection with the ΔΔRD1 strain, which lacks the extended RD1 region [40] (Fig 2A). Importantly, expression of espL by pGA44 complemented the phenotype to wild type levels (Fig 2A). These findings were further confirmed by colony forming unit (CFU) enumeration. While all of the strains were equally phagocytosed by THP-1 cells (S2B Fig), a major increase in the number of intracellular bacteria over one week was reported when H37Rv and the complemented strains, but not ΔespL, were used for infection (Fig 2B).
The crucial role played by the ESX-1 secretion system in inducing expression of cytokines of the innate immunity pathways was previously illustrated. In particular, EsxA secretion was found to be responsible for activating the cytosolic surveillance pathway based on cGAS-dependent sensing of DNA and the inflammasome [15–17]. Here, markedly reduced production of the pro-inflammatory cytokine IL-1β (Fig 2C), as well as decreased expression of type I interferon gene IFNB1 (Fig 2D), interferon-stimulated gene ISG15 (Fig 2E), and interleukin gene IL6 (Fig 2F), were noted after THP-1 infection by the ΔespL strain. On the contrary, H37Rv and the complemented strain ΔespL/pGA-espL elicited production of cytokines belonging to both the cGAS-STING-type I Interferon (IFN) and inflammasome axes. Overall, these data indicate that EspL is a key player in M. tuberculosis virulence, interaction with the immune system and, likely, in ESX-1 secretion, as explained below.
The secretion profile of the mutant was examined in parallel to that of the wild type H37Rv and of the complemented mutant strains. While the proteins were produced by all of the strains (Fig 3A), EsxA and EsxB were not detectable in the secreted fraction by immunoblot when EspL was missing, whereas EspA and EspD levels were greatly reduced (Fig 3B). As a consequence of the compromised secretion, accumulation of EsxA and EsxB occurred inside ΔespL cells (Fig 3A). Interestingly, EspB was found to be released into the culture supernatant in the absence of EspL (Fig 3B and S3 Fig), confirming that its secretion is not dependent on EsxA, EsxB, EspA or EspD, as reported earlier [41]. Therefore, the severe attenuation of ΔespL, described above, correlates with lack of secretion of the major virulence factor EsxA.
In-depth analysis of the culture filtrates of strains H37Rv, ΔespL, ΔespL/pGA-espL was performed by mass-spectrometry (S4 Fig, S2 and S4 Tables). While confirming the findings described above, these additional experiments showed that EspE, EspF and EspH (and other ESX-1 substrates such as EspC) are underrepresented in the secreted fraction of the mutant strain, indicating that ESX-1-dependent secretion activity is affected in ΔespL.
To gain insight into EspL function, the localization of the protein in sub-cellular fractions was studied. Total extracts from strain H37Rv were separated into cytosolic, membrane and capsular proteins, in addition to culture filtrate preparations. Anti-EspL antibodies identified a protein, with the apparent molecular weight of EspL, mainly in the cytosol and, to a lesser extent, in the membrane. However, partial contamination of the membrane fraction could not be excluded (S5 Fig). EspL was undetectable in the culture filtrate. Control antibodies against RpoB, Rv3852 and EsxB recognized their cognate antigens in the cytosolic/membrane, membrane only or cytosolic/secreted fractions, as expected [42]. EspL could thus exert its function in the cytosol or as a membrane-associated protein.
Since PFAM [43] and recent experimental work [44] predicted the presence of an YbaB-type DNA-binding domain [45–47] in EspL, we hypothesized that the protein may influence gene expression through binding to DNA. The transcriptome of the mutant strain was then analyzed and compared to that of the wild type by RNA-seq. Despite the low cut-off value (2-fold change), none of the M. tuberculosis genes was found to be deregulated in ΔespL, except for whiB6, whose expression level was decreased by 3-fold on average (S1 Table), and espL itself, which was not detected in the knock-out mutant. Genes belonging to the ESX-1 cluster, as well as genes which are part of other ESX loci (ESX-2 to ESX-5) were expressed at similar levels in the mutant as compared to the wild type. Curiously, genes that were reported to be included in the WhiB6 regulon in the related species Mycobacterium marinum [33] were not found to be deregulated by RNA-seq in M. tuberculosis ΔespL (S1 Table). These findings were confirmed independently by qRT-PCR, which also proved that re-introduction of espL into the complemented strain was necessary and sufficient to restore espL and whiB6 mRNA levels to normal (S6 Fig). Thus, EspL seems to control expression of whiB6 either directly or indirectly.
Intrigued by the discoveries reported earlier, we examined the impact of constitutive expression of whiB6 in the ΔespL mutant. Levels of whiB6 mRNA were increased by approximately 4-fold in the ΔespL strain carrying pGA-whiB6, as compared to wild type. All of the tested genes (i.e. espB, esxA, espE, espF, espH and espA), which are part of different transcriptional units [37,48], were induced (Fig 4A), therefore indicating that WhiB6 works as an activator of ESX-1 genes in M. tuberculosis. However, despite the increased expression of virulence-related genes, ΔespL/pGA-whiB6 displayed the same attenuation as ΔespL, ΔespL/pGA44 and ΔΔRD1 (Fig 4B). In other words, the lack of cytotoxicity caused by deletion of espL could not be bypassed by ectopic over-expression of whiB6. EspL is therefore essential for M. tuberculosis virulence.
Inspired by the work of Stoop and colleagues [49], we thoroughly analyzed the proteome of the espL knock-out mutant and compared it to that of the wild type, of the complemented strain and of the mutant expressing whiB6 in trans. Results are reported in Fig 5, Fig 6, S7 Fig, S3 Table and S5 Table. As expected, EspL was detected in the wild type and in the complemented strains only (Fig 5A). On the other hand, WhiB6 levels could only be measured in the strain over-expressing whiB6, indicating that this transcriptional regulator is poorly expressed in wild type conditions (Fig 6E). No significant difference was noted for EspB, EspA, EspC and EspD in the proteome of ΔespL (Fig 6D, 6A, 6B and 6C), whilst a small though statistically valid increase in EsxA and EsxB levels was reported (Fig 5B and 5C), thus corroborating the data obtained by immunoblot (Fig 3B). The most relevant variation in protein levels was noticed for EspE, EspF and EspH. Their abundance was greatly reduced in the mutant strain and complemented to wild type levels in ΔespL/pGA-espL (Fig 5D, 5E and 5F). Proteomic data contrasted with RNA-seq and qRT-PCR results, which proved that espE, espF and espH mRNAs in the espL knock-out strain were unaltered compared to the wild type (S1 Table and S6 Fig).
EsxA, EsxB, EspB and EspE amounts increased when whiB6 was provided ectopically (Figs 5B, 5C, 6D and 5D), thus reflecting the qRT-PCR assay in Fig 4A. However, EspF and EspH levels did not reach those of H37Rv upon WhiB6 expression in ΔespL (Fig 5E and 5F), in contrast to their transcripts which were induced by WhiB6 (Fig 4A). These results suggest that an additional, post-transcriptional control regulates the abundance of EspF, EspH and, most likely EspE, in M. tuberculosis. In the latter case, transcriptional upregulation bypassed the destabilization provoked by lack of EspL in strain ΔespL/pGA-whiB6, thus generating increased protein abundance. This did not happen in the case of EspF (Fig 5E) and EspH (Fig 5F). The espE cistron is probably more efficiently translated as its gene is the first in the operon and therefore less subject to mRNA degradation. EspL contribution is more evident for EspF and EspH, as the transcriptional increase caused by WhiB6 is not sufficient to avoid the destabilizing effect on the proteins provoked by lack of EspL.
Further confirmation to these findings was obtained by constructing strains that constitutively expressed HA-tagged EspE in the H37Rv and ΔespL backgrounds. Complementation by espL and expression of whiB6 were achieved by using the physiological mycP1 promoter, which represents the natural promoter of espL according to Cortes and colleagues [37]. Expression of espE.HA was therefore independent from EspL and from WhiB6 as it was placed under control of the PTR promoter [39]. The wild type phenotype was restored in strain ΔespL/pGA-espE.HA + espL, as shown in S8A and S8B Fig, thereby confirming that expression of EspE.HA did not cause abnormal behavior. In line with what has been described before, expression of whiB6 did not complement the lack of virulence in ΔespL/pGA-espE.HA + whiB6 (S8B Fig) but did increase the transcriptional levels of espE and esxA (S8C Fig), the latter result mirrored by the detection of a strong signal for the EsxA protein in Fig 7. Transcription of espE.HA was measured by qRT-PCR and confirmed as constitutive, almost identical in all of the strains, independently of the presence or absence of EspL and WhiB6 (S8A and S8C Fig). However, immunoblot analysis demonstrated that EspE.HA amounts in ΔespL/pGA-espE.HA were dramatically reduced (Fig 7), despite the presence of the espE.HA transcript (S8A Fig). Conversely, ΔespL/pGA-espE.HA + espL (expression of espL in trans) produced levels of EspE.HA equal to those in H37Rv/pGA-espE.HA. Providing whiB6 only (ΔespL/pGA-espE.HA + whiB6) did not restore the phenotype (Fig 7). Therefore, the effects mediated by EspL and WhiB6 were uncoupled here: WhiB6 was proved to act at the transcriptional level, whereas EspL was demonstrated to exert its function post-transcriptionally, presumably on protein stability.
Taken together, these results highlight a new role for EspL in stabilizing EspE, EspF and EspH protein amounts.
To identify EspL interacting partners, strains carrying HA-tagged EspL were made. Expression of N- or C-terminally tagged EspL in ΔespL was verified by immunoblot (S9A Fig) and the ability of the modified proteins to complement the attenuation profile was checked (S9B Fig). Total cell extracts of strains ΔespL/pGA-espL.HA and ΔespL/pGA-HA.espL were employed in immunoprecipitation experiments using anti-HA antibodies. Mass spectrometry analysis of the precipitated material demonstrated that EspD was significantly enriched in the pulled-down fractions of ΔespL/pGA-HA.espL and of ΔespL/pGA-espL.HA, together with HA.EspL and EspL.HA (S6 Table). Other proteins were detected but their abundance was not increased in the immunoprecipitated samples compared to the Input and to the untagged strains H37Rv and ΔespL (S6 Table). A second readout, i.e. immunoblot, was exploited to validate these data independently. As shown in Fig 8, both EspD and EspL levels were highly enriched upon anti-HA immunoprecipitation in both ΔespL/pGA-HA.espL and ΔespL/pGA-espL.HA as compared to the Input control and strain H37Rv. On the other hand, RpoB and GroEL2 levels were as expected. To conclude, EspL and EspD may interact directly or be part of a multiprotein complex inside M. tuberculosis cells.
The data presented here demonstrate the essentiality of EspL for ESX-1-dependent virulence and for stabilizing the intracellular levels of EspE, EspF and EspH in M. tuberculosis. ESX-1-dependent secretion in ΔespL was severely compromised, with undetectable levels of EsxA, EsxB, EspA and EspD in the culture filtrates. Conversely, secretion of EspB was not affected, confirming previous data generated by our group [41]. In line with the secretion profile, virulence and innate cytokine production were compromised when THP-1 cells were infected by the espL knock-out strain. In this regard, ΔespL behaves like the ESX-1-null mutants ΔRD1 [13] and ΔΔRD1 [40], which fail to stimulate the innate immune response [15]. These phenotypes tally with those previously reported for an espL transposon mutant in a clinical isolate of the W/Beijing family, which was shown to have lost the ability to arrest phagosomal maturation [50], and with the need for espL in order to fully complement an espB mutant in M. marinum [51]. While these phenotypic traits are most likely attributable to lack of secretion of the main ESX-1 substrates, whether all of them are directly caused by lack of EspL or are mediated by EspE, EspF or EspH is currently unknown and additional research is required.
Although the mechanistic details of these functions remain unknown, a role for EspL as a chaperone protein can be proposed. Indeed, the presence of heterodimeric complexes, where one protein acts as a chaperone for the other, is not unusual in the ESX-1 system [24,52]. A direct effect of EspL on transcription of espE, espF and espH was ruled out by RNA-seq and further confirmed by qRT-PCR. On the other hand, proteomics identified EspE, EspF and EspH as the only proteins whose abundance was highly affected by espL deletion. Based on these findings, an interaction between EspL and EspE, EspF and EspH could be hypothesized. However, those proteins were not detected by mass-spectrometry and immunoblotting analysis of immunoprecipitated material from strains expressing EspL.HA or HA.EspL.
EspL-mediated stabilization of EspE, EspF and EspH levels might therefore occur by other means. Interestingly, compelling evidence was obtained for EspL interacting with EspD, which itself is known to act as a stabilizer [53], further suggesting the existence of a “chaperone complex” which contributes to regulating ESX-1 activity post-transcriptionally and/or post-translationally. Curiously, while EspD stabilizes EspA and EspC [53], EspL performs the same task on EspE, EspF and EspH, whose genes are paralogs of espA-espC-espD [9], although EspL and EspD are different in size and sequence.
Secretion of EspD deserves additional discussion. While EspL is necessary for secretion of the ESX-1 substrates EsxA, EsxB, EspA, EspC and for extracellular release of EspD, this was previously reported to be independent of a functional ESX-1 apparatus [53]. These observations are consistent with the findings presented here as the requirement of EspL for secretion of EspD does not imply that EspD be released through the ESX-1 system. Of note, espL is only present in the ESX-1 cluster and its function may serve more than one type VII secretion system and target EspD to alternative machineries, namely ESX-2, ESX-3 or ESX-4.
A model for ESX-1-dependent secretion can be proposed based on the current knowledge and on the data presented here (Fig 9). EspL forms a chaperone complex with EspD and this in turn stabilizes the EspA-EspC, EspE-EspH dimers and EspF. The chaperone complex may target EspA-EspC to the secretion machinery in the inner membrane, where co-dependent secretion with EsxA-EsxB takes place.
Another interesting finding was the discovery of whiB6 as the only deregulated gene in the ΔespL transcriptome. Despite reduced expression of whiB6, no difference in the mRNA levels of the genes belonging to the WhiB6 putative regulon [33] was observed. This can be ascribed to the culture conditions used in these experiments, as it was reported that WhiB6 senses reducing conditions in M. marinum, and regulates transcription accordingly, thanks to its Fe-S cluster [33]. Additionally, the 3-fold deregulation of whiB6 may not be mirrored by deregulation of its own regulon. Nonetheless, when WhiB6 was provided in trans, expression of most of the ESX-1 substrates or components was increased. Thereby, WhiB6 controls transcription of the ESX-1 genetic locus in M. tuberculosis too. The latter conclusion further supports the work by Solans and colleagues [35], who discovered that a single nucleotide insertion in the promoter region of whiB6 determines the response to the transcriptional regulator PhoP and thus modulates EsxA levels.
A recent report described similar an even more pronounced deregulation of whiB6 and of the WhiB6-dependent genes in M. marinum lacking eccCb1 [54]. In that case, the existence of a negative feedback loop connecting the ESX-1 core complex in the membrane to ESX-1 gene expression was postulated [54]. This is consistent with our findings in M. tuberculosis as EspL localizes mainly in the cytosol but also in the membrane fraction. Another analogy with the eccCb1 mutant in M. marinum lies in the EspE and EspF protein levels, which also seem to be subjected to post-transcriptional control [54]. Of note, we demonstrated that virulence cannot be restored in ΔespL by expression of WhiB6 in trans.
Altogether, our data indicate an important role for EspL in M. tuberculosis pathogenesis and encourage further investigations into the contributions of EspE, EspF and EspH to virulence and to ESX-1-dependent secretion. EspF was previously shown to reduce M. tuberculosis virulence in the mouse model when deleted [40]. EspH was recently identified as necessary for secretion of EspE and EspF in M. marinum and capable of binding EspE, thus acting as a potential chaperone [55]. Furthermore, essentiality of EspH for phagocytic infection as well as for granuloma formation in the zebrafish larvae model was reported [55]. On the contrary, little is known about the role of EspE. Conservation of the genes for espH, espD and espL, in the greatly down-sized genome of M. leprae [56] suggests a conserved function and this justifies future investigation.
Mycobacterium tuberculosis strains (described in S7 Table) were grown at 37°C in 7H9 medium (Difco) supplemented with 0.2% glycerol, 0.05% Tween 80 and 10% albumin-dextrose-catalase (ADC, Middlebrook) or on 7H10 plates supplemented with 0.5% glycerol and 10% oleic acid-albumin-dextrose-catalase (OADC, Middlebrook). Sauton’s liquid medium was used for culture filtrate analysis. Streptomycin (20 μg/ml), kanamycin (20 μg/ml), hygromycin (50 μg/ml) or 2.5% sucrose were added when necessary. Experiments involving M. tuberculosis were performed in a Biosafety Level 3 (BSL3) laboratory, according to the national and international guidelines (Authorization number A070027/3). For cloning purposes, One Shot TOP10 chemically competent Escherichia coli (Invitrogen) were grown in Luria–Bertani (LB) broth or on LB agar with hygromycin (200 μg/ml), kanamycin (50 μg/ml) or spectinomycin (25 μg/ml).
Chemical reagents were obtained from Sigma-Aldrich, unless otherwise stated. Restriction and modification enzymes were purchased from New England Biolabs. Plasmid vectors are described in S8 Table. Oligonucleotides were synthesized by Microsynth. Sequences are available upon request.
One kb up- and downstream regions of the espL gene were PCR-amplified, ligated in-frame with the AvrII site and cloned into the PacI and AscI sites of pJG1100 [38], resulting in the suicide vector pJG1100-espL-UP/DOWN. The complementing plasmid pGA44-espL was constructed by cloning the espL gene into vector pGA44 [39], under control of the PTR promoter. Deletion of the full-length espL gene was achieved by homologous recombination using plasmid pJG1100-espL-UP/DOWN. After transformation of M. tuberculosis H37Rv, the first recombination event was selected on 7H10 plates, supplemented with hygromycin and kanamycin. Colonies were screened by colony PCR. Two clones that had undergone homologous recombination were grown in liquid 7H9 medium with no antibiotics, in order to promote the second recombination event and plasmid excision. Selection of the recombinant clones was performed by plating the bacteria on 7H10 plates supplemented with sucrose. The resulting colonies were tested by PCR to confirm deletion of espL from its native locus and further validated by whole genome sequencing.
M. tuberculosis strains were grown to mid-logarithmic phase and then diluted to an optical density at 600 nm (OD600) of 0.05 in 7H9 medium. OD600 was recorded at different time points to obtain the growth curves.
M. tuberculosis genomic DNA was extracted as previously described [57]. Libraries were prepared using the Kapa LTP Library Prep kit (Kapa Biosystems) according to the manufacturer’s recommendations. Cluster generation was performed using the Illumina TruSeq SR Cluster Kit v4 reagents and sequenced on the Illumina HiSeq 2500 using TruSeq SBS Kit v4 reagents. Sequencing data were demultiplexed using the bcl2fastq Conversion Software (v. 2.20, Illumina, San Diego, California, USA). Raw reads were adapter- and quality-trimmed with Trimmomatic v0.33 [58]. The quality settings were “SLIDINGWINDOW:5:15 MINLEN:40”. Preprocessed reads were mapped onto the M. tuberculosis H37Rv reference genome sequence (RefSeq NC_000962.3) with Bowtie2 v2.2.5 [59]. SNP calling was done using VarScan v2.3.9 [60]. To avoid false-positive SNP calls the following cutoffs were applied: minimum overall coverage of ten non-duplicated reads, minimum of five non-duplicated reads supporting the SNP, mapping quality score >8, base quality score >15, and a SNP frequency above 80%. All SNPs were manually checked by visualizing the corresponding read alignments. Sequencing data have been deposited to the Sequence Read Archive (SRA) under accession number SRP158673.
M. tuberculosis cultures were grown to OD600 of 0.3–0.4, harvested by centrifugation, pellets were resuspended in TRIzol Reagent (ThermoFisher) and stored at -80°C until further processing. Total RNA was extracted by bead-beating as previously described [39]. Integrity of RNA was checked by agarose gel electrophoresis, purity and amount of RNA were assessed using a Nanodrop instrument and Qubit Fluorometric Quantitation (ThermoFisher), respectively. SuperScript III First-Strand Synthesis System (Invitrogen) was used to generate randomly primed cDNA from 1 μg of RNA, according to the manufacturer’s recommendations. qPCR reactions were performed on an ABI 7900HT instrument, using Power SybrGreen PCR Master Mix (Applied Biosystems), according to the manufacturer’s instructions. The housekeeping gene sigA was used for normalization.
RNA was extracted from biological duplicates as described above. RNA-seq libraries were prepared from 1 μg of total RNA. The RNA samples were depleted of r-RNAs with the Illumina Ribo-Zero rRNA Removal Kit (Gram-Positive Bacteria) then used to generate sequencing libraries with the Illumina TruSeq Stranded mRNA reagents, omitting the polyA selection step (Illumina, San Diego, California, USA). Cluster generation was performed with the resulting libraries using the Illumina TruSeq SR Cluster Kit v4 reagents and sequenced on the Illumina HiSeq 2500 using TruSeq SBS Kit v4 reagents. Sequencing data were demultiplexed using the bcl2fastq Conversion Software (v. 2.20, Illumina, San Diego, California, USA).
Reads were processed and mapped to the reference genome sequence as described above. Counting reads over features was done with featureCounts [61] from the Subread package v1.4.6. Annotation was taken from TubercuList release R27 (https://mycobrowser.epfl.ch/releases). Differential gene expression analysis was done using DESeq2 [62]. RNA-seq data have been deposited to the Gene Expression Omnibus (GEO) repository under accession number GSE118994.
M. tuberculosis cells, grown to mid-exponential phase in 15 ml cultures, were pelleted by centrifugation, washed once in PBS (Phosphate Buffered Saline) supplemented with 0.05% Tween 80 and the pellets stored at -80°C until further use. Total lysates were obtained by sonication in lysis buffer (100 mM Tris pH 8, 2% SDS, cOmplete mini EDTA free Roche) and boiled at 100°C for 1 h. Proteins were quantified by using the Pierce BCA Protein Assay kit and 30 μg were submitted for mass spectrometry analysis. For the analysis of the culture filtrates, M. tuberculosis cultures were grown in 7H9 to mid-logarithmic phase. The culture medium was then replaced by Sauton’s supplemented with 0.05% Tween 80 and growth was continued for 3 d. Finally, bacteria were pelleted and resuspended in Sauton’s medium without Tween 80 for collection of the culture filtrates. These were filtered through 0.22 μm Steriflip Millipore Express Plus Membranes (Millipore), concentrated 100x using Amicon Ultracel-3K centrifugal filters (Millipore), quantified by using the Pierce BCA Protein Assay kit and 30 μg were submitted for mass spectrometry analysis.
Each sample was digested by Filter Aided Sample Preparation (FASP) [63] with minor modifications. Dithiothreitol (DTT) was replaced by Tris(2-carboxyethyl)phosphine (TCEP) as reducing agent and Iodoacetamide by Chloracetamide as alkylating agent. A combined proteolytic digestion was performed using Endoproteinase Lys-C and Trypsin. Acidified peptides were desalted on C18 StageTips [64] and dried down by vacuum centrifugation. For LC MS/MS analysis, peptides were resuspended and separated by reversed-phase chromatography on a Dionex Ultimate 3000 RSLC nanoUPLC system in-line connected with an Orbitrap Fusion Lumos Mass-Spectrometer (Thermo Fischer Scientific). Database search was performed using MaxQuant 1.6.0.1 [65] against the TubercuListR27 database (http://tuberculist.epfl.ch/). Carbamidomethylation was set as fixed modification, whereas oxidation (M), phosphorylation (S,T,Y) and acetylation (Protein N-term) were considered as variable modifications. Label Free Quantification (MaxLFQ) was performed by MaxQuant using the standard settings [66]. Perseus [67] was used to highlight differentially quantified proteins. Reverse proteins, contaminants and proteins only identified by sites were filtered out. Biological replicates were grouped together and protein groups containing a minimum of two LFQ values in at least one group were conserved. Empty values were imputed with random numbers from a normal distribution. The average LFQ values for the different proteins in the different strains were obtained from columns DM-DO, DP-DR and DS-DU in S2 Table, from columns EK-EM, EN-EP, EQ-ES, ET-EV in S3 Table. The difference between these numbers represents the “difference in protein abundance”. Significant hits were determined by a volcano plot-based strategy, combining t test p-values with ratio information [68]. Significance curves in the volcano plot corresponding to a SO value of 0.5 and 0.05 FDR (for culture filtrates) and to a SO value of 0.1 and 0.05 FDR (for cell lysates) were determined by a permutation-based method. Further graphical displays were generated using homemade programs written in R [69]. Raw data obtained from mass spectrometry experiments have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifiers PXD010929 (total cell lysate) and PXD011466 (culture filtrate).
Culture filtrates were obtained as described above for mass spectrometry analysis, quantified by using the Qubit Fluorometric Quantitation device (ThermoFisher) and loaded on SDS-PAGE 12–15% NuPAGE gels (Invitrogen) for immunoblot analyses. Bacterial pellets were washed once in Tris-Buffered Saline (TBS, 20 mM Tris-HCl pH 7.5, 150 mM NaCl) and stored at -80°C until further processing. Cells were sonicated in TBS supplemented with a protease inhibitor tablet (cOmplete mini EDTA free, Roche) for 15 min and the protein solution was then sterilized by filtration through a 0.22 μm filter (Pall Life Sciences) to remove residual intact cells. Protein samples were quantified using Qubit. Equal amounts of protein preparations (10 μg for cell lysates, 15–20 μg for culture filtrates) were loaded on SDS-PAGE 12–15% NuPAGE gels (Invitrogen) and transferred onto PVDF membranes using a semidry electrophoresis transfer apparatus (Invitrogen). Membranes were incubated in TBS-Tween blocking buffer (25 mM Tris pH 7.5, 150 mM NaCl, 0.05% Tween 20) with 5% w/v skimmed milk powder for 2h at 4°C prior to overnight incubation with primary antibody. Membranes were washed in TBS-Tween three times at room temperature, and then incubated with secondary antibody for 3 h before washing again. Signals were detected using Chemiluminescent Peroxidase Substrate 1 (Sigma-Aldrich).
Polyclonal anti-EspL, anti-EspB, anti-EspA [53], anti-EspD [53] antibodies were produced by Dr. Ida Rosenkrands (Statens Serum Institut, Copenhagen, Denmark). Monoclonal anti-RpoB antibodies were purchased from NeoClone, polyclonal anti-EsxB antibodies from Abcam, monoclonal anti-HA antibodies conjugated to Horseradish Peroxidase (HRP) from Cell Signaling. Polyclonal anti-Rv3852 antibodies were generated by Eurogentec [42]. The following reagents were obtained through BEI Resources, NIAID, NIH: monoclonal anti-Antigen 85, monoclonal anti-GroEL2 and polyclonal anti-EsxA antibodies.
Cell fractions were obtained as described previously [42]. Briefly, M. tuberculosis was grown in Sauton’s medium with 0.05% Tween 80 to mid-exponential phase, cells were collected by centrifugation, supernatants were filtered through 0.22 μm Steriflip Millipore Express Plus Membranes (Millipore) and concentrated 100x using Amicon Ultracel-3K centrifugal filters (Millipore) to obtain the culture filtrate fraction. The pellet was treated with 0.25% Genapol-X080 for 30 min at room temperature, followed by centrifugation at 14,000 g for 10 min. The proteins in the resulting supernatant were precipitated with Trichloroacetic acid (TCA), yielding the capsular fraction. The remaining pellet was subjected to sonication to break the cells, sterilized by filtration through a 0.22 μm filter (Pall Life Sciences) followed by ultra-centrifugation at 45,000 rpm for 1h at 4°C. The supernatant contained the cytosolic fraction, while the pellet was enriched with membrane proteins. Analysis of the protein fractions was carried out by immunoblot as described above.
M. tuberculosis cells in 30 ml cultures were pelleted by centrifugation, washed once in PBS (Phosphate Buffered Saline) supplemented with 0.05% Tween 80 and the pellets stored at -80°C until further use. Total lysates were obtained by sonication in TBS-T (25 mM Tris pH 7.5, 150 mM NaCl, 0.05% Tween 20), followed by filtration through 0.22 μm filters (Pall Life Sciences). Fifty microliters of Monoclonal Anti-HA Agarose Antibody beads (Sigma-Aldrich) were incubated with approximately 1 mg of bacterial extract in Spin-X centrifuge tubes (Costar) for 4 h at 4°C on an orbital shaker. The resin was washed four times in PBS and the immunoprecipitated material was eluted from the beads in PBS-SDS sample buffer (100 mM Tris HCl pH 6.8, 200 mM dithiothreitol, 4% SDS, 0.2% bromophenol blue, 20% glycerol) during a 5 min incubation at 95°C. Immunoprecipitated proteins were analyzed either by mass spectrometry as described [24] or by immunoblot. A mock (no antibody) control was run in parallel with agarose beads only.
THP-1 cells (ATCC-TIB202, LGC Standards GmbH, Germany) were cultured in RPMI1640 (Life Technologies) supplemented with 10% (v/v) Fetal Calf Serum (Life Technologies) and 1% sodium pyruvate (Life Technologies). Cells were differentiated in 96- or 12-well plates by addition of 4 nM phorbol 12-myristate 13-acetate (PMA) for 24 h at 37°C in 5% CO2. Differentiated cells were then infected with M. tuberculosis as follows. Bacteria were grown to exponential phase (OD600 between 0.4 and 0.8), washed once in 7H9 medium, resuspended in 7H9 to an OD600 of 1, equivalent to 3 x 108 bacteria/ml. The required volume of bacterial suspension was then added to RPMI1640 for infection of human THP-1 cells at the multiplicity of infection (MOI) reported in the text. Plates were sealed with gas-permeable sealing film and incubated at 37°C under 5% CO2. Intracellular bacteria were released by the infected cells by addition of 0.5% Triton-X. The suspension was serially diluted in 7H9 and plated on 7H10 plates. Colony forming units (CFU) were counted after incubation at 37°C for 4–5 weeks. PrestoBlue Assay (Thermofisher) to evaluate cell viability was performed according to the manufacturer’s instructions. Fluorescence was measured using a Tecan Infinite M200 microplate reader.
Cell culture supernatants from infections in 96-well plates were removed from infected cells 24 h post-infection. Supernatants were filtered through NANOSEP centrifugal devices (Pall Life Sciences) and assayed for human IL-1β (BD Biosciences) according to the manufacturer’s instructions.
RNA from infected cells in the 12-well format was extracted by using Qiagen RNeasy kit according to the manufacturer’s instructions 24 h post-infection and reverse-transcribed using the RevertAid First Strand cDNA Synthesis kit (Fermentas). Quantitative PCR analysis was performed on an ABI 7900HT instrument. All gene expression data are presented as relative expression to GAPDH.
Statistical analyses were performed in GraphPad PRISM by one-way or two-way ANOVA followed by Tukey’s multiple comparison test.
|
10.1371/journal.pntd.0007046 | Differential replication efficiencies between Japanese encephalitis virus genotype I and III in avian cultured cells and young domestic ducklings | Japanese encephalitis virus (JEV) genotype dominance has shifted to genotype I (GI) from genotype III (GIII) in China as demonstrated by molecular epidemiological surveillance. In this study, we performed a serological survey in JEV-non-vaccinated pigs to confirm JEV genotype shift at the sero-epidemiological level. The average ratio of GI/GIII infection was 1.87, suggesting co-circulation of GI and GIII infections with GI infection being more prevalent in pigs in China. To gain an insight into the reasons for this JEV genotype shift, the replication kinetics of seven recently-isolated JEV isolates including three GI strains and four GIII strains were compared in mosquito C6/36 cells, chicken fibroblast cells (DF-1) and porcine iliac artery endothelial cells (PIEC). We observed that GI strains replicated more efficiently than GIII strains in DF-1 and PIEC cells, particularly in DF-1 cells with titers reaching 22.9–225.3 fold higher than GIII strains. This shows an enhanced replication efficiency of GI viruses in avian cells. To examine this enhanced replication efficiency in vivo, young domestic ducklings were used as the animal model and inoculated with GI and GIII strains at day 2 post-hatching. We observed that GI-inoculated ducklings developed higher viremia titers and displayed a comparatively longer viremic duration than GIII-inoculated ducklings. These results conform to the hypothesis of an enhanced replication efficiency for GI viruses in birds. There are 36 amino acid differences between GI and GIII viruses, some of which may be responsible for the enhanced replication efficiency of GI viruses in birds. Based on these findings, we speculated that the enhanced replication of GI viruses in birds would have resulted in higher exposure and therefore infection in mosquitoes, which could result in an increased transmission efficiency of GI viruses in the birds-mosquitoes-birds enzootic transmission cycle, thereby contributing to JEV genotype shift.
| Japanese encephalitis virus (JEV) causes encephalitis in humans and reproductive disorder in pigs. The enzootic transmission cycle of JEV is maintained in nature by several species of mosquitoes and vertebrates including birds and pigs. In recent years, JEV genotype I (GI) replaced genotype IIII (GIII) as the dominant genotype in Asian countries. Genotype shift has an impact on disease control, and understanding the reasons for this shift will offer valuable insight into avenues for future disease control. Therefore, we compared the replication efficiencies of GI and GIII viruses in vitro and in vivo. We observed that GI viruses show higher replication titers in avian cells and higher viremia levels in young domestic ducklings than GIII viruses, suggesting an enhanced replication efficiency of GI viruses in birds. Based on these findings, we speculated that the enhanced replication of GI viruses in birds could provide increased mosquito infection, leading to an increase in the birds-mosquitoes-birds transmission cycle, thereby contributing to JEV genotype shift.
| Japanese encephalitis virus (JEV) is a zoonotic flavivirus that causes encephalitis in humans and reproductive disorders in pigs in the Asian pacific region [1,2]. The genome of JEV is single-stranded positive-sense RNA consisting of a short 5’ untranslated region, a single open reading frame, and a longer 3’ untranslated region. The single open reading frame encodes a polyprotein that is subsequently cleaved by both cellular and viral proteases into three structural proteins (capsid (C), pre-membrane/membrane (PrM), and envelope (E)) and seven nonstructural proteins (NS) (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) [3].
The JEV enzootic transmission cycle is maintained in nature by several species of mosquito vectors and vertebrate hosts. JEV is transmitted predominantly by Culex mosquitoes, but several other genera may participate in certain circumstances [4]. Mosquitoes transmit JEV from a viremic vertebrate to a susceptible vertebrate including humans, birds, pigs and other mammals by bite. After infection by JEV-infected mosquitoes, many domestic and wild bird species demonstrate varying degrees of viremia. Some of which, including young domestic ducklings and chicks, as well as ardeid wading birds develop a level of viremia sufficient to infect mosquitoes and are thus considered the amplifying hosts for JEV transmission [4–6]. Among mammal species susceptible to JEV infection, pigs are the only mammals responsible for JEV transmission, because JEV-infected pigs develop a level of viremia that remains high enough to infect mosquitoes for up to 4 days [4].
JEV is phylogenetically divided into five genotypes (genotype I to V) based on the nucleotide sequence of the E gene [7,8]. Genotype III (GIII) has historically been the main causative agent of Japanese encephalitis (JE) and was the dominant genotype throughout most of Asia from 1935 through the 1990s. Genotype I (GI) was isolated in Cambodia in 1967 and remained undetectable until 1977 when a new isolate was collected in China. Notably, molecular epidemiological surveillance of JEV isolates collected during the last 20 years revealed that GIII has been gradually replaced by GI, showing a genotype shift with GI as the dominant genotype in Asian countries [9–10].
Previous analysis of the differences between the two genotypes at genetic and epidemiological levels suggested that GI displaced GIII probably by achieving a replication cycle that is more efficient but more restricted in its host range [11]. This hypothesis was partially supported by an observation that a GI isolate has significantly higher infectivity titers in mosquito C6/36 cells than two GIII isolates [10]. It is known that pathogenicity and infectivity vary among JEV strains. We therefore evaluated the prevalence of GI and GIII infection in pigs in China and used seven recently-isolated JEV strains including three GI strains and four GIII strains to compare their replication efficiency in vitro and in vivo, with the aim of gaining insight into the reasons for the JEV genotype shift.
All animal experiments were approved by the Institutional Animal Care and Use Committee of Shanghai Veterinary Research Institute (IACUC No: Shvri-po-2016060501) and performed in compliance with the Guidelines on the Humane Treatment of Laboratory Animals (Ministry of Science and Technology of the People’s Republic of China, Policy No. 2006 398).
Seven recently-isolated JEV strains including three GI strains (SD12, SH2 and SH7 strains) and four GIII strains (SH1, SH15, SH19 and N28 strains) were used in this study. The basic information for these JEV strains is shown in Table 1. All JEV strains were isolated from aborted pigs or mosquitoes during 2015 and 2016 and were passaged fewer than seven times in cultured cells, including three passages for plaque purification. The genotypes of the JEV strains were identified using the sequence of the E gene, as described previously [12].
A total of 2272 porcine blood samples were collected from JEV-non-vaccinated pigs at pig farms and slaughterhouses located in 12 provinces in China in 2016, including Jilin, Inner Mongolia, Xinjiang, Qinghai, Ningxia, Hebei, Jiangsu, Shanghai, Hubei, Hunan, Guangdong and Guangxi. Porcine serum samples were stored at -80°C immediately after centrifugation. The detailed information of the porcine serum samples are shown in S1 Table.
Serum samples collected from pigs were screened by a commercial enzyme-linked immunosorbent assay (ELISA) kit specific to JEV infection (Wuhan Keqian Biology, Wuhan, China). The seropositive samples were further examined by antibody-sandwich ELISA to distinguish GI and GIII infection, as described previously [13]. Briefly, a 96-well ELISA plate was coated with 100 μl per well of porcine serum diluted at 1:100 as well as the positive and negative serum controls at 37°C for 90 min and was blocked with 5% skimmed milk in phosphate-buffered saline (PBS) containing 0.05% Tween 20. JEV SD12 (GI) and N28 (GIII) viruses were heat-inactivated in a water-bath at 56°C for 30 min and added into the serum-coated wells at 105 plaque forming units (PFU) per well. Following incubation at 4°C overnight, the diluted (1:3000) mouse anti-JEV antibodies were dispensed into each well and incubated at 37°C for 60 min. The bound antibodies were detected with horseradish peroxidase-conjugated goat anti-mouse IgG (Santa Cruz Biotechnology, Santa Cruz, CA, USA) and subsequently with 3,3',5,5'-tetramethylbenzidine. The optical density (OD450) of each well was measured at 450 nm. GI and GIII infection were differentiated by comparing the OD450 values produced by GI and GIII viruses [13].
Aedes albopictus C6/36 cells were maintained in RPMI 1640 medium (Thermo Fisher Scientific, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific) at 28°C. Chicken fibroblast cells (DF-1) and porcine iliac artery endothelial cells (PIEC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Thermo Fisher Scientific) supplemented with 10% FBS at 37°C in an atmosphere containing 5% CO2. For JEV infection, C6/36 cells grown on plates were infected with GI or GIII virus at a multiplicity of infection (MOI) of 0.01 and incubated at 28°C for 2h. Following washing with PBS, the cells were cultured in RPMI 1640 medium containing 2% FBS at 28°C for the indicated times. DF-1 and PIEC cells grown on plates were infected with GI or GIII virus at 0.1 MOI and incubated at 37°C for 2 h. Following washing with PBS, the cells were cultured in DMEM supplemented with 2% FBS at 37°C for the indicated times. The supernatants were sampled at the indicated intervals and stored at -80°C. The JEV titers in the supernatants were measured by 50% tissue culture infectious dose (TCID50) assay, as described previously [14].
Specific-pathogen-free Shaoxing ducklings (Anas platyrhyncha var. domestica) purchased from Harbin Veterinary Research Institute, Chinese Academy of Agricultural Sciences, were inoculated with JEV strains at day 2 post-hatching. Briefly, the ducklings were divided randomly into GI and GIII strain-inoculated groups (n = 10) and inoculated subcutaneously with 10,000 PFU of JEV per animal [5]. After inoculation, all ducklings were monitored for 7 days and blood samples (0.15 ml) were taken from the jugular vein once daily from 2 days post-inoculation (dpi) to 7 dpi for detection of viremia. The levels of viremia were measured by TCID50 assay, as described previously [14].
The amino acid sequences of JEV strains were obtained from GenBank (S2 Table). Multiple sequence alignments were performed using the DNASTAR Lasergene 7.1 (MegAlign). Phylogenetic tree was generated by the neighbor-joining method using MEGA version 6.06.
Student’s t-test or two-way analysis of variance (ANOVA) were used for significance analysis. A p value of <0.05 was considered significant.
Although JEV genotype shift has been demonstrated by molecular epidemiological analysis of JEV isolates in China [9,15], the serological evidence for the genotype shift in both humans and pigs was lacking in China. Pigs excluding those used for breeding are not vaccinated for JEV in China and thus are an ideal model for the serological survey and prevalence detection of GI and GIII infection. A total of 2272 serum samples were collected from JEV-non-vaccinated pigs at pig farms or slaughterhouses located in 12 provinces, of which 854 samples were seropositive for JEV infection, as screened by the ELISA kit (S1 Table). The prevalence of JEV infection varied among the 12 provinces, ranging from 25.0% to 58.7% with an average prevalence of 37.6% (Fig 1A and S1 Table). This prevalence rate was in line with a previous surveillance study [15]. The average prevalence of JEV infection in the second half year was 47.0% significantly higher than (32.0%) in the first half year (p = 0.0179) (Fig 1A), suggesting an increased prevalence of JEV infection after mosquito season.
To distinguish between GI and GIII infection in the seropositive samples, a previously established antibody-sandwich ELISA was performed, which categorized GI and GIII infection by comparing the OD450 values between GI and GIII viruses [13]. A sample producing a greater OD450 value against GI virus than GIII virus was considered as GI infection, reversely, a sample producing a greater OD450 value against GIII virus than GI virus was classified as GIII infection. Among the 854 seropositive samples, 116 were undistinguishable because their OD450 values were nearly identical. Of the remaining 738 samples, 443 and 295 were classified into GI and GIII infection, respectively, with an average GI/GIII infection ratio of 1.87 (Fig 1B and 1C and S1 Table), showing co-circulation of GI and GIII infections with GI infection being dominant in pigs.
No significant difference in the ratio of GI/GIII infection was detected between the first and second half years (Fig 1B). The prevalence of GI and GIII infections was further analyzed at the province level. The ratio of GI/GIII infection varied among the 12 provinces, with the highest ratio of 3.21 in Qinghai and the lowest ratio of 1.23 in Jiangsu (Fig 1D). Taken together, these data indicated that GI infection was dominant for pigs in China, confirming the genotype shift suggested by the molecular epidemiological analysis of JEV isolates.
Mosquitoes, birds and pigs are the primary hosts of JEV and play essential roles in maintaining the JEV transmission cycle [4]. Previous studies hypothesized that GI displaced GIII by achieving an increased replication efficiency in various hosts [10,11]. Thus, we compared the replication kinetics of GI and GIII viruses in mosquito C6/36 cells, chicken DF-1 and porcine PIEC cells. These cells were susceptible to JEV infection and used in this study as in vitro cell models of mosquito, bird and pig hosts, respectively. The C6/36, DF-1 and PIEC cells were inoculated with seven recently-isolated JEV strains including three GI strains (SD12, SH2 and SH7 strains) and four GIII strains (SH1, SH15, SH19 and N28 strains) and their replication titers in the supernatants were measured. No significant difference in replication titers for GI versus GIII were observed in C6/36 cells (Fig 2A). However, notable differences were observed in PIEC and DF-1 cells between GI and GIII strains. GI strains showed significantly higher replication titers than GIII strains in PIEC cells at 42 (p<0.0001), 48 (p = 0.0003) and 54 hours post-infection (hpi) (p = 0.0002) (Fig 2B). More significant differences in replication titers between GI and GIII strains were observed in DF-1 cells (Fig 2C). The average replication titers of the GI strains were 22.9, 51.0, 103.3, 225.3, 192.5, 97.7 and 63.4 fold higher than those of the GIII strains at 24 (p = 0.0005), 36 (p = 0.003), 48 (p<0.0001), 60 (p<0.0001), 72 (p = 0.0005), 84 (p<0.0001) and 96 hpi (p = 0.0004), respectively. These data indicate that GI viruses have an enhanced replication efficiency in chicken and porcine cells when compared with GIII viruses.
Given that the enhanced replication of GI strains was observed in chicken cells, we wanted to determine whether this enhanced replication was repeated in an animal model. Young domestic ducklings develop a detectable level of viremia after JEV infection and are considered an amplifying host contributing to the JEV transmission cycle [5,16]. We therefore used young domestic ducklings as an animal model to compare the replication efficiency between GI and GIII strains. Shaoxing ducklings were subcutaneously inoculated at day 2 post-hatching with JEV strains including three GI strains (SD12, SH2 and SH7 strains) and four GIII strains (SH1, SH15, SH19 and N28 strains), and the levels of viremia were measured. Most of the JEV-inoculated ducklings developed a detectable viremia level starting from 2 or 3 dpi and remained viremic for 1–4 days depending on the strain (Fig 3). No significant difference in viremia levels between GI- and GIII-inoculated ducklings were observed at 2 and 3 dpi (Fig 3A and 3B); however, significant differences were detected at 4 and 5 dpi. GI-inoculate ducklings developed significantly higher viremia levels than GIII-inoculated ducklings at 4 (p = 0.0007) and 5 dpi (p = 0.0309) (Fig 3C and 3D). The viremic rates were similar between GI and GIII-inoculated ducklings (Fig 3E), while the viremic duration of GI-inoculated ducklings was notably, but not significantly (p = 0.0525), longer than GIII-inoculated ducklings (Fig 3F). These data indicate that GI-inoculated ducklings develop higher viremia levels than GIII-inoculated ducklings, suggesting an enhanced replication efficiency of GI viruses in birds.
Minor mutations in JEV proteins are associated with changes in JEV replication and host fitness [17,18], we therefore performed a multiple alignment on amino acid sequences to detect the amino acid variations between the GI and GIII strains. The amino acid sequences of 19 GI strains and 20 GIII strains (S2 Table) were downloaded from GenBank and the amino acid variations were compared. There are 36 amino acid differences present in viral proteins including three structural proteins (C, PrM and E) and seven nonstructural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) (Table 2 and S1 Fig), some of which may be responsible for the difference in replication efficiency between GI and GIII viruses.
Notably, there were 4, 6, 5 and 6 variations respectively located in the E, NS1, NS3 and NS5 proteins that play important roles in JEV replication. The E protein is the major structural protein containing a receptor-binding domain and neutralization epitopes and plays major roles in mediating virus entry and pathogenicity [17,19]. The amino acid variation in the E protein may influence the cell tropism, penetration into cells and virulence. NS1 is a multifunctional glycoprotein that is present in different cellular locations including the intracellular membranes and cell surface as well as in sera as a secreted lipo-particle, and serves a central role in viral replication, eliciting the immune response, and inhibition of the complement system [20,21]. The amino acid variation in NS1 may also effect its roles in viral replication, immune modulation and immune evasion. NS3 is a multifunctional protein that possesses the enzymatic activities of serine protease, helicase and nucleoside 5’-triphosphatase [22,23]. The amino acid variation in NS3 may result in an altered enzymatic activity in the processing of the viral precursor polyprotein and the replication of viral genomic RNA. NS5 is the largest protein and consists of the methyltransferase (MTase) and RNA-dependent RNA polymerase (RdRP) [24]. MTase is involved in methylation of the 5’ RNA cap structure and RdRP is the key enzyme for viral replication. In addition, NS5 contributes to the blocking of interferon signaling pathways [25]. The amino acid variation in NS5 may change the enzymic activities of MTase and RdRP as well as the antagonization of the interferon response.
Molecular epidemiological surveillance of JEV isolates demonstrated that GI replaced GIII as the dominant genotype in China [9,15,26]; however, this genotype shift has not been confirmed by serological evidence. Vaccination with the SA14-14-2 live attenuated JE vaccine (GIII) results in a difficulty in distinguishing GI and GIII infection in humans using serological surveillance methods. We therefore studied the seroprevalence of GI and GIII infection in JEV-non-vaccinated pigs using the antibody-sandwich ELISA [13] and found that the average ratio of GI/GIII infection was 1.87 among 738 porcine serum samples collected from 12 provinces. These results suggest co-circulation of GI and GIII infections with GI infection dominant in pigs, confirming the genotype shift in China. However, these data generated by the antibody-sandwich ELISA [13] were somewhat provisional and future studies should further validate this approach using sera from animals infected with specific GI and GIII viruses. In addition, there were 116 porcine serum samples undistinguishable for GI and GIII infection, which may be attributable to that the animals were probably infected with both genotype viruses.
Previous analysis of the differences between GI and GIII viruses at genetic and epidemiological levels suggests that GI displaced GIII probably by achieving an increased replication efficiency in hosts [10,11]. To test this speculation, we used seven recently-isolated JEV strains including three GI strains and four GIII strains to compare their replication kinetics in mosquito C6/36, chicken DF-1 and porcine PIEC cells. These viruses were passaged fewer than seven times during isolation and plaque purification to avoid artificial mutations in viral proteins. No significant differences in replication titers were observed in C6/36 cells, but significant differences were observed in DF-1 and PIEC cells, between GI and GIII strains. Particularly, the replication titers of GI strains in DF-1 cells where they were 22.9–225.3 fold higher than those of GIII strains, indicating that GI viruses had higher replication efficiency in avian and porcine cells.
Although both birds and pigs are amplifying hosts in the JEV transmission cycle, JEV genotype shift also occurs in some endemic countries, including India [27], Malaysia [28] and Korea [29], where pig-breeding is not common. This suggests that birds including herons and ducks, but not pigs, are potential hosts contributing to genotype shift [29,30]. It is known that birds, especially wading birds, are considered the primary enzootic hosts of JEV, and play an essential role in epizootic viral amplification in some areas [4]. Young domestic ducklings are susceptive to JEV infection and develop viremia sufficient for mosquito infection [5]. In addition, large numbers of duck farms located near ponds and lakes where mosquitos breed, provide an abundant pool of amplifying hosts for JEV infection and facilitate the birds-mosquito-birds transmission cycle. We therefore used young domestic ducklings as an animal model to confirm the increased replication efficiency of GI strains observed in avian cells. As expected, GI-inoculated ducklings developed higher viremia titers and relatively longer viremia duration than GIII-inoculated ducklings, confirming that the replication efficiency of GI viruses was higher in birds. Taken together, these in vitro and in vivo data suggest that birds including young domestic ducklings may be the major contributing factor in genotype shift. A previous comparative analysis of replication kinetics between GI and GIII isolates in North American avian species indicated that GI viruses showed higher viremia titers than GIII viruses in several avian species including mallards, house finches and ring-billed gulls [6]. This observation further supports our hypothesis that birds might contribute to JEV genotype shift.
Numerous factors, including the replication efficiency of GI and GIII viruses in hosts, the availability and abundance of amplifying hosts, the contact rates among amplifying hosts and mosquito vectors, and climatic and environmental parameters, are considered to play a role in JEV genotype shift [6,31]. Based on our findings together with the previous observation that GI viruses show higher viremia titers than GIII viruses in several avian species [6], we speculate that the enhanced replication efficiency of GI viruses in birds would have provided mosquitoes more chances to be infected, which led to an increased transmission efficiency of GI viruses in the birds-mosquitoes-birds enzootic transmission cycle, and eventually the displacement of GIII viruses by GI viruses as the dominant genotype.
GI viruses also showed higher replication titers than GIII viruses at the late stage of replication in porcine cells, suggesting that pigs may also be involved in JEV genotype shift. We have inoculated sixty-day old and antibody-negative piglets (n = 5) with JEV SD12 strain (GI) and N28 strain (GIII) to examine the difference in viremia development. No significant differences in viremia levels was observed between GI- and GIII-inoculated piglets (S3 Table). However, we could not exclude the role of pigs in JEV genotype shift using this result because of the limited numbers of GI and GIII strains used for experimental inoculation.
A previous observation described a GI isolate (JE-91 strain) showing significantly higher replication titers than GIII isolates (Tiara and Matsunaga strains) in mosquito C6/36 cells in a short period from 24 to 48 hpi [10]. This observation is partially in contrast with our findings where no significant difference in replication titers between GI and GIII strains was observed in mosquito C6/36 cells. However, our findings are in agreement with an in vivo observation that no significant differences in replication kinetics and dissemination is observed between mosquitoes infected with the GI isolate (JE-91 strain) and the GIII isolate (Tiara strain) [32]. Our data together with these in vivo data suggest that mosquitoes might not play a crucial role in JEV genotype shift.
Multiple alignment of amino acid sequences between GI and GIII viruses indicated that there are 36 amino acid differences present in several viral proteins, including E, NS1, NS3 and NS5, which play important roles in JEV replication, pathogenicity and immune modulation. Minor mutations in JEV proteins are associated with changes in JEV replication and host fitness [17,18]. The amino acid mutations in GI viruses might be responsible for the enhanced replication efficiency of GI viruses in avian cells and young domestic ducklings. This hypothesis is currently under investigation in our laboratory.
In conclusion, a serological survey of JEV infection in JEV-non-vaccinated pigs demonstrated co-circulation of GI and GIII infections with GI infection dominant in pigs in China, despite the fact that these results generated by the antibody-sandwich ELISA were somewhat provisional. Comparative analysis of the replication kinetics of GI and GIII strains indicated that GI strains replicated more efficiently than GIII strains in avian and porcine cells, particularly in avian cells with titers reaching 22.9–225.3 fold higher than GIII strains. In addition, GI-inoculated ducklings developed higher viremia titers and displayed a comparatively longer viremic duration than GIII-inoculated ducklings. These observations suggested that there is enhanced replication efficiency of GI viruses in birds compared with GIII viruses. There are 36 amino acid differences between GI and GIII viruses, some of which may be responsible for the enhanced replication efficiency of GI viruses in birds. Based on these findings, we speculate that the enhanced replication efficiency of GI viruses in birds could provide mosquitoes more chances to be infected, which would lead to an increased transmission efficiency of GI viruses in the birds-mosquitoes-birds enzootic transmission cycle, and eventually the displacement of GIII viruses by GI viruses as the dominant genotype.
|
10.1371/journal.pntd.0003677 | Evaluation of Inapparent Dengue Infections During an Outbreak in Southern China | Few studies evaluating inapparent dengue virus (DENV) infections have been conducted in China. In 2013, a large outbreak of DENV occurred in the city of Zhongshan, located in Southern China, which provided an opportunity to assess the clinical spectrum of disease. During the outbreak, an investigation of 887 index case contacts was conducted to evaluate inapparent and symptomatic DENV infections. Post-outbreak, an additional 815 subjects from 4 towns with, and 350 subjects from 2 towns without reported autochthonous DENV transmission, as determined by clinical diagnosis, were evaluated for serological evidence of dengue IgG antibodies. Between July and November 2013, there were 19 imported and 809 autochthonous dengue cases reported in Zhongshan. Of 887 case contacts enrolled during the outbreak, 13 (1.5%) exhibited symptomatic DENV infection, while 28 (3.2%) were inapparent. The overall I:S ratio was 2.2:1 (95% CI: 1.1-4.2:1). Post-outbreak serological data showed that the proportion of DENV IgG antibody detection from the 4 towns with and the 2 towns without reported DENV transmission was 2.7% (95% CI: 1.6%-3.8%) and 0.6% (95% CI: 0-1.4%), respectively. The I:S ratio in the 3 towns where clinical dengue cases were predominately typed as DENV-1 was 11.0:1 (95% CI: 3.7-∞:1). The ratio in the town where DENV-3 was predominately typed was 1.0:1 (95% CI: 0.5-∞:1). In this cross-sectional study, data suggests a high I:S ratio during a documented outbreak in Zhongshan, Southern China. These results have important implications for dengue control, implying that inapparent cases might influence DENV transmission more than previously thought.
| In this report, we evaluated individuals with symptomatic and asymptomatic dengue virus (DENV) infections during a 2013 DENV outbreak in Southern China, as well as performed post-outbreak serological testing for DENV IgG antibodies, to better understand DENV transmission. These findings suggest a high rate of asymptomatic cases, which has important implications for future dengue control.
| Dengue is one of the most significant mosquito-borne diseases in the world. During the past three decades, the geographical spread of both the mosquito vectors and viruses have led to the global resurgence of epidemic dengue. The World Health Organization (WHO) has estimated that 3.6 billion people live in dengue-endemic areas and that 50 million dengue infections occur annually, with over 2 million causing dengue hemorrhagic fever (DHF) and 21,000 resulting in death [1]. More recent work, which considers both symptomatic and asymptomatic dengue infection, has estimated the global burden of dengue to be much higher, at 390 million infections per year [2].
The clinical manifestations of dengue virus (DENV) infection can be classified as inapparent, undifferentiated febrile illness, classic dengue fever, or the more severe forms, DHF and dengue shock syndrome (DSS). This clinical disease spectrum becomes very important when developing an appropriate surveillance strategy to detect DENV infections. Particularly, challenges can arise when individuals experience mild or asymptomatic infections, as most surveillance programs could easily miss these subclinical cases. Previous surveys conducted in DENV endemic regions have suggested that asymptomatic cases occur more frequently than symptomatic ones, and that the inapparent-to-symptomatic (I:S) ratio varies greatly [3–10]. Given that detectable viremia has been reported among inapparent cases by RT-PCR and virus isolation [11], and that silent circulation of DENV among humans has also been previously documented [4,12], it is possible that asymptomatic DENV infections could cause new foci of disease or eventually an epidemic in non-endemic regions [13]. Thus, it is critical that we fully understand the epidemiology of inapparent dengue infections in order to better develop control strategies to prevent such events.
The one Chinese study conducted in 2009, during an outbreak of DENV-3, the authors estimated the incidence rate of inapparent DENV infections in rural areas throughout Southeastern China to be 28%, but did not attempt to estimate an I:S ratio [14]. Outside of China, a study was conducted during a 2008–2009 dengue epidemic in Australia, where researchers serologically evaluated blood donors to estimate the I:S ratio for DENV infections, which they determined to be 0.59:1 (range 0.18–1.0) [15]. This ratio was markedly lower than similar studies conducted in other endemic regions [3–10]. In 3 other prospective studies that evaluated travelers in non-endemic regions, the I:S ratios were estimated to be 0.75:1, 1.8:1, and 3.0:1 [16–18]. While there have been multiple of such studies looking at inapparent and symptomatic DENV infection ratios, to our knowledge, no such studies have been conducted in China where DENV is a common viral threat in the southern parts of the country.
Re-emergence of dengue in Mainland China was first reported in 1978. Since then, multiple DENV outbreaks have occurred, primarily in Guangdong Province, Southern China [19]. Given there is currently no available evidence to support the presence of any epidemic foci in Mainland China, most researchers purport that the high prevalence of dengue is due to imported cases [20–22]. However, the impact of inapparent infections on the emergence of DENV transmission may call this hypothesis into question if substantiated with appropriate epidemiological data. Therefore, during the 2013 DENV outbreak in Zhongshan, Guangdong Province, China, we conducted a cross-sectional study in order to better understand the dengue virus infection spectrum and to estimate the I:S ratio.
Study methods were reviewed and approved by the Zhongshan Center for Disease Control and Prevention Institutional Review Board. All study participants provided informed consent. The aims of our study were explained, and written informed consent was obtained from all participants. For children less than 16 years of age, consent from the parent or guardian was obtained.
Zhongshan is located within the Pearl River Delta Region of Guangdong Province, specifically on the west bank of the estuary of the Pearl River. It is geographically connected with Guangzhou (the capital of Guangdong Province) on the north and in vicinity to Hong Kong and Macao. Zhongshan covers an area of 1,800 sq km, consisting of 4 urban and 20 suburban towns (Fig. 1) with a population of 3.15 million. The city is located in a subtropical humid zone with an annual average temperature around 23.0°C and an annual rainfall of 1,791 mm. Vector surveillance data show Aedes albopictus to be the predominant mosquito species found throughout the region, which emerge in highest abundance during the rainy season between May and October [23,24]. Since the re-emergence of dengue in 1979, periodic outbreaks, with interepidemic intervals of 2–7 years, have since occurred [25,26]. Sparse detection of dengue IgG antibodies through serological surveillance of healthy individuals and a low number of clinical cases suggest that no outbreaks occurred during 2007–2012 [23].
The study was divided into 2 phases (Fig. 2). In phase 1, active dengue surveillance was initiated on July 17, 2013 at select hospitals in Zhongshan. This was after a cluster of 15 dengue laboratory confirmed cases, occurring within 6 days of each other, were reported in Huangpu, which is located in northern Zhongshan. Based on the WHO surveillance recommendations, individuals were screened for DENV if they met a case criterion of acute febrile illness, fever (38°C or above), and at least 2 of the following symptoms: headache, retro-orbital pain, myalgia, arthralgia, rash, flush, hemorrhagic manifestations, leucopenia or thrombocytopenia. If it was within 5 days of fever onset, patients were tested using a commercially available RT-PCR kit (Shanghai ZJ Bio-Tech Co., Ltd.) to identify DENV RNA. After 5 days, a commercial IgM and IgG ELISA kit (Zhongshan Bio-Tech Co., Ltd.) were instead used to detect IgM and IgG antibodies against DENV. According to the Chinese national criteria for dengue diagnosis (WS216-2008), those who were either RT-PCR positive or IgM positive were classified as confirmed dengue cases.
The first confirmed dengue case within a community, without reported travel history to an epidemic focus within 14 days before symptom onset, was eligible to enroll as an index case. Once an index case was identified, local public health workers would visit the area surrounding the home of the index case within 5 days to invite individuals with potential co-exposure to participate in the study. A co-exposure was defined as any persons living in the same household as the index case, or a neighbor living within 100 meters of the index case’s home. Upon enrollment, participants were consented and administered a questionnaire to evaluate if they had experienced any dengue-like symptoms in the previous 30 days. Axilla-temperature was then recorded and 5ml of blood collected. Samples collected from individuals who reported no dengue-like symptoms were then tested for IgM and IgG antibodies against DENV by ELISA (Zhongshan Bio-tech) and samples collected from those who did report dengue-like symptoms were tested for dengue viral RNA using RT-PCR. Clusters were then stratified into the following categories based upon the household characteristics of the index case: construction site, factory, migrant rental, or newly-built community.
Phase 2 was then conducted in April 2014, right before the usual onset of dengue cases. Six towns were selected in total. Huangpu, Guzhen, Xiaolan, and Dongfeng each had documented transmission in 2013, while the towns of Sanxiang and Tanzhou had no autochthonous confirmed cases the same year. Sequential sampling was conducted in each town among apparently healthy individuals seeking a routine physical at local hospitals. A total of 1,167 participants were recruited in this manner. For each participant, a blood sample was collected and tested for IgM and IgG antibodies against DENV using an ELISA kit (Zhongshan Bio-tech), and a questionnaire was administered to retrospectively evaluate if dengue-like symptoms were experienced between June and November of 2013.
According to the dengue surveillance guide of China, imported dengue cases were defined as confirmed dengue cases with a travel history to dengue endemic countries in the past 14 days. In the cluster study, symptomatic dengue was defined as acute febrile illness plus an IgM or RT-PCR positive, while an inapparent dengue infection was defined as having no febrile illness and a positive dengue test (IgM, or RT-PCR). Dengue infections that were diagnosed by ELISA were further categorized as primary infection if the IgM/IgG ratio was ≥ 1.4 and secondary infection if the ratio was < 1.4. In the follow-up serosurvey, symptomatic infection was defined as individuals who retrospectively reported febrile illness during the outbreak period and had an IgG positive test, while an inapparent DENV infection was defined as having no febrile illness during the outbreak period and an IgG positive dengue test.
For cluster contacts, the infection rate was calculated using the equation: infection rate = (No. of recent infection / No. of cluster contacts)*100%. The inapparent to symptomatic (I:S) infection ratio was equal to the proportion of inapparent infection divided by the proportion of symptomatic infection for a given subgroup (RatioIS = PI/PS; PI = the proportion of inapparent DENV infection; PS = the proportion of symptomatic DENV infection). Confidence intervals (95%) for the I:S infection ratio were calculated using pI/ps×exp(±1.961n×pI+1n×ps). Statistical difference was determined if the 95% confidence intervals were not overlapping for the same variable. Statistical difference was determined if the 95% confidence intervals were not overlapping for the same variable.
Data for post-outbreak DENV IgG seroprevalence among towns, gender, age groups, and the reported incidence rate was analyzed. Chi-square testing and Spearman correlation analysis were conducted using SPSS 18.0.
Between July 12th and November 28th, 2013 there were 19 imported and 809 autochthonous dengue cases reported to Zhongshan CDC (Fig. 3). During this time, dengue cases peaked twice, once in July when the outbreak was first reported and a second time in October. The reported cases were distributed throughout 108 communities in 22 towns. The incidence rate varied substantially between towns, with the highest rate of 29.7 per 100,000 occurring in Huangpu (Fig. 1). Cases were more frequent among adults aged 50 years or above, while they were less frequent among children 0–14 years of age (Table 1).
Of the total reported dengue cases, 71 were identified and selected as index cases, distributed across 16 towns, with a median age of 33 years, and a similar composition of males and females (Fig. 1). For each cluster, 2–66 contacts (median 8) were enrolled, totaling 887, consisting of 43% females and an average age of 32 years, with 85% aged 15 years or above.
Serological analysis of the 887 cluster contacts showed 41 (4.62%) positive for DENV by either RT-PCR or IgM ELISA, indicating acute or recent dengue infection (95% CI: 3.24%-6.00%). Of these 41, 33 had IgM and IgG test, and only 1 (3.0%) was classified as a secondary infection. Thirteen (1.5%) of the 887 cluster contacts were identified as having a symptomatic DENV infection, while 28 (3.2%) were inapparent. The overall I:S ratio was 2.2:1 (95% CI: 1.1–4.2:1).
Table 2 summarizes the variation of the I:S ratio based on different characteristics of the cluster contacts. Contacts aged 50–91 years had the highest I:S ratio, compared to other age groups, though no significant difference between age groups was found. Three cases were detected among children aged 0–14 years, all of which were symptomatic. A similar I:S ratio was found for both male and female contacts.
When stratifying cluster contacts by type of DENV identified in the original index case, contacts of an enrolled index case positive for DENV-1 had a higher I:S ratio compared to contacts of an enrolled index case positive for DENV-3, though this difference was not considered significantly different. When stratified by type of housing, the I:S ratio among contacts from newly-built communities was the lowest (<1). Lastly, higher I:S ratios were associated with higher numbers of contacts in a particular cluster.
In April 2014, a total of 1,167 subjects from 6 towns were enrolled and serologically evaluated for IgG antibodies against DENV. The overall proportion of DENV IgG antibodies among subjects from the 4 towns with (Huangpu, Guzhen, Xiaolan, and Dongfeng) reported dengue circulation during 2013 was 2.7% (95% CI: 1.6%-3.8%), while the proportion among subjects from the 2 towns without (Sanxiang and Tanzhou) reported DENV transmission was 0.6% (95% CI: 0–1.4%). Table 3 shows the reported dengue cases, incidence rate, and DENV type from the 2013 outbreak and Table 4 shows the post-outbreak DENV IgG antibody results by participants’ town. Previous exposure (detection of DENV IgG antibodies) was significantly associated with participants’ town (Fisher exact χ2 = 15.24, p<0.01) and age (p = 0.02). Based on the Spearman correlation analysis, the proportion of DENV IgG antibody detection by town was positively correlated with the reported incidence rate (r = 0.88, p = 0.02). Subjects from Huangpu and Dongfeng had the highest proportion of DENV IgG antibody detection, while those from Sanxiang and Tanzhou had the lowest. The proportion of DENV IgG antibody detection among female subjects from Huangpu was significantly higher than that of males, while the proportion among males and females from Dongfeng were the same. The highest proportion of age-specific DENV IgG antibody detection among subjects from Huangpu and Dongfeng were in the 0–14 year age group. Additionally, there was no increasing trend of DENV IgG antibody detection by age, suggesting no cumulative exposure.
Of the 24 subjects who were positive for DENV IgG antibodies, none reported having dengue prior to 2013. While four subjects did report having acute febrile symptoms with headache between July and November 2013, only one met the WHO dengue case definition (described above) and was classified as a confirmed case. Assuming the baseline proportion of DENV IgG antibody detection by town was the same as Sanxiang and Tanzhou prior to 2013, the estimated dengue infection I:S ratio during the outbreak in Huangpu, Guzhen and Xiaolan, where DENV-1 circulated, was 11.0:1 (95% CI: 3.7-∞:1), and 1.0:1 (95% CI: 0.5-∞:1) in Dongfeng where DENV-3 circulated.
Prospective cohort studies are a common method for investigating the natural history of dengue infection in endemic regions [3–7,9]. However, it is often not feasible to conduct such studies in non-endemic regions, as it can be difficult to anticipate when and where an outbreak might occur. In this report, we closely investigated index clusters during a DENV outbreak, as well as conducted post-outbreak serological testing for DENV IgG antibodies This allowed us to better understand dengue transmission and to estimate the I:S ratio in a non-endemic city in Southern China. Study data show that adults accounted for the majority of the reported dengue cases and that individuals aged 50 years or more had the highest incidence rate. Additionally, serological analyses following the outbreak show that subjects from Sanxiang and Tanzhou, where no autochthonous dengue was confirmed in 2013, had the lowest proportion of DENV IgG antibody detection.
Previously conducted prospective cohort studies have often shown higher rates of inapparent versus symptomatic infections, with I:S ratios varying by geographic area, epidemiologic context, immunological status of patients, and types of circulating DENV [3–7,9,16,17]. Though our sample size was limited, our investigation does suggest that communities where DENV-3 primarily circulates have a lower I:S ratio than those where DENV-1 is the predominant virus type, which is consistent with a study conducted in Kamphageng Phet, Thailand [5]. Our investigation also suggests a higher I:S ratio among index clusters compared to those using similar study designs conducted in endemic regions [8,10,11,27]. This could be due to differences in DENV vector and host immunity. In contrast to Aedes aegypti found in endemic regions, Aedes albopictus is the only known vector for dengue in Zhongshan [23], and is thought to be a more susceptible vector of DENV with potential to transmit at low titers resulting in less clinically overt or severe disease [28]. Such was found by Gubler et al. from a comparative analysis of two dengue outbreaks caused by the same serotype. Patients with low levels of viremia were more commonly associated with the outbreak where dengue transmission was less explosive and clinical outcomes less severe [29]. Study data also show that the majority of individuals enrolled in this study experienced a primary DENV infection. When compared to other studies where the majority of the study population had secondary infections, the I:S ratio is markedly higher [3,10,27].
Post-outbreak serological data demonstrates a higher I:S ratio in towns where DENV-1 was the predominant circulating virus type, which was in stark contrast to what was found during the cluster investigation. This outcome could be due to differences in the sampling methods, which introduces the potential for recall bias as some mild febrile cases in the index cluster investigation were classified as inapparent infections. Yoon et al. found similar results in I:S ratios when comparing a prospective cohort with outcomes of the index cluster investigation [10].
It was interesting that no increasing trend of DENV IgG antibody detection by age was found, suggesting no cumulative exposure and that the majority of infections occurred during the 2013 outbreak. This finding is consistent with a previous study conducted by Zhongshan CDC showing sparse detection of DENV IgG antibodies between 2007 and 2012 [23]. The proportions of DENV IgG antibody detection among the 4 towns with confirmed dengue transmission were 7–64 times higher than the incidence rate calculated just from originally reported cases. Coupled with the I:S ratio estimates, it seems likely that a large number of inapparent infections and subclinical cases occurred during the outbreak, which could greatly influence the transmission dynamics of DENV in these areas [30,31].
There were several limitations worth noting. During the index cluster investigation, we were unable to perform a second dengue test following the outbreak, which could have resulted in a higher overall infection rate and subsequently a lower I:S ratio. We did not evaluate how distance between a contact and index cluster might influence DENV transmission. Also, estimates of dengue infection risk may have been impacted by some contacts living within 100 meters of the index cases refusing to participate. These refusal rates were estimated to be between 1% and 10% of the total number of possible enrollees. Lastly, some inapparent and symptomatic infections among individuals following the outbreak may have been misclassified due to the recall bias of the retrospective questionnaire and the inability to determine acute infections with an IgG antibody test.
Despite these limitations, to our knowledge, this is the first report of I:S ratios in China, allowing for a better understanding of the role of inapparent infections in driving DENV transmission in non-endemic regions where Aedes Albopictusis the only vector.
|
10.1371/journal.ppat.1007347 | Scavenger receptor-C acts as a receptor for Bacillus thuringiensis vegetative insecticidal protein Vip3Aa and mediates the internalization of Vip3Aa via endocytosis | The vegetative insecticidal proteins (Vip), secreted by many Bacillus thuringiensis strains during their vegetative growth stage, are genetically distinct from known insecticidal crystal proteins (ICPs) and represent the second-generation insecticidal toxins. Compared with ICPs, the insecticidal mechanisms of Vip toxins are poorly understood. In particular, there has been no report of a definite receptor of Vip toxins to date. In the present study, we identified the scavenger receptor class C like protein (Sf-SR-C) from the Spodoptera frugiperda (Sf9) cells membrane proteins that bind to the biotin labeled Vip3Aa, via the affinity magnetic bead method coupled with HPLC-MS/MS. We then certified Vip3Aa protoxin could interact with Sf-SR-C in vitro and ex vivo. In addition, downregulation of SR-C expression in Sf9 cells and Spodoptera exigua larvae midgut reduced the toxicity of Vip3Aa to them. Coincidently, heterologous expression of Sf-SR-C in transgenic Drosophila midgut significantly enhanced the virulence of Vip3Aa to the Drosophila larvae. Moreover, the complement control protein domain and MAM domain of Sf-SR-C are involved in the interaction with Vip3Aa protoxin. Furthermore, endocytosis of Vip3Aa mediated by Sf-SR-C correlates with its insecticidal activity. Our results confirmed for the first time that Sf-SR-C acts as a receptor for Vip3Aa protoxin and provides an insight into the mode of action of Vip3Aa that will significantly facilitate the study of its insecticidal mechanism and application.
| Bacillus thuringiensis Vip3A has potential in control of Lepidopteran pest and has been used in transgenic plants. However, studies of the insecticidal mechanisms of Vip3A are rare, and none of their definite receptors have been reported so far, which seriously restricts the study of its insecticidal mechanism and application. This work identified and confirmed the scavenger receptor class C like protein (Sf-SR-C) acts as the receptor of Vip3Aa protoxin, demonstrated that Sf-SR-C mediates the toxicity of Vip3Aa to Sf9 cells in an internalized manner. These results extend our understanding of SR-C proteins in insects and explain the specificity of Vip3Aa insecticidal activity, which strongly support it as a safe biopesticide. More importantly, it suggests the insecticidal mechanism of Vip3Aa different from the well-known “pore formation” model, “signal transduction” model, as well as newly found “necrosis” model of Cry toxins, which will significantly promote the relevant study of Vip3Aa. Last but not least, because scavenger receptors play a crucial role in innate immunity, our results provide relevant insights into host-pathogen interactions.
| Microbial insecticides, as substitutes for chemical pesticides, are alternatives for insect control in crops. Bacillus thuringiensis (Bt) is the most extensively used biopesticide worldwide due to its ability to produce insecticidal crystal proteins (Cry and Cyt toxins)[1–3]. The classical pore-forming model is the widely accepted mode of action of the three-domain crystal protein (3d-Cry) [1]. A signaling pathway model of the Cry toxin’s action has also been reported [4, 5]. Recently, Fengjuan et al. showed Cry6Aa could trigger the Caenorhabditis elegans death by necrosis signaling pathway [6]. In spite of differences, all three models agree that binding to host specific receptors is a key step in the process involved in cytotoxicity. Several types of receptors for Cry toxins have been reported, such as aminopeptidase N (APN), the cadherin-like proteins, alkaline phosphatases, and ABC transporter [1, 7, 8]. Bt has been used successfully to control many crop pests by transgenic plant or traditional spray approaches, however, many pests are not sensitive to Cry toxins and a number of cases of insect resistance to Cry toxins have been reported as a result of laboratory or field selections [1–3].
Vegetative insecticidal proteins (Vip), which are produced by Bt during its vegetative stages, share no sequence or structural homology with known Cry proteins, and have a wide spectrum of specific insecticidal activity, especially against lepidopteran pests [9]. Vip3 toxins have a different insecticidal process compared with Cry proteins, indicating they are likely to complement and extend the use of Bt insecticidal proteins. A synergistic effect of the toxins in Spodoptera frugiperda, Spodoptera albula, and Spodoptera cosmioides larvae was observed when Vip3Aa and Cry1Ia10 were combined [10]. Moreover, reports showed that transgenic cotton and corn co-expressing Vip3A and Cry1Ab, or Vip3A and Cry1Ac, provided excellent control of target insect species [3, 11–14] and no cross-resistance between Vip3A and Cry proteins was observed [3, 11, 12]. However, compared with Cry toxins, studies on the insecticidal mechanisms of Vip3A are scarce. Lee et al. proposed pore forming as the principal Vip3A mode of action [15]. Our previous work demonstrated that Vip3Aa induces apoptosis in cultured S. frugiperda (Sf9) cells [16]. Recently, Hernandez-Martinez et al. found that Vip3Aa could induce apoptosis in Spodoptera exigua midgut epithelial cells [17]. Reports also showed that Vip3A can not bind to the APN and cadherin-like protein [15]. Instead, it binds to proteins of susceptible insect’s midgut, which are distinct from the known Cry receptors [15, 18]. So far, almost nothing is known on Vip definite receptors except for their molecular weight.
Previously, we and Singh et al. have found Vip3A protoxin has cytotoxicity to S. frugiperda cells (Sf9 cells and Sf21 cells) [16, 19]. Therefore, we speculated that there are receptors for Vip3Aa in Sf9 cells membrane. In this study, to identify the receptors of Vip3Aa protoxin, we analyzed the Sf9 cells membrane proteins that bind to the biotin labeled Vip3Aa, via the affinity magnetic bead method coupled with nano-HPLC electrospray ion trap mass spectrometry (HPLC-MS/MS). We paid more attention to the scavenger receptor class C like protein (Sf-SR-C) from the 70 identified proteins due to class C scavenger receptors (SR-C) are membrane proteins and have only been identified in insects [20]. We investigated whether Sf-SR-C is the receptor of Vip3Aa both in vitro and ex vivo. Furthermore, we detected which domain of Sf-SR-C participates in the interaction with Vip3Aa, and validate whether Sf-SR-C mediates the internalization of Vip3Aa, since we observed the presence of Vip3Aa in the cytoplasm of Sf9 cells. Our data confirmed Sf-SR-C acts as the receptor of Vip3Aa, demonstrated the complement control protein (CCP) domain and MAM domain of Sf-SR-C interact with Vip3Aa, and further revealed that endocytosis of Vip3Aa mediated by Sf-SR-C correlates with its insecticidal activity. These results will significantly promote the study and application of Vip3Aa.
In addition to the significant virulence effect of Vip3Aa to S. frugiperda Sf9 cells [16, 19], we also found that Vip3Aa-RFP (a fusion protein of Vip3Aa protoxin and red fluorescence protein) could bind to the Sf9 plasma membrane as shown by colocalization with FITC-phalloidin and internalize in endosomes, while the RFP itself could not (Fig 1A and S1A Fig). Thus, to identify the receptors of Vip3Aa protoxin in Sf9 cells, biotin labeled Vip3Aa (Bio-Vip3Aa) (S1B Fig) was incubated with the extracts of Sf9 cell membrane proteins, immunoprecipitated with Streptavidin Mag Sepharose, and detected by Coomassie brilliant blue staining (Fig 1B, a). The rest of the bands were analyzed by HPLC-MS/MS after the band corresponding to Vip3Aa was excised (Fig 1B, b). Protein sequence database searching of the MS/MS spectra revealed that the bands represented 70 proteins (S1 Dataset), which included 33 ribosomal proteins (in which ribosomal protein S2 had been reported as an interacting partner protein of Vip3A by Singh et al. [19]) and 37 other proteins, including Sf-SR-C (S1C Fig).
At present, class C scavenger receptors (SR-C) have only been identified in insects [20] and only described in Drosophila melanogaster [21, 22]. First, the SR-C like gene was cloned from the cDNA of Sf9 cells and named as the Sf-SR-C gene (GenBank accession no. KX925839). We then purified the extracellular sequence of Sf-SR-C (aa 20–558) (Sf-SR-C-N) with a glutathione-S-transferase (GST) tag (GST-SR-C-N) (Fig 1C). A GST pulldown assay demonstrated that GST-SR-C-N could bind to Vip3Aa-Flag, but could not bind to the control protein Cry1Ac (Fig 1D). To assess the binding affinity between Vip3Aa protoxin and GST-SR-C-N, we used a microscale thermophoresis assay (MST) in which biomolecular interactions are quantitated by examining the motion of the molecules along a microscopic temperature gradient induced by an infrared laser [23, 24]. The estimated dissociation constant (Kd) was 190 ± 75 nM (Fig 1E). To further test whether full-length Sf-SR-C can interact with Vip3Aa, Sf-SR-C was then transiently expressed in Sf9 cells with a V5 tag after ligation into plasmid pIZT/V5-His (pIZT-SR-C) (Fig 1F). Immunoprecipitation analysis using the anti-V5 antibody showed that Vip3Aa-Flag could be co-immunoprecipitated with Sf-SR-C-V5 (Fig 1G). In the control experiment, we could not detect Cry1Ac after it was incubated with the lysate of Sf9 cells transfected with pIZT-SR-C (Sf9-pIZT-SR-C cells). Ligand blotting was used to detect the specific binding of Sf-SR-C to Vip3Aa. As shown in Fig 1H, Vip3Aa-Flag could bind to Sf-SR-C-V5 and excess Vip3Aa (200-fold) competed for Vip3Aa-Flag binding with Sf-SR-C-V5, which further indicated that Vip3Aa and Sf-SR-C can bind specifically. In addition, via the affinity magnetic bead method, immunoblotting revealed that the Sf-SR-C-V5 from the lysate of Sf9-pIZT-SR-C cells could interact with biotin-labeled Vip3Aa-Flag (S1D Fig). In contrast, Sf-SR-C-V5 could not interact with control biotin labeled ChiB-flag (Chitinase B secreted by Bt). These results indicated Vip3Aa protoxin can interact with Sf-SR-C in vitro.
To verify the role of Sf-SR-C in Vip3Aa protoxin binding to Sf9 cells in more detail, we attempted to generate Sf9 cells in which the expression of endogenous Sf-SR-C gene was reduced. Two plasmids, pIZT-SRi1 and pIZT-SRi2, which can generate fragments of double-stranded RNA (dsRNA) from the Sf-SR-C gene (S2A and S2B Fig) [25], were stably transfected into Sf9 cells, respectively, which resulted in the generation of Sf-SRi1 and Sf-SRi2 cell lines. As the quantitative real-time reverse transcription PCR (qRT-PCR) result shown in Fig 2A; the expression level of the Sf-SR-C gene was markedly reduced in the Sf-SRi1 and Sf-SRi2 cells compared with the Sf9 cells and the cells stably transfected with pIZT/V5-His (Sf-pIZT cells). Consistent with this, a CCK-8 cytotoxicity assay results showed that the cytotoxic effects of Vip3Aa on the Sf-SRi1 and Sf-SRi2 cells were also clearly reduced compared with those on Sf9 cells and Sf-pIZT cells (Fig 2B).
Next, we carried out co-localization assays to detect the interaction between Vip3Aa and Sf-SR-C. After treating the Sf9 cells with Vip3Aa-RFP for 6 h, we monitored the Vip3Aa and Sf-SR-C distribution using immunofluorescent staining. The anti-Sf-SR-C-N polyclonal antibody and Alexa Fluor 488-conjugated anti-rabbit antibody were used to show the location of Sf-SR-C in Sf9 cells. As shown in Fig 2C, most of the dots of Vip3Aa-RFP were co-located with Sf-SR-C, especially in the dots that were Sf-SR-C-rich. In the control experiment that the anti-GST polyclonal antibody was used, we detected almost no green fluorescence.
We also observed that Vip3Aa-RFP has almost no affinity for Drosophila S2 cells (S2 cells) (S3A Fig). We therefore cloned the gene of Sf-SR-C into plasmid pAc5.1/V5-HisB (pAc-Sf-SR-C) and transiently transfected it into S2 cells (S2-Sf-SR-C cells) to examine the specific interaction of Vip3Aa and Sf-SR-C in S2 cells. The ribosomal S2 protein of Sf9 cells (Sf-S2) was also heterologously expressed into the S2 cells (S2-Sf-S2 cells) as a control (S3B Fig). The Dylight 488 conjugated anti-V5 antibody was used to show the heterologously expressed protein in S2 cells. After treating the S2 cells with Vip3Aa-RFP for 12 h, immunofluorescent staining showed that Vip3Aa-RFP could clearly bind to the S2-Sf-SR-C cells, and the dots of Vip3Aa-RFP were co-located with the dots that were rich in Sf-SR-C (Fig 2D and S3C Fig), similar to the phenomenon that Vip3Aa-RFP binds to Sf-SR-C in Sf9 cells. In contrast, we didn’t detect the interaction between Vip3Aa-RFP and S2-Sf-S2 cells, nor did we find the binding of RFP to S2-Sf-SR-C cells. In addition, the cytotoxicity assay showed Vip3Aa protoxin is more toxic to S2-Sf-SR-C cells than to S2-Sf-S2 and S2 cells (Fig 2E) (The transfection efficiency was about 30%). Taken together, these results revealed that Sf-SR-C could also interact with Vip3Aa protoxin ex vivo.
Vip3Aa has a high affinity for IOZCAS-Spex-II-A cells (Spodoptera exigua cells) (S3A Fig) and shows a significant toxic effect to S. exigua [10]. We also cloned two partial sequences with similarity to the Sf-SR-C gene from the total cDNA of S. exigua cells (GenBank accession no. KY829113 and MF969248). Therefore, we attempted to use ingestion of bacterially expressed dsRNA to reduce the expression of the S. exigua larvae midgut SR-C gene (Se-SR-C) to detect whether it affected the toxicity of Vip3Aa to the larvae. The vector pET-Se-SRi, which expresses a partial dsRNA of the Se-SR-C gene (S2C Fig), was transformed into bacterial strain HT115 (DE3), which lacks RNase III activity to express dsRNA fragment of Se-SR-C (HT-pET-Se-SRi) [26] (S4A Fig). The vector pET-Hypi, which expresses a partial dsRNA of a hypothetical protein (Hyp) (GenBank: PCG66164.1), and the blank plasmid pET28a were transformed into the HT115 strain as control (HT-pET-Hypi and HT-pET28a). The qRT-PCR results showed that after continuous feeding of the S. exigua larvae with the strains for 7 days (Fig 3A and S4B Fig), the transcription level of the Se-SR-C gene of the larvae fed with the HT-pET-Se-SRi strain was effectively inhibited compared with the control (Fig 3B). The larvae were then exposed to Vip3Aa and the strains for another 5 d to detect the survival rate. The bioassay results shown in Fig 3C indicated that the mortality rate of the larvae in which the Se-SR-C gene was silenced was clearly lower than that of the control, which suggested that reducing the expression of the Se-SR-C gene in S. exigua lavae decrease their sensitivity to Vip3Aa.
Vip3A has high insecticidal activity against Lepidopteran rather than Dipteran [9]. To further examine the interaction of Sf-SR-C with Vip3Aa in an insect that is insensitive to Vip3Aa (S4C Fig), we constructed transgenic Drosophila that overexpressed Sf-SR-C using the esg-Gal4 tub-Gal80ts system. In this system, the esg-Gal4 driver is mainly active in the midgut cells of Drosophila and Gal4 is under the control of a temperature sensitive Gal80 that allows the conditional induction of the UAS-linked Sf-SR-C gene [27]. After culturing at 25 °C for 4 d, the fly strains were shifted to 29 °C (Gal4 ‘‘on”) or 18 °C (Gal4 ‘‘off”) for 3 d (Fig 3D). The about 2-day-old larvae were then treated with Vip3Aa or dialysis buffer for 48 h and the survival rates were detected (Fig 3D and S4D Fig). As shown in Fig 3E, the Drosophila larvae that overexpressed Sf-SR-C in their midgut (esgts>SR-Cvk33) (Red) had a significantly higher mortality rate after exposure to Vip3Aa compared with the control group, which was treated with dialysis buffer. In the group of esgts (Green) and UAS-SR-Cvk33 (Purple), which could not express Sf-SR-C, Vip3Aa showed no obvious toxicity to the larvae compared with the control. Moreover, shutdown of the expression of Sf-SR-C in the gut epithelia of the larvae (18 °C treated) eliminated the toxicity of Vip3Aa to the larvae (Blue). These results further indicated Sf-SR-C is the receptor for Vip3Aa, which causes the death of sensitive insects.
From BLASTP analysis, we found that the protein sequence of Sf-SR-C was not similar to the SR-C from D. melanogaster (dSR-CI) (only about 27% sequence identity). However, the extracellular sequence of Sf-SR-C has four structural domains that are similar to dSR-CI, including the CCP, MAM, somatomedin B, and Ser/Thr rich domains (Fig 4A). To further investigate which domain of Sf-SR-C mainly participates in the interaction with Vip3Aa protoxin, the extracellular sequence of Sf-SR-C was divided into three parts (SR-F-1, SR-F-2, and SR-F-3) (Fig 4A) and expressed as fusion proteins with GST. Dot blotting analysis revealed that GST-SR-F-1 (aa 20–138), which contains the CCP domain (aa 26–76), and GST-SR-F-2 (aa 139–320), which is the MAM domain, could bind to Vip3Aa-Flag, while GST and GST-SR-F-3 (aa 321–558) could not (Fig 4B). Furthermore, excess Vip3Aa (500-fold) competed for Vip3Aa-Flag binding with GST-SR-F-1 and GST-SR-F-2 (Fig 4B). Moreover, pulldown experiments also revealed that GST-SR-F-1 and GST-SR-F-2 could directly interact with Vip3Aa-Flag (Fig 4C). These results indicated that the binding of SR-F-1 and SR-F-2 with Vip3Aa-Flag was specific. GST-SR-F-1 contained regions other than the CCP domain; therefore, we further purified Sf-CCP (CCP domain of Sf-SR-C (aa 20–80) with a His-tag) to detect the interaction with Vip3Aa-Flag, and the Dm-CCP (CCP domain (aa 20–80) of dSR-CI (GenBank: U17693.1)) was used as a control. Both dot blotting analysis (Fig 4D) and pulldown assays (Fig 4E) verified the physical interaction between Sf-CCP and Vip3Aa-Flag. The results also showed that the Dm-CCP domain could not bind to Vip3Aa-Flag, which further validated the specific binding between Vip3Aa and Sf-SR-C. Furthermore, MST was also applied to assay the binding affinity of Vip3Aa protoxin with Sf-CCP and MAM domains (Fig 4F and 4G). The determined Kd values were 2.19 ± 1.55 μM and 463 ± 117 nM, respectively. These results certified the CCP and MAM domains of Sf-SR-C could bind to Vip3Aa protoxin.
As Figs 1A and 2C showed above, we observed red dots in the cytoplasm of Sf9 cells after exposing them to Vip3Aa-RFP, which suggested the internalization of Vip3Aa. We first used several inhibitors of endocytosis to test whether Vip3Aa-RFP could enter the Sf9 cells via endocytosis [28, 29]. As shown in Fig 5A and 5B, dynasore, which is an inhibitor of dynamin, could significantly inhibit Vip3Aa-RFP entry into Sf9 cells. The known macropinocytosis inhibitors, amiloride, cytochalasin D, LY294002, and wortmannin also reduce the number of red dots inside the Sf9 cells. However, among two inhibitors of clathrin-mediated endocytosis (chlorpromazine and monodansylcadaverine) and two inhibitors of clathrin-independent endocytosis (nystatin and cholesterol-oxidase), only monodansylcadaverine could reduce the number of red dots in Sf9 cells; the others had no effect on the number of Vip3Aa-RFP dots in the Sf9 cells compared with the control. These results suggested Vip3Aa enter Sf9 cells through dynamin-dependent and macropinocytosis-related endocytosis.
One of the main functions of scavenger receptors (SRs) is endocytosis [20]. Thus, we hypothesized that Vip3Aa enters the Sf9 cells via endocytosis mediated by Sf-SR-C. To further verify Sf-SR-C mediated the internalization of Vip3Aa, purified anti-Sf-SR-C-N polyclonal antibodies were incubated with the Sf9 cells for 1 h and the cells were then co-incubated with Vip3Aa-RFP for another 6 h. The results showed that the number of red dots in the cytoplasm of Sf9 cells was reduced visibly after treatment with the anti-Sf-SR-C-N polyclonal antibody, while there are many red dots in cells treated with anti-GST polyclonal antibodies (Fig 6A and 6D). We also quantified the number of red dots in the Sf-SRi1 and Sf-SRi2 cell lines. Compared with the Sf-pIZT cells, the internalization of Vip3Aa-RFP also reduced markedly in the Sf-SRi1 and Sf-SRi2 cells (Fig 6C and 6D). Furthermore, because we found that Vip3Aa can bind to GST-SR-F-1 and GST-SR-F-2, Vip3Aa-RFP combined with an excess of GST-SR-F-1 and GST-SR-F-2 (20-fold) were exposed to Sf9 cells, respectively. The competitive binding assay showed the amount of red dots in the Sf9 cells treated by Vip3Aa-RFP and GST-SR-F-2 was significantly decreased compared with the control cells treated with Vip3Aa-RFP and GST (Fig 6B and 6D). However, in the case of GST-SR-F-1, such phenomenon did not occur, which suggested the MAM domain might play more critical role in the internalization of Vip3Aa than the CCP domain. These results indicated the Sf-SR-C mediates the internalization of Vip3Aa via endocytosis.
The above results showed that silencing of Sf-SR-C gene could clearly reduce the toxicity of Vip3Aa to Sf9 cells (Fig 2B) and also reduce the amount of Vip3Aa entering into Sf9 cells (Fig 6B and 6D), which suggested the amount of Vip3Aa entering cells is directly related to its toxicity. Therefore, we carried out further experiments to verify this speculation. Firstly, we have demonstrated that the endocytosis inhibitor dynasore could significantly inhibit the internalization of Vip3Aa, without affecting the binding of Vip3Aa to Sf9 cells (Fig 5A and 5B and S5 Fig). Through cytotoxicity assay (Fig 7A), we further found that dynasore (4μM) markedly decreased the toxicity of Vip3Aa to Sf9 cells while reducing the entry of Vip3Aa into cells. Dynasore alone did not affect the survival of Sf9 cells. In addition, in the Fig 5A and 5B, we found that the DMSO (0.1%) had a tendency to increase the number of Vip3Aa into Sf9 cells. So we explored the highest concentration of DMSO that did not cause toxicity to Sf9 cells. As shown in Fig 7B and 7C, we found that when the concentration of DMSO increased to 0.6% (v/v), it could clearly increase the number of Vip3Aa entering Sf9 cells. Moreover, the cytotoxicity assay also showed that DMSO increased Vip3Aa's toxicity to Sf9 cells while promote the internalization of Vip3Aa (Fig 7D), and DMSO alone had no obvious toxicity to Sf9 cells. These results further demonstrated that the internalization of Vip3Aa is directly related to its toxicity. Taken together, our results indicated the induced mortality of Vip3Aa in Sf9 cells correlated with its endocytosis mediated by Sf-SR-C.
Vip3Aa proteins have been studied for more than 20 years since they were first found by Estruch et al. in 1996 [30]. They are considered as novel insecticidal toxins secreted by Bt because they have different insecticidal properties compared with known Cry toxins and have a broad insecticidal spectrum within Lepidoptera [9]. To date, more than 138 Vip proteins have been found and were divided into four categories according to the classification of Bt Toxin Nomenclature Committee [31]. However, there has been no report of a definite receptor for Vip toxins up to now. In this paper, via HPLC-MS/MS, 70 potential binding proteins of Vip3Aa, including ribosomal protein S2 and actin, were identified (S1 Dataset). Singh et al. identified ribosomal protein S2 as a toxicity-mediating interacting partner protein of Vip3A in Sf21 cells [19]. However, as an intracellular protein, S2 protein is not likely to be a receptor of Vip3A. That maybe why Singh et al named it interacting partner protein, not a receptor. Our results also showed Vip3Aa could not bind to the S2-Sf-S2 cells, which heterologously expressed Sf-S2 into the S2 cells, and had no obvious cytotoxicity to them (Fig 2D and 2E). It suggests that S2 protein is not a receptor for Vip3Aa. Actin was identified as a novel Cry1Ac binding protein in Manduca sexta midgut through proteomic analysis [32]. For the same reason, it is unlikely that this protein is serving as a receptor for Vip3Aa. We speculate that the Vip3Aa may interfere with the function of the ribosome and actin after entering the cells.
Estruch et al. mentioned that a 48-kDa protein from Agrotis ipsilon with homology to tenascins may act as the receptor of Vip3A in their patent [33]. However, they did not provide any supporting data and no subsequent reports proved their speculation. Furthermore, consistent with previous reports [9], we did not find the receptors for Cry toxins such as APN or cadherin-like proteins in the 70 proteins we identified, which suggested Vip3Aa share no binding sites with Cry toxins. Moreover, we also demonstrated that Cry1Ac could not bind to Sf-SR-C (Fig 1D and 1G). These results further strengthen the viewpoint that Vip3 toxins and Cry toxins have different mechanisms of action, which makes Vip3 toxins good candidates for combination with Cry toxins in transgenic plants to prevent or delay insect resistance and to broaden the insecticidal spectrum.
In this study, we provide in vitro, ex vivo, and bioassay evidences for the first time confirming that the SR-C-like protein Sf-SR-C from Sf9 cells is the receptor for Vip3Aa. Scavenger receptors are cell surface receptors that typically bind multiple ligands and promote the removal of non-self or altered-self targets. SRs are classified into 10 classes [20]. At present, the vast majority of SRs have been identified and studied in mammals; however, SR-C have only been found in insects and have only been described in Drosophila [21, 22]. In mammalian cells, SRs play a crucial role in maintenance of host homeostasis, and also participate in host immune responses and metabolism. They can recognize and bind to a broad spectrum of ligands, including modified and unmodified host-derived molecules or microbial components [21, 34]. However, researchers also found that pathogens have evolved mechanisms to subvert SRs’ function to infect host cells [34]. For example, hepatitis C virus [35], enterovirus 71 (EV71) [36], and coxsackievirus (CVA7, CVA14 and CVA16) [37] utilize class B receptors to infect host cells. Chlamydia trachomatis uses the lipid transfer activity of SR-B1 for survival in host cells [38]. Even the class B scavenger receptor CD36, which has been implicated in the clearance of several bacterial and protozoan pathogens, has been reported to be co-opted by mycobacteria [39]. In D. melanogaster, SR-CI was identified as the recognition receptor for acetylated low-density lipoprotein [22] and bacteria [40]. In addition, Philips et al. found that Peste in D. melanogaster (a CD36 homolog) is involved in the uptake of mycobacteria into host cells [41]. In this study, we provide another example, in which the bacterial toxin Vip3Aa can exploit Sf-SR-C of Sf9 cells to kill host cells.
In addition, the Vip1 and Vip2 proteins which were first found in Bacillus cereus are regarded as binary toxins. Vip1 is speculated as the binding component and triggers endocytosis, and Vip2 enters the cell and exerts its toxic effect [9]. Vip3Aa has no sequences similarity to Vip1 or Vip2, however, our results certified Vip3Aa can entry into the Sf9 cells by itself via the endocytosis mediated by Sf-SR-C.
In insects, there has been little research into the mode of action of SR-C. However, in mammals, one of the main functions of SR proteins is endocytosis, which can trigger a series of signaling pathways [21, 34]. More interestingly, SR function is increasingly linked to apoptosis in a wide variety of cell types. Binding of fucoidan ligand by the macrophage SR-A1 triggers endocytosis by caveolae-dependent pathways, which stimulates apoptosis via a p38 MAPK and JNK dependent intracellular signaling pathway [42]. In vascular cells, thrombospondin-1 activation of SR-B2 triggers downstream signaling through p38 MAPK and caspase dependent pathways with increased apoptosis [43]. In addition, SR-E1 function is linked to apoptosis in the endothelium, vascular smooth muscle cells, macrophages, epithelial cells and neurons. [44,45]. As mentioned above, some pathogens can utilize the function of SRs to invade the host cell. Some toxins secreted by bacteria can also entry into host cell via endocytosis to exert their toxic effects. Diphtheria toxin, an exotoxin secreted by Corynebacterium diphtheriae and causes the disease diphtheria in humans, is believed to enter toxin-sensitive mammalian cells by receptor-mediated endocytosis and inhibit protein synthesis of host cells [46, 47]. In this way, it acts as a RNA translational inhibitor and results into cell apoptosis. Receptor-mediated endocytosis is required for efficient expression of toxicity. Once endocytosis was inhibited, the cytotoxicity of diphtheria toxin was blocked accordingly [46, 47]. Our results indicated that the toxicity of Vip3Aa to Sf9 cell correlated with its endocytosis mediated by Sf-SR-C (Figs 5, 6 and 7). It suggested that internalization is essential for Vip3Aa to exert its toxic effects.
Endocytosis is mentioned in Cry5- Caenorhabditis elegans system. In that case, however, endocytosis is a protection strategy utilized by worms to against the toxin Cry5 [48]. As to the “signal transduction” model, endocytosis is not an indispensable step [4]. In newly found “necrosis” model, Cry6A toxin is also internalized into intestinal cells, but cell death induced by Cry6Aa does not depend on the apoptotic mechanism [6]. These further implied the mode of action of Vip3Aa toxins different from that of Cry toxins. However, the more detailed mechanisms of how Vip3Aa kills Sf9 cells after interacting with Sf-SR-C and the follow-up connection with our previous results that Vip3Aa can induce apoptosis in Sf9 cells [16], are complex and interesting and will require further study.
In cytotoxicity assays, we used Vip3Aa protoxin. However, we found that the purified Vip3Aa protoxin was unstable. As shown in S1B Fig, in the biotin labeled Vip3Aa (lane 3), we can see the emergence of the activated Vip3Aa like protein (about 66 kDa). After incubating Vip3Aa protoxin with Sf9 cells, western blotting revealed that the activated Vip3Aa like protein was also apparent in the medium (S6 Fig). Therefore, it is difficult to exclude the existence of activated Vip3Aa in the process of toxicity testing. Lee et al. have already demonstrated that the activated Vip3Aa has the pore formation activity. In contrast, the full-length Vip3Aa protein was unable to form pores [15]. They proposed formation of ion channels as the principal mode of action of activated Vip3Aa. However, in this work, we have demonstrated that full-length Vip3Aa could bind to the Sf-SR-C receptor and endocytosis of Vip3Aa correlates with its toxicity.
In addition, the activated Vip3Aa protein was considered to correspond to the C terminus of the Vip3Aa protoxin (from amino acid 199 to the end) [9, 33]. We also purified the activated Vip3Aa protein (Vip3Aa-199) (S7A Fig). Through cytotoxicity assay, we found that although the activated Vip3Aa is also toxic to Sf9 cells, it is obviously less than that of Vip3Aa protoxin (S7B Fig). So we think that the Vip3Aa protoxin plays a major toxic role in cytotoxicity assays. Furthermore, we found that endocytosis of Vip3A was almost completely inhibited after treated with dynasore (Fig 5B). Meanwhile, the toxicity of Vip3Aa was decreased clearly but not reduced accordingly (Fig 7A). It implied that endocytosis is critical for Vip3Aa to exert its toxic effects, but it may not be responsible for all the toxicity of Vip3Aa. Recently, Tabashnik et al. proposed a new model for Bt mode of action named “dual model”, where both the protoxin and activated Cry toxin forms can kill insects, with each form exerting its toxic effect via a different pathway [49]. This contrasts with what “classical model” in which protoxins are inactive. Whether Vip3Aa protoxin and the activated toxin use the different mechanism of action and whether Vip3A have other mechanisms of insecticide need further study.
To date, SR-C has only been described in Drosophila. The present study cloned and identified another SR-C gene in S. frugiperda. Moreover, we also cloned two other fragments from S. exigua cells, which have high sequence and structural homology with Sf-SR-C. This indicated that SR-C also exists in S. exigua and further extends the range of SR-C in insects. Our results showed that SR-C can be detected in Sf9 cells and S. exigua cells, as well as in Drosophila cells. However, only the former two types of cells have high affinity for Vip3Aa (Fig 1A and S3A Fig). We hypothesized that subtle differences in the sequence and the three-dimensional structure of the protein might influence their interaction with Vip3Aa. Consistent with our conjecture, Vip3Aa-Flag can bind to Sf-CCP but not to DM-CCP (Fig 4D and 4E). Furthermore, from the sequence alignments, we found that Sf-SR-C has no sequence and structural homology with known proteins from vertebrates, as well as with known Cry toxins receptors. The results presented here provide a plausible molecular basis for the lack of toxicity of Vip3A toxins toward non-target insects and vertebrates, and strongly support its use as a safe biopesticide. In addition, because SRs play a crucial role in innate immunity and in the pathogenesis of various diseases in mammals [34], our study might extend our understanding of SR-C proteins and provide other avenues for studying host-pathogen interactions.
What’s more, although reducing the expression of the Se-SR-C gene clearly reduced the toxicity of Vip3Aa to the larvae compared with that of the control (Fig 3C), the effect sizes between the larvae of Se-SR-C gene silencing and the control groups were not as obvious as expected. This implied that there may be other receptors for Vip3Aa contributing to the overall toxicity. Just like several receptors for Cry toxin have been discovered [1, 7, 8], some reports also found Vip3Aa could bind to different molecular weight proteins in the brush border membrane vesicles of susceptible insects, such as the 55, 65, 80, 100 and 110 kDa proteins [15, 50–52], which further indicated the existence of different kinds of receptors for Vip3Aa. Moreover, we have identified 36 other proteins besides the ribosomal proteins and Sf-SR-C from the extracted Sf9 cell membrane proteins which could interact with the Vip3Aa. Whether or not there are other receptors play roles, sequentially or simultaneously, in killing insect process, further in-depth studies are needed.
In conclusion, the present study identified and confirmed Sf-SR-C as the receptor for Vip3Aa, proved the CCP and MAM domains of Sf-SR-C interact with Vip3Aa, analyzed the binding specificity between Vip3Aa and Sf-SR-C, and certified Sf-SR-C mediate the internalization of Vip3Aa via endocytosis. Our results contribute to the understanding of the mode of action Vip3Aa and significantly facilitate the further study of its insecticidal mechanism and application.
E. coli DH5α for plasmid constructions and E. coli BL21 (DE3) for protein purification were cultured at 37 °C in lysogeny broth (LB) or agar. Bt9816C was previously isolated and maintained in our laboratory for generation of Vip3Aa [53]. The Drosophila S2 cells, S. frugiperda Sf9 cells and S. exigua cells (IOZCAS-Spex-II-A) were maintained and propagated in Sf-900 II SFM (Invitrogen) or SFX-Insect (HyClone) culture medium at 27 °C. Spodoptera exigua and Drosophila strains were used for the bioassays.
Drosophila genotypes used were: esgts: esg-Gal4, tub-Gal80ts, UAS-GFP/cyo; Tm2/Tm6B. UAS-SR-Cvk33: SP/Cyo; 10×UAS-SR-Cvk33/Tm6B. esgts>SR-Cvk33: esg-Gal4, tub-Gal80ts, UAS-GFP/Cyo; UAS-SR-Cvk33/Tm2.
To ectopically express Sf-SR-C in Drosophila, the primers pJF-Sf-SR-C-F and pJF-Sf-SR-C-R were used to clone the Sf-SR-C gene, which was then recombined with the linearized vector pJFRC2-10XUAS-IVS-mCD8::GFP (Addgene plasmid #26214) using a ClonExpress II One Step Cloning Kit (Vazyme). Transgenic lines were established through microinjection of the transgene DNA into embryos of PhiC31-mediated chromosome-integrated Drosophila strains PBac {y[+]-attP-3B} VK00033 [54].
Primary antibodies: Mouse anti-Flag (Cell Signaling 8146), rabbit anti-V5 (Cell Signaling 13202), rabbit anti-His (Cell Signaling 12698), anti-Sf-SR-C-N polyclonal antibodies were generated by immunizing rabbits with purified GST-SR-C-N. Secondary antibodies: goat anti-mouse IgG-HRP conjugate (Santa Cruz sc-2005), goat anti-rabbit IgG-HRP conjugate (Cell Signaling 7074), rabbit anti-GST (Polyclonal, Bioss bs-2735R). The primary antibodies and secondary antibodies were used at 1: 1000 for western blotting. For immunostaining assays, Anti-V5-Dylight 488 conjugate (Invitrogen MA5-15253-D488, 1:200) and Alexa Fluor 488 goat anti-rabbit IgG (Cell Signaling 4412, 1:200) were used.
Inhibitors: Dynasore (dynamin inhibitor, TargetMol T1848, 7.5μM), chlorpromazine (clathrin-mediated endocytosis inhibitor, Millipore 215921, 40μM), monodansylcadaverine (clathrin-mediated endocytosis inhibitor, Sigma D4008, 150μM), nystatin (sequesters cholesterol, Millipore 475914, 20μM), cholesterol-oxidase (oxidize cholesterol, Millipore 228230, 4unit/ml), amiloride (macropinocytosis inhibitor, Millipore 129876, 150μM), Cytochalasin D (macropinocytosis inhibitor, Millipore 250225, 500nM), LY294002 (broad PI(3)K inhibitor, Cell Signaling 9901s, 50μM), and wortmannin (broad PI(3)K inhibitor, Cell Signaling 9951s, 2μM). Cells were treated with the inhibitors for 1 h at 27°C before Vip3Aa-RFP was added.
For the expression of Vip3Aa protoxin, the vip3Aa gene was cloned in pET-28a(+) vector (Novagen) using oligonucleotide primer Vip-F and Vip-R (plasmid pET-Vip), resulting in a His6 fusion. For the expression of Vip3Aa-RFP, the vip3Aa gene and rfp gene were amplified using oligonucleotide primer pairs Vip-RFP-Up-F and Vip-RFP-Up-R, and Vip-RFP-Do-F and Vip-RFP-Do-R, respectively. Then the two gene fragments were ligated into the pET-28a (+) vector (Novagen) using a pEASY-Uni Seamless Cloning and Assembly Kit (TransGen) after digesting the vector with NcoI and XhoI (plasmid pET-Vip-RFP), resulting in a His6 fusion. The plasmids used to express Vip3Aa-Flag (pET-Vip-flag) was constructed similarly to plasmid pET-Vip-RFP, using oligonucleotide primer Vip-flag-F and Vip-flag-R. The plasmids used to express RFP (pET-RFP), Sf-CCP (pET-Sf-CCP) and Dm-CCP (pET- Dm-CCP) were constructed similarly to plasmid pET-Vip, using oligonucleotide primer pairs RFP-F and RFP-R, Sf-CCP-F and Sf-CCP-R, and Dm-CCP-F and Dm-CCP-R, respectively.
For the expressing of Sf-SR-C-N fused with glutathione-S-transferase (GST), the Sf-SR-C-N gene was amplified using primer SR-C-N-F and SR-C-N-R. The amplification product was inserted into the pGEX-6P-1 (GE Healthcare) vector using a pEASY-Uni Seamless Cloning and Assembly Kit (TransGen) after digesting the vector with BamHI and XhoI (plasmid pGEX-SR-C-N). The plasmids used to express SR-F-1 (pGEX-SR-F-1), SR-F-2 (pGEX-SR-F-2), and SR-F-3 (pGEX-SR-F-3) were constructed similarly to plasmid pGEX-SR-C-N, using oligonucleotide primer pairs SR-F-1-F and SR-F-1-R, SR-F-2-F and SR-F-2-R, and SR-F-3-F and SR-F-3-R, respectively.
Plasmids were transformed into E. coli BL21 (DE3) (Invitrogen) for expression and purification [55]. His-tagged proteins was purified by using cOmplete His-Tag Purification Resin (Roche), whereas GST- tagged proteins was purified by using GST-Sepharose affinity column (GE Healthcare). The purified protein was dialyzed in buffer containing 25 mM Tris-Hcl (pH 8.0), 150 mM NaCl and 5% glycerol at 4 °C with three buffer changes. The purified Vip3Aa is used for cytotoxicity assays.
All the primers and plasmids used in this study are shown in S1 and S2 Tables.
MST was used to determine the binding affinity between Vip3Aa protoxin and Sf-SR-C protein fragments. Briefly, purified proteins were dialyzed into 25 mM Hepes (pH 7.5), 150 mM NaCl, and 0.05 (v/v) % Tween-20. The purified Vip3Aa was labeled with the Monolith NT Protein Labeling Kit (Cat # L008) according to the supplied labeling protocol. 10 nM labeled Vip3Aa were incubated with 0.3 nM to 10 μM Sf-SR-C protein fragments for 20 min at RT respectively. Samples were then loaded into standard treated capillaries and analyzed with a NanoTemper Monolith NT.115 Pico (NanoTemper Technologies GmbH, Munich, Germany) at 25°C. Furthermore, the laser power was set to 10% and the LED power was set to 60%. Normalization of the fluorescence signal and fitting to the Hill equation were performed using the software MO Affinity Analysis v2.2.2 (NanoTemper). For each sample, the whole procedure was performed three times to yield independent triplicates.
Total RNA was isolated from cultured cells or S. exigua midgut using RNAiso Plus (Takara). cDNA was synthesized using a Transcriptor High Fidelity cDNA Synthesis Kit (Roche). Quantification of the cDNA was carried out using SYBR Premix Ex Taq II (Takara) and analyzed by using StepOne software (Applied Biosystems) as previously described [55]. The actin gene acted as the endogenous control. The relative amount of cDNA was calculated according to the 2−ΔΔCT method [56]. Data were analyzed from three independent experiments and are shown as means ± SD.
Plasmids used for Sf-SR-C gene silencing experiments were constructed as described by Katsuma et al. [25]. Fragments of the Sf-SR-C gene (nucleotides [nt] 294 to 803, dsRNA1s) and 400 bp from the complementary strand of the Sf-SR-C gene (nt 693 to 294, dsRNA1as) were amplified by using the primer sets SRi1-Up-F and SRi1-Up-R (for dsRNA1s) or SRi1-Do-F and SRi1-Do-R (for dsRNA1as). dsRNA1s was designed to be 110 bp longer than the dsRNA1as. dsRNA1s and dsRNA1as were inserted in tandem into the pIZT/V5-His vector using a pEASY-Uni Seamless Cloning and Assembly Kit (TransGen) after digesting the vector with KpnI and AgeI (pIZT-SRi1). In the same way, we constructed pIZT-SRi2 using the primer sets SRi2-Up-F and SRi2-Up-R for dsRNA2s (nt 1081–1590) or SRi2-Do-F and SRi2-Do-R for dsRNA2as (nt 1480–1081). We generated stable Sf-SR-C gene silencing Sf9 cells lines by transfection with pIZT-SRi1 or pIZT-SRi2 using the Cellfectin II reagent (Invitrogen) and PLUS Reagent (Invitrogen). At 2 d after transfection, zeocin (500 μg/mL) was added into the medium. Two to three weeks after drug selection, we examined the expression level of the Sf-SR-C gene by qRT-PCR analysis by using the primers SR-RT-F and SR-RT-R. The vectors pIZT-SR-C, pAc-SR-C, and pAc-Sf-S2, which were used to express the Sf-SR-C or Sf-S2, were transfected into Sf9 cells or S2 cells to express the targeted proteins using Cellfectin II reagent and PLUS Reagent.
The plasmid pET-Se-SRi and pET-Hypi were constructed as the pIZT-SRi1 by using the primer sets pET-SRi-Up-F and pET-SRi-Up-R for dsRNA3s (nt 1–870), pET-SRi-Do-F and pET-SRi-Do-R for dsRNA3as (nt 718–1), Hypi-Up-F and Hypi-Up-R for dsRNA4s (620bp), and Hypi-Do-F and Hypi-Dp-R for dsRNA4as (500bp). Then the dsRNA3s and dsRNA3as or dsRNA4s and dsRNA4as were inserted in into the pET28a vector. The pET-Se-SRi and pET-Hypi were transformed into the HT115 (DE3) strain, which lacks RNase III activity for dsRNA expression, as described by Tian et al. [26].
The purified Vip3Aa protoxin was labeled with biotin using an EZ-Link Sulfo-NHS-SS-Biotinylation Kit. (Thermo Scientific). The membrane proteins of Sf9 cells were extracted using a ProteoExtract Transmembrane Protein Extraction Kit (Novagen). Streptavidin Mag Sepharose beads (50 μL) (GE Healthcare) were washed and incubated with 0.2 mg biotin labeled Vip3Aa (Bio-Vip3Aa) for 1 h at 4 °C and washed three times with TBS to remove unbound proteins. The Vip3Aa tagged beads were then incubated with 1 mL of extracted Sf9 cell membrane proteins for 3 h at 4 °C and washed five times with washing buffer (TBS + 2 M urea). The precipitants were boiled with SDS loading buffer and subjected to SDS-PAGE. After cutting out the band representing Vip3Aa, the remaining bands were sent for LC-MS/MS (tandem mass spectroscopy) analysis.
The targeted sample was resolved by SDS-PAGE and transferred onto a Polyvinylidene fluoride (PVDF) membrane (Millipore). Primary antibody and HRP-coupled secondary antibody were used to detect the sample. The membrane was visualized using Immobilon Western chemiluminescent HRP Substrate (Millipore).
Cells were collected and lysed in 0.5 ml radio immunoprecipitation assay buffer (Cell Signaling 9806S) for 30 min on a rotor at 4 °C. After centrifugation at 12 000× g for 15 min, the lysate (30μL) was co-incubated with Vip3Aa-Flag (10 μg) for 2 h at 4 °C. The sample was immunoprecipitated with 5 μL anti-V5 antibody overnight at 4 °C, and 40 μL of protein G agarose beads (Santa Cruz) were washed and then added for additional 4 h. Thereafter, the precipitants were washed five times with washing buffer (3.2 mM Na2HPO4, 0.5 mM KH2PO4, 1.3 mM KCl, 135 mM NaCl, pH 7.4), and the immune complexes were boiled with loading buffer for 6 min and then analyzed by western blotting.
Five microliter of different regions of the Sf-SR-C protein (0.1 nmol) were dotted onto a PVDF membrane. After blocking with 5% skimmed milk in phosphate buffer solution with 0.05% tween-20(PBST), the membrane was incubated in Vip3Aa-flag (100 nM) for 1 h at room temperature, and washed at least three times using PBST. Vip3Aa without Flag-tag (500-fold excess) was used in the competition assays. The following steps are consistent with western blotting.
Ten microliter of Sf9-pIZT-SR-C cells lysate were subjected to SDS-PAGE and then transferred to PVDF membrane. After blocking with 5% skimmed milk in PBST, the membrane were incubated in Vip3Aa-flag (100 nM) for 2 h at room temperature, and washed at least three times using PBST. Vip3Aa without Flag-tag (200-fold excess) was used in the competition assays. The following steps are consistent with western blotting.
Different parts of the Sf-SR-C protein fused with (GST) (0.4 nmol) were incubated with GST-Sepharose affinity beads at 4 °C for 3 h and then washed three times with PBS to remove unbound proteins. Beads were incubated with Vip3Aa-flag (100 nM) and washed five times with PBS. The precipitated components were boiled with sample buffer for 10 min and analyzed by western blotting.
Cell viability assays were performed using the CCK-8 Counting Kit (Dojindo). Briefly, cells with a density of 5 × 104 cells per ml were seeded into 96-well culture plates separately. After overnight incubation, the cells were treated with Vip3Aa protoxin (50 μg/mL) for 48 h. WST-8 reagent was then added to each well. After incubating at 27 °C for 2 h, the absorbance was measured in microplate reader (PerkinElmer) at 450 nm. Treatment with sterile dialysis buffer was used as a control. All tests were performed in triplicate and were repeated at least three times. Cell viability (%) = average absorbance of treated group / average absorbance of control group × 100%.
Cells were grown to 60–80% confluence in Laser confocal culture dishes. After treatment, cells were washed three times with PBS to remove unbound ligands, and fixed with freshly prepared 4% paraformaldehyde at 37 °C for 30 min. For co-localization experiments, cells were then permeabilized (0.2% Triton X-100) and immunostained (primary and secondary antibodies were diluted in 5% skimmed milk powder). Cellular cortical actin and nuclei were labeled for 30 min with fluorescein isothiocyanate (FITC)-phalloidin (Sigma) and DAPI (Sigma) respectively. Cell images were captured using a Zeiss.LSM710 confocal microscope.
S. exigua:
Drosophila:
Experiments were performed at least three times independently. All statistical data were calculated with SPSS software. (v.22.0). For comparisons of the means of two groups, two-tailed t test was used. For comparisons of multiple groups with a control group, one-way ANOVA method was used. Significance of mean comparison is annotated as follow: ns, not significant; *P<0.05; **P<0.01; ***P<0.001.
|
10.1371/journal.pgen.1005444 | Trans-Reactivation: A New Epigenetic Phenomenon Underlying Transcriptional Reactivation of Silenced Genes | In order to study the role played by cellular RNA pools produced by homologous genomic loci in defining the transcriptional state of a silenced gene, we tested the effect of non-functional alleles of the white gene in the presence of a functional copy of white, silenced by heterochromatin. We found that non-functional alleles of white, unable to produce a coding transcript, could reactivate in trans the expression of a wild type copy of the same gene silenced by heterochromatin. This new epigenetic phenomenon of transcriptional trans-reactivation is heritable, relies on the presence of homologous RNA’s and is affected by mutations in genes involved in post-transcriptional gene silencing. Our data suggest a general new unexpected level of gene expression control mediated by homologous RNA molecules in the context of heterochromatic genes.
| We discovered a new epigenetic phenomenon we called trans-reactivation. We found that genes, unable to produce a functional coding transcript, but with the potential of transcribing other RNA’s within their gene body, strongly reactivate the transcription of a wildtype copy of the same gene silenced by heterochomatin. This new epigenetic phenomenon is heritable, relies on the presence of diffusible RNAs able to carry and transfer epigenetic information and is affected by mutations in genes involved in Post-Transcriptional Gene Silencing. Our data strongly suggest that homologous non-coding RNA can reactivate the expression of genes silenced by heterochromatin, thus defining a new unpredicted level of gene expression control in the context of heterochromatic genes.
| In recent years it has become increasingly evident that the expression of eukaryotic genomes is far more complex than it had been previously explored. An emerging body of evidence, coming from next generation sequencing approaches, is showing that the genomes of all studied eukaryotes are almost entirely transcribed, generating an enormous number of non-coding RNAs (ncRNAs) [1]. The ENCODE project showed that at least 90% of analyzed eukaryote genome is transcribed in different cell types, indicating that there is a huge reservoir of RNA molecules with potentially unexplored biological function [2,3,4]. These RNAs are remarkably different in their number, size, subcellular localization, and mechanisms of action, and many are essential to finely control gene expression as well as genomic plasticity [5].
It is becoming increasingly clear that some ncRNAs are part of epigenetic regulatory networks with striking evolutionary conservation [6,7]. For example, dynamic changes in chromatin function are frequently transacted by nuclear RNA signaling pathways. Although the evolutionarily conservation and precise molecular mechanisms are poorly understood, a differential recruitment of a hierarchy of chromatin modifying complexes to specific loci by RNAs sets precise transcriptional states leading to differentiation [8,9,10,11,12]. Moreover, the unusual epigenetic phenomena of paramutation [13,14,15], trans-induction [16] and transvection [17,18] observed in a variety of higher eukaryotes involve the activity of ncRNAs that ‘rewrite’ the transcriptional state of an allele, in processes that apparently escape classic Mendel’s laws of genetic inheritance. These epigenetic phenomena are clear examples of ncRNA-directed regulatory processes that transfer epigenetic information both across cells, between tissues and across generations, though the mechanisms underlying these phenomena still remain elusive.
In order to unveil the role played by cellular RNA pools produced by homologous genomic loci in changing the transcriptional state of a silenced gene, we used classic Position Effect Variegation (PEV) assays in the model system D. melanogaster [19,20] and tested the effect of non-functional alleles of the white gene in the presence of a functional copy of white, silenced by heterochromatin (wm4h). Surprisingly, we found that several non-functional white alleles, unable to produce the main wildtype white coding transcript but with the potential of transcribing other RNA’s from the white locus, trans-reactivate the variegating wm4h line thus increasing eye pigmentation. Strikingly, the presence of non-functional white alleles cause an increase in the wm4h gene transcript as well as an opening in the chromatin structure at the wm4h locus. Remarkably, this new epigenetic phenomenon is heritable, relies on the presence of diffusible homologous RNA’s, and is affected by mutations in genes involved in post-transcriptional gene silencing. Overall, our data strongly indicate that trans-reactivation is a new epigenetic phenomenon that positively control gene expression in the context of heterochromatin through homologous RNA molecules.
The white (w) gene encodes an ABC transporter essential for the red pigment transportation in the compound Drosophila eye. The In(1)wm4h X chromosome inversion places the fully functional wildtype euchromatic w gene adjacent to a region of pericentromeric heterochromatin, creating the variegated white-mottled 4 (wm4h) allele that is characterized by the random inactivation of w by the heterochromatin spreading from the inversion chromosome breakpoint (Fig 1A, upper panel). This cell autonomous transcriptional inactivation once established is clonally inherited, and it is responsible for an eye with a variegated expression of the red pigment, constituting an example of the genetic phenomenon known as Position Effect Variegation (PEV) [20].
Genetic screens have shown that a large number of mutations alter PEV phenotypes, resulting in the isolation of enhancers E(var)s or suppressors Su(var)s of variegating phenotypes. The vast majority of these modifiers were originally isolated in Drosophila as dominant mutations that suppressed or enhanced eye color variegation caused by the wm4h allele [21]. The molecular characterization of those mutants have shown that the products of E(var)’s and Su(var)’s are structural components of chromatin, or enzymes that covalently modify chromatin proteins [19]. However, to date the effect exerted by non-functional w alleles on the wm4h chromatin-dependent variegation is unknown. Therefore, we took advantage of the classic wm4h PEV assay to analyze if non-functional alleles of the w gene were able to modify eye color variegation when in trans-heterozygosis with the wm4h allele (Figs 1A, lower panel, and S1A and S1C).
We screened 19 loss-of-function or hypomorphic w alleles (w*) for genetic interaction with wm4h, and measured eye red pigment to quantify the strength of the genetic interaction [22]. The pilot screen resulted in the isolation of 14 suppressors (74% of total w* alleles screened) able to increase wm4h eye color pigmentation (Fig 1A and S1 Table). Among the tested hypomorphic alleles that behaved as robust suppressors of wm4h variegation, the wsey allele interacted more strongly (Figs 1B and S2). Moreover, loss-of-function alleles of w, including the widely used w1118 allele, showed a weak but highly reproducible suppression effect (Figs 1C and S3). Notably, F1 wm4h/Y males derived from the crosses did not show any increase in eye color pigmentation (Fig 1B and 1C), indicating that the genetic locus responsible for the increased eye color pigmentation in the wm4h/w* trans-heterozygous females is carried by the w* bearing X chromosome. Reciprocal crosses using multiple wm4h balanced lines (S1B, S1D, S1E and S1F Fig) and classic recombination mapping (S1G and S1H Fig) confirmed the effect of the w* loci tested in increasing eye color pigmentation of wm4h. Interestingly, deletions spanning the w locus (Δw) did not show any modification of the wm4h variegation (Fig 1D and S1 Table). Moreover, none of the w* alleles tested was able to reduce the red pigmentation of wm4h, while 5 of them (26%) had no effect (Fig 1A, lower panel, and S1 Table). The pilot screen we conducted suggested that the presence of white genomic homologous sequences of non-functional w* alleles could reactivate in trans-heterozygosis the wm4h heterochromatic silenced locus.
To further characterize the effect of wsey and w1118 alleles (from now on used as representative examples of w* alleles giving a strong and weak interaction with wm4h, respectively), we first decided to measure the strength of their suppression on the wm4h variegation. Interestingly, in the presence of a strong E(var) mutation encoding for a factor contributing to the opening of chromatin [23], both w1118 and wsey retain their ability to increase eye pigmentation of wm4h, though w1118 with a weaker strength (Figs 2A and S4A). This data indicate that the suppressing w* alleles are able to increase eye pigmentation even when the wm4h allele is in a strongly silenced heterochromatin context.
Next we wanted to test the ability of w* alleles to behave as suppressors in other PEV assays, a feature that is shared by classic Su(var)s encoding for factors responsible for the establishment and maintenance of inactive chromatin [24]. We choose the Sbv PEV assay, characterized by the T(2;3)Sbv translocation that juxtaposes the dominant Sb mutation to centric heterochromatin, resulting in mosaic flies with both short (Sb) and normal bristles (Fig 2B). As expected, when the Sbv stock was crossed with a strong Su(var) [25] a significant suppression was observed in the frequency of Sb bristles (Figs 2B, upper graph, and S4B). However, both w1118 and wsey as well as the w locus deletion (Δw) were not able to increase the number of short bristles when crossed with Sbv. These results suggest that the reactivating w* alleles do not likely encode for a factor that, beyond its role in eye pigmentation, it is also responsible for chromatin structure organization. Thus, our data strongly indicate that the mechanism underlying wm4h suppression by w* alleles is very likely unrelated to classic Su(var)’s function.
The suppression effect of w* alleles we observed over wm4h is highly reminiscent of transvection, a well-studied epigenetic phenomenon that results from the interaction between an allele on one chromosome and the corresponding allele on the homologous chromosome. Transvection may lead to both gene activation or repression, and is strongly dependent upon chromosome pairing [18]. However, because of the large In(1)wm4h inversion, the wm4h bearing chromosome is never paired with the w* containing sister chromatid in the suppressed trans-heterozygous females (Fig 1A), thus suggesting that the effect we observed could not be assimilated to classic transvection. However, to exclude intra-chromosomal long distance pairing or X chromosome specific effects, we tested the ability of w* alleles to de-repress variegated wildtype w insertions in pericentric and telomeric autosomal chromosome heterochromatin [26,27,28]. In all lines tested the strong wsey and in some cases also the weak w1118 allele, but not the Δw deletion, consistently increased eye pigmentation of variegated wildtype w pericentric or telomeric autosomal insertions (Figs 2C–2F and S4C). Our data strongly suggest that the increase in eye pigmentation exerted by w* alleles over wm4h is independent from the type of heterochromatin tested, it does not rely on intra-chromosomal pairing and it is chromosome independent.
To explore the correlation of increased eye pigmentation in trans-heterozygous interacting females (wm4h/w*) and transcriptional activation of the w gene, we conducted semi-quantitative RT-PCR on total RNA extracted from adult heads. To measure the amount of full-length wildtype white coding mRNA we designed primers against the 3’ of the transcript that are able to amplify the fully transcribed spliced and unspliced mRNA products (Fig 3A). As expected, the RNA extracted from wildtype flies shows a robust amplification of w, while flies carrying the w1118 loss-of-function allele or the Δw deletion do not show detectable levels of white transcripts (Fig 3A, compare lanes 1 with 2 and 7). Moreover, while the wm4h stock produced very low levels of wildtype coding white transcripts (Fig 3A, lane 3), the hypomorphic wsey flies produced only modest levels of unspliced white transcripts (Fig 3A, lane 4). Remarkably, when w1118 or wsey are introduced in trans with wm4h, we observed a highly reproducible and robust amplification of white transcripts (Fig 3, lanes 5 and 6). These data strongly correlate the levels of eye pigmentation, we observed in the trans-heterozygous w*/wm4h adults, with the levels of expressed coding white transcripts.
However, an increased white transcript stability in w*/wm4h trans-heterozygous could in theory also explain the observed high levels of amplification of coding white transcripts. Therefore, we tested the ability of mutants in genes known to be involved in maternal and zygotic fly mRNA stabilization for their ability to modify the levels of eye pigmentation normally scored in w*/wm4h adults. However, none of the mutants tested, including pumilio (pum) and smaug (smg), were able to significantly change eye pigmentation in our PEV assay (S5A Fig). These data strongly suggest that the increased levels of eye pigmentation observed in w*/wm4h trans-heterozygous females, cannot be directly correlated with general changes in white mRNA stability (S5A Fig).
While, for all the interacting loss-of-function w* alleles, we could be sure that any white coding transcript we detect could only be produced from the wildtype wm4h locus, the same conclusion cannot be drawn for the interacting hypomorphic w* alleles. Therefore, we decided to use genomic probes from the white genomic locus to conduct fluorescent in situ hybridization (FISH) analyses on polytene chromosomes, for monitoring the level of accessibility of the interacting w* and wm4h chromatin loci. Polytene chromosomes represent a special structural organization of Drosophila salivary glands, consisting of polyploid interphase nuclei, which originate by repeated rounds of DNA replication without cell division. Drosophila polytene chromosomes have proven to be an invaluable cytogenetic tool to examine chromosome structure. Moreover, DNA FISH probes hybridization is dependent on the level of replication of specific genomic regions on polytene chromosomes, which in turn depends on the overall level of transcriptional activity present on that locus [24].
Exploiting this special feature of polytene chromosomes, we designed genomic probes covering both the w locus and the coding sequences of the hsp70 gene as internal positive control for FISH probe chromatin accessibility. As expected, wm4h homozygous female chromosomes gave a FISH signal for the hsp70 gene but failed to show a detectable band from the wm4h pericentric chromatin region because this heterochromatic locus is under-replicated [24] (See arrowhead and asterisk in Fig 3B). On the other hand, homozygous chromosomes for the w1118 and wsey alleles or wildtype flies gave detectable FISH signals, for both the w and hsp70 loci (see arrowheads in Figs 3C and 3D and S5B). Remarkably, wm4h/w1118 and wm4h/wsey but not wm4h/∆w trans-heterozygous female chromosomes, on top of the expected signals coming from the w and hsp70 loci, gave a highly reproducible FISH band originating from the pericentric wm4h chromatin locus (see arrowhead in the vicinity of the asterisk in Figs 3E and 3F and S5B). This analysis strongly suggests that FISH signals detected from the wm4h locus in trans-heterozygous w*/wm4h chromosomes are the result of increased DNA replication, consistent with an increase in transcription at the wm4h locus, as also found for other classic Su(var)s [24]. In conclusion, our data strongly indicate that the increase in eye pigmentation, observed in the wm4h/w* trans-heterozygous combinations, is strongly correlated with a reopening of wildtype heterochromatic wm4h locus by the presence of homologous genomic sequences present in the interacting loss of function and hypomorphic w* alleles. We called the ability of w* alleles to re-reactivate wild type white at the wm4h locus: trans-reactivation.
The epigenetic interaction between the w* and wm4h alleles, leading to an increase in white transcription as a consequence of chromatin reopening at the silenced wm4h locus, is a phenomenon highly reminiscent of the epigenetic switch occurring in paramutation. In this epigenetic phenomenon, identified in plants, mice, and recently also in flies, the paramutating allele has the ability to change the activity state of its partner allele on the homologous chromosome (usually to a silent state) [14,15,29]. This effect is dependent on homologous sequences present in the two interacting alleles, is heritable and once established, is independent from the presence of the paramutating allele. To investigate if the reactivation of the wm4h locus was heritable through meiosis, we crossed F1 trans-reactivated wm4h/w* females with w*/Y males to look for trans-generational inheritance of trans-reactivation in the F2 progeny (Figs 4A and S6A).
In the F2 progeny we expect to score wm4h/w* females (that should continue to show increased levels of eye pigmentation), w*/Y males (deficient of red eye pigment), and finally wm4h/Y males where the F1 trans-reactivated wm4h allele, inherited from the F1 mothers has segregated from the w* trans-reactivating allele. If F2 wm4h/Y males have normal variegated eyes, the F1 trans-reactivated wm4h allele had lost its trans-reactivating potential during meiosis. On the other hand, if F2 wm4h/Y males showed increased eye pigmentation, this would be a strong indication that the trans-reactivated state of the F1 wm4h allele is transmitted in the germline and inherited in the F2 generation (Fig 4A).
Remarkably, both w1118 and wsey trans-reactivating alleles when segregated away from the F1 activated wm4h allele were able to generate F2 wm4h/Y males with strong eye pigmentation (Fig 4B and 4C). However, when the same F1 trans-heterozygous wm4h/w* females were crossed with Δw /Y males not bearing any homologous genomic w sequence (Figs 4D and S6B), the resulting F2 wm4h/Y males and wm4h/Δw females failed to trans-reactivate showing levels of eye variegation indistinguishable from parental (P) wm4h stocks (Fig 4E and 4F). Remarkably, when wild type w+ carrying flies are tested for their ability to trans-reactivate in F2 wm4h males, we did not observe any trans-reactivation (S6C Fig). However, when we crossed homozygous trans-reactivated wm4h/wm4h female with trans-reactivated wm4h/Y male, where the trans-reactivating w* allele was no longer present in both lines, we could reproducible score only trans-reactivated progeny up to the F5 (Fig 4G and 4H).
Although, trans-reactivation also works in reciprocal crosses where the trans-reactivating w* allele is carried by the mother (S1E and S1F Fig), our data show that wm4h trans-reactivation is dependent upon the presence of non-functional w* homologous genomic sequences present in at least one gamete (the sperm, in this set of experiment). Moreover, since fully functional coding wild type w+ alleles are unable to trans-reactivate, our data also indicate that the w* reactivating alleles likely encode trans-reactivating factors able to maintain the trans-reactivated state of wm4h coming from the other gamete, a feature that does not meet the classic definition of paramutation. Indeed, paramutated alleles maintain their state independently from the presence of the paramutating allele. However, crosses in which both parents have exclusively wm4h trans-reactivated alleles strongly maintain the trans-generational inheritance of trans-reactivated wm4h, making trans-reactivation a new epigenetic phenomenon, distinct from paramutation, causing the trans-generational inheritance of trans-reactivated wm4h.
A long-standing question in the understanding of the mechanisms underlying epigenetic memory inheritance is whether a genetic information could be transferred from one allele to another through physical pairing of homologous chromosomes, or via a diffusible intermediate carrying genetic information. Although, our data indicate that trans-reactivation of w* over wm4h is independent from classic chromosome pairing (Fig 2C–2F), some of the features of w* trans-reactivation are highly reminiscent of the process known as trans-inactivation observed for the brown (bw) locus in flies, and of trans-induction occurring at the globin gene cluster in mammals, two epigenetic phenomena opposite in their final transcriptional outcome, but highly dependent on long-range chromosome physical interaction [16,30].
Indeed, in trans-inactivation when two bw alleles are paired, heterochromatin insertion in one causes both alleles to associate with the heterochromatic centromere, suppressing brown expression [30]. On the other hand, in trans-induction intergenic transcription of the globin cluster specific for the erythroid cells can be induced in non-erythroid cells by the expression of transiently transfected globin genes [16]. Physical interaction of homologous chromatin region of the affected alleles appears to inactivate, in the case of trans-inactivation, or activate, in the case of trans-induction, chromosomal transcription. Indeed, the activating effect of w* alleles over wm4h could be explained by the two homologous inverted loci engaging on a long-range physical interaction. Under this model, the silent wm4h locus could dilute out heterochromatic factors over the non-functional w* homologous genomic sequences, causing a partial de-repression of wm4h, thus explaining the opening of the chromatin locus, as well as the increased white transcription and eye pigmentation.
In order to determine if w* trans-reactivation is mediated by a long range physical interaction between the homologous w* and wm4h loci or instead through diffusible factors produced by the homologous genomic sequences present in the w* alleles, we made use of the unique reproductive features of flies able to generate adults through the process of gynogenesis [31,32]. Gynogenesis is like parthenogenesis in that diploid zygotes inheriting all chromosomes from their mother can develop without a genetic contribution from fathers. However, even though gynogenetic diploid eggs do not require paternal chromosomes, they do require the physical penetration of the sperm into the mature egg and the contribution of paternal diffusible nuclear and cytoplasmic factors in order to initiate zygotic development (Fig 5A). In particular, when crossing the paternal effect lethal ms(3)K81 males with gynogenetic gyn2, gyn3 females, gynogenesis occurs and the only possible progeny resulting from this cross are gyn2, gyn3 females (Fig 5A).
Therefore, we asked what happened to eye color pigmentation in the progeny resulting from wm4h gynogenetic gyn2; gyn3 eggs fertilized by ms(3)K81 males carrying the strong trans-reactivating wsey allele (Fig 5B). If a long range physical interaction between the male-carrying wsey and female-carrying wm4h chromosomes is required for trans-reactivation we should observe no increase in eye pigmentation in the wm4h; gyn2; gyn3 F1 female progeny, because the father chromosomes do not contribute to the gynogenetic zygote (Fig 5B). However, if diffusible factors carrying epigenetic information are present in the ms(3)K81 males encoding the trans-reactivating wsey allele, then diploid wm4h; gyn2; gyn3 F1 female progeny should show trans-reactivated eyes (Fig 5B). Remarkably, F1 wm4h; gyn2; gyn3 females, not carrying any genomic contribution from the father, show strong trans-reactivated eyes when fertilized by ms(3)K81 males carrying the trans-reactivating wsey allele (Figs 5C and S7A). However, no effect on eye pigmentation was observed in the F1 progeny of gynogenetic females crossed with ms(3)K81 males carrying the Δw deletion (Figs 5D and S7B).
The data resulting from the gynogenesis experiment strongly rule out the possibility that trans-reactivation is due to any type of long-range physical genomic interactions between the wm4h and w* alleles. Instead, our data strongly indicates that diffusible factors, present in the gamete carrying the homologous trans-reactivating w* allele, are responsible for the phenomenon of trans-reactivation of wm4h.
The diffusible factors able to mediate trans-reactivation should carry a genetic information coming from the w* allele, epigenetically transfer this information to the wm4h locus, and amplify this information in the zygote in order to exert their effect later in the fully differentiated adult eye tissue. The high diffusibility, self-amplification property, and capacity to epigenetically transfer information make RNA a great candidate for such a trans-reactivating diffusible factor. We therefore looked at the effect of known mutations in genes involved in the generation, modification and processing of a variety of family of small ncRNA in the process of trans-reactivation (S7C and S7D Fig).
With the single exception of spd-E allele, mutations in genes encoding for factors involved in piRNA (armi, piwi, aub, hsp83, and ago2) and siRNA (dcr2 and r2d2) biogenesis enhanced the trans-reactivating effect of w* over wm4h (Fig 6A). On the other hand, mutations in genes encoding factors involved in the production of miRNA (ago1 and dcr1) suppress trans-reactivation, with the exception of ago2 that is also involved in the piRNA pathway (Fig 6A). Recently, a non-coding RNA expressed from a human pseudogene was reported to regulate the corresponding protein-coding mRNA by acting as a decoy for microRNAs [33]. This study raised the questions about the potential ability of non-coding transcripts to act as ‘sponges’ to attract miRNAs, thus boosting the expression of miRNA target genes. If the trans-reactivating w* alleles produced miRNAs sponge RNA’s we should expect that mutations in genes affecting miRNA biogenesis should enhance the effect of the trans-reactivating w* alleles over wm4h. However, our data clearly show that mutations in components of the miRNA processing machinery suppress trans-reactivation (Fig 6A and 6B), making it very unlikely that trans-reactivation could be explained by a miRNAs sponge effect. Remarkably, mutation in dnmt2, an RNA-dependent Methyl Transferase (RdMT) very recently shown to be involved in the phenomenon of paramutation in mice [34,35], has also strong suppressing effect on trans-reactivation (Fig 6A). Overall, our genetic interaction data strongly indicate that altering the biogenesis, the post-translational modification and processing of a variety of small ncRNA families interfere at various levels the efficiency of trans-reactivation (Fig 6B), strongly supporting the involvement of RNA molecules in the onset and maintenance of this new epigenetic phenomenon.
In order to identify unique specific RNA species produced by the w* alleles during the process of trans-reactivation of wm4h we conducted high throughput RNA sequencing analysis of total as well as small RNA enriched fractions from larval Malpighian tubules, an abundant fly larval tissue that express high levels of the w gene. However, the analysis of short RNAs did not result in the identification of specific or enriched RNA species in the trans-reactivated wsey/wm4h line when compared to the wsey or wm4h lines alone (lower panel Fig 6C). On the other hand, the analysis of long RNA species clearly showed that the trans-reactivated wsey/wm4h line overall produces more coding RNA from the w locus than the single wsey or wm4h lines alone (upper panel Fig 6C and 6D), in line with our previous semi quantitative RT-PCR analysis (Fig 3A). In conclusion, our RNA sequencing data did not allow us to identify specific or discrete RNA species present in the wsey trans-reactivating allele that could explain the onset of trans-reactivation.
However, if RNA species produced at the homologous w* alleles could influence in trans the transcription of the silenced wm4h locus, we expect that injection of wm4h eggs with total RNA extracted from w* individuals should result in trans-reactivated adult flies. Remarkably, when we injected wm4h eggs with total RNA extracted from wsey testis (Fig 6E) we obtained adults with trans-reactivated eyes. However, wm4h eggs injected with RNA derived from Δw testis had no effect (Fig 6E), strongly indicating that RNA molecules produced by the homologous wsey allele are responsible for the trans-reactivation of the wm4h locus. In conclusion, we propose that the RNA produced by the w* alleles from the w locus can induce in trans the activation of the silenced wm4h locus through a yet to be defined mechanism in which post transcriptional gene silencing and RNA methylation factors are involved.
Over the last decade we have come to appreciate that the eukaryotic chromatin is full of coding and non-coding overlapping transcripts, and that a lot of the genetic information is transacted by coding and non-coding RNAs. Indeed, the vast majority of genomes of all metazoans are transcribed into complex patterns of ncRNAs. Recent observations strongly suggest that ncRNAs contribute to the complex transcriptional networks needed to regulate cell function [36,37]. However, despite a rich variety of mechanisms by which RNA acts by repressing transcription have been identified, fewer studies have explored the role of RNA in the positively regulation of transcription. RNA is an integral component of chromatin [38] and many transcription factors and chromatin modifying enzymes have the capacity to bind RNA or complexes containing RNA or RNA-binding proteins [8]. Moreover, RNA-directed processes help to establish chromatin architecture and epigenetic memory [10]. Indeed, ncRNAs can act locally to regulate the epigenetic state of the nearby chromatin, often recruiting either repressing or activating chromatin remodeling complexes [8]. Given the high degree of sequence and locus specificity information, RNA could recruit chromatin remodelers and modifiers in a sequence-specific manner. Indeed, recent observations are beginning to reveal that RNA molecules can stimulate gene transcription. These RNA activators employ a wide array of mechanisms to up-regulate transcription of target genes, often functioning as DNA-tethered activation domains, as coactivators or modulators of general transcriptional machinery [39].
In search for a possible role played by cellular RNA pools produced by homologous genomic loci in changing the transcriptional state of a silenced gene, we found that non-functional alleles of the classic fly marker white could trans-reactivate the expression of wild type copy of white silenced by heterochromatin. We called this new epigenetic phenomenon trans-reactivation. This process is heritable over many generations, it relies on the presence of diffusible RNAs, and it is affected by mutations in genes involved in post-transcriptional gene silencing. Our data strongly suggest that the presence of a gene that does not produce a full coding transcript but only spurious transcription could trans-reactivate the expression of a functional copy of the same gene silenced by heterochromatin, defining a new unpredicted level of gene expression control in the context of heterochromatic genes.
The mechanisms of action and many features of trans-reactivation appear to be different from the previously characterized epigenetic reprogramming events occurring in paramutation, transinduction and transvection. In trans-reactivation genomic regions, transcribing RNAs unable to produce a full coding transcript, can positively influence the transcriptional state of an homologous locus silenced by heterochromatin. However, how can the information present in the non-functional RNA molecules be ‘read’ and telecasted to specific silenced homologous genomic regions? RNA molecules have the potential to pair with other RNA molecule or with DNA. Possibly through their ability to form paired structures, RNA may function as a sensor and/or a messenger of homology. RNA can form duplexes with sequence homology with genomic promoter sequences thus promoting transcription [40]. Indeed, there is a great deal of evidence that enhancers and other promoter regulatory sequences are transcribed in the cell in which they are active, and that enhancer transcription is often needed to activate the transcription on their target gene [41]. Moreover, given that mutations in genes encoding for factors involved in posttranscriptional gene silencing modulate trans-reactivation, it remains to be determined whether some sequence specific small ncRNAs, that we failed to identify, may explain the mechanism of trans-reactivation. Furthermore, we do not know why some non-functional w* null or hypomorphic alleles failed to trans-reactivate their homologous silenced locus. Indeed, differences in trans-reactivation ability between different w* alleles will need future investigation, and may provide key information to understand the mechanism of action of trans-reactivation. However, independently from the mechanistic details of trans-reactivation, we are convinced that we identified a new epigenetic phenomenon of allelic communication that acts through diffusible RNA molecules. Nevertheless, much work need to be done to study the evolutionary conservation and the general ability of trans-reactivating RNAs in resetting the epigenetic and transcriptional state of homologous genetic loci.
Flies were raised and crossed at 20° on K12 Medium (US Biological). Unless otherwise stated, strains were obtained from Bloomington, Szeged, and DGRC Stock Centers. Recombination mapping of the trans-reactivating effect of w* alleles is described in S1 Fig. The white alleles used for the trans-reactivation assays are listed in S1 Table. The w1; gyn-21;gyn-31 and ms(3)K811/TM3, Sb1, Ser1 lines were a generous gift from Dr. Mia Levine. The wm4h; the CyORoi/Sco—wm4h; PrDr/TM3 Sb,Ser—E(var)3−101- lines were donated by Dr. Gunter Reuter. The 39C12 and 118E-10 variegating autosomal lines were a gift from Dr. Sarah Elgin. The cn1 P{PZ}AGO104845/CyO—ry506, r2d21/CyGFP—aubHN2/CyO—y1 w67c23;P{EPgy2}AGO2EY04479/ Cy—Armi/TM3 ser, GFP—ru1 st1 spn-E1 e1 ca1/TM3, Sb1 Ser1—Dcr1/TM6c—Hsp83Scratch/TM3 Sb,Ser and T(2;3)SbV⁄ TM3,Ser lines were obtained from Dr. Maria Pia Bozzetti and Dr. Laura Fanti. The Ago2null, Dcr2null /CyO, piwi06843/CyO, aubQC42/CyO lines derive from Dr. Valerio Orlando Lab and the P-819 and A4-4 lines from Dr. Stephane Ronsseray lab. Dr Frank Lyko provided the Dnmt2Δ99/CyO line.
The eye pigment assay was performed in triplicate for each of the three biological replicate we conducted, as previously described [22]. Eyes from representative individuals of the genotype tested were photographed with an Olympus SZ61 stereomicroscope. For the dominant Stubble variegation (SbV) assay ten pairs of major dorsal bristles were analysed in 20 flies. The extent of Sb variegation was represented as the mean of Sb and WT bristles per genotype tested in three biological replicate.
Cytological preparations and FISH analysis were carried out as previously described [42] The P(CaSpeRA) plasmid containing the genomic sequences of the mini white as well as the hsp70 genes was used as a template for generating the FISH probe. FISH signals were acquired with a Leica DM 4000B microscope (Leica).
Total RNA was purified from 20 heads of adult flies of the desired genotype, using TRIzol reagent following the manufacturer’s (Invitrogen) recommended protocol. First strand cDNA was synthesized using MuLV reverse transcriptase (Applied Biosystems). Semi-Quantitative PCR was conducted using specific primer pairs that amplify the fourth, fifth and sixth exons of the white gene (forward, 5’- CTATAGGTCATATCTTGTTTTTATTGGCAC-3’) (Reverse 3’- CGTGGGTGCCCAGTGTCC-5’) and the following parameters: denaturation 2 min at 94°C, followed by 20 cycles of 30” at 94°C, 30” at 53°C, 2’ at 72°C and final extension at 72°C for 3’. The PCR products were analyzed by agarose electrophoresis and the images were acquired and quantified with the ChemiDoc XRS imager (BioRad). The Act5C RNA was amplified as loading control with the primer pairs (forward, 5’-CCTCGTTCTTGGGAATGG-3’, Act5C reverse, 3’- CGGTGTTGGCATACAGATCCT-5’) using the same PCR cycle parameters.
Total long RNAs were extracted from malpighian tubuli tissues using the miRNeasy Mini Kit, following an RNeasy MinElute Cleanup Kit (Qiagen) step for the enrichment of small RNAs. The 100bp paired-end sequencing was performed at IGA using a HiSeq2000TM machine. Long RNAs were mapped against the Drosophila genome (assembly BDGP Release 5) using TopHat (PUBMED 19289445) with default parameters. Mapped reads were associated with gene annotations (FlyBase 5.12), and transcript levels of annotated genes, expressed as Fragments per Kilobase of exon Per Million fragments mapped (FPKM), were estimated by CuffCompare (PUBMED 20436464). For small RNAs, sequencing adapters were trimmed from sequence reads with custom scripts. Resulting sequences were mapped against the Drosophila genome using Bowtie (PUBMED 19261174) allowing at most two substitutions within the first 20 base pairs.
wm4h; CyORoi/Sco 3h starved flies were allowed to lay eggs for 50 minutes on apple juice agar plates with yeast. Approximately 100–150 decorionated eggs were injected with 1–2 pl of an injection solution containing 5 mM NaCl, 5% Texas Red conjugated dextran (Molecular Probes), and 1 ng/μl of total RNA extracted and purified from testis of either Δw or wsey Drosophila males.
|
10.1371/journal.pgen.1000785 | The Bifidobacterium dentium Bd1 Genome Sequence Reflects Its Genetic Adaptation to the Human Oral Cavity | Bifidobacteria, one of the relatively dominant components of the human intestinal microbiota, are considered one of the key groups of beneficial intestinal bacteria (probiotic bacteria). However, in addition to health-promoting taxa, the genus Bifidobacterium also includes Bifidobacterium dentium, an opportunistic cariogenic pathogen. The genetic basis for the ability of B. dentium to survive in the oral cavity and contribute to caries development is not understood. The genome of B. dentium Bd1, a strain isolated from dental caries, was sequenced to completion to uncover a single circular 2,636,368 base pair chromosome with 2,143 predicted open reading frames. Annotation of the genome sequence revealed multiple ways in which B. dentium has adapted to the oral environment through specialized nutrient acquisition, defences against antimicrobials, and gene products that increase fitness and competitiveness within the oral niche. B. dentium Bd1 was shown to metabolize a wide variety of carbohydrates, consistent with genome-based predictions, while colonization and persistence factors implicated in tissue adhesion, acid tolerance, and the metabolism of human saliva-derived compounds were also identified. Global transcriptome analysis demonstrated that many of the genes encoding these predicted traits are highly expressed under relevant physiological conditions. This is the first report to identify, through various genomic approaches, specific genetic adaptations of a Bifidobacterium taxon, Bifidobacterium dentium Bd1, to a lifestyle as a cariogenic microorganism in the oral cavity. In silico analysis and comparative genomic hybridization experiments clearly reveal a high level of genome conservation among various B. dentium strains. The data indicate that the genome of this opportunistic cariogen has evolved through a very limited number of horizontal gene acquisition events, highlighting the narrow boundaries that separate commensals from opportunistic pathogens.
| The accessibility of complete bacterial genome sequences has provided important changes to the field of microbiology by significantly enhancing our understanding of the physiology, genetics, and evolutionary development of bacteria. Bifidobacteria are among such microorganisms, being mammalian commensals of biotechnological significance due to their perceived role in maintaining a balanced gastrointestinal (GIT) microflora. Bifidobacteria are therefore often applied as health-promoting or probiotic components in functional food products and represent a growing area of scientific interest. However, within the genus Bifidobacterium not all species provide beneficial effects on the host's health. In fact, the Bifidobacterium dentium species is considered an opportunistic pathogen since it has been associated with the development of dental caries. In this manuscript, we describe the complete genetic make-up of the B. dentium Bd1 genome and discuss functions that explain how this microorganism has adapted to the oral human cavity and imparts a cariogeneous phenotype. Moreover, we performed comparative genomic analyses of B. dentium genome with other bifidobacterial genomes in order to trace genetic differences/similarities between the opportunistic oral pathogen B. dentium Bd1 and closely related intestinal bifidobacteria.
| Bifidobacteria are relatively abundant inhabitants of the gastrointestinal tract (GIT) of humans and animals [1]. Many bifidobacterial species, in conjunction with other members of the intestinal microbiota are believed to contribute to host nutrition, while also impacting on intestinal pH, cell proliferation and differentiation, development and activity of the immune system, and innate and acquired responses to pathogens [2]–[8]. These perceived beneficial health effects have driven commercial exploitation of bifidobacteria as live components of many functional foods and therapeutic adjuncts. However, bifidobacteria have also been isolated from the human oral cavity, where their presence is linked to the progression of tooth decay: bifidobacteria have been detected in high numbers in infected dentine from carious lesions in children [9] and have been associated with childhood dental caries [10]. B. dentium can be found as part of the microbiota implicated in human dental caries [10]–[16]. In recent surveys of oral bifidobacteria associated with coronal caries in adults and children [17] and root caries in adults [18], B. dentium was the most frequently isolated Bifidobacterium species, representing approximately eight percent of the culturable bacteria isolated from active carious lesions. This species is capable of acidogenesis to produce a final pH in glucose-containing media below pH 4.2 [19], sufficient to cause extensive demineralisation of tooth tissues [20]. B. dentium may therefore significantly contribute to the pathogenesis of dental caries which is one of the most common chronic diseases, remaining untreated in many underdeveloped countries where dental pain is often alleviated only by the loss or extraction of the affected tooth [21].
The ecological plaque hypothesis was formulated to explain the composition and phenotypic properties of the microbiota associated with caries initiation and progression [22]. This hypothesis envisages that caries is the result of environmental changes, particularly as a result of reduced intra-oral pH as a consequence of bacterial fermentation of dietary carbohydrates. When this occurs in the oral cavity, it selects for a microbiota which is more aciduric and more acidogenic than that present in the absence of caries. The environmental change results in a significant alteration in the composition of the commensal microbiota, with taxa including bifidobacteria, lactobacilli, Actinomyces and streptococci proliferating [11].
Complete genome sequences of relatively few human intestinal commensal bifidobacteria have been determined, being largely motivated by their perceived health-promoting activity. These include Bifidobacterium longum subsp. longum NCC2705, B. longum subsp. longum DJO10A and B. longum subsp. infantis ATCC15697, Bifidobacterium animalis subsp. lactis DSM10140 and B. animalis subsp. lactis ADO11 [23]–[27]. Here, we describe the sequence analysis of the B. dentium Bd1 genome. This strain was originally isolated from human dental caries [28]. Analysis of the predicted proteome together with comparisons of the genome sequence to those of intestinal bifidobacteria revealed that this bacterium has undergone specific genetic adaptations for colonization and survival in the oral cavity.
The genome of B. dentium Bd1 is one of the largest bifidobacterial genomes reported to date, with a single circular chromosome consisting of 2,636,368 base pairs (Figure 1). The average GC content of 58.54% is similar to that of other sequenced bifidobacterial genomes and is consistent with the range of G+C mol% values for the Actinobacteria [1]. For protein-encoding DNA regions, the G+C contents of codon positions 1, 2 and 3 were determined to be 61%, 43%, 74%, respectively, the latter value somewhat deviating from the expected value (70%), as based on a survey of 696 eubacterial and 56 actinobacterial genomes or bifidobacterial genomes (NCBI source) (Figure S1 and data not shown).
The genome of B. dentium Bd1 possesses 55 tRNAs and four rRNA operons, which are located in proximity of the oriC. While B. dentium contains tRNAs for every amino acid, the corresponding genes for aminoacyl-tRNA synthetases for asparagine and glutamine appear to be absent. An alternative route is a pathway described for Fusobacterium nucleatum [29] that utilizes Gln-and Asn-tRNA amidotransferases, which amidate misacylated Gln-tRNA or Asn-tRNA charged with Glu or Asp to produce Gln-tRNA Gln or Asn-tRNA-Asn, respectively. Homologous genes that specify subunits for the Gln-and Asn-tRNA amidotransferase are present on the Bd1 genome and are also present on other sequenced bifidobacterial genomes (data not shown).
Identification of protein-coding sequences revealed 2143 open reading frames (ORFs) with an average length of 1059 bases and constituting 89% of the genome, the remainder representing intergenic regions with an average length of 143 bp. This latter value is lower than those calculated for other known bifidobacterial genomes, whose combined average intergenic region length is 191 bp, indicating that B. dentium Bd1 has a more compact genome. Such results are not biased due the methods used for B. dentium Bd1 genome annotation (see Text S1), since it employed an ORF identification protocol, with cut-off values that are similar to those used for the annotation of the so far published bifidobacterial genome sequences [23]–[27]. The ORFs are organised in a typical bacterial configuration, so that transcription is frequently in the same direction as DNA replication. A functional assignment was made for 78.5% of the predicted ORFs, while homologs with no known function from other bacterial species were identified for an additional 13% of the B. dentium Bd1 ORFs. The remaining 8.4% appears to be unique to B. dentium. The ATG start codon is preferred (78.9% of the time), while GTG and TTG are less frequently used start codons at 18.9% and 2%, respectively.
The presumed origin of replication (oriC) of the B. dentium Bd1 chromosome, including the adjacent and conserved dnaA, dnaN and recA gene configuration, was identified on the basis of common features to corresponding regions in other bacterial chromosomes [30]–[31]. The oriC was located proximal to the dnaA gene, in an AT-rich sequence containing characteristic DnaA boxes, while the position of the replication terminus (terC) was inferred using GC skew analysis (Figure 1).
The predicted B. dentium Bd1 proteins were functionally categorized and the proportions in each category were compared with those of other bifidobacterial genomes (Figure 2). It is notable that approximately 14% of the genes identified in the B. dentium Bd1 genome encode proteins that are predicted to be involved in carbohydrate metabolism and transport. Such an extensive genetic adaptation to carbohydrate metabolism is shared, to a similar degree, with enteric bifidobacteria, and likely represents a specific genetic adaptation of bacteria residing in the GIT, apparently both in the upper region (the oral cavity) as well as in the distal tract (the colon) of the GIT.
Furthermore, 3D-structure prediction of 1955 of the 2143 deduced proteins that constitute the predicted proteome of B. dentium Bd1 using the Fugue fold recognition method allowed a functional attribution of these predicted structures by means of the SCOP domain annotation into superfamilies (Figure 3). Such an analysis was also performed for the intestinal B. longum subsp. longum NCC2705 strain. Notably, both genomes possess a similar protein superfamily content distribution except for proteins assigned to the Toxin-defence group: the B. dentium Bd1 genome encodes nine times more proteins with this annotated function as compared to the B. longum subsp. longum NCC2705 genome (Figure 3). There are 18 genes encoding predicted sensor histidine protein kinases (HPK), 15 of which are located adjacent to a putative response regulator-encoding gene (one of which is separated by just a single gene), distributed throughout the Bd1 chromosome, which is somewhat more than one would predict based on its genome size and the number present in other, similarly sized bifidobacterial genomes (ranging from 5 to 17) [32]. This suggests that the relative abundance of two-component systems (2CSs) in a micro-organism is an indicator of its ability to sense dynamic environmental cues and to modulate appropriate physiological responses, a notion also exemplified by the high number of 2CSs found in Bacteroides thetaiotaomicron [33].
Comparative genomics of intestinal bifidobacteria may elucidate genomic regions involved in the maintenance of physiological homeostasis that is attributed to these bacteria. The genomic structure of B. dentium Bd1 is highly syntenic with that of the recently sequenced genome of B. dentium ATCC27678 (accession no. ABIX00000000; Figure 4) with an average nucleotide identity of 99% across these two genomes. This Bd1 versus ATCC27678 genome comparison was scrutinized for the identified ORFs with particular consideration of nucleotide changes occurring at particular positions for every codon in the Bd1 genome. The highest substitution rate occurred at the third codon position (36.3% vs. 29% at the second nucleotide and 34.7% at the first nucleotide). Furthermore, a survey of DNA sequence similarity at intergenic regions between both strains revealed a lower level of DNA conservation (98%) compared to that identified between coding regions (99%). Thus, in the B. dentium taxa, as is generally the case for Eubacteria, the intergenic regions have experienced a higher rate of nucleotide substitution compared to that of the coding regions.
Furthermore, sequence identity varied between the ORFs shared by both genomes (2133); with the large majority displaying an identity of 100%, and just 19 ORFs showing a similarity of less than 95%. The two genomes were shown to contain an identical repertoire of prophage-like and IS elements, although a small number of nucleotide differences (e.g., deletions or substitutions) were noticed for two IS elements, isblo10 and isblo3-2, and for the prophage-like elements Bdent-1 and Bdent-2, suggesting that these strains are very closely related but genetically distinct.
Points of disruption of the gene conservation between the two B. dentium genome sequences corresponds to the presence or absence of integrated elements (IS elements) and genes with a predicted function in sugar metabolism (Figure 4). For example, the B. dentium ATCC27678 genome contains a putative rhamnosyltransferase-encoding gene, which is located close to an IS element, and which is lacking in the corresponding position on the B. dentium Bd1 genome. This suggests that this genetic element in B. dentium ATCC27678 was acquired by HGT or it may be lost due to the presence of this mobile element. The conserved gene order was not limited to B. dentium ATCC27678, but can be expanded to B. adolescentis ATCC15703 (Figure S2). The degree of alignment between the genomes of different bifidobacterial species varied depending on the phylogenetic distance of the genomes being compared. Thus, the alignments of B. dentium Bd1 with the genomes of B. longum subsp. longum NCC2705, B. longum subsp. longum DJO10A or B. longum subsp. infantis ATCC15697 display an clearly reduced colinearity, resulting in an X-shaped plot diagram (Figure S2). This is indicative of multiple large rearrangements around the origin-terminus axis of the genome following divergence from a common ancestor [34]. When the genome of a Bifidobacterium strain outside the B. longum and B. adolescentis phylogenetic groups (e.g., B. animalis subsp. lactis ADO011) was thus aligned, sequence identities were restricted to very small genome segments (Figure S2). The fact that long-range genome alignments of Bifidobacterium could not be produced at the DNA level is a significant indication of profound intra-genus diversity, similar to that found for the lactobacilli [35], but contrasting for example with DNA-DNA interspecies alignments among other genera belonging to the Actinobacteria [1].
Genome alignments using PROmer allows the reconstruction of broad phylogenetic relationships between prokaryotic genomes [35]–[37]. A previous phylogenomics analysis based on 123 protein sequences representing the minimal core proteins of the Actinobacteria phylum highlighted the relatedness of B. longum subsp. longum NCC2705 to propionibacteria, Leifsonia and Tropheryma [1]. Here, we included in such an analysis the B. dentium Bd1 and other bifidobacterial genome sequences published to date. Interestingly, the resulting neighbour-joining tree revealed a clear evolutionary split of these bifidobacterial sequences with those derived from Propionibacterium acnes, Leifsonia xyli subsp. xyli and Tropheryma whipplei (Figure S3), and indicating that bifidobacteria are derived from a deep ancestor of the Actinobacteria phylum.
A comparative study was undertaken to determine putative orthology between the B. dentium Bd1 CDSs with those of five other completely sequenced bifidobacterial genomes, resulting in 908 putative orthologs that were shared between all these genomes (Figure S4). The most common functional classes represented by these core proteins were those involved in housekeeping functions including information processing, DNA replication, repair, cell division, transcription, translation and secondary metabolite biosynthesis, transport and catabolism. Proteins belonging to functional categories representing sugar and amino acid metabolism, and uptake by ABC transporters were the second largest commonly found group, emphasizing their apparent importance to bifidobacteria (Figure 3). When the genome sequence of B. dentium ATCC27678 was included in this analysis a total of 692 CDSs were found that have no matches in currently available bifidobacterial genomes, thus representing B. dentium-specific proteins. Over half of these are hypothetical proteins, whereas the remainder have their best matches in sequenced members of the Actinobacteria and/or Firmicutes, including bacteria of the oral microbiota such as Actinomyces spp., S. mutans and Treponema denticola [38]–[39]. Notably present among these B. dentium-specific genes are two adjacent ORFs (BDP_1871- BDP_1872) with homology to the hip operon found in Enterobacteria, which allows increased survival following various stress conditions [40]. Thus, the B. dentium hip operon may positively influence persistence in the oral environment upon exposure to stress conditions, e.g. the fluctuating acid environment that accompanies caries initiation. A subset, i.e. 181, of the 692 B. dentium-specific proteins are conserved in both B. dentium genomes but do not have significant matches in other currently available genome sequences, and may thus be responsible for certain unique adaptive properties. When these hypothetical proteins were scanned against a database of structural profiles using the FUGUE program, which can recognize distant homologues by sequence-structure comparisons [41], we identified a number of potential homologs with a significant Z-score subdivided in various clusters according to their predicted gene function (Figure S5). Interestingly, this analysis revealed that part (13.33%) of these B. dentium-specific proteins clustered in the toxin/defence family, suggesting that these proteins provide protection against host defensins, such as cationic and cysteine-rich peptides [42].
The coexistence of B. dentium within the oral biofilm which consists of over 900 taxa may facilitate exchange of genetic information that is mediated by mobile genetic elements. Such elements are present in virtually all bacterial genomes, and in some organisms the associated genes may contribute to the metabolism or pathogenic potential of the organisms. Analysis of the B. dentium Bd1 genome revealed the presence of conventional mobilome candidates that may have been acquired through Horizontal Gene Transfer (HGT). Analysis of G+C content (G+C), amino acid usage [43], BLASTP best-match and codon preference of the B. dentium Bd1 chromosome indicated that considering its total DNA content just 93,300 bp of the B. dentium Bd1 genome display a significant deviation (>2-fold difference) from the average values of the paramethers as indicated above and may have been recently acquired by HGT (Figure 5). This suggests that in contrast to other bifidobacterial genomes [44], HGT is not the main force driving genome evolution in B. dentium species.
Some representative mobile elements as well as DNA regions acquired by HGT will be discussed below. In silico analysis of the B. dentium Bd1 genome revealed the presence of two prophage-like elements designed Bdent-1 and Bdent-2, which exhibit a close phylogenetic relationship with phages infecting bacteria belonging to the Firmicutes phylum (Ventura et al, AEM in press, published now). The B. dentium Bd1 chromosome harbours seven insertion sequences (IS), belonging to three IS families, ISL3, ISL10, ISSdel and IS3 like (Table 1), a number which is much lower than those in other sequenced bifidobacterial genomes (data not shown).
Additional putative mobile elements identified in the B. dentium Bd1 genome are represented by two Clustered of Regularly Interspersed Short Palindromic Repeats (CRISPR) loci, named CRISPR1 and CRISPR2, with adjacent CRISPR-associated cas genes, CRISPR-Cas1 and CRISPR-Cas2, respectively. When CRISPR-Cas1 and CRISPR-Cas2 were compared to identified lactic acid bacteria CRISPR loci [45], they clustered into two different CRISPR families, Blon1 and Llel1, respectively (Figure S6), suggestive of two independent HGT events. CRISPRs represent the most widely distributed prokaryotic family of repeats [46],[47], and act as defence systems against invasion of foreign genetic material, in particular phages [48],[49].
The genome variability among different strains of B. dentium was investigated by Comparative Genomic Hybridization (CGH) experiments using B. dentium Bd1-based microarrays. We determined which and how many ORFs from the sequenced B. dentium Bd1 strain did or did not hybridize with total genomic DNA extracted from ten B. dentium strains from different origins (dental caries from adult or child, from saliva and from fecal samples). Overall, DNA from the tested B. dentium strains failed to efficiently hybridize to between 1% and 12% of the probes from the reference B. dentium strain Bd1. These values are small compared to those described for other bacterial species [50]–[56], including bifidobacteria, such as B. longum subsp. longum [57]. Such findings suggest that the B. dentium genome is only slowly evolving compared to other bacteria, including bifidobacterial species residing in the distal tract of the human GIT. Nevertheless, CGH cannot identify regions present in the tested strains but absent from the B. dentium Bd1 strain, while it also does not analyze the synteny of the genome. Consequently, caution needs to be employed when applying the term “divergent” to CGH studies. When projected on the genome map of B. dentium Bd1, the CGH results highlight clustering of conserved and variable ORFs (Figure 6). The region between the origin of replication and the terminus of replication in the clockwise direction represents the largest genome segment of relative high gene conservation (denoted as I ). In contrast, the region between the replication terminus and the origin of replication in the clockwise direction was shown to be a major area of genetic diversity (II in Figure 6). According to the Bd1 gene annotation, the types of genomic diversity thus identified can be assigned to two classes (i) mobile DNA that constitutes the B. dentium mobilome previously identified by in silico analyses; (ii) plasticity regions of B. dentium genome, which may underlie specific adaptations of the investigated strains, and which could represent laterally acquired DNA or remnants of ancestral DNA that have not (yet) been lost. Plasticity regions which are preferred sites for acquisition of strain-specific DNA are well recognized in the genomes of pathogens like H. pylori [58], where array-based CGH has similarly been used to highlight regions involved in adaptation to different pathological roles [59]. Five large DNA segments, which are conserved in B. dentium Bd1 and in the closely related strains ATCC27678 and LMG10585, clearly represent mobile DNA: two prophage-like elements, Bdent-1 and Bdent-1, the CRISPR elements and the cytosolic proteins (BDP_1391–1394) (Figure 6). In some strains, stretches of hybridizing prophage genes matched individual modules of the prophage (Figure 6, Bdent-2 prophage-like element), an observation which agrees with the hypothesis of modular phage evolution [60]. Within the variable regions of the CGH map, indicated as plasticity regions, genes associated with bacterium-environment interaction and metabolic abilities appear to be particularly enriched. These include the eps clusters, a putative fimbrial-biosynthesis gene cluster and membrane-associated transporters. The eps clusters of the Bd1 strain are associated with the dTDP-rhamnose biosynthesis locus, and represent the largest genome segment with substantial inter-strain genetic variability.
When the CGH data are expressed on a log2 scale according to the mean ratios of the normalized results, a major peak is noticed at the same position for all the B. dentium strains tested, indicative of very similar DNA sequences (Figure 6A–6L). In addition to the inset of Figure 6, which globally quantifies the similarity of the test strains versus B. dentium Bd1, a clustering of the microarray data was performed in order to extract qualitative information about the presence of each gene. A phylogenetic tree based upon these CGH scores identified B. dentium ATCC27678 and B. dentium LMG 10585 as the closest relatives of B. dentium Bd1 (Figure S7). Moreover, CGH clustering produced four groups of B. dentium strains based on varying levels of genetic diversity, and largely corresponding to their ecological origin (Figure S7). As a complement to the CGH analysis, we performed multilocus sequence analyses for the same strains, using the genes for the Clp ATPase (clpC gene), two F6P-phosphoketolases (xfp gene), DnaJ chaperone (dnaJ1), and DNA-directed RNA polymerase B' subunit (rpoC gene) as phylogenetic markers [61]. As expected, the phylogenetic tree produced from these concatenated sequences confirmed the clustering of GCH based data (Figure S7).
Homologs of all the enzymes necessary for the fermentation of glucose and fructose to lactic acid and acetate through the characteristic “fructose-6-phosphate shunt” [62], as well as a partial Embden-Meyerhoff pathway were annotated in the B. dentium Bd1 genome. These metabolic pathways are important for generation of pyruvate and re-oxidation of NADH, as well as for synthesis of an additional ATP molecule per glucose during the conversion of pyruvate to acetate. The enzymes responsible for pyruvate metabolism identified in the B. dentium genome include xylulose 5-phosphate/fructose-6-phosphate phosphoketolase, pyruvate dehydrogenase, pyruvate formate-lyase, phosphotransacetylase, acetate kinase, lactate dehydrogenase. B. dentium Bd1 possesses an incomplete tricarboxylic acid (TCA) cycle, which lacks oxoglutarate dehydrogenase, fumarase and malate dehydrogenase. The primary role of these TCA enzymes is most likely the production of precursors for amino acid and nucleotide biosynthesis. Since B. dentium Bd1 can be cultured anaerobically with urea, arginine and cysteine as the sole nitrogen sources (unpublished data), it was not surprising that genes required for the biosynthetic pathways of all amino acids were identified in the genome. The way in which cysteine is synthesized is unclear, as the genes involved in sulphate/sulphite assimilation are not present in the B. dentium Bd1 genome. It may synthesize cysteine in a manner similar to that suggested for B. longum subsp. longum NCC2705 using homologs of the genes for cysteine synthase/cysthathione beta synthase, O-acetylhomoserine aminocarboxypropyltransferase and cystathionine γ-synthase, and utilizing a reduced sulphur-containing compound as a sulphur source [23]. Genes encoding complete biosynthetic pathways for purines and pyrimidines from glutamine, as well as for riboflavin, thiamine and folate were identified, while no homologues were present for pathways to produce biotin, pyridoxine, cobalamin, panthotenate and niacin/nicotinic acid, which are also variably distributed in the genomes of other sequenced bifidobacterial genomes (data not shown).
Comparative analysis against the Transport and Classification Database [63] predicts that the B. dentium Bd1 genome contains 771 genes encoding (components of) transport systems, accounting for almost 34% of the total number of ORFs (Figure S8A).
Transport in B. dentium Bd1 is largely carried out by transporters or carriers (e.g., uniporters, symporters and antiporters) and by P-P-bond hydrolysis-driven transporters. A large proportion of the identified transporters are ATP-dependent, as expected for a microorganism lacking an electron transport chain [64]. Annotated solute-transporting ATPases include P-type, F-type and ABC-type. The P-type ATPases are predicted to be involved in the transport of calcium and potassium, whereas the F-type ATPases (e,g., F0F1ATPases) use an electrochemical gradient of H+ or Na+ to synthesize ATP, or hydrolyze ATP to reverse the electrochemical gradient [65]. As described below, a single predicted H+-transporting ATP synthase-ATPase is encoded by the B. dentium Bd1 genome. We identified 298 predicted ABC-type ATPases, of which about 70% are categorized as importers, representing the most abundant transport category, and accounting for almost 13% of all B. dentium Bd1 gene products. The ABC transporters identified have a predicted specificity for a wide variety of substrates, including amino acids, carbohydrates, oligopeptides, osmoprotectants (e.g., proline/glycine, betaine, choline), inorganic ions (e.g., Fe3+, Co2+, Mn2+, phosphate, nitrate, sulphate, and molybdenum) and antimicrobial peptides.
The vast majority of carbohydrate-modifying enzymes encoded by B. dentium Bd1 are predicted to be intracellular and so the uptake of sugars with a low degree of polymerization is a key component of B. dentium carbohydrate metabolism. The genome of B. dentium Bd1 encodes at least 167 ABC transport systems for dietary carbohydrates (Table S1). The B. dentium Bd1 genome also specifies two phospoenolpyruvate-phosphotransferase systems (PEP-PTS), consisting of the two general energy-coupling components, enzyme I (EI) and a heat-stable protein (Hpr), and two different sugar-specific multiprotein permeases known as enzyme II (EII).
The human oral cavity is a complex microbial ecosystem, the composition of which may vary depending on the frequency and nature of food ingestion with consequent fluctuations in biofilm pH. Compared to the distal bowel, where organisms are presented with a relatively consistent stream of molecules that cannot be metabolized or degraded by the more proximal microbiota, the oral cavity microbiota is exposed to the full contents of the ingested foods. Thus, possessing extensive catabolic abilities for carbohydrates is a potent energy-harvesting mechanism for B. dentium Bd1. Genomic data combined with our own data suggests that B. dentium Bd1 has a significantly larger arsenal of genes allowing for breakdown of sugars, also called glycobiome [44], as compared to other bifidobacterial species [23]–[25] or other characterized members of the oral microbiota (Figure 7C and 7D). Classification according to the Carbohydrate Active Enzymes (CAZy) system of Coutinho & Henrissat (1999) showed that the Bd1 genome specifies 117 carbohydrate-active genes including glycoside-hydrolases (GH), glycosyl-transferases (GT) and glycosyl-esterases (CE), which are distributed in 27 GH families, seven GT and three CE families (Figure 7A and 7B). The majority of the identified GH enzymes from B. dentium Bd1 are predicted to be intracellular, with a putative cellulase (BDP_2148) and a xylosidase (BDP_0236) predicted to be the only extracellular GH enzymes. Members of GH families that had previously not been detected in bifidobacterial genomes are GH78 (BDP_2152) and GH94 (BDP_2127), which are predicted to be involved in the metabolism of fucose (GH78) and cellobiose (GH94). Furthermore, the Bd1 genome encodes a wide variety of enzymes to ferment different pentose sugars (e.g., xylose, ribose and arabinose).
The fermentation abilities of this oral strain are clearly broader than those of the phylogenetically related enteric B. adolescentis ATCC15703 (Figure S9), probably reflecting the ecological niche it occupies, which apparently contains a higher variety of available sugars as compared to the distal regions of the gastrointestinal tract.
In addition to the transient food components, the human oral cavity is coated with large amounts of viscous secretion produced by the acinar cells of the salivary glands. This secretion consists predominantly of a heterogeneous population of glycoproteins, commonly referred to as salivary mucins [66]. These large, heavily glycosylated glycoproteins play a major role in the maintenance of viscoelastic properties of saliva, participate in the formation of the protective oral mucosal mucus coat and tooth enamel pellicle [67]. Salivary mucins are comprised of 20–22% protein, 0.2% covalently bound fatty acids, and 68–72% carbohydrate [66]. Notably, the carbohydrate component consists mainly of fucose, mannose, galactose (Gal), N-acetylglucosamine (GlcNAc) and N-acetylgalactosamine (GalNAc). The B. dentium Bd1 genome contains an extensive gene repertoire that appears to be dedicated to the metabolism of the backbone of mucin-containing carbohydrate structures, such as Galβ-1,3-GalNAc or Galβ-1,4-GlcNAc disaccharides. This repertoire includes genes for predicted enzymes such as a glucosaminidase and β-galactosidase that could be involved in the removal of monomeric carbohydrates from mucins. Moreover, the presence of a gene encoding a putative fucosidase enzyme (BDP_2152) indicates that B. dentium Bd1 can probably degrade fucose-containing glycans, such as those present in salivary mucins [68].
Salivary glycoproteins also contain considerably quantities (3.8–4%) of sialic acid and sulphate which decorate the surface of the mucin sugar backbone [66]. Interestingly, a predicted O-sialoglycoprotein endopeptidase (BDP_1212) and a sialic-acid specific acetylesterase (BDP_0122) were annotated in the genome of B. dentium Bd1. However, genes encoding a sialidase and additional enzymes to degrade sialic acid, which have been identified in other bifidobacterial genomes such as B. longum subsp. infantis ATCC15697 [25], do not appear to be present in the genome of B. dentium Bd1.
Dental caries is initiated by demineralization of the tooth surface due to the action of organic acid formed by dental plaque bacteria, arising from their fermentation of dietary carbohydrates. After fermentable carbohydrate intake, the plaque pH may decrease below the critical pH of 5.5, at which point human enamel undergoes demineralization, within minutes, and may remain acidified for several minutes up to several hours [69],[70]. This rapid acidification may not only cause demineralization of tooth surface but also temporarily inhibit bacterial growth in the oral biofilm. Thus, a high level of inherent acid tolerance appears to be crucial for the cariogenicity of oral microorganisms [71]. When the intracellular pH maintained by B. dentium under varying external pH conditions was experimentally compared to those of other caries-associated oral bacteria, such as Str. mutans and Lactobacillus paracasei, B. dentium Bd1 displayed a superior ability to keep a more neutral internal pH compared to these two other bacteria (Figure S10A and Nakajo, Takahashi and Beighton, personal communication). Moreover, when B. dentium Bd1 was cultivated in a synthetic medium at different pH values the growth of Bd1 was not significantly reduced by the highest level of acidity tested (pH 4), a value which can be reached in the oral cavity after food ingestion (Figure S10B). Notably, other closely related bifidobacteria which occupy a different ecological niche (intestinal vs. oral) do not exhibit this aciduric property (Figure S10B and data not shown). Higher levels of inherent tolerance of oral bacteria to acidification have been related to the presence of a membrane-bound, acid-stable, proton-translocating F1F0 ATPase system whose activity has been considered crucial in maintaining the intracellular pH at 7.5 [71]. In the B. dentium Bd1 genome, the F1F0-ATPase is encoded by the atp operon, but this system is also encoded by other bifidobacteria [72]. However, the genome of B. dentium Bd1 contains two adjacently located genes (BDP_1749 and BDP_1750) encoding a glutamate decarboxylase (GadB) and a glutamate/gamma-aminobutyrate anti-porter (GadC), not present in other bifidobacterial genomes so far published [23]–[25], and known in other bacteria to form a glutamate-dependent acid resistance system 2 (AR2) [73].
Characteristics contributing to the ecological fitness in the oral cavity, such as utilization of different diet-derived carbohydrates, and stress tolerance to antimicrobial compounds and acidic environments, should be discernible in B. dentium Bd1. To determine if B. dentium Bd1 functionally responds to stressful stimuli, we performed transcriptional profiling studies using Agilent arrays (Agilent, Palo Alto, Ca., USA) that contain oligonucleotides representing 2114 of the 2143 predicted B. dentium Bd1 protein-encoding genes.
Although B. dentium is not an invasive, life-threatening pathogen it plays a role in tooth tissue destruction and infects tooth dentine, and there are a number of ORFs that code for potential colonization or virulence factors, such as adhesins, exoenzymes, protease- and cytokine-modulating molecules, as well as putative hemolysins (see Table 2 for an overview of putative virulence factors of B. dentium Bd1, some of which will be discussed below). The latter are similar to hemolysins from oral pathogens, including Hemolysin A from S. mutans, a coiled-coil myosin-like protein. Among the putative B. dentium Bd1 surface antigen proteins, the BDP_0164 protein displays 51% similarity to the T. denticola pathogen-specific surface antigen, and is flanked on one side by an ORF (BDP_0163) involved in iron metabolism (high-affinity iron permease) and on the other side by a gene (BDP_0165) encoding an integral membrane protein with high similarity to a protein encoded by the oral pathogen Fusobacterium nucleatum. This suggests that the gene cluster (BDP_0163-BDP_0165) is involved in iron acquisition and adhesion.
Oral bacteria can adhere to salivary agglutinin, other plaque bacteria, extracellular matrix and epithelial cell-surface receptors [76]. In the most intensely studied oral pathogen, S. mutans, two major adhesins mediate this attachment: cell-surface or adhesion proteins, such as SpaA adhesins [77], and sucrose-derived glucans (e.g., gbpB). A homolog (BDP_2059; 32% identity) of the gene encoding a major adhesin of viridans streptococci, SpaA, was identified in the genome of B. dentium Bd1 (Table 2). SpaA binds to human salivary agglutinin, collagen and cells of certain oral pathogens, such as Actinomyces naeslundii [78],[79]. Notably, the BDP_2059 putative adhesin also contains a domain that is similar to a domain of a Streptococcus gordonii protein which mediates strong lactose-inhibitable coaggregation [80]. Furthermore, part of an operon that is required for the synthesis of cell wall polysaccharides in S. mutans UA159, i.e. rgpA, rgpB, rgpD and rgpC [81],[82], exhibits clear homology with B. dentium Bd1 ORFs BDP_1864, BDP_1864a, BDP_2047 and BDP_2048, respectively. In S. mutans these genes play not only a crucial role in binding to human oral tissues [83],[84] but they also participate in serotype determination [85]. These genes also show a strong divergence in G+C content relative to the remainder of the genome, indicating that this region has been acquired by horizontal gene transfer.
The B. dentium Bd1 genome also specifies surface proteins with domains that resemble those (Pfam number 31902) responsible for inter-bacterial aggregation by a choline-binding domain. Such choline-binding motifs are present in the ligands of the most important pneumococcal virulence proteins, encoded by pspC and pspA, which presumably act as adhesins that bind to host factors such as IgA and factor H [86]. Notably, and in contrast to the available chromosomes of enteric bifidobacteria, the B. dentium Bd1 genome harbors five adjacent genes that encode proteins containing such predicted choline-binding domains (BDP_2045, BDP_2054, BDP_2056, BDP_2059 and BDP_2061).
Furthermore, a large number of predicted surface and extracellular proteins that may be involved in host attachment and interaction were identified in a similar fashion as described for other oral pathogens [87],[88] (Table 2).
Other potential adhesion and virulence factors include glycoprotein-binding fimbriae that, in the oral cavity, may mediate the recognition of and adhesion to salivary proline-rich proteins that bind to tooth and mucosal epithelial cell surfaces. They may also bind to cell wall polysaccharides of certain oral bacteria [89]–[94]. So far bifidobacteria have not been shown to possess any fimbria-like structures on their cell surface, although homologs of fimbrial subunits have been identified in intestinal bifidobacteria [23],[44]. Notably, four loci encoding homologs of known fimbrial subunits FimA, FszB and FszD were identified in the B. dentium Bd1 genome (Figure 9). FimA (BDP_0535 and BDP_1224) displays high identity to FimA homologs found in oral commensals such as A. naeslundii and A. odontolyticus [91],[95].
Finally, B. dentium Bd1 encodes a number of putative proteases that may contribute to virulence by their ability to degrade host proteins for bacterial nutrition [96],[97] (Table 2).
We report here the B. dentium Bd1 genome sequence that constitutes the first genome based analysis of a bifidobacterial taxon recognized as an opportunistic pathogen. Although a large number of bacteria coexist in the oral cavity and upper respiratory tract in humans they have evolved to form a microbial community with complex physical and biochemical interactions. The analysis of the B. dentium Bd1 genome and comparisons with other bifidobacteria residing in the human intestine has revealed insights into the particular evolution and adaptive responses of the opportunistic pathogen B. dentium to the oral cavity. B. dentium, unlike its intestinal relatives that often are claimed to promote the health-status of their host, contributes to the destruction of the dentition. The elucidation of a putative“cariogenic gene suite” would provide salient targets to test the relative contribution of B. dentium to a core cariogenic microbiome, its impact on microbial colonization and succession, and its phylogenetic distribution. It remains unknown if these genomic features constitute a unique competitive strategy evolved in B. dentium. As such, it differs from other known bifidobacteria in several aspects of its basic physiology and its adaptation to an ecological niche. As our genome analysis shows, B. dentium can metabolize a much larger variety of carbohydrates than other Bifidobacterium species sequenced so far and greater than the range of oral streptococcal including S. mutans. Genes encoding putative virulence factors associated with adhesins, acid tolerance, defense against toxic substances and capacity in utilizing saliva-derived components, represents genetic evidence of the capacity of B. dentium to colonize the oral cavity and to proliferate within active carious lesions. The genome sequence, when explored using functional genomics approaches, will permit the analysis of genes involved in colonization, survival, growth and pathobiology of B. dentium in this unique polymicrobial environment.
The strain used in this study B. dentium Bd1 is equivalent to the type strain of B. dentium species (ATCC27534 or LMG11045 or DSM20436 or JCM 1195).
The genome sequence of B. dentium Bd1 was determined by shotgun sequencing and subsequent gap closure (Agencourt Genomic Services, MA, USA). The Bd1 genome was sequenced to approximately 10-fold coverage and assembled with Phred [98], Phrap and the Staden package [99]. Automated gene modelling was achieved using multiple databases and modelling packages as described previously [100]. Additional information on sequencing, bioinformatic, and functional genomics analyses are provided in Text S1.
The sequence reported in this article has been deposited in the GenBank database (accession number CP001750).
|
10.1371/journal.pbio.1002263 | Sleep-Dependent Reactivation of Ensembles in Motor Cortex Promotes Skill Consolidation | Despite many prior studies demonstrating offline behavioral gains in motor skills after sleep, the underlying neural mechanisms remain poorly understood. To investigate the neurophysiological basis for offline gains, we performed single-unit recordings in motor cortex as rats learned a skilled upper-limb task. We found that sleep improved movement speed with preservation of accuracy. These offline improvements were linked to both replay of task-related ensembles during non-rapid eye movement (NREM) sleep and temporal shifts that more tightly bound motor cortical ensembles to movements; such offline gains and temporal shifts were not evident with sleep restriction. Interestingly, replay was linked to the coincidence of slow-wave events and bursts of spindle activity. Neurons that experienced the most consistent replay also underwent the most significant temporal shift and binding to the motor task. Significantly, replay and the associated performance gains after sleep only occurred when animals first learned the skill; continued practice during later stages of learning (i.e., after motor kinematics had stabilized) did not show evidence of replay. Our results highlight how replay of synchronous neural activity during sleep mediates large-scale neural plasticity and stabilizes kinematics during early motor learning.
| Sleep has been shown to help in consolidating learned motor tasks. In other words, sleep can induce “offline” gains in a new motor skill even in the absence of further training. However, how sleep induces this change has not been clearly identified. One hypothesis is that consolidation of memories during sleep occurs by “reactivation” of neurons engaged during learning. In this study, we tested this hypothesis by recording populations of neurons in the motor cortex of rats while they learned a new motor skill and during sleep both before and after the training session. We found that subsets of task-relevant neurons formed highly synchronized ensembles during learning. Interestingly, these same neural ensembles were reactivated during subsequent sleep blocks, and the degree of reactivation was correlated with several metrics of motor memory consolidation. Specifically, after sleep, the speed at which animals performed the task while maintaining accuracy was increased, and the activity of the neuronal assembles were more tightly bound to motor action. Further analyses showed that reactivation events occurred episodically and in conjunction with spindle-oscillations—common bursts of brain activity seen during sleep. This observation is consistent with previous findings in humans that spindle-oscillations correlate with consolidation of learned tasks. Our study thus provides insight into the neuronal network mechanism supporting consolidation of motor memory during sleep and may lead to novel interventions that can enhance skill learning in both healthy and injured nervous systems.
| The cardinal features of motor skill learning are enhanced speed and automaticity of motor execution with preserved accuracy [1–3]. Motor learning is known to progress through a series of stages: an early stage accompanied by rapid improvements in accuracy with continued variability of movement kinematics, followed by consolidation of these processes and transition to a later stage of learning, in which kinematics are largely stabilized but slow improvements in accuracy continue to occur [4–6]. The underlying neural basis by which kinematics become stabilized during early motor learning is not well understood. Human studies suggest that non-rapid eye movement (NREM) sleep is essential for this consolidation and, in addition, results in additional gains in skilled motor performance [7–14]. Even brief naps during the day can mediate these “offline” motor improvements [11,15,16], including faster movements and reduced variability in timing.
There is evidence that NREM sleep promotes offline gains. Prior studies have described a relationship between motor learning, local slow-wave oscillations [17], the expression of immediate-early plasticity-related genes [18] and stabilization of dendritic spines [19]. However, how large-scale patterns of neural activity drive plasticity of specific motor circuits to result in enhanced motor performance is unknown. Based primarily on studies of single-unit activity conducted during hippocampal-dependent behaviors [9,13,16,17,20–25], we hypothesized that reactivations of task-related emergent neural firing during NREM sleep may be related to subsequent neural plasticity and associated offline behavioral gains. This hypothesis is largely consistent with theoretical models for how NREM sleep promotes learning more generally [20–22,26–29], but to our knowledge, there is little experimental support of this during procedural memory formation.
Microelectrodes were implanted (tetrode and microwire arrays were used in different animals, see Materials and Methods) into the lateral part of the caudal forelimb area of rats, the region most strongly associated with fine motor control of the distal forelimb, and the region directly involved in plasticity following skilled motor learning [30–32] (S1 Fig). Five days after electrode placement, animals began skilled motor training (Fig 1A and 1B). Skilled motor learning was conducted using the Whishaw forelimb reach-to-grasp task [33,34]. We chose this task both due to homology to skilled learning tasks in humans [35,36] and the extensive evidence that this task is associated with multiple levels of neural plasticity, including changes in Long-Term Potentiation (LTP) [37], spine growth [38–40] and motor map plasticity [30,41,42]. Neural activity was monitored during the following sequence of blocks: a “baseline” sleep block (Sleep1), a skilled motor learning session (Reach1), a sleep block (Sleep2), and a subsequent learning block (Reach2) (Fig 1C). Thus, we were able to compare both how task-related neural activity was modulated after sleep (i.e., by comparing Reach1 and Reach2) as well as how motor learning affects neural activity during sleep (i.e., by comparing Sleep1 and Sleep2).
Motor skill learning is typically assessed across two dimensions: speed and accuracy [1,13,43,44]. We examined both here, measuring accuracy as percent success in retrieving the pellet and speed as the overall time the animal took to perform the full reach-grasp-retract motor sequence (Fig 2). Online changes in skilled motor performance were quantified by comparing the first 20 trials (hereafter “Reach1early”) to the last 20 trials (hereafter “Reach1late”) of the first learning session; offline gains were measured by comparing the last 20 trials from the initial learning session (i.e., “Reach1late”) with the first 20 trials from the subsequent reach block (hereafter “Reach2early”). Across all animals, we found significant online improvements in accuracy (Fig 2D, p < 0.001, Wilcox rank-sum test) in Reach1early versus Reach1late; see Materials and Methods) but without improvements in speed (Fig 2C, overall ANOVA F(2,98) = 6.6, p < 0.01; post-hoc t tests comparing Reach1early with Reach1late, p = 0.3) during the initial training session (Reach1). Following sleep, however, animals executed the entire movement sequence considerably faster (Fig 2C, post-hoc t test comparing between Reach1Late and Reach2early, p < 0.001), with no decrements in accuracy (Fig 2D, p = 0.3, Wilcox rank-sum test between Reach1Late and Reach2early). Thus, sleep appeared to increase movement efficiency and was associated with no decrement in accuracy.
To further probe the effects of sleep on motor performance, we also analyzed online and offline changes in the trajectory of forelimb movements. To perform this analysis, we calculated whether there was a change in movement trajectories either during online learning or after sleep (S2 Fig). Trajectories were calculated both using an “external frame of reference” (i.e., relative to the end-point position of the pellet) and an “internal frame of reference” (i.e., relative evolution of the trajectory after the start of movement). During online training, we found that animals changed their external frame trajectory, suggesting greater orienting towards the pellet at the start of the movement, without significantly altering the kinematics of the trajectory itself. In contrast, after sleep, while the externally referenced trajectory remained unchanged, the internally referenced trajectory kinematics appeared to be significantly changed. This suggests that different aspects of kinematics are modified during online versus offline processing.
To investigate how changes in neural activity underlie the offline gains in motor efficiency described above, we compared the task-related activity during Reach1early, Reach1late, and Reach2early. For each neuron, we calculated the peri-event time histogram (PETH, smoothed using a Poisson-based Bayesian-adaptive regression model [45]) time-locked to reach-onset (Fig 3A). Reach onset was defined by the start of physical movements (i.e., identical to those used to calculate the behavioral metrics above) and not based on external cues. During each of the early, late, and post-sleep blocks, we estimated the degree of modulation (i.e., the peak firing divided by the pre-reach baseline) and the time to peak firing. Even in the very earliest learning block (Reach1), 92/102 neurons showed some evidence of task modulation, defined as at least a 2-fold increase in firing rate compared to the baseline (S3 Fig). As with other studies that have examined neural recordings during forelimb reach tasks in rodents [46,47], we found that single units demonstrated time-locked activity across many phases of reach (S3 Fig; See figure legend for further discussion on the distribution of neural activity observed). Three units were excluded from further analyses because of very sparse firing, which made it difficult to properly estimate PETH and/or timing information.
We next analyzed online and offline changes in the task-related modulation of neural activity. Interestingly, we found a strong and specific effect of sleep in changing both the timing and magnitude of task-related activity at a single-unit level (example shown in Fig 3A). During online learning, the relative timing to reach-onset (Fig 3B, Kolmogorov-Smirnoff, p = 0.9) and task-related modulation (Fig 3C, overall repeated-measures ANOVA F(2,98) = 5.5, p < 0.01; p = 0.9, paired t test between Reach1 and Reach2) did not change significantly. However, after sleep, there was a strong and highly significant change in the time to peak of the PETH (i.e., “temporal coupling shift,” Fig 3B, Kolmogorov-Smirnoff p < 0.001) and a smaller but significant change in task-related modulation (Fig 3C, p < 0.05, paired t test comparing modulation for Reach2early to both Reach1 blocks). Across all neurons, we also analyzed the mean shift following sleep. We found that there was a 200 ± 65 ms shift in the temporal coupling of neural activity to reach onset after sleep (repeated-measures ANOVA F(2,98) = 7.7, p < 0.001; p < 0.001 paired t test comparing modulation for Reach2early to both Reach1 blocks). S4 Fig shows the entire distribution of shift in neural firing for neurons during both the online and offline periods, demonstrating a widespread effect across most neurons we recorded from.
Prior reports in humans have demonstrated that sleep is required for offline performance gains [9,13]. To study whether sleep is essential for the behavioral and neural changes described above in our experimental paradigm, we performed a follow-up experiment in which a new set of animals were given 2 h of sleep-restriction between Reach1 and Reach2 (see Materials and Methods). To assess how this modified offline changes in behavior and neural firing, we compared the last 20 trials in Reach1 with the first 20 trials in Reach2 (Fig 4A). We found significant differences in offline changes in movement speed (Fig 4B “motor speed,” p < 0.001, 2-sided t test comparing the sleep and sleep restriction groups). We also found significant differences in the offline changes in the temporal shifts of peak task-related activity when compared to animals allowed to sleep (Fig 4B “neural speed,” p < 0.01, 2-sided t test). Fig 4C also shows the cumulative distributions of the individual timings of neural PETHs both before and after the sleep restriction (Kolmogorov-Smirnoff p = 0.7 between Reach1late and Reach2early). Thus, our data suggests that in the absence of sleep there were no changes in movement or neural speed, i.e., no offline gains at either a behavioral or neural level. It is also important to note that the sleep-restriction paradigm did not impair performance at either a behavioral or neural level; these animals were neither significantly slower (p = 0.7, one-sample t test comparing last 20 trials pre versus first 20 trials post sleep-restriction, n = 5 animals) nor significantly less accurate (p = 0.15, one-sample t test comparing accuracy in the last 20 trials before versus first 20 trials after sleep-restriction, n = 5 animals). Moreover, the population-tuning curve was not significantly different (Fig 4C). Thus, we did not observe any significant changes in performance (i.e., either a decrement or offline gains) after the period of sleep-restriction, but likewise, no offline gains in neural or motor speed.
The experiments described above were conducted with animals that were allowed to sleep twice (i.e., Sleep1 and Sleep2), with offline behavioral changes occurring after the second sleep block. The sleep-restriction control described above indicates that such offline gains do not occur simply as a result of the passage of an equivalent period of time awake. However, it is possible that the effects observed (e.g., the changes in movement speed after Sleep2) occur simply as a result of the animals having more overall sleep before performing the task. In other words, because animals in our experimental group are able to sleep for two sessions, it is possible that what we interpret as offline gains are simply a product of normal practice-related improvements (i.e., during online training) that occur when animals are well-rested. To address this issue, we allowed more sleep prior to the initial training block. Thus, animals were allowed to undergo both Sleep1 and Sleep2 blocks before any training occurred (S5A Fig). Animals were awoken for 20 min between those two blocks and were given 50 pellets to generally mimic the task structure of the first set of experiments. Animals were then given 150 trials of reach training to assess whether having extended sleep prior to training would result in improvements in motor speed during the training sessions itself. We found that extra sleep prior to training was not sufficient to induce significant changes in speed during training (S5B–S5D Fig, p = 0.36, paired t test, comparing the first 20 and last 20 trials, n = 5 animals). Animals were also tested the next day (i.e., after 24 h), to observe whether they showed similar offline gains as described in our early experiments. Indeed, these animals showed offline improvements in speed (p < 0.0001, paired t test comparing last 20 trials from Day 1 and the first 20 trials on Day 2, n = 5 animals) with maintenance of accuracy (p = 0.1, paired t test comparing last 20 trials from Day 1 and the first 20 trials on Day 2, n = 5 animals). This control experiment further confirms the role of offline processes in mediating speed improvements during skill acquisition.
Having demonstrated large-scale, sleep-dependent changes in neural activity and related motor behavior, we next examined what neural processes may be mediating these effects during the sleep block. We hypothesized that replay of task-related neural ensembles during NREM sleep drives the offline behavioral gains and neural modulation previously described. To investigate this, we used principle components analysis to identify task-related patterns of neural activity (i.e., neural ensembles) and then probed replay of these ensembles during sleep using methods described previously [29,48,49]. Across animals, mean time spent in NREM sleep during the Sleep1 block was 28.6 ± 11.3 min, and the mean time spent in NREM sleep during Sleep2 block was 24 ± 4.1 min (paired t test = 0.6).
Ensemble reactivation during sleep was quantified by applying a template created using principle components analysis (PCA) of the task-related neural activity. In other words, PCA was used to create templates that captured patterns of synchronized neural activity during task performance (S6 Fig). PCA resulted in a number of principle components (hereafter termed “ensembles”) that reflected patterns of common variance across the recorded single-units, with each component comprised of weights that reflected the contribution of each neuron to that particular ensemble. To represent the activity of a particular ensemble the traditional method is to multiply the weights from each neuron in a particular ensemble with the Z-scored activity matrix [48]. This same method can be used during sleep to assess the degree to which this ensemble is being reactivated (S6 Fig) [29,50]. Specifically, the ensemble defined from the task was multiplied by the Z-scored neural activity recorded during sleep blocks, resulting in a one-dimensional vector that represented the “activity” of that ensemble during the sleep blocks before and after (Fig 5A). Reactivation was thus defined as increased “activity” of the ensemble during the sleep-block after learning compared to the sleep-block prior to learning (Fig 5A–5C). In this relatively rapid motor task (~1 s), the first ensemble captured more variance than any other component and we therefore focused our analysis on it. Prior reports using this method have found that weaker ensembles (i.e., those with lower eigenvalues) show limited evidence of reactivation [50]. When we did examine the second ensemble, we found considerably more variability in terms of reactivation across animals (i.e., only some animals showed evidence of reactivation of this ensemble).
After learning, activation strength was significantly stronger during NREM sleep blocks for these task-related ensembles in comparison to the sleep block that occurred before learning (Fig 5B, p < 0.001, Wilcox non-parametric sign-rank test). We originally hypothesized a significant relationship would occur between reactivation of neural ensembles and subsequent temporal shifts of cortical neurons. To evaluate whether such a relationship existed, we calculated the ensemble reactivation for each unit. Reactivation was defined by calculating the difference in activation strength between Sleep1 and Sleep2 for that ensemble (i.e., Sleep1Activation−Sleep2Activation), multiplied by the principle component (PC) score for each neuron in that ensemble, resulting in that neuron’s “weighted” reactivation score. Interestingly, we found that the degree of reactivation at a single neuron level during sleep strongly and significantly predicted the increase in temporal coupling to reach onset that occurred upon awakening (Fig 5C, r = -0.41, pearson correlation, p < 0.001, analysis conducted across all neurons, n = 4 animals). This indicates that neurons that experienced the strongest reactivation during the NREM sleep block also experienced the greatest shift in temporal coupling to reach onset during the skilled reach task upon awakening.
To better understand the relationship between reactivation events and task-related neural activity, we performed two additional analyses (S7 Fig). First we performed an analysis to demonstrate that the reactivations truly reflect a specific pattern of task-activity—in other words, that the same neurons active during the task are highly active during the replay event. To perform this analysis, we calculated PETHs for each neuron during the ensemble replay (binned at 25 ms, and including data 250 ms before and 250 ms after each replay event, using only the top 10% of activation strengths) and compared this with the PETHs and PC ensemble created from activity during the reach task (S7A and S7B Fig). Importantly, this binning allows us to estimate variability in firing rates across neurons during reactivation events, but not temporal variability across neurons. Each PETH was then sorted according to the PC weight extracted during the task block. As predicted, there was good correspondence between the PC weight and the degree to which these neurons were firing during the reactivation (S7C Fig). This provides evidence that reactivations represent synchronous co-activation specifically of task-related neurons; in other words, variation in firing rates during the task are observed as variations in synchronous firing during reactivation events during sleep.
Prior studies have distinguished “reactivation,” observed as synchronous activity of task-related ensembles during subsequent sleep periods [50], and “replay” which involves a recurrence of sequential activity during subsequent sleep epochs. To assess the degree to which there is replay, we next examined the microstructure of reactivation events at a single millisecond resolution. To perform this analysis, we used a previously described template matching technique [51]. To create the task-template, we extracted the PETH both 250 ms prior to and 1,000 ms after the reach onset in each animal. This activity pattern was binned at different resolutions (50 ms, 125 ms, 250 ms, and 1,250 ms), to create bin templates ranging in size from 25 bins down to 1 bin (S7D Fig). Importantly, each binning resolution contained less temporal structure (i.e., the 1,250 ms bin does not retain any temporal information). The template matrix (neurons x bin) was then correlated with single-unit activity patterns that occurred during reactivation events identified above. Reactivation events were kept at a 1 ms resolution (i.e., no binning) for the template matching procedure. We found a general increase in the degree of template correlation after learning (p < 0.0001 across every template studied, S7D Fig). However, the highest degree of correlation and the greatest change post occurred with the largest bin size (i.e., with the least temporal information). This suggests that even while there is a small but significant change in the temporal structure of reactivation events, they are more linked to synchronous activation of neurons that fired during task performance (another example shown in S7E Fig).
Prior studies have suggested that both spindles and slow-wave oscillations may mediate offline gains in motor performance [9–11,13]. To further probe the relationship between ensemble reactivations and these phenomena, we calculated the event related local field potential (LFP) locked to the top tenth percentile of reactivation events after sleep (Fig 6). We first calculated the event-triggered average (ETA) of LFP filtered at slow/delta-oscillations (i.e., 0.5–4 Hz). We next subdivided spindles into slow and fast frequencies because of prior evidence suggesting that fast-spindles in particular are specific to offline gains [10,51]. We thus examined the ETA of slow-spindles (filtered at 9–12 Hz) and fast spindles (filtered at 13–16 Hz). Interestingly, we found an increase in the locking of fast but not slow spindles with reactivation events after learning (Fig 6A and 6B, * above indicates significant post-hoc differences using a paired t test). Likewise, we saw a change in the association of these events with slow oscillations (Fig 6C). We probed phase-locking in two different ways. First, we calculated the coefficient of variance across events for those time points that were significantly different in Sleep1 versus Sleep2. As expected, we found a significant reduction in the coefficient of variation (CV) across these time points for fast-spindles and slow-oscillations (Fig 6D, p < 0.0001, ranksum test). Because there were no significant differences in the slow-spindle frequencies, to assess CV we used the same time points as used for fast-frequencies. We found a slight increase in the CV (Fig 6D, p < 0.05 ranksum test), suggesting that the increased locking is specific to fast-spindle oscillations and not a general phenomenon. We also calculated the instantaneous phase at the slow and fast spindle oscillation, and the slow-wave oscillations at t = 0 (i.e., when the reactivation event occurred), using circular statistics comparisons [52]. Prior to learning, there was no significant phase-relationship between fast-spindles and reactivation events. However, after learning, the mean phase (in radians) of the fast-spindle oscillation at the time of these reactivation events -2.06, (95% confidence interval [CI] of .5542). There was not a significant phase-relationship between reactivation events and slow spindle oscillations. Finally, there was evidence for a slight but significant phase-shift after learning in coupling with slow-oscillations. Prior to learning, the mean phase of the slow oscillation at the time of reactivation was -2.395 (95% CI .0953, using circstats toolbox); after learning, the mean phase at the time of reactivation was -2.07 (95% CI .077, using circstats toolbox), a highly significant difference (p < 0.0001 ANOVA using circstats toolbox). These analyses demonstrate that, independent of changes in the mean amplitude, after learning there was significant phase-coupling to the spindle-oscillations, and a significant phase shift relative to the slow-wave oscillation, and these changes partly explain the results observed in Fig 6A–6D.
These results suggest that following learning, high reactivation events are more strongly coupled to both fast spindle oscillations and slow oscillations (Fig 6E). To probe this further, we calculated the instantaneous analytic amplitude of the local field potential filtered in the fast-spindle frequency at the trough of the reactivation-triggered slow-oscillation for reactivation event. Across events, we found a strong and significant increase in the analytic amplitude at this time-point (Fig 6F, p < 0.001, Wilcox ranksum). Finally, using an automated detection algorithm, we assessed whether there was a significant change in the proportion of distinct spindle events associated with these high strength reactivation events. Specifically, we analyzed the LFP filtered in the fast-spindle frequency time-locked to high-reactivation events pre/post learning, to assess whether there was an increase in the proportion of spindles associated with the highest reactivation events after learning. We found that after motor learning, spindles were 60% more likely to occur in association with these high reactivation events compared to before learning. (Fig 6G, p < 0.001, Wilcox ranksum).
We next analyzed patterns of activity during “later-stages” of learning, specifically defined here as continued practice on the skilled motor task on subsequent days. Prior research has divided motor learning into an early phase associated with the establishment of gross kinematic patterns and rapid gains in motor performance, and a later phase associated with overall kinematic stability and slower incremental gains in skill acquisition [5]. We have so far demonstrated that reactivation of task-related ensembles occurred after the initial motor learning session, when animals appeared to first form and then consolidate a novel kinematic trajectory. To explore whether changes in reactivation continues to occur through later stages of motor practice, we investigated whether task-related neural activity continued to experience an increase in the reactivation strength on subsequent days of motor learning (Fig 7A). For this analysis, data was gathered from three animals across two additional days of motor training. As there were no significant differences between day 2 and 3 on any of the parameters assessed below, data was pooled across these two days for the purposes of this analysis.
On subsequent sessions of motor learning (i.e., conducted on days 2–3), there were no further offline changes in movement speed/efficiency (example of one animal Fig 7B, p > 0.8 comparing Reach1Late and Reach2Early across all animals), indicating overall stabilization of the kinematic pattern. Despite this, there was continued evidence of learning, i.e., accuracy improved by 33% during the reach training in these subsequent sessions p < 0.05 Wilcox rank-sum test. The stable kinematics and slow improvements in accuracy are largely consistent with these subsequent training days being part of the late/slow-period of motor learning.
During this later phase of learning, we found no evidence of increased reactivation strength of task-related ensembles during sleep (Fig 7C, sign-rank p = 0.2). To assess whether this lack of an effect was significantly different compared to changes observed on day 1 of motor learning, we next calculated the overall reactivation (rank-ordered Sleep2–Sleep1 activations for each animal) on day 1 and compared this with reactivation on subsequent days of training. In addition, because prior analyses [29,50] have suggested that there is a highly skewed distribution of ensemble activations during sleep (i.e., many low-value activations), we further subdivided reactivation differences according to the overall percentile strength and compared across deciles (Fig 7D). This analysis demonstrates the highly skewed nature of these re-activations while also demonstrating the lack of an effect in later motor learning periods. Finally, across behavioral sessions in these animals over the first 3 d, we found a significant correlation between the degree of reactivation observed (using the top 10% reactivations pre versus post learning) versus changes in the motor speed before and after sleep (Spearman correlation, Rho = -0.76, p < 0.05, Fig 7E).
Many previous studies have examined the role of sleep in promoting offline gains in motor skill performance, with NREM sleep in particular promoting various types of motor learning [9,13,14,16,17,53,54]. However, little was known about the neurophysiological processes by which NREM sleep mediated motor memory consolidation. Here we show that reactivation of task-related neural patterns during NREM sleep is explicitly related to both performance improvements and plasticity of neural responses. We specifically found that these high reactivation events were closely linked to an increase in fast spindle oscillations and became slightly phase-shifted relative to the slow oscillations. Finally, we found that neural reactivation is very specific to early motor learning, and not simply a reflection of motor practice; NREM-sleep reactivation was not evident during later stages of motor learning once kinematic patterns had been stabilized. These results suggest that task-related neural reactivation during NREM sleep plays a key role in stabilizing the basic motor pattern during motor learning, with subsequent improvements not dependent on large-scale reactivation during sleep.
We show here that sleep-dependent reactivation of neural ensembles occurs in the context of procedural learning. Prior demonstrations of reactivation in cortex have been described in visual [21] and prefrontal [20,50,55,56] cortex. Importantly, these studies occurred in the context of hippocampal-dependent behavior and primarily in coordination with replay events from the hippocampus [20,21,50,55,56]. Another demonstration of sleep-dependent reactivation, by our group, occurred in the context of neuroprosthetic learning [29]. While neuroprosthetic learning is beginning to be explored in more detail [57–59], it is unclear whether this learning represents declarative, procedural, or some more abstract form of learning that is fundamentally different than either of the above. We thus identify the role of neural reactivation measured at single-neuron resolution for motor memory consolidation.
Prior research has divided motor learning into an early phase, associated with the establishment of gross kinematic patterns and rapid gains in motor performance, and a later phase, associated more strongly with subtle refinements of kinematic patterns and more incremental changes in skill acquisition [1,2,5]. We show here that large-scale reactivation of neural ensembles is associated with kinematic improvements and related changes in neural activity patterns. Indeed, our study suggests that offline processing during sleep may play a key role in the consolidation of motor memories, thus allowing animals to transition to a later phase of learning that is more strongly associated with more subtle motor refinements and increased automation [60].
We also found that after sleep there was greater temporal binding of single-unit activity in motor cortex to the initiation of the reach movement. We hypothesize that the faster activation of neurons after sleep are related to the binding of separate motor programs encoding specific parts of the complex movement (i.e., “reach, grasp, retract”), into one integrated motor program. While our evidence suggests that motor cortex is the final output of this program, it may not be where this program is ultimately “stored;” in other words, the ensembles controlling the different motor actions may be bound together by intrinsic connectivity within motor cortex or may be activated through distributed circuits that occur across cortico-striatal or prefrontal circuits [61]. It is important to point out that in hippocampal-dependent replay, systems consolidation theory would suggest that memories are being transferred from subcortical to cortical representations. In motor learning, it may well be that memories are being transferred from cortical to subcortical representations, perhaps representing a fundamentally different role for reactivation. Further research will be required to assess these ideas.
Studies conducted in both rodents and humans have demonstrated that early motor learning and motor cortical activity itself involves a distributed set of circuits including attentional/prefrontal regions, as well as cerebellum and dorso-medial striatum (i.e., the associative, prefrontal-connected portions) [3,12,60,62–68]. By contrast, later phases of skill learning seem to be more strongly associated with changes within motor cortex [31,41,64,64] and between dorso-lateral striatum (i.e., the “motor” striatial circuit) [62,67–68]. This suggests that ensemble reactivation during sleep, which seems to occur most strongly in association with this early learning phase, may occur through distributed reactivation across cortical and subcortical regions involved in various aspects of these motor actions.
This theory that motor ensembles are being driven by large-scale distributed networks during early motor learning processes, with stabilization associated with transfer of procedural memory into motor cortex, is consistent with what is known about how hippocampal associative circuits drive cortical replay in NREM sleep following declarative memory paradigms [22–24]. Moreover, this theory is also consistent with a recently proposed “active systems consolidation theory” [26,28] that suggests that spindle-dense Stage-2 sleep [28] serves to synchronize global cortical processes, thus mediating long-range consolidation of synaptic plasticity [18,19] and memory transfer to distant cortical regions [26]. Further studies involving dual recordings from disparate cortical regions during early/late motor learning sessions and during sleep will be required to demonstrate that ensemble reactivation truly is associated with active consolidation across a distributed set of cortical/subcortical brain regions.
Interestingly, we found a specific involvement of fast spindles occurring during slow-wave oscillations in the reactivation of task-related neural ensembles. While largely consistent with a body of work demonstrating the involvement of fast spindles in motor learning [10,11,51], this was quite distinct from what had been observed across the two previous studies demonstrating cortical reactivation in rodents [29,50]. In those studies, cortical reactivations seemed to be coupled most strongly to the peak negativity of the slow-oscillation [29,50] and in the case of prefrontal cortex, also to hippocampal sharp-wave ripples [50], which themselves occurred near the peak negativity of slow-wave oscillations and prior to spindle oscillations. By contrast, here we find that after motor learning, reactivation events seemed to be particularly time-locked to fast spindles that occur a short time after the trough of slow delta waves. This difference may represent an important and fundamental difference between motor and other forms of learning. Indeed, one recent study has demonstrated that during the trough of slow-wave oscillations, cortical neurons are driven strongly by hippocampal circuits whereas during spindle events hippocampal circuits are suppressed and processing is driven strongly by thalamo-cortical circuits [69]. Given that the bulk of non-cortical motor processing (i.e., from the cerebellum and striatum) are transmitted back to the cortex through the ventral thalamus [69], it is certainly plausible that emergent task-related ensembles during NREM sleep may be locked to spindle events, particularly if this reactivation is being driven by distributed mechanisms as previously postulated.
Together, our results shed light on the neural processes associated with offline gains in skilled motor performance. We have identified a specific neural correlate of the widely observed sleep-dependent improvement in movement efficiency and linked them to sleep dependent reactivation of activity patterns established during online earning. Our results particularly emphasize the importance of sleep during early motor learning when motor sequences are initially established. This phenomenon might be most relevant to early skill building during musical or sports training, or during early child development, when essentially all skills across both sensory and motor domains are new [54,70]. In addition, these results suggest a potential mechanism by which NREM sleep may serve to enhance plasticity and functional recovery following brain injury [71].
This study was performed in strict accordance with guidelines from the USDA Animal Welfare Act Regulations and United States Public Health Science (PHS) Policy. The protocol was approved by the San Francisco VA Medical Center Institutional Animal Care and Use Committee (IACUC, Protocol Number 13–006). We used 15 adult Long–Evans male rats (approximately 8 wk old; see S1 Table for complete details of animals, units, etc.). Animals were kept under controlled temperature and a 12–h light, 12–h dark cycle with lights on at 06:00 A.M. Probes were implanted during a recovery surgery performed under isofluorane (1%–3%) anesthesia. The post–operative recovery regimen included administration of buprenorphine at 0.02 mg/kg b.w. and meloxicam at 0.2 mg/kg b.w. Dexamethasone at 0.5 mg/kg b.w. and Trimethoprim sulfadiazine at 15 mg/kg b.w. were also administered post–operatively for 5 d. All animals were allowed to recover for 5 d prior to start of experiments.
We recorded extracellular neural activity using both tungsten microwire electrode arrays (MEAs, n = 3 rats, Tucker–Davis Technologies or TDT, FL) and tetrodes (n = 4 rats, Neuronexus, Michigan). Arrays were implanted in the caudal forelimb area of primary motor cortex (M1), centered at 3–4 mm lateral to bregma, 0.5 mm anterior to bregma to target upper limb primary motor cortex (M1) (S1 Fig). Final localization of depth (1,000–1,500 μm) was based on quality of recordings across the array at the time of implantation. We recorded spike and LFP activity using a 128–channel TDT–RZ2 system (Tucker–Davies Technologies). Spike data was sampled at 24,414 Hz and LFP data at 1,018 Hz. ZIF–clip based analog headstages with a unity gain and high impedance (~1 GΩ) were used. Only clearly identifiable units with good waveforms and high signal-to-noise were used. MEA recordings were sorted offline using PCA-based algorithms followed by manual cluster-cutting using TDT’s OpenSorter software. Tetrodes were sorted using “UltraMegaSort” toolbox (available online at https://physics.ucsd.edu/neurophysics/software.php), a set of MATLAB based scripts for tetrode sorting described in detail previously [29,72]. Specifically, a voltage-based threshold was set based on visual inspection for each channel that allowed for best separation between putative spikes and noise (typically this threshold was 4.5–5 standard deviation [SD] away from the mean). Snippets of data that crossed threshold were time-stamped as events, and waveforms for each event were peak aligned. K-means clustering was then performed across the entire data matrix of waveforms (30 samples/ch x 4 chs x # of waveforms). Automated sorting was performed by: (1) first over clustering waveforms using a K-means algorithm (i.e., split into many mini-clusters), (2) followed by a calculation of interface energy (a nonlinear similarity metric that allows for an automated decision of whether mini-clusters are actually part of the same cluster), and (3) followed by aggregation of similar clusters. Such aggregation allows for a reduction in the total numbers of clusters that need to be manually inspected. Automated sorting was followed by manual inspection and sorting of spikes (including further merging or splitting of automatically identified clusters and removing significant outliers based on Gaussian distribution of PC space), using feature space, auto-correlations, cross-correlations and linear discriminant analysis to determine which clusters represent single units and to prevent over-sorting (S8 Fig). Trial-related timestamps (i.e., trial onset, trial completion and timing of when animals reach the pellet) were sent to the RZ2 analog input channel using an Arduino digital board and synchronized to neural data.
Prior to surgery, animals were handled and acclimated to behavioral boxes and oriented to the pellet tray for 1 wk, at the end of which they were evaluated on 10 trials of the Whishaw forelimb reach to grasp single pellet task to determine handedness. This was followed by electrode implantation on the contralateral motor cortical hemisphere as described above. Five days after electrode implantation, animals were food-restricted for 2 d, followed by feeding animals a fixed amount during the course of training (2 average sized food pellets/day). Whishaw forelimb-reach was conducted using a clear plexiglass chamber, with a 1.5 cm slit for animals to place their forelimb through in order to reach a 45 gm pellet on a shallow dish 1.5 cm away from the front of behavioral chamber, using an automated chamber described in more detail in [34] (Fig 1A and 1B). Animals typically performed from 100–150 reaches in the first reach block (Reach1) and 50–75 in the re-test block (Reach2). All reaches were videotaped for post-hoc analysis of accuracy, kinematics, and dynamics.
All behavioral sessions began in the morning and consisted of 2 h of spontaneous recording (to record a “baseline” sleep period, Sleep1); motor skill learning (Reach1); a second 2-h block of spontaneous recording (Sleep2); and finally a “re-test” motor skill block to assess for changes in behavior/neural activity after sleep (Reach2).
Sleep-restriction experiments were conducted similarly to the experiments described above, with the exception that during the second 2-h block of spontaneous recording (termed Sleep2 above), animals were kept awake. Specifically, sleep-restriction sessions began in the morning and consisted of 2 h of spontaneous recording (to record a “baseline” sleep period, Sleep1), motor skill learning (Reach1), a second 2-h block of sleep-restriction (Sleep Restriction), and immediately after this a re-test motor skill block to assess for changes in behavior/neural activity after sleep (Reach2).
For the sleep restriction experiments, animals were kept in the behavioral box in which they conducted initial training sessions. The animals were closely observed for any behavioral evidence of sleep. In addition, the LFP was monitored in real-time to detect any evidence of sleep signatures. If detected, we gently tapped the box to keep awake. Tapping was typically required less than 1x/min early during the restriction period; and by the end of the restriction period had escalated to around 3 taps/min to keep animals awake. The entire paradigm was carried out identical to the training paradigm. Specifically, neural data was recorded during a 2-h block of spontaneous activity, during which time animals were allowed to sleep. Subsequently, animals performed the skilled motor learning task. After this, we performed sleep restriction for a 2-h period, after which animals were re-tested.
Data analysis was performed using a combination of custom written scripts in MATLAB and toolboxes developed for neural analysis.
We compared changes in task performance between and across sessions. Specifically, we compared the performance change between early and late trials by comparing changes in behavior between the first 20 and last 20 trials in block 1, and the effects of sleep by comparing the last 20 trials in block 1 with the first 20 trials in block 2. Three different aspects of learning were measured: accuracy (here defined as successful retrieval of the pellet into the chamber), speed (defined as time from the beginning of the reach to the pellet, and through to the execution of retract movement), and finally similarity of movements, assessed by calculating the Pearson correlations between movement trajectories in X-Y space. For this analysis, within each block (i.e., Reach1early, Reach1late, and Reach2early, we correlated movements in both the X-direction and Y-directions and averaged this together to get a trial × trial correlation matrix. We then calculated the mean correlation across all trials within the different blocks. Across all animals, statistical changes in accuracy were assessed by assigning, for each trial, a “1” for correct trials in which animals successfully retrieved the pellet and a “0” for incorrect trials, followed by logistic regression analysis of the overall distribution of 1’s and 0’s across groups. Changes in speed (measured as changes in time for the overall reach trajectory) were assessed using ANOVA/post-hoc Fisher’s test; and changes in Pearson correlations were calculated using ANOVA/post-hoc Fisher’s test after first Z-transforming correlation coefficients.
In addition to the above analyses, we also conducted trajectory analyses in “state-space” using two different procedures. First, trajectories were performed without any processing, termed here an “external frame of reference.” For this analysis, we analyzed the distribution of Y-coordinates relative to X-coordinates, as a way of determining overall differences in the state-space of the trajectory. Statistical analysis of these distributions across the different groups of trials for each X-coordinate (binned into units of 4; Reach1Early Reach1Late and Reach2) was performed using bootstrap techniques (confidence intervals were estimated by performing random sampling with replacement 2,000 times for each X-coordinate being estimated). We also performed an analysis in which we specifically looked at displacement over time by referencing each trajectory to its initial starting point. In this way, we measured changes in movement relative to an “internal” frame of reference.
Identification of NREM sleep epochs was performed by visual assessment of LFP during spontaneous recordings in 10-s increments (S6 Fig). During any period denoting sleep, if there was a sustained reduction >2 s in the amplitude of the slow-wave activity below threshold during a continuous epoch we excluded these segments. Neural activity from sleep epochs during spontaneous recordings was concatenated together, in order to analyze ensemble activations specifically during NREM sleep. All sleep-related analyses were constrained to the minimum amount of sleep achieved in either sleep epoch, in order to ensure that analyses were not biased by different amounts of sleep post-learning vs pre-learning.
Spindle detection was used using automated algorithms to detect such oscillations, adapting methods previously described [76]. Spindles were then detected using a threshold of 2.5 SD of the signal, with start and finish times calculated as the time points 1.5 SD of the signal. Events were identified as spindles only if they were longer than 400 ms and shorter than 3 s. Automatically detected spindles/delta-oscillations were visually inspected to ensure the algorithm was correctly detecting these events. We also tested more stringent criteria (3 SD of the signal for example); results reported here (an increase in spindle oscillations after learning) do not depend on the specific parameters chosen.
We performed either non-parametric tests (Wilcox signrank/ranksum) or one–way ANOVA with post-hoc Fisher’s for most comparisons, as noted in the text. Logistic regression was used to identify changes in binary measures of success rate during learning or after sleep.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.