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.pbio.0060125 | A Regulatory Code for Neuron-Specific Odor Receptor Expression | Olfactory receptor neurons (ORNs) must select—from a large repertoire—which odor receptors to express. In Drosophila, most ORNs express one of 60 Or genes, and most Or genes are expressed in a single ORN class in a process that produces a stereotyped receptor-to-neuron map. The construction of this map poses a problem of receptor gene regulation that is remarkable in its dimension and about which little is known. By using a phylogenetic approach and the genome sequences of 12 Drosophila species, we systematically identified regulatory elements that are evolutionarily conserved and specific for individual Or genes of the maxillary palp. Genetic analysis of these elements supports a model in which each receptor gene contains a zip code, consisting of elements that act positively to promote expression in a subset of ORN classes, and elements that restrict expression to a single ORN class. We identified a transcription factor, Scalloped, that mediates repression. Some elements are used in other chemosensory organs, and some are conserved upstream of axon-guidance genes. Surprisingly, the odor response spectra and organization of maxillary palp ORNs have been extremely well-conserved for tens of millions of years, even though the amino acid sequences of the receptors are not highly conserved. These results, taken together, define the logic by which individual ORNs in the maxillary palp select which odor receptors to express.
| Odors are detected by olfactory receptor neurons (ORNs). Which odor an individual neuron detects is dictated by the odor receptors it expresses. Odor receptors are encoded by large families of genes, and an individual neuron must thus select the gene it expresses from among many possibilities. The mechanism underlying this choice is largely unknown. We have examined the problem of receptor gene choice in the fruit fly Drosophila, whose maxillary palp contains six functional classes of ORNs, each expressing different odor receptor genes. By comparing the DNA sequences flanking these genes in 12 different species of Drosophila, we have identified regulatory elements that are evolutionarily conserved and specific to each odor receptor. Genetic analysis of these elements showed that some act positively to dictate expression in a subset of ORNs, while others act negatively to restrict the expression of a receptor gene to a particular ORN class. We identified a transcription factor, Scalloped, that mediates repression. We were surprised to find that the odor response spectra of these neurons have been well-conserved for tens of millions of years, even though the amino acid sequences of their receptors have diverged considerably.
| Odor discrimination is based on the differential activities of olfactory receptor neurons (ORNs), which in turn depend on the odor receptors that the ORNs express [1,2]. This raises an intriguing problem: how do individual ORNs select, from among a large repertoire, which receptor genes to express?
Two models—a deterministic model and a stochastic model—are often proposed to explain the problem of receptor gene choice [3]. In the deterministic model, different receptor genes contain different combinations of cis-acting elements, and an individual gene is selected in those ORNs with corresponding transcription factors. In the stochastic model, individual receptor genes are selected by an unknown, singular entity or process that can act on only one gene at a time.
In mammals, the expression of an individual receptor is restricted to a particular zone of the olfactory epithelium, but within a zone, the choice of one receptor by a neuron is widely believed to be accomplished via a stochastic mechanism, followed by negative-feedback inhibition [4–6]. Only a single allele of an OR gene is expressed in an ORN [7], a property that has recently been found to be widespread among 4,000 autosomal genes surveyed in the human genome [8]. A 2.1-kb region called the H element, defined by its high homology between human and mouse, was shown to be required for normal expression of several OR genes adjacent to it [4]. Further analysis of the H element suggested an elegant model in which it also acts as a trans-acting enhancer element that allows stochastic activation of a single OR gene in each neuron [5]; however, recent data have favored a model in which the primary function of the H region is to act in cis, as one of many cis-regulatory elements required for OR expression in the mouse [4,9,10]. These results focus attention on the question of how cis-regulation might underlie the strikingly sophisticated problem of receptor gene choice.
The fruit fly Drosophila melanogaster contains two olfactory organs, the antenna and the maxillary palp, each covered with olfactory sensilla (Figure 1A). Each sensillum contains ORNs, usually two, combined according to a strict pairing rule. In the antenna, each ORN class is restricted to a zone of the antennal surface, with zones showing varying degrees of overlap (Figure 1A and [11]). In the maxillary palp, physiological data showed that different types of sensilla, and by extension, different classes of ORNs, appear to be largely if not completely coextensive, as if the maxillary palp constituted a single zone [12].
There are 60 Odor receptor (Or) genes, most of which are expressed in either the antenna or the maxillary palp [13–16]. Each receptor is expressed in ORNs of a single functional class; ∼37 ORN classes have been defined [11,12,17–19]. Most ORN classes express a single receptor [19–22].
In an earlier study, we identified two regulatory elements that are required for organ-specific expression of receptor genes [23]. Within an organ, we found no evidence for a negative-feedback mechanism. However, we identified a cis- regulatory element required for receptor expression in one ORN class. These findings suggested the possibility that neuron-specific odor receptor choice in Drosophila may depend on a sophisticated combinatorial code of cis-regulatory elements, as opposed to a stochastic mechanism followed by a negative feedback mechanism. The results thus laid a foundation for a systematic investigation of the most challenging aspect of the problem: how different receptors are expressed in different ORNs of an individual organ. The maxillary palp was chosen because it offers the virtue of numerical simplicity. It contains ∼120 ORNs, which are housed in three types of sensillum: pb1, pb2, and pb3. Each sensillum contains two ORNs: pb1 contains pb1A and pb1B; pb2 contains pb2A and pb2B; pb3 contains pb3A and pb3B. The odor response profile of each ORN has been defined and a receptor-to-neuron map has been established [12,21]. Seven Or genes are expressed in the maxillary palp, with two genes coexpressed in the pb2A neuron.
We systematically identified novel regulatory elements that dictate the proper expression of the maxillary palp Or genes in the correct ORNs, that is, elements that underlie the receptor-to-neuron map. These elements were identified by using a phylogenetic approach, much as the H element was identified through a comparison of two species. We compared the regulatory regions of orthologs from two Drosophila species whose genomes have been sequenced, and we identified elements that are evolutionarily conserved and that are specific to individual maxillary palp Or genes. Analysis of these elements across all 12 sequenced Drosophila genomes identified six that are conserved particularly highly. Functional analysis of these six elements reveals that some act positively to express individual Or genes in a subset of ORNs, and some act negatively to restrict the expression of individual Or genes to a single ORN class. Repression can be mediated via upstream or downstream regions, and in one case is mediated by the transcription factor Scalloped. Some elements are also used in other chemosensory organs, and some are conserved upstream of genes required for ORN axon targeting, sorting, and guidance.
Taken together, the data support a model in which the receptor-to-neuron map is constructed via a system of molecular zip codes. Or genes contain three classes of regulatory elements: elements that specify expression in the correct organ, positive elements that activate Or genes in a subset of ORN classes within an organ, and negative elements that restrict expression to a unique ORN class within that organ. We propose that the concerted action of these three classes of elements thus solves a formidable biological regulatory problem. We carried out a functional analysis of the D. pseudoobscura maxillary palp. Surprisingly, we found a remarkable degree of conservation in the response spectra of the ORNs over tens of millions of years of evolution. The receptor-to-neuron map is also conserved.
We examined the spatial organization of ORN classes in the maxillary palp. First, an anti-Elav antibody was used to illustrate the distribution of the entire population of ORN nuclei of the maxillary palp (Figure 1B). Second, we carried out a multiple-label experiment to differentially mark ORNs of the three types of sensilla: ORNs of the pb1A class were labeled in green, pb2B in yellow, and pb3A in red. The three classes of ORNs show extensive spatial overlap (Figure 1C). These results are consistent with the intermingling of sensillum types that are observed when recordings are taken from sensillar shafts [12]. The spatial overlap of ORN nuclei indicates that the identity of an ORN and, by extension, its choice of a receptor gene, are not dictated solely by its spatial position in a field.
We previously compared the upstream regions of the two Or genes coexpressed in pb2A to identify regulatory sequences shared by these two genes, but not by any other maxillary palp Or gene [23]. To identify upstream regulatory elements for the other five maxillary palp Or genes, we used a different strategy based on phylogenetic analysis.
D. melanogaster and D. pseudoobscura diverged tens of millions of years ago [24] and contain orthologous receptor genes. We examined the upstream regions of orthologous Or genes for conserved elements shared by the members of each orthologous pair, but not by any of the other maxillary palp Or genes. Accordingly, we identified all conserved upstream sequences greater than 6 base pairs (bp) in length for each pair of orthologs using DOT-PLOT analysis (Figure S1A), and from these conserved elements we selected those that were specific to each gene. The analysis was focused on the 500 bp that are upstream of the translational start site, because in a previous study, this extent of DNA was sufficient to confer faithful expression to a GAL4 reporter gene in the case of each of two maxillary palp Or genes analyzed in detail [23]. One pair of orthologs, Or85d and its D. pseudoobscura counterpart, was exceptionally well-conserved in the 500-bp upstream region, showing 80% identity. To identify discrete conserved elements within the region upstream of Or85d, we expanded our analysis to include a more divergent species, D. virilis.
Conserved, gene-specific elements were identified for each of the five Or genes analyzed (Figure 1D). The number of such elements varies: Or59c contains one, whereas Or42a contains six. In the special case of Or85d, two elements are shared by D. virilis and D. melanogaster upstream of Or85d, but are not found upstream of any other maxillary palp Or gene.
To identify the best candidate for a regulatory element for each of these receptor genes, we used a powerful bioinformatic approach that takes advantage of the recent sequencing of the genomes of ten other Drosophila species: D. simulans, D. sechellia, D. yakuba, D. erecta, D. ananassae, D. persimilis, D. willistoni, D. virilis, D. mojavensis, and D. grimshawi. The upstream regulatory regions of the orthologous receptor genes from all 12 species were aligned (Figure S1B) using the genome browser at the University of California Santa Cruz, and each of the elements was mapped onto the alignment. Using this approach, we were able to identify the gene-specific element with the highest sequence conservation for each of the receptor genes (Figures 1E and Figure S1); in the case of Or42a, two elements were nearly identical in their extent of conservation, and we have analyzed both.
To determine whether the evolutionarily conserved, gene-specific elements have a regulatory function, we tested them in vivo using two complementary approaches, one based on a loss of function and one on a gain of function. For each gene, we analyzed the element with the highest sequence conservation. We did not analyze Or85d elements because we lacked a faithful Or85d-GAL4 driver.
Or46a is expressed in the pb2B neuron, and its upstream region contains two conserved, gene-specific elements (Figures 1D and 1E and Figure S1). One of these elements, 46a1, is more highly conserved. It is 10 bp long, its sequence shows 93% identity across the 12 species, and its position is conserved. A 1.9-kb region of DNA upstream of Or46a drives faithful expression of a GAL4 reporter in pb2B (Figure 2A and [21]). However, when the 46a1 element is mutated, the 1.9-kb region no longer drives expression (Figure 2B). In most cases, no cells are labeled; in rare cases, a single ORN is labeled (n = 0.52 ± 0.24 cells/maxillary palp; n = 8 independent lines examined; n > 10 maxillary palps examined per line). The simplest interpretation of these results is that the 46a1 element is necessary for Or46a expression in pb2B.
We then asked whether the 46a1 element can drive expression in the context of a minimal promoter. We placed four copies of 46a1 upstream of a TATA box and found that this small construct can in fact drive expression in maxillary palp cells (Figure 2C). Many, if not all, of the cells could be identified as ORNs, because they contain dendrites and axons; their identity is considered further below. Expression from this artificial promoter could also be detected in a small subset of neurons in the main gustatory organ, the labellum (unpublished data).
Or71a is expressed in pb1B. Its upstream region contains multiple gene-specific elements, of which the longest and best conserved is 71a3, consisting of 16 bp and showing 97% sequence identity. This element was tested in the context of the Or71a 5′ + 3′ construct, which contains sequences both upstream and downstream of Or71a [21]. This construct drives faithful expression of GAL4 when the 71a3 element is intact (Figure 2D and [21]), but not when it is mutated (Figure 2E; n = 0.25 ± 0.1 cells/maxillary palp; n = 8 independent lines examined; n > 20 maxillary palps examined per line). When multiple copies of 71a3 were placed upstream of a TATA box, the construct drove GAL4 expression in maxillary palp cells that can be identified as ORNs by virtue of their dendrites and axons (Figure 2F). Low levels of expression could also be detected in a small subset of cells in the labellum (unpublished data).
Or59c is expressed in pb3A, and its upstream region contains a single gene-specific conserved element, 59c1, which is 11 bp long and shows 97% sequence identity across nine species (Figure 2G); the region containing the 59c1 sequences could not be identified in three of the most distantly related species, D. virilis, D. mojavensis and D. grimshawi. We have tested its function by placing multiple copies upstream of a TATA box and found that this minimal promoter drove robust expression of GAL4 in the maxillary palp (Figure 2H). Expression was not detected in the labellum.
Earlier studies have shown that the expression of a subset of the maxillary palp Or genes requires the POU domain transcription factor Acj6 [25], which is expressed in all ORNs of the maxillary palp [46]. Acj6 also controls axon targeting specificity of a subset of maxillary palp ORNs . The 46a1, 71a3, and 59c1 elements do not contain predicted Acj6 binding sites (Bai L, Carlson JR, unpublished results), and the transcription factors that act on these sequences are unknown. To test whether the factors that act on these neuron-specific elements are dependent on acj6, we examined the expression of the minimal promoter constructs in an acj66 background.
In the acj66 mutant, although the expression of the Or46a-GAL4 driver is lost, which is consistent with the loss of Or46a mRNA observed previously [13], the expression of the 46a1 minimal promoter construct is still strong (Figure 2I and 2J). These results suggest that the factors that direct expression from the 46a1 motif are independent of acj6 for their expression and function (Figure 2K). An alternative possibility is that another transcription factor can compensate for the loss of acj6.
Expression of the Or71a-GAL4 driver can be detected in acj6, and the expression of the 71a3 minimal promoter construct can also be detected (Figure 2L and 2M). These results suggest that the factors binding to 71a3 do not require acj6 for their expression or function (Figure 2N).
In the case of Or59c, we find that acj6 is required both for expression of the gene and for the minimal promoter (Figure 2O and 2P). These results suggest that acj6 is required directly or indirectly for the expression of the 59c1 binding factor or for its function at the 59c1 site (Figure 2Q).
Or42a is expressed in pb1A, and 4.1 kb of upstream DNA drives faithful expression of GAL4 in maxillary palp ORNs [21]. Two elements are nearly identical in their high conservation: 42a4 (98%) and 42a6 (98%), and we tested the function of both elements in vivo. 42a6 maps only three bp from 42a5 (Figure 1D). We constructed a small deletion that eliminates both 42a6 and 42a5 elements, and we found no effect on Or42a-GAL4 expression (unpublished data).
The longer of the two most highly conserved elements at Or42a, 42a4, contains an inverted repeat: AGTGTAAAAGTTTACACTT. We were surprised to find that mutation of this element led to a 2-fold increase in the number of labeled maxillary palp cells, from 18.2 ± 1.8 (n = 9 maxillary palps) to 33.2 ± 3.7 (n = 9 maxillary palps quantified from two independent lines; n = 8 independent lines examined, n > 20 maxillary palps examined/line) (Figure 3A–3C). The simplest interpretation of this result is that 42a4 is a negative regulatory element that represses Or42a in a subset of ORNs. To test this interpretation, we first carried out a double-label experiment using probes for the endogenous Or42a mRNA and for the green fluorescent protein (GFP) that is driven by the mutant promoter via GAL4. We found that all Or42a+ cells express GFP, but that GFP is also expressed in an additional subset of cells (Figure 3D).
To identify the cells that ectopically express GFP, we undertook a series of additional double-label experiments. We found that the GFP+ cells do not express Or59c mRNA, indicating that they are not pb3A neurons (Figure 3E; 0% of the GFP+ neurons are Or59c+; n = 8 maxillary palps); nor are they paired with cells that express Or59c mRNA, indicating that they are not pb3B neurons. In another experiment, GFP+ cells did not label with an Or33c probe (only 3% of the GFP+ neurons appear Or33c+; n = 8 maxillary palps), indicating that they are not pb2A neurons; however, GFP+ cells were often found paired with Or33c+ cells (arrowheads), indicating that many GFP+ cells are pb2B neurons (Figure 3F). The identity of these GFP+ cells as pb2B neurons was confirmed directly in another double-label experiment using a probe for Or46a mRNA (Figure 3G; 94% of the cells labeled with Or46a mRNA were GFP+; this value is the mean of values determined from n = 8 maxillary palps).
The simplest interpretation of these results is that positive regulatory elements in the Or42a upstream region are capable of driving expression not only in the pb1A neuron but also in the pb2B neuron. The 42a4 element represses expression in pb2B neurons, thereby restricting expression to a single ORN class, pb1A.
The ectopic expression of an Or42a promoter in Or46a+ neurons suggested a relationship between these two genes. Further evidence for a relationship came from analysis of the minimal promoter containing multiple copies of 46a1 (Figure 2C). This promoter drove GFP expression in more ORNs than could be accounted for by Or46a+ neurons alone. A double-label experiment showed that while most of the GFP+ cells are in fact Or46a+, some are Or42a+ (Figure 3H).
The reciprocal relationship between Or42a and Or46a misexpression suggests that Or42a may contain an unidentified positive regulatory element, 42ax, that is similar in sequence to 46a1, with both sites able to bind a transcription factor present in both pb1A and pb2B. To test this interpretation, we examined the 500 bp upstream region of Or42a for an element similar, but not identical, to 46a1 (GACATTTTAA). We identified a sequence, TATATTTTAA, identical to 46a1 at the 8 underlined positions, at −455 bp. Moreover, these two sequences share an ATTTTA core, which has been shown to function as a binding site for basic helix-loop-helix transcription factors at other loci. TATATTTTAA is not found upstream of any other maxillary palp Or genes. This 42ax sequence is conserved in sequence (80% identity) and location in seven of the 12 Drosophila species. It will be interesting to identify the transcription factor that binds 46a1 and then test directly its binding to 42ax.
When DNA upstream of Or59c was fused to GAL4, expression of the reporter GFP was not faithful (Figure 4A; n = 5 independent lines); the same result was obtained when upstream regions of varying lengths were used (either 2.1 kb, which extends to the next upstream gene, or 5.2 kb, which includes upstream coding sequences). Double-label experiments using an Or59c probe revealed misexpression in many Or59c– cells; moreover, many Or59c+ cells did not express GFP. Some of the misexpressing cells are the neighboring pb3B neurons, which can be seen to be paired with Or59c+ pb3A cells (arrowheads in Figure 4A; 75% of the cells neighboring the Or59c+ cells were GFP+, n = 9 palps). To identify the other ORNs that ectopically express the Or59c-GAL4 construct, we carried out double-label experiments with other Or genes. Misexpression was also observed in pb1A cells, which express Or42a (96% of the Or42a+ cells misexpressed GFP, n = 9 palps), but not in the pb1B cells (Figure 4B), nor in the pb2A or B cells (Figure 4C). In summary, misexpression is specific to pb1A and pb3B.
Because neither of the varying lengths of upstream DNA sequences were sufficient to restrict GAL4 expression to the Or59c+ cells, we added 3′ sequences to the construct. Initially, 500 bp of DNA taken directly from the region immediately downstream from the Or59c stop codon was added downstream of the GAL4 coding region. Between the downstream sequences of Or59c and the GAL4 coding region was the Hsp70 3′ untranslated region (UTR), which is present in the GAL4 vector and which is often present in promoter-GAL4 analysis.
This Or59c 5′ + 3′ construct showed much less misexpression in Or59c− cells (Figure 4D). The total number of GFP+ cells declined from 49.7 ± 1.3 to 27.3 ± 2.1 (SEM; n = 10 in each case). However, some misexpression remained, and only 62% of the Or59c+ neurons were GFP+. We then removed the Hsp70 3′ UTR sequences, such that the Or59c downstream sequences were in close proximity to the 3′ end of the GAL4 coding region and the Or59c 3′ UTR is used. This construct drove faithful expression (Figure 4E; n = 8 independent lines examined). Thus, there is a negative regulatory element downstream of Or59c that restricts expression of this gene to pb3A neurons, and either there is a requirement that the native 3′ UTR be used, or else there is a regulatory factor that acts on this element in a context-dependent fashion in order to achieve this negative regulation. We note with interest that the inclusion of the downstream sequences, without the Hsp70 sequences, also drove expression in Or59c+ neurons that had previously failed to express the reporter, suggesting that the downstream sequences are required for positive as well as negative regulation of Or59c.
Inspection of the sequences downstream of Or59c that repressed misexpression revealed a binding site for the transcription factor Scalloped (Sd), AAATATTT [26] (Figure 5A). This site is well-conserved among a number of other species (Figure S2A). Sd has been shown to be expressed in olfactory organs [27]. To confirm and extend the description of sd expression we used an enhancer trap line, sdETX4 [27], and confirmed that sd is expressed in a subset of cells in the maxillary palp (Figure 5B and 5C).
To test whether sd represses Or59c, we carried out in situ hybridizations to the maxillary palp of a hypomorphic sd mutant, sd1 (Figure 5D). We found a 40% increase in the number of Or59c+ neurons (Figure 5E). By contrast, there was no increase in the number of Or42a+ neurons (Figure 5D and 5E). There was, however, an increase in the number of Or85d+ cells, and we note with interest that there is another type of Sd binding site, TAAAATTA [26], 737 bp downstream from the stop codon of Or85d.
The Or59c-GAL4 construct that contains only upstream sequences, Or59c 5′, misexpresses in two ORN classes, the neighboring pb3B cell (Or85d+) and pb1A (Or42a+), as shown above in Figure 4. We asked whether sd is expressed in these two ORN classes. Using an Or59c probe, which labels the pb3A cell, we found that sd is in fact expressed in neighboring cells (Figure 5F), but not in pb1A cells, which express Or42a (Figure 5G). These results suggest that Sd may repress the Or59c gene in pb3B. If so, we would expect that in an sd mutant, we would observe cells that coexpress Or59c and Or85d. We tested this possibility by carrying out double-label in situ hybridizations in two different hypomorphic alleles of sd, sd1, and sdSG29.1 [28]. In both alleles, we found Or59c+ Or85d+ cells (Figure 5H), but not Or59c+ Or42a+ cells (unpublished data). Thus repression of Or59c in the neighboring pb3B cell requires both a Sd binding site and Sd.
Since Sd represses Or59c in pb3B, why doesn't Sd also repress Or85d in pb3B, given that both Or genes have Sd binding sites? The simplest explanation is that the two Sd binding sites are distinct. There are several potential interacting partners with which Sd may interact to form a functional transcription factor [26,29], and the pb3B cell may contain a partner necessary for repression at the Or59c binding site but not a partner necessary for repression at the Or85d binding site. If a faithful Or85d-GAL4 construct becomes available, it will be interesting to replace the Or85d-type Sd binding site with the Or59c-type Sd binding site, to determine whether the Or59c-type site confers repression in the pb3B cell.
We note that Or85d-GAL4 constructs containing only the 5′ regions of Or85d, which lack the Sd binding site, drive misexpression in a number of non-neuronal cells of the maxillary palp (Figure S2B). Most of the labeled cells lack dendrites and axons, and when labeled with a membrane-bound GFP, as opposed to with RNA probes that label the cell bodies, these cells appear larger than ORNs. These results suggest that Sd may interact with a binding partner in non-neuronal cells to repress Or85d expression in these cells.
Or42a is expressed in the larval olfactory system as well as in the maxillary palp [21,30]. The Or42a-GAL4 construct shows expression in one ORN in each of the bilaterally symmetric larval olfactory organs, the dorsal organs (Figure 6A). We also observed expression in two neurons of the labellum, the taste organ on the adult head (Figure 6A). To determine whether the conserved elements identified in our analysis of maxillary palp receptor choice can act in these other chemosensory organs, we examined Or42a-GAL4 constructs in which these elements were mutated. A mutation that affects both 42a6 and 42a5, which did not affect expression in the maxillary palp, had no effect on expression in these other organs. However, mutation of 42a4, which relieved repression of Or42a in other maxillary palp ORNs, also relieved repression of Or42a-GAL4 in the larval olfactory organs and the labellum (Figure 6B): in both cases supernumerary neurons were labeled. In the labellum, ∼8–10 pairs of neurons were labeled. These results suggest that the molecular mechanisms underlying receptor gene choice in the maxillary palp overlap with those specifying receptor expression in other chemosensory organs.
In this study we have identified and functionally characterized a number of regulatory elements that operate in directing the formation of the receptor-to-neuron map of D. melanogaster. Because the newly defined elements we have analyzed here are conserved in sequence and position among Drosophila species, we predicted that the programmed regulation leading to the formation of receptor-to-neuron maps would be conserved as well. To test this prediction, we carried out a physiological analysis of the D. pseudoobscura maxillary palp. Although each of the seven Or genes expressed in the maxillary palp has an ortholog expressed in the D. pseudoobscura maxillary palp ( as described [21] and unpublished data), we expected that their odor response profiles would have diverged a great deal over the course of tens of millions of years. We did not know a priori whether we would be able to correlate D. pseudoobscura ORNs with D. melanogaster counterparts.
We were surprised to find that the profiles of the maxillary palp ORNs are remarkably well conserved between these two species (Figure 7). Despite the tens of millions of years of separation, each ORN class in D. melanogaster has a counterpart in D. pseudoobscura, and their responses to a panel of ten diverse odorants are strikingly similar. Not only are the magnitudes of the responses well conserved, but the modes of the responses, i.e., excitation versus inhibition, are conserved. For example, both the pb2B ORN of D. melanogaster and its D. pseudoobscura counterpart are excited by 4-methyl phenol and inhibited by 3-octanol. The orthologous receptors show amino acid identity as low as 59% in the case of Or71a (Figure S3), and in no case exceeded 84%, the identity determined for Or42a. Thus pb1B in D. melanogaster, which expresses Or71a, shows the same specificity for 4-methyl phenol and 4-propyl phenol as the corresponding ORN in D. pseudoobscura, although Or71a is only 59% identical between the two species.
The conservation of odor response spectra allows us to determine that the stereotyped pairing of ORNs is also conserved in the two species. These results suggest that not only are the response spectra of the odor receptors conserved with respect to a diverse panel of odorants, but that the program of receptor gene expression is also conserved between these distantly related species.
Given the success in identifying gene-specific elements required for the expression of individual Or genes in individual classes of ORNs, we asked whether the same approach could be used to identify sensillum-specific elements required uniquely by the Or genes that are expressed in the neighboring ORNs of a common sensillum. We searched for sensillum-specific elements conserved in the upstream regions of D. melanogaster and D. pseudoobscura Or genes. Only one element, AAATCAATTA, was found upstream of all orthologs expressed in a particular sensillum type (Figure S4A and [23]). Mutational analysis of this element in the Or42a promoter did not, however, appear to affect expression (Figures S4B–S4E). Furthermore, expression was not affected by mutation of the more proximal of the two copies of this element in the Or71a upstream region (unpublished data). These results suggest that this element is not required for expression in the pb1 sensillum.
We have analyzed the problem of how individual ORNs select which receptor genes to express, a fundamental problem that underlies all odor coding. In Drosophila, the foundation of olfactory perception is a stereotyped receptor-to-neuron map. The developmental process by which this map is constructed has been examined here using an analysis of evolutionary conservation as a point of departure.
We identified conserved, gene-specific elements flanking five maxillary palp receptor genes. Functional analysis of the six most highly conserved elements confirmed that elements upstream of four of these genes act either positively or negatively in gene regulation, thereby validating the experimental approach. Two elements did not appear to be required for normal gene regulation; however, it is possible that they act in a redundant fashion or that they mediate a response to such epigenetic factors as feeding status, mating status, or circadian rhythm, which we did not examine.
The elements varied in length from 7 to 19 bp; some of the longer ones could be composite sites that bind more than one factor. Several of the sites contain AT-rich cores, reminiscent of binding sites for certain classes of transcription factors including POU domain proteins. One element, 42a4, contains two iterations of an octamer, in an inverted repeat. Two elements, 46a1 and 71a3, overlap with a Dyad-1 element, CTA(N)9TAA, a positive regulatory element that is required for normal maxillary palp expression and that is found upstream of all of these maxillary palp Or genes [23]. The close juxtaposition of regulatory elements suggests an interaction among the regulatory proteins that they bind.
Our strategy for identifying these elements required that each be specific to a single maxillary palp Or gene. The identification of these elements reveals that each gene contains at least one unique element that is not shared with any other maxillary palp Or genes. This need not have been the case: the system could alternatively have been composed entirely of nonunique regulatory elements, each shared by multiple genes, but in unique combinations. In any case, in the maxillary palp the combinatorial code of cis-acting elements appears to include both unique and shared elements (e.g., Dyad-1).
The regulatory elements and the logic by which they operate are summarized in Figure 8. Positive regulatory elements direct expression in subsets of maxillary palp ORNs. Negative regulatory elements restrict this expression to a single ORN class. Overall, the correct expression pattern is determined by the interplay of positive and negative elements.
The negative regulation we have observed is highly specific. When the 42a4 element was ablated, Or42a misexpression was observed specifically in pb2B. One possible interpretation is that pb2B and pb1A, the cell that normally expresses Or42a, share a positively acting transcription factor that other ORNs lack. Thus the two ORNs with contexts that are permissive for Or42a expression are not neighboring ORNs that share a sensillum, but ORNs in different sensilla, with very different odor response profiles. Reciprocally, a positively acting element upstream of Or46a, which is expressed in pb2B, drives expression not only in pb2B but also in pb1A. This connection between pb1A and pb2B suggests a developmental relationship that remains to be defined in mechanistic terms.
This study has concentrated on receptor gene choice in the maxillary palp, on account of its numerical simplicity. Does a system of molecular zip codes also underlie the process of receptor gene choice across the entire odor receptor repertoire? In addition to the seven maxillary palp receptors, the Or gene family contains 53 other members expressed in the antenna or the larval olfactory system [19,30–32]. Using a comparative bioinformatic approach, we performed a large-scale analysis of sequence conservation in the 500 bp upstream of each of 42 Or genes across all 12 Drosophila species (Figure S5 and Text S1). We found great diversity in the number, lengths, and distribution of highly conserved upstream regions. Within the most highly conserved of these regions we identified a variety of elements that are shared among subsets of Or genes (Figure S6A and S6B). This analysis, then, reveals a combinatorial structure to the organization of shared elements upstream of these receptor genes. This pattern supports a model in which a combinatorial code of positive and negative regulatory elements dictates the proper expression of each Or gene.
What kind of proteins accomplish this regulation? In C. elegans, several kinds of transcription factors have been elegantly shown to play roles in specifying ORN identity and receptor expression [33]. In the mouse, a LIM-homeodomain protein, Lhx2, is required for normal ORN differentiation and expression of OR genes [34,35]. In Drosophila the POU domain protein Acj6 is required for the expression of a subset of Or genes [36]. We have also shown that Sd, a TEA domain-containing transcription factor, is critical in restricting the expression of some Or genes to their proper ORNs. Sd has been shown to act as a repressor in other systems and in fact is required for normal taste behavior in both larvae and adults [37]. Another aspect of receptor gene choice depends on proteins of the Notch pathway: receptor choice in neighboring ORNs of a sensillum appears to be coordinated via asymmetric segregation of regulatory factors from a common progenitor [23,38].
Some elements that are essential to odor receptor gene choice are also located upstream of genes required for axon guidance and sorting (Figure S7 and Text S1). The presence and positions of these elements have been conserved for tens of millions of years of evolution. The presence of Or regulatory elements upstream of ORN axon-guidance genes could reflect a relationship between receptor gene choice and axon targeting. In addition to selecting particular Or genes for expression, ORNs send axons to particular glomeruli in the antennal lobe of the brain. ORNs that express the same Or gene send axons to the same glomerulus [16,19]. Thus the olfactory system contains both a stereotyped receptor-to-neuron map and a stereotyped connectivity map in the antennal lobes. The tight coordination between receptor gene choice and axonal projection could in principle arise in part from overlap in the mechanisms underlying these processes. In mammals, odor receptors play a role in ORN targeting [39–41]. In Drosophila, ORN targeting does not require the receptors [20], but could require the regulatory apparatus used to express the receptors. Acj6 provides an example of a link between the two processes: it acts both in receptor expression and ORN axon targeting (Figure 2) [13,25]. Moreover, we have found that Acj6 is required for the activity of one of the regulatory elements identified here.
We found a remarkable similarity of function between the maxillary palp ORNs of two species that diverged more than tens of millions of years ago. We had expected that over this time interval, the odor specificities of the ORNs would have diverged markedly to serve differing needs of the two evolving species. Instead, every ORN class showed strikingly similar responses, with few exceptions. The results show that two odor receptors can differ a great deal in amino acid sequence and still exhibit a very similar odor specificity.
The organization of the organ in the two species is also identical, in that corresponding ORNs are combined according to the same pairing rules. This high degree of conservation suggests a critical role for the maxillary palp in odor coding and in the generation of olfactory-driven behavior. The conservation of regulatory elements and organization also suggests that the two species use common mechanisms to specify the receptor-to-neuron map.
The regulatory challenge confronted by the Drosophila olfactory system represents an extreme among problems of gene regulation. It requires the storage and deployment of a great deal of information. Our data support a model in which Or gene expression is controlled by a system of molecular zip codes. Each Or gene contains elements that dictate expression in the proper olfactory organ [23], positive regulatory elements that specify expression in a subset of ORN classes, and negative regulatory elements that restrict expression to a single ORN class. This logic and the components that execute it have solved such a challenging problem with such efficiency that they have apparently been well conserved for tens of millions of years.
Drosophila stocks were raised at 25 °C. Wild-type flies were Canton-S unless otherwise indicated. sd1 and sdETX4, referred to here as sd{PlacZ}, were obtained from the Drosophila Stock Center (Bloomington, Indiana). sdSG29.1 was a gift from S. Cohen. D. pseudoobscura was from the Drosophila Species Resource Center (Tucson, Arizona). w; UAS-mCD8-GFP/CyO;UAS-mCD8-GFP was used as a source of GFP unless otherwise indicated. All DNA constructs were sequenced and then injected along with Δ2,3 transposase plasmid into w1118 flies. Multiple transgenic lines, in most cases eight, were generated and tested for each construct.
To identify gene-specific conserved sequences in the upstream maxillary palp Or genes, we used ClustalW alignments and DOT-PLOT analysis (MacVector ). To map identified cis-elements to sequences and identify overrepresented motifs, the DNA-PATTERN (STRINGS) and OLIGO-ANALYSIS programs were used at the RSA tools website (http://rsat.scmbb.ulb.ac.be/rsat/). For the identification of conserved sequences the Drosophila genome browser at http://genome.ucsc.edu/ was used. A multiple alignment was constructed using MULTIZ from the best-in-genome pairwise alignments generated by BLASTZ. Large-scale predictions of conserved elements were obtained from the multiple alignments using the PhastCons program with the most-conserved option. Shared elements were identified using the OLIGO-ANALYSIS program at the RSA tools website.
The wild-type Or42a 4.1-kb promoter-GAL4 construct has been described previously [21]. In (42a4)-GAL4, the 42a4 element, which contains an inverted repeat (AGTGTAAANNTTTACACT), was mutated to (AGTG–––TTTGGATCC), resulting in a deletion within the first half-element and the substitution of a BamHI recognition sequence in the second half-element (italicized). This was accomplished by PCR amplification of two promoter fragments, one terminating immediately upstream of the TAAA in the first octamer of the 42a4 element, and the second wasa fragment extending from immediately downstream of this element to the start codon of Or42a. Primers for these PCR reactions contained the BamHI restriction site in place of ACACTT. The PCR products were AT-cloned into pGEM-T Easy. Subsequent ligation of the two PCR fragments resulted in the desired replacement in the context of the Or42a 4.1-kb promoter-transgene.
In the (42a5+6)-GAL4 construct a small deletion was designed to delete both the 42a5 and the 42a6 elements, which are separated by 3 bp (TGTGAACGATTGCAGCCTG). This was achieved by using a similar approach as for 42a(4)-GAL4, but in this case the two primers, containing BamHI sites at their ends, were designed to start immediately upstream of 42a6 and immediately downstream of 42a5. Ligation of the appropriate fragments led to the replacement of the entire 19-bp region, comprising the two elements, by a BamHI site.
In the (46a1)-GAL4 construct the 46a1 element (GACATTTTAA) was mutated by replacing the first six bases with a BamHI restriction site. This was achieved using a PCR cloning strategy similar to the ones described for the Or42a constructs.
In the (71a3)-GAL4 construct the 71a3 element (TGAATTTTAATTGAAA) was mutated to (GCTAGCTTAATTGAAA) by replacing the first six bases with a NheI restriction site using a PCR cloning strategy similar to the one described earlier, resulting in the desired mutation in the context of the Or71a 5′ + 3′-GAL4 construct.
We note that 46a1 and 71a3 each overlaps with a Dyad-1 motif, CTA(N)9TAA, a positive regulatory element that is required for expression of Or genes in the maxillary palp [23]; the mutations of 46a1 and 71a3 were designed so as not to affect the Dyad-1 motif.
The Or59c 2.1-kb promoter-GAL4 construct has been described in [21] and has been shown to express in a large number of non-endogenous cells in the palp. The Or59c 5′ + 3′-GAL4 was constructed by cloning a 0.5-kb fragment of DNA that lies immediately downstream of the Or59c stop codon into the SpeI/BamHI site that is positioned downstream of the GAL4-hsp70 3′ UTR in pG4PN. The 0.5-kb fragment was PCR-amplified from Canton-S genomic DNA with primers designed to add a Spe1 site to the 5′ end and a BamHI site to the 3′end.
The Or59c (5′ + 3′ direct)-GAL4 construct was made in several steps. First the 0.5-kb fragment of DNA immediately downstream of the Or59c stop codon, described above, was cloned into the BamHI/Spe1 site of pSK+ to generate plasmid pSK3'. Second, the GAL4 coding region was cloned as a HindIII fragment into pSK3' to yield pSKGAL4. The Or59c 5′ region was excised from the Or59c 2.1kb-GAL4 vector using KpnI/NotI(blunted) and it was KpnI/blunt cloned into the Kpn1/Apa1(blunted) site of the pSKGAL4 plasmid. Finally the KpnI/SpeI fragment from this plasmid was ligated with the KpnI/SpeI fragment of pG4PN to yield Or59c complex-GAL4.
Complementary pairs of oligonucleotides were designed such that upon annealing, they would yield a double-stranded DNA fragment that includes multiple copies of the corresponding conserved elements and overhangs on either side for EcoR1 restriction enzyme sites. These fragments were cloned directly into the EcoR1 site of the pPTGAL Drosophila transformation vector [42]. The 46a1 sequence was GACATTTTAAATGCCCTAATGACATTTTAAATGCCCTAATGACATTTTAAATGCCCTAATGACATTTTAA. The 71a3 sequence was CTAATTGAATTTTAATTGAAACGTCACTAATTGAATTTTAATTGAAACGTCACTAATTGAATTTTAATTGAAACGTCA. The 59c1 sequence was GCAAACTGTAATTAGAGGACCGCAAACTGTAATTAGAGGACCGCAAACTGTAATTAGAGGACCGCAAACTGTAATTAGAGGACC. We note that the constructs for 46a1 and 71a3 contain Dyad-1 motifs, but these motifs are not sufficient to drive expression in the maxillary palp [23]. The underlined sequences indicate the gene-specific elements, and the italicized sequences indicate the Dyad-1 sequences. For each minimal promoter construct, at least two independent lines were examined, and n > 20 maxillary palps were examined for each line.
In situ hybridization and immunohistochemical localization were performed as in [21]. Mouse anti-βGAL antibody (1:1000), and rabbit anti-GFP (1:250) were obtained from Promega.
To generate a high-resolution map of the nuclei of the three sensilla types (Figure 1C), (42a4)-GAL4/UAS-GFP; UAS-GFP/+ was used to label pb1A and pb2B in green, and Or46a and Or59c in situ hybridization probes were used to label with red the pb2B and pb3A cells, respectively. Thus pb1A was labeled green, pb2B was labeled yellow (red and green); pb3A was labeled red. Confocal Z-stacks consisting of nine optical sections of each palp were analyzed in Photoshop. Positions of the labeled nuclei were manually marked with the corresponding color at each optical plane, and the 9 stacks were compressed to generate a 2-D representation of all the labeled neurons.
Odors were delivered and action potentials were recorded as described previously [20] and in Text S1.
|
10.1371/journal.ppat.1006180 | The Malaria Parasite's Lactate Transporter PfFNT Is the Target of Antiplasmodial Compounds Identified in Whole Cell Phenotypic Screens | In this study the ‘Malaria Box’ chemical library comprising 400 compounds with antiplasmodial activity was screened for compounds that perturb the internal pH of the malaria parasite, Plasmodium falciparum. Fifteen compounds induced an acidification of the parasite cytosol. Two of these did so by inhibiting the parasite’s formate nitrite transporter (PfFNT), which mediates the H+-coupled efflux from the parasite of lactate generated by glycolysis. Both compounds were shown to inhibit lactate transport across the parasite plasma membrane, and the transport of lactate by PfFNT expressed in Xenopus laevis oocytes. PfFNT inhibition caused accumulation of lactate in parasitised erythrocytes, and swelling of both the parasite and parasitised erythrocyte. Long-term exposure of parasites to one of the inhibitors gave rise to resistant parasites with a mutant form of PfFNT that showed reduced inhibitor sensitivity. This study provides the first evidence that PfFNT is a druggable antimalarial target.
| The emergence and spread of Plasmodium falciparum strains resistant to leading antimalarial drugs has intensified the need to discover and develop drugs that kill the parasite via new mechanisms. Here we screened compounds that are known to inhibit P. falciparum growth for their effects on the pH inside the parasite. We identified fifteen compounds that decrease the pH inside the parasite, and determined the mechanism by which two of these, MMV007839 and MMV000972, disrupt pH and kill the parasite. The two compounds were found to inhibit the P. falciparum formate nitrite transporter (PfFNT), a transport protein that is located on the parasite surface and that serves to remove the waste product lactic acid from the parasite. The compounds inhibited both the H+-coupled transport of lactate across the parasite plasma membrane and the transport of lactate by PfFNT expressed in Xenopus oocytes. In addition to disrupting pH, PfFNT inhibition led to a build-up of lactate in the parasite-infected red blood cell and the swelling of both the parasite and the infected red blood cell. Exposing parasites to MMV007839 over a prolonged time period gave rise to resistant parasites with a mutant form of PfFNT that was less sensitive to the compound. This study validates PfFNT as a novel antimalarial drug target.
| The most virulent malaria parasite, Plasmodium falciparum, was responsible for the majority of the 214 million malaria cases and 438,000 malaria-attributable deaths estimated to have occurred in 2015 [1]. The parasite has developed resistance to most of the drugs deployed against it [2], and it is imperative that new drugs be developed to protect or replace existing therapies. Significant progress has been made in recent years in developing new antimalarials [3]; however, many of the compounds under development are structurally related to previous or current antimalarial drugs and there is an urgent need to identify novel lead-drug compounds that act on hitherto unexploited parasite targets and create improved options for resistance-deterring combination therapies. Such therapies should, ideally, contain at least two drugs that act on separate targets [4].
Recent large-scale whole cell phenotypic screens have uncovered tens of thousands of novel inhibitors of the in vitro growth of asexual P. falciparum parasites in human erythrocytes [5–7]. In a bid to further research into the novel antiplasmodial chemotypes, the Medicines for Malaria Venture (MMV) compiled the open access ‘Malaria Box’, a collection of 400 structurally-diverse compounds selected from the drug screen hits for which the mechanisms of action were not known [8]. Determining the mechanisms of action of these antiplasmodial compounds has the potential to uncover aspects of P. falciparum biology that can be exploited as drug targets.
The mechanism by which P. falciparum generates ATP presents potential vulnerabilities. In particular, the (disease-causing) asexual intraerythrocytic stages of P. falciparum rely primarily on glycolysis for energy metabolism [9]. Human erythrocytes infected with trophozoite-stage parasites devour glucose up to 100 times faster than uninfected erythrocytes [10], and generate large quantities of lactate, which is exported via a lactate:H+ symport mechanism [11]. Both glucose uptake and lactate export are likely to be essential for maintaining the energy requirements, intracellular pH and osmotic stability of these parasite stages. The P. falciparum hexose transporter (PfHT) mediates the uptake of glucose into the parasite and has been of interest as a drug target for some time [12, 13]. Recently, a compound from the Malaria Box was found to inhibit PfHT potently (with a 50% inhibitory concentration (IC50) of ~ 50 nM) while also displaying a high degree of selectivity for PfHT over the human glucose transporter GLUT1 [14].
A transport protein that mediates the efflux of lactate from the intraerythrocytic malaria parasite has recently been identified and characterised [15–17]. The protein belongs to the microbial formate nitrite transporter (FNT) family, localises to the parasite plasma membrane (as well as to the membrane bounding the parasite’s internal digestive vacuole), and transports lactate, as well as a variety of other monocarboxylates, in a pH-dependent manner that is consistent with H+-coupled transport [16, 17]. The FNT family is structurally unrelated to the monocarboxylate transporters that export lactate from human cells [16, 17].
In this study we have screened the Malaria Box for compounds that alter the intracellular pH of asexual trophozoite-stage parasites. Fifteen of the 400 compounds were found to acidify the parasite cytosol. Of these, two compounds were found to exert their effects on the parasite’s cytosolic pH by targeting the H+-coupled efflux of lactate via PfFNT.
In a previous study we showed that 28 compounds from the Malaria Box, when added to isolated asexual P. falciparum parasites, give rise to a cytosolic alkalinisation, together with an increase in the cytosolic [Na+] [18]. The data are consistent with the compounds inhibiting the putative Na+/H+ ATPase PfATP4 [19]. Here, we screened the remaining 372 Malaria Box compounds for their effects, at 1 μM, on parasite cytosolic pH. The screen entailed adding a 1 μL aliquot of a 1 mM stock of each compound to a 1 mL suspension of trophozoite-stage P. falciparum parasites that had been isolated from their host erythrocytes by brief exposure to saponin and loaded with the pH-sensitive fluorescent dye BCECF.
The fluorescence of the parasite suspension was monitored continuously. If, on addition of a compound, no change in fluorescence was observed after 2 min, another compound was added to the same cell suspension. A maximum of 9 compounds were added successively to a single batch of cells before concanamycin A (100 nM), an inhibitor of the parasite’s H+-extruding V-type ATPase that has been shown previously to acidify the cytosol in P. falciparum [20], was added as a positive control to confirm that a pH change was still detectable in those cells (Fig 1A).
Fifteen of the 372 Malaria Box compounds tested were found to give rise to an immediate-onset gradual decrease in fluorescence ratio (i.e. a decrease in the ratio of the fluorescence measured using two excitation wavelengths: 495 nm [pH-sensitive numerator] and 440 nm [pH-insensitive denominator]). In each case the decrease in fluorescence ratio resulted from a decrease in the pH-sensitive numerator, consistent with the cells having undergone cytosolic acidification (Fig 1B). The 15 compounds that induced an acidification in the initial screen (and for which structures are shown in S1 Table) were all re-tested in at least two further pH experiments and their effects on the cytosolic pH of the parasite confirmed.
The 15 compounds were also tested for their effects (at 1 μM and 5 μM) on the pH of the parasite’s digestive vacuole. This organelle is acidic, with an estimated pH value of approximately 5 [21–23]. Seven of the 15 compounds dissipated the pH gradient across the digestive vacuole membrane (causing its alkalinisation; S1 Table; S1 Fig).
Of the 372 Malaria Box compounds tested for their effects on cytosolic pH, 23 gave rise to effects on the fluorescence ratio that were considered likely to be ‘optical effects’ rather than genuine pH changes. Such effects arise from either an intrinsic fluorescence of the compound or an interaction between the compound and the fluorescent dye. In each case the change of the fluorescence ratio was abrupt rather than gradual (see example in Fig 1C). Furthermore, in most (20 out of 23) cases the change in fluorescence ratio resulted in full or in part from a change in the nominally pH-insensitive fluorescence emanating from excitation at 440 nm (the denominator). After an optical effect was encountered, concanamycin A (100 nM) was added to the cells to determine whether a pH change was still detectable (which, in each case, it was, ruling out the possibility that the compound that caused the optical effect had also dissipated the pH gradient across the plasma membrane), and the cells were replaced with a new batch of cells before screening further Malaria Box compounds.
One mechanism by which a compound might induce a cytosolic acidification is through inhibition of one (or more) of the plasma membrane proteins that mediate the efflux of H+ from the parasite. Lactate, produced in large quantities by the parasite as the end-product of glycolysis, is excreted via a coupled lactate:H+ transporter [11, 16, 17], raising the possibility that this protein might be a candidate target for some of the compounds identified as hits in the initial screen. Compounds that inhibit the lactate:H+ transporter might be expected to: (i) have no effect on cytosolic pH in parasites suspended in a glucose-free saline, as under this condition glycolysis is not operating, and there is therefore no production of lactate; (ii) prevent the lactate:H+ transporter-mediated cytosolic acidification that results from the exposure of a parasite to a large inward lactate gradient. Although the physiological role of the transporter is to mediate the net export of lactate, down the normally-outward lactate concentration gradient, the transporter is bidirectional and, under conditions of an imposed inward lactate concentration gradient (as results from the addition of a high concentration of lactate to the extracellular medium) it mediates a net influx of lactate:H+, resulting in a cytosolic acidification [11].
When tested at concentrations of 2.5–5 μM, only two of the 15 pH-lowering hits, MMV007839 and MMV000972 (which are structurally similar to one another; S1 Table) demonstrated both features that would be expected from a complete inhibition of lactate:H+ transport (Fig 2). The addition of MMV007839 or MMV000972 to isolated parasites suspended in a glucose-containing medium (in which glycolysis was active) resulted in cytosolic acidification (Fig 2B and 2C). The exposure of parasites to the solvent alone (0.1% v/v DMSO) had no effect on pH (Fig 2A). When isolated parasites are suspended in glucose-free medium, glycolysis ceases, the intracellular ATP concentration decreases to zero, and the parasite is unable to maintain the activity of the plasma membrane V-type H+-ATPase which serves to generate and maintain an inward H+ electrochemical gradient [24, 25]. As a result, the resting pH decreases to 7.0–7.1 (i.e. close to the extracellular pH) compared with 7.2–7.3 in glucose-replete cells. On addition of MMV007839 or MMV000972 to isolated parasites suspended in glucose-free medium (in which lactate production is eliminated) there was no change in the resting pH (Fig 2E and 2F), consistent with the acidification seen in glucose-replete cells being due to the inhibition of lactate efflux.
On restoration of glucose to parasites suspended in glucose-free medium (in the absence of inhibitors) pH increased as ATP levels, and hence the activity of the H+-extruding V-type H+-ATPase, were restored (Fig 2D). When the same manoeuvre was applied to parasites exposed to either MMV007839 or MMV000972, there was an initial acidification (consistent with lactate production having been restored and with the compounds inhibiting the efflux of the newly-generated lactate:H+; Fig 2E and 2F). However, after ~ 2 min the cells underwent a gradual alkalinisation, restoring the cytosolic pH to approximately 7.1 (Fig 2E and 2F). The basis for this recovery is not understood; no such recovery was observed when the compounds were added to parasites maintained continuously in the presence of glucose (Fig 2B and 2C).
MMV007839 and MMV000972 were then tested for their ability to inhibit the lactate:H+ transporter-mediated acidification of the parasite cytosol that normally occurs when lactate is added to the extracellular medium. This was tested at 4°C (Fig 2G–2I) to reduce the ability of the H+-extruding V-type H+-ATPase to counteract the lactate-mediated pH change [11]. At 4°C the acidification of the parasite cytosol induced by the addition of MMV007839 and MMV000972 was less pronounced than that seen at 37°C (cf. Fig 2B and 2C and Fig 2H and 2I), reflecting the reduced rate of glycolysis and hence reduced rate of lactate production at the lower temperature. The addition of 10 mM L-lactate to parasites (following a pre-treatment with solvent (0.25% v/v DMSO), which had no effect on cytosolic pH) induced an abrupt acidification of the parasite cytosol (Fig 2G), consistent with lactate entering the parasite rapidly in symport with H+. On addition of 10 mM L-lactate to parasites that had been pre-treated with MMV007839 or MMV000972, there was no such abrupt acidification (Fig 2H and 2I). Rather, the slow acidification mediated by the compounds continued at the same rate as that seen prior to the addition of L-lactate. For MMV007839 the rates of acidification before and after the addition of lactate were 0.011 ± 0.003 pH unit/min (mean ± SEM) and 0.011 ± 0.004 pH unit/min (n = 3; P = 0.9; paired t-test). For MMV000972 the rates were 0.012 ± 0.003 pH unit/min and 0.015 ± 0.006 pH unit/min (n = 3; P = 0.5). These data are consistent with the two MMV compounds inhibiting the lactate:H+ transporter and, thereby, the influx of lactate:H+ into the parasite.
The pH data obtained with the isolated parasites provide indirect evidence that MMV007839 and MMV000972 inhibit lactate:H+ symport across the parasite plasma membrane but do not reveal the molecular identity of the target. To gain information on the possible molecular target of the compounds, we cultured a recently-cloned P. falciparum Dd2 parasite line in the presence of increasing concentrations of MMV007839 until parasites showing resistance to the growth-inhibiting effect of this compound emerged. Two independent cultures (referred to here as ‘MMV007839-selected cultures A and B’) were initially maintained in the presence of 100 nM MMV007839, and the concentration subsequently increased incrementally over the course of six weeks.
The parasites from both cultures were tested, along with the Dd2 parental line, for their sensitivity to growth inhibition by MMV007839 and MMV000972. Dd2 parental parasites had an IC50 value for MMV007839 of 0.14 ± 0.02 μM and an IC50 value for MMV000972 of 1.8 ± 0.2 μM. Parasites from both MMV007839-selected cultures were highly resistant to both compounds, with IC50 values > 280-fold higher than that of the Dd2 parent for MMV007839, and > 30-fold higher than that of the Dd2 parent for MMV000972 (P < 0.001; paired t-tests; Table 1). The responses of parasites in the MMV007839-selected cultures to chloroquine and artemisinin were not significantly different from those of the Dd2 parental parasites (P > 0.1; Table 1).
Genomic DNA was extracted from both of the MMV007839-selected cultures as well as from the parental Dd2 culture. The entire PfFNT gene was then amplified by PCR and the coding regions sequenced. Parasites from each of the MMV007839-selected cultures were found to have a mutation that was not present in the Dd2 parental parasites or in the 3D7 reference sequence. In both MMV007839-selected cultures (A and B), the mutation in the coding sequence was G319A, which introduces a Gly107Ser mutation into the PfFNT protein. The finding of a mutation in the lactate:H+ transporter PfFNT in the MMV007839-selected cultures is consistent with PfFNT inhibition being the mechanism of action of MMV007839 and MMV000972. The parental and MMV007839-selected Dd2 parasites also had two polymorphisms in the PfFNT coding sequence relative to the 3D7 strain (PlasmoDB ID PF3D7_0316600): a C475G change, which codes for a His159Asp mutation, and a A756G change (synonymous).
The effects of MMV007839 and MMV000972 on lactate transport across the parasite plasma membrane were investigated by measuring the uptake of L-[14C]lactate by saponin-isolated trophozoite-stage parasites in the presence and absence of the compounds. As noted above, although the physiological role of the transporter is to facilitate the net export of lactate from the parasite, the transporter is bidirectional, and measuring the unidirectional flux of L-[14C]lactate into the parasite provides a convenient means of monitoring the activity of the transporter. It has been shown previously that the uptake of external L-[14C]lactate by isolated parasites is enhanced at lower pH values, consistent with the transporter being H+-coupled [11]. The L-[14C]lactate uptake experiments were therefore performed on isolated parasites suspended in an acidic (pH 6.1) medium, at 4°C (to slow the transport process). The level of intracellular L-[14C]lactate accumulation was determined 20 s after the addition of the radiolabel to parasites. Under these conditions (i.e. pH 6.1, 4°C) the 20 s incubation falls within the initial linear phase of uptake of the radiolabel and the measured uptake therefore reflects the lactate influx rate (our own preliminary experiments and [11]).
In isolated 3D7 parasites to which had been added DMSO (0.4% v/v, as a solvent control) the L-[14C]lactate ‘distribution ratio’ (i.e. the intracellular L-[14C]lactate concentration divided by the extracellular L-[14C]lactate concentration) at 20 s reached a value of 6.7 ± 0.4 (mean ± SEM; n = 4; S2 Fig). As has been reported previously [16], the broad-specificity anion transport inhibitor NPPB (50 μM) slowed L-[14C]lactate influx, reducing the distribution ratio at 20 s to 0.4 ± 0.1 (P < 0.001; one-way ANOVA with post hoc Tukey test; S2 Fig). MMV007839 (0.5 μM) inhibited L-[14C]lactate influx to a similar degree (P < 0.001; S2 Fig), with 2 μM MMV007839, as well as 0.5 μM and 2 μM MMV000972, causing complete inhibition of L-[14C]lactate uptake (P < 0.001; S2 Fig).
A comparison of the uptake of L-[14C]lactate into the PfFNTGly107Ser mutant parasites (from MMV007839-selected culture B) and their parental parasites revealed that transport was slowed significantly in the mutant strains (P < 0.001; unpaired t-test; Fig 3A). In the PfFNTGly107Ser mutant parasites the L-[14C]lactate distribution ratio, as measured at 20 s, reached a value of 2.1 ± 0.2 (mean ± SEM; n = 7), compared to a value of 5.3 ± 0.5 (n = 9) in the parental parasites. This finding suggests that the Gly107Ser mutation in PfFNT causes some impairment of its function. NPPB (100 μM) reduced the rate of L-[14C]lactate influx into both PfFNTGly107Ser mutant and parental parasites (Fig 3A; P < 0.001; unpaired t-tests).
The concentration dependence of the inhibition of L-lactate influx by the two MMV compounds in the PfFNTGly107Ser mutant parasites (MMV007839-selected culture B) was compared to that in the parental line. Both compounds were potent inhibitors of L-[14C]lactate uptake by Dd2 (parental) parasites (Fig 3B and 3C). The IC50 values for MMV007839 and MMV000972 in these experiments were 158 ± 42 nM (mean ± SEM; n = 7) and 49 ± 14 nM (n = 3), respectively. In the PfFNTGly107Ser mutant parasites the efficacy of MMV007839 and MMV000972 at inhibiting L-[14C]lactate transport was significantly reduced (Fig 3B and 3C; P ≤ 0.001; unpaired t-tests) with IC50 values of 16.3 ± 4.4 μM (n = 3) and 15.2 ± 1.9 μM (n = 4), respectively.
The finding of a reduced rate of L-[14C]lactate transport across the plasma membrane of PfFNTGly107Ser mutant parasites compared to parental parasites (Fig 3A) raised the question of whether the Gly107Ser mutation in PfFNT might be associated with a reduction in parasite fitness. To investigate this, we compared the growth rates of two PfFNTGly107Ser mutant parasite clones (generated from MMV007839-selected culture B by limiting dilution) with that of the parental Dd2 clone. We performed competition experiments in which mutant and parental parasites were mixed in an approximately 1:1 ratio, and their relative proportions monitored over time. These experiments revealed that the proportion of mutant and parental parasites remained approximately constant over the course of three weeks (S3 Fig). Thus, the Gly107Ser mutation in PfFNT was not associated with a decrease in the growth rate of asexual blood-stage parasites, suggesting that the parasite can withstand a substantial reduction in the rate of PfFNT-mediated lactate transport before its viability is compromised.
To determine directly whether PfFNT is inhibited by MMV007839 and MMV000972, and if so whether the Gly107Ser mutation reduces the sensitivity of PfFNT to inhibition by the compounds, we expressed native PfFNT and PfFNTGly107Ser in Xenopus laevis oocytes.
Consistent with our observations in parasites (Fig 3A), which revealed that the rate of L-[14C]lactate transport into PfFNTGly107Ser mutant parasites was only 39 ± 5% (mean ± SEM; n = 7) of that observed in the parental parasites, Xenopus oocytes expressing the mutant PfFNTGly107Ser protein were found to be significantly impaired in their ability to transport L-[14C]lactate relative to oocytes expressing native PfFNT (Fig 4). In the absence of MMV007839 or MMV000972, the PfFNT-mediated component of L-[14C]lactate transport into oocytes expressing PfFNTGly107Ser was only 52 ± 7% (mean ± SEM; n = 4) of that measured in oocytes expressing native PfFNT (P < 0.001; one-way ANOVA with post hoc Tukey test).
MMV007839 and MMV000972 inhibited L-[14C]lactate uptake via both the native PfFNT and PfFNTGly107Ser transporters (Fig 4, S4 Fig), whilst causing no changes in L-[14C]lactate transport into the control (non-injected) oocytes (S4 Fig). Much higher concentrations of the compounds were required to inhibit L-[14C]lactate transport via PfFNTGly107Ser compared to those found to inhibit transport via native PfFNT (Fig 4). The IC50 values for the inhibition of PfFNT-mediated L-[14C]lactate transport by MMV007839 were 22.5 ± 1.4 nM (mean ± SEM; n = 4) for native PfFNT and 1.3 ± 0.2 μM for PfFNTGly107Ser (P < 0.001; one-way ANOVA with post hoc Tukey test). For MMV000972, the IC50 values for native PfFNT and PfFNTGly107Ser were 50 ± 1 nM and 6.3 ± 0.9 μM, respectively (P < 0.001). These results provide direct evidence that PfFNT is inhibited by MMV007839 and MMV000972 and that the Gly107Ser mutation greatly reduces the sensitivity of PfFNT to inhibition by these compounds.
The specificity of the inhibition by the MMV compounds of the transport of L-[14C]lactate via PfFNT was investigated by testing the effects of the compounds on [3H]chloroquine transport by oocytes expressing a chloroquine-transporting isoform of the P. falciparum chloroquine resistance transporter [26] (PfCRT; from the chloroquine-resistant Dd2 strain) and on [3H]hypoxanthine transport by oocytes expressing the P. falciparum equilibrative nucleoside/nucleobase transporter PfENT1 [27, 28]. Neither compound had a statistically significant effect on the transport of [3H]chloroquine via PfCRT (S5 Fig). Neither compound inhibited [3H]hypoxanthine uptake by PfENT1-expressing oocytes (S5 Fig), although 1 μM MMV000972 caused a slight increase in [3H]hypoxanthine uptake by PfENT1-expressing oocytes (P < 0.05; one-way ANOVA with post hoc Tukey test).
The effect of MMV007839 (the more potent of the two MMV compounds at inhibiting parasite growth) on the metabolite profile of erythrocytes infected with mature 3D7 trophozoites was assessed using a previously-described untargeted LC-MS approach [29]. The compound was added at approximately 20× IC50 (Table 1; 6 μM) and samples were taken for metabolite profiling at 1 h, 3 h and 6 h after the addition. Exposure of 3D7 parasites to 6 μM MMV007839 for 1 h or 3 h did not significantly affect their viability (P > 0.14; paired t-tests; S6 Fig), assessed by exposing parasitised erythrocytes to the inhibitor for the specified time, then removing the inhibitor and culturing the parasites for a further three days and measuring the final parasitaemia. By contrast, exposure for 6 h resulted in a reduction in parasitaemia (as measured three days after the exposure) to 49 ± 7% (mean ± SEM; n = 3) of control levels (P = 0.007; S6 Fig).
The levels of a number of metabolites within infected erythrocytes were found to be affected by MMV007839 (Fig 5; S1 Data). For those metabolites for which differences in abundance between MMV007839-treated and control samples were statistically significant, the relative levels (i.e. the level of the metabolite in MMV007839-treated cells relative to that in untreated cells) are shown in Fig 5B. Lactate was the metabolite that showed the greatest increase in abundance in MMV007839-treated infected erythrocytes relative to untreated infected erythrocytes (a 6-fold increase at each time point; Fig 5A and 5B; S1 Data), consistent with the MMV compound inhibiting the efflux of lactate from the parasite. There was also a significant elevation of pyruvate (the metabolite directly upstream of lactate), the glycolytic metabolite sn-glycerol-3-phosphate, and a number of peptides, in MMV007839-exposed cells (Fig 5A and 5B).
The levels of a number of metabolites were decreased in MMV007839-treated infected erythrocytes relative to untreated infected erythrocytes. These included NADH and the pyrimidine precursors orotate and orotidine-5-phosphate (Fig 5A and 5B).
The levels of a range of metabolites in the extracellular medium were also analysed (Fig 5C). Extracellular lactate and α-ketoglutarate levels were decreased in MMV007839-treated samples relative to untreated samples, whereas the levels of pyruvate, phosphoenolpyruvate and 3-phosphoglycerate were increased. Phosphoenolpyruvate and 3-phosphoglycerate were the only metabolites for which differences in abundance between MMV007839-treated and untreated samples were statistically significant (S1 Data).
The MMV007839-induced accumulation of lactate (and glycolytic intermediates) within the parasite might be expected to induce cell swelling, through osmotic effects. To test this we investigated the effect of MMV007839 on the volume of both isolated (3D7) parasites and intact parasitised erythrocytes, using a Coulter Multisizer. Addition of MMV007839 (1 μM) to a suspension of isolated parasites caused the parasites to swell, increasing their volume by 13.1 ± 2.8% (relative to their starting volume; equating to a volume increase of 5.1 ± 0.8 fL; mean ± SEM; n = 3; P < 0.001) within 10 min and maintaining a similar volume for the next 30 min (Fig 6A). MMV007839 (1 μM) also caused erythrocytes infected with 3D7 trophozoites to swell, increasing their volume by 7.2 ± 1.3% (relative to their starting volume; equating to a volume increase of 5.4 ± 1.0 fL; mean ± SEM; n = 3; P < 0.01) within 10 min and maintaining a similar volume for the next 50 min (Fig 6B).
In this study we provide multiple lines of evidence that two structurally-related compounds from the MMV Malaria Box, MMV007839 and MMV000972, kill asexual blood-stage P. falciparum parasites via inhibition of the lactate:H+ transporter PfFNT. Compounds with this mechanism of action have not been described or exploited as antimalarial drugs previously. Our findings highlight the potential of PfFNT as an antimalarial drug target.
The most direct line of evidence that the two compounds target PfFNT comes from experiments in which PfFNT was studied in isolation from other P. falciparum transporters in the Xenopus oocyte. In this heterologous expression system, MMV007839 and MMV000972 inhibited L-[14C]lactate transport via the native form of PfFNT with IC50 values of 23 nM and 50 nM, respectively, while not inhibiting two other P. falciparum transporters (PfCRT and PfENT1). The IC50 value for PfFNT inhibition by MMV007839 is somewhat lower than the IC50 values we obtained for parasite growth inhibition by this compound (260 nM for 3D7 parasites and 140 nM for Dd2 parasites). The IC50 value for parasite growth inhibition by MMV000972 (1.8 μM for Dd2 parasites) was much higher than that obtained for inhibition of native-PfFNT-mediated L-[14C]lactate transport. MMV007839 and MMV000972 have also been tested against 3D7 parasites by other groups, yielding IC50 values of 283–442 nM and 2.6 μM, respectively (http://www.mmv.org/research-development/malaria-box-supporting-information). Given that MMV007839 and MMV000972 inhibit PfFNT with comparable potency when expressed in oocytes, the much higher IC50 value for parasite growth inhibition by MMV000972 compared to MMV007839 might result from a lower concentration of the former compound in the parasitised erythrocyte or the parasite itself.
Parasites selected for resistance to growth-inhibition by MMV007839 had a Gly107Ser mutation in PfFNT. This mutation, which was observed in two independent cultures within six weeks of first adding the compound, decreased the susceptibility of parasites to MMV007839 and MMV000972 by > 30-fold while not affecting their sensitivity to chloroquine or artemisinin. The fact that the MMV007839-selected parasites were cross-resistant to MMV000972 was not surprising in light of the high degree of structural similarity between the compounds and our finding that they both inhibit PfFNT.
The fact that we were able to generate parasites with high-level resistance to MMV007839 and MMV000972 within six weeks is a potential cause for concern with regard to the potential of such compounds as antimalarials. However, this would not necessarily deter further investigations into exploiting PfFNT as a drug target. Many compounds for which resistance has been generated in vitro are undergoing development [3], such as the protein translation inhibitor DDD498 [30] and various compounds that have been proposed to target PfATP4 [31–33]. A relationship between the propensity of parasites to develop resistance in vitro and the likelihood of parasite resistance emerging in vivo is not firmly established. Factors such as the drug’s rapidity of action [34] and the growth rate and transmissibility of drug-resistant parasites also contribute to the overall threat of clinical resistance [35]. Furthermore, carefully selected dosing regimens and partner drugs may reduce the likelihood of selecting for resistance-conferring mutations in the target protein in vivo.
Our direct studies of PfFNT inhibition were complemented by in situ studies with P. falciparum parasites. The parasite studies provided multiple lines of evidence (based both on pH measurements and L-[14C]lactate flux measurements) that both MMV compounds abolish L-lactate transport across the parasite plasma membrane. Experiments with L-[14C]lactate showed that MMV007839 and MMV000972 inhibited the transport of L-lactate across the parasite plasma membrane with IC50 values of 158 nM and 49 nM respectively in wild-type Dd2 parasites. These IC50 values are similar to those obtained for the inhibition of L-[14C]lactate uptake by PfFNT expressed in Xenopus oocytes. This suggests that PfFNT may be the only transporter that can mediate the transport of L-lactate across the parasite plasma membrane at an appreciable rate, at least in trophozoite-stage parasites. The parasite does encode two transporters with homology to human monocarboxylate transporters, however neither showed a capacity to transport lactate when expressed heterologously in yeast [17].
The potency with which MMV007839 and MMV000972 inhibited L-[14C]lactate transport by parasites was significantly reduced in the resistant parasites bearing the Gly107Ser mutation in PfFNT. Using the Xenopus oocyte system, this finding was attributed directly to a reduced inhibitor sensitivity of the mutant PfFNTGly107Ser protein compared to the native protein. The mutated residue is adjacent to T106, which has been predicted to form part of the PfFNT transport channel [17]. The mutation may reduce the affinity with which MMV007839 and MMV000972 bind to the transporter, or otherwise prevent the compounds from inhibiting its function. The mutation was associated with an approximately 50–60% reduction in L-[14C]lactate transport by both parasites and PfFNT expressed in Xenopus oocytes. This reduction in lactate transport was not associated with a decrease in the growth rate of asexual blood-stage parasites, suggesting that the rate of lactate export from parasites must be inhibited by more than 50% to affect their viability under the conditions tested here.
The metabolite profiles of MMV007839-treated and untreated parasitised erythrocytes provided further evidence in support of the mechanism of action proposed here. Lactate levels were specifically elevated in MMV007839-treated infected erythrocytes compared to untreated infected erythrocytes, consistent with decreased efflux from infected erythrocytes. The level of lactate secreted into the medium was also reduced in MMV007839-treated cultures relative to untreated cultures (although this decrease did not reach statistical significance), while pyruvate, phosphoenolpyruvate and 3-phosphoglycerate secretion was increased in the MMV007839-treated infected erythrocytes. Collectively, these analyses indicate that MMV007839 induces a build-up of lactate, its immediate precursor (pyruvate), and other glycolytic intermediates. This build-up is partly countered by increased secretion of partly oxidised intermediates and accompanied by significant swelling of both the parasite and the intact infected erythrocyte. The observed accumulation of lactate and pyruvate is a distinctive phenotype that has not been observed previously when infected erythrocytes were treated with a range of antimalarials including glycolysis inhibitors [29]. The metabolic perturbations were observed after 1 h, 3 h and 6 h exposures to MMV007839. As parasite viability was not affected at the early time points, these changes cannot be attributed to cell death.
NADH levels were significantly lower in MMV007839-treated compared to untreated infected erythrocytes. This might be explained by the consumption of NADH in the reaction that produces sn-glycerol 3-phosphate, which was significantly elevated within the MMV007839-treated cells. Significant changes in intracellular abundance were also observed for a range of other metabolites, including a number of (likely haemoglobin-derived) small peptides and pyrimidine precursors. The basis for these changes was not investigated, although they may be a secondary consequence of the MMV007839-induced acidification of the parasite cytosol or disruption to energy metabolism. Previous work has demonstrated that inhibiting the P. falciparum glucose transporter not only perturbs glycolysis but also impacts other metabolic pathways including pyrimidine biosynthesis and haemoglobin digestion [29]. Studies in which the fates of isotope-labelled metabolites are tracked would be required to resolve the mechanistic basis for the full range of metabolic perturbations resulting from inhibition of PfFNT. Thus while the changes in the levels of glycolytic intermediates observed in MMV007839-treated parasitised erythrocytes are fully consistent with inhibition of PfFNT, it remains possible that changes in the levels of other metabolites could have resulted from off-target effects of MMV007839 rather than as a consequence of downstream effects of PfFNT inhibition.
A recently published study in which extensive toxicity testing was performed on the Malaria Box compounds revealed that MMV007839 and MMV000972 display toxicity in some assays [36]. The compounds themselves are therefore unlikely to be suitable as clinical candidates. Further screening for compounds that inhibit PfFNT will be required to develop suitable clinical candidates that might be used to exploit this target in the field.
In summary, our screen of the Malaria Box for compounds that perturb the pH inside the parasite led to the discovery of two compounds that kill parasites via inhibition of a novel target, PfFNT. The mechanisms of action of the remaining pH-decreasing hits are yet to be determined, and further studies on these compounds may uncover additional targets. Another transport protein that mediates the translocation of H+ across the plasma membrane, PfATP4, has emerged recently as a particularly vulnerable drug target [37], and compounds believed to target this pump give rise to a cytosolic alkalinisation. Thus, pH assays can be used as a primary screen for the identification of additional compounds that inhibit either PfFNT or PfATP4.
The Malaria Box [8] was provided by MMV. Upon receipt of the Malaria Box, each compound was diluted to 1 mM in DMSO and aliquoted into multiple plates. Information about the compounds is accessible via the ChEMBL-NTD database (https://www.ebi.ac.uk/chemblntd). Additional quantities of MMV007839 and MMV000972 were obtained from MMV and Vitas-M Laboratory, respectively.
The use of human blood in this study was approved by the Australian National University Human Research Ethics Committee. The blood was provided by the Australian Red Cross Blood Service without disclosing the identities of the donors.
Ethical approval of the work performed with the Xenopus laevis frogs was obtained from the Australian National University Animal Experimentation Ethics Committee (Animal Ethics Protocol Number A2013/13) in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes.
The 3D7 strain of P. falciparum (isolated in the Netherlands but likely to be of African origin) was used throughout this study, except for the MMV007839 resistance selection studies, which were initiated with a clone of Dd2 (Thai origin) that had been generated previously by limiting dilution [38], and the subsequent experiments performed to characterise the resistant parasites. The 3D7 strain and the Dd2 clone were checked for mycoplasma contamination prior to their use in this study and were found to be mycoplasma-free. Parasites were cultured in human erythrocytes [39], with continuous shaking [40], and were synchronised by sorbitol treatment [41]. The culture medium was RPMI 1640 containing 25 mM HEPES (Gibco) supplemented with 11 mM additional glucose, 0.2 mM hypoxanthine, 20 μg/mL gentamicin sulphate and 3 g/L Albumax II.
Prior to cytosolic pH, digestive vacuole pH, parasite volume and L-[14C]lactate uptake measurements, mature trophozoite-stage parasites (approximately 34–40 h post-invasion) were functionally isolated from their host erythrocytes by brief exposure (of cultures at approximately 4% haematocrit) to saponin (0.05% w/v, of which ≥ 10% was the active agent sapogenin) [42]. Unless stated otherwise the parasites were then washed several times in bicarbonate-free RPMI 1640 supplemented with 11 mM additional glucose, 0.2 mM hypoxanthine and 25 mM HEPES (pH 7.10), and maintained in this medium at a density of 1 × 107–3 × 107 parasites mL-1 at 37°C until immediately before their use in experiments.
Cytosolic pH was measured as described previously [25] at 37°C (unless stated otherwise) using saponin-isolated trophozoites loaded with the pH-sensitive fluorescent dye BCECF (Molecular Probes). In most experiments parasites were suspended in ‘Experimental Saline Solution’ (125 mM NaCl, 5 mM KCl, 1 mM MgCl2, 20 mM glucose, 25 mM HEPES; pH 7.10 unless stated otherwise). For experiments in which parasites were deprived of glucose, glucose-free Experimental Saline Solution was used (135 mM NaCl, 5 mM KCl, 1 mM MgCl2, 25 mM HEPES; pH 7.10). Parasites suspended at a density of 1 × 107–3 × 107 parasites mL-1 were excited successively at 440 and 495 nm, with emission recorded at 520 nm, using either a PerkinElmer LS 50B Fluorescence Spectrometer or a Cary Eclipse Fluorescence Spectrophotometer. The relationship between the ratio of the two measurements (495 nm/440 nm) and the cytosolic pH was calibrated as described previously [25] with one exception: for experiments performed at 4°C, the previously published calibration method yielded abnormally high values for resting pH, and the calibration was therefore performed using Experimental Saline Solution (adjusted to various pH values) in place of the high-K+ calibration solutions.
The pH inside the digestive vacuole (pHDV) was monitored at 37°C using isolated parasites containing fluorescein-dextran (10,000 MW; Molecular Probes) in their digestive vacuoles. The parasites were prepared as described previously [43] and suspended in Experimental Saline Solution (pH 7.10) at a density of 1 × 107–3 × 107 parasites mL-1. Fluorescence ratio measurements were obtained with a PerkinElmer Life Sciences LS 50B fluorometer with a dual excitation Fast Filter accessory using excitation wavelengths of 440 and 495 nm and an emission wavelength of 520 nm [44].
The uptake of L-[14C]lactate (PerkinElmer) by isolated 3D7 trophozoites was determined essentially as described previously [16]. Experiments were performed at 4°C on parasites suspended in pH 6.10 Experimental Saline Solution. Parasites were preincubated with either DMSO or inhibitors (at the concentrations stated in the relevant Figure legends) for 1 min before a 200 μL sample of the suspension was added (at ‘time zero’) to an equal volume of pH 6.10 saline (with or without inhibitors as specified in the Figure legends) layered over a 200 μL dibutyl phthalate/dioctyl phthalate (5:4 v/v) oil mix (in a microcentrifuge tube) and containing L-[14C]lactate (giving a final concentration after the cell suspension was added of 1.3 μM).
At 20 s after combining the parasites and L-[14C]lactate, uptake was terminated by centrifuging the sample (15,800 × g, 2 min in a rapid-acceleration Beckman Microfuge E), thereby sedimenting the parasites below the oil layer. A 10 μL aliquot of the supernatant solution was taken from above the oil layer in each tube to enable the determination of extracellular L-[14C]lactate concentrations. The remaining supernatant solution was discarded and residual radioactivity on the sides of each tube was removed by rinsing the tube four times with water before aspirating most of the oil. The cell pellets were lysed using Triton X-100 (0.1% v/v) and deproteinised with trichloroacetic acid (2.5% w/v final) before the radioactivity in each sample was measured using a scintillation counter [42].
In each experiment the amount of radioactivity in the cell pellets that was attributable to ‘trapped’ extracellular L-[14C]lactate was estimated under conditions in which the amount of L-[14C]lactate taken up by the parasites was minimised by using the previously-described [16] lactate transport inhibitor NPPB. Parasites were preincubated with NPPB (100 μM) for 1 min then combined with a L-[14C]lactate solution to which unlabelled L-lactate and NPPB had been added (yielding a final lactate concentration of 10 mM and a final NPPB concentration of 100 μM), then centrifuged immediately.
‘Distribution ratios’ (the intracellular L-[14C]lactate concentration divided by the extracellular L-[14C]lactate concentration) were calculated as described previously [16]. The extracellular L-[14C]lactate concentration for each sample was determined using the aliquots of supernatant solution taken from above the oil layer. The intracellular L-[14C]lactate concentration for each sample was determined from the radioactivity incorporated into the cell pellet (after subtraction of the radioactivity attributable to the trapped extracellular L-[14C]lactate), the parasite number, and the previously determined estimate of the water volume of a saponin-isolated parasite [42].
Parasite proliferation was measured in 96-well plates using a fluorescent DNA-intercalating dye [45]. The assays were initiated with erythrocytes infected with predominantly ring-stage parasites, and the starting parasitaemia was 0.5–1% and the haematocrit was 1%. The experimental protocol and procedure for data analysis were the same as those outlined previously [46] with two modifications: the duration of the assay was 68–72 h, and a supramaximal concentration of an antiplasmodial compound (either 5 μM chloroquine, 0.5 μM artemisinin, 12.8 nM KAE609 or 150 μM MMV007839) was used for the ‘zero growth’ control. The concentration of DMSO in the assays did not exceed 0.2% (v/v).
The volume of saponin-isolated 3D7 trophozoites and of erythrocytes infected with 3D7 trophozoites was measured using a Beckman Coulter Multisizer 4 fitted with a 100 μm ‘aperture tube’. In the case of isolated parasites, the cells were washed and resuspended (at 37°C) in pH 7.10 Experimental Saline Solution. The electrolyte solution within the aperture tube was the same as the Experimental Saline Solution except that it lacked glucose. For experiments on infected erythrocytes, the infected cells were separated from uninfected erythrocytes using a Miltenyi Biotec VarioMACS Magnet [47, 48] and the cells were maintained at 37°C in bicarbonate-free RPMI supplemented with 25 mM HEPES, additional glucose (11 mM) and 0.2 mM hypoxanthine, and pH-adjusted to 7.40. For these measurements the electrolyte solution within the aperture tube differed only from this medium in that it was not supplemented with hypoxanthine or additional glucose. For each measurement of cell volume, approximately 20,000 pulses (each corresponding to the passage of a single cell through the aperture) were recorded. The median volume of the cells within each sample was determined by fitting a log Gaussian distribution curve to the population data.
A clone of the chloroquine-resistant Dd2 strain of P. falciparum was used to generate MMV007839-resistant parasites. This clone was generated previously by limiting dilution and was used successfully to generate resistance to two (PfATP4-associated) Malaria Box compounds [18]. Two independent cultures (each containing ~ 5 × 108 parasites) were exposed to increasing concentrations of MMV007839 for ~ 6 weeks, starting with a concentration of 100 nM on Day 0, and increasing to 125 nM (approximately the IC50 value for the parental Dd2 line) on Day 3. For MMV007839-selected culture A, the drug concentration was increased by 5 nM increments on Day 18, Day 21, and every two days thereafter. For MMV007839-selected culture B, the drug concentration was increased by 15 nM increments every two days starting on Day 18. The cultures were provided with fresh medium and blood and diluted to reduce the parasitaemia as needed.
Genomic DNA was extracted from saponin-isolated parasites (from the MMV007839-selected bulk cultures) using a QIAGEN Plant DNeasy kit. To sequence exons 1–3, the entire PfFNT gene was PCR-amplified using KOD Hot Start DNA polymerase and primers 1 and 2 (S2 Table). The PCR product was extracted using a GeneJET Gel Extraction Kit (Thermo Fisher Scientific). Sequencing was performed at the ACRF Biomolecular Resource Facility (The John Curtin School of Medical Research, Australian National University) using primers 1 and 3–7 (S2 Table). To sequence the fourth (and final) exon, the 3’ end of the gene and part of the 3’ UTR was PCR amplified using the primers 3 and 8 (S2 Table), with primer 8 also used for sequencing.
For fitness assays, two clones were obtained from MMV007839-selected culture B by limiting dilution, performed essentially as described previously [38]. The presence of the G319A mutation in the PfFNT coding sequence (giving rise to the Gly107Ser mutation in PfFNT) was confirmed in both clones.
Erythrocytes infected with 3D7 trophozoites were separated from uninfected erythrocytes using a magnet supplied by Colebrook Bioscience [47, 48]. The resulting cells, which consisted of > 95% infected erythrocytes, were then allowed to recover for 0.5–1 h at 37°C in ‘complete medium’ (RPMI 1640 supplemented with 0.5% Albumax II, 25 mM HEPES, 100 μM hypoxanthine and 10 μg/mL gentamycin). At ‘time zero’, MMV007839 (6 μM) was added to three cell suspensions (each containing 1 × 108 cells at 0.4% haematocrit), while three identical cell suspensions were left untreated. The cell suspensions were incubated at 37°C under controlled atmospheric conditions (5% CO2 and 1% O2 in N2). At each of three time points (1 h, 3 h and 6 h), one MMV007839-treated cell suspension and one untreated cell suspension were used to extract metabolites. To extract metabolites, cell suspensions were first pelleted by centrifugation (14,000 × g, 30 s). Aliquots (5 μL) of the supernatant media were collected and the remaining supernatant media discarded. The cells were then washed with 1 mL of ice-cold PBS. To extract metabolites, 200 μL volumes of 80% acetonitrile (in H2O; containing 5 μM [13C]aspartate as an internal standard) were added to the cell pellets and samples of extracellular media. The samples were vortexed briefly then centrifuged (14,000 × g, 10 min, 4°C), and the resulting supernatant solutions were transferred to vials for LC-MS analysis.
The metabolites were separated on a SeQuant ZIC-pHILIC column (5 μm, 150 × 4.6 mm, Millipore) with a 1200 series HPLC system (Agilent), using a flow rate of 0.3 mL/min with 20 mM ammonium carbonate in water (A) and 100% acetonitrile (B) as the mobile phase. A binary gradient was set up as follows: 0.5 min: 20% A and 80% B, 30 min: 80% A and 20% B, 31 min: 95% A and 5% B, 35.5–45 min: 20% A and 80% B. Detection of metabolites was performed on an Agilent Q-TOF mass spectrometer 6550 operating in negative ESI mode. The scan range was 85–1200 m/z between 5 and 35 min at 0.9 spectra/second.
LC-MS.d files were converted to.mzXML files using MS convert and analysed using MAVEN [49]. Following alignment, metabolites were assigned using exact mass (< 5 ppm) and retention time (compared to a standards library of 150 compounds run on the same day). The area top for each positively assigned metabolite was integrated and used to calculate the ratio of the metabolite concentration between MMV007839-treated and untreated controls. Log2 ratios (+ MMV007839/- MMV007839) across the time series were then plotted using the heatmap script in R.
The Xenopus laevis oocyte expression system was used to assess the effects of MMV007839 and MMV000972 on PfFNT (from the 3D7 strain), PfENT1 (from the FAF6 strain) and PfCRT (from the Dd2 strain). The oocyte expression vectors containing PfFNT, PfENT1 and PfCRT were made previously [16, 26, 50]. The Gly107Ser mutant form of PfFNT was generated through site-directed mutagenesis of the oocyte expression vector containing wild-type (3D7) PfFNT using primers 9 and 10 (S2 Table). Oocytes were harvested from adult female Xenopus laevis frogs and prepared as described previously [51]. cRNA was made using the mMessage mMachine T7 transcription kit and the MEGAclear kit (Ambion) and microinjected into oocytes as outlined elsewhere [51]. The amount of cRNA injected (per oocyte) was 30 ng for PfFNT, 10 ng for PfENT1, and 20 ng for PfCRT. The uptake of radiolabelled substrates was measured 2–5 days post-injection at 27.5°C in ND96 buffer (containing 96 mM NaCl, 2 mM KCl, 1 mM MgCl2, 1.8 mM CaCl2, 10 mM MES and 10 mM Tris-base; pH-adjusted to the value specified in the relevant Figure legends). L-[14C]lactic acid (Na+ salt; 150.6 mCi/mmol) and [3H]hypoxanthine monochloride (14 Ci/mmol) were purchased from Perkin Elmer, and [3H]chloroquine (20 Ci/mmol) was purchased from American Radiolabeled Chemicals. Each experiment used ten oocytes for each condition tested. The influx of the radiolabelled substrate was halted by removing the reaction buffer and washing the oocytes twice in ice-cold ND96 buffer (3.5 mL). The oocytes were then transferred to separate wells of a 96-well plate, lysed with 10% SDS, and the radioactivity measured as described previously [52].
For measurements of cell volume, one-way ANOVAs were carried out with the pre-normalised data using ‘experiment’ as a ‘blocking factor’, to prevent differences in the starting volume between independent experiments (which may have resulted from differences in the average age of the parasites on the different days) from eroding the precision of the test. Post hoc comparisons were then performed using the least significant difference test.
For the LC-MS data, two-way ANOVAs were performed for both the extracellular and intracellular metabolite datasets. Between-subject testing was performed and the results corrected for multiple hypothesis testing using the False Discovery Rate determined by MetaboAnalyst [53].
For all other comparisons, P values were obtained using either one-way ANOVAs followed by post hoc Tukey tests or t-tests (paired or unpaired as appropriate), as stated in the relevant sections. All tests were two-sided.
|
10.1371/journal.ppat.1000348 | Bim and Bmf Synergize To Induce Apoptosis in Neisseria Gonorrhoeae Infection | Bcl-2 family proteins including the pro-apoptotic BH3-only proteins are central regulators of apoptotic cell death. Here we show by a focused siRNA miniscreen that the synergistic action of the BH3-only proteins Bim and Bmf is required for apoptosis induced by infection with Neisseria gonorrhoeae (Ngo). While Bim and Bmf were associated with the cytoskeleton of healthy cells, they both were released upon Ngo infection. Loss of Bim and Bmf from the cytoskeleton fraction required the activation of Jun-N-terminal kinase-1 (JNK-1), which in turn depended on Rac-1. Depletion and inhibition of Rac-1, JNK-1, Bim, or Bmf prevented the activation of Bak and Bax and the subsequent activation of caspases. Apoptosis could be reconstituted in Bim-depleted and Bmf-depleted cells by additional silencing of antiapoptotic Mcl-1 and Bcl-XL, respectively. Our data indicate a synergistic role for both cytoskeletal-associated BH3-only proteins, Bim, and Bmf, in an apoptotic pathway leading to the clearance of Ngo-infected cells.
| A variety of physiological death signals, as well as pathological insults, trigger apoptosis, a genetically programmed form of cell death. Pathogens often induce host cell apoptosis to establish a successful infection. Neisseria gonorrhoeae (Ngo), the etiological agent of the sexually transmitted disease gonorrhoea, is a highly adapted obligate human-specific pathogen and has been shown to induce apoptosis in infected cells. Here we unveil the molecular mechanisms leading to apoptosis of infected cells. We show that Ngo-mediated apoptosis requires a special subset of proapoptotic proteins from the group of BH3-only proteins. BH3-only proteins act as stress sensors to translate toxic environmental signals to the initiation of apoptosis. In a siRNA-based miniscreen, we found Bim and Bmf, BH3-only proteins associated with the cytoskeleton, necessary to induce host cell apoptosis upon infection. Bim and Bmf inactivated different inhibitors of apoptosis and thereby induced cell death in response to infection. Our data unveil a novel pathway of infection-induced apoptosis that enhances our understanding of the mechanism by which BH3-only proteins control apoptotic cell death.
| Infection with various pathogens results in the inhibition or activation of apoptotic cell death [1]. Whereas viral pathogens frequently inhibit host cell apoptosis, many bacteria kill immune or epithelial cells by apoptosis allowing them to subvert immune reactions or to invade tissues, respectively. The obligate human specific bacterium Neisseria gonorrhoeae (Ngo), the causative agent of the sexually transmitted disease gonorrhea, induces apoptosis in genital epithelia. Since induction of apoptosis requires the firm attachment of the gonococci to host cells [2], exfoliation of infected epithelial cells covered with adherent bacteria has been suggested as the immediate cellular responses against infection [3],[4]. This detachment-associated apoptosis of infected cells resembles anoikis, a special form of apoptosis that is induced by absent or inappropriate cell–matrix interactions [5].
Bcl-2 family proteins control mitochondrial outer membrane permeabilization (MOMP), which is the critical step in many forms of apoptosis [6],[7]. The Bcl-2 family consists of pro- and antiapoptotic members that share homologies within their Bcl-2 homology domains (BH). The antiapoptotic Bcl-2 family proteins harbor BH1-4 domains and presumably act by sequestering and inhibiting proapoptotic Bcl-2 members [8]. Proapoptotic Bcl-2 family proteins can be further subdivided into the branch of pore forming, multidomain BH1-3 proteins (like Bak and Bax) and the BH3-only branch (including Bim, Bmf, Bid, Bad, Noxa and Puma) [9],[10]. Active BH3-only proteins cause conformational changes within Bak and Bax, which subsequently homooligomerize and form pores in the outer mitochondrial membrane [11],[12]. MOMP culminates in the release of proapoptotic proteins like cytochrome c, leading to the activation of caspases and caspase-independent death effectors [13].
The mechanisms through which BH3-only proteins activate Bak or Bax are not fully understood. BH3-only proteins may release Bak and Bax from inhibition by anti-apoptotic Bcl-2 protein [14]. Alternatively, the group of BH3-only proteins may include two subgroups, namely survival antagonists that neutralize BH1-4 proteins, and death agonists that activate BH1-3 proteins [15],[16]. A competition of death agonists and survival antagonists for the binding to BH1-4 proteins has been reported [17]. Upon binding of survival antagonists, death agonists are released from their sequestration by BH1-4 proteins and hence freed to act directly on BH1-3 proteins.
Cytotoxic stimuli activate BH3-only proteins by a variety of distinct mechanisms such as p53-dependent transcriptional regulation (Puma and Noxa [18],[19]), proteolytic cleavage (Bid [20]), dephosphorylation (BAD [21]) or phosphorylation (Bim and Bmf). Under normal circumstances, Bim and Bmf are sequestered via dynein light chains (DLC) to the actin and tubulin cytoskeleton, respectively, which prevents them from activating Bak and Bax [22],[23]. Previous work has suggested that phosphorylation of Bim and Bmf within their DLC-binding sites is mediated by Jun-N-terminal kinase-1 (JNK-1) facilitating the release of both proteins from the cytoskeleton during anoikis [24],[25].
Here, we analyzed the signaling pathways upstream of Bak and Bax in Ngo-infected cells. Unexpectedly, both Bim and Bmf were found to act in concert to induce apoptosis of Ngo infected cells. Our data suggest a role of Ngo infection-induced cytoskeletal reorganization in the initiation of apoptosis pathways.
Exfoliation of epithelial cells has previously been described to be caused by Ngo infection [3],[4],[26]. Since this process resembles anoikis, we further investigated the connection between Ngo-induced cell detachment and apoptosis. HeLa cells were infected with Ngo VPI (N242), a clinical isolate and morphological changes were correlated with the activation of caspases. Detachment from the culture support was visible as soon as 6 to 9 h post-infection (Figure 1A) concomitant with the proteolytic maturation of caspase 3 (Figure 1B). To test whether caspase activity is required for exfoliation, cells were infected in the absence or presence of the pan-caspase inhibitor Z-VAD-fmk and then were analyzed by electron and fluorescence microscopy. In the presence of Z-VAD-fmk, Ngo-infected cells continued to detach yet remained otherwise intact and hence failed to disintegrate by apoptosis (Figure 1C and 1D, Video S1) while the activation of caspases 3 and 7 was blocked during the entire duration of the experiment (Figure 1E and 1F). Detachment was further analyzed by acquiring z stacks of infected cells by laser scanning confocal microscopy and subsequent 3-dimensional remodeling (Figure 1G and Video S2). The detachment and induction of apoptosis is not a general response of these cells to infection stress since the same cell line exhibits marked apoptosis resistance as consequence of infection with C. trachomatis [27],[28]. These results demonstrate that caspases are required for the apoptotic disassembly of Ngo-infected cells but not for their detachment.
To test whether the observed effect is specific for HeLa cells and the bacterial strain VPI (N242), we tested other gonococcal derivatives for their capacity to induce exfoliation and cell death. Eleven clinical gonococcal isolates from different patients isolated from blood, urethra, cervix, vagina or urine were analyzed. The capacity of these strains to adhere to HeLa cells correlated well with the induction of exfoliation and apoptosis (Table 1; Figure S1), suggesting that exfoliation and apoptosis induction is a common effect of adherence to HeLa cells. We also investigated whether the effect is specific for HeLa cells. N242 induced significant apoptosis in ME180 and Hep2 cells. From the genetically defined derivatives of the laboratory strain MS11, only strain N920 which forms pili induced apoptosis in ME180 cells, whereas strains expressing Opa57 (N1163, not shown) or Opa52 (N309) failed to induce significant apoptosis in HeLa, Hep2 and ME180 cells (Figure S2). N242 is a clinical isolate expressing 5 different Opacity-associated (Opa) proteins required for the binding of these bacteria to heparane sulfate proteoglycanes (HSPG), integrins or ‘carcinoembryonic antigen-related cell adhesion molecules’ (CEACAMs) of the host cell ([29]; for review see [30]). We next tested whether laboratory strains gain the capacity to induce detachment of HeLa cells expressing Opa protein receptors. The MS11 strain MS11 N927 (Opa−; PorBIA) failed to induce detachment of HeLa cells permanently expressing CEACAM 1 or CEACAM 3 respectively (Figure S3) [31]. In contrast, the MS11 strain N1163 (Opa57; PorBIA) induced detachment of infected cells in the CEACAM 1 expressing HeLa cells but not in the cell line expressing CEACAM 3 (Figure S3). These data suggested an interaction-, and in addition a receptor-specific mechanism underlying detachment and apoptosis induction.
Since infection of HeLa cells with N242 caused the most prominent effects, we focused on this system to further investigate the mechanisms underlying infection-induced apoptosis. We have previously demonstrated that infection with Neisseria induced the activation of Bak and Bax and finally apoptotic cell death [32]. To delineate the signaling pathway leading to the activation of Bak and Bax, we systematically depleted BH3-only proteins in a RNA interference miniscreen. The knockdown of the siRNAs was validated by quantitative real-time PCR (>75% knockdown at the mRNA level) and immunoblot analysis (>75% knockdown at the protein level) (Figure 2A and 2B). Knockdown of Bim and Bmf (but not that of Bid or Bad) resulted in a significant reduction of effector caspase activity, as measured with a fluorogenic caspase 3/7 substrate or by immunochemical detection of proteolytically mature caspase 3 (Figure 2C and 2D, and Figure S4). Bim and Bmf knockdown specifically inhibited the caspase activation induced by Ngo (Figure 2C and 2D, and Figure S5), yet had no effect on caspase activation induced by the genotoxic agent cisplatin (Figure S5A). Ngo infection failed to induce Puma, Noxa and any of the tested mRNAs for BH3 only proteins (Figure 2E, and Figure S6), although cisplatin was able to activate the transcription of both the Puma and Noxa genes (Figure S5B and Figure S5C). Accordingly, the depletion of Puma or Noxa did not affect caspase-3 activation in Ngo-infected cells (Figure 2F). These data demonstrate that Bim and Bmf are specifically required for the Ngo-triggered activation of caspases.
In healthy cells, Bim and Bmf are sequestered to the cytoskeleton by binding to dynein light chains. In response to cytotoxic stimuli, that induce cytoskeletal rearrangements, Bim and Bmf may act as stress sensors in thus far that they are released from the cytoskeleton and induce the activation of Bak and Bax [22],[23]. Accordingly, cytoskeleton fractions obtained from Ngo-infected cells generally contained less Bim and Bmf than those from non-infected control cells (Figure 2G and Figure S7A). A similar result was obtained when cytoskeleton and cytosol were separated by sucrose gradient centrifugations. Bim and Bmf from infected samples shifted from heavier cytoskeleton containing to lighter fractions (Figure S7B), indicating a release of these proteins from the cytoskeleton in infected cells. Addition of Z-VAD-fmk did not prevent the release of Bim and Bmf from the cytoskeleton (Figure 2G), demonstrating that this phenomenon occurs independently of caspase activity.
Active JNK-1 is reportedly sufficient for the release of Bim and Bmf from the cytoskeleton [24]. We have previously shown that JNK-1 is activated already 30 minutes after infection with Neisseria, leading to NFκB activation and proinflammatory responses [33]. Although the short-term effects of JNK-1 activation can be cytoprotective, prolonged JNK activation induces apoptotic cell death [34]. Phosphorylated, active JNK-1 could be detected for the whole period of Ngo infection up to 15 h (Figure 3A), correlating with a reduced electrophoretic mobility of Bim and Bmf at later timepoints (Figure 3B and Figure S8). ERK seems not to be involved in the signaling as there was no activation upon infection (Figure S9). Silencing of JNK-1 with validated siRNAs (Figure 3C) prevented the shift in the size of Bim and Bmf (Figure 3B, 3C, and 3D), indicating a role of JNK-1 in post-translational modification of Bim-L and Bmf in Ngo-infected cells.
The caspase activity of Ngo-infected cells depleted of JNK was reduced to the same level as that of cells subjected to the knockdown of Bim or Bmf (Figure 3E). Moreover, the frequency of cells with apoptotic chromatin condensation was reduced In JNK-1-depleted as compared to control cells (Figure 3F). JNK-1 depletion also partially inhibited the Ngo-induced release of Bim and Bmf from the cytoskeleton (Figure 3G). In addition inhibiting JNK-1 by means of a chemical inhibitor partially reduced a translocation of Bim and Bmf from heavier to lighter fractions in sucrose gradients (Figure S7C). In conclusion, JNK-1 depletion can prevent the post-transcriptional modification of Bim and Bmf, reduce their loss from the cytoskeleton fraction and inhibit apoptosis of Ngo-infected cells.
The Rho-GTPases are central regulators of cytoskeletal changes initiated by extracellular signals. Most prominent, Rho and Rac have been shown to be involved in neisserial uptake and phagocytosis [35]. Therefore, we reanalyzed the link between Rac and JNK signaling during Ngo-induced apoptosis [36]. The knockdown of Rac-1 or its inhibition with the pharmacological agent NSC23766 abolished the reorganization of the cytoskeleton initiated by Ngo infection (Figure 4A). Rac inhibition caused a significant reduction in caspase activation, apoptotic nuclear fragmentation (Figure 4B and 4D), JNK-1 phosphorylation (Figure 4D) and Bim and Bmf cytoskeletal release (Figure S7D). These experiment place Rac upstream of JNK-1 and all JNK-1-dependent apoptotic events affecting Ngo-infected cells.
The role of Bim and Bmf in the activation of Bak and Bax – direct activation or deinhibition? - is still a matter of controversy [8],[15],[17]. Irrespective of their exact of mode of action the function of Bim and Bmf in Ngo-induced apoptosis cannot be redundant because depletion of either of them prevented the induction of apoptosis by Ngo infection. To unravel the mechanisms of this non-redundancy, we assessed the activation of Bak and Bax by means of antibodies that recognize their exposed N-termini and hence their activated conformation. SiRNA- and shRNA-mediated knockdown of Bim as well as siRNA-mediated knockdown of Bmf prior to Ngo infection prevented the activation of both Bak and Bax (Figure 5A and 5B and Figure S4), underlining the essential need of both Bim and Bmf in this pathway.
In certain apoptosis pathways like Ngo infection or cisplatin induction, Bak and Bax become activated in a hierarchical manner, with Bak acting upstream of Bax [32],[37]. The activation of Bak involves its release from antiapoptoticBcl-2 analogues such as Mcl-1 [38]. Combined silencing of Mcl-1 and Bim, but not that of Mcl-1 and Bmf or Mcl-1 knockdown alone reestablished the apoptotic program triggered by Ngo infection (Figure 5C), suggesting that Bim acts as a specific Mcl-1 antagonist in this system. Combined silencing of Bcl-XL plus Bmf, but not that of Bcl-XL and Bim or Bcl-XL alone also reestablished Ngo-induced apoptosis (Figure 5C). In contrast, Bcl-2 co-silencing had no apoptosis-sensitizing effect on either Bim- or Bmf-depleted cells. Potential off target effects within the same protein family could be excluded by systematic cross analysis. In particular, Bim- and Bmf silencing did not cause deregulated expression of anti-apoptotic members of the Bcl-2 family (Figure S10), ruling out an indirect effect of Bim and Bmf depletion by overexpression of apoptosis inhibitors. We concluded from these data that Bim and Bmf activate apoptotic pathways by functionally sequestering Mcl-1 and Bcl-XL, respectively.
Neisseria gonorrhoeae is a highly adapted human pathogen that utilizes multiple adhesins to interact with host cell receptors to trigger cytoskeletal reorganization, invasion or phagocytic uptake, intraphagosomal accommodation, nuclear reprogramming of host cells, cytokine/chemokine release and finally host cell apoptosis [39]. By investigating the apoptotic pathway involved in the infection-induced activation of Bak and Bax, we discovered an unexpected connection between pathogen-induced cytoskeletal reorganization and apoptosis. Attachment of bacteria initiated the activation of Rac-1 leading to rearrangement of the cytoskeleton (which is presumably required for exfoliation) and the activating phosphorylation of the stress kinase JNK-1. JNK-1 then participated in the activation of the BH3-only proteins Bim and Bmf that together facilitate Bak- and Bax-dependent apoptosis.
Besides the well characterized isolate N242 [29], several other clinical isolates induced exfoliation and apoptosis indicating that gonococci trigger similar pathways leading to cell death. Our preliminary data on the initial trigger of cell detachment leading to cell death unveiled a role of specific adhesin – receptor interactions. N242 induced exfoliation and cell death in different cell lines tested. These effects very likely depend on the interaction of one or more of the expressed Opa proteins with a yet uncharacterized receptor. Although derivatives of strain MS11 failed to induce apoptosis in HeLa cells, a similar efficient response as with N242 was observed with derivative N1163 (Opa57; PorBIA) upon infection of HeLa-CEACAM1 but not in HeLa-CEACAM3. Interestingly, CEACAM-1 has been shown to be upregulated in primary ovarian surface epithelial cells by gonococcal infection suggesting that the interaction with this receptor has in vivo relevance [40]. Moreover, the specificity for one CEACAM-recombinant cell line over the other is interesting, because both have been demonstrated to be susceptible for infection with Opa57-expressing gonococci [41]. It is therefore likely that particular adhesin-receptor interactions determine the detachment and apoptosis induction as consequence of this cell – pathogen interaction. This assumption would be in agreement with several reports on the inhibition of apoptosis by gonococcal infection [42]. In one of these studies, Bim was downregulated upon infection of epithelial cells with a piliated gonococcal derivative [43], supporting a central role of Bim in life-death decisions as consequence of gonococcal infections.
Numerous bacterial pathogens induce the reorganization of the host cell cytoskeleton, often initiating the active uptake of bacteria [44]. Nevertheless, the activation of apoptosis is not a common outcome of such bacterial infections. Our data suggest that downstream of cytoskeletal reorganization, the prolonged activation of JNK-1 is required for lethal signaling. It is interesting to note that short term activation of JNK induces antiapoptotic and proinflammatory responses in the host cell infected with Ngo [33],[45]. JNK may therefore exert a dual function during Ngo infection, first by protecting the cell for a short period post-infection and then by triggering the exfoliation of the infected cells.
We show here that JNK was required for Bim- and Bmf-dependent apoptosis during infection, consistent with the previously described JNK-specific phosphorylation of Bim and Bmf within their dynein binding domains [24]. Accordingly, the release of Bim and Bmf from the cytoskeleton as well as their reduced electrophoretic mobility was reduced in JNK-1-depleted cells (Figure 3G).
The exact mode of BH3-only activity is still being discussed. Here we show that both Bim and Bmf are essential to induce Bak and Bax activity for Ngo-triggered apoptosis. As the double knockdown of Bim and Mcl-1 re-sensitized cells for apoptosis, the action of Bmf alone seems to trigger apoptosis efficiently in the absence of Mcl-1. Likewise, apoptosis could be rescued in the absence of Bmf by co-knockdown of Bcl-XL, suggesting that the action of Bim alone suffices to induced apoptosis in the absence of Bcl-XL. In this scenario, both Bim and Bmf need to be activated for efficient induction of cell death due to their joint capacity to inhibit two different anti-apoptotic Bcl-2 homologues (see model in Figure 6). As a result, this study furnishes yet another example for the complex relationship between antagonizing pro- and antiapoptotic Bcl-2 family proteins.
HeLa cells (human cervix carcinoma, later diagnosed as adenocarcinoma) ATCC CCL2 and HeLa cell lines expressing recombinant CEACAM receptors [31] were grown in RPMI 1640 (Gibco) supplemented with 10% heat inactivated FCS in the presence of 5% CO2. The cells were routinely passaged every 2–3 days and the passage number never exceeded 20 passages before a new batch with a low passage number was used. Cells were seeded 24 h before infection and were washed several times with RPMI before infection. HEK 293T cells (immortalized human embryonic kidney) ATCC CRL-11268 for the production of virus and HEp-2 (ATCC CCL-23) were grown in DMEM (Gibco) supplemented with 10% heat inactivated FCS. and ME-180 (HTB-33) were grown in McCoys 5a supplemented with 10% heat inactivated FCS respectively under the same conditions.
The following N. gonorrhoeae strains and derivatives were used in this study: The clinical isolate Ngo strain VP1 (N242; PorBIA; P−; Opa27; Opa27,5;Opa28; Opa29; Opa30; LPS type L1) [29]; Ngo strain MS11 derivatives N302 (PorBIB; P−; Opa−), N920 (PorBIA; P+; Opa−), a PorBIA derivative of N917 (PorBIB; P+; Opa−) and N927 (PorBIA; P−; Opa−) have been described [46],[47]. MS11 derivative N1163 (PorBIA; P−; Opa57) is a PorBIA derivative of strain N313 [41]. Clinical gonococcal isolates from Germany were obtained anonymously from the strain collection of the National Reference Laboratory for Meningococci hosted by the Institute for Hygiene and Microbiology at the University of Würzburg. Species confirmation for those strains was obtained at the Reference Laboratory by standard biochemical tests and partial 16S rRNA sequencing. Gonococci were grown on GC agar base plates (Becton Dickinson, Difco and Remel) supplemented with Proteose Pepton Nr. 3 (Difco) and 1% vitamin mix for 14–20 h at 37°C in 5% CO2 in a humidified atmosphere. Infections were routinely performed in the absence of FCS at a multiplicity of infection (MOI) of 1. If not indicated else wise the respective assays were carried out after 15 h of infection with N242.
For the inhibition of caspases, cells were pre-incubated with 50 µM Z-VAD-fmk (Bachem) for 15 min prior to infection and throughout the experiment. Cisplatin was used at a concentration of 50 µM for 20 h in supplemented media.
5×105 cells per sample were harvested in 100 µl loading buffer (60 mM Tris-HCl pH 8.0, 6% SDS, 10 mM DTT, 6% β-mercaptoethanol, 40% glycerol, and 0.1% bromophenol blue) and 20 µl of the protein lysates were separated and transferred as described before [32]. The following antibodies and sera were used in this study: anti-β-Actin (Sigma); anti-Bad (Cell Signaling); anti-Bak NT (Upstate); anti-Bak (Ab-1) (Millipore);anti-Bax NT (Upstate); anti-Bax (6A7) (BD Pharmingen); anti-Bid (Cell Signaling); anti-Bim (Sigma); anti-Bmf (Cell Signaling); anti-cleaved Caspase-3 (Cell Signalling); anti-JNK-1 (Santa Cruz); anti-pJNK-1 (Cell Signalling) and anti-Mcl-1 (BD Pharmingen). Equal loading was routinely confirmed by appropriate loading controls.
Caspase 3 and 7 activities were measured by CaspACE assay. Control and infected cells were collected and stained with 10 µM CaspACE (Promega) in growth media at 37°C, 5% CO2 for 20 min. After staining, cells were washed twice with PBS and immediately subjected to FACS analysis.
Immunoprecipitation was carried out as described earlier [32]. Solubilized cells were precleared and incubated with 2 µg of anti-Bax (6A7) or anti-Bak (Ab1) antibody at 4°C for 2 h. Immunoprecipitates were collected by incubating with protein G-Sepharose (Amersham) for 2 h. The pellets were washed intensively with lysis buffer and resuspended in sample buffer before analysis by Western blotting using anti-Bax NT and anti-Bak NT antibodies as described above.
5×105 HeLa cells were transfected with 1 µg of siRNA (Quiagen) using RNAiFect transfection kit (Quiagen) according to the manufacturer's instructions. The gene silencing was routinely validated by real time PCR as previously described [48] and by Western blot analysis 72 h post-transfection. The sequence targeted by siBad; siBim, siBid; siBmf; siJNK; siNoxa; siPuma; siMcl-1; siBcl-XL, Bcl-2 and siRac-1 were: ACGAGTTTGTGGACTCCTTTA; CGGAGACGAGTTTAACGCTTA; TAGGGACTATCTATCTTAATA; CACCGGCTTCATGTGCAGCA; AAGAAGCUAAGCCGACCAUUU; TGGGCTATATACAGTCCTCAA; CAGCCTGTAAGATACTGTATA; CGGGACTGGCTAGTTAAACAA; TCCATTATAAGCTGTCGCAGA; ATGCATTTCCTGGAGAATATA and GAGCTTTGAACAGGTAGTGAA respectively.
Stable shRNA-expressing HeLa cells were generated as described before in Kepp et al., 2007. The DNA coding for the shRNA was cloned into pLVTH-M vectors from which it integrated upon viral transfer into the genome of target cells. GFP was used as a marker to select for stable clones. The following sequence was targeted to silence Bim by shRNA expression: TAAGATAACCATTCGTGGG. The efficiency of gene silencing was validated by Western blot analysis and quantitative realtime PCR. The cells transduced with the empty vector were used as controls.
Cells seeded on coverslips were infected for 15 h, fixed in 3.7% PFA and permeabilized using 0.1% Triton X-100. Nonspecific binding was blocked by using 1% goat serum. The samples were stained using anti-actin (Sigma) and anti-tubulin (molecular probes) followed by detection with fluorochrome-coupled secondary antibodies (Jackson Immuno Research) using a Leica confocal microscope with TCS software or a Zeiss immunofluorescence microscope with ACT software. 3dimensional remodeling was performed using Metamorph and Imaris software.
For apoptosis quantification, fixed cells were stained with 1 µg/ml Hoechst 33342 (Invitrogen) for 10 min followed by intense washing with PBS. A minimum of 5 fields per slide was analyzed for chromatin condensation using a Zeiss immunofluorescence microscope.
Control, infected and infected zVAD treated cells were fixed 15 h post-infection with 2.5% glutaraldehyde, post-fixed with 0.5% osmium tetroxide and contrasted using tannic acid and uranyl acetate. Specimens were dehydrated in a graded ethanol series and embedded in Polybed. Ultrathin sections were analyzed in a Leo 906E transmission electron microscope (Leo GmbH).
To analyze cytoskeleton-associated proteins 1×106 cells were incubated for 15 min in HBSS (Gibco). All lipidic membranes were destabilized for 5 min at 4°C by incubation with 5 ml high detergent containing Buffer M (1 mM EGTA, 4% PEG 6000, 100 mM PIPES pH 6.9) containing 0.5% Triton X-100. The cytoplasmic and compartmental proteins containing supernatant was removed and the cytoskeleton was washed with cold Buffer M. The remaining proteins were collected in sample buffer and analyzed by Western blotting.
The averages and standard errors of the mean as well as the t-tests have been calculated using MS Excel. Significance is indicated with ** p < 0,01 and * p < 0,05. If not indicated else wise in the figure legend the data represents at least 3 independent experiments.
|
10.1371/journal.pgen.1007036 | Functional divergence of chloroplast Cpn60α subunits during Arabidopsis embryo development | Chaperonins are a class of molecular chaperones that assist in the folding and assembly of a wide range of substrates. In plants, chloroplast chaperonins are composed of two different types of subunits, Cpn60α and Cpn60β, and duplication of Cpn60α and Cpn60β genes occurs in a high proportion of plants. However, the importance of multiple Cpn60α and Cpn60β genes in plants is poorly understood. In this study, we found that loss-of-function of CPNA2 (AtCpn60α2), a gene encoding the minor Cpn60α subunit in Arabidopsis thaliana, resulted in arrested embryo development at the globular stage, whereas the other AtCpn60α gene encoding the dominant Cpn60α subunit, CPNA1 (AtCpn60α1), mainly affected embryonic cotyledon development at the torpedo stage and thereafter. Further studies demonstrated that CPNA2 can form a functional chaperonin with CPNB2 (AtCpn60β2) and CPNB3 (AtCpn60β3), while the functional partners of CPNA1 are CPNB1 (AtCpn60β1) and CPNB2. We also revealed that the functional chaperonin containing CPNA2 could assist the folding of a specific substrate, KASI (β-ketoacyl-[acyl carrier protein] synthase I), and that the KASI protein level was remarkably reduced due to loss-of-function of CPNA2. Furthermore, the reduction in the KASI protein level was shown to be the possible cause for the arrest of cpna2 embryos. Our findings indicate that the two Cpn60α subunits in Arabidopsis play different roles during embryo development through forming distinct chaperonins with specific AtCpn60β to assist the folding of particular substrates, thus providing novel insights into functional divergence of Cpn60α subunits in plants.
| Chaperonins are large oligomeric complexes that are involved in the folding and assembly of numerous proteins in various species. In contrast to other types of chaperonins, chloroplast chaperonins are characterized by the hetero-oligomeric structure composed of two unique types of subunits, Cpn60α and Cpn60β, each of which is present in two or more paralogous forms in most of higher plants. However, the functional significance underlying the wide array of subunit types and complex oligomeric arrangement remains largely unknown. Here, we investigated the role of the minor Cpn60α subunit AtCpn60α2 in Arabidopsis embryo development, and found that AtCpn60α2 is important for the transition of globular embryos to heart-shaped embryos, whereas loss of the dominant Cpn60α subunit AtCpn60α1 affects embryonic cotyledon development. Further studies demonstrated that AtCpn60α2 could form functional chaperonins with AtCpn60β2 and AtCpn60β3 to specifically assist in folding of the substrate KASI, which is important for the formation of heart-shaped embryos. Our results suggest that duplication of Cpn60α genes in higher plants can increase the potential number of chloroplast chaperonin substrates and provide chloroplast chaperonins with more roles in plant growth and development, thus revealing the relationship between duplication and functional specialization of chaperonin genes.
| Chaperonins are a class of molecular chaperones that are characterized by a barrel-shaped architecture formed by two stacked oligomeric rings consisting of several subunits of approximately 60 kDa. Two types of chaperonins have been identified: type I chaperonins are found in eubacteria, chloroplasts, and mitochondria; and type II exist in archaea and eukaryotic cytosol. The main difference between them is that type I chaperonins require co-chaperonins consisting of seven 10 kDa subunits for substrate encapsulation, whereas type II chaperonins have a built-in lid that plays the same role [1]. The GroEL/GroES complex in Escherichia coil has been studied extensively as the prototype of type I chaperonins. GroEL is a homo-oligomer which consists of two stacked heptameric rings. The folding cycle in GroEL/GroES has been surveyed in detail, and the canonical view suggested that the complex operates through the asymmetric “bullet” cycle. In this asymmetric cycle, unfolded/misfolded substrates first bind to the hydrophobic cavity lining of one ring (cis ring), utilizing the exposed hydrophobic residues, and then the binding of ATP causes large conformational changes of the cis ring that further trigger the binding of co-chaperonins. Binding of co-chaperonins initiates further conformational changes and caps the cis ring, and consequently encapsulates the substrates into an expanded cavity with a hydrophilic lining, which assists the substrates to refold into their native states. Following the hydrolysis of ATP in the cis ring, ATP and other non-native proteins bind to the opposite ring (trans ring), resulting in the dissociation of refolded substrates, ADP, and co-chaperonins in the cis ring [2–4]. Moreover, an alternative model, known as the symmetric “football” model, was recently also proposed. In this model, the exchange of ADP to ATP is extremely rapid in the presence of abundant substrate protein, resulting in formation of a symmetric “football” intermediate that has GroES bound to both rings and can assist in protein folding simultaneously in both rings. This “football” intermediate would be reverted to the “bullet” conformation upon ATP hydrolysis [5–7]. In addition, although Escherichia coil only contains one chaperonin gene, a survey of 669 complete bacterial genomes showed that nearly 30% contain two or more chaperonin genes, and a degree of subfunctionalization has occurred in the chaperonin subunits encoded by these duplicated genes [8]. Moreover, in the previous study, Wang and coworkers found that some specific mutations of GroEL can improve the folding of GFP, but the mutated GroEL has a reduced ability to function as general chaperones, suggesting a conflict between the increased ability of GroEL to fold particular substrates and its general ability to fold a wide range of substrates, and this conflict would be resolved by duplication and variation of chaperonin genes [9].
Chloroplast chaperonin (Ch-Cpn60) was first found as a homolog of GroEL that could bind to the chloroplast Rubisco large subunit and assist the assembly of Rubisco, a key rate-limiting enzyme in the process of carbon dioxide fixation [10]. In contrast to GroEL, ch-Cpn60s contain two different types of subunits, Cpn60α and Cpn60β, which only share approximately 50% identity. Ch-Cpn60s composed of Cpn60α and Cpn60β are considered to be the native form of chloroplast chaperonins in vivo, because ch-Cpn60s purified from Pisum sativum, Brassica napus, Arabidopsis thaliana and Spinacia oleracea were all shown to be hetero-oligomers consisting of nearly equal amounts of Cpn60α and Cpn60β [11–13]. In Arabidopsis thaliana, there are two Cpn60α genes and four Cpn60β genes, which encode three dominant subunits: AtCpn60α1, AtCpn60β1 and AtCpn60β2; and three minor subunits: AtCpn60α2, AtCpn60β3 and AtCpn60β4 (the nomenclature used in this article is in accordance with The Arabidopsis Information Resource database). Among them, AtCpn60α1 and AtCpn60α2 share only about 57% identity, and AtCpn60β1/2/3 share 90%-95% identity, while AtCpn60β4 is only 60% identical to the other AtCpn60β subunits [14–15]. AtCpn60α1 was the first chaperonin gene studied in detail, and its mutant, schlepperless (slp), showed retardation of embryo development before the heart stage, and defective embryos with highly reduced cotyledons [16]. Then a T-DNA insertion mutant of AtCpn60α2, emb3007, showed the embryo development arrested at the globular stage in the SeedGenes database (http://www.seedgenes.org/), suggesting that AtCpn60α2 is also possibly an embryo-defective gene [17–18]. A T-DNA mutant lacking the AtCpn60β1 transcript, len1, had impaired leaves and showed systemic acquired resistance (SAR) under short-day condition [19]. It was also reported that a weak mutant of AtCpn60α1 and a strong mutant allele of AtCpn60β1 both showed impaired chloroplast division and reduced chlorophyll levels, and the AtCpn60β1 AtCpn60β2 double mutant led to an albino seedling similar to slp, suggesting that AtCpn60β1 and AtCpn60β2 are redundantly required for normal chloroplast function, together with AtCpn60α1 [20]. In addition, a recent report also showed that the ch-Cpn60 containing the AtCpn60β4 subunit played a specific role in the folding of NdhH, a subunit of the chloroplast NADH dehydrogenase-like complex (NDH), indicating that the particular type of AtCpn60β subunit could contribute to the folding of some specific substrates [21].
Embryogenesis is the beginning of plant development. During Arabidopsis embryo development, chloroplast biogenesis is a temporary process. Proplastids in the whole embryo first begin to differentiate into chloroplasts at the transition stage, and then the mature chloroplasts degenerate to undifferentiated eoplasts during seed maturation [22–24]. For decades, through forward and reverse genetic screens in Arabidopsis, numerous chloroplast proteins crucial for embryo development were discovered. Interestingly, nearly all embryo defects caused by chloroplast dysfunction displayed premature arrest at the globular stage, indicating that the formation of impermanent chloroplasts in Arabidopsis embryos is especially crucial for the transition of globular embryos to heart-shaped embryos [25–27].
Here, we provided genetic evidence to show that the two AtCpn60α genes in Arabidopsis affect the embryonic development at different stages. Further studies revealed that CPNA2 could form a functional chaperonin with AtCpn60β2 and AtCpn60β3 subunits to specifically assist the folding of KASI (β-ketoacyl-[acyl carrier protein] synthase I), and KASI could not be folded by the functional chaperonin containing CPNA1. Moreover, we found that the KASI protein level was largely reduced due to loss-of-function of CPNA2, and the reduction in KASI protein level possibly caused the abnormality of the cpna2 embryos. Our results showed that Cpn60α2 and Cpn60α1 in Arabidopsis can form functional chaperonin complexes with specific AtCpn60β subunits to assist the folding of particular substrates, indicating that functional divergence of chloroplast Cpn60α subunits has occurred in higher plants.
To elucidate the molecular mechanisms that control embryo development in Arabidopsis, we ordered the stock CS76507, which is a set of 10,000 T-DNA lines, from Arabidopsis Biological Resource Center (ABRC, http://abrc.osu.edu/). From the stock, we obtained a mutant that showed obvious seed abortion, with a frequency of 25.55% (n = 2270). We found that the T-DNA insertion of this mutant is located in the first exon of AT5G18820 (Fig 1A) using thermal asymmetric interlaced PCR [28]. We named the gene CPNA2 because it encodes the chaperonin subunit AtCpn60α2, and named the mutant cpna2-2 (The emb3007 mutant described previously was designated as cpna2-1 here). PCR analysis of cpna2-2/+ progeny showed that no homozygous mutant plant existed, and the ratio of heterozygote to wild type was nearly 2:1 (S1 Table). Reciprocal crosses between heterozygote and wild-type plants further demonstrated that the transmission efficiency of gametophytes was not affected by loss-of-function of CPNA2 (S1 Table). Moreover, we obtained another T-DNA mutant (cpna2-3, SALK_144574) from ABRC, and found that the mutant cpna2-3/+ also showed seed abortion, with a frequency of 26.09% (n = 1196) (Fig 1A and 1B).
To further confirm that CPNA2 is responsible for seed abortion in cpna2-2/+ and cpna2-3/+ plants, we performed a complementation test. A genomic fragment, including CPNA2, 1548 bp upstream of the start codon and 673 bp downstream of CPNA2, was introduced into cpna2-2/+ and cpna2-3/+. The result showed that the fertility was restored in the siliques of CPNA2pro:gCPNA2 transgenic plants (Fig 1B). Together, these results indicated that seed abortion in Arabidopsis could be caused by loss-of-function of CPNA2.
To investigate the cause of seed abortion of cpna2-2/+ and cpna2-3/+, we first examined the processes of embryo development in the seeds of cpna2-2/+ and cpna2-3/+ using the whole mount clearing technique. We could not distinguish abnormal embryos in the siliques from the zygote stage to the globular stage. However, when most embryos in the seeds of cpna2-2/+ and cpna2-3/+ reached the transition stage, some embryos showed the irregular globular shape and the start of abnormal cell division (Fig 1C). While wild-type embryos progressed into further stages, the abnormal embryos still stayed at the globular stage, and finally degraded along with the collapse of seeds (Fig 1C). The phenotype of these abnormal embryos is consistent with the mutant emb3007 in SeedGenes (http://www.seedgenes.org/; [17–18]). In addition, since CPNA2 was predicted to encode a chloroplast chaperonin subunit and chloroplast chaperonins had been reported to be involved in the folding and assembly of many chloroplast proteins [29], we wondered whether the abortion of cpna2 embryos was due to impaired chloroplast development. To investigate this, we first examined the subcellular localization of CPNA2 in mesophyll protoplasts isolated from transgenic plants carrying the 35Spro:CPNA2-GFP construct. As shown in Fig 2A, the protoplasts containing CPNA2-GFP fusion protein displayed GFP signals that overlapped with chlorophyll autofluorescence, confirming that CPNA2 is located in chloroplasts. Then we observed the ultrastructure of chloroplasts using transmission electron microscopy. In the 6 DAP siliques of cpna2-2/+ plant, we observed that mature chloroplasts in wild-type embryos contained organized thylakoid membranes stacked into grana (Fig 2B and 2C). In contrast, only abnormal chloroplasts that lacked thylakoid membranes and contained a deeply stained mass were found in cpna2-2 mutant embryos (Fig 2D and 2E). Collectively, these results suggested that loss-of-function of CPNA2 impeded the process by which proplastids differentiate into mature chloroplasts during embryo development, thereby causing the arrest of the cpna2 embryos and seed abortion of cpna2-2/+ and cpna2-3/+.
It had been reported that CPNA2 has an extremely low signal in all tissues and developmental stages using the Genevestigator program, and the CPNA2 protein could not be detected in proteomics studies [14,30]. To further investigate the expression pattern of CPNA2, we examined CPNA2 transcript levels in different Arabidopsis tissues using quantitative real-time PCR (qRT-PCR). The result demonstrated that CPNA2 is expressed in all tissues and is especially highly expressed in the 5 DAP (day after pollination) siliques (Fig 3A). Then, we performed a GUS assay in the transgenic plants carrying the CPNA2pro:GUS construct, and observed a very strong signal in the SAM and a weaker signal in vascular bundles of 7 DAG (day after germination) seedlings, which largely declined in 14 DAG seedlings (Fig 3B and 3C). No signal was found in mature leaves, flowers, inflorescences and siliques, possibly due to low abundance and dispersion of GUS signals.
To investigate the expression pattern of CPNA2 during embryo development, we obtained transgenic plants carrying the CPNA2pro:H2B-GFP construct, and observed the fluorescent signal in the dissected embryos. No GFP (green fluorescent protein) signal was detected in the globular embryos, but GFP signals began to sporadically appear in the transition stage embryos (Fig 3D). When the embryos reached the heart stage, fluorescent signals were located on the adaxial sides of cotyledons (Fig 3D). Next, the signals were mainly detected in cotyledons of the torpedo and cotyledon stages, and were still strongest on the adaxial sides of cotyledons (Fig 3D). Together, these results showed that CPNA2 is highly expressed in the SAM of early seedlings and embryonic cotyledons.
Hill and Hemmingsen reported that CPNA2 is the paralog of CPNA1 [15], thus it is possible that they have redundant functions. However, a CPNA1 mutant, slp (schlepperless), showed an embryo-defective phenotype that mainly appeared at the heart stage and thereafter, which is different from the cpna2 mutants [16]. To confirm the previous study, we obtained another AtCpn60α1 mutant (cpna1, SALK_006606) (S1A Fig), and then observed embryo development in the siliques of cpna1/+. As expected, nearly a quarter of the examined embryos (25.71%, n = 579) showed the abnormal phenotype. The abnormal embryos had highly reduced cotyledons, a larger angle between cotyledons, and developed more slowly from the torpedo stage (S1B Fig), which was similar to the phenotype of the slp embryos.
As shown above, cpna1 and cpna2 mutants had very different embryo-defective phenotypes, which could be caused by functional divergence and/or different expression patterns of the two genes. To investigate the expression pattern of CPNA1 during embryo development, we obtained transgenic plants carrying the CPNA1pro:H2B-CFP construct, and observed CFP (cyan fluorescent protein) signals in the dissected embryos. The fluorescent signal was originally detected in protoderm cells at the globular stage, and then concentrated in the SAM at the transition stage (Fig 3E). When the embryos developed into heart, torpedo, and cotyledon stages, most CFP signals were specifically redistributed on the adaxial sides of cotyledons (Fig 3E).
As the above results demonstrated, CPNA1 has a similar expression pattern to CPNA2 during embryo development, while it is expressed more widely. This indicates that the different embryo-defective phenotypes of the cpna1 and cpna2 mutants are likely to be caused by functional divergence of the two genes, but not due to differences in expression pattern. Moreover, these findings also suggested that CPNA1 mainly plays a role at the torpedo stage and thereafter, whereas CPNA2 is crucial to reach the heart stage for Arabidopsis embryos.
Ch-Cpn60s had been considered to be hetero-oligomers consisting of equal amounts of Cpn60α and Cpn60β, based on several studies conducted in Pisum sativum, Brassica napus, Arabidopsis thaliana, and Spinacia oleracea [11, 12, 13, 31]. Moreover, because CPNA2 and CPNA1 could play different roles during embryo development, we wondered which AtCpn60β subunits could interact with CPNA2 or CPNA1 to form specific chaperonins. Using AtPID (Arabidopsis thaliana Protein Interactome Database) [32], we first predicted the functional partners of CPNA2 and CPNA1. The result showed that AtCpn60β1 and AtCpn60β2 had much higher scores than AtCpn60β3 and AtCpn60β4 among the predicted functional partners of CPNA1 (S2 Table). In contrast, AtCpn60β3 and AtCpn60β2 were the top two predicted functional partners of CPNA2, and AtCpn60β3 had a far higher score than the other candidates (S2 Table). These results implied that CPNA2 and CPNA1 could possibly interact with different AtCpn60β subunits to form specific functional chaperonins.
To further clarify the functional partners of CPNA2 and CPNA1, we obtained T-DNA insertion mutants of AtCpn60β1 (CPNB1), AtCpn60β2 (CPNB2), AtCpn60β3 (CPNB3), and AtCpn60β4 (CPNB4) from ABRC (S2A Fig). Homozygous mutant plants of all AtCpn60β genes could be obtained and had normal fertility. Through reverse transcription PCR (RT-PCR) analysis, we also confirmed the complete loss of the corresponding transcripts in the cpnb2, cpnb3 and cpnb4 mutants, and the enormous reduction of the CPNB1 transcript in the cpnb1 mutant (S2B Fig). We then crossed these mutants pairwise to obtain double heterozygous plants. In all double heterozygous plants, we only observed aberrant seeds in siliques of the cpnb1/+ cpnb2/+ and cpnb2/+ cpnb3/+ plants (S2C Fig), implying that the cpnb1 cpnb2 and cpnb2 cpnb3 double homozygous embryos were possibly abnormal.
We also obtained cpnb1/+ cpnb2 and cpnb2 cpnb3/+ plants in the self-crossed progenies of the cpnb1/+ cpnb2/+ and cpnb2/+ cpnb3/+ mutants, respectively. As expected, siliques of the cpnb1/+ cpnb2 and cpnb2 cpnb3/+ plants contained numerous aberrant seeds, at a frequency of 25.43% (n = 1050) and 25.64% (n = 1166), respectively, suggesting that the cpnb1 cpnb2 and cpnb2 cpnb3 embryos were probably abnormal. To clarify the cause for the abnormality of the cpnb1 cpnb2 and cpnb2 cpnb3 embryos, we examined the developmental processes of these double homozygous embryos in the aberrant seeds. In siliques of the cpnb1/+ cpnb2 plants, the cpnb1 cpnb2 embryos had a similar shape to wild-type embryos at the heart stage, apart from slightly smaller cotyledons and a larger angle between cotyledons. However, the cpnb1 cpnb2 embryos began to display highly reduced cotyledons at the torpedo stage and thereafter compared to the wild-type embryos (Fig 4A). In contrast, we observed that the majority of the cpnb2 cpnb3 embryos (70.61%, n = 228) were arrested at the globular stage in the siliques of the cpnb2 cpnb3/+ plants, while the other cpnb2 cpnb3 embryos showed various phenotypes, including remarkable retardation of embryo development, and variation of cotyledon shape and number (Fig 4B).
Overall, we found that cpnb1 cpnb2 embryos had a very similar phenotype to cpna1 embryos, echoing the previous study that showed that cpnb1 cpnb2 seedlings were analogous to cpna1 seedlings [20]. In addition, although the penetrance was incomplete possibly due to partial complement of CPNB1, the majority of the cpnb2 cpnb3 embryos were shown to phenocopy the cpna2 embryos. Combined with the prediction from AtPID (S2 Table), this result further suggested that CPNA1 is likely to form a functional chaperonin with CPNB1 and CPNB2 during Arabidopsis embryo development, whereas CPNB2 and CPNB3 were the functional partners of CPNA2. Moreover, since the cpnb1/+ cpnb3/+ plants showed normal seed development, it seemed that CPNB2 was a versatile AtCpn60β subunit which could form functional chaperonins with both CPNA1 and CPNA2, and was sufficient to support embryo development alone.
As the previous results demonstrated, we found that CPNA2 and CPNA1 have nonredundant functions during embryo development. Moreover, it had been widely reported that some chaperonins containing specific subunits had unique substrates and played an important role under certain circumstances [21, 33, 34, 35]. Based on these results, it is possible that the chaperonin containing CPNA2 has some specific substrates that cannot be folded by the chaperonin containing CPNA1, thus affecting the specific developmental process of Arabidopsis embryos.
To examine this possibility, we first introduced two chimeric genes encoding an HA (influenza hemagglutinin protein epitope) tag fused to the C-terminus of CPNA2 or CPNA1 into cpna2-2/+ and cpna1/+ mutants, respectively. The CPNA2pro:CPNA2-HA and CPNA1pro:CPNA1-HA constructs fully restored the fertility of cpna2-2/+ and cpna1/+ plants, respectively (Fig 5A), indicating that HA-tag has no effect on the functions of CPNA2 and CPNA1. Since expression of the CPNA2 gene in vegetative tissues is very low [14], we also obtained transgenic lines carrying the 35Spro:CPNA2-HA or 35Spro:CPNA1-HA construct, and conducted Co-IP assay in 7 DAG transgenic and wild-type seedlings using the μMACS HA isolation kit (Miltenyi Biotec). The immunoprecipitates were separated by SDS-PAGE and then analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) (Fig 5B). The MS analysis showed that KASI, a protein involved in de novo fatty acid synthesis, had more peptides and much higher scores in CPNA2 immunoprecipitation fractions than in CPNA1 and WT immunoprecipitation fractions (Table 1 and S1 Data). These results suggested that KASI was a possible specific substrate of the chaperonin containing CPNA2. Moreover, the MS data also showed that CPNB3 was not detected in CPNA1 immunoprecipitation fractions, but had high scores in CPNA2 immunoprecipitation fractions (Table 1), echoing the previous genetic results demonstrating that CPNB3 is a functional partner of CPNA2 but not CPNA1.
To further confirm that KASI is a specific substrate of the chaperonin containing CPNA2, we also performed a proteinase K protection assay. This assay takes advantage of the formation of a highly stable cis-ternary complex consisting of substrate, chaperonin and co-chaperonin in the presence of ADP [36, 37]. This cis-ternary complex could sequester a substrate into the cavity of a chaperonin, thus protecting the substrate from digestion by proteinase K. Recombinant CPNA1, CPNA2, CPNB1, CPNB2, CPNB3, Cpn20 (a co-chaperonin subunit in Arabidopsis), and KASI proteins were overexpressed in Escherichia coli and then purified on Ni-NTA agarose resin. Subsequently, the various chaperonins consisting of AtCpn60α and AtCpn60β subunits were reconstituted according to previously described method [31]. After purification by gel-filtration chromatography, the reconstituted chaperonins were used in proteinase K protection assays of denatured KASI protein. As shown in Fig 5C, 5D and 5E, KASI was digested by proteinase K in the presence of Cpn60A1-B1, Cpn60A1-B2 or Cpn60A1-B3, indicating that KASI was not likely to form stable cis-ternary complexes with Cpn20 and chaperonins containing CPNA1. In contrast, in the presence of Cpn60A2-B2 or Cpn60A2-B3, almost all of the KASI protein was protected from digestion by proteinase K (Fig 5G and 5H), which was likely due to the formation of the stable cis-ternary complexes. Taken together, these results showed that denatured KASI protein could only be captured and refolded by Cpn60A2-B2 and Cpn60A2-B3, but not by chaperonins containing CPNA1, confirming that KASI is a specific substrate of the functional chaperonins containing CPNA2.
In addition, as shown in Table 1, CPNB1 had very high scores in both CPNA2 immunoprecipitation fractions, indicating that CPNB1 could likely interact with CPNA2. To know whether CPNB1 was a functional partner of CPNA2, we also reconstituted the chaperonin consisting of CPNA2 and CPNB1 (Cpn60A2-B1) and performed a proteinase K protection assay of denatured KASI using Cpn60A2-B1. As shown in Fig 5F, KASI was not protected from digestion by proteinase K in the presence of Cpn60A2-B1, indicating that Cpn60A2-B1 was not a functional chaperonin that can assist the folding of KASI. This result showed that although CPNB1 could interact with CPNA2 when the CPNA2 protein was overexpressed in vivo, it could not form a full-functional chaperonin with CPNA2 alone, providing further evidence supporting the above genetic findings that suggested CPNB1 was not a functional partner of CPNA2.
As demonstrated above, KASI is a specific substrate of the functional chaperonin containing CPNA2. To know whether the level of KASI protein was reduced in the cpna2 homozygous mutant, we first rescued the development of cpna2-2 homozygous embryos using a ABI3pro:CPNA2-HA construct. Since the ABI3 gene is specifically expressed in seeds [38, 39], we obtained cpna2-2 homozygous seedlings from ABI3pro:CPNA2-HA transgenic lines (Fig 6A). The abnormal seedlings had the white cotyledons and could not develop true leaves even on the fourteenth day after germination (Fig 6A). Genotypic analysis of the abnormal seedlings also confirmed that they were cpna2-2 homozygous mutants partially rescued by the ABI3pro:CPNA2-HA construct (Fig 5B). Then we examined KASI protein levels in 7 and 14 DAG seedlings of WT and cpna2-2 by immunoblotting. As shown in Fig 6C and 6D, KASI protein levels in 7 and 14 DAG seedlings of cpna2-2 were reduced to approximately one-tenth of the level in contemporaneous WT seedlings. This result indicated that the KASI protein level could be largely reduced due to loss-of-function of CPNA2 in vivo.
In addition, Wu and Xue reported that KASI was crucial for embryo development and KASI deficiency resulted in disrupted embryo development before the globular stage [40]. Therefore, it is possible that the decline of KASI protein level in cpna2 embryos causes abortion of the mutant embryos. To examine this possibility, we constructed a CPNA2pro:amiR-KASI vector to specifically reduce the expression level of KASI in the embryos of transgenic lines at the transition stage and thereafter. In T1 generation transgenic plants, we chose three lines for follow-up studies. As shown in Fig 7A, we found that the expression levels of KASI in lines 18 and 23 were reduced by almost half, whereas the expression level of KASI in line 16 remained unchanged. Then we carefully examined embryo development in ovules of wild type, line 16, line 18, and line 23. In the ovules of line 16, embryos developed similarly to wild type (Fig 7B). In contrast, when wild-type embryos reached the heart stage, almost all the embryos of line 18 and line 23 still stayed at the globular stage, and reached the heart stage when normal embryos entered the torpedo stage (Fig 7B). Embryos in the ovules of line 18 and line 23 ultimately could reach the cotyledon stage (Fig 7B), and the fertility of line 18 and line 23 plants was not affected. Moreover, we also counted the percentages of all embryonic morphologies in the 3, 4, 5, and 7 DAP siliques of wild type and transgenic lines (S3 Table), further confirming the delayed embryo development in lines 18 and 23, consistent with the morphological observations.
By observing the embryo development of the KASI knock-down transgenic lines, we found that reduction of the expression level of KASI clearly delayed the process by which globular embryos develop into heart-shaped embryos. This finding suggested that the KASI level is crucial to reach the heart stage for Arabidopsis embryos, implying that abnormality of the cpna2 embryos is likely to be caused by a decrease of correctly folded KASI protein in the mutant embryos.
To investigate the evolutionary relationship of Cpn60α1 and Cpn60α2, we obtained protein sequences of CPNA1 and CPNA2, excluding the transit peptides predicted by TargetP [41], and then searched for homologous proteins in various species using BLAST (Basic Local Alignment Search Tool, http://blast.ncbi.nlm.nih.gov/Blast.cgi). By sequence alignment, we constructed the phylogenetic tree of Cpn60α1 and Cpn60α2. The tree showed that Cpn60α1 orthologs exist in monocotyledons, dicotyledons, gymnosperms (Picea sitchensis), bryophytes (Physcomitrella patens) and algae (Chlamydomonas reinhardtii), whereas Cpn60α2 orthologs form a separate cluster and only exist in monocotyledons and dicotyledons (S3 Fig). The result suggested that Cpn60α1 is more primitive than Cpn60α2, and Cpn60α2 probably originated from gene duplication and variation of Cpn60α1 in angiosperms.
As the above results showed, the functional chaperonin containing CPNA2 could specifically assist in the folding of KASI and could play a unique role during Arabidopsis embryo development. To clarify the structural basis of functional specialization of CPNA2, we first analyzed the three-dimensional (3-D) structures of GroEL, CPNA1, and CPNA2 through homology modeling. As shown in Fig 8A, compared with GroEL and CPNA1, CPNA2 lacked strands 2 and 3. To confirm the importance of strands 2 and 3, we also did homology modeling of the CPNA2 orthologs in Brassica napus, Vitis vinifera, and Morus notabilis. Surprisingly, these orthologs all possess strands 2 and 3 (Fig 8B), suggesting that the lack of strands 2 and 3 is not likely to be necessary for functional specialization of CPNA2. Moreover, Ile 150 (I150) and Asp 398 (D398), which are crucial for ATP/ADP binding in GroEL, are also highly conserved in CPNA1 and CPNA2 (Fig 8A).
In previous studies, a few positions in the apical domain of GroEL had been proposed as substrate binding sites, which are highly conserved in many species [42–44]. To know whether these positions led to functional specialization of CPNA2, we first examined the corresponding positions in the homologous proteins of CPNA1 and CPNA2 by protein sequence alignment (S4 Fig). The result showed that the highly conserved hydrophobic residue Ala 259 in the orthologs of CPNA1 is converted to a hydrophilic residue Glu or Ser in the orthologs of CPNA2, while the conserved positively charged residue Arg 267 in Cpn60α1 is mostly replaced by an uncharged residue Gln or Asn in Cpn60α2 (S4 Fig). Furthermore, Glu 259 and Gln 267 of the CPNA2 protein are also unique in all the Arabidopsis ch-Cpn60 subunits (Fig 8C), implying that the two positions are likely the cause for functional specialization of CPNA2. To confirm this conjecture, Glu 259 and Gln 267 of CPNA2 were converted to Ala 259 and Arg 267, respectively, through site-directed mutagenesis, and then a CPNA2pro:CPNA2, CPNA2pro:CPNA2E259A, CPNA2pro:CPNA2Q267R or CPNA2pro:CPNA2E259A/Q267R construct was introduced into cpna2-2/+ plants. Unexpectedly, cpna2-2/+ plants carrying the CPNA2pro:CPNA2, CPNA2pro:CPNA2E259A, CPNA2pro:CPNA2Q267R or CPNA2pro:CPNA2E259A/Q267R construct all had normal fertility (Fig 8D), suggesting that the two conserved residues in the orthologs of CPNA2 are not crucial for the functional specialization of CPNA2. These results indicated that CPNA2 is likely to utilize a few new positions to bind KASI, and that detailed structural information of the chaperonin containing CPNA2 is required to further elucidate the mechanism of KASI folding.
Chloroplast chaperonins are composed of two types of chaperonin subunits, Cpn60α and Cpn60β, which is different from the chaperonins in bacteria and mitochondria. Several previous studies demonstrated that ch-Cpn60s consisting of nearly equal amounts of Cpn60α and Cpn60β were the native form of chloroplast chaperonins in vivo, although the Cpn60β subunit could also form the chaperonin complex in reconstitution experiments alone [11, 12, 13, 31]. Among the four Cpn60β subunits in Arabidopsis, AtCpn60β1, AtCpn60β2, and AtCpn60β3 have more than 90% identity, while AtCpn60β4 shares only 60% identity with the other three AtCpn60β subunits. However, the homo-oligomers reconstituted with AtCpn60β1, AtCpn60β2 or AtCpn60β3 have unique physicochemical properties, different preferences for various co-chaperonins, and distinct abilities of folding substrates, implying the functional divergence of Cpn60β1/2/3 in Arabidopsis [45].
In this study, we analyzed the phenotypes of different combinations of AtCpn60β double mutants, and found that cpnb1 cpnb2 and cpnb2 cpnb3 double mutant embryos phenocopied cpna1 and cpna2 embryos, respectively. This finding suggested that CPNA1 plays a role in embryo development together with CPNB1 and CPNB2, while CPNA2 can function with CPNB2 and CPNB3. Moreover, we found that CPNB3 could not be detected in the CPNA1 immunoprecipitation fractions, whereas it was abundant in the CPNA2 immunoprecipitation fractions (Table 1). This result indicated that CPNA1 had far lower affinity for CPNB3 compared with CPNA2, further confirming that CPNB3 is the functional partner of CPNA2 but not CPNA1, consistent with the genetic results. In addition, we found that the chaperonin complex reconstituted with CPNA2 and CPNB1 could not protect KASI from proteinase K (Fig 5F), thus suggesting that CPNB1 is the functional partner of CPNA1 but not CPNA2, consistent with the genetic results. These findings provided evidence that different AtCpn60α subunits could bind specific AtCpn60β subunits as their functional partners, indicating the functional divergence of Cpn60α subunits in Arabidopsis.
Moreover, we also found that although CPNB1 and CPNA2 do not appear to function together in the folding of KASI, CPNB1 was abundant in the CPNA2 immunoprecipitation fractions (Table 1). It was also reported that AtCpn60β1/2/3 and AtCpn60α1 usually formed the native chaperonin together in vivo [21, 30], even though we did not detect CPNB3 in CPNA1 immunoprecipitation fractions possibly due to low affinity of CPNB3 for CPNA1. These results suggested that all the AtCpn60β1/2/3 subunits are usually mixed into native chaperonins containing the Cpn60α subunit in vivo, although they have different affinity for specific Cpn60α subunit. Moreover, although it had been reported that the native chaperonin containing Cpn60β4 in Arabidopsis is composed of seven Cpn60α1, two Cpn60β4, and five Cpn60β1/2/3 [21], the exact proportions of AtCpn60β1, AtCpn60β2, and AtCpn60β3 in native chaperonins are difficult to determine due to similar molecular weights and high identity of AtCpn60β1/2/3. Further study is still needed to clarify the stoichiometry of subunits in native ch-Cpn60s of Arabidopsis.
Although multiple chaperonin genes are present in a high proportion of prokaryotes and eukaryotes, the biological significance of duplication and variation of chaperonin genes has yet to be fully elucidated. Recently, a study on the type II chaperonin of Sulfolobales showed that three different chaperonin subunits (α, β, γ) could form three types of chaperonins at different temperatures, and specific chaperonins could fold a distinct range of substrates to adapt to environmental changes [33]. Moreover, it was also reported that GroEL1 in Mycobacterium smegmatis specifically interacts with KasA (a key component of type II Fatty Acid Synthesis) to affect mycolic acid synthesis and biofilm formation, whereas GroEL2 provides the housekeeping chaperone function [35]. In the field of chloroplast chaperonins, Zhang and coworkers recently determined the crystal structure of the apical domains of Cpn60α and Cpn60β1 in Chlamydomonas reinhardtii, and elucidated the structural basis for why Cpn60α and Cpn60β subunits have different affinity for substrates and co-chaperonins [46]. Additionally, in line with the divergence of protein sequence, the Cpn60β4 subunit in Arabidopsis has a unique structure and the chaperonin containing Cpn60β4 could specifically assist the folding of NdhH [21]. These findings revealed that duplication and variation of chaperonin genes could extend the function of chaperonins in various species.
Here, we found that KASI, a protein involved in de novo fatty acid synthesis, was far more abundant in CPNA2 immunoprecipitation fractions than in CPNA1 immunoprecipitation fractions, implying that KASI was likely to be a specific substrate of the chaperonin containing CPNA2. To confirm this conjecture, we conducted the proteinase K protection assay of KASI in the presence of different chaperonin complexes. It was shown that both Cpn60A2-B2 and Cpn60A2-B3 could perfectly protect KASI from digestion by proteinase K, whereas all the chaperonins containing CPNA1 could not protect KASI. This result further showed that the functional chaperonins containing CPNA2 could specifically assist in the folding of KASI, suggesting that CPNA2, a minor Cpn60α subunit, has a unique function in Arabidopsis. Moreover, we also examined two conserved positions proposed as substrate binding sites in the orthologs of CPNA2, and found that they were not responsible for the functional specialization of CPNA2. Hence, detailed structural analysis of the chaperonin containing CPNA2 is required to further elucidate the mechanism of KASI folding.
Additionally, it was reported that chaperonins in various species had a wide range of substrates [1], therefore we cannot exclude the possibility that the chaperonin containing CPNA2 has other specific substrates in addition to KASI. Moreover, since we conducted Co-IP assay in 7 DAG seedlings but not in embryos due to technology limitations, it is possible that there are some unknown specific substrates of the chaperonin containing CPNA2 that only exist in embryos. The detection of more specific substrates would further contribute to the functional elucidation of CPNA2 in Arabidopsis.
A number of genes involved in de novo fatty acid synthesis are essential for early embryo development in Arabidopsis. GURKE, a gene encoding the acetyl-CoA carboxylase ACC1, is required for partitioning the apical part of globular embryos in Arabidopsis [47], and loss-of-function of CAC1A, a gene encoding the biotin carboxyl-carrier protein BCCP1, obviously delayed embryo development from the early globular stage [48]. Moreover, KASI deficiency was also found to result in arrested development of most kasI embryos before the globular stage, and delayed development of few kasI embryos [40]. Additionally, in Arabidopsis microarray data sets [49], we found that the expression level of KASI has a dramatic increase when embryos reach the heart stage, implying that KASI is likely to be crucial for the transition of globular embryos to heart-shaped embryos. To confirm this conjecture, we specifically reduced the expression level of KASI in the embryos at the transition stage and thereafter by transforming the CPNA2pro:amiR-KASI vector. In the transgenic lines, we observed that the process of globular embryos reaching the heart stage was delayed, confirming that KASI plays an important role in the formation of heart-shaped embryos.
Since cpna2 embryos are arrested at the globular stage and loss-of-function of CPNA2 could result in a significant decrease in the KASI protein level that is crucial for the transition of globular embryos to heart-shaped embryos (Figs 1 and 6), the arrest of cpna2 embryos is likely due to the reduction of the well-folded KASI protein level in mutant embryos. However, we did not find any CPNA2pro:amiR-KASI transgenic line in which embryo development is arrested at the globular stage, perhaps because it would be difficult to obtain transgenic lines in which the expression of KASI is nearly knocked out since KASI is an embryo-lethal gene as reported by Wu and Xue [40]. Additionally, as previously mentioned, other unknown specific substrates of the chaperonin containing CPNA2 might exist in Arabidopsis embryos, and these substrates are also possible to play an important role in embryo development. Hence the elucidation of why cpna2 embryos are arrested at the globular stage still needs further research.
During plastid development, proplastids develop highly organized thylakoid membrane to differentiate into mature chloroplasts, and formation of the thylakoid membrane requires coordinated synthesis and assembly of proteins, pigments, and glycerolipids. In chloroplasts, de novo fatty acid (FA) synthesis produces 16:0 and 18:0 FAs that are the building blocks of membrane glycerolipid production [50]. Therefore, FA synthesis is crucial for the formation of the thylakoid membrane and for chloroplast biogenesis.
In this study, we found that CPNA2 deficiency could result in a significant decrease in the KASI protein level (Fig 6). KASI is a key condensing enzyme involved in de novo FA synthesis [50], and therefore, it is likely that CPNA2 deficiency also disrupts FA synthesis in chloroplasts, thus impeding formation of the thylakoid membrane and chloroplast biogenesis. In accordance with this conjecture, we found that abnormal chloroplasts in cpna2-2 embryos lacked thylakoid membranes and contained a deeply stained mass (Fig 2D and 2E), indicating that chloroplast biogenesis in cpna2-2 embryos is severely disrupted. Moreover, the result of GUS staining showed that CPNA2 is highly expressed in the SAM of Arabidopsis seedlings (Fig 3B and 3C), implying that CPNA2 may also play an important role in chloroplast biogenesis in the SAM. This idea was further supported by the result that the cpna2-2 homozygous seedlings could not develop true leaves (Fig 6A). Taken together, these results suggest that CPNA2 is crucial for chloroplast biogenesis, and thus affects the developmental processes of Arabidopsis embryos and seedlings.
In the process of biological evolution, gene duplication and variation usually extend the function of original genes to adapt to environmental changes. In this study, we found that CPNA2 in Arabidopsis belongs to a unique type of Cpn60α subunits that only exist in angiosperms. Functional chaperonins consisting of CPNA2 and specific Cpn60β subunits could specifically assist in the folding of KASI, and play an important role in the transition of globular embryos to heart-shaped embryos in Arabidopsis. This neofunctionalization of Cpn60α subunits in Arabidopsis provides a novel insight into the significance of multiple Cpn60α genes in plants, and reveals the relationship between duplication and functional specialization of chaperonin genes.
The Columbia ecotype of Arabidopsis thaliana was used as the wild type in this study. The T-DNA insertion mutants were obtained from ABRC (Arabidopsis Biological Resource Center), including CS76507, SALK_144574 (cpna2-3), SALK_006606 (cpna1), SAIL_852_B03 (cpnb1), SALK_014547 (cpnb2), SALK_099972 (cpnb3) and SALK_064887 (cpnb4). The cpna2-2 mutant was obtained from the stock CS76507 using thermal asymmetric interlaced PCR [28]. The T-DNA flanking sequences of the mutants were determined by PCR using specific primer of T-DNA left border (LB) and specific genomic primers (LP and RP). All plants were grown in a greenhouse under long-day condition (16 h light/8 h dark) at 22°C.
To construct CPNA2pro:gCPNA2, the 5000 bp CPNA2 genomic fragment was amplified from wild-type genome, and then cloned into pCambia1300 vector (Cambia).
To construct CPNA2pro:GUS, the promoter of CPNA2 was amplified and cloned into pCambia1381Xb vector (Cambia).
To construct CPNA2pro:H2B-GFP, CPNA1pro:H2B-CFP, 35Spro:GFP and 35Spro:CPNA2-GFP, we first obtained the pC1300-GFP and pC1300-CFP vectors using the operation procedure described by Ren et al. [51]. Then the promoters of CPNA2 and CPNA1 were amplified and inserted into the above vectors to produce CPNA2pro:GFP and CPNA1pro:CFP, while the 35S promoter was cloned into pC1300-GFP to produce 35Spro:GFP. Finally, the H2B coding sequence was amplified and cloned into CPNA2pro:GFP and CPNA1pro:CFP to obtain CPNA2pro:H2B-GFP and CPNA1pro:H2B-CFP, while the CPNA2 coding sequence was inserted into 35Spro:GFP to obtain 35Spro:CPNA2-GFP.
To construct 35Spro:CPNA2-HA, 35Spro:CPNA1-HA, CPNA2pro:CPNA2-HA, ABI3pro:CPNA2-HA and CPNA1pro:CPNA1-HA, the 35S promoter, the CPNA2 promoter, the CPNA1 promoter and ABI3 promoter were first cloned into pCambia1300 vector, respectively, to produce pC1300-35pro, pC1300-CPNA2pro, pC1300-CPNA1pro and pC1300-ABI3pro. Then the CPNA2 coding sequence fused to HA tag (CPNA2-HA) and the CPNA1 coding sequence fused to HA tag (CPNA1-HA) were amplified and cloned into pC1300-35pro to produce 35Spro:CPNA2-HA and 35Spro:CPNA1-HA, while CPNA2-HA and CPNA1-HA were cloned into pC1300-CPNA2pro and pC1300-CPNA1pro, respectively, to produce CPNA2pro:CPNA2-HA and CPNA1pro:CPNA1-HA. Moreover, CPNA2-HA was also cloned into pC1300-ABI3pro to produce ABI3pro:CPNA2-HA.
To construct CPNA2pro:amiR-KASI, we first obtained one amiRNA sequence (TGATGTAATTTACCTCCGCAG) designed for targeting the KASI gene using the Web MicroRNA Designer (WMD3; http://wmd3.weigelworld.org/cgi-bin/webapp.cgi) [52]. Then the amiRNA foldback fragment was generated by overlap extension PCR using four specific primers provided by WMD3 and pRS300 vector as a template. Finally, the amiR-KASI foldback fragment was inserted into the pC1300-CPNA2pro vector to produce CPNA2pro:amiR-KASI.
To construct CPNA2pro:CPNA2, CPNA2pro:CPNA2E291A, CPNA2pro:CPNA2Q299R and CPNA2pro:CPNA2E291A/Q299R vectors, CPNA2E291A, CPNA2Q299R and CPNA2E291A/Q299R were amplified from the CPNA2 coding sequence by site-directed mutagenesis using overlap extension PCR. Then these mutant sequences together with the CPNA2 coding sequence were cloned into the pC1300-CPNA2pro vector to obtain CPNA2pro:CPNA2, CPNA2pro:CPNA2E291A, CPNA2pro:CPNA2Q299R and CPNA2pro:CPNA2E291A/Q299R.
After sequencing, all the constructs were transformed into Arabidopsis plants using the floral dip method [53]. After screened on Murashige and Skoog medium with 10 mg/L hygromycin, positive transformants were identified by PCR and used for subsequent analysis. All the primers for cloning were listed in S4 Table.
Fresh ovules were first dissected from siliques using two needles and cleared with Hoyer’s solution following the protocol described by Yadegari et al. [54]. Then the embryos in the cleared ovules were observed under the Olympus TH4-200 microscope with differential interference contrast (DIC) optics and photographed by a SPOT Xplorer Camera (Diagnostic Instruments).
GUS staining was conducted according to the method described by He et al. [55]. The various tissues of CPNA2pro:GUS plants were incubated in GUS solution for 2 to 3 days at 37°C, and then observed by Olympus SZX12 stereomicroscope and photographed with a digital camera (Cool SNAP, RS Photometric).
To observe the fluorescent signals of embryos in the CPNA2pro:H2B-GFP and CPNA1pro:H2B-CFP plants, fresh embryos were isolated from ovules through enzymolysis (1% cellulose and 0.8% macerozyme dissolved in 13% mannitol, enzymolysis for 0.5 h at 37°C), mounted in 10% glycerol, and then observed under a confocal microscope (Fluoview1000; Olympus). The images were obtained under EGFP fluorescence channel (excitation, 488 nm; emission, 505–530 nm) and ECFP fluorescence channel (excitation, 440 nm; emission, 505–530 nm).
Total RNA of various Arabidopsis tissues were extracted using Trizol reagent (Sigma) and then reverse-transcribed into cDNA with a Reverse Transcription System (TOYOBO). The cDNAs of rosette leaves of the cpnb1-4 homozygous mutants were used as the templates for PCR analysis with the gene-specific primers. qRT-PCR of CPNA2 was performed using TransStart Top Green qPCR SuperMix (TransGen, China) with a Rotor-Gene 6000 machine (Corbett Research) and the relative expression levels normalized to GAPDH were analyzed by the double standard curves method as described previously [56]. qRT-PCR of KASI was performed using TransStart Top Green qPCR SuperMix (TransGen, China) with a Bio-rad CFX Connect machine (BIO-RAD) and the relative expression levels normalized to GAPDH were analyzed by the comparative CT method as described previously [57,58]. Three biological and three technical replicates of each sample were made for qRT-PCR analysis. Primers used in the experiments were listed in S4 Table.
The mesophyll protoplasts of 35Spro:GFP and 35Spro:CPNA2-GFP transgenic plants were isolated according to the method described previously [59], and then observed under a confocal microscope (Fluoview1000; Olympus). A 488 nm argon ion laser line was used for excitation of GFP and chlorophyll, while 505–530 nm and 650–675 nm emission filters were used for capturing GFP and chlorophyll autofluorescence, respectively.
The wild-type and cpna2 embryos in 6 DAP siliques of cpna2-2/+ plants were fixed, embedded and sectioned as described by Deng et al. [59]. The ultrathin sections were examined and photographed under a transmission electron microscope (Hitachi HT7700).
Chaperonin-substrate complexes were isolated from the 35Spro:CPNA2-HA and 35Spro:CPNA1-HA transgenic plants with the μMACS HA isolation kit (Miltenyi Biotec) according to the procedure described previously [21]. In brief, intact chloroplasts were first isolated from 7 DAG seedlings of the transformants by the method described previously [60]. Then the freshly isolated chloroplasts were ruptured in lysis buffer (50 mM Tris-HCl pH 8.0, 0.01% Tween 20, 10 mM MgCl2, 20 mM glucose, 30 U/ml hexokinase) plus protease inhibitor cocktail (Biotool). After lysis of chloroplasts, ADP (Sigma) was added into the lysates to reach a concentration of 10 mM, and then the lysates were centrifuged at 20,000 g for 10 min. The supernatants were transferred to new tubes and then NaCl was added into the supernatants to reach a final concentration of 150 mM. After incubating with 50 μl anti-HA Microbeads for 2.5 h at 4°C, the mixture was transferred to columns placed in a magnetic field. After rinsing four times with 200 μl washing buffer I (50 mM Tris-HCl pH 8.0, 1% Triton X-100, 0.5% Sodium deoxycholate, 150 mM NaCl, 5 mM ADP), twice with 200 μl washing buffer II (50 mM Tris-HCl pH 8.0, 1% Triton X-100, 150 mM NaCl, 5 mM ADP) and once with washing buffer III (25 mM Tris-HCl pH 7.5, 5 mM ADP), the immunoprecipitates were then eluted with 50 μl elution buffer (50 mM Tris-HCl pH 6.8, 50 mM DTT, 1% SDS, 1 mM EDTA, 0.005% bromophenol blue, 10% glycerol).
After elution, the immunoprecipitates were in-gel digested and analyzed by mass spectrometry as described by Wang et al. [61] with minor modification. In brief, the total protein was loaded to the gel and SDS-PAGE was conducted. After electrophoresis, the gel was stained, sliced and in-gel digested by trypsin, and then the desalted peptides were dissolved in 0.1% formic acid/2% acetonitrile/98% H2O, loaded onto a C18 trap column (Thermo Scientific), and subsequently eluted from the trap column over the self-packed C18 analytic column in a 120 min gradient. The LC-MS/MS analysis was performed by using a Q Exactive HF instrument (Thermo Scientific) equipped with an Easy-nLC 1000 system. MS data was acquired and submitted to Proteome Discoverer 1.4 (Thermo Scientific) to perform protein identification and quantitation utilizing its integrated SEQUEST HT search engine and Percolator algorithm. The peptide mass tolerance was set to 10 ppm and 20 mmu for MS/MS. Carbamido methylation of cysteine was set as a fixed modification, and oxidation of methionine and deamidation of N, Q as a dynamic modification. A high confidence dataset with less than 1% FDR (false discovery rate) was used for peptide filtering. Files from the samples were searched against the Arabidopsis proteome database of Swiss-Prot (http://www.uniprot.org/).
The coding sequences (CDS), excluding the portion of the transit peptides, of CPNA1, CPNA2, CPNB1, CPNB2, CPNB3, Cpn20, and KASI in Arabidopsis were cloned into pET-28a (+) vector (Novagen). Then all the proteins were overexpressed in E. coli expression strain BL21 following induction with isopropyl β-D-thiogalactoside (IPTG), and the BL21 cells were harvested and resuspended in lysis buffer (50 mM Na2HPO4 pH 8.0, 0.3 M NaCl, 1% Triton X-100, 5% glycerol, 2 mM PMSF). Following sonication of the BL21 cells, the fusion proteins were purified by High Affinity Ni-NTA Resin (Genscript).
For the proteinase K protection assay, Cpn60s were reconstituted with Cpn60α and Cpn60β according to the method described previously [31,62]. 15 μM Cpn60α, 15 μM Cpn60β and 10 μM Cpn20 were mixed in the incubation buffer (50 mM Tris-HCl pH 8.0, 0.3 M NaCl, 10 mM MgCl2, 16 mM KCl, 2 mM DTT and 5 mM ATP), and then incubated for 2 h at 30°C. After centrifugation, the supernatant fraction of the reconstitution mixture was collected and loaded on a Enrich Size Exclusion Column 650 (Bio-Rad). Then the reconstituted Cpn60s were purified and collected by gel-filtration chromatography.
The proteinase K protection assay was performed according to the procedure as described previously [62] with some modification. The various purified Cpn60s (1 μM) and denatured substrate KASI (0.64 μM) were incubated in refolding buffer (50 mM Tris-HCl pH 7.5, 50 mM KCl, 5 mM MgCl2, 1.9 μM Cpn20 and 1 mM ADP) for 20 min at 25°C, and then proteinase K (sigma) was added to a final concentration of 2.0 μg/mL. After incubation for 0, 5, 10, 15 and 20 min at 25°C, the proteolysis was stopped by adding PMSF (2 mM). Finally, the content of the substrates in reaction mixtures were analyzed by SDS-PAGE and immunoblotting with His-tag antibody (Genscript).
Total protein in 7 and 14 DAG seedlings of WT and cpna2-2 homozygous mutant carrying ABI3pro:CPNA2-HA vector was extracted according to the method described previously [59]. Then the concentrations of total protein were normalized by immunoblotting analysis of ACTIN using anti-ACTIN (ABclonal). 20 μl normalized protein samples were loaded on 12% SDS-PAGE gels and analyzed by immunoblotting using KASI antibody (ABclonal) to detect the KASI protein levels. The relative KASI protein levels in the seedlings of WT and cpna2-2 homozygous mutant were quantified by ImageJ software.
Multiple sequence alignment of Cpn60α proteins in various species was generated with ClustalX 1.83 [63]. Then the alignment result was used for building the phylogenetic tree with MEGA 5.1 [64]. The neighbor-joining method was used with a bootstrap (1000 replicates) test of phylogeny.
The predicted structural models of CPNA1, CPNA2, BnaC02g08340D, LOC100257653 and L484_018489 were obtained by SWISS-MODEL (http://www.swissmodel.expasy.org/), while the crystal structure of GroEL (1AON, Chain A) was used as the template. The finished models were visualized using Swiss-Pdb Viewer 4.1.0 [65].
Sequence data in this article can be found in TAIR (The Arabidopsis Information Resource) under these accession numbers: CPNA1 (AT2G28000), CPNA2 (AT5G18820), CPNB1 (AT1G55490), CPNB2 (AT3G13470), CPNB3 (AT5G56500), CPNB4 (AT1G26230), Cpn20 (AT5G20720), KASI (AT5G46290).
|
10.1371/journal.pntd.0007595 | Inhibition of Ebola Virus by a Molecularly Engineered Banana Lectin | Ebolaviruses cause an often rapidly fatal syndrome known as Ebola virus disease (EVD), with average case fatality rates of ~50%. There is no licensed vaccine or treatment for EVD, underscoring the urgent need to develop new anti-ebolavirus agents, especially in the face of an ongoing outbreak in the Democratic Republic of the Congo and the largest ever outbreak in Western Africa in 2013–2016. Lectins have been investigated as potential antiviral agents as they bind glycans present on viral surface glycoproteins, but clinical use of them has been slowed by concerns regarding their mitogenicity, i.e. ability to cause immune cell proliferation. We previously engineered a banana lectin (BanLec), a carbohydrate-binding protein, such that it retained antiviral activity but lost mitogenicity by mutating a single amino acid, yielding H84T BanLec (H84T). H84T shows activity against viruses containing high-mannose N-glycans, including influenza A and B, HIV-1 and -2, and hepatitis C virus. Since ebolavirus surface glycoproteins also contain many high-mannose N-glycans, we assessed whether H84T could inhibit ebolavirus replication. H84T inhibited Ebola virus (EBOV) replication in cell cultures. In cells, H84T inhibited both virus-like particle (VLP) entry and transcription/replication of the EBOV mini-genome at high micromolar concentrations, while inhibiting infection by transcription- and replication-competent VLPs, which measures the full viral life cycle, in the low micromolar range. H84T did not inhibit assembly, budding, or release of VLPs. These findings suggest that H84T may exert its anti-ebolavirus effect(s) by blocking both entry and transcription/replication. In a mouse model, H84T partially (maximally, ~50–80%) protected mice from an otherwise lethal mouse-adapted EBOV infection. Interestingly, a single dose of H84T pre-exposure to EBOV protected ~80% of mice. Thus, H84T shows promise as a new anti-ebolavirus agent with potential to be used in combination with vaccination or other agents in a prophylactic or therapeutic regimen.
| There are no approved vaccines or treatments to combat infections with ebolaviruses, which cause Ebola virus disease (EVD), an often rapidly fatal disease characterized by fever and bleeding that results in death in up to ~90% of cases. Ebolaviruses are among the most pathogenic viruses that cause human disease and represent a threat to global public health. Outbreaks of EVD occur periodically in African countries and can be exported elsewhere, with recent outbreaks including one ongoing in the Democratic Republic of the Congo and the largest ever in Western Africa in 2013–2016. There is therefore a great need to develop new vaccines and treatments that target ebolaviruses. We examined whether a lectin (carbohydrate-binding protein), predicted to bind to carbohydrates present on the surface of many viruses and thereby interfere with infection, could block ebolavirus infection and be used for prevention and/or treatment of EVD. We found that the protein blocked ebolavirus infection in cell cultures and, moreover, protected a significant proportion of ebolavirus-infected mice from death, even when administered only once before exposure to virus as a preventive. The protein hence shows promise as a potential agent to prevent and/or treat EVD.
| Filoviruses, which include the five species of Ebolavirus, are among the direst of all human pathogenic viruses, causing severe disease in humans and nonhuman primates that is often fulminant and rapidly fatal. Ebola virus disease (EVD) has a case fatality rate of 25–90%, with an average case fatality rate of 50% [1]. Since the discovery of ebolaviruses in 1976, there have been sporadic outbreaks of EVD in African countries, the longest and deadliest of which was the 2013–2016 Western African Ebola epidemic in Guinea, Liberia, and Sierra Leone, with approximately 28,000 cases and 11,000 deaths [2]. The latest outbreaks of EVD occurred from May to July of 2018 in the Democratic Republic of the Congo, with another unlinked outbreak ongoing in the same country as of August 2018, underscoring the unabating potential of ebolaviruses to re-emerge. Due to their high fatality rates and potential to be weaponized, filoviruses are considered category A priority pathogens and bioterrorism agents by the National Institute of Allergy and Infectious Diseases and the Centers for Disease Control and Prevention, respectively, and represent a serious threat to global health.
Unfortunately, there is currently no licensed vaccine or treatment for this deadly disease despite fervent efforts to develop preventive and therapeutic agents. Though progress has been made in developing anti-ebolavirus agents, challenges still remain. A number of ebolavirus vaccines have demonstrated safety and at least some immunogenicity in people [3,4], but the only ebolavirus vaccination trial to date to obtain clinical efficacy data was the Ebola Ça Suffit! trial, which took place during the 2013–2016 epidemic. This study employed ring vaccination of people epidemiologically linked to patients with EVD. Clinical trial subjects were vaccinated with the recombinant vesicular stomatitis virus vaccine (rVSV-ZEBOV), a live attenuated vaccine expressing the EBOV glycoprotein (GP) in a VSV vector [5]. Although the trial reported 100% efficacy in ring-vaccinated individuals and rVSV-ZEBOV is being used successfully in the current (August 2018 to manuscript submission) outbreak, there are challenges with its administration and it is not yet clear whether the vaccine will provide long-term protection [6–8]. Three experimental agents demonstrating at least some pre-clinical efficacy were used in the 2013–2016 outbreak [9], including ZMapp, a cocktail of three monoclonal antibodies against EBOV [10]; favipiravir, a viral RNA polymerase inhibitor [11]; and in two patients the nucleotide analog prodrug remdesivir [12,13], but neither these nor other antiviral therapeutic agents have yet been proven effective in clinical trials. Both ZMapp and vaccines require a cold chain, which would complicate distribution of these agents in the event of an outbreak given the poor infrastructure of many affected areas, and ZMapp is costly and requires administration by skilled health care workers. Given the paucity of licensed anti-ebolavirus vaccines and therapeutics, there is a critical need to develop new and potent agents for EVD prevention and treatment.
Lectins, or carbohydrate-binding proteins, have been viewed as candidate antiviral agents since many recognize glycans on the surfaces of viruses that are not commonly present on most normal human cells [14,15]. However, clinical use of lectins has been slowed by their mitogenicity, or ability to cause the proliferation of cells, particularly immune cells, that could potentially result in an unwanted immune reaction. We recently rationally engineered a lectin derived from bananas, banana lectin (BanLec), to retain broad-spectrum antiviral activity while removing mitogenicity with a single amino acid substitution, resulting in H84T BanLec (H84T) [16]. H84T binds to high-mannose N-glycans and has antiviral activity against a number of high-mannose-expressing viruses, including HIV-1 and -2, hepatitis C virus, and influenza A and B virus, among others. Ebolaviruses express, in addition to other glycans, a high level of high-mannose N-glycans on their surface glycoprotein (GP) molecules [17–19], envelope proteins important for viral entry [20]. Considering that H84T has demonstrated antiviral activity against other viruses containing high-mannose and that high-dose treatment with mannose binding lectin, another lectin with specificity for mannose, has proven efficacious in limiting EVD in vivo [21], in the present study we sought to characterize whether H84T has anti-ebolavirus activity and whether H84T could in turn protect against EVD in vivo. We demonstrate that H84T inhibits ebolavirus activity in vitro, protects mice from ebolavirus infection, and may exert its anti-ebolavirus effect by blocking both the entry and transcription/replication phases of the viral life cycle.
HEK293T/17 cells (ATCC CRL-11268) were obtained from the University of Virginia Tissue Culture Facility and were maintained in Dulbecco’s Modified Eagle Medium (DMEM) containing 1% sodium pyruvate, 1% L-glutamine, 1% antibiotic-antimycotic (all Gibco Life Technologies), and 10% supplemented calf serum (Hyclone). Vero E6 (ATCC CRL-1586, Manassas, VA), HeLa (ATCC CCL-2), and Huh 7 (human hepatocellular carcinoma) cells were maintained following recommended protocols.
H84T BanLec was prepared in E. coli as previously described [16], except that non-His-tagged protein was used for the majority of the study (in all experiments except those depicted in Figs 1, 2B and 7) and purified on a Sephadex G-75 column instead of on a Ni-NTA agarose column. Briefly, cleared bacterial lysates were added to the Sephadex column equilibrated with PBS and the column washed with PBS until the OD of the flow-through at 280 nm was < 0.02. The protein was then eluted with 0.2 M methyl-α-D-glucopyranoside. WT and D133G BanLec were prepared as previously described [16] and were His-tagged. For all lectins, endotoxin levels were tested with the Pierce LAL Chromogenic Endotoxin Quantitation Kit (Thermo Fisher Scientific). To remove endotoxin, 1 M glucose was added and pooled eluates containing the protein were passed through Mustang E filters (Pall). Following endotoxin removal to < 0.1 endotoxin units/mg of protein, the Vivaspin 20 centrifugal unit with 3K MWCO was used to remove glucose and concentrate H84T in water. Protein and endotoxin concentrations were then tested using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific) and LAL assay, respectively.
Ribavirin and toremifene citrate were purchased from Selleck.
KZ52-scFv, which recognizes EBOV GP, was expressed and purified by the Antibody Engineering and Technology core at the University of Virginia. To do so, a gene fragment consisting of codon-optimized DNA sequences for the KZ52 variable heavy chain linked to the KZ52 variable light chain by a 15 amino acid linker (Gly4Ser)3 was synthesized by BioBasic. The gene was cloned into a pET-based vector containing a 6x histidine tag and a pelB leader sequence for periplasmic expression of protein. E. coli B strain competent cells engineered to promote disulfide bond formation in the cytoplasm that is suitable for T7 promoter driven protein expression (designated SHuffle T7 Express: fhuA2 lacZ:T7 gene1 [lon] ompT ahpC gal λatt::pNEB3-r1-cDsbC (SpecR, lacIq) ΔtrxB sulA11 R(mcr-73::miniTn10—TetS)2 [dcm] R(zgb-210::Tn10—TetS) endA1 Δgor Δ(mcrC-mrr)114::IS10) were transformed with the pET-KZ52 construct and grown on ampicillin-LB agar plates.
The transformed cells were grown using rich medium, 2xYT, with ampicillin, at 37°C for 16 h. The next day, pre-warmed 2xYT medium was prepared in 2 L conical flasks (300 mL per flask) with antibiotics, inoculated with 10 mL of the overnight culture and then grown at 37°C (250 rpm) until the optical density at 600 nm had reached 0.5. Cells were then placed on ice for 30 min and IPTG was added to a final concentration of 0.5 mM. The cultures were grown at 25°C (250 rpm) for an additional 16 h. Bacterial cells were pelleted by centrifugation for 20 min (11,000 x g, 4°C), the supernatants were discarded, and the pellets were frozen at -20°C until needed.
Ebola virus/H.sapiens-tc/GIN/2014/Makona-C05 (EBOV/Mak, GenBank accession no. KX000398.1) and mouse-adapted Ebola virus/Mayinga (GenBank accession no. KY425637.1) (ma-EBOV) were propagated as previously described [23]. Virus stock and challenge inoculum titers were determined by plaque assay on Vero E6 cells. All procedures using live EBOV were performed under BSL-4 conditions.
Animals were housed in a facility accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. All experimental procedures were approved by the National Institute of Allergy and Infectious Diseases, Division of Clinical Research, Animal Care and Use Committee, and were in compliance with the Animal Welfare Act regulations, Public Health Service policy, and the Guide for the Care and Use of Laboratory Animals recommendations.
trVLPs were prepared under BSL-2 conditions as previously described [25–29]. Briefly, HEK293T/17 cells were seeded in 6-well plates and were transfected at 50–75% confluency with pCAGGS-NP; pCAGGS-VP35; pCAGGS-VP30; pCAGGS-L; a tetracistronic mini-genome plasmid encoding EBOV GP, EBOV VP24, EBOV VP40, and Renilla luciferase; and pCAGGS-T7 polymerase using TransIT-LT1 (Mirus). 24 h post transfection, the medium in each well was replaced with 2 mL fresh medium containing 5% serum and 25 mM HEPES (Gibco Life Technologies). 72 h post transfection, the cell medium containing trVLPs was collected, pooled, and cleared of cellular debris by centrifugation (800 x g, 5 min). trVLPs were stored at 4°C until use.
VLPs were prepared as previously described [26,27,30–32]. Briefly, 80% confluent HEK293T/17 cells were transfected with cDNAs encoding EBOV Mayinga GP, VP40, mCherry-VP40, and βlam-VP40. The cell medium was collected 24 and 48 h post transfection, pooled, and cleared of cellular debris by centrifugation (800 x g, 5 min). VLPs in the cleared medium were pelleted through a 20% sucrose cushion by centrifugation (112,398 x g, 2 h) and resuspended in 10% sucrose-HM (20 mM HEPES, 20 mM MES, 130 mM NaCl, pH 7.4). All VLP preps were assessed by western blot analyses and titered on HEK293T/17 cells to confirm entry competency. VLPs were frozen in single use aliquots at -80°C until use.
Infection of HEK293T/17 cells by trVLPs was assayed as described [25–29]. In short, to prepare target cells, HEK293T/17 cells were seeded in opaque white 96-well plates (Genesee Scientific). The next day, the cells were transfected with pCAGGS-NP, pCAGGS-VP35, pCAGGS-VP30, pCAGGS-L, and pCAGGS-Tim1 using polyethylenimine (PEI, Polysciences Inc). 18–24 h post transfection, target cells were pre-treated for 1 h at 37°C in a 5% CO2 incubator with the indicated concentration(s) of the indicated drug(s) diluted in cell medium (medium alone for mock). For some experiments (as noted), the trVLPs (instead of cells) were pretreated in medium containing the indicated concentration(s) of the indicated drug(s) or KZ52-scFv (or medium alone for mock) for 1 h at 37°C.
To assess trVLP infection, the pretreatment solution (or medium above cells for experiments where the trVLPs were pretreated) was removed and replaced with trVLPs diluted in growth medium containing the indicated concentration(s) of the indicated drug(s) or the pretreated trVLPs (as noted). The cells were then incubated for 48 h at 37°C in a 5% CO2 incubator, after which time luciferase was measured using Renilla-Glo substrate and a GloMax plate reader (both Promega). Percent inhibition was calculated as [1 - (treated trVLP signal-cell background) / (untreated trVLP signal-cell background)] x 100. Cytotoxicity was calculated as [1 - (treated cell signal-background) / (untreated cell signal-background)] x 100.
HEK293T/17 cells were seeded in 24-well plates. When 50–75% confluent, the cells were pretreated for 1 h at 37°C in a 5% CO2 incubator with the indicated concentration(s) of the indicated drug(s) (medium alone for mock). Transfection complexes composed of (per well) 25 ng pCAGGS-NP, 25 ng pCAGGS-VP35, 15 ng pCAGGS-VP30, 200 ng pCAGGS-L, 50 ng of a monocistronic mini-genome plasmid encoding Renilla luciferase, 50 ng pCAGGS-T7 polymerase, and 1.5 μL TransIT-LT1 were diluted in Opti-MEM1 (OMEM, Gibco Life Technologies) and added directly to the cells. 24 h post transfection, the medium in each well was replaced with fresh growth medium containing 5% serum, 25 mM HEPES, and the indicated concentration(s) of the indicated drug(s). 48 h post transfection, the medium above the cells was removed and Renilla-Glo was added. Once the cells were lysed (by the Renilla-Glo reagent), lysates were transferred to opaque white 96-well plates (Genesee Scientific) and luminescence was measured by GloMax. Percent inhibition was calculated as [1 - (treated p1cis signal-cell background) / (untreated p1cis signal-cell background)] x 100.
VLP entry assays were performed as previously described [26,27,30–32]. Briefly, HEK293T/17 cells were seeded in 96-well plates coated with fibronectin (Millipore) and were grown overnight at 37°C in a 5% CO2 incubator. When the cells were 80–90% confluent, the cells were pretreated with the indicated concentration of H84T Banana Lectin (H84T) diluted in OMEM (water for mock) for 1 h at 37°C in a 5% CO2 incubator. For some experiments (as indicated), the VLPs (instead of the cells) were pretreated in OMEM containing the indicated concentration of H84T or KZ52-scFv for 1 h at 37°C. The pretreatment solution (or medium above the cells for experiments where the VLPs were pretreated) was removed and VLPs were bound to the cells by spinfection (250 x g) for 1 h at 4°C. The cells were incubated for 3 h at 37°C in a 5% CO2 incubator, the β-lactamase substrate CCF2-AM (Life Technologies) was loaded into cells for 1 h at RT, and the cells were incubated overnight at RT. The cells were then lifted, fixed, and analyzed by flow cytometry as described in detail in Ref. 30. Percent inhibition was calculated as [1 - (treated virus signal-CCF2-AM background) / (untreated virus signal-CCF2-AM background)] x 100.
HEK293T/17 cells were seeded in 6-well plates. When ~50–75% confluent, the cells were pretreated for 1 h at 37°C in a 5% CO2 incubator with the indicated concentration of H84T (water for mock). Transfection mixes comprised of (per well) 0.5 μg EBOV VP40 cDNA, 0.5 μg full-length EBOV Mayinga GP cDNA, 10 μL Lipofectamine 2000 (Invitrogen), and 250 μL OMEM were prepared and added drop-wise to the cells. Following a 24 h incubation at 37°C in a 5% CO2 incubator, the supernatant (containing budded VLPs) was harvested and cleared of cellular debris by centrifugation. The VLPs in the clarified medium were pelleted through a 20% sucrose cushion and were resuspended overnight at 4°C in HM buffer. Concurrently, the VLP producer cells were lysed in 1X cold RIPA buffer (150 mM NaCl, 1 M Tris pH 8.0, 1% IGEPAL, 0.1% SDS, 0.5% sodium deoxycholate). The lysates were then cleared of debris by centrifugation and a BCA was performed to determine total protein concentration.
Gels of cell lysates (10 μg) or budded VLPs (from the supernatants) were run, protein was transferred to nitrocellulose, and the membranes were blocked for 1 h at RT. The membranes were then probed overnight for EBOV VP40 and tubulin (loading control) using 1:500 rabbit α-VP40 (IBT Bioservices) and 1:1000 mouse α-tubulin (Sigma) antibodies, followed by washing. EBOV VP40 and tubulin were detected using α-rabbit IR800 and α-mouse IR680 infrared secondary antibodies and an Odyssey CLx infrared imager (all LI-COR Biosciences). Bands corresponding to VP40 and tubulin were quantified using Image Studio Lite (LI-COR Biosciences), and the percentage of VP40 in the supernatant was calculated as the signal intensity of the VP40 band in the supernatant divided by the total signal intensity of the VP40 bands in the supernatant + lysate.
In a pilot test, we found that H84T BanLec (H84T) impedes infection by HIV-1 pseudovirions bearing the glycoprotein (GP) from the Mayinga variant of Ebola virus (EBOV). Based on these preliminary findings, we tested the effects of three doses of H84T on the replication of EBOV (Makona variant), the causative agent of the 2013–2016 Western African epidemic of EVD, in Huh 7 cells (derived from human liver) and Vero E6 cells (derived from monkey kidney), cells that are frequently used for in vitro studies of EBOV infection [23–32]. As seen in Fig 1, H84T blocked the replication of Makona EBOV in Huh 7 and Vero E6 cells, yielding 96 and 67% inhibition, respectively, at the highest concentration tested (20 μM). We next performed 8-point dose response studies to further assess the effects of H84T. As seen in Fig 2A, H84T inhibited replication of Makona EBOV by ~50% at the highest concentration (20 μM) tested in both Huh 7 and Vero E6 cells. Before testing H84T in the mouse model, we tested its effects on infections by mouse-adapted EBOV (ma-EBOV), which is derived from Mayinga EBOV [33]. Inhibition of ma-EBOV appeared somewhat stronger, reaching ~70–90% inhibition, with IC50 values of ~6 and ~1 μM, respectively, in Huh 7 and Vero E6 cells (Fig 2B).
Given that H84T inhibited the replication of Makona EBOV and ma-EBOV in cell cultures, we tested its ability to block EBOV infection in vivo in a murine model of disease. Initial dosing strategies were designed based on prior experience with efficacy against influenza A virus as well as pharmacodynamic data in mice (E. Covés-Datson and D. Markovitz, manuscript in review). As seen in Fig 3A, in a first study, two H84T dosing regimens (Groups 6 and 7) led to partial (~50%) protection: when H84T was administered at a dose of 50 mg/kg intraperitoneally (IP) 6 h prior to virus exposure and then again either on days 2, 4, 6, and 8 or on days 3, 6, and 9 (i.e. with either one or two days in between doses). No vehicle-treated mice survived. Other dosing regimens (Fig 3A: Groups 1, 3, 4, and 5) yielded less or no protection. In a second study, H84T was administered 6 h prior to exposure and then again on days 3, 6 and 9, with either the prior (50 mg/kg) or lower (5, 10, or 25 mg/kg) doses. A slightly higher level of protection (~77%) was observed at the 50 mg/kg dose (Fig 3B: Group 1) compared to that seen in the first study (~50%) (Fig 3A: Groups 6 and 7). Using this dosing schedule, as the dose of H84T was reduced there was a decrease in the survival benefit (Fig 3B: Groups 2–4). In the second experiment, we also tested the 50 mg/kg dose of H84T when administered in alternative regimens. Interestingly, the same level of protection (~77%) was observed when a single 50 mg/kg dose of H84T was administered 6 h prior to virus exposure (Fig 3C: Group 8) as when four 50 mg/kg doses of H84T were administered, 6 h prior to exposure and then again on days 3, 6, and 9 post-exposure (Fig 3B: Group 1). Alternative regimens that lacked a 50 mg/kg dose of H84T 6 h prior to challenge resulted in no protection with the exception of Group 9 (15.4% protection), which received the first dose at 14 h post-virus exposure (Fig 3C: Groups 9–13). Consistent with the findings in Fig 3A, less frequent dosing with H84T and a dose at 6 h prior to virus exposure resulted in higher levels of protection. Therefore, the findings from the second test of H84T in EBOV-challenged mice (displayed in Fig 3B and 3C) confirmed and extended those from the first test (shown in Fig 3A), demonstrating that a single dose of H84T provides reasonably effective prophylaxis against EVD.
In the case of HIV-1, wild-type (WT) BanLec binds to high-mannose sugars on the gp120 envelope glycoprotein and blocks HIV-1 infection by inhibiting virus attachment to and fusion with host cells [34]. We therefore tested if H84T might exert its inhibitory effect against EBOV by blocking EBOV entry into host cells. To do this we directly compared the ability of H84T to block infection by transcription- and replication-competent virus-like particles (trVLPs), a surrogate assay for authentic EBOV infection [25–29], and to block entry of EBOV VLPs; both assays can be conducted under BSL-2 conditions. The VLP entry assay employs filamentous entry-reporter VLPs containing the EBOV matrix protein (VP40) and glycoprotein (GP) as well as the entry reporter, VP40 β-lactamase [30]. Entry reporter VLPs enter host cells (monitored by cleavage of a fluorescent β-lactamase substrate), but do not perform further stages of the EBOV life cycle, as they do not contain a viral genome or replication machinery proteins. trVLPs are morphologically similar to entry reporter VLPs and virus (i.e., they are filamentous) and they also contain EBOV VP40 and GP. However, trVLPs additionally contain a 4-cis mini-genome (encoding Renilla luciferase, VP40, GP, and VP24, flanked by 5’ leader and 3’ trailer sequences from the EBOV genome) as well as EBOV proteins that drive EBOV replication: NP, L, VP30 and VP35. After trVLPs enter a cell, they replicate the 4-cis mini-genome and transcribe its encoded genes including Renilla luciferase, which is easily monitored by a Renilla luciferase assay (see Fig 1 in Watt and Hoenen [25]). When trVLPs and VLPs were pre-treated with H84T (and then H84T was maintained in the cultures), the engineered lectin strongly inhibited trVLP infection, with an apparent IC50 of ~5 μM, but showed minimal inhibition of VLP entry (Fig 4A). We also tested the effects of pre-treating the target cells with H84T (and then maintaining it throughout the experiment). H84T behaved similarly, showing strong inhibition of trVLP infection but no significant inhibition of VLP entry (Fig 4B). Hence, whether the VLPs/trVLPs or the target cells were pre-treated with up to 25 μM H84T, the engineered lectin did not inhibit the signal in the VLP entry assay but had a strong inhibitory effect on trVLP infection.
Where studied, low molecular weight inhibitors that block Ebola entry by targeting host cell pathways block trVLP infection and VLP entry with similar potencies [26,27]. Expecting H84T to be an EBOV entry inhibitor, we were therefore surprised that it inhibited trVLP infection much more strongly than VLP entry (Fig 4A and 4B). To explore this observation further, we first tested higher concentrations of H84T. As seen in Fig 4C, higher concentrations of H84T (100 and 250 μM) did inhibit VLP entry, but as seen previously (Fig 4A and 4B) considerably less potently than trVLP infection. In parallel with the experiment depicted in Fig 4C, we tested the effects of a single chain form of the KZ52 mAb (i.e., a virus-targeted biological inhibitor) on trVLP infection and VLP entry. As seen in Fig 4D, KZ52-scFv also inhibited trVLP infection more strongly than VLP entry, but the differential effect was not as strong as seen with H84T.
Given that H84T strongly blocked trVLP infection with only relatively weak activity on VLP entry, we tested whether it inhibits post-entry phases of the EBOV life cycle. We first assessed the effects of H84T in an assay that measures genome transcription and replication independently of EBOV entry. To do this we compared the effects of H84T on the Renilla luciferase read-outs from ‘1-cis’ (monocistronic) and ‘4-cis’ trVLP assays [25,28]. Unlike the 4-cis assay described above (referred to above as the trVLP infection assay), which evaluates the entire EBOV life cycle (including entry), the 1-cis assay strictly measures mini-genome transcription and replication. For the 1-cis assay, cells are co-transfected with a monocistronic mini-genome containing a Renilla luciferase reporter (flanked by the EBOV 5’ leader and 3’ trailer sequences) along with cDNAs encoding T7 polymerase, and the EBOV replication proteins NP, VP30, VP35, and L. Hence, within the transfected cells, the mini-genome is transcribed and Renilla luciferase is produced (see Fig 2 in Hoenen et al. [28]).
As expected, ribavirin, a known EBOV replication inhibitor [27], blocked genome replication (1-cis assay) as strongly as it inhibited the trVLP full life cycle (4-cis assay) (Fig 5; panels A and B represent two independent experiments). The entry inhibitor toremifene [29,31,32] blocked the full life cycle assay (4-cis, which requires VLP entry and subsequent steps of the EBOV life cycle), the expected behavior of an entry inhibitor, while having only a small effect (0–18% inhibition) on the 1-cis genome transcription/replication assay. In parallel experiments, H84T exerted a stronger effect (31–76% inhibition) than toremifene on genome transcription/replication (1-cis assay), but the inhibition of transcription/replication caused by H84T was weaker than its corresponding effects on the full (4-cis) life cycle assay.
Since H84T exerts a stronger effect on the trVLP full life cycle assay (requiring entry, replication/transcription, and assembly/budding) with weaker effects in the VLP entry (Fig 4) and transcription/replication (Fig 5) assays, we next asked whether it affects the other major stage of the EBOV life cycle involving new particle assembly, budding, and release from producer cells. To do this, we monitored the effects of H84T on the budding of EBOV VLPs containing EBOV VP40 and EBOV GP (but no other EBOV proteins), using an assay described by Loughran et al. [35]. There was no apparent effect of H84T on the budding of EBOV VLPs (Fig 6), suggesting that H84T does not exert its inhibitory effect against EBOV via inhibition of virus assembly, budding, or release.
We next sought to further investigate the unexpected observation that H84T blocks trVLP infection more strongly than VLP entry, reasoning that inhibition of trVLP infection by H84T could potentially be due, in part, to carbohydrate-independent effects. Thus, we compared the anti-trVLP infection activity of H84T and WT BanLec to that of D133G BanLec (D133G) (Fig 7), a variant of banana lectin that is identical to H84T except that, instead of the H84T mutation, it contains a single amino acid mutation in one of its two carbohydrate binding sites and therefore is not thought to efficiently bind to carbohydrates [16]. Indeed, in accordance with decreased binding to carbohydrates, D133G completely lacks antiviral activity against HCV, HIV-1, and influenza A and B viruses [16; and E. Covés-Datson and D. Markovitz, manuscript in review]. As expected based on findings in Fig 4, we found that H84T strongly inhibits trVLP infection. WT BanLec exhibited somewhat stronger inhibition of trVLP infection, which is consistent with previous in vitro studies comparing the activity of WT and H84T BanLec [16]. Surprisingly, D133G was also able to inhibit trVLP infection, though its inhibitory effect was diminished compared to that of H84T or WT BanLec. Cytotoxicity on 293T cells was minimal for all three lectins tested. The finding that D133G inhibits some trVLP infection was unexpected given that it does not even minimally inhibit HIV-1, HCV, or influenza A or B virus replication at doses at which H84T blocks these viruses. It appears that some carbohydrate binding is required for the inhibitory effect of BanLec against EBOV, as the carbohydrate-binding WT and H84T BanLec block trVLP infection more potently than does D133G BanLec. However, that D133G, a carbohydrate binding site mutant, inhibits trVLP infection suggests that at the concentrations used, BanLec may exert some carbohydrate-independent effects, e.g., against EBOV replication. Alternatively, we cannot rule out that D133G might retain weak binding to carbohydrates, which could account for its corresponding weak inhibition of trVLP infection.
Outbreaks of EVD continue to wreak havoc on a global scale and are capable of causing many thousands of deaths and widespread disruption in the regions where the virus emerges. These outbreaks often also spread panic and fear in other regions over the possibility of imported cases, which occurred during the 2013–2016 Western African Ebola epidemic. At the time of this writing, there is an ongoing Ebola epidemic in the Democratic Republic of the Congo that began in August 2018 and is as of yet uncontrolled, in part due to military conflict in the region; this outbreak follows on the heels of another unrelated outbreak in the same country in May-July 2018. Efforts to control Ebola outbreaks have historically relied on general infection control techniques, since to date there is no licensed vaccine or treatment against ebolaviruses. Several vaccine candidates have demonstrated safety and varying degrees of immunogenicity in people, but thus far only the rVSV-ZEBOV vaccine is being used clinically. On the therapeutic side, although some promising candidates exist, including ZMapp and remdesivir, no anti-ebolavirus treatments have yet been licensed, and there remains a need to develop additional potential anti-ebolavirus therapies, ideally ones in ample supply and not requiring a cold chain.
In this work, we first demonstrated that an engineered banana lectin, H84T BanLec (H84T), is active against EBOV in cell cultures. It inhibits both the Makona strain of EBOV, the strain responsible for the 2013–2016 epidemic, which was the largest in history, and the mouse-adapted Mayinga strain of EBOV in the two cell lines tested.
In a murine model of EBOV, we first found that a 50 mg/kg dose of H84T administered IP 6 h pre-exposure to the virus and then every 2 or 3 days thereafter for 3 to 4 additional doses protected ~50% of mice from EBOV infection. Dosing with 50 mg/kg of H84T more frequently was less efficacious, as when H84T was administered twice before exposure to the virus or every day (for 9 days), mouse survival was significantly reduced. In a second experiment, we found that H84T was most efficacious, protecting 77% of mice from otherwise lethal infection when administered 6 h pre-exposure to the virus and then again on days 3, 6 and 9 post-exposure. Lower doses were less efficacious as compared to the 50 mg/kg dose when administered in the same dosing schedule (i.e., 6 h before virus exposure and then on days 3, 6, and 9). Most interestingly, a single dose administered 6 h pre-exposure to the virus also yielded 77% survival. These latter findings suggest that H84T, in conjunction with vaccination, could have potential as a prophylactic agent for health care workers before beginning to treat EVD patients or for warfighters before going off to war. We have previously found that H84T can be toxic to mice when administered at high concentrations (200 mg/kg) (E. Covés-Datson and D. Markovitz, manuscript in review) which, along with the long half-life of H84T, likely explains why some dosing schedules that included more frequent doses of H84T were not as efficacious in the present study. In any case, it appears to be contraindicated to administer H84T to EBOV-infected mice on consecutive days.
When we examined the mechanism of action of H84T against EBOV, we found that at lower micromolar concentrations H84T inhibits EBOV infection, but not viral entry, as indicated by the fact that H84T robustly inhibited infection by trVLPs while having no or minimal effects on VLP entry. This was surprising since H84T is predicted to bind to EBOV GP, the EBOV entry protein. At very high concentrations, above 25 μM, H84T was able to inhibit entry of EBOV VLPs, but not as efficiently as it inhibited trVLP infection, which measures the full viral life cycle and was inhibited at much lower concentrations of H84T. The same higher concentrations of H84T blocked transcription and/or replication of the mini-genome (i.e., in the 1-cis assay), but, as seen for VLP entry, not as potently as they inhibited trVLP infection, suggesting that inhibition of the full life cycle was stronger than inhibition of EBOV genome transcription/replication alone. However, even high concentrations of H84T did not block VLP budding or release. Taken together, these findings indicate that H84T may inhibit EBOV infection by blocking both viral entry and transcription and/or replication. The observed post-entry effects may be due, in part, to non-carbohydrate binding activities of BanLec, as seen with the D133G mutant. Moreover, given that the inhibitory effect of H84T was considerably more potent for the overall life cycle than for either viral entry or genome transcription and/or replication, it is possible that these anti-EBOV effects are synergistic.
That H84T has demonstrated broad-spectrum activity against many different strains of HIV-1 and 2, HCV, and influenza A and B raises the possibility that H84T may also be broad-spectrum against the different species of pathogenic Ebolavirus, all of which possess GP molecules that express high-mannose [18] and so would be potential targets for binding by H84T. Most of the vaccines and several therapeutics in development, including ZMapp, are specific for one particular species of ebolavirus [36]. Thus, these agents currently in development would not likely be suitable for broad-spectrum empiric management of suspected EVD, such as would be required for immediate use in the event of a filovirus outbreak before identification of the virus had occurred. Although activity against only one ebolavirus was tested in the present study, if H84T indeed exhibits activity against additional species of Ebolavirus, it could be particularly useful as such an empiric treatment.
Our data demonstrate that H84T may have potential in people as an anti-ebolavirus prophylactic, perhaps in conjunction with vaccination as, remarkably, a single dose of H84T partially protects mice from otherwise lethal EBOV infection and as H84T has a long serum half-life (E. Covés-Datson and D. Markovitz, manuscript in review). H84T could also potentially be used therapeutically, perhaps especially when combined with other anti-ebolavirus agents, which if synergistic, could lower the amount of H84T needed. Remaining questions include why H84T is more potent against other viruses than it is against EBOV and what its precise mechanism of action is on a molecular scale. Whereas the IC50 concentration of H84T against EBOV appears to be in the low micromolar range, H84T inhibits HIV-1 and -2, HCV, and some subtypes of influenza A with concentrations in the low nanomolar range [16]. The higher concentration required to inhibit Ebola virus could potentially be attributed to the fact that ebolaviruses possess a very high number and density of glycans on their surfaces as compared to other viruses against which H84T has stronger activity [37–39]. In addition, ebolavirus GPs are heavily O-glycosylated [18], in contrast to the glycoproteins of HIV-1 and influenza A, e.g., which contain only N-glycans [38,40], and H84T generally does not recognize O-glycans [16]. If the mechanism of action of H84T requires its carbohydrate-binding activity, which appears to largely but not entirely be the case, then it is reasonable to hypothesize that a larger number and density of glycans on the surface of EBOV would require a higher concentration of H84T to bind to the glycans, though of course this would need to be tested. An additional, non-mutually exclusive possibility is that it is more difficult to block EBOV entry because EBOV can use TIM and TAM family members [41] in addition to lectins to gain entry into cells.
It is exciting to consider that H84T may inhibit both viral entry and genome transcription and/or replication, as this represents a distinct mechanism of action compared to that of WT BanLec against HIV-1 (primarily attachment inhibition) [34] and to that of H84T against influenza A virus (primarily fusion inhibition; E. Covés-Datson and D. Markovitz, manuscript in review). We hypothesize that the anti-viral entry effect may rely on binding to GP, but exactly how a high-mannose-binding lectin inhibits genome transcription/replication is an open question. It is clear, however, that H84T holds promise as a potential anti-ebolavirus prophylactic and/or therapeutic agent since it inhibits two variants of EBOV in cell cultures, protects mice from EBOV infection, may have a mechanism of action targeting both entry and transcription/replication, and is an inhibitor that could potentially be used in combination with other anti-ebolavirus agents that work through different mechanisms [29].
|
10.1371/journal.pntd.0003978 | Transcription Profiling of Malaria-Naïve and Semi-immune Colombian Volunteers in a Plasmodium vivax Sporozoite Challenge | Continued exposure to malaria-causing parasites in endemic regions of malaria induces significant levels of acquired immunity in adult individuals. A better understanding of the transcriptional basis for this acquired immunological response may provide insight into how the immune system can be boosted during vaccination, and into why infected individuals differ in symptomology.
Peripheral blood gene expression profiles of 9 semi-immune volunteers from a Plasmodium vivax malaria prevalent region (Buenaventura, Colombia) were compared to those of 7 naïve individuals from a region with no reported transmission of malaria (Cali, Colombia) after a controlled infection mosquito bite challenge with P. vivax. A Fluidigm nanoscale quantitative RT-PCR array was used to survey altered expression of 96 blood informative transcripts at 7 timepoints after controlled infection, and RNASeq was used to contrast pre-infection and early parasitemia timepoints. There was no evidence for transcriptional changes prior to the appearance of blood stage parasites at day 12 or 13, at which time there was a strong interferon response and, unexpectedly, down-regulation of transcripts related to inflammation and innate immunity. This differential expression was confirmed with RNASeq, which also suggested perturbations of aspects of T cell function and erythropoiesis. Despite differences in clinical symptoms between the semi-immune and malaria naïve individuals, only subtle differences in their transcriptomes were observed, although 175 genes showed significantly greater induction or repression in the naïve volunteers from Cali.
Gene expression profiling of whole blood reveals the type and duration of the immune response to P. vivax infection, and highlights a subset of genes that may mediate adaptive immunity.
| Plasmodium vivax malaria is a debilitating, occasionally life-threatening, and economically burdensome disease in Central Latin America, where 70%- 80% of the population lives with the risk of infection. We performed a gene expression profiling experiment taking advantage of a previously described sporozoite challenge experiment in Cali, Colombia that reported more severe malaria symptoms in subjects who have never experienced malaria. We show that no major differences are seen in the transcriptomes of uninfected naïve and semi-immune volunteers prior to infection, but differential expression of both neutrophil and interferon-related genes was evident at onset of malaria. Several hundred genes showed a stronger response in the naïve individuals just as parasites appear in the peripheral blood, and these fall into several pathways of interest. These findings show how information from gene expression profiling of whole blood can reveal the type and duration of the immune response to P. vivax infection, and highlights a subset of genes that may mediate adaptive immunity in chronically exposed individuals.
| One of the features of Plasmodium species that make them such pernicious parasites is their ability to avoid the host immune system [1,2]. While this is achieved in part by virtue of their complex life cycle that includes intra-erythrocyte cycling and periodic sequestration in various tissue compartments [2], it is also clear that Plasmodium infection causes short- and probably long-term modification of host immune function. Molecular methods are shedding some light on the mechanisms behind these modifications. For example, it is now clear that exposed individuals generally do mount an antigen response to Plasmodium antigens that persists [3,4], and that several biochemical pathways are engaged, including interferon and cytokine signaling, membrane lipid modification, and reactive oxygen species metabolism [5]. Host factors including genetic variation, both within and between populations, play a role in modulating immunity in malaria, as does the microbiome [6–9].
An important factor influencing the clinical course of disease is prior exposure to malaria. Adults and older children tend to experience reduced prevalence of malaria infection and have less severe symptoms [10,11]. Nevertheless the mechanisms responsible for host resistance to malaria are still poorly understood. As a prelude to evaluation of vaccine efficacy in a Colombian population, we recently carried out a challenge experiment in which we described the responses of immunologically naïve and semi-immune individuals to deliberate infection with Plasmodium vivax through mosquito bites [12]. All nine volunteers from a malaria endemic region near the town of Buenaventura were weakly positive for IgG antibodies to sporozoites or blood stage proteins prior to the experiment, and after challenge eight of them showed increased antibody titers against blood stages. Similarly, five of seven naïve volunteers from the city of Cali converted to sero-positivity that was generally maintained for at least four months. While there was no significant difference in the time to first appearance of blood stage parasite assessed by thick blood smears (12 to 13 days in both groups) or by polymerase chain reaction (PCR) (around 9 days), the naïve volunteers experienced classical early malaria symptoms, whereas the semi-immune volunteers were for the most part nearly asymptomatic, at least at the day of diagnosis when curative prophylaxis was administered [12].
In order to begin to characterize the molecular basis for this difference in clinical course of disease as a function of prior exposure to malaria, we report here two types of transcriptome profiling of peripheral blood samples from the Colombian challenge experiment volunteers. First we used targeted measurement of a set of 96 highly informative transcripts by nanoscale Real Time PCR (RT-qPCR) [13] in order to generate a time course of the infection transcriptional response. Second, we used RNASeq [14] on a subset of six volunteers contrasting baseline and incident malaria, to ask whether (i) there is a difference in immune profiles between naïve and semi-immune individuals in the absence of infection, and (ii) patent infection results in a differential transcriptional response that may hint at the molecular basis of long-term immunity. We also contrasted our findings with those of cross-sectional studies, concluding that history of exposure is just one of many factors mediating host–parasite interactions in malaria.
The experimental design protocol of this research was approved by the Institutional Review Boards (IRB) at the Malaria Vaccine and Drug Development Center (CECIV, Cali) and Centro Médico Imbanaco (Cali). Volunteers were adults and were extensively informed about the risks of participation. Before signing the written consent, all volunteers had to pass an oral or written exam related to the trial and its risks. Clinical trial was registered under registry number NCT01585077. It is described in more detail in Arévalo-Herrera et al. [12], which reports the clinical responses to malaria challenge. Sixteen Duffy-positive (Fy+) male and female volunteers (9 semi-immune, previously exposed to malaria, from Buenaventura and 7 immunologically naïve with respect to malaria, from Cali; 10 men and 6 women) were enrolled.
Volunteers where invited to the vaccine center two days (day -2) prior the challenge day (day 0) for physical examination and blood sample collection. Fig 1 summarizes the blood sampling strategy. Blood samples used for the RT-qPCR experiment were collected on day -2 (pre-challenge), day 5, day 7, day 9, on the day of first detection of Plasmodium by thick smear test (day 12–13), and on month 4. RNASeq analysis, also approved by the Georgia Tech IRB, was performed for 12 individuals (six each from Buenaventura and Cali) for two of the timepoints, namely the diagnosis day and baseline (pre-challenge day).
For each sample, approximately 1 mL of blood in 2 mL of buffer was collected into a Tempus tube, which preserves whole blood RNA at 4°C indefinitely. Whole blood mRNA was extracted using Tempus Blood RNA Tube isolation kits provided by the manufacturer Applied Biosystems, and the sample quality was determined based on the Agilent Bioanalyzer 2100 RNA Integrity score (RIN). Two samples had RIN approximatley 4 but these were not outliers in the analysis and there was no indication that RNA degradation influenced the results meaningfully.
Reverse Transcription followed by quantitative PCR (RT-qPCR) was performed using Fluidigm 96×96 nanofluidic arrays targeting a set of 96 transcripts that are broadly informative of the major axes of variation for peripheral blood gene expression from Preininger et al. [15] at six timepoints (Pre-challenge, day 5, day 7, day 9, Diagnosis (Dx) and month 4). The RT-qPCR was completed in three steps: (1) Total whole blood RNA was converted to single stranded cDNA using polyT priming of reverse transcription, (2) the 96 targeted genes were pre-amplified in a single 13-cycle PCR reaction for each sample following conditions outlined in the manufacturer’s protocol by combining cDNA with the pooled primers and EvaGreen Mastermix (Fluidigm BioMark), and (3) qPCR reactions were performed for each sample and individual gene on each sample on a 96×96 array with 30 amplification cycles. Average Ct value was calculated at a point in which every reaction is in the exponential phase to ensure accuracy and precision of amplification. In order to make the analysis more easily comparable with traditional transcript abundance measures such as those obtained with microarrays or RNASeq, each Ct value was subtracted from 30, setting missing values to 0. Since small Ct values correspond to high transcript abundance, this subtraction yields values ranging from 0 (no expression) to 30 (very high abundance). All measurements are reported in S1 Table.
Library preparation for RNASeq was performed using the Illumina TruSeq Low Throughput (LT) RNA Sample Preparation Protocol. Short read sequencing was performed in rapid run mode with eight samples per lane on an Illumina HiSeq 2500, generating 100 bp paired-end libraries with an average of 15 million paired reads per sample, and then sequencing on an Illumina HiSeq2100 at Georgia Institute of Technology.
The raw RNASeq reads (Fastq files) for each sample were tested using FastQC software analysis to check the quality of the data (S2 Table) and then aligned to the reference human genome (hg19 / GRCh37 assembly) using Bowtie as the short read aligner, and splice junctions were identified using TopHat2 in the Tuxedo protocol [16]. After alignment, estimation of transcript abundance measures as fragments per kilobase of exon per million aligned fragments (FPKM) values was performed using Cufflinks [16]. As a quality control for high variance associated with low abundance, genes with FPKM greater than 2.5 averaged across the 24 samples were retained for downstream analyses, representing 6,154 genes.
FPKM values were then transformed to logarithm base 2 to guarantee that the data were more normally distributed and to simplify the interpretation of the scale of differential expression (each unit difference corresponds to a two-fold difference in abundance). The supervised normalization of microarray (SNM) procedure was then used to normalize the data with the R package from Bioconductor [17], fitting location and timepoint as the biological variables, and individual as the adjustment variable (fit but not removed). All downstream analyses were performed on this normalized data set provided as S3 Table.
Many gene expression profiling experiments start with analysis of Principal Components, but since these are study specific, we also performed an analysis focused on large sets of genes that have been found to consistently covary in peripheral blood, namely blood informative transcript (BIT) axes analysis [15]. This analysis focuses on 9 common axes of variation that we detected in all human peripheral blood gene expression datasets that we have examined. They are related to 28 modules of co-expressed genes described by Chaussabel and colleagues [18] and found to be dysregulated in various immune diseases, but which collapse into 9 larger Axes of transcripts that covary in healthy adults. Each axis includes from several hundred to several thousand genes that gene set enrichment analysis suggests are involved in particular immune functions, broadly speaking, T cell signaling (Axis 1), reticulocyte number (Axis 2), B cell signaling (Axis 3), and inflammation (Axis 5) or specific immune or physiological responses (Interferon signaling, Axis 7).
Blood informative transcript (BIT) axes analysis was performed by generating the first PC for the 10 genes that are most strongly correlated with each of the 9 Axes reported in [15]. Principal component one (PC1) for each of these 10 sets of BIT provide a summary axis score, to which all of the other genes in the Axis positively correlate, nominally with a Pearson correlation coefficient greater than 0.5 as listed in S4 Table. This Axis score is then contrasted with respect to the covariates of interest (primarily time-point, location, and the interaction between them, but in exploratory analyses gender, parasitemia, and individual) using standard parametric t-tests or analysis of variance, or with linear regression.
Most statistical analyses of both the Fluidigm and RNASeq datasets were performed in JMP Genomics version 5 (SAS Institute, NC), starting with the Basic Expression Workflow, which performs principal components analysis (PCA), and computes a weighted total contribution of the covariates of interest to the axes (principal variance components analysis, PVCA). Linear regression was then used to assess the relationship between the individual covariates and PC, and/or analysis of variance was used to detect differential expression between locations or timepoints. We fit models with timepoint and location as fixed effects, and with individual as a random effect, in order to control for the effect of differential responses among individuals within each sample. No differences in the significance of the fixed effects were observed compared to models without individual, implying that this source of variation is minimal, and we report the significance of the full model.
A Benjamini-Hochberg 5% false discovery rate was used to select differentially expressed genes. Volcano plots contrast the significance (negative log10 of the p-value, NLP) against the fold difference (normalized log2 Ct or FPKM units) between specific conditions. Hierarchical clustering was performed using Ward’s method, and approximately unbiased boostrap support AU values were computed with the R program pvclust [19].
For the RNASeq there were 6 individuals of each gender, but they were asymmetric with respect to location, since 5 females were from Buenaventura and 5 males from Cali. Gender did not however account for a significant proportion of the major PC of gene expression. To confirm this, we fit models with gender, time and location for each Axis, and in each case the gender term was non-significant and the other terms were unaffected.
In order to identify whether location influences the axes of variation in another study, we reanalyzed data from Idaghdour et al. [9] who characterized whole blood transcriptomes of infants from the West African Republic of Benin, infected with Plasmodium falciparum. They reported on 61 healthy controls from a hospital in the city of Cotonou, and 92 cases drawn approximately equally and without bias with respect to parasitemia levels from Cotonou and the village of Zinvié, located 36 km from Cotonou (GEO accession number GSE34404). They identified parasitemia as the major factor influencing transcript abundance overall, but also described a location effect that is considered with respect to the BIT axes here.
The RNASeq dataset has been deposited into the Gene Expression Omnibus archive (GEO) under accession number GSE67184 and RT-qPCR data accession number GSE67470.
The first objective of this study was to compare the time course of transcriptional changes during response to infection, between naïve and semi-immune volunteers. There were 16 volunteers in all, 7 from Cali who had not previously been exposed to malaria, and 9 from Buenaventura, a village in an endemic region for the disease, all of whom had experienced between 2 and 5 mild bouts of malaria. Fig 1 summarizes the peripheral blood sampling scheme from 14 volunteers at Day 5 following exposure and again at Day 7, from 16 volunteers at Day 9 when PCR later confirmed initial appearance of blood-stage parasites, from 11 volunteers on Days 12 or 13 when parasitemia was diagnosed in thick blood smears, and from 14 volunteers four months after the initiation of the experiment. As mentioned above, there were no significant differences between the two groups either in the length of the pre-patent period or the level of parasitemia attained before administration of a curative cocktail of anti-malarial drugs.
Whole blood gene expression was monitored in each of the 85 samples using a Fluidigm nanoscale RT-qPCR array targeting 96 genes referred as “blood informative transcripts” (BIT) (S5 Table). These BIT consistently capture the covariance of over half of the genes expressed in blood, specifically serving as biomarkers for 10 conserved axes of variation. Across all of the gene expression measurements, 30% of the variance was among individuals, and just 6.5% between the timepoints, with very little differentiation between the naïve and pre-immune volunteers (Fig 2). The remainder of the variance was due to random biological or technical noise, or to the covariance of gene expression along the Axes.
We confirmed that most of the genes were co-regulated in this dataset by observing a strong correlation of expression for each of the 10 BIT for each Axis, and then generated Axis scores as the first principal component of the variance of those 10 BIT. Only two of the Axes were differentially expressed among timepoints in the Fluidigm data (Fig 3A and 3B). Axis 5 is related to innate immune signaling and neutrophil number, and seems to decline at Diagnosis, surprisingly, implying a mild reduction in inflammatory gene activity. Axis 7 represents Type 1 interferon induction and is, as expected, elevated at diagnosis, reflecting a transient specific immune response. Both axes had returned to close to baseline levels three months after recovery. No other gene expression differences detected by this targeted RT-qPCR analysis were associated with time or population. These results are consistent with previously observed stable maintenance of peripheral blood gene expression profiles in healthy adults. A caveat to this analysis is that it is possible that other genes not included in the targeted set of probes do change in expression prior to the diagnosis of parasitemia, or alternatively do not return to baseline after recovery.
In order to obtain a more comprehensive picture of the changes in gene expression as parasites first appear in the blood, we performed RNASeq on 6 volunteers each from Cali and Buenaventura, both at Baseline and Diagnosis. An average of 15 million paired-end 100bp short read alignments to the human reference genome were obtained for each sample, allowing us to estimate transcript abundance for each of 6,154 genes. Analysis of variance was used to contrast gene expression relative to population and timepoint, and to assess the interaction between these two factors. Fig 2B shows that 25% of the total variance was among individuals, similar to the Fluidigm observation, and that very little differentiation was seen between populations. However, just over one third of the variance was between Baseline and Diagnosis samples, implying a much greater response to infection than suggested by the RT-qPCR data, though it should be noted that when only contrasting the two most different timepoints, this contrast was expected to account for more of the variance.
The differential expression of Axes 5 and 7 was confirmed by the RNASeq data (Fig 3D and 3E), which also suggested divergence of Axis 2 (Fig 3C). Individual variability in response was minimal, as inclusion of individual as a random effect in the models had no effect on the proportion of variance due to either the time-of-diagnosis or population sampled. Up-regulation of Axis 2 is likely to be a sign of elevated erythropoiesis since they are enriched for erythrocyte-related function [15] and reanalysis of the dataset reported by Whitney et al. (2003) [20] shows that the genes in this axis are highly correlated with reticuloycte count, suggesting a mild physiological response to loss of red blood cell function even in the early stages of malaria. Interestingly, the increased resolution of RNASeq suggests differential responses of Axes 5 and 7 between the naïve and semi-immune populations. Specifically, the neutrophil and TLR-signaling associated with Axis 5 is much weaker in the naïve individuals (Fig 3D, solid blue points, p = 0.0008, though the location×timepoint interaction term is not significant, p = 0.13), whereas the induction of interferon signaling is variable in semi-immune volunteers (Fig 3D, open red circles), two of whom showed no response. The directional trends were the same in the Fluidigm data, but less apparent.
Consistent with timepoint rather than population explaining a large proportion of the variance, gene-specific differential expression analysis revealed more than 250 transcripts up- or down-regulated at the experiment-wide threshold of p<10-5 (Fig 4A), but only two transcripts more highly expressed in Buenaventura and none in Cali (Fig 4B). Approximately 50 genes show more than 2-fold up-regulation at Diagnosis relative to Baseline yet are less significant than many of the orange-colored genes (Fig 4A, green-colored genes). The reason is that these genes are even more highly upregulated in a subset of individuals, namely the naïve (Cali) volunteers. In fact, 175 genes show a significant timepoint-by-population interaction effect at p<0.05 (ANOVA, Fig 4C; S6 Table). These are represented in the heat-map in Fig 4C, showing two-way hierarchical clustering of transcripts in samples, two-thirds of the genes are actually down regulated at Diagnosis day (red sample labels, top). Interestingly, there was a marked distinction between the two timepoints (Fig 4C) in the sense that the Baseline samples were intermingled with respect to whether they were from the naïve or semi-immune populations, whereas the Diagnosis ones showed a near-perfect separation with respect to pre-immune exposure (bootstrap support 78%). In other words, most of the genes showing an interaction effect were more strongly up- or down regulated in the naïve than semi-immune individuals. An exception was a Baseline sample from a Cali volunteer (number 306), which clustered with the Diagnosis set but still showed a robust response to malaria infection along with moderate thrombocytopenia and leukopenia, as did Cali 310 who was not an outlier.)
Given the importance of cytokines to regulation of the immune response, we specifically analyzed the expression of many of the genes in the RNASeq dataset that are related to Interleukin (IL), interferon (IFN), tumor necrosis factor (TNF), and transforming growth factor (TGF) signaling. This analysis revealed three groups of samples, and three clusters of genes (Fig 5). Once again, the Baseline and Diagnosis samples were separated, excluding the outlier Cali 306 Baseline sample and two others, but in this case there was no clear separation relative to pre-infection malaria status. One cluster of 14 genes, including IL32 and IL8, was not differentially expressed. Another cluster of 23 genes, including the IL4R, IL6R, and IL7R and IL17R receptors, was upregulated at Baseline, particularly strongly in three volunteers (Cali 302 and Buenaventura 341 and 375). The third cluster of 19 genes, including TNF, IL1B and IL15, showed the opposite tendency, namely up-regulation at Diagnosis, particularly strongly in two samples (314 from Cali and 324 from Buenaventura). These results imply that there is strong co-regulation of the cytokine response and infection, but that this is not mediating the differential response between naïve and semi-immune individuals. This is somewhat surprising, especially given that the experience of fever was significantly different between the two populations, who might have been predicted to differ with respect to the pyrogenic cytokines IL1, IL6, IL8 and TNF.
Closer examination of the differentially expressed genes between Baseline and Diagnosis suggested a complex pattern of cross-regulatory interactions. The up- and down-regulated cytokines for example both include pro- and anti-inflammatory peptides and their receptors. Similarly, there appear to be counter-balancing signal transduction profiles: JAK1 and RAF1 are both strongly down-regulated in all volunteers at Diagnosis, whereas IL6ST and SOS1 are up regulated.
Among the 175 genes showing a significant timepoint-by-population interaction effect, namely a stronger response at diagnosis in the immunologically naïve individuals, there are several types of gene functions of interest (Table 1). These include lysosomal components (CTSH, RILP), regulators of macrophage activity (CD163, MMP25, SIRPA, TBC1D14, TNFSF13), splicing factors (EIF2C4, SNRPB2, SNRPG), lipid biosynthesis (DGAT2, LPPR2), solute carriers (S100P, SLC6A6, SLC11A1, SLC7A7), signal transduction (G3BP1, GAB3, MAPK13, TLE3) and Cell Cycle and DNA damage response (ATM, PRKDC, ARID4A). Some genes with an interaction effect showed stronger down-regulation in Cali (Fig 6A, ATM), or stronger down-regulation in Buenaventura (Fig 6B, EIF2C4), compared with one that showed a similar up-regulation at both locations (Fig 6C, ATP1B3).
Finally, we reanalyzed an infant malarial gene expression dataset from Benin [9]. All samples were collected within a period of 10 weeks in the Spring of 2010, and transcript abundance data was generated on Illumina HumanHT-12 BeadChips for 155 individuals (61 controls from Cotonou, 24 high parasitemia from the village of Zinvie, 52 low parasitemia from Zinvie, and 18 from the city of Cotonou). Critical differences relative to our study include (i) comparison with P. falciparum rather than with P. vivax infection, (ii) infants versus young adults comparison, and (iii) cross-sectional rather than Baseline vs Diagnosis analysis. Nevertheless, a significant correlation (Fig 7A and 7B) was observed between parasitemia and two Axes of variation, Axes 1 and 5. However, in this case there was activation of the innate immunity/inflammation genes as parasite burden increases. Axis 1, which is enriched for T-cell signaling activity [15], was strongly reduced as parasitemia increased, but like Axis 5, not significantly affected in the infants with low parasitemia. From 32 genes showing a significant interaction effect between timepoint and population in our challenge experiment, 12 were nominally differentially expressed between malaria patients in the city of Cotonou and rural village of Zinvie in Benin.
The core result of this study was that gene expression was significantly altered at the time of malaria diagnosis, particularly in the immunologically naïve volunteers. Although the targeted expression profiling is less comprehensive and less sensitive than the RNASeq, it suggests that there is minimal transcriptional change in peripheral blood prior to patent infection, and that individual profiles return to baseline within a few months of parasite clearance. No obvious difference in the transcriptomes of uninfected naïve and semi-immune volunteers was seen, but several hundred genes showed a stronger response in the naïve individuals. We cannot however conclude that prior immune exposure is the only reason for this difference as other lifestyle factors that distinguish the inland city of Cali from the oceanside town of Buenaventura, (where there is likely a larger proportion of African ancestry) may also play a role. However, the data is strongly suggestive of a long-term modulation of the malaria immune response involving multiple molecular pathways.
Some studies have suggested that clinically immune individuals infected with P. vivax show lower levels of inflammatory and regulatory cytokines, than individuals infected with P. falciparum malaria [21]. Nevertheless, the down-regulation of multiple genes related to innate immunity, inflammation, and neutrophil abundance, all correlated with Axis 5, observed here was unexpected. A large cross-sectional study of infants with malaria conducted in the West African Republic of Benin [9] documented a strong up-regulation of the same genes, although reanalysis of their data shown in Fig 7A suggests that is only true in the presence of high levels of parasitemia. Even more surprisingly, the reduction in inflammatory gene expression was stronger in the naïve than semi-immune volunteers. One possibility is that there is a transient reduction in relative neutrophil counts and inflammatory gene expression as the parasite first appears in the bloodstream, just as the lymphoid cells begin to amplify their response, and this is corrected as parasite levels increase and neutrophilia occurs a few days into the infection [22,23].
An observation that is consistent with published data is the strong induction of an interferon response in association with blood-stage malaria [24,25]. It is unclear whether this induction was stronger in Cali or Buenaventura, since a couple of the Cali volunteers had unusually high baseline interferon-related gene expression captured by Axis 7. It does appear that a few of the semi-immune individuals did not mount an interferon response, consistent with the absence of overt clinical symptoms and implying that their immunological memory was able to deal with at least the early stage of infection without mounting the kind of major immunological response observed in the naïve volunteers. This in turn implies that the presence of blood stage parasites alone is not the only determinant of whether or not an individual mounts an interferon response. The overall cytokine profile shifts reported in Fig 5 did not correlate with the clinical profile differences, which could be explained by the host immunity level that can vary due to the acquired immunity throughout repeated exposure [26]. Presumably larger sample sizes and longitudinal profiling during disease will identify associations between gene expression and physiological response, which is also likely to involve other tissues.
On the other hand, multiple classes of gene activity do seem to be differentially activated between naïve and semi-immune volunteers. These include various signal transduction molecules, genes related to macrophage activity, and other cellular processes that are known to influence immune responsiveness including lipid synthesis and lysosomal function concordant with Portugal et al. [27] who suggest that as children develop exposure-dependent immunity to P. falciparum, the responses reduce pathogenic inflammation and boost anti-parasite mechanisms. The aforementioned study in Benin again provides a potential comparison, since it included the contrast between children in the city of Cotonou with the rural village of Zinvié. Differences in human peripheral blood gene expression according to lifestyle are prevalent [28], but it is nevertheless interesting that, of the 32genes showing a significant interaction effect between timepoint and population in our challenge experiment, 12 were nominally differentially expressed between malaria patients from the two locations in Benin, compared with no more than three expected. Fig 7B shows that Axis 1 (related to T-cell signaling) is down-regulated with high parasitemia, and consistently reduced in the village of Zinvié. This Axis was not affected in our study, but collectively these observations of context-dependent alterations in gene expression provide further evidence that immune history is an important mediator of the differential clinical profiles observed among individuals.
There is also considerable interest in the use of gene expression profiling to identify genes that may mediate robust vaccine responses. Recent study reports on influenza and yellow fever have highlighted individual genes that are required for vaccine effectiveness, but have also suggested that baseline profiles of immune cell types may provide better predictors of antibody production [8,29]. Various properties of Plasmodium suggest that this organism may present a more difficult scenario for dissecting the molecular basis of vaccine responses, but we consider the results reported here to be an encouraging baseline establishing that differential responses to a malaria challenge can be detected by gene expression profiling. It will be interesting to see whether pre-immune exposure influences the molecular basis of vaccination with irradiated sporozoites in the next phase of this study.
This study shows that differential gene expression is particularly strong in naïve volunteers in comparison to semi-immune individuals at the time of malaria diagnosis. One way to interpret this result is that it provides a molecular signature of tolerance of, as opposed to resistance to, the pathogen [30]. In the presence of chronic exposure, the host immune system moves toward an equilibrium where pathogen is tolerated by mounting a measured immune response, without requiring complete sterile immunity that would likely have a greater physiological impact on the infected individual. This in turn implies that gene expression profiling of lymphocytes can be used to identify the type and duration of the immune signals that may be biomarkers for vaccine immunogenicity, and to establish how semi-immune exposure modifies their activation.
|
10.1371/journal.ppat.1001064 | A Novel CCR5 Mutation Common in Sooty Mangabeys Reveals SIVsmm Infection of CCR5-Null Natural Hosts and Efficient Alternative Coreceptor Use In Vivo | In contrast to HIV infection in humans and SIV in macaques, SIV infection of natural hosts including sooty mangabeys (SM) is non-pathogenic despite robust virus replication. We identified a novel SM CCR5 allele containing a two base pair deletion (Δ2) encoding a truncated molecule that is not expressed on the cell surface and does not support SIV entry in vitro. The allele was present at a 26% frequency in a large SM colony, along with 3% for a CCR5Δ24 deletion allele that also abrogates surface expression. Overall, 8% of animals were homozygous for defective CCR5 alleles and 41% were heterozygous. The mutant allele was also present in wild SM in West Africa. CD8+ and CD4+ T cells displayed a gradient of CCR5 expression across genotype groups, which was highly significant for CD8+ cells. Remarkably, the prevalence of natural SIVsmm infection was not significantly different in animals lacking functional CCR5 compared to heterozygous and homozygous wild-type animals. Furthermore, animals lacking functional CCR5 had robust plasma viral loads, which were only modestly lower than wild-type animals. SIVsmm primary isolates infected both homozygous mutant and wild-type PBMC in a CCR5-independent manner in vitro, and Envs from both CCR5-null and wild-type infected animals used CXCR6, GPR15 and GPR1 in addition to CCR5 in transfected cells. These data clearly indicate that SIVsmm relies on CCR5-independent entry pathways in SM that are homozygous for defective CCR5 alleles and, while the extent of alternative coreceptor use in SM with CCR5 wild type alleles is uncertain, strongly suggest that SIVsmm tropism and host cell targeting in vivo is defined by the distribution and use of alternative entry pathways in addition to CCR5. SIVsmm entry through alternative pathways in vivo raises the possibility of novel CCR5-negative target cells that may be more expendable than CCR5+ cells and enable the virus to replicate efficiently without causing disease in the face of extremely restricted CCR5 expression seen in SM and several other natural host species.
| SIV causes AIDS in macaques, like HIV-1 does in humans, but not in its natural host species such as sooty mangabeys (SM). It is therefore important to understand infection in natural hosts, including the mechanisms and cellular targets of infection. SIV replication in SM is thought to exclusively use CCR5 as its entry coreceptor, which mediates viral entry in conjunction with CD4 and is the main determinant of target cell tropism. However, other molecules also function as SIV coreceptors in vitro. We discovered that inactivating mutations in the CCR5 gene are common among SM and, furthermore, homozygous mutant animals lacking functional CCR5 still become infected and have high viral loads. Ex vivo, SM lymphocytes can be infected independently of CCR5, and several alternative entry coreceptors are used by SIV from both CCR5+ and CCR5-null animals. Thus, SIV infection in SM is mediated by other coreceptors in addition to CCR5, suggesting that these molecules together may determine tropism and cell targeting in vivo. These results provide new insight into an important model of nonpathogenic natural host infection, and identify a novel role for alternative entry pathways suggesting a potentially broader range of target cells in vivo than currently recognized.
| HIV-1 emergence into the human population resulted from cross-species transmission of SIVcpz from chimpanzees (Pan troglodytes), which itself resulted from transmission and subsequent recombination of SIVs infecting primates on which chimpanzees prey [1], [2]. Similarly, both simian AIDS caused by SIVmac/smm in rhesus macaques (RM; Macaca mulatta) and HIV-2 infection of humans originated from cross-species transmission of SIVsmm from naturally infected sooty mangabeys (SM; Cercocebus atys) [3], [4], [5]. In marked contrast to pathogenic infections leading to AIDS in non-natural hosts, infection in natural host species including SM is typically non-progressive [6], [7], [8]. Importantly, the benign nature of SM infection in vivo is not due to overall restricted viral replication, as both nonpathogenic natural host and pathogenic nonnatural host infections are characterized by robust viremia [9], [10], [11], [12]. This observation indicates that immunodeficiency virus replication and pathogenesis are not inextricably linked. Thus, understanding natural host infection has become a high priority for identifying key features of infection in vivo that regulate pathogenesis and, potentially, identify opportunities to modulate disease apart from or in addition to suppressing overall virus replication through pharmacologic or immune mechanisms.
HIV and SIV entry into target cells is initiated by binding of the viral envelope glycoprotein (Env) to cell surface CD4, followed by structural changes that enable interactions with a seven transmembrane G protein coupled cell surface receptor that then triggers fusion. HIV-1 isolates use CCR5 or CXCR4 or both, and in vitro use other molecules infrequently. The restricted expression of CCR5 mainly on memory CD4+ T cells, but broader expression of CXCR4 on both memory and naïve subsets, is thought in part to underlie the accelerated disease progression seen in individuals in whom CXCR4-using HIV-1 variants emerge late in the course of infection [13], [14], [15], [16]. In contrast, SIV strains use CCR5 almost universally and very rarely use CXCR4. Sooty mangabeys express very low levels of CCR5 on their CD4+ T cells, a mechanism by which replication might be regulated in vivo and restrict transmission and pathogenesis [17], [18]. However, most strains of SIV use a number of other alternative coreceptors in in vitro assays, such as CXCR6 (STRL33), the orphan receptors GPR1 and GPR15, and several others [19], [20], [21]. Despite the efficient use of such alternative entry pathways by SIVmac and SIVsmm isolates in transfected cells, infection and cell targeting in vivo is generally thought to be dependent on CCR5 [22]. Notably, however, the proportion of CD4+ T cells depleted and/or infected at a given time in macaques and mangabeys may exceed the proportion of cells with detectable CCR5 expression, raising the possibility that other pathways in addition to CCR5 might be utilized [23], [24].
Although both natural and non-natural host infections result in high level virus replication, several distinguishing features provide clues as to possible causes for the distinct outcomes. It is long believed that in addition to CD4+ T cell destruction, pathogenesis involves an inability to effectively replenish these populations [25], [26]. In pathogenic rhesus macaque infection, damage to the CD4+ T central memory (Tcm) subpopulation appears to play a central role in the inability of infected animals to replenish CD4+ T effector and effector memory (Tem) cells depleted by infection [27]. It has been recently found that cell-associated viral loads in CD4+ Tcm are considerably lower in SM than RM, despite equivalent or higher Tem infection levels, which might enable better immune cell homeostasis in infected SM (G.S. and M.P; unpublished observations). Another difference is the presence of chronic generalized immune activation in infected humans and RM, whereas natural hosts display generalized immune activation during acute infection that then rapidly resolves [11], [28], [29], [30]. Chronic generalized immune activation may contribute to accelerated T cell turnover and ultimate depletion, and is believed to result in large part from translocation of gut microbial products due to gastrointestinal barrier damage during acute infection [31]. However, vigorous virus replication and extensive CD4+ T cell depletion in gut mucosal tissue occur in both natural and non-natural hosts [24], [32], [33], [34]. A potentially important difference is Th17 CD4+ T cells, which play a critical role in defense against bacteria at mucosal sites, and are lost in HIV-1 and SIVmac rhesus macaque infection but spared in infected natural hosts [35], [36], [37], [38]. Thus, the factors regulating CD4+ T cell subset targeting in vivo may be central to defining the outcome of infection in natural or non-natural hosts.
Interestingly, evolution appears to have favored mutations in the CCR5 gene that abrogate surface expression of this molecule. An allele containing a 32 base pair frameshift deletion in human CCR5 that abrogates cell surface expression (CCR5Δ32) is present at a frequency of 10% in the Caucasian population, resulting in about 18% heterozygous and 1% homozygous individuals. CD4+ cells from individuals homozygous for CCR5Δ32 are resistant to infection by CCR5-using HIV-1 isolates in vitro but permissive for strains that can use CXCR4, and the essential role for CCR5 in HIV-1 transmission and infection is demonstrated by the finding that individuals homozygous for CCR5Δ32 are highly resistant to infection [39], [40]. Furthermore, heterozygous individuals can be infected but show lower viral loads and slower disease progression, in association with lower levels of CCR5 expression [41], [42]. Red-capped mangabeys (RCM; Cercocebus torquatus) are the natural host of SIVrcm, and a 24 base pair in-frame deletion (CCR5Δ24) that also abrogates surface expression is present at an allelic frequency of 87% [43]. As a result, ≥70% of RCM are homozygous for the mutation and do not express CCR5, and the one known exception to exclusive CCR5 dependence by SIV in vivo is SIVrcm, which uses CCR2b for entry and cannot use CCR5. The same CCR5Δ24 mutant allele was also reported in two different populations of SM, which are closely related to RCM, but with a low allelic frequency of 4% and no animals homozygous for the allele were found [43], [44].
In this study, we identified a novel 2 base pair deletion and frameshift mutation in SM CCR5 (CCR5Δ2) that results in a lack of surface expression and coreceptor function. The mutation is present at a 26% allele frequency in the large Yerkes National Primate Research Center (YNPRC) SM colony, and together with the previously-described Δ24 allele results in 8% of animals lacking functional CCR5. However CCR5-null SM are susceptible to natural and experimental SIVsmm infection and exhibit robust viral replication. This is the first clear evidence for in vivo alternative coreceptor use by SIVsmm in its natural hosts, which provides an explanation for the efficient use of alternative coreceptors by the SIVsmm/mac family of viruses. This data also suggests that cell targeting and tropism in sooty mangabeys is linked to expression and use of both CCR5 and additional alternative entry pathways, and identifies a third example of convergent evolution resulting in nonfunctional mutant CCR5 alleles among primate species.
We amplified the CCR5 coding sequence from SM genomic DNA, cloned it into an expression vector and analyzed several clones by sequence analysis. Cloning from four independent PCR reactions amplifying smCCR5 from one animal (FVq) resulted in the identification of two distinct alleles in each of the PCR amplifications (Figure 1). One was a wild-type allele, similar to published smCCR5 sequences (smCCR5 wt), and the other was a novel allele containing a two base pair deletion at nucleotides 466 and 467 of the coding sequence (smCCR5Δ2), which corresponds to the fourth transmembrane domain (TM4) of the smCCR5 protein (Figure 1A). This deletion causes a frameshift that results in a predicted protein with 110 missense amino acids prior to termination at residue 265 (Figure 1B). In addition to this deletion, the smCCR5Δ2 allele contains two nucleotide substitutions compared with the wild-type allele identified. The first is a 436T>G substitution resulting in a L146V amino acid change, which is also found in several published wild-type smCCR5 sequences. The second nucleotide substitution is 538C>T, but is masked in CCR5Δ2 due to the frameshift.
In addition to premature truncation of the protein, the mutation results in several charged amino acids predicted within TM4, as well as loss of both disulfide bonds (between the N-terminus and third extracellular loop (ECL), and between the first and second ECLs, respectively) that maintain secondary structure. These features suggest that the mutant protein is unlikely to be configured in a manner to allow proper membrane placement and either normal signaling or SIV/HIV entry coreceptor function.
There are several other mutations in primate CCR5 genes, including the well-studied 32 base pair frameshift mutation in human CCR5 (CCR5Δ32) [39], [40], [42] and a 24 base pair deletion (CCR5Δ24) that is common in RCM and also reported at low frequency in SM [43], [44]. Therefore, we examined the relationship between this SM CCR5Δ2 deletion and the Δ24 RCM/SM and human Δ32 deletions (Figure 1A & B). The smCCR5Δ2 deletion occurs in the same region of TM4, and overlaps the Δ24 deletion, which is characterized by multiple G-T-G repeats. In contrast, the human CCR5Δ32 deletion occurs approximately 90 bases downstream of smCCR5Δ2, within the second extracellular loop of the protein. Interestingly, the SM and RCM Δ24 alleles also contain the 436T>G and 538C>T substitutions seen smCCR5Δ2, which result in amino acids identical to those in human CCR5 at those respective sites (valine at position 146 and serine at position 180; Figure 1B).
Because the CCR5Δ24 mutation was previously reported to be present in animals at YNPRC [44], animals were screened for its presence by PCR and several Δ24 carriers were identified (Figure S1). We therefore generated a clone of the CCR5Δ24 coding region by PCR of genomic DNA from one heterozygous animal. Sequence analysis of this CCR5Δ24 clone was similar to the sequence previously described [43], [44], except for a non-coding 1026G>T substitution.
Previous studies have shown that proteins encoded by the mutant human CCR5Δ32 and RCM/SM CCR5Δ24 are not expressed on the cell surface [40], [44]. To test whether the smCCR5Δ2 mutant gene gave rise to a protein expressed on the cell surface, we transfected 293T cells with wild-type and mutant SM CCR5 expression plasmids, along with human CCR5, and measured surface expression by flow cytometry. Staining utilized mAb 3A9, which cross-reacts with both human and SM CCR5 and, importantly, recognizes an epitope in the N-terminus that is upstream of the mutation and should be detected if the predicted protein were expressed.
As shown in Figure 2, surface expression of CCR5 was readily detected on cells transfected with SM or human wild-type CCR5. In contrast, neither the Δ2 nor Δ24 SM mutant CCR5 alleles gave rise to detectable surface expression. We have so far been unable to assess intracellular expression because of high nonspecific intracellular staining with 3A9 and other anti-N-terminal CCR5 antibodies in all cells tested so far (data not shown). From this data, we conclude that the frameshift mutation in smCCR5Δ2 results in a truncated protein that is not expressed on the cell surface. In addition, our observation that the CCR5Δ24 mutant allele is not expressed at the cell surface is consistent with previous reports that this deletion abrogates cell surface expression even though it is not a frameshift [43], [44]. We also asked if the mutant protein might have a dominant negative effect on wild-type CCR5 expression, but found no change in CCR5 staining if the wild-type CCR5 plasmid was co-transfected along with the Δ2 or Δ24 alleles (Figure S2).
Next we determined the prevalence of the smCCR5Δ2 allele in SM housed at the Yerkes National Primate Research Center (YNPRC), which is the largest captive colony of SM in the world (n = 202). All animals were initially screened using a discriminatory PCR assay that specifically identifies the smCCR5 wild-type and Δ2 alleles (Figure S1A). Animals were also screened for the Δ24 allele using primers that discriminate Δ24 from Δ2 and wild-type alleles based on amplicon size (Figure S1B). Results were then verified by direct bulk sequencing of genomic DNA amplicons that confirmed the presence or absence of homozygous genotypes, or demonstrated frameshifting with sequence overlap in heterozygous animals (Figure S1C).
The result from this analysis is shown in Table 1. Five of six possible genotypes were identified: 50.5% of the SM carried two wild-type CCR5 alleles; 37.6% were heterozygous for wild-type and Δ2 alleles, and 4% of the SM were heterozygous for the wild-type and Δ24 alleles. Notably, 6.4% of the SM were homozygous for the smCCR5Δ2 allele, and 1.5% carried both Δ2 and Δ24 mutant alleles (Table 1). Thus, nearly 8% of animals carry two CCR5 mutant alleles encoding defective CCR5 proteins. We did not identify any SM that were homozygous for the Δ24 allele.
In this SM population, analysis of allelic frequencies showed 26% of alleles carried the novel Δ2 deletion, 3% carried the Δ24 deletion, and 71% were wild-type. Of note, the 3% frequency we found for the Δ24 allele is similar to the 4% allelic frequency described 12 years ago among SM housed at YNPRC [44]. We then used the allelic frequencies to calculate a predicted genotype distribution (Table 1 and Table S1). There was close agreement between predicted and observed genotypes (p = ns; Chi-square test), suggesting that the CCR5 alleles are in equilibrium in this population (of which ∼60% are SIV-infected), without evidence of selective pressure favoring or disfavoring any of the genotypes.
The YNPRC colony is the largest population of SM in the US and the close match between predicted and observed CCR5 genotype distributions suggested an absence of selective pressure for or against any specific genotype. However, we wished to determine the frequency with which these alleles and genotypes were present in other SM populations, so we analyzed genomic DNA obtained from 29 animals housed at the Tulane National Primate Research Center (TNPRC). Of note, many of the monkeys that founded the TNPRC colony originally came from YNPRC in the 1980s, but have been housed and bred separately since then. In this smaller population the Δ2 allele was present at a frequency of 19% and the Δ24 allele had a frequency of 5%. As shown in Table 2 and Table S2, the observed genotype frequencies in this population also do not differ from those predicted by allele frequencies.
We also asked whether the CCR5Δ2 allele was present in SM in Africa. For this analysis we amplified CCR5 genes from fecal-derived host DNA samples from 33 wild-living animals in the Tai forest of Cote d'Ivoire [45]. Five animals carried both the CCR5Δ2 and wild-type alleles, and in one animal only the Δ2 allele was detected. An additional 2 animals carried both Δ24 and wild-type alleles, while the remaining animals revealed only wild-type sequences. Because this analysis utilized fecal DNA containing limiting quantities of host DNA, we can only be certain that both alleles were captured if animals were found to be heterozygous. Thus, while it is not possible to determine a precise allele frequency, these genotypes suggest a minimum allele frequency of 9% for CCR5Δ2 and 3% for CCR5Δ24 in this population. This result indicates that the Δ2 allele is also present in wild-living SM in Cote d'Ivoire, although likely at a lower frequency than in captive animals at YNPRC.
Since overexpression studies in transfected cells suggested that neither this common smCCR5Δ2 nor the less common Δ24 proteins are expressed on the cell surface, we examined the relationship between genotypes and CCR5 expression on primary SM CD4+ and CD8+ T cells. To address this point, we analyzed CCR5 expression data that was available from animals housed at the YNPRC collected during periodic surveys between 2004 and 2009, focusing on uninfected animals, and grouped individuals according to their CCR5 genotypes. Since neither Δ2 nor Δ24 CCR5 alleles express following transfection, the two mutations were combined for the purposes of this analysis into a homozygous wild-type, heterozygous, and homozygous deletion allele groups (Figure 3).
As previously reported [18] and as shown in Figure 3, CCR5 expression is markedly greater on SM CD8+ T cells than CD4+ T cells. In both populations there was a gradation in the percentage of cells staining positive for CCR5 that correlated with CCR5 genotype status. The difference was particularly evident for CD8+ T cells, and the percentage of CCR5+/CD8+ T cells was 9.9%, 5.4% and 0.8% for wild-type, heterozygous and homozygous mutant groups, respectively (Figure 3A; p<0.0001 by Kruskal-Wallis test). This finding indicates that the CCR5Δ2 genotype acts as a determinant of CCR5 expression on CD8+ T cells. There was also a strong linear relationship between wild-type gene dosage and CCR5 expression for CD8 cells (R2 = 0.99996; p = 0.009 by Pearson's 2-tailed correlation coefficient), consistent with the absence of a dominant negative effect by the mutant alleles in transfected cells (Figure S2). A trend was also evident for CCR5 staining on CD4+ T cells among the three groups (2.8%, 1.7% and 1.2% in the wild-type, heterozygous and homozygous mutant groups, respectively), which did not reach statistical significance (p = ns by Kruskal-Wallis test) but is consistent with the notion that, although overall very low, CCR5 expression on CD4+ T cells is also regulated by CCR5 genotype.
Representative FACS plots showing CCR5 expression on CD4+ and CD8+ T cells from a homozygous wild-type, heterozygous and homozygous mutant animal are shown in Figure 3B. We think the low level (∼1%) of cells within the CCR5 gate for homozygous mutant animals likely represents background staining, rather than low levels of N-terminal expression given the result of transfection studies (Figure 2). Unfortunately, antibodies available that are directed at other epitopes in human CCR5 do not recognize SM CCR5.
In humans, the CCR5Δ32 homozygous genotype provides powerful protection against HIV-1 infection [39], [40], [42]. Therefore, we asked if animals homozygous for CCR5-null alleles were infected by SIVsmm. In order to more properly assess natural susceptibility, we restricted this analysis to YNPRC animals that were naturally infected (n = 120) and those with documented SIV-negative status (n = 72), and excluded animals in the colony known to have been experimentally infected (n = 10).
As shown in Table 3, we found that among SM with the CCR5 homozygous wild-type genotype, 65% were naturally infected while 35% were uninfected, and 62% of CCR5 heterozygous animals were infected and 38% were uninfected. Unexpectedly, among animals homozygous for CCR5-null alleles (n = 14), 50% were naturally infected and 50% were seronegative. Thus, animals lacking functional CCR5 genes are susceptible to natural SIVsmm infection. The slightly lower prevalence of SIV infection among animals in the CCR5-null group was not statistically significant (p = ns; Chi-square test). We also compared the distribution of genotypes within the SIV-negative and naturally-infected SIV populations (Table 4). Similarly, there were no significant differences in genotype distribution between the SIV serostatus groups (p = ns for each genotype group; 2-sample proportions test), although a slightly lower proportion of homozygous CCR5 mutant animals was seen in the infected animals compared with SIV-negative animals (5.8% vs. 9.7%; p = ns). Therefore, the CCR5-null genotype does not prevent natural acquisition of SIVsmm infection. Furthermore, neither heterozygosity nor homozygous null genotype appears to significantly influence SM susceptibility to SIVsmm infection in vivo, and while we cannot absolutely rule out a small effect, any protection that might be afforded by CCR5-null status would be slight. This result stands in marked contrast to the profound protective effect of CCR5Δ32 homozygosity in humans.
We also determined the genotypes of 10 animals at YNPRC that were infected experimentally. Five of these animals were homozygous for the CCR5 wild-type gene, three were heterozygotes (all W/Δ2) and two were homozygous for smCCR5Δ2. Furthermore, all but one of the TNPRC animals studied are SIVsmm-infected, including a mix of both natural infections and experimental inoculation done in earlier decades. Among the infected animals were three homozygous CCR5-null animals, while the one uninfected animal was heterozygous (W/Δ2). Therefore, SM naturally deficient in CCR5 expression are susceptible to experimental as well as natural SIVsmm infection, confirming that non-CCR5 entry pathways can mediate SM natural host infection.
We next asked if CCR5 genotype affected plasma viral loads in infected animals. Based on previously collected data, the log10 viral load (mean ± SEM) was calculated for animals in each genotype group; for animals with multiple data points available, their mean viral load (log10) was used (Figure 4). This analysis showed robust viral loads in all genotype groups, with a modest but statistically significant gradient in VL dependent on the presence of a wild-type CCR5 allele (p = 0.005 by Kruskal-Wallis test). Animals possessing two wild-type alleles exhibited the highest viral load (4.83±0.10 log10), heterozygotes showed an intermediate level (4.65±0.10 log10), and infected animals with the CCR5-null genotype had the lowest VL (4.37±0.15 log10). Thus, there is approximately 0.5 log10 difference in viral load in animals with two wild-type compared with two CCR5-null alleles. The two important points here are that homozygous mutant animals have vigorous viral replication despite lacking functional CCR5, and simultaneously exhibit a small but significant difference in plasma viral load associated with increasing CCR5 gene dosage.
CD4 counts remain stable during chronic SIVsmm infection in the vast majority of animals, but some exceptions have been noted [46], [47], so we next asked if SIV-infected animals exhibit differences in CD4 counts depending on their CCR5 genotype. However, there was no significant difference in CD4+ T cell levels among the genotype groups, whether assessed based on absolute counts (Figure S3A; p = ns, Kruskal-Wallis) or on the basis of CD4 percentage (Figure S3B; p = ns, ANOVA). These results suggest that CD4+ T cell levels are maintained similarly in SIV-infected SM regardless of CCR5 genotype.
We then asked if there was any chance that the smCCR5Δ2 mutant allele, when expressed along with CD4, could support SIV infection in vitro. We considered it unlikely but thought it was necessary to test directly given the staining patterns by CCR5 mAb 3A9 in Δ2 homozygous primary mononuclear cells (Figure 3). We also considered it important to test using SIVsmm from a CCR5Δ2 homozygous infected animal, in case viral adaptation might have enabled use of an N-terminal region alone. Therefore, we generated pseudotype virions carrying Env glycoproteins cloned directly from plasma virus of an SIVsmm-infected CCR5Δ2 homozygous animal (FNp), along with pseudotypes carrying Envs cloned from plasma of an infected wild-type animal (FFv). Use of plasma virus ensured that these Envs were derived from actively replicating virus. Target 293T cells were co-transfected with CD4 plus plasmids encoding wild-type smCCR5, smCCR5Δ2, human CCR5 or an empty vector as a control, then infected with the primary SIVsmm pseudotypes. Of note, pseudotype virions carrying the FNp 5.1 Env were considerably less infectious than the other Envs and very large amounts of this virus were required to achieve equivalent infectious inocula, which generally resulted in high levels of background for this Env.
As shown in Figure 5, wild-type smCCR5 and human CCR5 support infection of all SIVsmm variants to similar levels. In contrast, the smCCR5Δ2 allele does not support infection by any of the viruses. Importantly, CCR5Δ2 does not function as a coreceptor for Env variants from CCR5-null animals. Thus, SIVsmm infections in CCR5-null animals are mediated through pathways independent of CCR5.
The data presented here indicates that SIVsmm must be able to use entry pathways other than CCR5 for replication in vivo. It is long known that many SIV strains use a number of alternative coreceptors in addition to CCR5 in vitro, although these viruses rarely use CXCR4. We therefore tested the ability of SIVsmm envelopes to mediate infection through human CCR2b, CCR3, CCR8, GPR1, GPR15 (BOB), CXCR6 and CXCR4. This analysis employed the uncultured SIVsmm envelope glycoproteins cloned from plasma of the CCR5-null (FNp) and homozygous wild-type (FFv) infected animals.
As shown in Figure 6, all SIVsmm envelope glycoproteins tested mediated entry into cells expressing CD4 in conjunction with GPR15 and CXCR6, while GPR1 was also used but less efficiently. In contrast, none of the SIVsmm envelope glycoproteins could use CXCR4 as a coreceptor, nor CCR3 or CCR8 (data not shown). Interestingly, SIVsmm envelope glycoproteins failed to use CCR2b, an entry pathway employed by SIVrcm in RCM that typically lack CCR5 due to a high prevalence of the Δ24 mutation [43]. Furthermore, the patterns of alternative coreceptor use were similar for envelope glycoproteins derived from CCR5-null and CCR5 wild-type animals, indicating that alternative coreceptor utilization is a feature shared by SIVsmm regardless of whether the host animal expresses CCR5. While absolute luciferase production varied among experiments, GPR15 and CXCR6 typically supported levels of infection similar to that mediated by CCR5 (Figure S4), although transfected targets likely represent maximum levels of potential utilization relative to primary cells that would express these molecules at physiological levels.
We next investigated the role of non-CCR5 pathways in SIVsmm infection of primary SM cells, utilizing both the specific CCR5 antagonist maraviroc and CCR5-null PBMC derived from CCR5Δ2 homozygous animals. Maraviroc blocks chemokine signaling and HIV-1 Env entry through human and rhesus macaque CCR5 [48], [49], but blocking of SM CCR5 coreceptor function has not been reported. Therefore, we first tested the effect of maraviroc on SIVsmm entry through smCCR5 in transfected cells. As shown in Figure 7A, maraviroc blocked SIVsmm pseudotype infection of target cells expressing CD4 and smCCR5, reducing luciferase expression to the level seen with target cells expressing CD4 alone (data not shown), indicating complete blocking of smCCR5-mediated entry by maraviroc. In contrast, maraviroc did not inhibit SIVsmm entry mediated by GPR15 (Figure 7A), nor did it affect entry by VSV-G pseudotypes (data not shown), confirming that blocking is not a nonspecific effect.
Next we asked if maraviroc would affect productive infection of SM primary PBMC by infectious isolates of SIVsmm. PBMC from an uninfected CCR5 wild-type (FAk) and a CCR5-null (FAz) animals were activated with PHA for three days, and then infected in the presence and absence of maraviroc with two SIVsmm primary isolates (M923 and M951), which were both derived from CCR5 wild-type infected animals. Viral replication was measured as SIV gag p27 levels in supernatants collected periodically post-infection (Figure 7B).
We first noted that both SIVsmm isolates were able to productively infect primary PBMC in vitro from the CCR5-null SM (FAz). In these cells, there was no difference in replication associated with CCR5 blocking by maraviroc, as expected given the lack of functional CCR5 encoded by the mutant genes. This result indicates that SIVsmm infection of CCR5-null PBMC can occur independently of CCR5. We then tested CCR5 wild-type PBMC (FAk), and found that SIVsmm established productive infection in both the absence and presence of maraviroc. Furthermore, there was no difference in the level of infection achieved when CCR5 was blocked, based on p27 antigen production. This finding indicates that SIVsmm efficiently enters even CCR5-expressing SM primary PBMC through pathways independent of CCR5.
Unlike HIV-1 in humans, CXCR4 use by SIV in SM is rare. However, a few exceptions have been noted in which CXCR4 use emerged following experimental infection, which was associated with profound CD4+ T cell loss although not clinical AIDS [47]. Therefore, to ask if restricted CCR5+ target cell availability due to genetic absence of the coreceptor might be linked to CXCR4 emergence, we genotyped two infected CD4-low SM previously described in which CXCR4-using SIVsmm variants emerged [47]. Neither animal possessed a CCR5-null genotype: one was CCR5 homozygous wild type and the other was heterozygous for the CCR5Δ2 allele. Thus, the fact that acquisition of CXCR4 use by SIVsmm can occur but is not associated with animals that genetically lack CCR5 is consistent with the notion that alternative pathway-supported entry in vivo is robust and lack of CCR5 does not serve as a driving force in the rare cases with emergence of CXCR4 use.
We unexpectedly found that 8% of sooty mangabeys in a large US captive population lack functional CCR5 due to the high prevalence (29%) of mutations in the CCR5 gene that abrogate cell surface expression and SIV coreceptor function. Despite their CCR5-null status, homozygous mutant animals are susceptible to natural as well as experimental SIVsmm infection and display viral loads only modestly lower than CCR5 wild-type animals. In vitro, SIVsmm enters primary SM lymphocytes independently of CCR5, and Envs from both wild type and CCR5-null infected animals use several alternative coreceptors in addition to CCR5, but do not use CXCR4. These data indicate that both CCR5 and alternative coreceptor pathways mediate cell entry and robust viral replication in vivo. The recognition that both CCR5-dependent and independent pathways are used in the SM natural host has significant implications for understanding viral tropism in vivo and CD4+ T cell subset targeting that may regulate the outcome of natural host infection, and raises important questions about entry coreceptor use in pathogenic non-human primate models of AIDS. This finding also explains the previously obscure reason for widespread and efficient use of alternative coreceptors among the SIVmac/smm family of viruses. Finally, it provides a third example of convergent evolution resulting in disruption of CCR5 function among primates, with consequences for virus-host interactions in SM that differ from both humans homozygous for CCR5Δ32 and red capped mangabeys homozygous for CCR5Δ24.
From the original descriptions of alternative coreceptor use by SIVmac/smm viruses, and subsequently by other SIV strains, the reason for conserved use of these pathways has remained elusive [20], [21], [50]. It has been repeatedly shown that various SIV isolates can infect primary human Δ32 homozygous PBMC independent of both CCR5 and CXCR4, indicating that alternative coreceptors are expressed in a manner that supports infection in primary lymphocytes ex vivo, at least in cells of human origin [51], [52], [53], [54], [55], [56]. Our data show for the first time that the alternative coreceptors are used by SIVsmm in primary simian T cells ex vivo and, more importantly, in vivo in the natural host from which the SIVmac/smm family derived. Several important questions are raised by this finding: (1) are alternative pathways also operative in SM with wild-type CCR5 expression, or are they only relevant if CCR5 is absent; (2) what essential role or selective advantage do they provide SIVsmm in SM infection in vivo that has led to conservation of alternative coreceptor entry pathway use; (3) does alternative coreceptor use define a novel population of CCR5-negative target cells that contributes to the ability of host and virus to coexist without disease, and; (4) what role do alternative pathways play in infection of macaques, the nonhuman primate model used to study AIDS.
Prior to this confirmation that alternative pathways are used in vivo, it seemed plausible that alternative coreceptor use by SIVsmm/mac was an in vitro epiphenomenon of little biological significance. Recognizing that they are operative in vivo, it seems more likely that conservation of their use among SIV isolates reflects some role that, if not completely essential, offers a selective advantage for the virus. Comparison of viral load data among the genotype groups showed robust replication in the absence of CCR5, but a step-wise increase associated with the presence of one or two functional CCR5 alleles, and ex vivo blocking studies with maraviroc confirmed that both wild-type and CCR5-null primary PBMC possess efficient CCR5-independent entry pathways. On one hand, if the relevant coreceptor(s) are expressed only on cells that also express CCR5, the use of multiple pathways in vivo might be an example of functional redundancy acquired by SIV, reminiscent in part of the functional redundancy of the chemokine/chemokine receptor system. On the other hand, if CCR5 and alternative pathways are expressed on distinct or only partially overlapping CD4+ T cell subsets, the 0.5 log10 VL difference between Δ/Δ and W/W animals may also be consistent with separate components of plasma viremia supported by CCR5 and non-CCR5 pathways (and an intermediate gene-dosage effect in the presence of one CCR5 allele). It is notable that the degree of depletion in SM gut CD4+ T cells exceeds the proportion that express detectable CCR5 [24], and it will be important to determine if this is due to infection and targeting of CCR5-negative cells in wild-type animals mediated by other coreceptors.
As to why alternative coreceptor use is conserved among SIVsmm and related strains, it seems unlikely to result from the 8% prevalence of CCR5-null animals in SM, and more likely reflects a unique role that provides an advantage over CCR5 alone in transmission, establishment of reservoirs or other aspects of infection. Sooty mangabeys overall express very low levels of CCR5 on CD4+ T cells, which has been proposed as an evolutionary adaptive response to “protect” critical target cells and minimize pathogenesis [17], [18]. If so, acquisition of alternative coreceptor use by the virus may have reflected a “counter-measure” to maximize replication capacity, although doing so in a manner that still retains the nonpathogenic nature of infection. Thus, the use of alternative entry pathways in vivo might enable infection of a novel CCR5-negative target cell population that is more expendable than CCR5+ cells, allowing the virus to replicate efficiently without causing disease in the face of extremely restricted CCR5 expression. It will therefore be important to define the distribution of cells infected in animals with and without functional CCR5.
An important question raised by these SM findings is whether alternative coreceptors are utilized in pathogenic infection of macaques, which is widely used to model human AIDS. While CCR5 clearly plays a principal role as evidenced by substantial albeit variable viral suppression by CCR5 antagonists [57], [58], levels of CD4+ T cell infection in rhesus macaques can also substantially exceed the proportion of cells that express detectable CCR5 [23]. While it is possible that infection of apparently CCR5-negative targets reflects entry mediated by CCR5 at levels below the threshold detectable by FACS, our findings revive the question of whether it may be mediated by additional entry pathways. Few studies have attempted to address alternative coreceptor use in vivo. One report showed that mutations that abrogated GPR15 use by SIVmac had little effect on replication or pathogenesis in rhesus macaques [22]. Similarly, when pigtail macaques (Macaca nemestrina) were infected with SIVmne (also derived from SIVsmm), serial isolates exhibited decreasing ability to use alternative entry pathways when assayed in vitro [51]. On the other hand, in cynomolgus macaques (Macaca fascicularis) infected with SIVsmm, animals with progressive disease showed retention or broadening of alternative coreceptor use while those without disease progression showed narrowing of alternative coreceptor use [52]. Thus, it remains to be determined whether alternative pathways support entry in particular subsets of CD4+ T cells in the macaque model in vivo. Of note, emergence of CXCR4 use is common in HIV-1 infection of humans but exceedingly infrequent in macaques infected with SIVmac. HIV-1 rarely uses alternative coreceptors efficiently and evolution to CXCR4 use by HIV-1 is believed to result, in part, from loss of CCR5+ target cells in late stage disease. Thus, the availability of efficient alternative entry pathways may be one reason that SIV rarely evolves to use CXCR4.
The SIVsmm Envs examined here use GPR15 and CXCR6 for entry quite efficiently in vitro, and use GPR1 somewhat less efficiently. This result is concordant with coreceptor use patterns of multiple other SIV isolates [22], [23], [30]. In human blood cells, CXCR6 is highly expressed on CD4+ and CD8+ memory but not naïve T cells, and on gamma-delta T and to a lesser extent NK cells [59], although others have reported expression by CD4+ naïve T cells as well [60]. Interestingly, CXCR6 expression is regulated by T cell activation in a pattern that is tightly linked with CCR5 expression in response to some stimuli, but markedly different in response to others [59]. GPR15 is also expressed on lymphoid and myeloid cells but with less information about specific distribution patterns [20], [21], [61]. Of note, both CXCR6 and GPR15 are also highly expressed in intestinal tissues [20], [62], raising the question of whether alternative coreceptor use may be involved in mucosal events that play a central role in infection. Thus, CXCR6 and GPR15 are particularly likely candidates for mediating SM infection independent of CCR5, and further studies are required to determine which one or ones are responsible for entry into primary SM PBMC ex vivo and infection in vivo.
One of the most important priorities in HIV/AIDS research at present is understanding why infected natural hosts remain healthy while rhesus macaques infected with SIV, humans infected with HIV-1 or HIV-2, and chimpanzees infected with SIVcpz develop AIDS. Many features are shared by non-pathogenic natural host and pathogenic non-natural host infection including sustained high level viremia, vigorous immune activation during acute infection, and extensive depletion of gut mucosal CD4+ T cells. In pathogenic infections it is believed that in addition to infection and loss of short-lived T effector and T effector memory cells, damage to long-lived CD4+ Tcm populations that impairs the capacity to maintain immune cell homeostasis is a critical factor in progressive immunodeficiency [26], [27], [63]. CCR5 levels are profoundly lower on SM CD4+ T cells compared with RM and humans, which may restrict the target cells available for infection in vivo [17], [18]. More recent findings indicate that CD4+ Tcm in SM have particularly impaired CCR5 expression upon activation, and this corresponds with markedly lower levels of cell-associated infection in Tcm compared with Tem cells in SM, whereas both populations are similarly infected in RM (G.S. and M.P; unpublished observations). Thus, in addition to the role of CCR5, it will be important to define the distribution and use of other coreceptors by SIVsmm in its natural host as well as in pathogenic rhesus macaque infection. Another prominent difference between pathogenic infection and nonpathogenic natural host infection is the presence of sustained high level generalized immune activation in rhesus and humans, whereas acute infection in natural hosts is associated with transient immune activation that rapidly resolves [11], [29], [35], [64], [65], [66]. Sustained immune activation during chronic infection may be an additional factor driving T cell turnover and depletion. A principal mechanism driving chronic immune activation is believed to be translocation of microbial products due to disruption of gut mucosal barrier integrity that occurs early in infection [31]. However, gut mucosal lymphocyte infection and CD4+ T cell depletion occurs in natural host as well as pathogenic host infection [24], [33]. One potentially critical difference is the loss in human and rhesus macaque infection, but preservation in natural hosts, of mucosal CD4 Th17 cells, which play a critical role in gut mucosal immune defense [35], [36], [37], [38]. Why gut Th17 CD4+ T cells are preserved in infected natural hosts but depleted in other hosts remains to be determined, but the critical role of entry coreceptors in determining tropism and cell subset infection in vivo suggest that both CCR5 and alternative coreceptor pathways must be defined in order to understand the targeting versus protection of critical CD4+ cell subsets.
In addition to the YNPRC and TNPRC SM colonies, we also found the Δ2 allele in SM in the Tai forest of Cote d'Ivoire, confirming its presence not just in captive but also in wild-living West African animals, albeit at a lower frequency. Sooty mangabeys at YNPRC were derived from multiple sources and their history is not well documented [67], so it is difficult to know for sure from what geographic regions in West Africa these animals descended. Of note, the Δ2 allele was not described in earlier studies that reported the Δ24 allele in SM from West Africa and the YNPRC colony [43], [44]. However, CCR5 alleles in those studies were screened by PCR amplicon size, which could discriminate a 24 bp size difference but is unlikely to distinguish a 2 bp difference between the wild type and Δ2 alleles.
The identification here of SM CCR5Δ2 brings to three the number of primates known to have a high prevalence of defective CCR5 alleles. Interestingly, each of the three examples of populations with homozygous mutant CCR5 individuals shows distinct patterns of host/virus interactions. For HIV-1 in humans, which have the lowest prevalence of defective CCR5 genes (∼1% among Caucasians), CCR5 use is a stringent requirement for establishment of new infections, and its absence in the host provides almost complete protection from infection even though late-stage variants can use CXCR4 and, occasionally, other coreceptors. In RCM, which have the highest prevalence of CCR5-null individuals (≥70%), SIVrcm has adapted to use CCR2b as a coreceptor and lost the ability to use CCR5 [43]. SIVsmm infection in SM, which have an intermediate prevalence of CCR5-null individuals (8% in this population), demonstrate an intermediate relationship, in which alternative coreceptors efficiently mediate both natural and experimental infection in the absence of CCR5, but CCR5 use by the virus is retained, perhaps because both CCR5 and alternative pathways together maximize replication in the majority of animals, while the non-CCR5 pathways are required in the CCR5-null animals.
It is somewhat unexpected that SIVsmm does not use CCR2b, which is the route of entry taken by SIVrcm in the absence of CCR5. SM and RCM are closely related and sometimes considered sub-species within the same species [43], [68]. The presence of the same CCR5Δ24 allele in RCM and SM has been interpreted as indicating an origin prior to separation of these populations, although it is uncertain if the remarkably high frequency CCR5Δ24 in the RCM population reflects selective pressure exerted by an environmental or infectious cause, or founder effect [43]. Another question raised by our results is whether the SM CCR5Δ2 and the SM/RCM CCR5Δ24 emerged independently, resulting in deletions in the same region due to multiple nucleotide repeats enabling recombination, or given the overlapping sites whether Δ2 emerged first and additional events led to Δ24. Since Δ2 is not expressed, it is unlikely that further deletions would lead to any additional selective advantage, and two separate recombination deletion events in the same region of the same gene seem more probable.
What selective pressures might have led to three independent primate CCR5 deletion alleles is uncertain. CCR5Δ32 has been present in the human population for at least 3000 years [69], far longer than HIV-1, and despite considerable speculation on infectious or other pressures, both what factors fueled its emergence and when it occurred remain enigmatic [70]. In contrast, SIV has been endemic in the SM and RCM populations for much longer, although whether its entry predated separation of the populations is uncertain [68]. The CCR5 mutation is currently not essential for protection from SIV-induced pathogenesis, but it is plausible that each of the mutations result from ancestral evolutionary pressure by pathogenic SIV infection. For SM, genetic abrogation of CCR5 expression may have been an additional, complementary response to control pathogenesis along with phenotypic CCR5 downregulation [18]. If so, in both SM and RCM the virus then acquired mechanisms to circumvent the restriction, by expanding coreceptor use for SIVsmm or switching for SIVrcm, yet the hosts then acquired other additional mechanisms to avoid pathogenesis. Alternatively, it may be that all three CCR5 mutations, human, SM and RCM, reflect evolutionary adaptation to some other as-yet unidentified infectious or other environmental factor acting similarly on all three types of primates. Nevertheless, whatever its origin and frequency, the SM CCR5Δ2 mutation here expands the pathways known to support infection in this important natural host model, the identity, distribution and utilization of which must be taken into account in understanding SM infection in vivo, and the similarities or differences from RM infection that determine outcome from infection.
All animal experimentation was conducted following guidelines established by the Animal Welfare Act and the NIH for housing and care of laboratory animals and performed in accordance with Institutional regulations after review and approval by the Institutional Animal Care and Use Committees (IACUC) at the Yerkes National Primate Research Center (YNPRC) or the Tulane National Primate Research Center (TNPRC). Studies were also reviewed and approved by the University of Pennsylvania IACUC.
These studies utilized blood cells from animals housed at the YNPRC or TNPRC. Peripheral blood mononuclear cells (PBMC) were isolated from whole blood using standard density gradient separation methods. For genomic analysis, approximately 0.8×106 PBMC were lysed in DNA lysis buffer (100 mM KCl; 0.1% NP40; 20 mM Tris pH 8.4; 0.5 mg/ml proteinase K; 200ul total volume) and used as a template for PCR amplification. For infection studies, cryopreserved PBMC were thawed under standard conditions and maintained at 106 cells/ml in RPMI supplemented with 10% FBS, 1% glutamine and 1% penicillin-streptomycin, stimulated for 3 days with 5 µg/ml of phytohemagglutinin (PHA; MP Biomedical), infected and then maintained in the same media in the presence of IL-2 (50 U/ml; Novartis). Analysis of wild SM genotypes was carried out on purified DNA derived from fecal specimens collected in Cote d'Ivoire, which were previously characterized by mitochondrial DNA sequence analysis to represent 33 distinct individuals [45].
Full-length SM-CCR5 genes were amplified by PCR from SM genomic DNA using high fidelity DNA polymerase (Phusion; Finnzymes) and primers based on conserved 5′ and 3′ regions of published SM CCR5 coding sequences (forward: 5′-ATG GAC TAT CAA GTG TCA AGT CCA ACC-3′; reverse: 5′-TCA CAA GCC AAC AGA TAT TTC CTG CTC C-3′). PCR reactions contained Phusion polymerase (1 unit) in HF buffer, primers (0.5 uM), dNTPs (0.2 mM) and 200 ng of purified genomic DNA as template in a 50 ul reaction volume. Thermocycling conditions: initial denaturation at 98°C for 45 seconds, followed by 20 cycles of 98°C for 10 seconds, 71°C for 30 seconds and 72°C for 90 seconds, with a final extension step of 72°C for 10 minutes. PCR amplicons were column purified (QIAquick PCR Purification kit; Qiagen) and then used in a second PCR reaction employing primers that incorporated a HindIII restriction site at the 5′ end of the coding region and a BamHI restriction site at the 3′ end of the CCR5 coding sequence (forward: 5′-GCT GCT ATA AGC TTC CAC CAT GGA CTA TCA AG-3′; reverse: 5′-AGC GAG CGG ATC CTC ACA AGC CAA CAG ATA-3′; restriction sites underlined). Thermocycling conditions for the second PCR reaction were: an initial denaturation step at 98°C for 45 seconds, followed by 5 cycles of 98°C for 10 seconds, 71°C for 45 seconds and 72°C for 60 seconds, followed by 15 cycles of 98°C for 10 seconds, 76°C for 45 seconds and 72°C for 60 seconds, and final extension at 72°C for 10 minutes. CCR5 amplicons were cloned into the expression plasmid pcDNA3.1+ (Invitrogen) using HindIII and BamHI, and screened by restriction analysis followed by sequence confirmation. Sequences of the smCCR5Δ2, smCCR5Δ24 and smCCR5 wild-type genes cloned here have been deposited in Genbank (accession numbers HM246694, HM246695 and HM246693, respectively).
A two-step PCR-based genotyping assay was developed that identifies CCR5 wild-type and CCR5Δ2 alleles based on differential primer annealing, using genomic DNA from lysates of SM PBMC. Genomic DNA was subject to PCR amplification using the first round primers described above, to amplify the entire CCR5 coding region, in reactions that contained Platinum Taq polymerase (1 unit; Invitrogen), 1.5 mM MgCl2, primers (0.5 uM each), dNTPs (0.2 mM), 1–3 ul DNA lysate and Taq buffer in 25 ul reaction volumes. Thermocycling conditions were: initial denaturation at 98°C for 45 second followed by 10 cycles of 94°C for 20 seconds, 68°C for 45 seconds, 72°C for 60 seconds and final extension at 72°C for 10 minutes. The product of this reaction (2.5 ul) was then used as a template for two separate second-round amplifications, each of which used a common downstream primer but different upstream primers specific for the wild-type and Δ2 alleles, respectively (forward CCR5 wild-type: 5′-ATC ACT TGG GTG GTG GCT-3′; forward CCR5Δ2: 5′-ATC ACT TGG GTG GTG CGT-3′; common downstream: 5′-GGT GTT CAG GAG AAG GAC AAT GTT G-3′). The second round PCR reaction used the same conditions as the first reaction. Products of the wild-type and Δ2 amplification reactions were visualized by 2% agarose gel electrophoresis and ethidium bromide staining, which demonstrated a 325 base pair product following amplification with wild-type or Δ2 primer pairs, or both for heterozygotes (Figure S1A).
Each animal was also screened for the CCR5Δ24 deletion allele with a two-step PCR-based assay. Products of the first round PCR reaction described above, were subjected to nested amplification with inner primers (forward: 5′-GGC TAT CGT CCA TGC TGT GT-3′; reverse: 5′-GAC CAG CCC CAA GAT GAC TA-3′) and thermocycling conditions as follows: initial denaturation at 94°C for 45 seconds, followed by 25 cycles of 94°C for 20 seconds, 59°C for 45 seconds, 72°C for 60 second, and a final extension at 72°C for 10 minutes. Products were visualized by 3% agarose gel electrophoresis and ethidium bromide staining, yielding a 205 base pair product from the Δ24 allele, which was easily distinguishable from the 227–229 bp product from the wild-type and CCR5 Δ2 alleles (Figure S1B).
As a secondary confirmation of genotypes established by PCR screening, direct bulk sequencing was carried out on PCR amplified genomic DNA. Amplification was done using the outer primer set as described above, except that Phusion high fidelity polymerase was used for 30 cycles. Products were column-purified and sequenced. Genotypes were verified by manual inspection to confirm the presence of uniform sequences for homozygous animals, or detection of expected frameshifts resulting in overlapping sequences for heterozygous animals (Figure S1C).
For fecal-derived samples, 0.5 ug of purified DNA (of which only a fraction reflected host-derived DNA) was PCR amplified for 35 cycles using first round outer primers as described above, and then subjected to nested amplification for 30 cycles using the inner primer set described above. Products were then subjected to direct sequence analysis.
SIVsmm Env-mediated entry was analyzed using luciferase-expressing reporter viruses pseudotyped with envelope glycoproteins of interest. Pseudotype viruses were generated by co-transfecting 293T cells with a plasmid encoding the NL4-3-based env-deleted luciferase-expressing virus backbone (pNL-luc-E−R+) [71] along with expression plasmids encoding SIVsmm, SIVmac, HIV-1 or VSV-g envelope glycoproteins. Cells were transfected overnight using Fugene (Roche) and washed the next day to remove residual transfection reagent. Supernatants were collected 2 days later, clarified by centrifugation and stored at −80°C until use. Pseudotype viruses were quantified based on HIV-1 Gag p24 antigen ELISA (PerkinElmer) and virion infectivity measured on U87 cells stably expressing CD4 and CCR5. Inocula were then standardized on the basis of infectivity in U87/CD4/CCR5 cells (1×106 relative light units; RLU).
The ability of pseudotype viruses to use different coreceptors was assayed in target 293T cells expressing CD4 and the coreceptor of interest. Target cells were prepared by co-transfection with expression plasmids carrying CD4 and the desired co-receptor (1ug of each plasmid) using Fugene. Cells were re-plated one day post-transfection at 2×104 cells/well in 96-well plates and then infected the following day with pseudotype reporter viruses using equivalent inocula (1×106 RLU) by spin inoculation for 2 hours at 1200G. Three days later cells were lysed (0.5% Triton X-100 in PBS) and infection quantified on the basis of luciferase production in target cells, determined by adding an equal volume of luciferase substrate (Promega) and measuring luciferase activity in RLU with a luminometer.
SIV envelopes used in pseudotype infections were generated from plasma of SIVsmm-infected CCR5 wild-type (FFv) and Δ2 homozygous (FNp) SM by single genome amplification (SGA) using methods and protocols previously described and cloned into pcDNA3.1 using Topo TA (Invitrogen) [72], [73]. Infectious SIVsmm strains M923 and M951 were isolated as previously described [74] and stocks were prepared in primary SM PBMC.
293T cells were transfected with wild-type or mutant forms of CCR5 using Fugene according to manufacturer's instructions. One day later cells were detached by incubation in PBS containing 2 mM EDTA, washed in FACS buffer (PBS containing 1% FBS and 0.1% sodium azide), and stained with the CCR5 monoclonal antibody, clone 3A9-[APC] (BD Pharmingen) or isotype-matched control. Cells were analyzed using a FACS Caliber flow cytometer (BD Biosciences) and FloJo software (Tree Star, Inc.) to determine CCR5 surface expression.
PHA-stimulated SM PBMC were plated in 96-well plates at 2.5×105 cells/well, incubated for 1 hour in the presence or absence of the CCR5 antagonist maraviroc (15 uM; Pfizer), and then infected with SIVsmm strains (M923 and M951) by spin inoculation (1200×g for 2 hours) followed by overnight incubation. The next day cells were washed in PBS and maintained in media containing IL-2 (50 U/ml), with or without maraviroc (15 uM). Cell supernatants were collected periodically for 3 weeks and replication measured by SIV Gag p27 antigen in cell supernatant by ELISA (Advanced BioScience Laboratories).
Clinical data on infected and uninfected SM housed at YNPRC has been reported previously for surveys carried out in 2004–5 and 2006–7 and 2008–9 [75]. CD3, CD4 and CCR5 expression on PBMC were analyzed by FACS and plasma viral loads were measured by real-time PCR as described [11], [75]. For purposes of analysis, any undetectable viral loads were set at 75 copies, the lower limit of detection. In cases where animals seroconverted during the period of observation, data from prior to infection was included with uninfected animals, while data from after infection was included with infected animals. If multiple measurements were available for individual animals for any parameters, mean values were utilized.
Virological and immunological data were compared between genotype groups using ANOVA or Kruskal-Wallis test followed by the Dunn's multiple comparison test for multiple groups. The two-sample proportions test and Chi-Square test were used for comparison between independent groups. Statistical tests were performed using Prism 4.0 software and OpenEpi: Open Source Epidemiologic Statistics for Public Health, Version 2.3. Data were considered significant when P-value was below 0.05.
|
10.1371/journal.pbio.2006872 | Candida albicans biofilm–induced vesicles confer drug resistance through matrix biogenesis | Cells from all kingdoms of life produce extracellular vesicles (EVs). Their cargo is protected from the environment by the surrounding lipid bilayer. EVs from many organisms have been shown to function in cell–cell communication, relaying signals that impact metazoan development, microbial quorum sensing, and pathogenic host–microbe interactions. Here, we have investigated the production and functional activities of EVs in a surface-associated microbial community or biofilm of the fungal pathogen Candida albicans. Crowded communities like biofilms are a context in which EVs are likely to function. Biofilms are noteworthy because they are encased in an extracellular polymeric matrix and because biofilm cells exhibit extreme tolerance to antimicrobial compounds. We found that biofilm EVs are distinct from those produced by free-living planktonic cells and display strong parallels in composition to biofilm matrix material. The functions of biofilm EVs were delineated with a panel of mutants defective in orthologs of endosomal sorting complexes required for transport (ESCRT) subunits, which are required for normal EV production in diverse eukaryotes. Most ESCRT-defective mutations caused reduced biofilm EV production, reduced matrix polysaccharide levels, and greatly increased sensitivity to the antifungal drug fluconazole. Matrix accumulation and drug hypersensitivity of ESCRT mutants were reversed by addition of wild-type (WT) biofilm EVs. Vesicle complementation showed that biofilm EV function derives from specific cargo proteins. Our studies indicate that C. albicans biofilm EVs have a pivotal role in matrix production and biofilm drug resistance. Biofilm matrix synthesis is a community enterprise; prior studies of mixed cell biofilms have demonstrated extracellular complementation. Therefore, EVs function not only in cell–cell communication but also in the sharing of microbial community resources.
| Candida albicans—the most common fungal pathogen in humans—often grows as a biofilm, resulting in an infection that is difficult to treat. These adherent communities tolerate extraordinarily high concentrations of antifungals due in large part to the protective extracellular matrix. The present study identifies extracellular vesicles (EVs) that are distinct to biofilms. These EVs deliver the functional extracellular matrix and are essential for resistance to antifungals. Our findings not only reveal a coordinated mechanism by which the defining trait of the biofilm lifestyle arises but also identify a number of potential therapeutic targets.
| Vesicles are released externally by cells of bacteria, archaea, and eukaryotes [1–3]. These extracellular vesicles (EVs) deliver cargo of RNA and protein that is protected by a surrounding lipid bilayer. Classes of EVs have been distinguished based upon their size, cargo, and mechanisms of biogenesis [1–3]. Functional analysis has shown that EVs play diverse biological roles in delivery of effectors to target cells. For example, during Drosophila wing development, secretion of the morphogenic effector Hedgehog in EVs is required for activation of many of its target genes [4]. For many bacterial pathogens, toxin delivery via EVs causes host cell damage or lysis [1]. In the case of the eukaryotic protozoan Trypanosoma brucei, EVs orchestrate community escape from sources of environmental stress [5]. The purpose of EV secretion is thus tailored to each organism's biology and environmental context.
Microorganisms exist predominantly in surface-associated communities called biofilms, which typically have high cell density and include an extracellular polymeric matrix [6]. Biofilm cells are notorious for their resistance to antimicrobial treatments [7], a property often determined by multiple mechanisms [8]. Our interest is in the eukaryotic microorganism Candida albicans, which poses a severe threat to hospitalized patients with vascular devices due to its capacity for biofilm formation [9, 10]. Candida species proliferate on the surface of these devices as a biofilm [11–13]. Candida biofilm cells resist available drug therapies [14], and thus, the only currently effective therapy is removal of medical devices, which is often impossible for critically ill patients [15]. One of the central determinants of C. albicans (mating type locus [MTL] a/α) biofilm drug resistance is a mannan–glucan complex in the extracellular matrix [16, 17]. Our findings reported here show that EVs promote assembly of the mannan–glucan complex that leads to drug resistance. We suggest that drug resistance of other microbial biofilms may also rely upon the efficient sharing of community resources as EV cargo.
We have reported that C. albicans biofilm extracellular matrix includes a significant phospholipid component [18], a finding that might indicate the presence of EVs in the matrix material. In support of this idea, we observed numerous <100-nm spheres on the surface of biofilm cells (Fig 1A) and embedded in the extracellular matrix (Fig 1B). EVs, isolated from biofilm [19, 20] and imaged by cryoTEM, were enriched for an exosome population based upon size [21] (Fig 1C), though other vesicle types may be included in the preparation. Time course studies revealed that vesicle production peaks at 48 h after biofilm initiation (Fig 1D). These kinetics paralleled the time course of both biofilm cell accumulation and matrix deposition [22]. Our results indicate that C. albicans, like many other microbes [1, 23], produces biofilm EVs.
EVs are known to be produced by free-living planktonic cells of numerous fungi, including C. albicans [1, 24, 25]. We assessed the similarity of biofilm and planktonic EVs through comparisons of their sizes and composition. The present observations with C. albicans are consistent with studies of Saccharomyces cerevisiae [26] revealing the production of two populations of planktonic EVs (Fig 1E). There is a 30–200-nm diameter population that corresponds in size to exosomes and a larger 200–1,000-nm diameter population that corresponds in size to microvesicles [26]. In contrast, biofilm EVs comprise predominantly a 30–200-nm diameter exosome-sized population (Fig 1F).
Proteomic analysis revealed that planktonic and biofilm EVs have a considerable proportion of distinct cargo, with 34% of the proteome being unique to the biofilm state (Fig 2A–2C and S1 Table). In addition, many proteins shared by vesicles from both sources were 10- to 100-fold more abundant in the biofilm EVs. Our results indicate that EVs produced by biofilms are distinct from those of planktonic cells.
The composition of biofilm EVs pointed toward two prospective roles in biofilm extracellular matrix biogenesis. First, vesicle composition shows a high degree of similarity with matrix composition protein (Fig 2D–2F) and polysaccharide content (Fig 2G and 2H), suggesting that vesicles may be a major source of matrix material. The protein comparison suggests that up to 45% of the proteins in the biofilm matrix may be delivered by vesicles (Fig 2F and S2 Table). Polysaccharide analysis revealed a predominance of mannan and glucan, two major matrix components, in vesicle cargo by gas chromatography, which identified both components in a percent ratio of 84.0 ± 1.6/3.2 ± 1.0 in vesicles and 44.3 ± 4.2/8.8 ± 1.2 in the matrix, respectively. The major mannan component of the complex displayed structural similarity to the biofilm matrix mannan–glucan complex by 1H NMR in Fig 2G and S3 Table and 2D 1H-13C NMR in Fig 2H and S4 Table, a determinant of biofilm associated drug resistance [27]. Thus, biofilm vesicles may deliver cargo that forms the extracellular matrix. Comparative analysis of the lipid composition of biofilm EVs and matrix revealed similarity in the sphingolipid and phospholipid components, particularly in phosphatidylcholine, phosphatidylinositol, and phosphatidylethanolamine (Fig 2I). However, the neutral lipid component in the extracellular matrix appeared distinct and likely reflects an additional vesicle-independent mechanism of delivery for the remaining lipid constituents.
A second possible role is that vesicle cargo has a catalytic function in matrix macromolecule synthesis. Specifically, one of the enriched functional ontology categories for the biofilm EV proteome was polysaccharide modification (Fig 2A–2F). These observations suggest that biofilm EVs may deposit cargo that contributes directly to matrix structure, and they may also provide catalytic activities that engage in matrix polysaccharide synthesis.
We sought to test our hypothesis that biofilm EVs function in matrix biogenesis. The size range of biofilm EVs suggests that they are exosomes [21], and in other eukaryotes, exosome production is governed by the endosomal ESCRT pathway [21]. In fact, we note that biofilm vesicle cargo includes ESCRT subunits Hse1 and Vps27 (S1 Table). We identified 21 C. albicans ESCRT subunit homologs to S. cerevisiae and created homozygous deletion mutants (Fig 3A and 3B). Sixteen of the mutants showed decreased vesicle production (Fig 3B). We note that exosome production depends upon only a subset of ESCRT subunits in other eukaryotes [3, 21] in keeping with our observations for C. albicans. The ESCRT mutants with reduced EV production enabled us to test whether biofilm vesicles have a role in biofilm matrix biogenesis and function.
We screened the ESCRT vesicle–defective mutants for biofilm matrix–associated phenotypes. All mutants produced a biofilm structure, but a subset had prominent defects. Seven of the ESCRT mutants exhibited hypersusceptibility to the antifungal fluconazole during biofilm growth (Fig 3C). The enhanced susceptibility biofilm phenotype was reversed in each of these ESCRT mutants for which a WT allele was introduced (despite multiple attempts, we did not successfully construct a VPS2 complemented strain). This change in drug susceptibility was biofilm specific, as planktonic susceptibility was similar in WT and these ESCRT mutants (MIC range 0.25–0.5 μg/ml). The clinical relevance of these observations was confirmed via demonstration of congruent drug-susceptibility phenotypes in the rat vascular catheter biofilm model [28] (Fig 3D). Our previous studies have shown that biofilm matrix sequesters antifungals to promote drug resistance [16, 17, 29], and we verified that the six of the seven drug-susceptible ESCRT mutants were also defective in fluconazole sequestration (Fig 3E). We speculate that the sole ESCRT mutant that did not exhibit altered drug sequestration (DOA4) may reflect a difference in vesicle cargo or perhaps a matrix-independent resistance mechanism. Drug sequestration has been linked to matrix quantity and presence of a mannan–glucan complex (MGCx). Each of the ESCRT mutants with vesicle and drug-susceptibility defects similarly displayed defects in matrix mannan and glucan quantity (Fig 3F). As the vesicles alone did not sequester antifungals (S1 Fig), we reason this phenomenon is due to vesicle matrix delivery. These extracellular matrix defects are also demonstrated visually by the absence of matrix that adorns WT biofilms for each of the seven vesicle mutants (Fig 3G).
We considered two models for the relationship between ESCRT function, biofilm EVs, and matrix biogenesis. One model is that biofilm EVs have a direct role in matrix biogenesis; ESCRT defects cause matrix defects by reducing the levels of vesicles or packaging of functionally relevant cargo. An alternative model is that EVs have no role in matrix biogenesis; ESCRT defects cause matrix defects due to indirect effects. The second model stems from the growing appreciation that ESCRT machinery, with its central role in organelle physiology, has impact on diverse aspects of cell biology [30]. We used a “vesicle add-back” protocol to test these models (Fig 4A). Specifically, if vesicles have a direct role in matrix biogenesis, then providing WT biofilm vesicles to a vesicle-defective ESCRT mutant should restore matrix production and matrix-associated phenotypes. Remarkably, the addition of the WT vesicles to drug-susceptible ESCRT mutants increased drug resistance dramatically (Fig 4B). Furthermore, the addition of WT biofilm vesicles restored biofilm matrix architecture and quantities of the key mannan–glucan components (Fig 4C and 4D). These results support the first model: a subset of ESCRT subunits promote matrix biogenesis and function through their role in biofilm EV production.
Among the proteins in biofilm EVs, several have previously defined roles in biofilm matrix biogenesis and specifically matrix polysaccharide modification (S2 Table) [16, 27]. We considered a model in which the presence of these proteins as vesicle cargo is central to their functional activity; they are “functional passengers.” An alternative model is that they are “coincidental passengers” in vesicles and that their true function is vesicle independent. For example, they may function in matrix biogenesis at intracellular sites or after conventional secretion into the extracellular milieu. We deployed our vesicle add-back protocol to test these models, using mutants in cargo proteins putative glycanosyltransferase (Phr1) and putative endo-beta-D-glucosidase (Sun41), which act in the glucan modification pathway (Fig 4E) [16, 31]. Remarkably, addition of WT vesicles to these drug susceptible cargo mutants restored drug resistance. Control studies in which vesicles from the Phr1 and Sun41 mutants were added to the respective mutants did not alter the fluconazole susceptibility phenotype. These results favor the functional passenger model—that cargo proteins function to confer biofilm drug resistance as vesicle components rather than through some vesicle-independent activity.
Our results indicate that biofilm growth of C. albicans results in a distinctive EV population and cargo. These findings echo studies of bacterial and eukaryotic cells that show that EV properties reflect environmental and developmental signals [1, 3]. Our findings also add a new facet to the understanding of EV function: whereas prior studies have shown a role for EVs in cell–cell signaling, our studies reveal a role for EVs in the sharing of community resources [27], that of biofilm matrix material. Matrix is a pivotal determinant of C. albicans biofilm drug resistance, and our results reveal EV-dependence for drug resistance both in vitro and in an animal biofilm infection model. Our findings suggest that EV-based therapeutics [32] may be a useful new platform for antibiofilm strategies.
All animal procedures were approved by the Institutional Animal Care and Use Committee at the University of Wisconsin according to the guidelines of the Animal Welfare Act, The Institute of Laboratory Animal Resources Guide for the Care and Use of Laboratory Animals, and Public Health Service Policy. The approved animal protocol number is DA0031.
The parent strain C. albicans SN152 (MTL a/α) was used to create homozygous deletion strains (S6 Table) using a SOE-PCR-based disruption cassette method, employing histidine and lysine auxotrophic markers [33]. PCR with primers listed in S5 Table was used to verify genotypes. Complementation of mutant strains with a single gene-of-interest copy used selection for arginine prototrophy. Transformants were selected on minimal medium with the corresponding auxotrophic supplements. Both planktonic and biofilm cultures were grown in RPMI 1640, buffered with 4-morpholinepropanesulfonic acid (MOPS) for all experiments described below [34].
One of four in vitro biofilm models was used, including a 96-well or 6-well polystyrene plate, polystyrene roller bottle, or glass coverslip. Biofilm drug susceptibility was assessed using the 96-well flat-bottom polystyrene plate assay [35–38]. Matrix composition assessment utilized the 6-well plate assay. Biofilm architecture was imaged using scanning electron micrograph (SEM) using a coverslip biofilm assay. Matrix biochemistry was determined from biofilms growing, using a rolling bottle system [34]. A minimum of three biological replicates were performed for each assay.
Biofilm matrix for matrix biochemical analysis was grown using a rolling bottle biofilm model [34]. After 48 h of growth, media was removed, and the Candida biofilms were dislodged from the roller bottle surface by spatula. The intact biofilm was then gently sonicated to remove matrix from cells (sonication with a 6-mm microtip at 20 kHz with an amplitude of 30% for 8 min), followed by centrifugation to separate fungal cells from the matrix. The isolated matrix was then lyophilized. Matrix was similarly isolated from 6-well biofilm plates [16].
EVs were isolated from both planktonic cultures and large-scale biofilms grown in polystyrene roller bottles [34]. The culture media was removed from the bottles, filter sterilized, and concentrated down to 25 ml using a Vivaflow 200 unit (Sartorius AG, Goettingen, Germany) equipped with a Hydrosart 30 kDa cut-off membrane. The sample was centrifuged at 10,000 × g for 1 h at 4°C to remove smaller cellular debris. The pellets were discarded, and the resulting supernatant was centrifuged again as described above. The resulting supernatant was then centrifuged at 100,000 × g for 1.5 h at 4°C. The supernatants were then discarded, and the pellet was then resuspended in phosphate-buffered saline (PBS) (pH 7.2). Next, the sample was subject to size exclusion chromatography on a HighPrep 16/60 Sephacryl S-400 HR column (GE Life Sciences) pre-equilibrated with PBS (pH 7.2) containing 0.01% NaN3. All chromatographic separation steps were performed at room temperature on the high-performance liquid chromatography ÄKTA-Purifier 10 system (Amersham Biosciences AB, Uppsala, Sweden).
EVs were quantified using a combination of imaging flow cytometry, image confirmation, and fluorescence sensitivity in low-background samples, as previously described [39, 40]. Prior to analysis, samples were stained with carboxyfluorescein succinimidyl ester (CSFE) and 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (Dil) at 37°C for 90 min. Excessive dye particles were removed from stained vesicles using illustra microspin G-50 columns (GE Healthcare). All samples were analyzed on the ImageStreamX Mk II flow cytometry system from Amnis Corporation (Seattle, Washington, United States) at ×60 magnification, with default low flow rate/high sensitivity using the INSPIRE software.
The mean particle size of the vesicles dispersions were determined using a Zetasizer Nano-ZS (Malvern Instruments, Malvern, United Kingdom). In order to obtain the optimum light scattering intensity, 10 μl of the vesicles suspension was added to 990 μl of PBS. All the measurements were carried out in triplicate at 25°C [41].
For SEM of biofilms, 40 μl of an inoculum of 108 cells/ml in RPMI–MOPS was added to the coverslips and incubated for 60 min at 37°C. 1 ml RPMI–MOPS was added to each well, and the plates were incubated at 37°C for 20 h. One ml fixative (4% formaldehyde, 1% glutaraldehyde in PBS) was then added to each well prior to incubation at 4°C overnight. Coverslips were then washed with PBS prior to incubation for 30 min in 1% osmium tetroxide. Samples were then serially dehydrated in ethanol (30% to 100%). Critical point drying was used to completely dehydrate the samples prior to palladium-gold coating. Samples were imaged on a SEM LEO 1530, with Adobe Photoshop 7.0.1 used for image compilation [27].
For cryoTEM, 3 μl of sample suspensions were pipetted onto a glow-discharged 200 mesh copper grid with a lacey carbon support film (EMS, 1560 Industry Road, Hatfield, Pennsylvania, 19440, US, #LC200-CU). Before sample application, the grid was mounted on a tweezer in the Vitrobot (FEI, 5350 NE Dawson Creek Drive, Hillsboro, Oregon, 97124, US, model MarkIII). In an automated sequence, excess fluid was blotted off, and the grid was plunge frozen in liquid ethane. Once frozen, the grid was mounted in a precooled cryo transfer sample holder (Gatan, 780 Commonwealth Drive, Warrendale, Pennsylvania 15086, US, model 626) and inserted into the TEM (Hitachi Ltd., 4026, Kuji-cho, Hitachi-shi, Ibaraki, 319–12, Japan, model HT7700). The samples were observed at 120 kV acceleration voltage, and the sample temperature was kept at −170°C.
Enzymatic “in liquid” digestion and mass spectrometric analysis was done at the Mass Spectrometry Facility, Biotechnology Center, University of Wisconsin–Madison. 200 μg of matrix proteins were extracted by precipitation with 15% TCA/60% acetone and then incubated at −20°C for 30 min. The matrix or vesicle preparation was centrifuged at 16,000 × g for 10 min, and the resulting pellets were washed twice with ice-cold acetone, followed by an ice-cold MeOH wash. Pelleted proteins were resolubilized and denatured in 10 μl of 8 M urea in 100 mM NH4HCO3 for 10 min, then diluted to 60 μl for tryptic digestion with the following reagents: 3 μl of 25 mM DTT, 4.5 μl of acetonitrile, 36.2 μl of 25 mM NH4HCO3, 0.3 μl of 1M Tris-HCl, and 6 μl of 100 ng/μl Trypsin Gold solution in 25 mM NH4HCO3 (Promega Co., Madison, WI). Digestion was conducted in two stages, first overnight at 37°C, then additional 4 μl of trypsin solution were added and the mixture was incubated at 42°C for an additional 2 h. The reaction was terminated by acidification with 2.5% TFA to a final concentration of 0.3% and then centrifuged at 16,000 × g for 10 min. Trypsin-generated peptides were analyzed by nanoLC-MS/MS using the Agilent 1100 nanoflow system (Agilent, Palo Alto, CA) connected to a hybrid linear ion trap-orbitrap mass spectrometer (LTQ-Orbitrap, Thermo Fisher Scientific, San Jose, CA) equipped with a nanoelectrospray ion source. Capillary HPLC was performed using an in-house fabricated column with an integrated electrospray emitter, as described elsewhere [42]. Sample loading and desalting were achieved using a trapping column in line with the autosampler (Zorbax 300SB-C18, 5 μm, 5 × 0.3 mm, Agilent). The LTQ-Orbitrap was set to acquire MS/MS spectra in a data-dependent mode as follows: MS survey scans from 300 to 2,000 m/z were collected in profile mode with a resolving power of 100,000. MS/MS spectra were collected on the five most abundant signals in each survey scan. Dynamic exclusion was employed to increase the dynamic range and maximize peptide identifications. Raw MS/MS data were searched against a concatenated C. albicans amino acid sequence database using an in-house MASCOT search engine [43]. Identified proteins were further annotated and filtered to 1.5% peptide and 0.1% protein false-discovery-rate with Scaffold Q+ version 3.0 (Proteome Software Inc., Portland, Oregon) using the protein prophet algorithm [44].
The C. albicans vesicle and matrix proteomes were analyzed using the Kyoto Encyclopedia of Genes and Genomes (KEGG) [45, 46]. Each protein predicted from the C. albicans genome assigned a KEGG Ontology ID (KOID) was obtained, and the specific pathway and superpathway membership information retained. This was then correlated with the experimental proteome data, and the number of proteins expressed within a given pathway was then determined. Tabulated proteins were presented as a percentage out of the total number of proteins predicted to belong to a given pathway from the C. albicans genome, as determined by KEGG. The visualization of relative quantities of biofilm proteins was also done using KEGG protein functional categorization. On the basis of this hierarchical classification scheme, Voronoi treemaps were constructed [47]. This approach divides screen space according to hierarchy levels in which the main functional categories determine screen sections on the first level, subsidiary categories on the second level, and so forth. The polygonic cells of the deepest level represented functionally classified proteins and were colored according to relative abundance of each protein that was determined based on total counts of corresponding trypsin-digested peptides.
Lipids were extracted from the desalted lyophilized EV or matrix powder with a mixture of CHCl3/MeOH (2:1, by vol) containing 0.1 g/l BHT. The sample was vortexed, incubated in the dark for 2 h at room temperature, and then centrifuged. The separated layer of organic solvents was removed, and the pellet was washed with 2 ml of CHCl3/MeOH (2:1, by vol) and centrifuged. The collected lipid extracts were combined and dried under a stream of nitrogen. After drying, the sample was reconstituted in 0.5 ml of CHCl3/MeOH (2:1, by vol.) and subjected to TLC separation on 20 cm × 20 cm silica gel Si60 plates. Neutral lipids were separated in hexane/ethyl ether/AcOH (90:20:1, by vol), which yielded triacylglycerols, sterol esters, free fatty acids, and a pool of immobile phospholipids. The latter group was scrapped off the plate, extracted from the silica gel, and subjected to another TLC separation in CHCl3/MeOH/AcOH/H2O (50:37.5:3.5:2, by vol). This step yielded four classes of glycerolipids (phosphatidylcholine, phosphatidylethanolamine, phosphatidylserine, and phosphatidylinositol) and one class of sphingolipids (sphingomyelins). Lipids were visualized under UV light after spraying plates evenly with a 0.2% solution of fluorescein in EtOH. All isolated lipid classes were scraped off their silica gel plates and re-extracted with CHCl3/MeOH (4:1, by vol) containing 0.1g/l BHT. Samples were vortexed, incubated overnight at room temperature, and then centrifuged in order to remove silica gel particles. 100 μl of 0.05 mg/ml pentadecanoic acid was added to each sample and the organic solvents were evaporated under nitrogen. Next, isolated lipids were subjected to methylation in the presence of 0.5 ml of 14% BF3 in MeOH. Vials containing the processed lipids were boiled. After cooling, the samples were mixed with 1 ml hexane and 0.5 ml H2O, vortexed, and centrifuged. The top hexane layer containing methyl ester derivatives was transferred to a new clean glass tube, dried under nitrogen, resuspended in 100 μl hexane, and transferred to GC vials. Fatty acid methyl esters were identified by gas chromatography using a Hewlett-Packard 5890 equipped with a capillary column coated with DB-225 (30-m length, 0.25-mm internal diameter, 0.25 μm; Agilent Technologies, Inc., Wilmington, Delaware). Peaks were identified by a comparison of retention times with a set of authentic fatty acid standards provided by Supelco. The abundance of fatty acids was calculated from the relative peak areas [18].
Delipidated vesicle and matrix pellets containing carbohydrates and proteins were washed twice with acetone, dried under a stream of nitrogen, and reconstituted in 3 ml of 20 mM bis-Tris/HCl (pH 6.5) loading buffer. Aliquots were chromatographically desalted on a HiPrep 26/10 Desalting column (GE Healthcare Life Sciences, Uppsala, Sweden) and then separated on an anion exchanger HiPrep 16/10 DEAE FF column (GE Healthcare Life Sciences) equilibrated with 20 mM bis-Tris/HCl (pH 6.5). Carbohydrate positive flow-through fractions were pooled together, lyophilized, resuspended in 15% acetonitrile in 150 mM ammonium bicarbonate, and applied to gel filtration on a HighPrep 16/60 Sephacryl S-300 HR column (GE Healthcare). All chromatographic separation steps were performed at room temperature on the high-performance liquid chromatography ÄKTA-Purifier 10 system (GE Healthcare Life Sciences).
Sugars were converted to alditol acetate derivatives according to the procedure described previously [48]. Monosugar alditol derivatives were identified and quantified by GLC-FID on a Shimadzu GC-2010 system (Shimadzu Co., Kyoto, Japan) using a (50% cyanopropylphenyl) methylpolysiloxane column (#007–225; 30 m × 0.25 mm with 0.25 μm film thickness,) (Quadrex Co., Woodbridge, Connecticut).
The samples were dissolved in 100 μl water and precipitated by addition of 900 μl EtOH. After centrifugation, the precipitate was dried, dissolved in D2O (99.9% D, Sigma-Aldrich), and lyophilized. The sample was then dissolved in 280 μl D2O (99.96% D, Cambridge Isotope Laboratories) containing 0.5 μl acetone and placed into a 5-mm NMR tube with magnetic susceptibility plugs, matched to D2O (Shigemi). NMR experiments were recorded at 65°C on an Agilent Inova-600 spectrometer equipped with a 5-mm cryoprobe. The 1-D proton experiment was acquired in 8 transients with water presaturation. The 2-D COSY experiment was collected with gradient enhancement in 400 increments of 8 transients each. The 2-D TOCSY and NOESY experiments were acquired with water presaturation in 128 increments of 16 transients each. Spinlock time in TOCSY was 80 ms, and mixing time in NOESY was 200 ms. The gradient-enhanced 1H-13C HSQC experiment with adiabatic 180° carbon pulses and multiplicity editing was acquired in 128 increments of 64 transients each, with a spectral width of 18091 Hz in the carbon dimension. The gradient-enhanced 1H-13C HMBC experiment with adiabatic 180° carbon pulses was acquired in 128 increments of 128 transients each, with a spectral width of 18,091 Hz in the carbon dimension. Chemical shifts were measured relative to DSS at 0 ppm in both proton and carbon scales by setting the chemical shift of internal acetone to 2.218 ppm (proton) and 33.0 ppm (carbon). Chemical shifts assignments reported in S3 and S4 Tables were performed based on literature values reported elsewhere [49].
In vitro biofilm drug susceptibility to the antifungal fluconazole (at a concentration of 1,000 μg/ml) was assessed using a tetrazolium salt XTT reduction assay [16]. The percent reduction in biofilm growth compared to untreated controls is reported. Assays were performed in triplicate, and the significance of differences were assessed by one-way analysis of variance (ANOVA) with the posthoc Bonferroni and Holm methods [50] The CLSI M27 A3 broth microdilution susceptibility method was determine fluconazole activity against planktonic Candida strains. A visual turbidity endpoint was 24 h of grown was utilized.
An external jugular vein rat catheter infection model was utilized for in vivo biofilm assessment [28, 37, 38]. Quantitative cultures of C. albicans after 24 h of in vivo growth was utilized to measure viable biofilm cell burden. For drug treatment experiments, fluconazole at a concentration of 250 μg/ml was instilled and dwelled in the catheter over a 24-h period. The post treatment viable burden of Candida biofilm on the catheter surface was compared to untreated control growth. Three replicates were performed for treatment and control conditions.
Radiolabeled fluconazole was used to measure drug concentration in intact biofilms, matrix, and inside biofilm cells using a 6-well biofilm plate assay [37, 51]. After 48 hrs of biofilm growth, plates were washed and then incubated with 8.48 x 105 cpm of 3H fluconazole (Moravek Biochemicals; 50 μM, 0.001 mCi/mL in ethanol). Unlabeled fluconazole (20 μM) in RPMI–MOPS was added for an additional 15-min incubation period and then washed to remove unbound fluconazole. Biofilm were collected with a spatula. Matrix and cells were isolated as described above. Intact biofilm, matrix, cell samples were added to a Tri-Carb 2100TR liquid scintillation analyzer after adding ScintiSafe 30% LSC mixture to each sample fraction. Three biologic and technical replicates performed. Values were compared to the reference strain using pairwise comparisons with ANOVA with the Holm-Sidak method.
A 100-μl sample of purified biofilm EVs (equivalent of 1000-ml biofilm culture) was used to assess fluconazole sequestration. Vesicles were mixed with an equivalent volume of the radiolabeled drug and incubated for 1 h at 37°C. The sample was centrifuged for 10 min at 14,000 × g followed by collection of supernatant and washed three times with 1 ml of PBS. The collected vesicle pellet was resuspended in 200 μl of PBS and added to a Tri-Carb 2100TR liquid scintillation analyzer after adding ScintiSafe 30% LSC mixture. Three replicates were used.
Biofilms were formed in the wells of 96-well microtiter plates, as described above. After a 5-h biofilm formation period, the biofilms were washed with PBS twice, and purified EVs at concentrations of 21804 ± 1711 EVs/ml were added. For treatment studies, after an additional hour of incubation, biofilm cultures were treated with fluconazole (1,000 μg/ml), followed by the drug treatment protocol described above. For biofilm matrix studies, the samples were incubated for an additional 24 hrs prior to either SEM imaging or matrix isolation for quantitative carbohydrate analysis.
|
10.1371/journal.ppat.1005466 | siRNA Screen Identifies Trafficking Host Factors that Modulate Alphavirus Infection | Little is known about the repertoire of cellular factors involved in the replication of pathogenic alphaviruses. To uncover molecular regulators of alphavirus infection, and to identify candidate drug targets, we performed a high-content imaging-based siRNA screen. We revealed an actin-remodeling pathway involving Rac1, PIP5K1- α, and Arp3, as essential for infection by pathogenic alphaviruses. Infection causes cellular actin rearrangements into large bundles of actin filaments termed actin foci. Actin foci are generated late in infection concomitantly with alphavirus envelope (E2) expression and are dependent on the activities of Rac1 and Arp3. E2 associates with actin in alphavirus-infected cells and co-localizes with Rac1–PIP5K1-α along actin filaments in the context of actin foci. Finally, Rac1, Arp3, and actin polymerization inhibitors interfere with E2 trafficking from the trans-Golgi network to the cell surface, suggesting a plausible model in which transport of E2 to the cell surface is mediated via Rac1- and Arp3-dependent actin remodeling.
| Alphaviruses, such as Chikungunya or Venezuelan equine encephalitis viruses, are significant human pathogens that cause arthritis or fatal encephalitis in humans. For productive infection of cells, alphaviruses rely on a repertoire of cellular host proteins, including trafficking factors that mediate transport of viral components across the cell. We have performed a functional screen to identify cellular factors that are crucial for this transport process. We show that Rac1, PIP5K1-alpha, and the Arp2/3 complex are cellular regulators of alphavirus infection. These factors are important for major cellular actin rearrangements that occur at a late stage of virus infection and are virus-induced. Concomitantly, these factors might be essential for trafficking of the viral E2 surface glycoprotein from the trans-Golgi network (TGN) to the cell surface. E2 was found to associate with actin, as well as to co-localize with Rac1, PIP5K1-α, and actin filaments. Late E2-containing vesicles, termed cytopathic vacuoles II (CPV-II), were also imaged along and at the end of actin filaments in alphavirus-infected cells.
| Viral infection requires extensive subcellular trafficking, including cell entry, delivery of the genome to replication sites, and transport of viral proteins to and assembly of viral particles at the plasma membrane for egress. To this end, viruses make use of different cellular cues and signals to hijack existing endocytic and secretory pathways, cellular motor proteins, and cytoskeletal filaments.
Here we examine cellular trafficking machineries utilized by alphaviruses. Alphaviruses (family Togaviridae) are single-stranded, positive-sense RNA viruses that produce enveloped virions. Chikungunya virus (CHIKV), eastern equine encephalitis virus (EEEV), Venezuelan equine encephalitis virus (VEEV), and western equine encephalitis virus (WEEV) are the most medically important human alphaviruses that cause debilitating arthritides (CHIKV) or encephalitides (EEEV, VEEV, and WEEV) [1–3]. For instance, since December 2013, spread of CHIKV in the Caribbean has caused tens of thousands of human infections [4].
The alphavirus genome consists of two open reading frames encoding nonstructural and structural polyproteins. Four nonstructural proteins (nsP1-4) are required for transcription and replication of viral RNA, and three main structural proteins (i.e., capsid protein C, envelope glycoproteins E2 and E1) are the main constituents of virions. Alphavirus replication occurs initially at the plasma membrane [5,6]. Replication complexes are subsequently internalized via an endocytic process that requires a functional actin-myosin network. Following endocytosis, replication complex-containing vesicles migrate via a microtubule-dependent mechanism to the perinuclear area where they form stable, large and acidic compartments termed cytopathic vacuoles (CPV)-I. CPV-I structures are derived from modified endosomes and lysosomes and are associated with the alphaviral nonstructural proteins and viral RNA [7–9]. In the late stage of alphavirus infection, trans Golgi network (TGN)-derived vacuoles marked with the E1/E2 glycoproteins become predominant [10,11]. In these membrane vacuoles (termed CPV-II), the viral glycoproteins are arranged in a tubular structure. CPV-II vacuoles are implicated in intracellular transport of alphavirus glycoproteins from the TGN to the site of budding on the plasma membrane prior to virus egress [8,12].
Results from small interfering RNA (siRNA) screens identified a number of host factors that possibly promote or restrict nonpathogenic alphavirus infection [13–15]. However, detailed mechanistic studies regarding the role of host factors in alphavirus trafficking have not been performed. In this study, we used an RNAi-based screen to identify and validate trafficking host factors required for infection by the pathogenic VEEV and other pathogenic alphavirus relatives. Mutagenesis-, chemical inhibitor- and imaging-based approaches were further used to validate and decipher the role of these factors in alphavirus infection.
siRNA pools targeting each of 140 human trafficking genes were transfected into HeLa cells. A non-targeting siRNA was used as a control. Cells were subsequently infected with VEEV (chosen as a prototype alphavirus for the screen) for 20 h and then fixed and stained with a VEEV E2 glycoprotein-specific antibody (Fig 1A). Staining was performed without permeabilization to detect only E2 present on the cell surface. Cell number and infection rate were determined using quantitative high-content image-based analysis (see Materials and Methods). The infection rate of control siRNA-transfected cells was optimized to yield, on average, 70–80%. Analysis of the results revealed that siRNAs against 51 host trafficking factors decreased VEEV infection rate by >30% (Z-score <-2) (S1 Table).
To confirm results of the primary screen and to rule out potential off-target effects of individual siRNAs, we performed a secondary screen of deconvoluted siRNA pools. A hit was considered validated if at least 2 siRNAs from the set of 4 individual siRNAs targeting the gene product reduced the VEEV infection rate by ≥30% and had a p-value of <0.05 versus control siRNA-transfected wells. Wells that had low normalized cell numbers (final cell number <70% of the control siRNA-transfected well) due to combined effects of siRNA toxicity and VEEV cytopathic effects were excluded from further analyses. Analysis of the results led to validation of 19 (61%) out of the 31 primary hits (S2 Table).
Importantly, the list of validated hits was enriched for crucial regulators of the actin cytoskeleton. In particular, knockdown of four subunits of the heptameric Arp2/3 complex, ARPC4, ARPC5, ARPC1B (S2 Table), and ACTR3 (actin-related protein 3; Arp3) (Fig 1B), significantly inhibited VEEV infection. In addition, Ras-related C3 botulinum substrate 1 (Rac1), and phosphatidylinositol 4-phosphate 5-kinase type 1-alpha (PIP5K1-α or PIP5K1A) were also identified as hits (Fig 1B). The Arp2/3 complex plays a central role in actin dynamics by controlling filament nucleation [16,17]. Rac1 is a member of the Rho GTPase family and among its many functions modulates actin cytoskeleton organization [18]. PIP5K1-α is a lipid kinase involved in the synthesis of the signaling molecule phosphatidylinositol-4,5-bisphosphate (PI4,5P2), which is a central regulator of the actin cytoskeleton in response to multiple signals [19].
Our siRNA results were further confirmed using single siRNAs against Rac1, Arp3, and PIP5K1-α from another source (Fig 1B, siRNAs 5–7). We also observed a ≈10 to >30 fold reduction in VEEV titer following knockdown of these host factors (Fig 1C). Finally, siRNA-mediated knockdown of Rac1, Arp3, or PIP5K1-α inhibited infection of CHIKV (S1A and S1B Fig). These results indicate that Rac1, Arp3, and PIP5K1-α play an important role in alphavirus infections.
To validate the role of Rac1 and Arp3 in VEEV infection, we tested whether the Rac1 inhibitors EHT1864 and NSC23766 [20,21] and the Arp3 inhibitors CK548 and CK869 [22] could block VEEV infection. Upon treatment of HeLa cells with either of these types of inhibitors, VEEV infection rates were reduced in a dose-dependent manner (Fig 2A and 2B). Similar results were observed when the Rac1 inhibitors EHT1864 or NSC23766 or the Arp3 inhibitor CK548, were tested in primary human astrocytes (Fig 2C and 2D). These inhibitors were also effective in reducing infection rates of other alphaviruses. EHT1864 inhibited infections by CHIKV and the closely related Sindbis virus (SINV), and CK548 decreased CHIKV, SINV, EEEV, and WEEV infection rates (S2A and S2B Fig). None of the treatment conditions in either assays resulted in cytotoxicity. Overall, our results further confirm the importance of host factors Rac1 and Arp3 in alphavirus infection.
To determine if the function of Rac1 in alphavirus infection required Rac1’s GTPase activity, we established tetracycline-inducible 293 Flp-In T-REx cell lines that express chloramphenicol acetyltransferase (CAT, used as a control), wild-type Rac1, constitutively active Rac1 (G12V), or dominant-negative Rac1 (T17N) (Fig 2E) [23,24]. Rac1 expression in these cells was induced with tetracycline for 24 h, followed by infection with VEEV, or a non-alphavirus control (Rift Valley fever virus; RVFV strain ZH501, hereafter, RVFV). Expression of both Rac1 mutant variants (G12V, T17N) reduced VEEV but not RVFV infection rates, whereas expression of wild-type Rac1 had no effect (Fig 2F, S2C Fig). Both Rac1 mutants also reduced VEEV titer in the media (S2D Fig). We also confirmed the importance of Rac1 GTPase activity during WEEV and CHIKV infection (S2F and S2G Fig). The inhibitory effects of both Rac1 mutant variants on alphavirus infection likely indicate that the role of Rac1 during infection requires completion of the GTP-GDP-exchange/GTP-hydrolysis cycle. Cycling between GTP- and GDP-bound states may be required for productive infection, and shifting the level of activity predominantly to either side may block signaling pathways that emanate from the turnover.
Rac1 also forms a complex with PIP5K1 kinases that are necessary for stimulation of PI4,5P2 synthesis and actin assembly [25]. PIP5K1-α directly binds Rac1 via the polybasic tail of Rac1. Specific mutations within this region, such as K186E, abrogate Rac1:PIP5K1-α binding in vitro [26]. To examine whether Rac1:PIP5K1-α complex formation is important for VEEV infection, we used the tetracycline-inducible 293 Flp-In T-REx cell line to expresses Rac1 variant K186E (Fig 2G). Once induced, these cells and control cells expressing CAT or wild-type Rac1 were infected with VEEV or RVFV. Expression of Rac1 K186E reduced VEEV but not RVFV infection rates (Fig 2H, S2E Fig). VEEV titer in the media was also reduced (S2D Fig). Finally, we confirmed the importance of Rac1:PIP5K1-α complex formation to infection with CHIKV (S2H and S2I Fig). These results suggest that binding of Rac1 to PIP5K1-α plays a role in alphavirus infections.
We used a multi-cycle VEEV in our screen. Consequently, Rac 1 and Arp3 could have acted at a number of stages of the VEEV lifecycle. To determine when Rac1 and Arp3 act, we first determined the time necessary for a single lifecycle (round) of VEEV TC-83 (live-attenuated vaccine strain) infection. We measured virus particle release from HeLa cells to the media at different time points post virus inoculation using qRT-PCR analysis. Virus particle release into the media was observed at 9 h post inoculation of HeLa cells (Fig 3A and S3A Fig, left panel), suggesting an approximately 9-h replication cycle for VEEV under these conditions. Expression kinetics of the late alphaviral gene product, E2, was also analyzed. E2 expression was detected as early as 7 h post virus inoculation (Fig 3B and S3A Fig, right panel). Experiments performed with virulent VEEV IC-SH3 yielded similar results on expression of E2 and C proteins at these time points (Fig 3C). We confirmed our results with a one step-like growth curve analysis using a high MOI (MOI = 10) and also measured intracellular viral RNA (vRNA) levels as a function of time. Significant increase in intracellular vRNA levels was found at 5 h post virus inoculation, suggesting that virus replication/transcription is initiated prior to this time point (Fig 3D).
To narrow down the lifecycle stage targeted by Rac1 and Arp3, we performed time-of-addition experiments using inhibitors of these host factors. This time-based approach determines how long the addition of a compound can be postponed before losing its antiviral activity in cell culture. For example, if an inhibitor that targets viral fusion is present at the time when virus entry and fusion occurs within the viral lifecycle, productive infection will be inhibited. In contrast, if this inhibitor is added after the entry/fusion process is completed, the inhibitor will no longer be effective in blocking infection.
As a positive control for infection inhibition, HeLa cells were pretreated with increasing concentrations of Rac1 or Arp3 inhibitors 1 h before addition of virus. Alternatively, inhibitors were added to the cells at different time points after virus inoculation (1, 3, 5, or 7 h, Fig 3E) but prior to virus release (9 h post inoculation). When the Rac1 inhibitor EHT1864 or the Arp3 inhibitor CK548 were added 1, 3, or 5 h after VEEV exposure, VEEV infection rates were reduced to that detected with the positive control condition (pretreatment). However, addition of inhibitors 7 h after virus inoculation had significantly less effect on infection, suggesting that the inhibitors lose their antiviral activity at this time. Similar results were obtained with VEEV TC-83 in the context of a single replication cycle (S3B Fig); both EHT1864 and CK548 inhibitors reduced VEEV TC-83 infection when they were added up to 7 h post inoculation. Furthermore, when the inhibitors were added to HeLa cells 5 h following VEEV inoculation, VEEV titer in the media was significantly reduced (approximately 80- to >7,000-fold reduction, S3C Fig). Since the inhibitors exhibited antiviral activity when they were added 5 h post virus inoculation but significantly lost their antiviral affect when they were added 7 h post virus inoculation, these results indicate that Rac1 and Arp3 most likely play a role in the VEEV life cycle sometime between 5 h and 7 h post virus inoculation. Since one lifecycle of the virus takes at least 9 h to complete, and since transcription/replication is initiated prior to 5 h post virus inoculation, these results indicate that these inhibitors act at a late stage of virus infection.
To further confirm that Rac1 and Arp3 do not act at earlier stages (entry and replication), we first utilized a VEEV cell entry surrogate system composed of retroviral pseudotypes (Moloney murine leukemia virus; MoMLV) encoding eGFP and carrying the viral envelope proteins [27,28]. HeLa cells pretreated with control siRNA or with siRNAs targeting Rac1 or Arp3 were transduced with MoMLV-VEEV or MoMLV-EBOV (non-alphavirus control). As previously reported, MoMLV-EBOV entry into HeLa cells was reduced following knockdown of Rac1 or Arp3 [29,30] (Fig 3F). However, Rac1 or Arp3 knockdown had no or minimal effect on MoMLV-VEEV transduction rates, indicating that envelope-mediated entry of VEEV is independent of these two proteins.
Next, we examined the effect of the various inhibitors on total E2 protein levels in the context of virus infection. None of the inhibitors had an effect on E2 protein levels as determined by western blot analysis (Fig 3G).
Finally, we tested the effect of Rac1 and Arp3 on alphavirus replication in infected cells by treating cells with siRNAs as described above or with inhibitors against Rac1 or Arp3. Intracellular vRNA copy numbers were determined by qRT-PCR. The siRNAs as well as the inhibitors had no significant effect on intracellular vRNA copy numbers (Fig 3H and 3I). Similar results were obtained when the inhibitors were tested for their effect on CHIKV replication using a previously published replicon system (S3D Fig [31]). Overall, these results indicate that Rac1 and Arp3 function after virus entry and replication, but prior to budding and release.
As mentioned above, Rac1, Arp3, and PIP5K1A all affect cellular actin dynamics [16–19]. Previous studies have demonstrated a role for actin in alphavirus infection [32,33]. For example, in the early stages of infection of another alphavirus, Semliki Forest virus, replication complexes are internalized via an endocytic process that requires a functional actin-myosin network [7]. However, our time-of-addition experiments suggest that Rac1 and Arp3 play a role later in infection. We therefore investigated whether actin dynamics might play an additional role at later stages of infection.
To this end, we performed time-of-addition experiments (similar to the ones described above) with actin polymerization inhibitors. Cells were either pretreated with increasing concentrations of inhibitors before addition of virus (positive control) or preincubated with virus and subsequently treated with inhibitors at different time points after infection (Fig 4A and 4B).
Compared to the positive control condition (pretreatment), the actin polymerization inhibitors, latrunculin A and cytochalasin D, were less effective in inhibiting VEEV infection when they were added 1 h after virus inoculation (Fig 4A and 4B). This loss of antiviral activity is possibly due to the previously described role of actin in internalization of alphavirus replication complexes [7]. Inhibition of VEEV infection rates remained similar if actin polymerization inhibitors were added up to 5 h after virus inoculation. However, additional loss of antiviral activity was observed when the inhibitors were added at 7 h post virus inoculation. These results suggest that actin polymerization inhibitors target two separate steps in VEEV’s life cycle, one early in infection and one late in infection.
To further validate our results that actin might play a role in the later stages of the alphavirus lifecycle, we tested the effect of various doses of actin polymerization inhibitors (latrunculin A, cytochalasin B and D) or a microtubule-depolymerizing agent (nocodazole) on VEEV infection rate when added at various time points post virus inoculation. HeLa cells and primary human astrocytes were inoculated with VEEV first, and inhibitors were added 3 (HeLa) or 5 (astrocytes) h later. Disruption of actin dynamics by the actin polymerization inhibitors reduced VEEV infection rates and VEEV titer in a dose-dependent manner without cytotoxicity (Fig 4C–4E). Although some nocodazole-mediated inhibition of viral infection was observed, inhibition was not as marked as that observed with actin polymerization inhibitors and was accompanied by increased cytotoxicity (Fig 4C and S4A Fig). Phalloidin and tubulin staining demonstrated that the actin and microtubule cytoskeleton morphology was indeed disrupted upon treatment with these inhibitors (S4B and S4C Fig). These results further imply that actin polymerization might have an essential role in later stages of VEEV infection.
To determine if the actin polymerization inhibitors (latrunculin A and cytochalasin D) might block viral replication or E2 expression at later stages of infection, we inoculated cells with VEEV TC83 and treated them 5 h later with the inhibitors. Intracellular vRNA copy numbers were determined by qRT-PCR 11 h after virus inoculation. Alternatively, cells were lysed and analyzed for E2 expression by immunoblotting. Both inhibitors had no significant effect on vRNA copy numbers and E2 expression levels (Fig 4F and 4G). Finally, no effect on virus replication was observed when the actin polymerization inhibitors were tested for their effect on a CHIKV replicon system (S4D Fig) [31]. Together, the data suggests that the role of actin in the later stages of infection does not involve viral replication or late gene expression.
To assess the possible role of actin in the later stages of alphavirus infection, we assessed temporal changes of actin rearrangements during the course of viral infection. HeLa cells were infected with VEEV, CHIKV, or RVFV (used as a control) and co-stained at the indicated time points with antibodies against viral proteins and phalloidin. Confocal microscopy revealed major changes in the actin-staining pattern within alphavirus-infected cells (VEEV, CHIKV), as indicated by the accumulation of actin in large structures in the cytoplasm (i.e., actin foci, indicated by asterisks in Fig 5A). These foci co-localized with the alphavirus envelope protein E2 (Fig 5A). In contrast, such actin rearrangements were not observed in RVFV- or mock-infected cells (Fig 5A).
Actin foci were further quantified (measured as the number of foci per cell) in mock-, VEEV-, CHIKV-, and RVFV-infected cells (Fig 5B). These foci were detected as early as 7 h after VEEV inoculation (Fig 5C) and could also be detected upon infection with other alphaviruses (EEEV, WEEV, and SINV, S5A Fig). We also tested whether alphavirus nsP1, which was previously shown to mediate disruption of actin stress fibers and induction of filopodia-like extensions [34], could induce generation of actin foci. Expression of VEEV TC83 nsP1 in HeLa cells did induce filopodia-like extensions. However, no actin foci were observed (S5B Fig). Overall, our results demonstrate that, as early as 7 h post inoculation with alphaviruses, infection causes major cellular actin rearrangements leading to the formation of actin foci that are not nsP1-dependent and that co-localize with the alphavirus envelope protein E2.
Because our data suggested that the timing of the effects of Rac1 and Arp3 and the formation of actin foci take place late in infection (Figs 3 and 5), we speculated that Rac1 and Arp3 proteins might play a role in this alphavirus-induced actin remodeling. To test this hypothesis, HeLa cells were treated with increasing concentrations of Rac1 or Arp3 inhibitors, infected with VEEV, and subsequently stained with fluorescent phalloidin and antibodies against E2. Treatment with either the Rac1 (EHT1864) or Arp3 (CK548) inhibitor significantly reduced the number of actin foci and the percentage of infected cells in a dose-dependent manner (Fig 5D and 5E). In fact, under these conditions actin foci were rarely observed in confocal images even in E2-positive cells. These observations clarify that Rac1 and Arp3 function upstream of the major actin rearrangements detected in VEEV-infected cells.
Since Rac1-PIP5K1-α complex formation plays a role in alphavirus infection (Fig 2) and because Rac1 inhibitor reduced actin foci formation in alphavirus-infected cells (Fig 5), we next examined whether both host factors could be observed on actin foci and/or filaments within alphavirus-infected cells. Basal-to-apical confocal section series of VEEV-infected HeLa cells are shown in Fig 5F. PIP5K1-α and Rac1 show increased co-localization with actin foci and E2 towards the apical area (S5C and S5D Fig). Both host factors are also detected along actin filaments, where they co-localize with E2 (Fig 5F, insets).
To better characterize the nature of the observed actin foci within infected cells, we performed sequential scanning of cells stained for actin and alphavirus E2 in both stimulated emission depletion (STED) microscopy and confocal microscopy imaging modes (for comparison, see S6A Fig). With improved resolution of STED microscopy, actin foci within infected cells were found to be clusters of filamentous actin with a diameter range of 5–11 μm (Fig 6A). Actin filaments within the clusters are seen with VEEV E2 puncta at their ends or along them (Fig 6A). On the cell periphery, E2 puncta are localized in proximity to actin filaments (Fig 6A). E2 puncta are also observed at the ends of actin filaments in primary human astrocytes, and in CHIKV-infected HeLa cells (Fig 6B).
In a series of basal- (Section 7) to-apical (Section 25) confocal sections, a single VEEV-infected cell can be seen with an actin cluster (S6B Fig). E2 co-localizes with the actin cluster, and cytoplasm/nucleus staining demonstrates that the generated actin cluster is localized within the cell (S6C Fig). In contrast, co-localization of E2 and microtubules was not significant (S6B Fig).
We also performed electron microscopic studies to examine the localization of cytoskeletal elements relative to alphaviral CPV-II structures. These structures compartmentalize the viral glycoproteins E1 and E2 and serve as transport vehicles for the glycoproteins from the TGN to the viral budding sites on the plasma membrane. Electron-microscopic studies of VEEV-infected cells (Fig 6C) show CPV-II structures alongside or at the end of thin filaments, which, based on size and morphology, most likely correspond to actin filaments [12]. CPV-I replication compartments are also present within these cells (Fig 6C, bottom right panel) [9].
Because alphavirus E2 co-localized with actin filaments in infected cells, we next tested whether VEEV E2 associates with actin. HeLa cells were infected with VEEV or RVFV (control) or left uninfected (mock). Virus envelope protein-binding factors were subsequently immunoprecipitated from cell lysates with antibodies to surface glycoproteins E2 (VEEV) or Gn (RVFV). Western blot analysis of the immunoprecipitated fraction (IP) showed enrichment of actin in E2 immunoprecipitates from VEEV-infected cells relative to mock-infected control (more than 4-fold increase by densitometry analysis, Fig 6D, left panel). Such an increase in immunoprecipitated actin was not observed or was minimal in Gn immunoprecipitates from RVFV-infected cells (1.5-fold or less increase by densitometry analysis, Fig 6D, middle panel). To confirm the E2-actin association, we repeated these immunoprecipitation assays using more stringent lysis and washing conditions and performed the reverse experiment using an antibody against actin to examine its ability to immunoprecipitate E2 from VEEV-infected cells. Our results show that antibodies against E2 immunoprecipitated actin (more than 8-fold increase by densitometry analysis) and antibodies against actin immunoprecipitated E2 (more than 4-fold increase by densitometry analysis) from VEEV-infected, but not from mock-infected cells (Fig 6D right panel). These results indicate that VEEV E2 either directly or indirectly associates with actin in lysates from infected cells. However, since our lysis buffer included detergent (NP-40), the observed association between E2 and actin was most likely not in the context of CPV-II structures.
E2 was mainly localized in perinuclear puncta in cells treated with the Rac1 and Arp3 inhibitors, whereas in DMSO-treated cells E2 was found throughout the cytoplasm and at the plasma membrane (Fig 5D). Previous studies have demonstrated that the alphavirus glycoproteins E1/E2 are transported from the TGN to the cell surface via TGN-derived vacuoles [12,35], suggesting that the observed puncta might represent TGN or TGN-derived vacuoles. We therefore hypothesized that Rac1- and Arp3-dependent actin remodeling in alphavirus-infected cells might be important for trafficking of E1/E2. To test this hypothesis, primary human astrocytes were treated with DMSO, EHT1864, or CK548 and then infected with VEEV. Cells were stained with antibodies against VEEV E2 glycoprotein and the TGN marker TGN46. VEEV E2 was primarily located at the cell surface in control DMSO-treated cells (Fig 7A, zoom 1). In some of the cells, E2 puncta co-localized with TGN46. However, upon treatment with the Rac1 or Arp3 inhibitors, E2 localization in TGN46-positive puncta was significantly enhanced (Fig 7A, zoom 2 and 3) and less E2 glycoprotein was observed at the cell surface. Quantification of TGN46-to-cell-surface ratio of E2 staining intensity in control- or compound-treated VEEV-infected astrocytes is shown in S7A Fig. Similar experiments performed in HeLa cells using the inhibitors and the TGN marker Golgi-localized, gamma ear-containing, ARF-binding protein 3 (GGA3) yielded comparable results (S7B Fig).
In addition, we developed a flow cytometry-based assay for detection of VEEV E2 on the plasma membrane. We examined cell-surface expression of E2 following treatment with actin polymerization, microtubules depolymerization, Rac1, or Arp3 inhibitors. HeLa cells were infected with VEEV and treated 5 h later with increasing concentrations of EHT1864, CK548, latrunculin A, cytochalasin D, or nocodazole. Cells were subsequently stained for surface expression of E2 and with the 7-amino-actinomycin D viability dye. Concomitantly, an aliquot of the cells of each treatment group was lysed and analyzed for total E2 expression in whole-cell lysates. None of the inhibitors significantly affected total protein levels of E2. However, the actin, Rac1, and Arp3 inhibitors decreased geometric mean fluorescence intensity of E2 on the cell surface in a dose-dependent manner (Fig 7B and 7C). In contrast, the microtubule inhibitor, nocodazole, had no effect on cell surface E2 expression. The effect of the actin, Rac1, and Arp3 inhibitors on E2 surface expression was specific as no or minimal effect was observed on surface expression of cellular CD44 (S7C Fig). Overall, our data suggest that actin, Rac1, and Arp3, but not microtubule, inhibitors might interfere with trafficking of E2 from the TGN or TGN-derived vacuoles to the cell surface.
To examine if the actin remodeling observed in alphavirus-infected cells is associated with any TGN membrane structures, we stained VEEV-infected cells with the TGN marker TGN46. Actin clusters were observed near TGN46 (S8A Fig) and VEEV E2 was detected on these actin clusters and co-localized with the TGN marker. Rac1 was also found to co-localize with the TGN marker and E2, whereas PIP5K1-α co-localized with E2 but not with TGN46 (S8B and S8C Fig).
Reorganization of the host cytoskeleton varies among infections with different viruses and can play a role in every stage of the viral life cycle. Examples include virion movement (surfing) towards entry sites, actin-enhanced endocytic entry pathways, and actin-based, filopodial extensions (termed tunnelling nanotubes) that act as bridges to facilitate virus spread (reviewed in [36–39]).
Here, using an siRNA screen, we identified trafficking host factors that are important for alphavirus infection and are crucial regulators of the actin cytoskeleton. To date, Rac1- and Arp2/3-mediated actin rearrangements have mainly been associated with virus uptake and entry [30,40–44]. Rac1 is predominantly known as a key regulator of the actin cytoskeleton at the plasma membrane [45]. There, Abelson interactor 1 (Abi1) and Wiskott-Aldrich syndrome protein (WASP) family verprolin-homologous protein (WAVE), but not neural (N)-WASP, are essential for Rac1-dependent membrane protrusion and macropinocytosis [46].
Recently, however, Rac1, the Arp2/3 complex, and actin have emerged as major factors in the secretory pathway in processes such as biogenesis and motion of Golgi-derived transport carriers to the plasma membrane [47–50]. During formation of TGN carriers, Rac1 functions downstream of ADP-ribosylation factor 1 (Arf1). Arf1 recruits clathrin/adaptor protein 1 (AP-1)-coated carriers and a complex composed of cytoplasmic fragile X mental retardation 1 (FMR1)-interacting protein (CYFIP), nucleosome assembly protein 1 (NAP1), and Abi1 to the TGN. Rac1 and its exchange factor Rho guanine nucleotide exchange factor 7 (ARHGEF7) bind CYFIP and trigger N-WASP- and Arp2/3-mediated actin polymerization necessary to tubulate clathrin-AP-1-coated carriers [51]. Therefore, during alphavirus infection, Rac1 could potentially be recruited to the TGN to mediate biogenesis of E2-containing vesicles and/or their transport from the TGN to the cell surface via actin (see model, Fig 7D). In support of this hypothesis, some of the host factors mentioned above, such as clathrin heavy chain 1 (CLTC), AP-1 subunits (AP1M1), and Arf1 were identified as hits in our primary and validation siRNA knockdown screens (S1 and S2 Tables). Furthermore, siRNAs targeting N-WASP reduced the infection rate of both VEEV and CHIKV (S1B Fig). Finally, during VEEV infection Rac1 was found to co-localize with E2 at the TGN (S8 Fig). Hence, Arf1 may function upstream of Rac1 to facilitate biogenesis and/or motion of E2 transport carriers from the TGN to the plasma membrane and that this transport is mediated by N-WASP.
Viruses have evolved specific egress pathways for transporting viral components to the plasma membrane, often using the cell’s secretory pathway via the endoplasmic reticulum, the Golgi, and even transport vesicles. Most exocytic transport of cellular secretory cargo to the plasma membrane relies on microtubules for long-range translocations [52,53]. The microtubule network is also emerging as the preferred cytoskeletal element recruited for transportation of components of certain viruses to the cell surface [54–57]. Examples are microtubule delivery of influenza A virus HA membrane glycoprotein to the apical surface of MDCK cells [58] and vesicular stomatitis Indiana virus glycoprotein G trafficking from the TGN-to-plasma membrane [59].
In contrast, our results demonstrate that transport of the alphavirus membrane glycoprotein E2 is at least in part dependent on actin and actin regulators (Rac1 and Arp3). We hypothesize that the coordinated activities of PIP5K1-α, Rac1, and the Arp2/3 complex might mediate alphavirus envelope E2 trafficking from the TGN to the cell surface via actin. Several results support this actin-dependent transport model (Fig 7D). First, time-of-addition experiments with Rac1 and Arp3 inhibitors demonstrated that both factors function at a late stage of virus infection (Fig 3). Second, within a similar time frame (concomitantly with E2 expression in infected cells) major actin rearrangements into clusters occur in alphavirus-infected cells (Figs 3 and 5). Super high-resolution fluorescence microscopy and electron microscopy show that CPV-II structures containing E1 or E1/E2, respectively, are localized along or at the end of actin filaments. Rac1 and PIP5K1-α also co-localize with E2 on actin foci (Fig 5). In infected cell lysates, E2 envelope protein was found to associate (either directly or indirectly) with actin (Fig 6). Third, Rac1 and Arp3 inhibitors blocked formation of virus-induced actin clusters (Fig 5). In cells treated with actin, Rac1, or Arp3 inhibitors, most of the E2 staining was found to localize with TGN markers, and E2 levels at the cell surface were reduced (Fig 7). We have not yet examined the role of actin, PIP5K1-α, Rac1, and the Arp2/3 complex in E1 trafficking. However, since E1 and E2 are oligomerized into trimeric complexes during transit to the plasma membrane in CPV-II structures [60], we speculate that these host factors will have a similar function in trafficking of both viral proteins.
Actin dynamics are involved in numerous aspects of intracellular transport. However, little is known regarding manipulation of these host machineries by pathogenic alphaviruses. Viruses can serve as unique tools to decipher how a particular cargo recruits actin filament tracks and the host factors and motors associated with these movements. Our results suggest a previously unidentified role of host factors Rac1, Arp3 and PIP5K1-α late in alphavirus infection via actin remodeling that possibly mediates transport of alphavirus envelope glycoproteins from the TGN to the cell surface. It is important to note that although our data indicate that actin plays a major role in alphavirus glycoprotein transport, our experiments do not exclude the existence of other, parallel, transport mechanisms mediated by intermediate filaments or microtubules.
Recombinant alphaviruses expressing tagged E2 could be useful to further substantiate our findings. However, until now, we have not succeeded in rescuing such viruses. Finally, our high-content siRNA screen reveals novel host regulators of alphavirus infection and potential therapy targets.
An arrayed library targeting 140 trafficking genes (Dharmacon Human ON-TARGETplus siRNA Library—Membrane Trafficking—SMARTpool, G-105500-05, Thermo Scientific,) was used to transiently reverse-transfected HeLa cells (10,000 cells per well, 96-well format) in triplicate at a 30-nM final concentration, using HiPerfect (Qiagen). Cells were washed on the following day and infected 24 h later with VEEV ICS-SH3 at an MOI of 0.5 for 20 h. Cells were fixed with 10% formalin (Val Tech Diagnostics) and stained for high-content quantitative image-based analysis. The screen was repeated three times. In 6 wells on each plate, cells were transfected with a negative control siRNA (NT, siCONTROL Non-Targeting siRNA #2, Dharmacon D-001210-02). The infection rate of control siRNA-transfected cells was optimized to yield, on average, 70–80%, following multiple virus replication cycles.
For the primary screen, siRNA pools were classified as hits if the average of triplicate wells showed that the percentage of VEEV-positive cells decreased by more than 30% compared to that observed with the control siRNA wells on the plate (Z-score <-2 SD). In the validation screen, the individual oligomers comprising each pool were placed into separate wells, and the screen was repeated. siRNA targets were considered validated if two or more of the individual oligomers were classified as hits compared to the control wells on the plate (similar parameters as above) and if the cell number was not less than 30% of the average of the negative control wells on the plate. Catalog numbers and sequences of siRNAs are provided in Table 1. The percent of infected cells relative to controls, as well as the normalized cell numbers (normalized to control siRNA) is provided in S1 Table.
HeLa (ATCC, #CCL-2), BHK-21 (ATCC, #CCL-10), and Vero cells (ATCC, #CCL-81) were maintained in Eagle’s minimum essential medium supplemented with 10% fetal calf serum. T-REx-HeLa cells expressing human wild type Rac1 fused to eGFP, and Flp-In 293 T-REx cells expressing human wild type Rac1, Rac1 G12V, Rac1 T17N, Rac1 K186E or CAT upon tetracycline induction were generated by using the T-REx System or the Flp-In T-REx system, respectively, according to the manufacturer's instructions (Life Technologies). Cells were induced to express wild-type human Rac1, variants thereof, or CAT in 96-well plates by adding tetracycline (1 μg/ml) to the growth medium. Normal human astrocytes were obtained from Lonza and maintained according to the provider's instructions. Plasmids encoding Rac1 variants (wild-type Rac1, Rac1 T17N or Rac1 G12V) fused to an avian myelocytomatosis (myc) protein tag were purchased from the Missouri S&T cDNA Resource Center (www.cdna.org). A plasmid encoding Rac1 K186E was generated by using the QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent Technologies). Sequences of the primers are provided in Table 1. A plasmid encoding pcDNA3-EGFP-Rac1-wt was obtained from Addgene.
Mouse monoclonal antibodies against CHIKV (2D21-1), EEEV (1C2), VEEV (1A4A-1), WEEV (9F12), and RVFV envelope glycoprotein Gn (4D4) and nucleoprotein (R3-ID8-1-1) were obtained from US Army research Institute of Infectious Diseases (USAMRIID) archives [61]. Goat antibody against VEEV capsid (C) or envelope protein was generously provided by AlphaVax (via Kurt Kamrud). Rabbit antibodies against Arp3, actin, N-WASP, GAPDH, FLAG, and HA were obtained from Sigma-Aldrich. Mouse monoclonal antibodies against actin, CD44, GGA3, and Rac1 were purchased from BD Transduction Laboratories. Rabbit antibody against α/β-tubulin was obtained from Cell Signaling Technology. Sheep anti-human TGN46 antibody was from AbD Serotec. Alexa Fluor-conjugated antibodies and phalloidin, Hoechst 33342, and HCS CellMask Red were obtained from Life Technologies. All chemical inhibitors were purchased from Sigma-Aldrich, with the exception of EHT1864 (Tocris Bioscience). Cells were incubated with inhibitors for 1 h before addition of viruses unless otherwise indicated in the figure legends.
Infections with VEEV IC-SH3, EEEV FL91-4679, WEEV CBA87, RVFV ZH501, and CHIKV AF15561 were conducted under Biosafety Laboratory 3 conditions. All alphaviruses were propagated in BHK-21 cells and purified via sucrose gradients. RVFV was propagated in Vero cells. Viral infectivity was titrated by plaque assays as previously described [62].
MoMLV-eGFP pseudotypes carrying the VEEV envelope proteins E1/E2 or Ebola virus envelope GP1,2 (control) were produced as previously described [27,28,63]. MoMLV-eGFP pseudotypes were added to siRNA-treated HeLa cells for 6 h. Cells were then washed and supplemented with growth medium. Cell transduction efficiency was determined 2 days later by measuring eGFP expression using an Opera confocal reader (PerkinElmer).
For CHIKV replicon assays, we used the previously described BHK-CHIKV-NCT cells, which contain the CHIKV replicon with two reporter genes, Renilla luciferase (Rluc) and EGFP [31].
BHK-CHIKV-NCT cells were seeded onto 96-well plates at densities of 2 × 104 cells/well, incubated overnight, and treated with the indicated compounds at various concentrations. After exposure for 48 h, the Rluc activity resulting from the translation of CHIKV-Rluc genomic RNA was determined from the lysates using a Rluc assay kit (Promega) with a Tecan microplate reader.
HeLa cells in 6-well plates were infected with VEEV TC-83 or RVFV MP12 (MOI = 1) for 8 h. Cells were lysed in a mild lysis buffer (50 mM Tris pH 7.4, 50 mM NaCl, 0.2 mM ethylenediaminetetraacetic acid (EDTA), and 1% Triton X-100) or a lysis buffer (25 mM Tris pH 7.4, 150 mM NaCl, 1 mM EDTA, 5% glycerol, and 1% NP-40) from Pierce Crosslink Immunoprecipitation supplemented with Complete protease inhibitor cocktail (Thermo Scientific Pierce). Cleared lysates were incubated overnight at 4°C with protein A/G beads (Thermo Scientific Pierce) and VEEV E2- or RVFV Gn-specific antibodies or with beads cross-linked to antibodies against VEEV E2 or actin. Cell lysate immunoprecipitates were analyzed by SDS-PAGE and immunoblotting using the indicated antibodies.
For western blot analyses, cells were lysed with RIPA lysis and extraction buffer supplemented with complete protease inhibitor cocktail (Thermo Scientific Pierce). Cleared lysates were analyzed by SDS-PAGE and immunoblotting using WesternBreeze chromogenic or chemiluminescent kits (Life Technologies) and the indicated antibodies. Densitometric analysis of western blots was performed with ImageJ [64].
Cells were grown on glass cover slips and inoculated with VEEV or CHIKV for 1 h. Cells were fixed 20 h (VEEV) or 48 h (CHIKV) later, permeabilized with 0.5% Triton X-100 (Sigma-Aldrich) in phosphate-buffered saline (PBS), blocked with 3% bovine serum albumin in PBS for 1 h. and stained using mouse anti-E2 antibodies (1:1,000 dilution), followed by ATTO 647N Goat Anti-Mouse IgG (Active Motif) (1:2,000 dilution). Actin was stained with Phalloidin ATTO 565 (Sigma-Aldrich) (1:80 dilution). Slides were mounted in ProLong Gold Antifade Reagent (Life Technologies) and dried overnight at room temperature before imaging. All confocal images were acquired on the Leica SP5 TCS 2C STED confocal system (Leica Microsystems) equipped with Leica’s inverted DMI 6000 microscope and STED 100x oil objective. Images were acquired at an imaging speed of 400 Hz, pin hole set to Airy1, line average of 6, and 1024 X 1024 formats. For STED of ATTO dyes, the pulsed Ti:SA infra red laser (Mai Tai, model # MAI TAI BB990, Spectra-Physics) was tuned to 740 nm.
HeLa cells grown on a MatTek dish (MatTek corporation, MA) were infected with VEEV TC83 (MOI = 5) for 20 h. Cells were fixed for 1 h in primary fixative (2.5% formaldehyde, 2.5% glutaraldehyde, 0.1 M sodium cacodylate, pH 7.4), washed three times in ice-cold 0.1 M sodium cacodylate buffer, and incubated with 1% osmium tetroxide in 0.1 M of sodium cacodylate for 1 h, washed three times with distilled water, stained and stabilized on ice with 2% uranyl acetate for 1 h and successively dehydrated on ice through a series of 22%, 50%, 75%, and 95% ethanols. The cells were then dehydrated three times at room temperature in 100% ethanol and infiltrated in well-mixed 50% ethanol and 50% Durcupan ACM resin (Fluka, Sigma-Aldrich) for 1 h with agitation. Cells were infiltrated twice by 100% Durcupan ACM for 3 h with agitation, after which the samples were placed in an oven and polymerized at 60°C for at least 48 h. The glass coverslip was peeled away from the bottom using a razor blade, and the selected area was cut out and glued to a block for sectioning. Thin sections (approximately 80 nm) were collected and pre-stained with 1% uranyl acetate and Sato lead before examination on a JEOL 1011 transmission electron microscope at 80 kV. Digital images were acquired using an AMT camera system.
Plasmid encoding HA-tagged PIP5K1α was generously provided by Dr. Richard Anderson (University of Wisconsin). Plasmid encoding FLAG-tagged nsP1 was generated in-house by PCR. Plasmids were transiently reverse-transfected into HeLa cells on glass coverslips (Fisher Scientific) using Lipofectamine LTX Reagent (Life Technologies). T-REx HeLa cells on glass coverslips were induced with tetracycline for 24 h to express Rac1-eGFP. VEEV-infected cells were fixed, permeabilized, and blocked as described for STED. After incubation with primary antibodies and fluorescent secondary antibodies, slides were mounted as described for STED and air-dried before imaging with a TCS-SP5 confocal/multiphoton microscope (Leica Microsystems). All confocal images represent a single plane acquired with a 100× oil objective. Similar experimental conditions were used for imaging studies of actin, tubulin, TGN46, and VEEV E2 in HeLa cells. Co-localization analysis of tubulin or actin with VEEV E2 was performed with the ImageJ program using the Interactive 3D Surface Plot plugin [64].
For analysis of the siRNA screen, cells were stained without prior permeabilization. Cells inoculated with CHIKV, EEEV, RVFV, WEEV or SINV or cells designated for phalloidin or TGN staining were permeabilized prior to blocking as described above. Cells were then stained with murine monoclonal antibodies against the indicated viral proteins (1:1,000 dilution) and, where indicated, against TGN46 or GGA3 (1:250 dilution). Subsequently, cells were stained with appropriate Alexa Fluor-conjugated antibodies (1:1,000 dilution), and Alexa Fluor 568 Phalloidin (1:100 dilution) where indicated. All infected cells were also stained with Hoechst 33342 and HCS CellMask DeepRed for nuclei and cytoplasm detection, respectively.
High-content quantitative imaging data were acquired and analyzed on an Opera quadruple excitation high sensitivity confocal reader (model 3842 and 5025; Perkin-Elmer), at two exposures using a ×10 air, ×20 water, or ×40 water objective lenses as described in [65]. Images were analyzed using Acapella 2.0, 2.6, 2.7 (Perkin-Elmer) scripts in Evoshell or the building-blocks interface in the Columbus image analysis server (PerkinElmer). Nuclei and cytoplasm staining were used to determine total cell number and cell borders, respectively. Mock-infected cells were used to establish a fluorescence intensity threshold for virus-specific staining. Quantification of virus-positive cells was subsequently performed based on mean fluorescent intensities in the virus-specific staining channel. Infection rates were then determined by dividing the number of virus-positive cells by the total number of cells measured. Detailed pipelines for image-based quantification of alphavirus-induced actin foci and TGN46-to-plasma membrane E2 staining intensity ratio are available upon request. At least 5,000 cells and up to 15,000 cells were analyzed per replicate in drug- or siRNA-treated cells. For actin foci analysis, 1,000–1,500 cells were used per replicate. For analysis of TGN46-to-plasma membrane E2 staining intensity ratio, 700 cells were used per replicate.
HeLa cells in 12-well plates were inoculated with VEEV TC-83 (MOI = 10) for 5 h. DMSO, EHT1864, CK548, nocodazole, latrunculin A, or cytochalasin D were subsequently added at the indicated concentrations. Five or 6 h later, cells were detached with Cell Dissociation Buffer (Life Technologies) and washed with flow buffer (PBS/0.5% bovine serum albumin/2mM EDTA). Cells were incubated with mouse anti-VEEV E2 or CD44 primary antibody (1:1,000 dilution in flow buffer) for 30 min on ice and then washed twice with ice-cold flow buffer. Cells were incubated for 20 min in the dark with Alexa Fluor 488 Goat Anti-Mouse IgG secondary antibody (Life Technologies) (1:5,000 dilution in ice-cold flow buffer) and with 7-amino-actinomycin D to exclude dead cells from analysis (1:500 dilution). Following two more washes with ice-cold flow buffer, cells were fixed in 1% paraformaldehyde. Cytometric collection was performed using a FACS Canto II (BD Biosciences), and data were analyzed using Flowjo v7.6.5 (TreeStar).
VEEV TC-83 RNA yields from the media and the cells and relative expression levels of PIP5K1-α in siRNA-treated HeLa cells were determined by qRT-PCR as previously described [65]. Serial 10-fold dilutions of the assayed (102 to 107 copies) virus were used as standards. Relative expression levels were determined by using the comparative cycle threshold method. Sequences of qRT-PCR probes/primers are provided in Table 1.
Data are representative of at least three independent experiments, and values are given as mean of triplicates ± standard deviation (SD) unless otherwise indicated. Statistical significance was determined by the paired Student’s t test.
|
10.1371/journal.pntd.0006255 | An outbreak of Leishmania major from an endemic to a non-endemic region posed a public health threat in Iraq from 2014-2017: Epidemiological, molecular and phylogenetic studies | Cutaneous leishmaniasis (CL) is a neglected worldwide, zoonotic, vector-borne, tropical disease that is a threat to public health. This threat may spread from endemic to non-endemic areas. Current research has exploited epidemiological, molecular and phylogenetical studies to determine the danger of an outbreak of CL in the borderline area between northern and central Iraq from 2014–2017.
For the first time, using sequence analysis of the cytochrome b gene, the occurrence of CL in the borderline area between northern and central Iraq was confirmed to be due to Leishmania major. The phylogenetic analysis indicated that it was closely related to the L. major MRHO/IR/75/ER strain in Iran.
In conclusion, the genotype confirmation of the L. major strain will improve our understanding of the epidemiology of the disease. This is important for facilitating control programs to prevent the further spread of CL. Furthermore, this area could be considered as a model for further research on the risk of global CL epidemics in other non-endemic countries where both reservoir hosts and sandfly vectors are present.
| Leishmaniasis refers to a disease with three main types of clinical manifestation in infected individuals including cutaneous, mucocutaneous and visceral forms. It is caused by several species of a parasite belonging to the genus Leishmania and is transmitted by a small blood-sucking insect called a sandfly. The disease is mostly confined to the majority of the poorest countries worldwide, including Iraq, and is categorized as a low priority public health concern. The risk of the disease is exacerbated especially when suitable environments assist the sandfly and reservoir host to breed and spread and help the parasite to transfer from high incidence areas to places free from the disease. Therefore, we investigated the risk of the CL form of the disease after an outbreak in a borderline between northern and central of Iraq using the most sensitive diagnostic techniques including PCR and gene sequencing. The epidemiological, molecular and phylogenetic analyses of the parasites were studied, and we found that the parasite species Leishmania major was associated with the outbreak. Phylogeny analysis confirmed that the identified strain of the parasite matched an Iranian strain. These results indicate the risk of the disease spreading from endemic to non-endemic areas.
| Leishmaniasis is considered to be a neglected tropical and zoonotic disease that spreads via phlebotomine sandfly vectors [1]. Leishmaniasis is a parasitic disease caused by intracellular protozoa which in humans has four clinical forms including cutaneous (CL), diffuse cutaneous (DCL), visceral (VL) and mucocutaneous (MCL) leishmaniasis and it is endemic in different parts of the world [2]. The morbidity associated with human CL is up to 1.2 million cases distributed worldwide resulting in extensive integumentary lesions [3].
There are two groups of CL, New World and Old World leishmaniasis, with only the latter group identified in the Middle East and it includes three main species; L. major, L. tropica and L. infantum [4]. Recent studies showed a high prevalence of CL in Iran [5, 6], Turkey and Syria [7]. Although Iraq shares long borders with these countries and leishmaniasis is endemic, the World Health Organization has not classified it as a country with a high burden profile [8].
In Iraq, several studies have been performed to diagnose Leishmania parasites from skin lesions of human patients by using different methods including histopathological examinations, direct smears, cultures and serological tests [9, 10]. Few studies have been conducted to exploit PCR in the characterization of the Leishmania strains in human cutaneous lesions [11] and VL-suspected patients [12] in central Iraq. Studies have been performed without conducting gene sequencing or phylogenetic analyses. However, in a US military base in Southern Iraq, a phylogenetic study investigated the prevalence of different Leishmania species in sandflies using molecular study and phylogenetic analysis [13].
Therefore, the aim of this study was to identify the genotype of the most prevalent CL strains in the region using cytochrome b gene amplification by PCR and sequencing.
An outbreak of leishmaniasis was clinically suspected for the first time in 2013 in areas belonging to the Kifri district in the Garmian Administration. The term (Garmian) is a Kurdish word which is used to denote a ‘hot and dry area’ indicating information about location and climate. The Garmian area is located in the southeast Kurdistan region of Iraq. It is in between the latitudes (34°15–33 = - 35° 11–05 =) above the equator and the longitudes (44° 29–41 = - 45° 54–20 =) of the eastern hemisphere. The Garmian includes the districts: Kalar, Kifri, and Khanaqin, and its total area is 6731.73 square kilometers. According to the official site of the general board of tourism of Kurdistan- Iraq in 2015, the total population of the central town of Garmian, Kalar, is about 250,000 residents [14]. In this region, there is an increasing concern about the cutaneous form of leishmaniasis which is publicly known as “Baghdad sore”. Since 2014 leishmaniasis has been considered a notifiable disease and every new case with a clinical manifestation of cutaneous lesions of leishmaniasis should be recorded officially by local authority officers as a transmissible disease before the patient receives treatment.
Northern and central Iraq have undergone economic and humanitarian crises due to the Iraqi civil war since 2014. Moreover, the topography of both territories is different. In addition, there has been only one updated map up to 2008 based on the last report of CL incidence in Iraq by the WHO [3]. Thus, in this study, a spot map of CL cases was updated in the borderline region using Landsatlook viewer (USGS Products, Data available from the U.S. Geological Survey). The map (Fig 1a and 1b) shows the outbreak of CL from an endemic area in the Kurdistan Region of Iraq (KRI) including Diyala province to a non-endemic area inside the KRI including the Garmian administrative region.
Data of cutaneous leishmaniasis (CL) in Iraq were collected from the WHO website [8, 15, 16], and a line graph showing annual numbers of CL cases from 1989–2015 was plotted using GraphPad Prism version 6.06 for Windows (GraphPad Software, La Jolla California USA, www.graphpad.com). Furthermore, in the area of the study (Garmian administration, Kurdistan Region, Iraq), new cases were referred by local health general practitioners to visit Kalar General Hospital in Garmian, Sulaimaniyah province to be clinically examined by dermatologists and receive Pentostam injection treatment (sodium stibogluconate). The data of clinically examined patients collected by the Department of Transmissible Diseases in Garmian from 2014–2017 were also analyzed.
Thirty samples were collected from lesions of new clinically suspected CL cases or from patients receiving early treatment in February, March and April 2017. The sample collection was performed by cleaning the skin lesions with cotton soaked in 70% ethyl alcohol and left to dry. This was followed by injecting 0.1 ml sterile normal saline into the active borders of the skin lesions using a 25-gauge insulin needle and then aspirating the fluid into sterile 1.5 ml tubes. The samples were directly preserved in 0.4 ml absolute ethanol, labeled and stored at room temperature for molecular study. Later on, the samples were submitted to the molecular laboratories of the University of Garmian which is based in the Kalar district.
Clinical samples were collected from patients who agreed to participate in this study and signed an informed consent form. The study was also approved by the Ethical Committee of the Department of Biology, College of Education, University of Garmian with permit number (85, 18/04/2017). After receiving permission from the General Directorate of Garmian Health (permit number 1550, 10/05/2017), the samples were transported to the molecular biology lab of Garmian University.
A pair of primers including Leishmania cytochrome b forward (LCBF): GGTGTAGGTTTTAGTTTAGG, Leishmania cytochrome b reverse (LCBR): CTACAATATACAAATCATAATATACAATT (Macrogen Co., Seoul, KR) were exploited for amplification of the Leishmania cytochrome b gene with a product size of 866 bp as previously used for identification of almost all species of Leishmania by PCR and DNA sequencing [17].
Total genomic DNA of the ethanol-preserved samples was extracted by a PrimePrep Genomic DNA Extraction Kit (from tissue). The ethanol was removed from the samples by using centrifugation and washing with normal saline. The pellets were mixed with 200 μl tissue lysis buffer (TL buffer) and 20 μl proteinase K and incubated at 56°C for approximately an hour until the samples were lysed. According to the manufacturer’s instructions, DNA was isolated using ethanol and buffers then eluted with 200 μl elution buffer (TE) provided by the company (GENET BIO CO., Daejeon, KR).
Conventional PCR was performed individually for each sample in 20 μl reactions containing 1x Prime Taq premix (2x) which contains Prime Taq DNA Polymerase 1 unit, 2x reaction buffer, 4 mM MgCl2, enzyme stabilizer, sediment, loading dye, pH 9.0 and 0.5 mM each of dATP, dCTP, dGTP, dTTP and 0.5 μM final concentration from each of the LCBF and LCBR primers. The PCR reaction conditions were 94°C for 3 min; 40x at 94°C for 1 min, 60°C for 1 min, 72°C for 2 min; 72°C for 5 min using a thermal cycler (Mastercycler nexus, Eppendorf AG, Hamburg, Germany).
PCR products were run at 110 V for 50 min on a 1.5% agarose gel in 1x TBE (87.5 mM Tris base, 89 mM boric acid, 3 mM EDTA) and stained with Prime safe dye (GENET BIO CO., Daejeon, KR). A total of 5 μl of PCR products from nine positive samples and 5 μl (5 pmoles) of forward or reverse primers for forward or reverse sequencing, respectively, were sequenced using the Sanger method (Macrogen Co., Seoul, KR) and edited by CodonCode Aligner (CodonCode Corporation, 101 Victoria Street, Centerville, MA 02632).
To the best of our knowledge, no phylogenetic data were found exploring CL strains in the area along the borderline between the northern region (Kurdistan Region) and the middle part of the country based on the NCBI search engine [18] on 28/06/2017 using these keywords (Leishmaniasis Iraq PCR).
Seven high-quality sequences were submitted to the NCBI GenBank using Bankit [18]. The high-quality sequences were determined based on having high single peaks using CodonCode Aligner. The cytochrome b gene sequences together with those from representative strains were aligned with CLUSTAL W software and examined using the program MEGA (Molecular Evolutionary Genetics Analysis) version 7. Phylogenetic trees were constructed by the neighbor-joining method with the distance algorithms available in MEGA version 7. The distances were calculated using the Kimura 2-parameter method. Bootstrap values were determined with 1,000 replicates of the data sets [19].
The CL outbreak occurred in areas between northern and central Iraq (Fig 1a). The data showed that cases of CL seemed to have originated from endemic areas like Diyala, Saladin, Mosul, and Kirkuk provinces. This outbreak may have been caused by infected people traveling from the center to the north of Iraq. It is worth mentioning that before toppling down the previous government of Iraq, i.e., before 2003, people from the southern and central provinces rarely visited Kurdistan due to strictly controlled borders (As shown in Fig 1b, interrupted blue line). After 2003, the situation changed and there was free public access into the Kurdistan Region of Iraq (KRI). Furthermore, since 2014, people have moved from the war zones of the central provinces to the Garmian region. Thus, this area can be regarded as a typical model region to understand the risk of leishmaniasis spread from endemic to non-endemic areas.
As shown on the map (Fig 1b), CL spread from endemic areas of the KRI borderline such as Jabara, Kokiz, Qara Tapa, Khanaqin, Jalawla and Tuz Khurmatu to non-endemic areas including Kalar, Kifri, and Rizgari districts was due to the arrival of refugees. This may have led to spread of the disease in other parts of Kurdistan including Sulaimaniyah, Irbil, and Duhok. Even so, temperature and humidity could be related to the distribution of the vectors and reservoirs [20]. Therefore, further study on the sandfly and animal reservoirs from both the endemic and non-endemic areas will help uncover the epidemiology of the disease. According to Alvar et al. [3], the WHO reported 1655 CL cases/ year from 2004–2008 in Iraq; the annual incidence rate was estimated from 8300 to 16,500 cases, although this number seems to be underestimated by the WHO. This could be due to a lack of diagnostic services, so the disease was not regarded as a major public health concern [21]. In addition, it is worth mentioning that not all cases were reported by the authorities for the following reasons: firstly, some infected people objected to receiving treatment due to painful intralesional injections. Secondly, some patients did not wish to visit hospitals since they use traditional medicine for treatment or believe that the lesions are self-curable. Finally, some places are far from health centers. Therefore, these data should be updated in Iraq and this will be important before introducing any CL control programs.
In the current study, the data show the prevalence of CL in Iraq from 1989–2015 (Fig 2). As the country went through several wars, internal conflicts, economic crises, and sanctions over the previous 27 years, massive fluctuations in the number of reported CL cases can be noticed. There was a sharp increase in the number of cases after the second gulf war in 1990. This trend remained high until 1997, which could be due to discontinuation of control programs and lack of healthcare services due to the economic blockade. The number of cases remained low from 1998 until 2003, possibly due to WHO interventions [15] and control programs or relative economic growth after the removal of the sanctions in 1997. Again from 2009 the number of cases increased; however, it remained relatively low until 2014, then a sharp rise was recorded with a peak in 2015. The last increase in the number of CL cases could be due to the civil war in Iraq starting in 2014 as the war led to displacement of millions of people, especially from endemic areas to non-endemic areas, as well as a deterioration in health services.
After the introduction of malaria control programs in Iraq, the number of CL cases decreased until their discontinuation in the mid-1960s [15]. Afterwards, massive fluctuations in the number of reported cases were noticed. The number of CL cases could be related to certain factors. One of the main factors is population displacement, which brings non-immune people to endemic areas and infected people to non-endemic areas. In addition, increased contact with reservoir animals and sandfly vectors, untreated patients, malnutrition, poor sanitation and environmental changes are other possible reasons for the increase in CL cases.
Most of the districts of the Garmian region belong to the Sulaimaniyah governorate which is regarded as one of the non-endemic areas for CL [22]. However, after the start of the war from mid-2014, large-scale emigration of people occurred in Iraq. Garmian, as one of the border regions, housed a large number of refugees from other parts of conflict zones in Iraq; especially people from the endemic areas of Diyala, Kirkuk, Saladin and Mosul arrived in the region. This may have changed the region from a non-endemic to endemic area. This argument is supported by the large increase in the number of CL cases over the last 4 years in the region as shown in (Fig 3). Further study is required to confirm whether the region has become endemic by recording new cases who have not visited any endemic areas.
Regarding the monthly prevalence of CL in Garmian, the number of recorded CL cases started to increase from November (Fig 4). The maximum number of CL cases was recorded in January and February. The recorded numbers decreased from March and remained low until October. These findings agreed with those reported for other parts of Iraq [10]. As the incubation period of the disease ranged from two to four months, the majority of the cases recorded during winter months were probably bitten by insects from the summer to early autumn seasons.
Of the thirty samples collected from suspected new cases and CL patients receiving early treatment, 15 samples were positive for PCR targeting the leishmanial cytochrome b gene on gel electrophoresis, showing single bands with a product size of about 850–900 bp (Fig 5). Of the 15 positive samples, sequences of seven samples showing strong signals were determined, and all the Leishmania parasites were identified as L. major with 100% similarity with the L. major strain MRHO/IR/75/ER cytochrome b gene (GenBank accession number KU680828) [23]. This is the first molecular record of the L. major strain in Iraq using sequence analysis. Nonetheless, confirming only 7 cases out of a total 30 samples should be considered as a limitation of direct molecular tools for investigation of the L. major strain epidemiologically. Nucleotide sequence data reported will appear in the GenBank database under the accession numbers MF-370217-MF-370223.
The cytochrome b gene sequences obtained in this study were subjected to phylogenetic analysis together with those from representative Leishmania strains. The phylogenetic data indicated that Leishmania strains in the borderline area between northern and central Iraq were closely related to the Iranian MRHO/IR/75/ER strain and other L. major strains of Old World CL (Fig 6). This result is not surprising as the region shares its border with Iran which is the nearest country to the endemic area of central Iraq. Particularly, Diyala province has had a commercial relationship with Iran through the border close to the Khanaqin district since 2003.
Microsatellite analyses using specimens from geographically isolated areas, Central Asia, the Middle East and Africa showed L. major has little genetic variation when compared to L. tropica [25, 26]. This may partly explain the slight variation in L. major identified in this study. Further large-scale genetic analysis using more sensitive methods such as microsatellite analysis will be necessary in the future.
The exact origin of the parasite is unknown although there has been a history of the disease in Iraq [27]. In addition, there has been a long history of pilgrims visiting from Iran to sacred shrines in central and southern Iraq via either the Garmian region or central provinces.
The reservoirs of L. major have not been identified in Iraq, but in Iran, four gerbil species were identified as main reservoirs including Rhombomys opimus, Meriones libycus, Meriones hurrianae and Tatera indica [28]. Nevertheless, we recently identified the L. major MRHO/IR/75/ER strain from cutaneous lesions of a dog in the study area [24].
Further studies will reveal the transmission vectors and reservoirs and aid in control of the outbreaks. To our knowledge, this is the first record of a phylogenetic study in Iraq concerning L. major causing CL prevalence. The findings are also significant for future creation of vaccines against the Leishmania strain in Iraq and it is important to understand the global prevalence and epidemiology of the Leishmania strains.
In conclusion, we identified the L. major strain in Iraq for the first time using PCR and DNA sequencing. In addition, phylogenetic study revealed the main Leishmania genotype causing health problems in the borderline area between the non- endemic area in the north and the endemic region of central Iraq. The identified parasite was similar to the MRHO/IR/75/ER strain which is endemic in Iran. Furthermore, we reported new CL cases in the borderline area between central and northern Iraq, particularly in the Garmian region. This region can be regarded as a model for further study of epidemic CL outbreaks to other non-endemic areas. This study also suggests that researchers conduct more studies regarding the threat of CL which may spread globally to countries where both reservoirs and sandflies are present.
|
10.1371/journal.pntd.0005814 | High prevalence of thiamine (vitamin B1) deficiency in early childhood among a nationally representative sample of Cambodian women of childbearing age and their children | Thiamine deficiency is thought to be an issue in Cambodia and throughout Southeast Asia due to frequent clinical reports of infantile beriberi. However the extent of this public health issue is currently unknown due to a lack of population-representative data. Therefore we assessed the thiamine status (measured as erythrocyte thiamine diphosphate concentrations; eThDP) among a representative sample of Cambodian women of childbearing age (15–49 y) and their young children (6–69 mo).
Samples for this cross-sectional analysis were collected as part of a national micronutrient survey linked to the Cambodian Demographic and Health Survey (CDHS) 2014. One-sixth of households taking part in the CDHS were randomly selected and re-visited for additional blood sampling for eThDP analysis (719 women and 761 children). Thiamine status was assessed using different cut-offs from literature.
Women were mean (SD) 30 (6) y, and children (46% girls) were 41 (17) mo. Women had lower mean (95% CI) eThDP of 150 nmol/L (146–153) compared to children, 174 nmol/L (171–179; P < 0.001). Using the most conservative cut-off of eThDP < 120 nmol/L, 27% of mothers and 15% of children were thiamine deficient, however prevalence rates of deficiency were as high as 78% for mothers and 58% for children using a cut-off of < 180 nmol/L. Thiamine deficiency was especially prevalent among infants aged 6–12 mo: 38% were deficient using the most conservative cut-off (< 120 nmol/L).
There is a lack of consensus on thiamine status cut-offs; more research is required to set clinically meaningful cut-offs. Despite this, there is strong evidence of suboptimal thiamine status among Cambodian mothers and their children, with infants <12 mo at the highest risk. Based on eThDP from this nationally-representative sample, immediate action is required to address thiamine deficiency in Cambodia, and likely throughout Southeast Asia.
| Thiamine is an often-overlooked micronutrient of concern in Cambodia and throughout Southeast Asia, where reports of beriberi are not uncommon due to a diet of thiamine-poor, white, polished rice. Thiamine plays a critical role in cellular energy generation, and also modulates neuronal and neuromuscular transmissions. Thiamine deficiency can progress to beriberi, which can be fatal. Although several recent studies have investigated thiamine status and/or beriberi in the region, this is the first nationally-representative biochemical thiamine data from any country in Southeast Asia. Unfortunately, there is a lack of clinically meaningful cut-offs to interpret these data. Upwards of 10 different cut-offs exist, but many are simply the lower bounds of a reference range, and therefore do not align with clinical symptoms of beriberi. Using the most conservative cut-off from the literature, 27% of mothers and 15% of children were thiamine deficient. More research is required to develop more useful, clinically meaningful thiamine status cut-offs. In addition, given the distinctive peak in Cambodian infant mortality data suggestive of infantile beriberi, immediate action is required to improve the thiamine status in Cambodia, and likely in other countries in Southeast Asia.
| Beriberi is a ‘forgotten disease’ [1–4] that remains a public health issue in Southeast Asia despite near eradication elsewhere [5–7]. Beriberi is caused by thiamine (vitamin B1) deficiency, and is most serious in infants due to the rapid growth and development that occurs during this time, and the relatively high thiamine needs compared to body size [4,8]. Breast milk thiamine concentrations reflect maternal dietary thiamine intake [9]. As such, poor maternal thiamine status during pregnancy and lactation puts infants at risk of developing beriberi [9–11], which can lead to death in hours of clinical presentation if not recognized or left untreated [6]. While infantile beriberi is the most serious outcome of thiamine deficiency, marginal thiamine status in the wider population causes fatigue, apathy, anorexia, and dizziness [12], and with these the potential for decreased school performance and/or economic output. In addition, Israeli children who consumed thiamine-deficient infant formula in infancy, but did not develop beriberi, exhibited retarded neurological, cognitive, and cardiological development at age 5–7 y [13], highlighting the importance of thiamine sufficiency in early life.
Beriberi remains a problem in Southeast Asia, in part, because non-parboiled [14], unfortified white rice is the dietary staple [8,12]. In Cambodia, white rice makes up an estimated 60% of daily dietary energy [15]. Although rice contains thiamine [8], it is found only in the outer husk and bran, the vast majority of which is removed during the milling process [14]. In most rice-consuming cultures, polished white rice is preferred [14,16] for several reasons: organoleptic qualities, white rice as a status symbol [12], and because removal of the lipid-rich outer bran increases shelf-life [14]. In Cambodia brown rice is a cultural dietary taboo; people were forced to eat brown rice during the Khmer Rouge regime [17].
Although there are several reports of infantile beriberi in Cambodia [5,18,19], there is a lack of accurate prevalence data. The World Health Organization has suggested that in the absence of reliable information on the prevalence of beriberi or on biochemical markers of thiamine status, infant mortality curves could be indicative of thiamine deficiency being a health concern, with a peak in infant mortality around 3–4 months of age being suggestive of a high prevalence of beriberi [20]. We analyzed infant mortality data from the Cambodian Demographic Health Surveys (CDHS) 2000, 2005, 2010 and 2014 and found indeed a peak in mortality around 3 months of age (Fig 1).
Thiamine status has traditionally been assessed using a functional indicator, erythrocyte transketolase activity coefficient [21,22], however, this method has several shortcomings including the inactivation of transketolase during sample processing and storage, poor inter-assay precision [23], and a tendency of this assay to underreport deficiency among chronically deficient individuals [24]. More recently, the biochemical assessment of the biologically active form of vitamin B1, thiamine diphosphate, in whole blood or in erythrocytes has been advocated [2]. Erythrocyte thiamine diphosphate concentrations (eThDP) measured by high performance liquid chromatography (HPLC) overcomes several downfalls of the functional assay, and correlates well with erythrocyte transketolase activity coefficient [23]. Coats and colleagues reported that Cambodian mother-infant dyads in Prey Veng province had significantly lower whole blood thiamine diphosphate concentrations, regardless of infant clinical beriberi diagnosis, compared to American controls [5]. Lower eThDP were reported in a representative survey of Cambodian women of childbearing age residing in urban Phnom Penh and rural Prey Veng provinces compared to a small convenience sample of purportedly thiamine-replete Canadian women from Vancouver [25,26].
Unfortunately there is currently a lack of consensus on the most appropriate cut-offs to define suboptimal status or deficiency using whole blood ThDP or eThDP, and there are no nationally representative data available on biochemical thiamine status of any population group in Southeast Asia. Therefore, the objective of this study was to determine eThDP among women of childbearing age and their children aged 6–69 mo who participated in the most recent 2014 Cambodian Demographic and Health Survey (CDHS) [27] and the linked National Micronutrient Survey [28] to determine the prevalence of thiamine deficiency in this population using various cut-offs.
This biochemical thiamine analysis was part of the 2014 Cambodian National Micronutrient Survey [28] (conducted June 2 –December 12, 2014), which was linked to the CDHS, a nationally-representative survey of adults aged 15–49 y and children from 24 Cambodian provinces [27]. Population proportionate to size sampling was used to select 611 villages from which 16,356 individual households were selected. Trained, Khmer-speaking enumerators visited all selected households and collected information on health outcomes including nutrition, fertility and family planning, morbidity and mortality, housing, and assets and wealth using a validated, standardized questionnaire [27]. One week to 2 months after the CDHS had visited the household, one sixth of the households were re-visited and biological samples were collected from mothers and their children. Full sampling and survey details can be found elsewhere [27,28]. With one sixth of the households re-visited, it was estimated that 935 mothers and children could be included in the micronutrient survey, but due to absence of mother or care-takers and refusal to participate, blood samples were collected from only 726 women and 781 children, and eThDP was measured in 726 and 761 samples, respectively. Pregnant women (n = 7) were excluded, leaving 719 maternal samples for analysis. Hemoglobin concentrations were measured by the CDHS survey, and not repeated during the micronutrient survey. As hemoglobin was measured in only half of the children and women, data on hemoglobin and anemia prevalence is only available for 441 women and 476 children.
The National Ethics Committee for Health Research, Phnom Penh, Cambodia granted ethical approval (057 NECHR 2014) for the 2014 Cambodian National Micronutrient Survey. Eligibility for participation included: having participated in the CDHS and have given permission to the CDHS team to be re-visited for the micronutrient survey, a child in the household aged 6–59 mo, neither mother nor child having evidence of severe or chronic illness, and mothers or care-taker providing written, informed consent. All data was anonymized.
Nurses collected non-fasting, morning-time blood samples into trace element-free, heparin-coated tubes (Vacuette, Greiner Bio One, Austria) at a central village location. Samples were stored in a dark cooler box and transported to the nearest Provincial Health Centre within 6 h of collection. Samples were centrifuged (3000 g for 10 min), plasma and buffy coat were thoroughly removed, and erythrocytes were separated into 500 μL aliquots and frozen at -20°C. Frozen samples were then transported to the Department of Fisheries Post-Harvest Technologies and Quality Control Laboratory, Fisheries Administration in Phnom Penh for storage at -20°C. Once the survey was completed the samples were batch shipped on dry ice to Abbot Laboratories in Singapore for eThDP analysis.
After arrival of the blood samples in Abbott Nutrition R&D Singapore Center, the samples were stored at -78°C. eThDP was measured using a modified method of Lu & Frank [29], as reported elsewhere [30]. Briefly, samples were thawed on ice in a dark room with amber light. Trichloroacetic acid solution was added to precipitate protein out. After centrifugation, the supernatant was collected, washed with methyl tert-butyl ether, and subject to ultra-high performance liquid chromatography (Agilent model 1290 system, Singapore) with a fluorescence detector (Agilent model G1321A, Agilent Technologies, Singapore) and an autosampler (Agilent model G4226A, Agilent Technologies, Singapore) that allowed online pre-column derivatization with potassium ferricyanide.
At least ten different thiamine status cut-offs are reported in literature for use in women and children [5,21,22,31–40]. Noteworthy are the Institute of Medicine (IOM) definitions of thiamine deficiency (eThDP <70 nmol/L) and marginal deficiency (70–90 nmol/L) [22], which are based on values from 68 healthy Dutch blood donors and laboratory staff aged 20–50 y [31]. In the original citation these are described as cut-offs for whole blood or red cells [41], causing confusion over the correct biological sample for their use. If the IOM cut-offs represent whole blood ThDP, then values should be corrected for hematocrit to obtain eThDP values [34]. For example, the Coats et al. noted that their lab uses a reference range of 80–150 nmol/L for whole blood ThDP, or, if divided by hematocrit, 150–290 nmol/L for eThDP [5]. The Institute of Medicine cut-offs have not been employed here due to confusion surrounding use for whole blood versus erythrocyte ThDP.
We have made a distinction between cut-offs being reported as falling below a ‘reference range’, or ‘deficient’ or ‘marginally deficient’, as it is unclear whether a value outside a reference range represents real deficiency. For example, a thiamine deficiency cut-off of eThDP <180 nmol/L was proposed by Mancinelli and colleagues to align with the 25th percentile eThDP of 103 healthy controls (45 men and 58 women, employees of University “La Sapienza” Hospital) in Rome, however none of these subjects were thiamine deficient [38]. In best practice, an average of 120 subjects are needed to generate accurate reference limits for a given biomarker [42]; this has not been the case for the majority of thiamine cut-offs.
We present four cut-off values describing suboptimal status below a reference range: the abovementioned eThDP <180 nmol/L [38]; eThDP <165 nmol/L, the lower bound of a 95% reference range (165–286 nmol/L) of 48 (25 men and 23 women) healthy hospital staff at Broadgreen Hospital, Liverpool, UK [32]; <150 nmol/L, corresponding to the eThDP reference range of the Mayo Medical Laboratories [5]; and <140 nmol/L, the lower limit of normal eThDP (cut-off of lowest 2.5%) of healthy blood donors in Christchurch, New Zealand; n unknown [33]. eThDP <148 nmol/L was used to categorize low thiamine status in two studies [35,36], and originated as the lower bound of normal range (50–150 ng/mL packed cells) from 21 healthy adults in Nashville, Tennessee [37]. Since the values of 148 and 150 nmol/L are close, we have used only <150 nmol/L as a cut-off in the current paper.
Marginal thiamine deficiency has been described using one cut-off, eThDP between 120–150 nmol/L, a cut-off reported in [21,39], but no details of these values are known. Two cut-offs for thiamine deficiency have been reported: eThDP <120 nmol/L, again that was used in [21,39], but the origins of this cut-off are unknown; and <118.5 nmol/L, which was used to categorize thiamine deficiency in [35], and is described as below the 95 percentile reference range (40–85 μg/L) among healthy black South African adults in [40]. As these values are close again, we have used <120 nmol/L as cut-off for thiamine deficiency.
Demographic characteristics were computed as mean (SD) or n (%), and eThDP as mean (95% CI). Children’s eThDP and thiamine status are categorized by age category, 6–12 mo, 13–24 mo, 25–36 mo, 37–59 mo, and ≥ 60 mo. A t-test was performed to compare women and children’s eThDP, and eThDP among residents in rural and urban areas; a one-way ANOVA was employed to compare eThDP among different wealth quintiles, and children’s age categories (with least significant difference post-hoc correction for multiple comparisons).
Linear regression models were built to measure the association between eThDP and various independent variables. Variables were included in the linear regression model if P < 0.20 in bivariate correlation, and were entered stepwise into separate models for women and children. The following variables were evaluated for inclusion in the model for both women and children: province, wealth quintile, population density (urban/rural), mother’s education, cigarette smoke in home, subsidized healthcare available for household, weight, height/length, age, and hemoglobin concentration. The model for women also included BMI, and the children’s model also included sex and birth order.
Results were considered significant at P < 0.05. Weight-for-age, height-for-age, BMI-for-age, and weight-for-height z-scores were calculated using WHO Anthro and WHO Anthro Plus software programs, otherwise all analyses were performed using SPSS for Macintosh version 23.0 (IBM, Armonk, NY, USA).
Demographic characteristics are shown in Table 1. Women were 30 (6) y, and the majority had a normal BMI (70%; 18.5–24.99 kg/m2), were married (93%), and had attended some formal schooling (93%). Children were 41 (17) mo and 46% were girls; 28% (n = 130) of children were wasted (weight-for-age z score < -2 SD), and 39% (n = 182) were stunted (height-for-age z score < -2 SD).
Children had a higher mean (95% CI) eThDP of 174 nmol/L (171–179 nmol/L) compared to women, 150 nmol/L (146–153 nmol/L; P < 0.001); Table 2. eThDP did not differ between children living in rural (173 nmol/L, 169–177 nmol/L) versus urban areas (180 nmol/L, 172–189 nmol/L; P = 0.14), however, rural women had lower eThDP (146 nmol/L, 143–150 nmol/L) compared to their urban peers (164 nmol/L, 157–171 nmol/L; P <0.001). Compared to higher wealth quintiles, eThDP was lower among both children (P = 0.04) and women (P < 0.001) in lower wealth quintiles. Young children aged 6–12 mo had significantly lower eThDP (144 nmol/L, 130–159 nmol/L) compared to older children aged 13–36 mo (176 nmol/L, 170–173 nmol/L; P < 0.001) and > 36 mo (177 nmol/L, 172–182 nmol/L; P < 0.001); eThDP in the latter two age groups did not differ (P = 0.90).
Table 3 shows the percentage of women and children below selected cut-offs. Using the most conservative cut-off for thiamine deficiency (eThDP < 120 nmol/L), 27% of mothers and 15% of children were thiamine deficient. Worrisome, 38% of infants were thiamine deficient.
The following variables were included in the eThDP prediction linear regression models: for children, age (mo), hemoglobin concentration (g/dL), and household wealth quintile; for women, household wealth quintile, household qualification for subsidized health, province of residence, population density (rural/urban), education level attended, and hemoglobin concentration (g/dL) were included. Hemoglobin concentration was the only predictor of children’s eThDP (adjusted R2 = 0.024, standardized β [95% CI], 0.161 [3.2–11.3 g/dL], P < 0.001). The model for women’s eThDP (adjusted R2 = 0.044) included wealth quintile (standardized β [95% CI], 0.209 [3.6–10.0], P < 0.001) and hemoglobin concentration (standardized β [95% CI], 0.114 [0.6–7.9 g/dL], P = 0.02).
Here we present the first nationally representative biochemical thiamine status data from a country in Southeast Asia, a region where beriberi still exists [5–7]. Despite variation in the prevalence of thiamine deficiency by cut-off, there is clear evidence of suboptimal thiamine status among women of childbearing age and their children in Cambodia. Of highest concern are infants aged 6–12 mo (n = 50), of whom 38% was classified as thiamine deficient by the most conservative cut-off, and up to 70% using the most liberal one (eThDP < 180 nmol/L). Although not included in this study, Cambodian infants aged 0–6 mo, who are at the highest risk for developing infantile beriberi [6,8] due to the relatively high thiamine needs compared to their body size [8], are likely to have a poor thiamine status too. Given the peak in infant mortality around 3 months of age in Cambodia, combined with the biochemical evidence of a high prevalence of thiamine deficiency in the population, we are convinced that infant beriberi is a highly under recognized cause of death in Cambodia, and that infants <12 mo of age are at the highest risk for thiamine deficiency. Indeed, Kauffman et al. estimated that infantile beriberi might be responsible for 6% of overall infant mortality in Cambodia [44]. Whereas other causes of infant mortality have been addressed, leading to a considerable decrease in infant mortality over the last 2 decades [27], the peak at 3 months of age has remained (see Fig 1).
Children had significantly higher eThDP than their mothers (174 versus 150 nmol/L; P < 0.001). This is consistent with a recent study in Cambodia in which we measured eThDP among women (18–45 y) and their children (6–59 mo) in Prey Veng province as part of a randomized controlled trial investigating thiamine-fortified fish sauce [45]. Women and children in the control group (who received only nutrition education; 92 mothers and 87 children) had mean (95% CI) eThDP of 184 nmol/L (169–198 nmol/L) and 213 nmol/L (202–224 nmol/L), respectively. Coats and colleagues reported similar values among Cambodian women of childbearing age: 141 and 150 nmol/L (thiamine diphosphate in whole blood, corrected for hematocrit) among mothers of infants with and without the clinical symptoms of infantile beriberi, respectively [5]. The fact that Coats et al. reported ThDP concentrations in infants with beriberi and without beriberi that are close or within to the lower reference ranges for ThDP (<140 and <150 nmol/L [5,33]) suggests that these cut-offs might be too conservative. Alternatively, perhaps eThDP (or whole blood ThDP) is not a good predicator of beriberi. However, reports of beriberi are not uncommon in Cambodia and throughout Southeast Asia [5–7,18,19,46,47], suggesting that thiamine deficiency is an issue. Therefore our biochemical data suggesting a high prevalence of thiamine deficiency suggest that current cut-offs can indicate populations at risk for thiamine deficiency even though the cut-offs might not be clinically meaningful.
These Cambodian values do differ greatly from older, previously reported values among older children and adults in Europe. The mean eThDP of British adolescent girls and boys (n = 54, 13–14 y) was 226.8 nmol/L and 206.1 nmol/L, respectively [35], which are similar to those reported for free-living British elderly women (247 nmol/L, n = 80) and men (218 nmol/L, n = 57) [36]. However, due to advances in HPLC equipment and sensitivity since these latter studies were published two decades ago [35,36,39], there is merit in investigating whether different thiamine deficiency cut-offs need to be developed for adults and children.
As shown in Table 3, there is wide variation in the prevalence of thiamine deficiency and/or suboptimal thiamine status depending on the cut-off employed, therefore it is difficult to determine the severity of low thiamine status as a public health concern in Cambodia and the wider region. However, even with the most conservative cut-offs, >25% of the mothers, and 38% of the infants were classified as deficient in our study, making thiamine deficiency a serious public health concern, which is also reflected in the peak in infant mortality around 3 months of age. Wilkinson et al. used the lowest 2.5% of healthy blood donors in Christchurch, New Zealand to set their deficiency cut-off of eThDP < 140 nmol/L, but astutely noted that “the lower limit of normal, seen in a healthy population, cannot be assumed to be the upper limit of abnormal” [33]. While the cut-offs shown in Table 3 may be helpful in categorizing potentially at-risk individuals, it is clear that more research is required to develop clinically meaningful thiamine deficiency cut-offs. Although a recent review has not found compelling human nutrition trials to prioritize an update of recommended dietary thiamine intakes [48], this may be due to a heavy focus on beriberi as an outcome. Thiamine is involved in important cell functions including glucose conversion and energy metabolism in the Kreb’s cycle and pentose phosphate pathway [49]. Thiamine also performs critical enzymatic functions in processes related to brain development and function, neuronal communication, as well as immune system activation, signaling and maintenance [50]. There is evidence that obesity may impair thiamine utilization and alter requirements: a recent American study reported thiamine deficiency in 15.5–29% of obese patients seeking bariatric surgery [51]. In addition, considerable evidence over the past century has linked thiamine deficiency to neurological problems including cognitive deficits and encephalopathy [52]. Even a short-term exposure to poor thiamine intake in early life may have long-term impacts on cognition [53,54]. It is difficult to establish new thiamine cut-offs without new human studies on these manifestations of thiamine deficiency that are distinctly different from beriberi. Perhaps a wider description of clinical syndromes of thiamine deficiency is needed, compiling all under one term, for example, Thiamine Deficiency Disorders, just as advancing insights into iodine deficiency in the early 1980s led to the use of the term Iodine Deficiency Disorders. And perhaps circulating thiamine concentrations might not sufficiently reflect thiamine status. We further urge future researchers to collect beriberi prevalence data from clinical settings with matched biochemical samples to better guide development of cut-offs that have clinical and/or physiological meaning.
This study has several strengths, but most notably it is the first nationally representative evaluation of eThDP in a country in Southeast Asia (while a national study in the Philippines using erythrocyte transkelotase activity is the most recent in the region in two decades [55]). Thiamine status was measured among women of childbearing age and children because, due to unequal household food distribution and higher needs relative to body size, these groups are at highest risk of nutritional deficiencies in low-income countries. These are also the most pertinent population groups for biochemical thiamine assessment because thiamine-deficient mothers confer a higher risk of infantile beriberi [9–11], and in turn mortality [6], to their children. There is evidence that increased thiamine intake improves biochemical thiamine status among mothers and their infants in Cambodia [30]. During a recent randomized controlled trial in Prey Veng we found that maternal consumption of thiamine-fortified fish sauce over 6 mo throughout pregnancy and lactation resulted in higher eThDP among mothers and infants, as well as breast milk thiamine, compared to a control sauce containing no thiamine [30], indicating that this group shows potential for improvement in eThDP. However, thiamine deficiency may be common among the elderly [56], and beriberi outbreaks among adult men have also been reported [47,57], likely due to higher thiamine needs with increased physical activity and a high-carbohydrate diet [22,49,58]. Therefore future studies should include the full range of population groups.
Dietary intake data was not collected in this study, therefore while it is well established that low dietary thiamine intake causes low biochemical thiamine status [22], and that there is little thiamine available in the Cambodian diet [15,59], we cannot provide direct causation for low eThDP in this population.
Consistent with recent studies of thiamine status in Cambodia, we report low eThDP among a nationally representative sample of Cambodian women and their young children. Thiamine status classification varies dramatically depending on the cut-off employed, from 27% to 78%, and 15% to 58% among mothers and children, respectively. More research is required to develop more useful, clinically meaningful thiamine status cut-offs. However, in view of the peak in infant mortality around 3 months of age suggestive of infantile beriberi, immediate action is required to develop interventions to increase thiamine intake in Cambodia and the wider Southeast Asia region where thiamine deficiency and beriberi remain a public health concern.
|
10.1371/journal.pntd.0001384 | Accuracy of Urine Circulating Cathodic Antigen (CCA) Test for Schistosoma mansoni Diagnosis in Different Settings of Côte d'Ivoire | Promising results have been reported for a urine circulating cathodic antigen (CCA) test for the diagnosis of Schistosoma mansoni. We assessed the accuracy of a commercially available CCA cassette test (designated CCA-A) and an experimental formulation (CCA-B) for S. mansoni diagnosis.
We conducted a cross-sectional survey in three settings of Côte d'Ivoire: settings A and B are endemic for S. mansoni, whereas S. haematobium co-exists in setting C. Overall, 446 children, aged 8–12 years, submitted multiple stool and urine samples. For S. mansoni diagnosis, stool samples were examined with triplicate Kato-Katz, whereas urine samples were tested with CCA-A. The first stool and urine samples were additionally subjected to an ether-concentration technique and CCA-B, respectively. Urine samples were examined for S. haematobium using a filtration method, and for microhematuria using Hemastix dipsticks.
Considering nine Kato-Katz as diagnostic ‘gold’ standard, the prevalence of S. mansoni in setting A, B and C was 32.9%, 53.1% and 91.8%, respectively. The sensitivity of triplicate Kato-Katz from the first stool and a single CCA-A test was 47.9% and 56.3% (setting A), 73.9% and 69.6% (setting B), and 94.2% and 89.6% (setting C). The respective sensitivity of a single CCA-B was 10.4%, 29.9% and 75.0%. The ether-concentration technique showed a low sensitivity for S. mansoni diagnosis (8.3–41.0%). The specificity of CCA-A was moderate (76.9–84.2%); CCA-B was high (96.7–100%). The likelihood of a CCA-A color reaction increased with higher S. mansoni fecal egg counts (odds ratio: 1.07, p<0.001). A concurrent S. haematobium infection or the presence of microhematuria did not influence the CCA-A test results for S. mansoni diagnosis.
CCA-A showed similar sensitivity than triplicate Kato-Katz for S. mansoni diagnosis with no cross-reactivity to S. haematobium and microhematuria. The low sensitivity of CCA-B in our study area precludes its use for S. mansoni diagnosis.
| We aimed to assess the accuracy of a commercially available rapid diagnostic test for the detection of an infection with the blood fluke Schistosoma mansoni in urine. In total, 446 school children from three different settings of south Côte d'Ivoire provided three stool and three urine samples. Stool samples were examined with the widely used Kato-Katz technique and analyzed with a microscope for S. mansoni eggs. Urine samples were examined with a filtration method for S. haematobium eggs and with a rapid diagnostic test for S. mansoni that is based on detecting circulating cathodic antigens (CCA). We used a commercially available test (designated CCA-A) and an experimental formulation (CCA-B). Examination of nine Kato-Katz thick smears per child revealed a prevalence of S. mansoni in the three settings of 32.9%, 53.1%, and 91.8%. The sensitivity of triplicate Kato-Katz from the first stool sample was comparable to a single CCA-A (47.9–94.2% vs. 56.3–89.6%), and significantly higher than the sensitivity of a single CCA-B test (10.4–75.0%). CCA-A showed a considerably lower specificity than CCA-B (76.9–84.2% vs. 96.7–100%). In the settings studied in south Côte d'Ivoire, the CCA-A test holds promise for the diagnosis of S. mansoni, whereas results with CCA-B were suboptimal.
| There is growing awareness, political commitment, and financial resources to control neglected tropical diseases (NTDs) [1]–[3]. Preventive chemotherapy, that is the repeated large-scale administration of drugs to at-risk populations, has become the key strategy for the control of several NTDs, including schistosomiasis [3]–[5]. Although the issue of diagnosis has received only token attention in the current era of preventive chemotherapy, its importance must be emphasized for rapid identification of high-risk communities warranting regular treatment, appraisal of drug efficacy, monitoring progress of control interventions, and improved patient management [6]–[8]. With regard to intestinal schistosomiasis due to Schistosoma mansoni and S. japonicum, the Kato-Katz technique is the most widely used diagnostic approach in epidemiological surveys [8], [9]. Although the Kato-Katz technique is relatively simple to perform, it requires a minimum of equipment (i.e., microscope, chemicals, and test kit material) and well-trained laboratory technicians [10]. Moreover, a shortcoming of the Kato-Katz technique is the only low-to-moderate sensitivity for S. mansoni diagnosis in low endemicity areas [11]–[13]. Hence, multiple Kato-Katz thick smears are required to enhance sensitivity [14], but this poses operational challenges and strains financial resources.
The detection of circulating antigen of S. mansoni in urine has been suggested as an alternative to the Kato-Katz technique [15]–[17]. Indeed, both circulating anodic antigen (CAA) and circulating cathodic antigen (CCA) can be detected in sera and urine of individuals infected with S. mansoni [18]. Both antigen-detecting assays are sensitive and specific and correlate with the presence and intensity of infection [19]. Antigen detection in urine using a rapid diagnostic test (RDT) based on an enzyme-linked immunosorbent assay (ELISA) technique is potentially useful and non-invasive and could change the management of infected individuals, particularly at the peripheral level in endemic countries where microscopes and qualified laboratory technicians are often not available [6], [7]. A point-of-contact (POC) CCA urine test has been developed for the diagnosis of S. mansoni [15], which is now commercially available as a RDT in cassette form. In view of promising results obtained thus far [17], [20], [21], the Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) initiated a multi-country study to assess the accuracy of a commercially available CCA cassette test for the diagnosis of S. mansoni.
The study reported here is part of this multi-country evaluation. We assessed the accuracy of a commercially available urine CCA cassette test (designated CCA-A) for S. mansoni diagnosis. Additionally, we employed an experimental formulation of the test (CCA-B). Nine Kato-Katz thick smears from each participant served as diagnostic ‘gold’ standard. In addition, our team employed the ether-concentration method on sodium acetate-acetic acid-formalin (SAF)-fixed stool samples for the diagnosis of S. mansoni, urine filtration for the identification of S. haematobium eggs, and Hemastix dipsticks for the detection of microhematuria in urine. The study was carried out in south Côte d'Ivoire, in three settings where S. mansoni is endemic at different levels, whereas S. haematobium co-exists in one of the settings.
The study protocol was approved by the institutional research commission of the Swiss Tropical and Public Health Institute (Basel, Switzerland) and was cleared by the ethics committees of Basel (EKBB; reference no. 377/09) and Côte d'Ivoire (reference no. 1993 MSHP/CNER). District health and education authorities, village chiefs, parents/legal guardians, and participating children were informed about the purpose and procedures of the study. Parents/legal guardians provided written informed consent for their children to participate. Additionally, all children assented orally. Participation was voluntary and children could withdraw at any time without further obligation. All parasitological results were coded and treated confidentially. At the end of the study, children attending the schools involved in this study were treated with praziquantel (single 40 mg/kg oral dose) and albendazole (single 400 mg oral dose) free of charge, irrespective of the child's helminth infection status [22].
In October/November 2010, we carried out a cross-sectional survey in three epidemiological settings in the district of Azaguié, south Côte d'Ivoire. Azaguié is located approximately 40 km north of Abidjan, the economic capital of Côte d'Ivoire. The settings were selected after a pre-screening done in 10 schools. For the pre-screening, in each school, 25 children were randomly selected. All children attending grades 3–5 (CE1, CE2, and CM1) were given a unique number, lots including all numbers were closed and placed in a box, and finally 25 lots per school were drawn. The selected children provided a single stool and a single urine sample, which were examined for S. mansoni with triplicate Kato-Katz thick smears and S. haematobium with a single filtration, respectively. Based on this pre-screening, we selected the following sites, according to SCORE guidelines: setting A, low S. mansoni endemicity (i.e., prevalence: 10–24%); setting B, moderate S. mansoni endemicity (prevalence: 25–49%); and setting C, co-endemic for S. mansoni and S. haematobium.
According to the literature, a single Kato-Katz thick smear for diagnosis of S. mansoni in low endemicity settings has a sensitivity of only 20–30% [23], [24]. However, since our study was to be carried out in both low and moderate endemicity settings, we assumed that a single Kato-Katz thick smear has a maximum sensitivity of 60%. The sensitivity of the CCA test is reported to be 80% or higher [15], [25]. Using these sensitivity estimates, a significance level of 5%, and a power of 80%, our sample size of complying children was calculated at 90. Assuming a compliance of 70% for the submission of each of three requested stool samples, the number of children to be included in each study setting was at least 199. To achieve this sample size, we selected by computer-based randomization 220 children aged 8–12 years from readily available school lists of Abbé-Begnini (setting A), Azaguié Gare (setting B), and M'Bromé/Makouguié (setting C).
The purpose and procedures of the study were explained to the village authorities, the school directors, and the teachers of the selected schools. Teachers were invited to prepare class lists, including names, sex, and age of the children attending grades 3–5. Next, the study was explained to the children in lay terms and they were provided with an information and consent sheet with further details of the study and children and parents' rights. Children who submitted a written informed consent from their parents/guardians and assented orally themselves were given a 125 ml plastic container labeled with a unique identifier (ID). Children were invited to return the containers filled with a fresh lime-sized morning stool sample the following day. Upon collection of the filled container, a new empty container was handed out for stool collection on the next day. This procedure was repeated over a week until most children had submitted a total of three stool samples. Each day, between 10:00 and 12:00 hours, participating children were provided with another empty container labeled with the respective ID for collection of urine samples.
Stool and urine samples were transferred to a laboratory at the Université de Cocody and processed the same day. From each stool sample, triplicate Kato-Katz thick smears were prepared, using 41.7 mg templates, following standard protocols [9]. In brief, triplicate Kato-Katz thick smears were prepared on microscope slides, labeled with a child's ID plus letter A, B, or C. Slides were allowed to clear for at least 30 min before quantitative examination under a microscope by experienced laboratory technicians. The number of S. mansoni and other helminth eggs (e.g., Ascaris lumbricoides, hookworm, and Trichuris trichiura) was counted and recorded for each species separately. For quality control, 10% of the Kato-Katz thick smears were re-examined by a senior technician.
In addition, from the second day stool sample, ∼1 g of feces was weighed into plastic vials containing 10 ml of a SAF solution. Within 8 weeks, the SAF-fixed stool samples were processed with the ether-concentration method, following a standard protocol [12], [26]. In brief, the stool-SAF solution was rigorously shaken and then poured through medical gauze placed on a plastic funnel into a conical glass tube. The conical tubes were centrifuged for 1 min at 500× g. Subsequently, the supernatant was discarded and 7 ml of 0.85% sodium chloride (NaCl) solution and 2–3 ml ether were added to the pellet. Tubes were closed with a rubber stopper, manually shaken for ∼30 sec and then centrifuged for 5 min at 500× g. This procedure leads to the separation of the suspension in four layers. The three top layers were discarded and the complete sediment layer was placed on a microscope slide, covered with a slip and subsequently examined under a microscope for helminth eggs (i.e., S. mansoni and soil-transmitted helminths) and intestinal protozoon cysts.
All urine samples were subjected to CCA-A (batch 32727) on the day of sample collection. The first urine sample was additionally subjected to CCA-B (batch 32686). Both CCA urine cassette assays were obtained from Rapid Medical Diagnostics (Pretoria, South Africa) and performed at ambient temperature, following the manufacturer's instructions. Briefly, one drop of urine was added to the well of the testing cassette and allowed to absorb. Once fully absorbed, one drop of buffer (provided with the CCA test kits) was added. The test results were read 20 min after adding the buffer. In case the control bands did not develop, the test was considered as invalid. Valid tests were scored as either negative or positive, the latter further stratified into 1+, 2+, or 3+ according to the visibility of the color reaction. All tests were read independently by two blinded investigators and in case of discordant results discussed with a third independent investigator until agreement was reached.
In addition to the CCA cassettes, each urine sample was subjected to a filtration method for S. haematobium egg counts and to a Hemastix dipstick (Siemens Healthcare Diagnostics GmbH; Eschborn, Germany) for microhematuria assessment on the day of sample collection. In brief, samples were shaken, and 10 ml of urine filtered through a 13-mm diameter small meshed filter (20 µm; Sefar AG; Heiden, Switzerland), which was then placed on a labeled slide and examined under a microscope for S. haematobium eggs [8]. For appraisal of microhematuria, a Hemastix dipstick was soaked in urine, left in the open air for 1 min, before scoring according to the manufacturer's instructions.
Data were entered twice in a Microsoft Excel spreadsheet, transferred in EpiInfo version 6.4 (Centers for Disease Control and Prevention; Atlanta, GA, USA) and validated. Statistical analyses were done with STATA version 10 (Stata Corp.; College Station, TX, USA).
Only those children who had complete data records were included in the final analysis (i.e., nine Kato-Katz thick smears, a single ether-concentration, three CCA-A, one CCA-B, three urine filtrations, and three Hemastix dipsticks). To obtain a standardized measure of infection intensity, expressed as eggs per gram of stool (EPG), for each individual, we calculated the arithmetic mean S. mansoni fecal egg counts (FECs) from the nine Kato-Katz thick smears and multiplied by a factor 24. Infection intensity of S. mansoni was classified into light (1–99 EPG), moderate (100–399 EPG), and heavy (≥400 EPG). Egg counts of S. haematobium were utilized to stratify into light (1–49 eggs/10 ml of urine) and heavy infection intensities (≥50 eggs/10 ml of urine) [4].
The strength of agreement between nine Kato-Katz thick smears and triplicate CCA-A, one CCA-B, and one ether-concentration for each endemicity setting was assessed by kappa statistics (κ), as follows: κ<0 indicating no agreement, κ = 0–0.2 indicating poor agreement, κ = 0.21–0.4 indicating fair agreement, κ = 0.41–0.6 indicating moderate agreement, κ = 0.61–0.8 indicating substantial agreement, and κ = 0.81–1.0 indicating almost perfect agreement [27], [28].
As proposed by the SCORE secretariat, the results from nine Kato-Katz thick smears were considered our ‘gold’ standard. We determined the sensitivity (proportion of true-positives detected by the test) and specificity (proportion of true-negatives detected by the test) of single and multiple tests. As with some of our previous work, we used a second ‘gold’ standard by considering a positive test result (regardless of the test) as true-positive [29], [30]. Hence, we combined results from all tests (i.e., nine Kato-Katz thick smears plus triplicate CCA-A, one CCA-B, and one ether-concentration) and therefore maximized specificity.
We employed an ordinal logistic regression approach, which is an extension of the general linear model to ordinal categorical outcomes to assess the correlation between CCA-A and CCA-B color reaction categories and S. mansoni FECs. The arithmetic mean FEC of three Kato-Katz thick smears per stool sample per day served as continuous explanatory variable, whereas the color reaction of the CCA test was considered as categorical outcome. This statistical procedure was also used to compare between the CCA test results considered as categorical outcome, and different infection intensity categories of S. mansoni (i.e., light, moderate, and heavy) utilized as categorical explanatory variables.
A logistic regression was performed to assess the association between CCA-A and CCA-B test results, expressed as binary outcome variable (negative/positive) with S. haematobium egg count as continuous explanatory variable and mircohematuria as categorical explanatory variable among children without a S. mansoni infection.
Non-overlapping 95% confidence intervals (CI) or p-values≤0.05 were considered as statistical significance.
Figure 1 shows the adherence of school children to provide multiple stool and urine samples for a suite of diagnostic tests for detection of S. mansoni and S. haematobium infection. Overall, 674 school children aged 8–12 years were enrolled with slightly more boys than girls (343 vs. 331). The number of children in settings A, B and C was 234, 220 and 220, respectively. At least one stool or one urine sample was provided by 223, 178 and 206 children in settings A, B and C, respectively. Overall, 465 children submitted three stool samples, which were subjected to triplicate Kato-Katz thick smears. Results from a single ether-concentration method were available for 555 children. Three CCA-A test results were available for 489 children, whereas 545 children had the first urine sample additionally subjected to a CCA-B test. Finally, three urine filtrations for S. haematobium diagnosis and three Hemastix dipstick tests for appraisal of microhematuria were done for 489 children.
Results on three stool samples (examined with nine Kato-Katz thick smears and a single ether-concentration) and three urine samples (examined with three CCA-A, one CCA-B, three urine filtrations and three Hemastix dipsticks) were available from a total of 446 children. Among them 48.7% (n = 217) were boys and the median age of the cohort was 10 years. All further analysis focused on this cohort of children.
Table 1 shows the number of children examined and those positive for S. mansoni and S. haematobium, as assessed by different diagnostic approaches, stratified by study setting.
As indicated in Table 2, our ordinal logistic regression analysis showed that for an increase of S. mansoni infection intensity by 1 EPG, the likelihood of a stronger color reaction of the CCA-A (odds ratio (OR) = 1.07) and the CCA-B (OR = 1.03) is significant (both p<0.001). When S. mansoni FECs were not considered as continuous, but stratified according to pre-set thresholds into no, light, moderate and heavy infection intensity, we found that for each increase in infection intensity category, the likelihood of a stronger color reaction of both CCA-A (OR = 36.5) and CCA-B (OR = 25.2) is highly significant (both p<0.001). Figure 2 shows the correlation between infection intensity classes according to pre-set thresholds [22] and the percentage of infected individuals as determined by a single or triplicate CCA-A and a single CCA-B.
Table 3 shows that, if only S. mansoni-negative children were included in a logistic regression analysis and adjustments were made for S. haematobium egg counts and infection intensity classes, no significant association between the CCA-A positivity rate and S. haematobium egg counts was found (OR = 1.09; p = 0.121). There was also no significant association between the CCA-A positivity rate and microhematuria classes detected (p>0.05). Due to the small number of children found positive with the CCA-B test no logistic regression analysis was performed.
For the rapid identification of populations at highest risk of schistosomiasis and other helminth infections that warrant preventive chemotherapy, as well as for monitoring progress of control interventions and new efforts toward elimination, assessment of drug efficacy, and improved patient management, the importance of an accurate diagnosis at the individual and population level must be emphasized [6], [7], [31]. The widely used Kato-Katz technique for the diagnosis of S. mansoni (and S. japonicum) has several shortcomings: in low endemicity settings this technique considerably underestimates the ‘true’ prevalence of infection [8], [11], [32]–[34]. Moreover, a minimum of equipment and well trained laboratory technicians are needed for quality results. Promising results have been reported with a CCA urine test for the diagnosis of S. mansoni in different settings [17], [35]. Some of the previous investigations, however, lacked a rigorous diagnostic ‘gold’ standard, as CCA test results were compared with singe or duplicate Kato-Katz thick smears from one or two stool samples [21], [36].
Within the frame of a SCORE-funded multi-country study, we have now assessed the accuracy of a commercially available CCA urine cassette assay (CCA-A, batch 32727) and an experimental formulation (CCA-B, batch 32686) provided by the same manufacturer and tuned to have a higher specificity, which was run in parallel with the commercially available test in three epidemiological settings of south Côte d'Ivoire. Results of the CCA tests were compared with nine Kato-Katz thick smears (three stool samples, each subjected to triplicate Kato-Katz thick smears). Additionally, we performed a single ether-concentration test using SAF-fixed stool samples. The influence of S. haematobium infection and presence of microhematuria on the performance of the CCA test was determined. In all three settings, a single CCA-A showed a similarly high sensitivity than triplicate Kato-Katz thick smears from a single stool sample, but both approaches missed a considerable number of infections when considering nine Kato-Katz thick smears as ‘gold’ standard. As expected, CCA-B showed a higher specificity than CCA-A, but the sensitivity of CCA-B was considerably lower than that of CCA-A. Indeed, a single CCA-B showed a significantly lower sensitivity than a single CCA-A, and triplicate Kato-Katz thick smears, particularly in settings A and B where the endemicity of S. mansoni was lower than in setting C. We were surprised by the low sensitivity of the ether-concentration method for S. mansoni diagnosis, which warrants follow-up investigations.
The CCA-A seems to be an appropriate test for the diagnosis of S. mansoni in our study area in south Côte d'Ivoire where the prevalence of S. mansoni is above 25% and no recent control efforts have been implemented. Importantly, the co-endemicity of S. haematobium did not influence the accuracy of the CCA-A for the diagnosis of S. mansoni. Additionally, a concurrent infection with soil-transmitted helminths showed no negative influence on the accuracy of the CCA urine test for S. mansoni diagnosis, confirming recent observations made by Shane and colleagues in a study done in Kenya [17]. Furthermore, our study did not reveal a significant association between CCA-A positive results and microhematuria, as determined by Hemastix dipsticks, which relaxes the manufacturer's indication that false-positive results can occur if an individual presents microhematuria. However, further studies in different settings are warranted to confirm that microhematuria or urinary tract infections are not negatively impacting on CCA test results. Also the ability of the CCA test to detect antigen of juvenile Schistosoma worms, which are not yet producing eggs, needs further investigation. Noteworthy, the sensitivity of 56.3% of a single CCA-A in the setting A with a S. mansoni prevalence of 32.9% (based on nine Kato-Katz thick smears) is considerably lower than the sensitivity of 96.3% detected with a single CCA cassette of the same manufacturer in a Kenyan setting with a similar prevalence (38.8%) [17]. This difference might be explained by our more rigorous diagnostic approach, i.e., triplicate instead of duplicate Kato-Katz thick smears of three consecutive stool samples as ‘gold’ standard and by working in a slightly lower endemicity area. The sensitivity of a single CCA-A for S. mansoni diagnosis increased from 56.3% (setting A) to 69.6% (setting B) and 89.6% (setting C) in parallel to increasing prevalence (32.9% to 53.1% and finally to 91.8%), and corresponding mean FECs (17.4 EPG to 62.4 EPG and finally to 482.8 EPG). These findings emphasize the impact of higher prevalences and infection intensities on the positivity rate of the CCA. The strong association between the intensity of the color reaction of the CCA-A band and S. mansoni infection intensities according to FECs by the Kato-Katz method in our studies is in line with previous reports of the CCA dipstick and cassette [37], [38]. The results of the experimental CCA-B formulation, which has been tested on a single urine sample from all children, are suboptimal. Indeed, only low sensitivities and a poor agreement with results of the Kato-Katz method were found, particularly in the lower endemicity areas (settings A and B). In our hands, despite high specificity, the CCA-B in its current formulation cannot be recommended for S. mansoni diagnosis in south Côte d'Ivoire.
The following issues speak for or against the application of the CCA-A versus the Kato-Katz method in helminth control programs or public health centers: at first view, in moderate-to-high-risk communities for S. mansoni infections as found in our study in Côte d'Ivoire (i.e., prevalence above 25%), the collection of a single stool sample and its examination with triplicate Kato-Katz thick smears seems to be an acceptable approach for S. mansoni diagnosis. The advantage of the Kato-Katz method is that it can concurrently detect other helminth species, such as the three main soil-transmitted helminths (i.e., A. lumbricoides, hookworm, and T. trichiura), which is not possible with the CCA. However, the Kato-Katz method requires a minimum of equipment, including a microscope, and well trained laboratory technicians who can identify helminth species-specific eggs in the thick smears. For application of the CCA-A, no additional equipment and only a minimum of training are needed. However, it only detects S. mansoni and no concurrent soil-transmitted helminth infections. The cost of a single cassette (approximately US$ 2) is currently still out of reach of people at highest risk of intestinal schistosomiasis (i.e., poor rural dwellers in sub-Saharan Africa) [17]. However, the cost of triplicate Kato-Katz thick smears are likely higher than a single CCA test [10]. From a convenience and logistical point of view, the collection of urine samples for the CCA is more straightforward than collection of stool for the Kato-Katz method. Indeed, urine production is more convenient for the patient and can be done without special efforts on the spot and at the same day resulting in high compliance rates, while stool production is inconvenient and collection can render a second consultation necessary and thus further exacerbate costs [8], [39].
The performance and sensitivity of the CCA test in low-risk communities (prevalence below 10%), identified by the application of multiple Kato-Katz thick smears on stool samples collected over multiple days, remains to be elucidated. Noteworthy, our study intended to test the CCA in a setting with a S. mansoni prevalence of 10–24% as requested by SCORE. However, we observed a considerable increase in the prevalence of S. mansoni when not only applying triplicate Kato-Katz from a single stool sample as in the pre-screening, but nine Kato-Katz thick smears overall from three stool samples: the observed prevalence increased from 17% to 34% in setting A, and from 36% to 54% in setting B. Due to this rigorous diagnostic approach we ended up with higher prevalences. Retrospectively, this had to be expected, as predicted by mathematical modeling and field observations [11], [13]. If the CCA test proves to be more sensitive than multiple Kato-Katz thick smears in settings characterized by low prevalence and intensity of S. mansoni infection intensities, it will be a most useful test. For example, in areas where intense helminth control efforts have diminished the prevalence and intensity of S. mansoni infections and control programs are focusing elimination, population screenings are necessary to identify remaining S. mansoni hot-spots for targeted anthelmintic treatment and other interventions. For these large-scale screenings the CCA-A would be an excellent tool due to its fast and easy application.
We conclude that in the current study area of south Côte d'Ivoire, where the prevalence and intensity of S. mansoni are still high, partially explained by the prior lack of control efforts, the CCA-A can become a useful method for S. mansoni diagnosis in health centers at the periphery and schistosomiasis control programs. On the other hand, while the specificity of the CCA-B test was high, its current formulation cannot be recommended for S. mansoni diagnosis. Clearly, there is a need to evaluate the CCA test in settings characterized by low S. mansoni prevalences and infection intensities to assess its potential role in schistosomiasis control programs progressing toward transmission control and local elimination and for reliable individual diagnosis.
|
10.1371/journal.pgen.1005480 | The Evolutionarily Conserved LIM Homeodomain Protein LIM-4/LHX6 Specifies the Terminal Identity of a Cholinergic and Peptidergic C. elegans Sensory/Inter/Motor Neuron-Type | The expression of specific transcription factors determines the differentiated features of postmitotic neurons. However, the mechanism by which specific molecules determine neuronal cell fate and the extent to which the functions of transcription factors are conserved in evolution are not fully understood. In C. elegans, the cholinergic and peptidergic SMB sensory/inter/motor neurons innervate muscle quadrants in the head and control the amplitude of sinusoidal movement. Here we show that the LIM homeobox protein LIM-4 determines neuronal characteristics of the SMB neurons. In lim-4 mutant animals, expression of terminal differentiation genes, such as the cholinergic gene battery and the flp-12 neuropeptide gene, is completely abolished and thus the function of the SMB neurons is compromised. LIM-4 activity promotes SMB identity by directly regulating the expression of the SMB marker genes via a distinct cis-regulatory motif. Two human LIM-4 orthologs, LHX6 and LHX8, functionally substitute for LIM-4 in C. elegans. Furthermore, C. elegans LIM-4 or human LHX6 can induce cholinergic and peptidergic characteristics in the human neuronal cell lines. Our results indicate that the evolutionarily conserved LIM-4/LHX6 homeodomain proteins function in generation of precise neuronal subtypes.
| The correct generation and maintenance of the nervous system is critical for the animal’s life. Dysregulation of these processes leads to multiple neurodevelopmental disorders. It has been a daunting challenge not only to identify the developmental mechanisms that determine neuronal cell fate, but also to understand the extent to which the mechanisms are evolutionarily conserved. Here, we describe a developmental mechanism that determines the fate of a specific cholinergic and peptidergic neuronal type in C. elegans. We show that the lim-4 LIM homeodomain transcription factor is necessary and sufficient to promote and maintain the specific cholinergic and peptidergic properties and functions via binding to unique DNA sequences. We also demonstrate that C. elegans lim-4 and human LHX6 show striking functional similarity; specifically, C. elegans LIM-4 or human LHX6 can induce cholinergic and peptidergic characteristics in human neuronal cell lines. Given the high conservation of these transcription factors, these developmental mechanisms are likely to be generally applicable in the nervous system of other organisms as well.
| The proper generation and maintenance of cells in the nervous system is essential for multi-cellular organisms. Each neuron achieves its identity by the acquisition of many distinct features, including appropriate synaptic contacts and expression of distinct sets of neurotransmitters. Fate determination and specification of neuronal cells largely relies on interactions between trans-acting transcription factors and cis-regulatory elements of their target genes [1, 2]. The same transcription factors may be used again after neuronal cell fate determination to maintain the neuron’s integrity [3]. However, it has been challenging not only to discover the transcription factors, but also to identify their regulatory mechanisms and target genes critically associated with determination and maintenance of neuronal cell fate.
In the nematode Caenorhabditis elegans, about 40% of nervous system (~120 neurons) appear to be cholinergic, including a subset of motor neurons in the ventral nerve cord and several sensory, motor and interneurons in the head, and many of these cholinergic neurons co-express neuropeptides [4, 5]. Several genes that function in terminal differentiation and specification of cholinergic neuronal fate have been identified, including the Olf/EBF transcription factor unc-3 for the A-, B-, and AS-type ventral nerve cord and SAB motor neurons, LIM homeobox transcription factor ttx-3 (ortholog of mammalian Lhx2/9) for the AIY and AIA interneurons, Paired-like homeobox gene ceh-10 for the AIY interneurons, and POU homeobox gene unc-86 for the IL2 sensory neurons, URA motor neurons and URB interneurons [6, 7, 8, 9, 10]. These transcription factors act as terminal selectors to directly or indirectly regulate expression of most terminal differentiation genes, such as the cholinergic gene battery but not that of pan-neuronal genes, and broadly affect terminal differentiation of each cholinergic neuron types [2]. Although the terminal selector transcription factors that are required for terminal differentiation of half of the cholinergic neurons have been identified, mechanisms and genes that differentiate other morphologically and functionally different cholinergic neuron types remain to be elucidated.
The SMB multimodal sensory/inter/motor neurons consist of two pairs of neurons that are located in the head and innervate the head neck muscles (Fig 1A). Their processes, which run in ventral or dorsal sublateral cords to the tail and have electric and chemical synaptic contacts to other neurons in the head, were proposed to sense the stretch of body and regulate head locomotion [11]. In fact, laser ablation of the SMB neurons caused increased reversal frequency and wave amplitude of forward locomotion [12]. These neurons utilize at least two neurotransmitters, acetylcholine and a FMRFamide-related peptide, FLP-12 [5, 13]. Genes or molecules that are pivotal for the generation or differentiation of these SMB neurons have not been identified.
Here, we show that the LIM homeodomain LIM-4 protein is necessary to drive expression of terminal differentiation genes, including the cholinergic gene battery and the flp-12 neuropeptide gene, but not pan-neuronal genes in the SMB neurons; consequently, in lim-4 mutants, the neuronal function of the SMB neurons is abolished. We find that LIM-4 maintains its own expression by autoregulation in the SMB neurons and ectopic expression of LIM-4 is sufficient to drive expression of the SMB marker in other cell types. Moreover, our promoter analyses and bioinformatic searches with the SMB marker genes identified a cis-regulatory motif that is necessary and sufficient to drive gene expression in the SMB neurons. We also show that two lim-4 human orthologs, LHX6 and LHX8, functionally substitute for lim-4 in C. elegans. Furthermore, expression of C. elegans LIM-4 or human LHX6 in the human neuroblastoma cell line induces cholinergic and peptidergic characteristics. We propose that there is an evolutionarily conserved role of lim-4/LHX6/LHX8 LIM homeobox genes as terminal selectors to differentiate cholinergic and peptidergic neuronal cells and provide insight into how neuronal characteristics such as neurotransmitter identity are acquired via trans-acting and cis-regulatory mechanisms.
To identify factors that specify the neuronal cell-fate of SMB, we performed a genetic screen to isolate animals in which the expression pattern of a terminal differentiation marker, flp-12p::gfp reporter, was disrupted exclusively in the SMB neurons. flp-12 encodes a FMRFamide-related neuropeptide and is expressed in a set of neurons that includes the SMB and SDQ neurons in adults [13]. Among mutants isolated from this screen, seven mutant alleles (named as lsk1,2,4,5,6,7, yn19) exhibited complete loss of flp-12 expression, while one mutant allele (lsk3) showed weak expression of flp-12 in all four SMB neurons at either adult (Fig 1B; S1 Fig; Table 1) or L1 larval developmental stage (Fig 1B; Table 1; S1 Table) animals. By contrast, expression of flp-12 in the SDQ and other neurons weakly expressing flp-12 was unaffected in all eight mutants (Fig 1B; S1 Fig), indicating that expression of flp-12 was specifically affected in the SMB neurons of these mutants.
From subsequent complementation test and three factor analysis, all mutations were found to be allelic to the previously identified lim-4(ky403) mutation. lim-4 encodes a LIM homeodomain protein that is required for specification of AWB and ADF chemosensory neuron identity [14, 15,16]. In lim-4(ky403) null mutants, AWB cell fate is changed to that of the AWC chemosensory neurons, thereby causing dye-filling defects in the AWB neurons [14]. Expression of flp-12 was also completely abolished in the SMB neurons of ky403 mutants (Fig 1B; Table 1). Like in the ky403 mutants, the AWB neurons failed to dye-fill in the lsk1-7 and yn19 mutants (S2 Fig; S2 Table). The molecular lesions of all eight mutants mapped to the coding region of the lim-4 gene (Fig 1C). Five mutant alleles (lsk1,4,5,6,7) had nonsense mutations that resulted in premature translation stop, suggesting that these mutations are null alleles. lsk2 had a mutation in the splice donor site after the 1st exon. yn19 and lsk3 had missense mutations within the coding region of the second LIM domain, resulting in C199Y and E207K substitutions, respectively. The cysteine residue (C199) is critical for forming a zinc finger motif in the LIM domain [15]. The glutamate residue (E207) resides in the LIM domain and is highly conserved through evolution (Fig 1C), suggesting that this residue is essential for LIM-4 function via protein-protein interactions. These findings indicate that LIM-4 has a role in regulating gene expression in the SMB neurons.
To determine the extent to which LIM-4 regulates gene expression in the SMB neurons, we examined additional SMB terminal differentiation genes, including odr-2 GPI-anchored cell surface protein [17], trp-1 TRPC channel [18], and cholinergic markers such as unc-17 vesicular acetylcholine transporter (VAChT) [19] and cho-1 choline transporter (ChT) (Fig 2A) [20]. In order to locate the SMB cell bodies, we used expression of ceh-17p::dsRed in the cell bodies of SIAV as a marker that is directly adjacent to the cell bodies of SMBD (S3 Fig) [21]. None of the SMB specific or cholinergic markers were expressed in the SMB neurons of lim-4 mutants while expression in other neuron types was generally not affected (Fig 2B and 2C; Table 1). We next tested expression of two well-characterized pan-neuronal gene markers, rgef-1 Ras guanine nucleotide releasing protein and unc-119 chaperone [6, 7]. Expression of these pan-neuronal genes was not altered in the SMB neurons of lim-4 mutants (Fig 2D; Table 1), indicating that the SMB cells may retain neuronal properties.
To determine whether the lim-4 SMB neurons adopted a different cell fate such as the structurally and/or functionally related sub-lateral nerve cord neurons including the SIA, SIB, and SMD neurons, we tested markers including ceh-17 for SIA [21], flp-22 for SMD [13] or ceh-24 for SIA, SIB, and SMD [22, 23]. None of these markers were ectopically expressed in lim-4 mutants (S4 Fig), suggesting that the cell fate of the SMB neurons is not transformed to that of the structurally and/or functionally related cell types.
The SMB neurons are generated from ABalpapap (SMBDL, SMBVL) or ABarappap (SMBDR, SMBVR) precursors, and three of their sister cells undergo programmed cell death before hatching (S5 Fig) [24]. We observed expression of pan-neuronal markers in SMB of adult lim-4 mutants (Fig 2D), suggesting that the SMB cells do not adopt the apoptotic fate of their sister cells. Based on these results, we conclude that LIM-4 activity does not initiate neuronal cell fate, but specifies SMB cell fate by regulating expression of terminal differentiation genes, thereby acting as a terminal selector transcription factor in the SMB neurons.
Wild-type animals move in sinusoidal waves of a consistent wave width and wavelength (Fig 3A) [25]. lim-4 mutants move in a coiled or loopy fashion (Fig 3A) [14]. To quantitate the loopy uncoordinated movement, the waveforms of these animals were measured by viewing tracks made in a bacterial lawn and compared to that of wild-type animals (Fig 3B). lim-4 mutants had significantly accentuated waveforms (Fig 3C). While the average wavelength for lim-4 null mutants (ky403 or lsk5) is similar or mildly decreased compared to that of wild-type animals, the average wave width for ky403 or lsk5 mutants (ky403: 359.46±13.51 μm, n = 30; lsk5: 374.27±16.47 μm, n = 30) is about 70% higher than that of wild-type animals (N2: 194.54±4.28 μm, n = 30) (Fig 3A and 3C). yn19 and lsk3 missense mutants similarly exhibited significantly larger wave width (yn19: 312.31±7.75 μm, n = 30; lsk3: 365.43±11.97 μm, n = 30) (Fig 3A and 3C), suggesting that yn19 and lsk3 mutations also fully eliminate the contribution of LIM-4 to locomotion.
To assess whether the loopy movement of lim-4 mutants is due to a functional defect of the SMB neurons, we ablated the SMB neurons by laser microsurgery. Consistent with a previous study [12], killing the SMB neurons resulted in a loopy or coiled movement phenotype; the average wave width was increased by over 50% compared to control animals and similar to that of lim-4 mutants (Fig 3D). We, however, noted that the SMB ablation did not result in as strong a loopy phenotype as the lim-4 mutations, suggesting that the mutations have additional effects on locomotion beyond elimination of SMB function. lim-4 is also expressed in the SAA neurons (see below) of which roles have been implicated in head locomotion [11]. Laser ablation of the SAA neurons did not cause a loopy or coiled movement, ruling out the possibility that defects of the SAA neurons result in movement defects of lim-4 mutants (Fig 3D). These results indicate that the SMB neurons function to regulate locomotion by modulating the wave width of the animal and that the loopy phenotype of lim-4 mutants is due to defects in the function of the SMB neurons.
lim-4 has previously been shown to be expressed in several neuronal types in the head of postembryonic animals; these neurons include the AWB, SIA, SAA, RID, RIV, and RMD neurons, but not the SMB neurons [14]. Like the SMB neurons, the SIA neurons project their processes into the sub-lateral nerve cords and their cell morphology and position are similar to those of the SMB neurons [11]. To determine whether lim-4 expression was mis-identified in the SIA neurons, we examined the expression pattern of lim-4p::gfp transgene (oyIs35) that includes 3.6 kb of upstream sequence [14, 26] and compared it to that of ceh-17p::dsRed, a SIA marker [21] or flp-12p::mCherry, a SMB marker [13], respectively (Fig 4A). We observed co-localization of lim-4 expression with that of flp-12 but not of ceh-17, indicating that lim-4 is expressed in the SMB neurons rather than the SIA neurons. In support of this re-assignment, the expression of other SMB markers such as odr-2p::gfp or trp-1p::gfp was completely abolished in the SMB neurons of lim-4 mutants, whereas expression of ceh-17 was not affected (Fig 2B; S4 Fig).
Expression of lim-4 was previously shown to be autoregulated in the AWB neurons but not in the other LIM-4-expressing neurons [14]. We confirmed that lim-4 expression in the AWB neurons was not seen in lim-4 mutants at L1 larval stage, whereas the expression of lim-4 in the other neurons including the SMB neurons, was detected (Fig 4B). However, lim-4 expression in the SMB neurons gradually decreased from the L1 larval stage until it became undetectable in the adult stage (Fig 4B; Table 1). Hence, lim-4 appears to be required to maintain its own expression in the SMB neurons but does not initiate its expression, further supporting its role as a terminal selector gene in the SMB neurons.
To determine whether lim-4 acts cell-autonomously within SMB, we tried to rescue lim-4 phenotypes by expressing a wild-type lim-4 cDNA driven under the control of lim-4pΔ3 promoter. The lim-4pΔ3 promoter includes minimal upstream regulatory sequences that drive transgene expression exclusively in the SMB neurons but not as strongly as the full promoter of lim-4 and more dominantly in the SMBD than SMBV neurons (see below), and was used to identify expression in SMB of genes tested in this study (S6 Fig). The gene expression and locomotion defects of lim-4 mutants were partially restored, while the dye-filling defects were still present and the normal average wavelength was not altered, indicating that LIM-4 acts in the SMB neurons to affect locomotion and transmitter specification (Fig 4C and 4D; S3 Table). Taken together, lim-4 is expressed and acts in the SMB neurons to specify the SMB cell-fate.
To determine when the activity of LIM-4 is required for the expression of the SMB markers and proper locomotive movement, we first expressed lim-4 with an inducible, ubiquitously expressed heat-shock promoter (hsp16.2) [27]. Upon transient supply of lim-4 gene activity at the fourth larval stage (i.e., long after the SMB neurons have differentiated in the embryo), expression of flp-12 was fully restored and the loopy phenotype of lim-4 mutants was rescued (Fig 4C–4F). These results demonstrate that post-developmental expression of LIM-4 is sufficient to restore the expression of the SMB markers and the function of the SMB neurons in lim-4 mutants. These data further indicate that the SMB neurons are not irreversibly switched to another cell-fate and demonstrate that loss of lim-4 does not result in irreversible developmental defects. The dye-filling defects of the AWB/ADF neurons in lim-4 mutants were partially rescued after multiple heat shocks (S3 Table) [14, 16].
We also used the inducible rescue assay to corroborate the prediction that lim-4 is continuously required to maintain the functional properties of the SMB neurons. To this end, we supplied lim-4 activity via the heat-shock promoter at L4 stage and then analyzed the animals after 14 hours at the young adult stage. In these animals, we found the locomotory defects to be partially rescued (Fig 4G). When assayed after a long time interval at 3 and 6 days (i.e., older adult stages), the animals again displayed a mutant phenotype indistinguishable from the control (Fig 4G), suggesting that the transient rescuing ability of the lim-4 gene activity has faded. These results demonstrate that lim-4 does not only initiate but also maintains the expression of the SMB terminal differentiation genes and, hence, the function of the SMB neurons.
To address whether expression of lim-4 is sufficient to induce the SMB identity in other cell-types, we first examined ectopic flp-12 expression upon transient supply of lim-4 gene activity via the heat-shock promoter at the embryonic stage. Although the heat-shock promoter should drive ubiquitous LIM-4 expression, ectopic flp-12 expression was not seen broadly elsewhere; interestingly, expression was limited in only one cell-type, the ALN neurons, in 25% of transgenic animals (n = 50) (Fig 5A). The ALN neurons are a pair of cholinergic oxygen-sensing neurons in the tail [5, 11, 28] that do not appear functionally or linearly related to the SMB neurons (S5 Fig).
We next attempted to express LIM-4 in the glutamatergic chemosensory neurons AWC using the promoter of the ceh-36 homeobox gene [26, 29]. Ectopic expression of unc-17 VAChT was detected in AWC (Fig 5B) while the flp-12 was not ectopically expressed in AWC (S7 Fig). We further induced broader ectopic expression of LIM-4 in the subset of glutamatergic neurons under a specific eat-4 glutamate transporter gene promoter that drives reporter expression in 11 (but not in AWC) out of 38 glutamatergic neuron classes in the hermaphrodites [30]. Expression of LIM-4 in these cells did not drive ectopic expression of cho-1 ChT or affect expression of eat-4 (S8 Fig), suggesting that expression of LIM-4 alone is not sufficient to generally induce cholinergic cell fate in a subset of glutamatergic neurons. These results suggest that lim-4 is partially sufficient to drive expression of the SMB markers in a context-dependent manner.
Homeodomain transcription factors generally bind well-defined DNA sequences to control transcription of target genes [31]. Systemic analysis of homeodomain DNA-binding specificities allowed prediction of the recognition motif of each homeodomain protein, and a cis-regulatory motif containing the consensus TAAT core DNA sequences was predicted to be the binding site for the homeodomain of LIM-4 and its mammalian and Drosophila homologs (LHX6/8 and Arrowhead, respectively) (Fig 6A; S9 Fig) [32, 33]. Indeed, LHX6 and LHX8 have been shown to directly bind to the predicted DNA sequences (ATAATCA) in the promoter regions of the Shh gene [34].
To identify cis-regulatory motifs required to drive expression of the SMB marker genes in the SMB neurons, DNA sequences within the promoters of the SMB markers were serially deleted and the resultant transgenic animals were examined for altered expression patterns. From these analyses, we first determined a minimal region within the flp-12 promoter for flp-12 expression. Deletion of a 150 bp sequence located ~162 bp upstream of the translation start sequence caused decreased gfp expression in the SMB neurons (Fig 6B; S10 Fig). Within the 150 bp region, we next found four AT- rich DNA sequences that are fully conserved in the promoters of the flp-12 orthologs in the related Caenorhabditis species (Fig 6B). Mutations of two AT-rich DNA sequences resulted in an almost complete loss of gfp expression, while mutations of the other two sequences did not affect the gfp expression (Fig 6B), indicating that the former two motifs are necessary for the expression of flp-12 in the SMB neurons. DNA sequences of these motifs (AAAATTG and ACAATAG) share limited sequence conservation with putative LIM-4 binding sequences, and will be referred to as SMB motifs (Fig 6A). To test whether these SMB motifs are sufficient to drive gene expression in the SMB neurons, we inserted three copies of the SMB motifs in the promoter of flp-7, which is normally expressed in the several head neurons, but not in the SMB neurons (Fig 6C) [13]. Transgenic animals expressing a flp-7p-SMB motif::gfp reporter construct still exhibited gfp expression in flp-7 expressing neurons, indicating that insertion of the SMB motifs within the regulatory region of flp-7 does not alter the flp-7 expression pattern. In addition, we observed consistent expression of flp-7 in the SMB neurons in 100% transgenic animals (n = 50) (Fig 6C), suggesting that these SMB motifs are necessary and sufficient to drive gene expression in the SMB neurons. These results are consistent with the hypothesis that LIM-4 directly binds the SMB motifs to regulate expression of flp-12 gene in the SMB neurons.
To define additional regulatory motifs for expression of SMB markers, we examined the promoter regions of two additional SMB markers, odr-2 and unc-17, and defined the regions essential for SMB expression. These regions in the odr-2 or unc-17 promoters contained the SMB motifs found in the flp-12 promoter (Fig 6D). We mutated these motifs in the context of the odr-2 and unc-17 reporter genes and found that these mutations reduced expression of the reporter genes (Fig 6D). These results demonstrate that distinct cis-regulatory motifs can determine cell-specific expression or something of this sort.
A cis-regulatory region in the lim-4 promoter that is required for lim-4 expression in the AWB neurons was previously identified [35]. We performed analogous deletion analysis experiments in transgenic animals to dissect the lim-4 promoter to identify motifs required for lim-4 expression in the SMB neurons (S11 Fig). As proof-of-principle, we also identified the lim-4 regulatory sequences for the AWB expression (S11 Fig). However, we could not identify simple cis-regulatory motifs in the lim-4 promoter required for the SMB expression (S11 Fig). Instead, multiple regions in the lim-4 promoter act in concert to regulate LIM-4 expression in the SMB neurons, suggesting a complexity of cis-regulatory motifs in the lim-4 promoter to ensure proper LIM-4 expression in the SMB neurons.
The mammalian genome contains two LIM-4 orthologs, LHX6 and LHX8. In mice, these genes are largely expressed in the developing and adult striatum and orchestrate specification of interneuron identities; specifically, LHX6 and LHX8 are required to determine GABAergic/peptidergic and cholinergic interneuronal cell fate, respectively. In addition, these genes have redundant function to regulate expression of shh in MGE neurons [34]. LIM-4 exhibits a high degree of protein sequence homology to LHX6 and LHX8 (in particular, 60% identical in its homeodomain) (Fig 7A) [14], suggesting a functional conservation of these proteins.
To test for functional homology, we first tried to rescue C. elegans lim-4 mutants by expressing human LHX6 or LHX8 cDNA under the control of the heat shock promoter. Similar to C. elegans lim-4 cDNA, human LHX6 or LHX8 cDNA fully restored altered locomotion of lim-4 mutants (Fig 7B; S12A Fig). Moreover, LHX6 also fully rescued the defect of flp-12 expression in lim-4 mutants while LHX8 did not rescue (Fig 7C; S12B Fig), indicating that LHX6 may have a higher degree of functional conservation to LIM-4 than LHX8.
LHX8 has been shown to be required for the development and maintenance of cholinergic neurons in mouse basal forebrain [36, 37]. Overexpression of LHX8 was sufficient to differentiate rat hippocampal neural stem cells or newborn neurons into cholinergic neuron types [38, 39] and induced expression of cholinergic markers in a human neuroblastoma cell line [40]. Whether LHX6 has a similar role in specification of cholinergic cell fate has not been explored. Thus, we tested whether overexpression of human LHX6 or C. elegans LIM-4 could promote expression of cholinergic markers in human neuroblastoma SH-SY5Ycells. These cells appear to mimic immature catecholaminergic neurons when untreated [41, 42], but can differentiate into various mature neuron-like phenotypes depending on the addition of differentiation-inducing agents [43]. We generated stable SH-SY5Y cell lines expressing either LHX6-GFP, lim-4-GFP, or empty vehicle and asked whether transfected cell lines express cholinergic markers. Expression of either LHX6-GFP or lim-4-GFP was detected predominantly in cell nuclei, supporting the action of LHX-6 and LIM-4 as transcription factors (Fig 7D; S13A Fig). Transfected cells were immunoreactive to VAChT or choline acetyltransferase (ChAT) antibodies and exhibited higher endogenous ChAT message levels, as assayed by quantitative reverse transcription polymerase chain reaction (RT-PCR), compared to that in cells transfected with an empty vehicle (Fig 7D; S13B Fig). Thus, human LHX6 and even C. elegans lim-4 are sufficient to promote expression of cholinergic markers in human cells. We also noted that cells expressing either LHX6-GFP or lim-4-GFP were morphologically different to cells bearing empty vehicle: empty vehicle bearing cells tended to grow in clusters and were round shape; by contrast, LHX6-GFP or lim-4-GFP expressing cells formed less clusters and appeared as spiky neuronal cells (Fig 7D; S13B Fig). These results, therefore, indicate that LHX6 or LIM-4 can induce differentiation of SH-SY5Y cells into cholinergic as well as neuronal phenotypes.
We further examined the morphology of transfected cells by electron microscopy. Untransfected or empty vesicle transfected SH-SY5Y cells exhibited typical shapes of mitochondria, endoplasmic reticulum (ER), and Golgi apparatus in the peri-nuclear region (Fig 7E). In either lim-4 or LHX6 transfected cells, however, we observed additional ultrastructural components, such as 100nm large-dense core vesicles, which may contain neuropeptides, near the Golgi apparatus (Fig 7E) [44]. Furthermore, in the spiky protruded region of lim-4 or LHX6 transfected cells, we also identified mitochondria and small synaptic vesicles that may represent axon terminals of neurons (Fig 7E). These results further support that expression of LIM-4 or LHX6 may produce synaptic vesicles containing large neuropeptides and small molecule neurotransmitters in human cell lines.
In these studies, we have identified an important regulator that controls terminal differentiation of a distinct neuronal cell-type in C. elegans. The SMB neurons appear to have mixed neuronal functions. Their long, unbranched and synapse-free processes along the body may serve as proprioreceptors to sense body stretch. In addition, they contact over 20 sensory, inter, or motor neurons via chemical or electrical synapses and may integrate additional extrinsic or intrinsic cues to regulate head muscle contraction [11]. The SMB neurons co-express a unique combination of neurotransmitters, acetylcholine and FLP-12 FMRFamide-like neuropeptides. Thus, an intriguing question is how the SMB sensory/inter/motor neurons acquire their unique characteristics. Our experiments show that the LIM-4 LIM homeodomain transcription factor is necessary and sufficient to promote and probably maintain SMB-specific properties and functions. In lim-4 null mutants, the neurotransmitter identity and neuronal function of the SMB neurons are completely lost but pan-neuronal features are not affected. Transient LIM-4 expression in lim-4 mutants not only restores the SMB characteristics and functions, but also induces ectopic expression of a SMB-expressed neurotransmitter in another cell-type. Our promoter analysis suggests that LIM-4 directly regulates expression of the SMB terminal differentiation marker via well conserved homeodomain binding sequences and also controls its own expression. Hence, we propose that lim-4 acts as a terminal selector gene to broadly specify the SMB neuronal identity [2, 45].
A few terminal selector genes that determine cholinergic cell-fate in C. elegans have been identified. For example, the Olf/EBF gene unc-3, the heterodimer of LIM homeobox gene ttx-3 and Paired-like homeobox gene ceh-10, and POU homeobox gene unc-86 regulate terminal differentiation of the A-, B-, and AS-type ventral nerve cord or SAB motor neurons, the AIY interneurons, and the IL2 sensory neurons, URA motor neurons and URB interneurons, respectively [6, 7, 8, 9, 10]. ttx-3 also acts as a terminal selector in the AIA interneurons [10]. These genes regulate expression of not only the cholinergic gene battery, including unc-17 (VAChT), but also other terminally differentiated cell-specific markers. Furthermore, these trans-acting factors appear to directly bind to the evolutionarily conserved cis-regulatory elements of most, if not all, their target genes. In the case of the unc-17 promoter region, distinct cis-regulatory target sites, such as the COE motif for UNC-3 and the AIY motif for TTX-3/CHE-10, are systemically organized (S15A Fig). In this study, we identified an additional cis-regulatory element, called the SMB motif, in the unc-17 gene (S15A Fig), suggesting that the elaborate cis-regulatory architecture ensures expression of cell-specific characteristics. Since additional terminal selector genes required for specification of over 50 uncharacterized cholinergic cell-types need to be identified, additional motifs must exist in the cholinergic gene battery such as unc-17 (S15A Fig).
Recent work has shown that distinct combination of 13 different terminal selector genes defines identity of 25 different glutamatergic cell-types in C. elegans, suggesting that the combinatorial codes of terminal selector transcription factors are a general theme for determining cell-type specificity [30]. In fact, in the AIY cholinergic neurons, two terminal selector transcription factors, ttx-3 and ceh-10, form a heterodimer that directly regulates expression of their target genes via a common cis-regulatory bipartite motif; mutations of each gene lead to complete loss of the AIY specific neuronal identity [6, 9]. We have tried to identify the putative binding partner(s) of LIM-4 in the SMB neurons by analyzing the expression pattern of the flp-12 reporter construct in mutants for which genes were previously reported to be expressed in the SMB neurons such as the fax-1 nuclear receptor and cog-1 Nkx6-type homeobox transcription factor [46, 47]. None of these mutations affects flp-12 expression in the SMB neurons (S4 Table). Therefore, it is not yet clear which transcription factors work in combination with lim-4 to control the terminal differentiation of the SMB neurons. Because our results demonstrate that lim-4 expression is initiated at an early developmental stage and then autoregulated afterward, we also tested the possibility that expression of fax-1 or cog-1 in the SMB neurons may regulate lim-4 expression in SMB. In chemosensory neuron types, the lin-11 LIM homeobox, ceh-37 Otx, mls-2 HMX/NKX homeobox and nhr-67 Tailless/TLX genes control expression of terminal selector genes in the AWA, AWB, AWC, and ASE neurons, respectively [26, 29, 48, 49]. However, lim-4 expression is not altered in fax-1 or cog-1 mutants (S4 Table), indicating that these SMB-expressed transcription factors may have more specific roles in development or differentiation of the SMB neurons. We will continue searching for the transcription factors that partner with LIM-4 or act upstream or downstream of LIM-4 to control terminal differentiation of the SMB neurons.
Previous studies show that lim-4 determines the proper cell-type specification of the AWB and serotonergic ADF neuron types [14, 16]. The AWB and ADF neurons are two classes of amphidial chemosensory neurons in the head of worms that detect volatile chemical repellants and putative food signals, respectively [50]. In lim-4 mutants, the AWB neurons lack AWB-specific characteristics and functions, such as expression of putative 7-TM receptor str-1, AWB-specific cilia and axon morphology, and abilities to take up lipophilic dyes, and instead acquire features and functions of the AWC olfactory neurons, such as expression of the putative 7-TM receptor str-2 and AWC-specific cilia and axon structures, suggesting that lim-4 acts as a cell fate switch between the AWB and AWC neurons [14]. In case of ADF cell-fate specification, lim-4 acts transiently in the precursor cells of ADF and regulates part of the terminal differentiation process; lim-4 mutants lack expression of a set of serotonergic markers, including tryptophan hydroxylase tph-1, but do not affect expression of a putative 7-TM receptor srh-142 [16]. The ceh-37 Otx gene is required for expression of lim-4 in AWB and srh-142 in ADF, suggesting that ceh-37 also differentially affects the AWB and ADF cell fates [26]. In addition, lim-4 has a role in regulating axon morphology of the SAA neurons but not expression of terminal differentiation markers [14]. We propose that the SMB neurons in lim-4 mutants do not adopt a functionally or lineage-related cell fate, but remained undifferentiated because they lose expression of most, if not all, terminal differentiation genes. Thus, lim-4 plays distinct roles in neuronal development in a context-dependent manner and acts as a bona fide terminal selector for the differentiation of SMB (S15B Fig).
The LIM homeobox gene family has a high degree of structural conservation through evolution amongst the homeobox gene superfamily [51–53]. However, their functional conservation among distantly related species is relatively unexplored. We demonstrate that C. elegans lim-4 and human LHX6 and LHX8 (also referred as L3 or LHX7) show striking functional similarity; LHX6 completely rescues locomotive defects and flp-12 expression phenotypes of lim-4 mutants, whereas human LHX8 restores only locomotion but not flp-12 expression in lim-4 mutants. Furthermore, expression of either LHX6 or lim-4 is sufficient to drive cholinergic differentiation in human neuroblastoma cells. The role of LHX8 in cholinergic cell-fate determination in the mammalian nervous system has been well characterized [36, 54, 55]. Deletion of the murine LHX8 (LHX7), causes a subtype of cholinergic interneurons to convert into another subtype of GABAergic interneurons [55]. However, the study of LHX6 function has focused on GABAergic fate specification [56, 57]. We have uncovered that LHX6 also plays a role in cholinergic cell fate determination in C. elegans or human neuroblastoma cells and acts as a terminal selector to control the differentiation of neuronal subtypes. Cholinergic neurons in the mammalian forebrain have crucial roles in locomotive and cognitive functions and thus, understanding and manipulation of cholinergic cell fate specification may be beneficial to identify therapeutic targets and methods for neurodiseases resulting from cholinergic neuronal dysfunction.
N2 Bristol strain was used as wild-type strain. Mutant strains and transgenic strains used in this study are listed in S5 Table. All strains were maintained at 20°C.
The flp-12p::gfp(ynIs25) integrated strain was used to performed EMS mutagenesis according to Sulston and Hodgkin (1988). Eight alleles (yn19, lsk1, lsk2, lsk3, lsk4, lsk5, lsk6, lsk7) in which GFP expression was completely abolished, were isolated from screening ~15,000 haploid genomes, found to be allelic to each other, and mapped on LG X. Based on three factor crosses using the double mutants unc-6 dpy-6, dpy-8 unc-6, and unc-2 dpy-8, yn19 was located approximately 2.4 MU downstream of unc-2 and 4.3 MU upstream of dpy-8. In this region, we did complementation tests with a lim-4(ky403) mutant and found that they were allelic. The molecular lesions were identified by sequencing amplification products of lim-4. All lim-4 alleles were outcrossed with N2 at least five times before phenotypic analysis. To observe locomotion phenotypes of lim-4 mutants (ky403, yn19, lsk3, lsk5), the integrated flp-12p::gfp array was removed by mating with N2 males.
For promoter analysis, promoter regions of odr-2 and unc-17 were amplified by PCR from N2 genomic DNA and were inserted into the pPD95.77 vector [27]. lim-4p::gfp [14] was gifted from Piali Sengupta. Promoter regions of each reporter construct were deleted with various digestion enzymes or PCR fusions. Mutagenesis was performed using QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer’s protocol.
For the constructs used to test for rescue, the hsp16.2 promoter was fused with the following cDNAs: lim-4 cDNA (kind gift form Oliver Hobert), human LHX6 cDNA (BC103937) and human LHX8 cDNA (BC040321) (Thermo Fisher Scientific). The lim-4p∆3 or odr-1 promoter were used to generate the lim-4p∆3::lim-4cDNA or odr-1p::lim-4cDNA constructs, respectively.
To generate the flp-7p-SMBmotif::gfp construct, three copies of SMB motif oligomers that have SacI enzyme site at 5’ and 3’ ends were synthesized and inserted into SacI site at -353 bp region of the flp-7 promoter.
otIs518; otIs534 (cho-1fosmid::yfp; eat-4fosmid::mChOpti) and the eat-4p∆5::gfp construct were kind gifts from Oliver Hobert. To express lim-4 cDNA in the subset of glutamatergic neurons, lim-4 cDNA was replaced with gfp to generate eat-4p∆5::lim-4cDNA. Then, 2.5 ng of eat-4p∆5::lim-4cDNA was injected into the otIs518; otIs534 strain with 50 ng rol-6 as an injection marker.
To express lim-4 cDNA in the AWC neurons during development stage, ceh-36p was fused with lim-4 cDNA and 5 ng was injected into vsIs48 (unc-17p::gfp) and ynIs82 (flp-12p::gfp) with 50ng of odr-1p::dsRed as an injection marker. As the control, 50 ng odr-1p::dsRed was injected into vsIs48 (unc-17p::gfp).
ceh-17p::dsRed was kindly gifted from Satoshi Suo.
Heat shocks were administered to fourth larva stage (L4) of the transgenic worms at 33°C twice for 30 minutes with an hour incubation at 20°C between heat shocks for recovery modified from [7]. After heat shocks, worms were incubated at 20°C for 14 hours to reach the young adult stage when lim-4 phenotypes were assayed. To observe ectopic expression of flp-12p::gfp in other cell types, heat shocks were administered at the embryo 2- or 3-fold stage two times at 37°C for 30 minutes with an hour incubation at 20°C between heat shocks. Heat shocks were administered at L4 animals to observe locomotion at days 1, 3, and 6. The L4 stage was counted as day 0.
The level of flp-12p::gfp expression in the SMB neurons was quantified as strong, weak, off. Strong was determined as robust expression in the SMB neuronal cell bodies and processes. Weak was defined as faint expression in cell bodies and no expression in the processes. Off was defined as no flp-12p::gfp expression in either cell bodies and processes. To measure wave width and wavelength of the worm tracks, Leica microscope software (Leica Application Suite Advanced Fluorescence Lite 3.5. Ink) was used. Wavelength was defined as distance between one peak and the next corresponding peak, and wave width was distance from the peak to the trough of the sine wave. The average of six consecutive wave width and wavelength from each worm track was quantified as the individual data.
Laser ablation experiments were performed as previously described [58]. L1 larvae of the integrated lim-4p::gfp (oyIs35) strain were anesthetized with 10 mM sodium azide and all four SMB or SAA neurons were killed by a nitrogen dye-pulsed laser (Photonic Instruments, St. Charles, IL). Animals were recovered for 3 days at 20°C. After performing locomotion assays, GFP expression of lim-4p::gfp reporter was observed to confirm laser ablations in the SMB or SAA neurons.
The conservation of the cis-regulatory motif from the promoter analysis of the flp-12, lim-4, odr-2, and unc-17 were examined by using the USCS genome browser (http://genome.ucsc.edu/). DNA sequences from the five different Caenorhabditis species were obtained from the UCSC website, and aligned using the ClustalW2 in EBI (European Bioinformatics Institutes; http://www.ebi.ac.uk/Tools/msa/clustalw2/) to identify cis-regulatory regions including SMB motifs. The position frequency matrix (PFM) of LIM-4, LHX6, and LHX8 predicted binding sites were derived from a web based tool, PreMoTF (http://stormo.wustl.edu/PreMoTF). Predicted conserved motif sequence logo was obtained from the Seq2Logo website (http://www.cbs.dk/biotools/Seq2Logo/).
Fluorescent microscopic images were taken with a Zeiss LSM700 Confocal microscope and were obtained using ZEN 2009 Light Edition software. For light microscopic images of worms, Leica High-performance Fluorescence Stereomicroscopy M205FA was used and Leica Application Suite Advanced Fluorescence Lite 3.5 software was used to measure the phenotype.
The SH-SY5Y human neuroblastoma cell line (ATCC) was cultured with 10% complete medium (1:1 mixture of DMEM and Ham's F12 medium and 10% supplemental fetal bovine serum, 100 U/ml penicillin, and 100 μg/mL streptomycin) in a humidified, 5% CO2-95% air, 37°C incubator.
HEK293T cells were purchased from American Type culture Collection (ATCC) and were cultured in DMEM with 10% FBS. cDNA for lim-4 or LHX6 was cloned into the lentiviral vector, pRetroX-IRES-ZsGreen1 (Clontech). Transformation was performed with lentivirus constructs and packaging vector into 293T cells using a lipofectamine 2000 reagent (Invitrogen). At 72 hours post transfection, the viral particles were harvested by filtration using a 0.45mm syringe filter.
Wells of glass bottom dishes (MatTek) were coated 0.1μM fibronectin (Sigma) for 1 h at 37°C. After removing the fibronectin solution, SH-SY5Y cells were seeded at density of 5×104 cells per well and grown to confluence in 50%. Viruses were added to cells at 37°C for 120 min followed by addition of an equal volume of DMEM and Ham's F12 medium with 10% fetal bovine serum, and incubation for a further 24 h. Cells then were washed with phosphate-buffered saline, and DMEM and Ham's F12 medium with 1% fetal bovine serum. After an additional 24 h in growth medium, cells were washed in 1:1 DMEM and Ham's F12 medium with 1% fetal bovine serum. Under these conditions, up to 90% of cells were infected.
Cells were fixed using 4% paraformaldehyde, followed by 0.1% triton-X permeabilization and incubation with antibodies. Fixed cells were incubated at 4°C overnight with the primary antibodies, including human-cross reactive rabbit anti-ChAT (Millipore) and human-cross reactive rabbit anti-VAChT (Synaptic Systems). Bound antibodies were visualized with Alexa Fluor 594 FluoroNanogold-anti-rabbit–conjugated secondary antibodies (Nanoprobes). Nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI, Sigma).
Total RNA was obtained from the cells by using a Trizol Reagent (Invitrogen) according to the manufacturer’s instructions. cDNAs were synthesized using High-Capacity cDNA reverse transcription kits (Applied Biosystems). Quantitative real-time RT PCR was performed using the SYBR Green PCR master mix kit (Applied Biosystems) on the ABI 7500 Real Time PCR System under the following conditions: Cycling conditions were 2 min at 50°C, 10 min at 95°C, followed by 40 cycles of 95°C for 15s and 60°C for 1 min. Primer sets were designed using primer express 3.0 software based on the human gene sequences from GenBank and are as follows: ChAT (sense: 5'-GGCTCAGAACAGCAGCATCA -3' and antisense: 5'-GAGACGGCGGAAATTAATGACA -3'); GAPDH (sense: 5'-ACCCACTCCTCCACCTTT GA-3' and antisense: 5'-TGTTGCTGTAGCCAAATTCGTT-3'). The housekeeping gene GAPDH was used as an internal standard. Reaction specificity was confirmed by melting curve analysis.
Control and transfected SH-SY5Y cells grown on the MatTek culture dish were fixed for 2 h at 4°C in PBS containing 2.5% glutaraldehyde. After three washes in PBS, the cells were postfixed with 1% osmium tetroxide on ice for 2 h and washed three times again in PBS. The cells were then embedded in Epon 812 mixture and polymerized in an oven at 60°C for 24 hours after dehydration in increasing concentrations of ethanol (50, 70, 80, 90, 95 and 100%) and propylene oxide series (20 min each). The embedded blocks were trimmed and sectioned on an ultramicrotome with a diamond knife and ultrathin sections were collected on Formvar-coated copper grids. The grids were stained with 2.5% uranyl acetate (7 min) and Reynolds lead citrate (2 min), and were viewed with a transmission electron microscope (Technai G2 Spirit Twin, FEI, USA) at 120 kV.
|
10.1371/journal.pntd.0004411 | Disease Risk Perception and Safety Practices: A Survey of Australian Flying Fox Rehabilitators | Interactions with flying foxes pose disease transmission risks to volunteer rehabilitators (carers) who treat injured, ill, and orphaned bats. In particular, Australian bat lyssavirus (ABLV) can be transmitted directly from flying foxes to humans in Australia. Personal protective equipment (PPE) and rabies vaccination can be used to protect against lyssavirus infection. During May and June 2014, active Australian flying fox carers participated in an online survey (SOAR: Survey Of Australian flying fox Rehabilitators) designed to gather demographic data, assess perceptions of disease risk, and explore safety practices. Responses to open-ended questions were analysed thematically. A logistic regression was performed to assess whether rehabilitators’ gender, use of PPE, threat perception, and years of experience predicted variation in their odds of being bitten or scratched. Eligible responses were received from 122 rehabilitators located predominantly on the eastern coast of Australia. Eighty-four percent of respondents were female. Years of experience ranged from <1 to 30 years (median 5 years). Respondents were highly educated. All rehabilitators were vaccinated against rabies and 94% received a rabies titre check at least every two years. Sixty-three percent of carers did not perceive viruses in flying foxes as a potential threat to their health, yet 74% of carers reported using PPE when handling flying foxes. Eighty-three percent of rehabilitators had received a flying fox bite or scratch at some point during their career. Carers provide an important community service by rescuing and rehabilitating flying foxes. While rehabilitators in this study have many excellent safety practices, including a 100% vaccination rate against rabies, there is room for improvement in PPE use. We recommend 1) the establishment of an Australia-wide set of guidelines for safety when caring for bats and 2) that the responsible government agencies in Australia support carers who rescue potentially ABLV-infected bats by offering compensation for PPE.
| Wildlife rehabilitators can encounter risks when handling animals, such as physical harm and exposure to infectious diseases. In Australia, rehabilitators that care for fruit bats may be exposed to Australian bat lyssavirus if bitten or scratched, which is fatal to people not vaccinated against rabies. We initiated a survey to assess rehabilitators’ perceptions of disease risk associated with fruit bats as well as rehabilitators’ safety practices. Despite an excellent rabies vaccination rate (100%), we found room for improvement in use of personal protective equipment. Supporting this, our regression analysis showed that use of protective equipment is associated with less chance of being bitten or scratched. Rehabilitators that are able to safely handle fruit bats can reduce risk to themselves, model good behaviour for onlookers, and protect animals from euthanasia. We recommend that carers develop Australia-wide guidelines for safety when rehabilitating bats and that the responsible government agencies in Australia support carers who rescue potentially lyssavirus-infected bats by offering compensation for costly protective equipment.
| Mainland Australia is home to four species of Pteropus fruit bats (flying foxes) which play an important ecological role by pollinating forest ecosystems and dispersing seeds as they forage for nectar, pollen, and fruit [1]. Though generally hardy animals, flying foxes occasionally sustain injuries due to extreme weather events [2] or manmade hazards (e.g. barbed wire, netting, power lines) [3]. A number of volunteer rehabilitators (commonly known as carers—the terms are used interchangeably here) care for injured, ill, and orphaned flying foxes, often as part of wildlife care groups [4]. A carer typically rehabilitates a flying fox until it can be returned to the wild; in cases of debilitating or lasting injury, a flying fox is euthanised or occasionally kept in permanent care.
Of concern to human health, and rehabilitator health in particular, is that flying foxes are reservoir hosts of zoonotic viruses—those passed from an animal to a human [5]. The most prominent of these in Australia are Australian bat lyssavirus (ABLV) and Hendra virus (HeV), both of which cause fatal disease in humans [5]; however, only ABLV is known to be transmitted directly from bats to humans. All four mainland species of flying fox (the black, Pteropus alecto; the grey-headed, P. poliocephalus; the little red, P. scapulatus; and the spectacled, P. conspicillatus) are reservoirs of ABLV and HeV [6–8]. ABLV infection has additionally been recorded in a species of insectivorous bat [7].
The four fatal human cases of HeV to date resulted from close contact with horses [9] and there is no evidence of direct flying fox to human transmission [10]. In contrast, ABLV can be transmitted directly from an infected bat to a human via a bite, scratch, or saliva contamination of broken skin or mucous membranes [11–13]. The clinical consequences of ABLV infection mirror those of classical rabies [14]. Guidelines recommend that wound care and rabies post-exposure prophylaxis be administered following any Category II exposure (nibbling of uncovered skin, minor scratches or abrasions without bleeding) or Category III exposure (scratch, bite, or saliva contamination of broken skin or mucous membranes) [15,16]. Rabies and ABLV are among the few viruses capable of causing clinical and pathological signs of disease in bats [17,18]. ABLV-infected bats often display neurological signs of infection and aggression and are frequently unable to fly [8], thereby increasing the opportunity for interaction with humans and other animals. Since the identification of ABLV in 1996, there have been three documented human fatalities (all occurring in Queensland) including one wildlife carer who had cared for both flying foxes and insectivorous bats [11–13].
Members of the Australian public are cautioned not to handle bats [19–21]; instead, bat rehabilitators are commonly called upon to rescue bats trapped in fencing or netting. These rescues, along with daily interactions such as treating injuries and hand-feeding pups, pose a bite and scratch hazard to carers. Carers are at special risk of ABLV infection as sick, injured, and orphaned bats have a significantly higher rate of ABLV infection than healthy bats [8]. Use of personal protective equipment (PPE) is recommended for rehabilitators [15,16] and rabies pre-exposure vaccination is typically required by care organisations. While state guidelines for the care of flying foxes exist [22–26], there is no set of unifying regulations in place across Australia, and safety practices vary between carers and care organisations.
No studies to date have comprehensively assessed risk perception, safety practices, and potential disease exposure in the Australian flying fox rehabilitator community. Studies of human-bat interactions in Australia incorporating carers have focused primarily on potential ABLV exposures [27–29]. More recent studies have examined the Australian public’s knowledge of and attitudes towards bats, including risk perception, but these were not designed to target bat rehabilitators [30–33]. Only two studies have specifically characterized the rehabilitator community. After two outbreaks of HeV (then known as equine morbillivirus) in Queensland in 1994, 128 bat carers were tested for antibodies to HeV and additionally asked to report their contact history (including bites and scratches) with flying foxes [10]. The study reported that some carers were concerned about the risk of HeV infection from flying foxes, but did not provide exact numbers. A 1998 survey explored demographics and motivations of flying fox rehabilitators, with a minor focus on risk perception [4]. Neither addressed safety practices, thus information regarding this aspect of care is especially deficient.
This study addresses the current lack of information about the flying fox rehabilitator community in Australia by presenting updated demographic data, assessing disease risk perception among carers (specifically focusing on viruses), and exploring the safety practices carers employ and the reasons underlying their actions.
A total of 21 email addresses for Australian flying fox and wildlife care organisations, as well as wildlife health interest groups, were identified via 1) online searches using various combinations of the following keywords: “flying fox”, “bat”, “carer”, “rehabilitator”, “wildlife”, and “Australia” and 2) referral. All organisations (S1 Table) were contacted via a solicitation email containing a link to an online survey (SOAR: Survey Of Australian flying fox Rehabilitators) hosted on SurveyMonkey from 8 May to 1 June 2014. The survey was open to Australian adults (aged 18 years and older) who had cared for flying foxes within the last twelve months. Participants were encouraged to share the survey link with rehabilitators unaffiliated with a care organisation.
The survey design was based in part on previous work [4] but was modified and updated to reflect the increased awareness of flying foxes as reservoir hosts for a variety of zoonotic viruses [34]. The survey was piloted with four individuals familiar with flying foxes but not involved in their care. Recommendations were incorporated before wider distribution to the target audience.
The survey (S1 File) included demographic questions on gender, age, state or territory of residence, level of education, and whether the respondent had ever completed a similar survey. Several questions addressed aspects of caring for flying foxes, such as motivations, years of experience, care organisation affiliation, and where flying foxes were housed while in care. Further questions focused on threat and risk perception. In the survey, “threat perception” was used to describe the implications of viral infections carried by bats on carer health, while “risk perception” was used in reference to questions relevant to the behaviour of carers to mitigate the risks associated with bites or scratches or potential exposure of pets. Carers were asked whether they felt that viruses found in flying foxes posed a potential threat to carer health. Carers also used ordered rating scales to rate the risk to human health posed by several hypothetical situations involving flying foxes. Response options ranged from “high risk” to “no risk” with an additional option of “don’t know.” Participants were questioned regarding their safety precautions, including rabies vaccination status, frequency of titre checks, whether these checks were self-initiated or required by a care organisation, preferred PPE, and whether they had ever been bitten or scratched by a flying fox. For certain multiple-choice questions participants were asked to further explain their choice in an open-ended response; participants were also provided room at the end of the survey to make additional comments.
Survey responses were exported from SurveyMonkey into Microsoft Excel as a.csv file. Responses to open-ended questions were manually spell-checked to ensure clarity between investigators, then analysed thematically [35]. Initial codes were generated and refined to classify responses; thematic maps were then created to sort codes into broader themes and sub-themes. Once themes were reviewed and refined, all responses were re-coded.
All statistical analyses were performed in the R statistical environment (version 3.1.1) [36]. A binary logistic regression was performed to determine whether rehabilitators’ gender, use of PPE, threat perception, and years of experience predicted variation in whether they had been bitten or scratched by a flying fox during their careers. The regression was performed using the glm function with a binomial error distribution and logit link function. Threat perception was assessed by carers’ responses to the question, “Do you feel that viruses found in flying foxes are a potential threat to the health of carers?”. Threat perception (yes/no), gender (male/female), and PPE (none/any) were categorical variables, while years of experience was included as a continuous covariate to account for the fact that rehabilitators with more experience would have had more occasions to be bitten or scratched. “Any” PPE was defined as all categories of PPE other than “nothing” (i.e. nitrile gloves, heavy gloves, or other PPE). Explanatory variables were checked for multicollinearity by calculation of variance inflation factors using the vif function in the car package [37]. Model performance was assessed by creating a receiver operating characteristic curve and calculating the area under the curve (AUC) with the roc function in the pROC package [38].
The study was approved by the CSIRO Health and Medical Research Human Research Ethics committee (protocol LR07/2014). Participants gave informed consent by reading a consent page and clicking a button to proceed with the survey.
A total of one hundred thirty-six survey responses were received. Participants’ responses were excluded if they did not complete the survey or did not fit the eligibility requirements, leaving 122 remaining responses (S2 File). The number of eligible rehabilitators reached by the online solicitation is unknown, and thus the response rate could not be determined; however, the number of responses is comparable to similar studies [4,10].
Selected demographic characteristics of rehabilitators are displayed in Table 1. Eighty-four percent (103/122) of carers were female; half of all carers (50%, 61/122) were 45–64 years old. Most responses were received from rehabilitators residing in New South Wales (47%, 58/122), Queensland (34%, 42/122), and Victoria (11%, 14/122). Fifty-five percent (68/122) of carers listed university or technical college as their highest level of education. Almost all (94%, 115/122) respondents indicated that they were affiliated with a care organisation; in total, 36 care organisations were represented. Nearly all rehabilitators (95%, 116/122) reported that they had never participated in a similar survey. Years of experience ranged from <1 to 30 years (median 5 years; not displayed in Table 1).
To investigate their motivations for caring, participants were asked, “What do you enjoy most about caring for/handling flying foxes?” and asked to choose two options from a list developed by Markus & Blackshaw [4]. Responses to this question are presented in Table 2. Returning the flying fox to nature (67%, 82/122) and helping to conserve the species (55%, 67/122) were the most popular choices. Among carers that listed “Other” as a motivation (14%, 17/122), the main themes were public outreach and education, close interaction with flying foxes, and a desire to help animals.
Safety practices employed by carers are summarized in Table 3. All rehabilitators (100%, 122/122) reported that they were vaccinated against rabies, which is used to protect against ABLV infection. Most carers (94%, 115/122) reported having their rabies titre checked at least every two years; for 58% (70/122), titre checks were required by their care organisation, while 38% (46/122) of carers initiated the checks. Of the five (4%, 5/122) rehabilitators who reported never having their titres checked, four had ≤ 2 years of experience. Most rehabilitators reported that flying foxes in their care were housed in a human residence (40%, 49/122) or a human and pet residence (30%, 36/122).
A little over a quarter of carers (26%, 32/122) reported that they typically used no protection to handle flying foxes. The ease of handling a flying fox was a high priority among these rehabilitators, with nearly all expressing the opinion that using gloves limited dexterity and reduced sensitivity. These limitations were felt to potentially increase the risk of a bite (e.g. due to not being able to feel the position of a flying fox’s head) or of harming the bat (e.g. by inadvertently applying excess force or pressure). Less commonly expressed was a belief that vaccination, experience, and training provided protection without a need for PPE.
Another quarter of carers (25%, 31/122) reported that they most commonly used nitrile or heavy gloves (nitrile or similar, 12/31; heavy gloves, 19/31). Of the 59 rehabilitators (48%, 59/122) who reported “Other” as their typical PPE, 18 (31%, 18/59) said that their choice of PPE depended on the situation and the flying fox being handled. Towels and blankets were the most popular alternative PPE listed; carers articulated that they provided a balance of personal safety and dexterity. Rehabilitators also reported using other types of gloves, arm protection (e.g. long sleeves, Neoprene arm protectors), and eye protection. There was a perception of low risk among carers who reported using “Other” or no PPE; it was not always specified whether this was a risk of being bitten or scratched or a risk of disease transmission.
Self-protection (e.g. against bites, scratches, and associated pain) was the most common motivation among rehabilitators who reported using some form of PPE. Protection of the bat was also a theme, with carers recognizing that receiving a bite or scratch would mean euthanasia of the bat (as guidelines for public health units recommend the testing, typically via a fluorescent antibody test on brain tissue, of any bat involved in a potential ABLV exposure [15]). Five rehabilitators reported using PPE specifically when being observed by members of the public to set a good example.
Although bats likely harbour a variety of pathogens (bacteria, parasites and viruses), only zoonotic viruses such as ABLV have been associated with human disease and were therefore the focus of the current survey. A majority of carers (63%, 77/122) reported that they did not feel that viruses found in flying foxes were a potential threat to carer health. When asked to elaborate, the key theme was that vaccination against rabies and regular titre checks eliminated any threat. Secondary themes included the importance of training, hygiene, and handling techniques, and perceptions of low prevalence of ABLV infection in flying foxes and low chance of disease transmission. Several carers expressed that they felt it was possible to recognize bats infected with ABLV. Direct bat-to-human transmission of HeV was not perceived as a threat. Among rehabilitators that did feel that viruses were a potential threat (37%, 45/122), themes included the potential lethality of ABLV infection (including a previous carer death) the high contact rate between carers and bats compared to the general public, and the potential for flying foxes to harbour other viruses. As in the former group, rehabilitators emphasized proper handling and training. Carers in both groups perceived certain categories of flying foxes to pose more of a threat to their health, namely adults, wild-caught bats, and “odd” or “suspicious” bats.
Participants additionally rated the risk to human health in a number of hypothetical situations (adapted from [30]) involving flying foxes (Fig 1, S2 Table). A member of the public handling a live flying fox was perceived as the riskiest situation (59%, 70/118, rated as high risk), while disposing of a dead flying fox was perceived as the least risky situation (47%, 55/118, rated as no risk). Responses of “Don’t know” were recorded for only two situations: a flying fox interacting with pets and disposing of a dead flying fox. Four respondents failed to assign a risk to one or more scenarios and were not included in Fig 1.
A binary logistic regression was performed to determine whether any groups of carers are more likely to be bitten or scratched when handling flying foxes. Specifically, this analysis examined whether carers’ use of PPE, gender, threat perception, and years of experience predicted variation in the response variable: whether a carer had been bitten or scratched by a flying fox in their career. Results of logistic regressions are typically reported in odds ratios (ORs), where an OR of 1 indicates that two groups have equal odds of experiencing an outcome of interest. An OR greater than 1 for a given group indicates that the group has higher odds of experiencing the outcome of interest than a second group. If the 95% confidence interval of the OR does not cross 1, this is generally considered a significant result [39].
Rehabilitators who wore no PPE had 9.58 times (95% CI: 1.83–177) the odds of being bitten or scratched compared to those who used any type of PPE (nitrile gloves, heavy gloves, or other protection; Fig 2, S3 Table). Wearing heavy gloves provided the best protection, followed by nitrile gloves and other protection (S4 Table). A carer’s gender, threat perception, and years of experience did not significantly predict variation in being bitten or scratched (Fig 2, S3 Table).
Most carers (83%, 101/122) indicated that they had been bitten or scratched by a flying fox during their career (females: 82%, 84/103; males: 89%, 17/19). The Communicable Diseases Network Australia (CDNA) recommendation for post-exposure management (PEM) comprises both wound management and receiving post-exposure prophylaxis [15]. Carers were asked to elaborate on how they responded to being bitten or scratched or how they would respond if they had not been bitten or scratched. For carers who had been bitten or scratched, this question was intended to assess actual post-exposure actions taken, but a number of carers gave hypothetical responses. Carers reported a range of PEM; at one extreme, carers reported that they would do nothing or ignore the wound, while at the other extreme, carers reported that they would wash the wound, apply an antiseptic or virucide, and receive a rabies booster shot. Other carers reported that they would practice wound care but not seek medical attention. A main theme was that carers’ responses depended on several factors, such as a bat’s health status and rescue or care history. Scratches were reported to be common, especially from orphaned flying foxes, and perceived to be less risky than bites. Some carers factored their rabies titre level or date of last vaccination into the decision to seek medical attention. Attitudes towards euthanasia of a flying fox that had bitten or scratched a rehabilitator (as per Public Health Unit guidelines) varied. While some carers reported that they would euthanise the bat themselves or contact health authorities to arrange euthanasia, others emphasised that they would never euthanise a bat, or only as a last resort.
One hundred twenty-two eligible responses were received for this survey, a number comparable to similar studies reported in 1998 and 2014 [4,10]. Demographic characteristics of carers in 2014 were similar to those reported in 1998 [4]; most carers were female and were between the ages of 45–64. While Markus & Blackshaw [4] did not directly ask rehabilitators’ educational level, the authors inferred from occupation that many were highly educated. Likewise, 55% (68/122) and 22% (27/122) of carers in this study listed university/technical college and postgraduate study, respectively, as their highest level of education. Rehabilitators were thus highly educated compared to the Australian population as of May 2014, of which 45.7% attained a bachelor’s degree, graduate or advanced diploma, or Certificate III/IV, and 5.2% received a postgraduate degree [40].
Compared to the 1998 survey [4], the present study had a greater geographic representation. No responses were received from Victoria, the Australian Capital Territory, Northern Territory, or South Australia in 1998, whereas rehabilitators from these states and territories made up 18% of respondents in the current survey. This may reflect increased ease of contacting carers through online methods, rather than any shift in range limits of flying foxes [41]. The high percentage of respondents from NSW and QLD is likely due to the large number of care organisations found in these two states, which in turn reflects the distribution of flying foxes along the eastern coast of Australia, and is unlikely to introduce significant bias to the results. While at least 36 care organisations are represented, the true number of organisations represented is probably higher, as some carers did not specify the organisation to which they belonged, while others indicated only an umbrella organisation rather than an individual branch.
Since 1998, a number of novel viruses have been identified in both frugivorous and insectivorous Australian bats (e.g. Cedar virus and other paramyxoviruses, Broome virus) [42–45]. However, disease threat perception rates associated with bats amongst carers have remained moderate. A smaller percentage of carers in 2014 compared to 1998 felt viruses found in flying foxes were a potential threat to carer health (2014: 37%, 45/122; 1998: 41%, 49/119), but this difference was not statistically significant (P = 0.2862, two-tailed Fisher’s exact test). Threat perception appears to be driven by a focus on ABLV. Given the low number of fatalities due to ABLV combined with high rates of rabies vaccinations and titre checks, rehabilitators may not view potential viral transmission as a threat.
Risk ratings of hypothetical situations involving flying foxes were generally intuitive (e.g. a member of the public handling a flying fox was likely perceived as the riskiest situation because average citizens are rarely rabies-vaccinated). Two situations are of particular interest. A flying fox interacting with pets had the second-highest number of “high risk” ratings. This is important considering that 30% of rehabilitators reported that flying foxes were cared for in a human and pet environment. Disposing of a dead flying fox was perceived as the least risky situation. Although the CDNA considers contact with a flying fox that has been dead for more than four hours to be low risk [15], rabies virus has been found to remain viable for more than four hours under favourable temperature and sunlight conditions [46]. Thus, rehabilitators disposing of flying fox carcasses should still consider using PPE.
Despite moderate threat perception of viruses in flying foxes, and moderate-to-low risk perception of hypothetical rescue situations, carers reported high frequency of PPE use when handling flying foxes. This discrepancy may be due to regulations imposed by care organisations, or, as some rehabilitators indicated, a desire to model appropriate behaviour for observing members of the public. Although we asked carers to report what they “typically” used to handle flying foxes, it is possible that carers vary their PPE depending on the threat they associate with a particular bat. Adult and wild-caught bats were perceived to pose more of a threat to carers, which corresponds with data on bats submitted for ABLV testing as part of a surveillance program between June 1996 and March 2002 [8]. However, the recent detection of ABLV in three juvenile flying foxes underscores that bats of all ages can be infected [47]. While age and origin of a bat are relatively easy to identify, “odd” or “suspicious” bats were also perceived to pose an increased threat. ABLV infection can cause a range of behaviours in infected bats, from paresis (weakness) to aggression, and carers may have different thresholds in considering a bat suspicious. A carer overconfident in her or his ability to diagnose ABLV infection, or unwilling to have a bat euthanised, might postpone or forgo seeking medical attention if bitten or scratched by a flying fox.
When carers in Queensland and New South Wales were tested for HeV antibodies in the mid-1990s, 74% reported having been bitten and 88% reported having been scratched by a flying fox [10]. These values are comparable to the 83% in this study who reported having been bitten or scratched. While this percentage does not give a sense of how many times a carer has been bitten or scratched, just a single exposure can be sufficient for ABLV infection. Our results suggest that rehabilitators should use PPE when handling flying foxes in order to reduce their odds of being bitten or scratched. However, we recognize that PPE is just one component involved in handling a bat, and that as carers emphasized, learning safe handling techniques is also important.
Carers ranged in the levels of PEM they reported, from taking no action to fully adhering to the CDNA guidelines (practicing wound care and receiving post-exposure prophylaxis). Many carers appeared to adjust their PEM based on the wound’s severity (e.g. bite or scratch), the perceived risk posed by the bat (e.g. age, behaviour), and knowledge of their rabies titre level.
Our results may be subject to selection bias because the survey was hosted online and was thus only available to people with internet access. There may be less representation from carers in remote areas and carers with lower socioeconomic status. In addition, because participants were recruited via email discussion lists, rehabilitators unaffiliated with a care organisation may be underrepresented. The lack of a centralized registry of carers prevents an estimate of whether the sample is representative of the carer community as a whole. Our results may also be subject to volunteer bias, as rehabilitators who have a high level of compliance with recommended safety measures may have been more likely to complete the survey. Because all experiences were self-reported, they may be subject to recall bias. Although the survey was anonymous to encourage honesty, participants may have underreported risky behaviours.
Some carers indicated that the wording of the hypothetical risk scenarios were unclear, as the risk to human health could be interpreted from the point of a carer or a member of the public. For this reason, we did not include a numeric measure of risk perception, as calculated by Young et al. [30], as an explanatory variable in the regression analyses.
This report describes the results of the first survey designed to explicitly gather data on disease risk perception and safety practices among Australian flying fox rehabilitators. We found that carers are highly aware of ABLV, but do not perceive viruses in general to pose a threat to their health. Rehabilitators in this study have many excellent safety practices, including a 100% vaccination rate against rabies, but there is still room for improvement in the use of PPE, both in overall use and in use of heavier-duty equipment that can offer better protection. One barrier to PPE use may be cost; since carers are typically volunteers, carers may not prioritize PPE given limited funds. PEM also presents an opportunity for improvement, as some carers report not practicing wound care or seeking post-exposure prophylaxis after a bite or scratch from a bat. We recommend the development of Australia-wide guidelines for safety when caring for bats. These guidelines, developed by carers, should emphasize the importance of proper PPE use to reduce the risk of being bitten or scratched, which will in turn protect bats from the threat of euthanasia. The guidelines should additionally provide recommendations on rabies vaccination and frequency of titre checks.
Although they work in a volunteer capacity, rehabilitators provide an important service to their communities by rescuing and rehabilitating flying foxes. Carers act as first responders in diverse situations, ranging from rescuing bats from barbed-wire fence to treating large numbers of bats in extreme heat waves, and are thus put at increased risk of zoonotic disease transmission. Bites and scratches are common in rehabilitators’ lifetimes, although this study did not measure how frequently these occurred. Given that Australian state and territory government agencies recommend that members of the public should rely on carers to handle bats [19–21], we recommend that these agencies in turn support carers. Through the HeV PPE rebate program, Queensland veterinarians are offered compensation for the cost of initial purchases of PPE, as well as for PPE used during an HeV investigation [48]. Similarly, state and territory government agencies could offer compensation to carers who rescue a potentially ABLV-infected bat. Such a program would help protect rehabilitators, and thus enable them to continue caring for the bats that are so vital to Australia’s forest ecosystem.
|
10.1371/journal.pbio.1001317 | The Evolution of Sex Is Favoured During Adaptation to New Environments | Both theory and experiments have demonstrated that sex can facilitate adaptation, potentially yielding a group-level advantage to sex. However, it is unclear whether this process can help solve the more difficult problem of the maintenance of sex within populations. Using experimental populations of the facultatively sexual rotifer Brachionus calyciflorus, we show that rates of sex evolve to higher levels during adaptation but then decline as fitness plateaus. To assess the fitness consequences of genetic mixing, we directly compare the fitnesses of sexually and asexually derived genotypes that naturally occur in our experimental populations. Sexually derived genotypes are more fit than asexually derived genotypes when adaptive pressures are strong, but this pattern reverses as the pace of adaptation slows, matching the pattern of evolutionary change in the rate of sex. These fitness assays test the net effect of sex but cannot be used to disentangle whether selection on sex arises because highly sexual lineages become associated with different allele combinations or with different allele frequencies than less sexual lineages (i.e., “short-” or “long-term” effects, respectively). We infer which of these mechanisms provides an advantage to sex by performing additional manipulations to obtain fitness distributions of sexual and asexual progeny arrays from unbiased parents (rather than from naturally occurring, and thereby evolutionarily biased, parents). We find evidence that sex breaks down adaptive gene combinations, resulting in lower average fitness of sexual progeny (i.e., a short-term disadvantage to sex). As predicted by theory, the advantage to sex arises because sexually derived progeny are more variable in fitness, allowing for faster adaptation. This “long-term advantage” builds over multiple generations, eventually resulting in higher fitness of sexual types.
| For well over a century, biologists have wondered why sex is such a common mode of reproduction, given the immediate 2-fold fitness cost entailed by the reduced number of offspring per parent. The most classic explanation is that sex is favoured because it helps to generate the variation necessary for adaptation. While theoretical models and indirect lines of evidence support this idea, there are no direct experimental data and it is far from obvious whether any such advantage could balance the considerable costs of sex. Using experimental populations of a facultatively sexual species of rotifer, we demonstrate that rates of sex evolutionarily increase as populations adapt to novel environments. We show that sex creates a diverse array of genotypes, including many that are quite unfit but also others that are very fit in the new environment. Though the average fitness of these sexually derived offspring is lower than that of asexuals, those well-adapted genotypes generated by sex contribute disproportionately to future generations, causing the genetic propensity for sex to ultimately increase.
| The pervasiveness of sex, given its varied and potentially large costs, is highly perplexing [1]–[10]. Numerous hypotheses have been proposed and sophisticated theoretical analyses have helped to define the conditions under which particular hypotheses may apply [3],[11]–[14]. Despite the importance of this problem, rarely have the hypotheses been tested by examining how key factors affect the evolution of sex [15]–[18].
Over a century ago, Weismann [19],[20] argued that sex might be beneficial because it helps generate the variation necessary for adaptation. While intuitively appealing, the idea is not necessarily correct as sex will increase the variance in fitness only if there is a preponderance of “negative genetic associations” such that good alleles are often found in genomes with bad alleles. It was later realized that such negative associations may develop under certain forms of nonlinear selection (as occurs when approaching an adaptive optimum [21]–[25]) or, perhaps more importantly, due to an interaction between directional selection and drift, known as the Hill-Robertson effect [6],[26]. For these more sophisticated reasons, Weismann's original conjecture is thought to be valid and is considered by many as the leading explanation for the evolutionary function of sex [27].
Rigorous theory shows that sex can facilitate adaptation [21],[24]–[26],[28], but the conditions under which this will translate into a net selective advantage for sex itself are more limited [21],[24],[29]–[34], especially given the infamous costs of sex [1],[3]. Indeed, a number of studies have demonstrated that sexual populations adapt faster than asexual populations [7]–[10],[35],[36]. Such studies imply a population- or group-level advantage to sex, though none of these studies directly competed sexual and asexual populations against one another during adaptation. Consequently, it is impossible to know whether any benefit to sex with respect to adaptation would have been outweighed by its immediate costs. More importantly, group-level advantages to sex cannot be used as evidence for the maintenance of sex within populations, as emphasized by John Maynard Smith [1] and George Williams [37].
Better support for adaptation providing a “gene-level” advantage to sex comes from survey studies showing that recombination tends to increase as an incidental by-product of directional selection on other traits [38],[39]. However, the evolution of recombination is not the same as the evolution of sex. The intrinsic costs of sex and recombination differ, and even ignoring these costs, theory shows that selection on recombination is often not an accurate predictor of selection on sex because of segregation effects [29],40,41. More evidence for adaptation favouring genetic shuffling comes from a recent study in C. elegans showing adaptation favours outcrossing over self-fertilization [42]. Though a related phenomenon, this is not direct evidence for the role of adaptation in maintaining sex. The contrast between selfing and outcrossing is not the same as the contrast between asex and sex because different types of genetic associations are involved. Further, the intrinsic costs of sex (relative to asexuality) differ from the intrinsic costs of outcrossing (relative to selfing). Despite these important differences, the recombination and outcrossing studies offer indirect evidence that adaptation can select for sex. However, direct experimental evidence for adaptation favouring sex is lacking.
Beyond the crucial step of empirically demonstrating the requirements necessary to cause an evolutionary increase in sex, a more thorough understanding requires identifying the population genetic mechanisms that drive the evolution of sex. A general theoretical framework divides the total selection on sex into components arising from “short-term” and “long-term” effects [30] (see [41],[43] for further discussion of these terms). The “short-term” effect of sex refers to the immediate fitness consequences of rearranging gene combinations. Sex does not directly change allele frequencies, but it does re-distribute alleles (i.e., breaks down genetic disequilibria). Whenever alleles interact to affect fitness (i.e., if there is dominance or epistasis), altering gene combinations will change fitness. For this reason, the mean fitness of sexual-derived progeny can differ from that of asexually derived progeny coming from the same set of parental genotypes. The short-term effect of sex results from alleles that promote sex being associated with different gene combinations than the alleles that promote asexual reproduction [10]–[12].
Regardless of whether there are gene interactions or not, the redistribution of alleles through sex can result in the variance of sexually derived offspring being different (higher or lower) than that of asexually derived offspring. If the sexually derived subpopulation has more variance than the asexually derived subpopulation, then the former will better respond to subsequent selection. Though sex does not immediately affect allele frequencies, it alters the genetic variance, which allows subsequent selection to cause allele frequencies to diverge between more versus less sexual lineages. The “long-term” effect of sex refers to selection on sex that results from genes that promote sex becoming associated with a different frequency of fitness-affecting alleles [10]–[12]. It is worth noting that the label “long-term” effect is somewhat misleading as long-term effects can arise over a single complete generation involving both reproduction and selection. While long-term effects can build in strength over multiple generations, it is not necessary to have hundreds of generations for this form of selection to alter the evolution of sex.
There is a myriad of hypotheses for the evolutionary maintenance of sex, but they can all be interpreted as providing an advantage to sex through either short- or long-term effects [11],[12]. Despite the importance of these general mechanisms to our understanding of selection on sex and the potential to study these effects by examining the effect of sex on the mean and variance in fitness, no empirical study has clearly linked the evolution of sex to either of these mechanisms.
Here we examine the Weismann hypothesis by evaluating whether sex is favoured during adaptation to a novel environment. We do this by (i) examining whether sex increases in frequency during adaptation and (ii) measuring the difference in fitness between naturally occurring sexual and asexual genotypes at various points during the course of adaptation. Finally, we test whether the advantage to sex arises from a short- or long-term effect by examining the effects of sex on the mean and variance in fitness at several points over the course of adaptation. This allows us to test the prediction that sex is favoured through a long-term advantage [24],[27],[31],[32].
To test Weismann's hypothesis at the within-population level, we used replicated experimental populations of the haplodiploid monogonont rotifer Brachionus calyciflorus. These rotifers are facultatively sexual, reproducing amictically at low densities but changing to mictic (sexual) reproduction in response to a chemical stimulus indicative of high density [44]. When stimulated, the amictic mothers produce daughters that develop into mictic females. Unfertilized mictic females produce haploid eggs that develop into males, and if young mictic females mate, her haploid eggs are fertilized and develop into resting eggs. Amictic females hatch from resting eggs when stimulated by environmental cues. Previous work with rotifers from this source population reveals there is substantial genetic variation in the strength of the stimulus needed to induce sex, thus allowing for the evolution of rates of sex [17]. Amcitic eggs develop within 1 d and the time the females start producing their first offspring is less than 24 h after hatching. Fertilized mictic eggs (resting eggs) from this population hatch spontaneously at a high rate under typical lab conditions (between 1 and 5 d after they are produced; see Material and Methods and [45]). From these observations, we approximate the mean time to complete an asexual generation to be ∼1.5 d. The “sexual cycle” takes ∼6 d but involves two generations (∼1.5 for the production of mictic females and then ∼4.5 d for sexually derived offspring to hatch and mature). Given that the overall rate of sexual reproduction is low, the average generation time is expected to be closer to 1.5 d than 4.5 d.
All of our replicate populations descended from a common natural source. However, 10 replicates came from subpopulations more recently adapted in the lab to one environment (“Environment A”), whereas 10 other replicates came from subpopulations more recently adapted to another (“Environment B”). The two environments differ in their algal food source and NaCl concentration. For our main experiment, 10 replicates (5 from each environment) serve as control (non-adapting) populations and are maintained under the environmental conditions to which they had already adapted. The remaining 10 populations are transitioned to the alternative environment (5 replicates A→B; 5 replicates B→A); we refer to these as “adapting” populations. This reciprocal experimental design offers the opportunity to infer the role of adaptation per se, rather than a particular environment, in affecting the evolution of sex. Population sizes are relatively large throughout the experiment (N≅3,500–7,500).
Over the course of 70 d (ca. 45 asexual generations) of evolution, there is clear evidence of adaptation in the populations that experience an environmental change. Population densities, which initially plummet during the transition to the alternate environment, increase to stable levels characteristic of well-adapted populations (Figure 1). Moreover, estimates of individual fitness show similar increases over time (Figure 2). In contrast, control (non-adapting) populations remain stable both in density and in fitness assay measures over this period.
As predicted by the Weismann hypothesis, rates of sex increase during the period of rapid adaptation. Later, sex declines as adaptation slows, presumably reflecting the intrinsic costs of sex outweighing the diminishing benefits of sex as the opportunity for adaptation declines. These temporal changes in sex are evident in two separate measures of sex. First, we use the fraction of fertilized mictic eggs (out of all eggs) as an in situ measure of sexual investment (fertilized mictic eggs are visibly distinct from other eggs). There is an obvious increase in the investment in sexual eggs during the period of rapid adaptation, followed by a decrease (see Figure 1 legend for statistics). This pattern cannot be explained by density effects directly triggering sex as the observed changes in sex go in the opposite direction from the well-known pattern for this species in which high density induces sex [44]. In contrast to the adapting populations, the percentage of fertilized mictic eggs in the control populations shows little change.
Our second measure is based on a controlled assay of the propensity for sex. Each week, 42 rotifers are isolated from each population and maintained individually under standardized conditions for three clonal generations. Third generation individuals are exposed to a specified concentration of a sex-inducing stimulus. We determine the fraction of individuals that are induced into sexual reproduction by this cue. In the adapting populations, we observe a significant increase in the propensity for sex during the early phases of adaptation, followed by a subsequent decline (Figure 3, see legend for statistics). In contrast, the propensity for sex declines monotonically in the control populations. On day 37, a second set of 10 adapting populations (five for each environment) was initiated from the control populations. We refer to these as the “Set 2 adapting populations.” Our data on this second set are less detailed and over a shorter period, but these populations also show a similar increase in sex.
Several lines of evidence indicate that the changes in the propensity for sex (Figure 3) are not due to a plastic response stimulated by moving into a new environment. First, the assays are always performed in the third generation after isolation into standardized conditions so these changes cannot be due to the immediate shock of changing environments. Second, the Day 0 data represent assays on third generation clonal descendants of rotifers that have just transitioned to the alternative environment. As there is no difference in sex between control and adapting populations at this initial time point, it is clear that sex is not a stress-induced response resulting from a mismatch between genotype and environment. An unlikely third possibility is that the stress of a novel environment accumulates over multiple generations to induce the delayed rise in sex observed in Figure 3. As described in Figure S1, we tested this by transferring rotifers to the alternative environment and propagating them clonally for an extended period as individual lineages to prevent changes due to selection. Compared to rotifers maintained in the original environments, there was no change in the propensity for sex either in the short term or after a 16-generation delay (which corresponds to the same time period as the rise in sex observed in Figure 3).
The results of our in situ measure of “investment in sex” (Figure 1) and our well-controlled “propensity for sex” assay (Figure 3) are reasonably congruent for the adapting populations, but there is a puzzling inconsistency with respect to the controls. In the control populations, the “propensity for sex” declines monotonically, whereas the “investment in sex” is low and relatively constant. Because the strength of the sex-stimulating cue used in the “propensity for sex” assay is much stronger than the expected strength of the cue experienced in situ in the control populations based on their densities, we do not expect to see the same magnitude of change in the two types of measures. Nonetheless, some corresponding decline in the “investment in sex” measure is expected but not observed. As the data are somewhat noisy, it is conceivable that we simply lack the statistical power to detect a decline.
In this system, as in most others, the products of sexual reproduction are not phenotypically identical to those of asexual reproduction (i.e., fertilized mictic eggs are different than amictic eggs). Consequently, it is a concern whether changes in sex are actually due to selection for sex rather than a by-product of selection for some correlated feature. However, this alternative interpretation is inconsistent with our results. The parallel responses of adapting populations in both environments, as well as the pattern of temporal change within environments (increases during adaptation followed by decreases as fitness plateaus), indicate that neither environment favours resting (fertilized mictic) eggs per se.
Nonetheless, it would be more compelling to show differences in fitness between sexual and asexual genotypes to provide direct evidence of selection on the genetic consequences of sex. For this purpose, we sample fertilized mictic and amictic eggs weekly from each population. Rotifers are propagated individually before we measure lifetime reproduction for multiple clones of the third generation of each genotype, allowing us to compare recently created sexual and asexual genotypes that all develop from the same type of egg. The results, representing fitness measures on ∼22,000 individuals, are presented in Figure 2. In the control populations from both environments, genotypes derived from sexual reproduction are much less fit than those from asexual reproduction. In contrast, when populations transition to a new environment, we initially find no difference in fitness between sexually and asexually derived genotypes. As adaptation proceeds, sexually derived genotypes become significantly more fit than asexually derived genotypes (days 21–35 for A→B; days 21–49 for B→A). As populations approach their new fitness equilibrium, the pattern reverses again and the sexual load characteristic of well-adapted populations begins to re-emerge (days 42–70 for A→B; days 63–70 for B→A; see Table S1 for statistical comparisons between sexuals and asexuals at each time point).
The assays described above reflect differences in fitness between naturally occurring sexually and asexually produced offspring. The genotypes isolated for this assay are appropriately biased in that sexual genotypes will tend to descend from lineages with more sex in their history than asexual genotypes. This difference between the genealogical histories of naturally occurring fertilized mictic and amictic eggs is what yields a measure of the net effect of sex, but this also precludes a more detailed understanding of the population genetic mechanisms responsible. We cannot tell whether an observed advantage of sex results from the immediate benefit of genetic mixing (“short-term advantage”) or the accrued benefit of past selection on genetic variation released by previous bouts of sex (“long-term advantage”).
To differentiate between short- and long-term effects as mechanisms driving the evolution of sex, it is necessary to examine how sex affects the fitness of progeny from an unbiased set of parents. By comparing sexually and asexually derived offspring from random sets of parents, we can determine how sex affects the distribution of offspring fitness values without the confounding effects of past selection associated with sexually inclined lineages. By exposing a random sample of rotifers from each population to an extremely strong sex stimulus that induces sex at a very high rate across a wide array of genotypes [17], we obtain sexually derived offspring from a largely unbiased sample of parents. We obtain asexually derived offspring from random samples of rotifers from each population kept at low densities. Eggs are isolated and maintained individually under standardized conditions for multiple clonal generations before replicate measures of lifetime reproduction are made for each genotype. (This procedure is illustrated in Figure S2, where it is contrasted with the assay procedure for measuring fitness from naturally occurring sexual and asexual genotypes.) We perform this type of assay for the first set of adapting populations at two time points, sampling parents on day 33 (shortly after the propensity for sex has peaked) and day 67 (when adaptation is near complete and the propensity for sex is in decline). For the second set of adapting populations, we sample parents somewhat earlier during the course of adaptation (16 and 30 d after their initiation, corresponding to days 53 and 67 on Figures 2 and 3). We have analogous data for control populations for each of these time points.
The distributions of sexually and asexually derived offspring fitnesses are shown in Figures S3 and 4; ratios comparing key properties of these distributions are shown in Figure 4. Sexually produced offspring have lower mean fitness than asexually produced offspring (t = −18.9, df = 16, p = 2.3×10−12 for adapting populations; t = −62.8, df = 26, p<2×10−16 for control populations; Day 67 data from the first set of adapting populations are not used in these comparisons as fitness has plateaued before this point). The lower average fitness of sexually produced offspring is predicted whenever non-additive gene action (dominance and/or epstasis) plays an important role in shaping patterns of genetic associations (disequilibria) [3],[11]. Bad combinations of alleles that have been eliminated by past selection can be recreated by sex, reducing mean fitness, a phenomenon that can be thought of as a “sexual load” and is sometimes called “genetic slippage” [46],[47].
Although the distributions of sexually derived offspring fitness have lower averages than the corresponding distributions for asexuals, the variances for sexuals are higher (t = 17.0, df = 16, P = 1.1×10−11 for adapting populations; t = 6.1, df = 26, P = 1.8×10−6 for control populations). This increased variance associated with sex reflects the existence of negative genetic associations likely generated either by epistasis or Hill-Robertson effects [6],[26],[32]. Sex and recombination are expected to dissipate these disequilibria, resulting in an increase in genetic variance.
The pattern of sex reducing the mean but increasing the variance, indicative of a short-term disadvantage but a long-term advantage to sex, is qualitatively similar in both adapting and control populations. Although the directions of the short- and long-term effects are the same between treatments, the relative magnitudes differ. In Environment A, sex reduces mean fitness by ∼30% in control populations but only by ∼20% in adapting populations (before populations reach near complete adaptation). A similar effect occurs in Environment B, where sex reduces mean fitness by ∼45% in control populations but only by ∼20% in adapting populations. Thus, the short-term disadvantage of sex is ∼30%–50% smaller in adapting populations. This difference between adapting and control populations is supported by formal comparisons (t = −3.8, df = 16, p = 0.001 for Environment A; t = −14.8, df = 16, p = 9.6×10−11 for Environment B).
The long-term effect results from a difference in genotypic diversity in fitness created by sexual reproduction relative to that resulting from asexual reproduction. This is often discussed in terms of differences in variance. As described above, sexual genotypes are more variable in lifetime reproduction than asexuals in both adapting and control populations. The relative increase in variance due to sex is greater in adapting populations than in control populations (t = −2.4, df = 16, p = 0.03 for Environment A; t = −5.4, df = 16, p = 5.9×10−5 for Environment B; Figure 4C,D).
There are potential problems with using the variance as a measure of the long-term effect of sex when the mean fitness of sexually and asexually derived offspring differs. High variance of sexuals may result from the production of low fitness genotypes. The generation of such variants is not useful for adaptation and thus cannot contribute to a long-term advantage to sex. Rather, we are interested in whether sex tends to produce particularly good variants. For this purpose, we compare the average fitness of the top 10% of sexually and asexually produced genotypes (Figure 4E,F). Sex generates significantly better genotypes in the top end of its fitness distribution than does asexual reproduction in adapting populations (t = 3.06, df = 16, p = 0.007), but the opposite is true in control populations (t = −13.3, df = 26, p = 4.2×10−13). Similar patterns are observed using the top 5%, 15%, or 25% (see Figure S5).
Above, we have discussed fitness distributions for sexually and asexually derived offspring obtained in two different ways (Figure S2). First, we isolated naturally occurring fertilized mictic and amictic eggs, and thus, these two types of eggs came from lineages with different histories of sex (Figure 2). Second, we generated sexual and asexual offspring from random sets of parents (Figures 4, S3, S4, S5). If we compare these assays at a similar time point during adaptation (close to day 30), there is a dramatic difference. In the first assay (from non-random parents; Figure 2), we find that sexually derived offspring have higher average fitness than asexually derived offspring. In the second assay (random parents; Figure 4), sexually derived offspring have lower average fitness. As discussed above, even though sex produces offspring that are less fit, on average, than asexually derived offspring, sex also generates some particularly high fitness genotypes. These genotypes contribute disproportionately to future generations, carrying alleles for sex with them. As a result of generating extreme variants and subsequent selection, alleles that increase sexual propensity become associated with alleles conferring adaptation. Consequently, naturally occurring sexuals eventually become more fit, on average, than asexuals because of this accrued benefit from past selection. The higher average fitness of sexuals observed in the first assay midway through the course of adaptation represents the long-term advantage realized. However, as time passes, beneficial alleles will eventually accumulate in less sexually inclined genotypes, and thus the advantage to sex will erode as populations approach an adaptive optimum and the influx of new beneficial alleles slows. The short-term disadvantages of sex, along with other costs of sex, can then drive an evolutionary decrease in sex.
Previous experiments have shown that sexual groups can adapt faster, thereby providing indirect evidence of a group-level advantage to sex (at least in the absence of intrinsic costs) [7]–[10],[35],[36]. For the first time, we demonstrate that the frequency of sex within a population rises over time during adaptation. These results are consistent with the idea that Weismann's hypothesis can provide an advantage to sex at the gene level that can be sufficiently strong to overwhelm the intrinsic costs of sex. Weismann's hypothesis and related theories [27] make a strong prediction that sex should be favoured during adaptation because of a long-term advantage, and we have found evidence supporting this mechanism. On the other hand, much of this body of theory [29],[31]–[34] does not make a clear prediction with respect to short-term effects; when Hill-Robertson effects are responsible for negative disequilibria, short-term effects could be positive, negative, or zero, but models invoking non-linear selection to generate negative disequilibria predict negative short-term effects [24]. In our study short-term effects appear to be substantial. The reduction in negative short-term effects that seems to accompany the transition to a new environment (possibly reflecting environment-specific epistasis) is somewhat unexpected and reduces the threshold for a long-term advantage to create a net benefit to sex.
Though our results provide direct support for the operation of the Weismann hypothesis, we have not shown quantitatively that the Weismann hypothesis alone can fully explain the observed evolution of sex. It is possible that other factors also contribute to these changes. Here we consider two alternatives, but our data do not provide strong support for either. In this system, as in most others, the products of sexual reproduction are not phenotypically identical to those of asexual reproduction (i.e., fertilized mictic eggs are different than amictic eggs). Consequently, some of the observed evolution of sex could be a by-product of selection for fertilized mictic eggs (rather for genetic mixing). However, this alternative interpretation is inconsistent with our results. The parallel responses of adapting populations in both environments, as well as the pattern of temporal change within environments (increases during adaptation followed by decreases as fitness plateaus), indicate that neither environment favours resting (fertilized mictic) eggs per se. Moreover, we have direct evidence that genotype, independent of egg type, is important during adaptation; naturally occurring, sexually derived genotypes are more fit than asexually derived genotypes even when both develop from the same egg type (Figure 2).
A second factor of possible importance to our results is differential selection between the sexes [48]–[51]. In this system, (sexual) males are haploid, potentially allowing for more efficient selection on recessive beneficial alleles than can occur in the absence of sex. Under this hypothesis, we would expect that when we experimentally force individuals through the sexual cycle, the resulting offspring should, on average, be more fit than with asexual reproduction because of the extra sieve of haploid male selection that occurs incidentally during the process of creating sexual offspring. In fact, we observe the opposite; sexually derived genotypes from random sets of parents are less fit on average than asexually derived genotypes (Figure 4). This should not be taken as evidence that haploid selection has no effect at all, but rather it suggests that haploid selection does not play a strong role.
The costs of sex are expected to be high in this system, and it is unclear whether the observed benefits can outweigh these costs. In this regard, it is worth considering three points. First, modifier alleles that increase the rate of sex by a small degree experience only a small fraction of the cost of sex [52]. Second, long-term benefits can be quite powerful, especially when the baseline rate of sex is quite low [30],[52], as it is in our system (5%–7%, Figure 1). A modifier allele that slightly increases the rate of sex only suffers the cost of sex in those generations where it induces sex but enjoys the benefit of having created a good genotype for many generations.
Third, the advantage gained by “high-sex” genotypes during adaptation is likely considerably larger than it appears. The observed advantage in fitness of naturally occurring, sexually derived genotypes over asexually derived genotypes during adaptation reaches 30%–50% (Figure 2), but this underestimates the difference in fitness between “high-sex” genotypes and “low-sex” genotypes. This is because the distinction between “high-sex” genotypes and “low-sex” genotypes with respect to degree of sex is quantitative; both types use both reproductive modes. Consequently, the naturally occurring fertilized mictic eggs will come from both “high-sex” and “low-sex” parental genotypes but be biased toward coming from the former. Conversely, the amictic eggs will come from both “high-sex” and “low-sex” parents but be biased toward the latter. Thus, the difference in fitness between genotypes isolated from fertilized mictic eggs versus those isolated from amictic eggs will clearly underestimate the true difference in fitness between “high-sex” and “low-sex” genotypes.
The two environments used here were used in a previous study of the evolution of sex. The main result of that study was that higher rates of sex were maintained when populations experienced spatial heterogeneity in selection [17]. However, that experiment also provided a hint of the Weismann effect as even the spatially homogenous (control) populations showed an initial increase in sex followed by a decline on a time scale similar to that observed here. Because both environments were novel compared to the source population of rotifers, it is likely that the initial increase was due to an advantage to sex during adaptation to those environments. Though adaptation itself was not measured in that study, those results are consistent with what we have reported there.
Despite its importance to theory, the effect of sex on the distribution of offspring fitness has been measured in only a handful of taxa [45],[53]–[57]. In several of those cases [54],[55],[57], sex has been observed to reduce the mean but increase the variance, suggesting that long-term advantages to sex may be reasonably common but in none of those previous cases were evolutionary changes in the rate of sex measured. As seen here, short-term disadvantages coupled with long-term advantages can occur in cases where sex increases (adapting populations) as well as in cases where sex continuously declines (controls). However, we found substantial differences in the magnitudes of these effects between adapting and control populations. Moreover, the direction of the “long-term effect,” rather than just the magnitude, differs between adapting and control treatments if one considers the top 10% rather than the variance (the use of the latter is based on a weak selection approximation [30]).
While our experiment is unique in being able to link a change in sex to short- and long-term effects, a number of details remain unknown. A long-term advantage is expected to exist when sex dissipates negative genetic associations. Are negative genetic associations built by non-linear selection [21],[24],[25] or Hill-Robertson effects [6],[26],[32]? Similarly, we do not know whether dominance or epistasis is responsible for the immediate consequences of sex (short-term effects). Such information will be important to help understand the relative importance of segregation and recombination in driving the evolution of sex.
For sex to have any effect genetically, there must be genetic variation within populations. Even in well-adapted populations, we see clear evidence of genetic variance; when sex is imposed on random samples of parents, there is a dramatic decline in fitness. What sort of variation is responsible for this effect? One simple explanation is that recessive deleterious alleles hitchhike to high frequency in a heterozygous state and can persist as long as populations reproduce asexually much of the time so that deleterious homozygotes are rarely produced. A second explanation is that multiple high-fitness co-adapted genotypes are maintained by some form of balancing selection such as frequency-dependent selection. When it occurs, sex and recombination breaks down these co-adapted genotypes, resulting in low fitness genotypes. Unlike the first explanation, this alternative can apply to both haploid and diploid systems and so has been invoked to account for sex-induced reductions in fitness in studies on haploid Chlamydomonas [8],[54],[55],[58].
Though our experiment is consistent with the main tenets of the Weismann hypothesis, it also demonstrates a well-known weakness of this idea. The advantage to sex observed here is brief on an evolutionary time scale. Perhaps if adaptive optima are continually shifting, selection for sex could be maintained indefinitely [24]. Do selective pressures in nature change sufficiently frequently to explain the observed levels of sex? This is an empirical issue requiring data from the field. Lab-based studies such as the one reported here are necessary to directly evaluate the potential of hypotheses and to test their underlying mechanisms. However, such studies alone cannot prove any hypothesis as the explanation for the ubiquity of sex in nature. Attempts to study the evolution of sex in the field [15],[18],[53],[59] will be needed to evaluate the importance of results from theory and lab experimentation.
The rotifers for this study descended from a population collected from sediment taken from Lake Onondaga, New York, in spring 2009 [45]. The populations used here were started from lab stocks that have previously been adapted to two different food conditions, Environments A and B (which we have previously called “low” and “high” food conditions [17]). These environments differ with respect to the algal suspension used to maintain the rotifers. Algae (Monoraphidium minutum, SAG 278-3, Algae Collection University of Goettingen) were taken from long-term chemostats to ensure constant food conditions over the course of the experiment. Chemostats were either run with a low nitrogen concentration in the medium = 160 µM (Environment A) or a higher nitrogen concentration in the medium = 1,000 µM (Environment B). The inorganic medium (nitrate as limiting N-source) was modified after [60], with additional 0.5 g/l NaCl to Environment A. The algae suspension for replacement of medium was prepared by diluting algae to concentrations of 2×106 cells/ml with the same inorganic medium used for the chemostats but lacking nitrogen.
Stocks were kept under either of the two food conditions for 11 mo prior to the start of the experiment (∼9.5 mo with low rates of migration between the two environments—heterogeneous populations in [17]—and no migration for the last 6 wk before the start of the experiment; during this period, populations consisted of approximately 8,000 to 10,000 individuals), and experimental populations were started from these pre-adapted populations (10 populations per environment). Populations were maintained as semi-continuous cultures by replacing 10% of each culture including rotifers and algae every second day with a respective algae solution. Rotifer, amictic, and resting egg densities were enumerated under a stereoscope each time food was replaced [17]. Experimental populations of Brachionus calyciflorus were kept at 25±1°C (12/12 D-L) in tissue culture flasks (Sarsted, 500 ml) and moved randomly three times per week on the three shelves of the incubator. For more detailed methods, see Text S1.
Replicate experimental populations were either maintained in the same environment to which they had previously adapted (non-adapting control populations; n = 5 per environment) or were transitioned to the other environment 10 d after the start of the experiment—that is, either from Environment A to B, or from B to A (adapting populations; n = 5 per environment). The transition occurred by substituting the other algae source during the regular food replacement schedule (see above). About 95% of the algae was replaced after 1 wk. Ten additional adapting populations (n = 5 per environment) were started at day 37 of the experiment. To create these populations, the 10% extracted media of the control populations on day 36 were pooled with others from the same environment and the following day were distributed among five new populations for each adapting population. The remaining volume was filled with fresh medium and the respective algae solution.
Sexual reproduction in Brachionus species is density dependent and stimulated by a chemical signal that is produced by the rotifers [61]. The propensity for sex was measured weekly and followed the protocol in [17]. Briefly, we isolated 42 asexual individuals from each population and individuals were transferred to single wells with 10 ml of food containing medium, so that each rotifer received the same food from which they were isolated. Individual rotifers were maintained under these conditions for two generations and one neonate of the third generation after isolation was individually transferred to a single cell of a 96-well plate with conditioned medium [17] containing the same food source from which they were isolated. The initial female was removed after they produced the first offspring and the offspring was scored as amictic or mictic by the type of offspring they produced. Sexual females produced only males (haploid) because they were unmated in the assay.
Ten fertilized mictic (resting) eggs and 10 amictic eggs were isolated weekly from each population and transferred to a single well of a 24-well plate for hatching (Figure S2). The two types of eggs can be distinguished by their morphology: amictic eggs are completely filled and have a pale gray colour, while resting eggs are only partially filled and have a much darker coloration. Rotifer females from amictic eggs hatched within 1 d after isolation, and females from resting eggs hatched within 1 to 5 d. To avoid differences that could occur because sexually derived genotypes develop from resting eggs whereas asexually derived genotypes develop from amicitc eggs, we maintained each genotype by clonal reproduction for two generations prior to fitness measurements (in the same food environment from which they were isolated). The first five offspring from the third generation (asexual) after isolation were used to measure lifetime reproduction (five individuals per genotype). Each individual was placed in an individual well, and each day, the number of offspring was recorded and the female was transferred to a new well with fresh medium and food until the female died. Lifetime reproduction was used as a measure for fitness.
Spontaneously occurring fertilized mictic eggs are expected to originate from a non-random subsample of the population. To examine the effects of sex on a more random sample of genotypes, we transferred 5% of the populations to a new flask, added additional food, and allowed the population to grow to high densities (Figure S2), inducing almost the entire population to switch to sexual reproduction (density >30 females per ml; all genotypes are expected to switch to sexual reproduction at this density; cf. [17], Figure S2). Another 5% were transferred to a flask containing a large volume of medium and food, and these subpopulations were kept at low densities to ensure only asexual reproduction (less than one female per ml). After 7 d, 20 resting eggs were isolated from the high-density subpopulations and transferred individually to single wells for hatching and fitness assays as described above. Similarly, 20 amictic eggs were transferred from the low-density subpopulations. This procedure was applied to samples collected on Days 33 and 67 for the first set of adapting populations and on Days 53 and 67 for the second set of adapting populations. For each of these time points (Days 33, 53, and 67), similar data were collected from the non-adapting control populations.
Multivariate statistical analyses were done in the R statistical environment [62]. Treatment (Control A, Control B, Adapting B→A, Adapting A→B) specific models (generalized mixed models GLMM using the lmer4 package [63]) were used to test for differences in the percentage of fertilized mictic eggs (Figure 1) and propensity to reproduce sexually (Figure 3) with time as a fixed effect and replicate population as a random effect (using binomial error structure). To test for the increase and decrease in sex in the adapting populations, quadratic and linear models were compared.
The effect of sex on the distribution of genotype fitnesses was examined as follows. All analyses were performed on genotypic mean values (from five clonal replicates per genotype). Mean fitness of sexually and asexually derived rotifers hatched from naturally occurring eggs isolated directly from the experimental populations (Figure 2) were compared using environment and time-point-specific generalized mixed models (GLMM) with reproduction mode (sexually or asexually) as fixed and replicate population nested in reproduction mode as random effect.
To examine the effects of sex on a more random sample of genotypes, the distributions of sexually and asexually derived offspring were compared with respect to mean, variance, and mean of the top 10%. In each case, the data were analyzed with a linear model on the difference between sexuals and asexuals, using population as the unit of replication. To evaluate the effect of sex within treatments, we examined the significance of the intercept in separate analyses for adapting and control populations (variables were coded such that the intercept reflects the average effect across environments and time). For adapting populations, only Day 33 data for Set 1 were used as fitness had plateaued before Day 67 (Figure 2). For Set 2, we used the average values from Days 53 and 67 for each population (these represent days 16 and 30 of adaptation for Set 2). We obtained qualitatively similar results, using a total evidence approach by combining p values [64] from individual paired t tests (sex versus asex) for each set in each environment. To directly compare the effects of sex between adapting and control populations, we analyzed the difference in log of fitness between sexuals and asexuals in a linear model including both adapting and control treatments. Variance was calculated as the variance among genotypic means.
|
10.1371/journal.pntd.0007560 | Potent and selective inhibitors for M32 metallocarboxypeptidases identified from high-throughput screening of anti-kinetoplastid chemical boxes | Enzymes of the M32 family are Zn-dependent metallocarboxypeptidases (MCPs) widely distributed among prokaryotic organisms and just a few eukaryotes including Trypanosoma brucei and Trypanosoma cruzi, the causative agents of sleeping sickness and Chagas disease, respectively. These enzymes are absent in humans and several functions have been proposed for trypanosomatid M32 MCPs. However, no synthetic inhibitors have been reported so far for these enzymes. Here, we present the identification of a set of inhibitors for TcMCP-1 and TbMCP-1 (two trypanosomatid M32 enzymes sharing 71% protein sequence identity) from the GlaxoSmithKline HAT and CHAGAS chemical boxes; two collections grouping 404 compounds with high antiparasitic potency, drug-likeness, structural diversity and scientific novelty. For this purpose, we adapted continuous fluorescent enzymatic assays to a medium-throughput format and carried out the screening of both collections, followed by the construction of dose-response curves for the most promising hits. As a result, 30 micromolar-range inhibitors were discovered for one or both enzymes. The best hit, TCMDC-143620, showed sub-micromolar affinity for TcMCP-1, inhibited TbMCP-1 in the low micromolar range and was inactive against angiotensin I-converting enzyme (ACE), a potential mammalian off-target structurally related to M32 MCPs. This is the first inhibitor reported for this family of MCPs and considering its potency and specificity, TCMDC-143620 seems to be a promissory starting point to develop more specific and potent chemical tools targeting M32 MCPs from trypanosomatid parasites.
| In recent years, the pharmaceutical company GlaxoSmithKline announced the disclosure of small collections of antiparasitic compounds to facilitate research and drug development for three of the main Tropical Neglected Diseases- i.e. Human African Trypanosomiasis, Leishmaniasis and Chagas Disease. These collections include new chemical entities with potential novel mechanisms of action that are likely to be active against a wide variety of targets. Taking advantage of these open access molecules, we successfully set up medium-throughput screening assays to find the first inhibitors of two metallocarboxypeptidases of the M32 family, a group of proteolytic enzymes proposed to play several roles in the biology of trypanosomatids including peptide catabolism, maintenance of parasite adaptive fitness and hydrolysis of bioactive peptides from the human host.
| Members of the Trypanosomatidae family comprise parasitic organisms that cause highly disabling and often fatal diseases in humans and animals. The species that are responsible for human infections are Trypanosoma brucei, which cause Human African trypanosomiasis (HAT), Trypanosoma cruzi, the etiological agent of Chagas disease (American trypanosomiasis), and Leishmania spp., which cause different forms of leishmaniasis. Together, these vector-borne diseases constitute a substantial public health problem for which there is not a satisfactory treatment [1]. Major side-effects, and in some cases low effectiveness, are common problems associated with existing therapy. This situation makes imperative the development of new chemotherapeutic options. In this context, new drugs based on unique aspects of parasite biology and biochemistry are of great interest, particularly in the case of emerging resistance to traditional treatments [2–4]. In this scenario, proteases have become popular targets as these enzymes play key functions in parasite biology; namely nutrition, cell cycle progression, invasion and pathogenesis, among others.
The M32 family of metallocarboxypeptidases (MCPs) contains a group of hydrolases, which although being broadly distributed among prokaryotic organisms, are only present in a few eukaryotes including some green algae and trypanosomatids [5]. This unique phylogenetic distribution, in particular the absence of M32 enzymes in metazoans, has been considered an attractive trait due to the high specificity/selectivity potential of this family for drug target development. Within the Trypanosomatidae family several conserved M32 MCPs have been characterized [5–10]. Nonetheless, the cellular or biological functions of these proteins are currently unknown, as well as their essentiality status. In T. brucei, the genome-wide study by Alsford et al. (2011) reported no significant lost-of-fitness after induction of T. brucei MCP-1 (TbMCP-1) RNAi in bloodstream and procyclic stages, as well as in the differentiation from procyclic to bloodstream forms [11]. More recently, however, it has been shown that TbMCP-1 null mutant strains display extended doubling times in culture, suggesting that this enzyme might contribute to the adaptive fitness of the bloodstream form [12]. On the basis of their biochemical properties and stage-specific expression, the L. major M32 carboxypeptidase has been implicated in the catabolism of peptides and proteins to single amino acids required for protein synthesis [7]. The restricted substrate preference of T. cruzi MCP-1 (TcMCP-1), plus its strong structural similarity to angiotensin I-converting enzyme (ACE), neurolysin and thimet oligopeptidase [8], have also pointed out a possible regulatory role of this family in the metabolism of small peptides. In fact, it has been shown that TcMCP-1 can produce des-Arg9-bradykinin [6], a peptide that promotes the process of cell invasion through B1 receptors by the T. cruzi trypomastigotes [13]. In this sense, two reports have suggested that M32 peptidases are secreted by trypanosomatids [14, 15], a fact that is in agreement with this hypothesis. In the current scenario, the availability of selective small-molecule modulators of M32 MCPs activity would be of great value to ask mechanistic and phenotypic questions in both biochemical and cell-based studies. However, no inhibitors have been reported to date for these enzymes or other members of this family.
Recently, a diverse collection of ~ 1.8 million compounds from the proprietary library of GlaxoSmithKline (GSK) has been run through whole-cell phenotypic screens against L. donovani, T. cruzi and T. brucei. As a result, three anti-kinetoplastid chemical boxes of ~200 compounds each were assembled and open sourced [16]. The guiding design criteria for these molecule sets were chosen to include structures from different chemical families that are likely to be active against a wide variety of targets. By taking advantage of this diversity, we identified the first inhibitors of the M32 family of MCPs within the GSK HAT and CHAGAS chemical boxes. As model enzymes of the M32 family we employed TcMCP-1 and TbMCP-1, which have similar basic amino acid preference at the P1´ position and share 71% of protein sequence identity [5, 6].
To evaluate compounds in the HAT and CHAGAS chemical boxes, we devised a continuous assay for each MCP, based on FRET (fluorescence resonance energy transfer) peptides. We carried out the optimization process in 384 well plates, the same format used for the screening of the compound collections. For the selection of the most suitable substrate for the HTS assay, we initially assayed six FRET peptides against both enzymes. These were recently designed considering subsite preferences (P1´-P4) of TcMCP-1 and TbMCP-1 [12]. However, because no peptide was completely satisfactory for both enzymes, we selected independent substrates, Abz-LKFK(Dnp)-OH and Abz-RFFK(Dnp)-OH, for TcMCP-1 and TbMCP-1 assays, respectively. After substrate selection, a convenient enzyme concentration in the assay was determined through the activity of 2-fold dilutions of TcMCP-1 and TbMCP-1 at a fixed substrate concentration (Fig 1A and 1B). Moreover, the Selwyn test [17] revealed no enzyme inactivation under the conditions tested (Fig 1C and 1D). Thus, for a wide range of enzyme concentrations (for both MCPs), the V0 vs. [E]0 curves showed a linear behavior (Fig 1E and 1F). In particular, for [TcMCP-1]0 < 0,34 nM and [TbMCP-1]0 < 1,53 nM, the rate of the substrate hydrolysis remained constant for at least 40 minutes, a suitable time to perform the screening (Fig 1A and 1B).
The best balance between TcMCP-1 activity on Abz-LKFK(Dnp)-OH substrate (estimated as dF/dt) and the time over which the reaction displayed linear kinetics was achieved at [TcMCP-1]0 = 0,17 nM. Under these conditions, the enzyme showed the typical hyperbolic behavior predicted by the Michaelis-Menten equation (Hill coefficient = 1,06) and an estimated KM value of 2,23 ± 0,28 μM (Fig A in S1 Text). Similarly, when the TbMCP-1 concentration was fixed at 1,25 nM we obtained a KM value on Abz-RFFK(Dnp)-OH substrate of 0,37 ± 0,06 μM (Hill coefficient = 1,03) (Fig A in S1 Text). To afford the best opportunity to find compounds with different inhibition modalities, we decided to employ balanced assay conditions (i.e. KM/[S] = 1)[18]. Using these conditions, preliminary characterization experiments of both optimized assays showed good general performance, with a dynamic range (μC+—μC-) higher than 15 RFU/sec, a μC+/μC- ratio ≥ 50, good reproducibility (VC < 5%) and a Z´ factor value in the range 0,6–0,8.
Using the same lot of substrate and enzyme, the 404 compounds present in the HAT and CHAGAS chemical boxes were screened at a single fixed dose (25 μM). Each plate included 24 positive and negative controls, plus 16 wells containing 31,25 mM EDTA (inhibition control) alternately located in columns 11, 12, 23 and 24. In general, for each MCP, both plates presented highly similar Z´ scores although best values were obtained for the TbMCP-1 assay presumably due to the lower background signal of the Abz-RFFK(Dnp)-OH substrate. To avoid the interference of highly fluorescent compounds, an auto-fluorescence cut-off value equal to 2x105 RFU was used to accept or discard a molecule from the subsequent analysis. Using this limit, ~19% of the compounds were eliminated for TcMCP-1 and TbMCP-1 assays. Statistics are summarized in Table 1.
As shown in Table 2, if we consider a cut-off value ≤ 3 standard deviations from the control mean (μc+ - 3σc+), 70 and 132 inhibitory molecules were retrieved for TcMCP-1 and TbMCP-1, respectively. To reduce the number of resultant hits, we explored other two thresholds focusing only in outliers: i) those compounds showing slopes >3σ standard deviations above the average of all slopes in the plate (control independent) and ii) those compounds showing an inhibition percentage >3σ standard deviations above the average for the plate (control dependent). Interestingly, both criteria retrieved exactly the same list of compounds for TcMCP-1 (n = 5) while for TbMCP-1 the intersection between this two groups was lower (2 out of 4 compounds).
In the secondary screening we decided to include all compounds that showed ≥ 40% of inhibition (TcMCP-1: 23 compounds; TbMCP-1: 27 compounds). To estimate IC50 for the resulting hits, two-fold serial dilutions, ranging from 7,5 pM to 62,5 μM, were analyzed against both recombinant MCPs using identical assay conditions as in the primary screening. Prior to the analysis of the complete dataset, we examined whether there was a correlation between the inhibition percentages in the primary (compound concentration 25 μM) and secondary screening, using only the data corresponding to a compound concentration of 31,5 μM. This was important to assess consistency of data, as both screening rounds were performed without technical replicates due to limitation of compound stocks. For TcMCP-1, 9 compounds presented similar behavior in both screenings (correlation coefficient r2 = 0,9868; slope = 1,146) (Fig 2A) whereas 7 molecules failed to reach ≥ 40% of inhibition threshold (n = 6) or displayed no inhibition (n = 1) (correlation coefficient r2 = -0,518; slope = 0,2595). Additionally, 7 compounds performed better in the secondary screening (correlation coefficient r2 = 0,5156; slope = 1,2749). For the T. brucei enzyme, consistent results in both assays were achieved only by 8 compounds (correlation coefficient r2 = 0,9349; slope = 1,080) (Fig 2B). About 45% of the samples did not repeat the ≥ 40% of inhibition criterium (n = 10) or did not inhibit (n = 2) TbMCP-1 (correlation coefficient r2 = 0,1163; slope = 0,3173). Finally, another 7 molecules performed better in the secondary screening than in the first round. Despite the observed round to round discrepancies (Table A in S1 Text), we decided to continue curve analysis for all the compounds, with the exception of the three that showed no inhibition at 31,5 μM during secondary screening.
For TcMCP-1, five compounds (TCMDC-143620, TCMDC-143422, TCMDC-143456, TCMDC-143209 and TCMDC-143385) showed an IC50 value ≤ 10 μM (Fig 3A and Table 3). In good agreement, the four more potent molecules (TCMDC-143620, TCMDC-143422, TCMDC-143456 and TCMDC-143209) also inhibited the T. brucei enzyme (Table 3). Compounds TCMDC-143385 and TCMDC-143172 (which display an IC50 ~10 μM for TcMCP-1) did not reach the 40% inhibition threshold in the TbMCP-1 primary screening and were left out from the secondary analysis. Other potent molecules, namely TCMDC-143409 and TCMDC-143323 were specific inhibitors of T. brucei enzyme or produced little inhibition on TcMCP-1 (< 30%) (Fig 3B and Table 3). The structure of the top-five inhibitors for each enzyme is shown in Fig 3C.
To first assess the possibility that these lead compounds have shared structural features that help explain their bioactivity profile, we performed three different clustering strategies: one using Tanimoto similarity (Fig B in S1 Text), one based on shared substructures (overlap of Maximum Common Subgraphs, MCS) (Fig C in S1 Text), and the third one based on shared physicochemical properties (Fig 4). Whereas the Tanimoto clustering was expected to be inconclusive based on the premises used to assemble the chemical boxes (one or two putative chemotypes per box [16]); the clustering based on physicochemical properties also showed no significant correlation between these properties and the observed IC50s. Similarly, MCS clustering provided no insights into candidate substructures guiding the activity or specificity of the compounds against each enzyme. In all three strategies, the clusters not only group up dissimilar potencies, but also mix compounds with different enzyme specificity.
To determine the number and type of Zinc-binding groups (ZBGs) among the compound leads, an MCS analysis was performed using an ad hoc curated [20, 21] database of ZBGs. From a total of 48 groups available in the database, only six of them were found among 24 of the 30 lead compounds: pyridine (14 compounds), sulfonamide (7 compounds), imidazole (4 compounds), pyrazole (3 compounds), diol (1 compound) and hydrazide (1 compound). The majority of compounds (24 out of 30) presented at least one ZBG in the structure. More specifically, 15 with a single group and 9 with two groups were found. All compounds and their corresponding ZBGs have been summarized in Fig D in S1 Text.
Considering the abundance of ZBGs and heteroatom-containing moieties in the hits, we evaluated the possibility of a nonspecific mechanism of inhibition (involving metal chelation) for the top-five inhibitors identified in the screening for each enzyme. Because M32 MCPs show a strong topological similarity with ACE [22], we chose this enzyme to estimate the IC50 value for each molecule. As done for the MCPs essays, ACE activity was analyzed employing a FRET substrate, Abz-FRK(Dnp)P-OH, at a concentration equal to the apparent KM of the enzyme ~3 μM [23]. Experiment set up is summarized in Figs E and F in S1 Text. For comparative purposes, captopril, a potent competitive ACE inhibitor, was included in the analysis (IC50 ~1 nM) (Fig 5A). Under these conditions, no inhibition could be detected for any of the compounds evaluated, thus suggesting that these molecules are not promiscuous metallocarboxypeptidase inhibitors (Fig 5B, 5C and 5D) but are instead specific inhibitors of M32 MCPs.
M32 MCPs have an unusual phylogenetic distribution (with trypanosomatids being among the few eukaryotic genomes encoding these enzymes). Hence M32 MCPs from parasites arose naturally as interesting candidates for drug target development. Furthermore, the current lack of knowledge about the cellular and/or physiological role(s) of these enzymes makes the identification of potent inhibitors a task of great significance, as these compounds may be used as molecular probes to potentially identify natural substrates, to recognize the specific pathways in which they are involved or, hopefully, to perform their chemical validation as drug targets. In this work, we describe the first drug-like inhibitors of TcMCP-1 and TbMCP-1, two closely related MCPs from the human pathogens T. cruzi and T. brucei, respectively. Our starting point were the GSK HAT and CHAGAS boxes, two small collections containing non-redundant, chemically diverse and highly bioactive compounds [16], which could facilitate future optimization efforts.
Although we initially aimed for a common assay for both MCPs, we soon realized that the use of different FRET substrates for each enzyme resulted in better general performance of the individual assays (considering signal robustness, temporal duration of linear kinetics, dynamic range, μC+/μC- ratio and Z´ factor). Surprisingly, the substrates that resulted most suitable for the developed HTS assays were not, in any case, those that showed the best values of kcat, KM and kcat/KM in their previous kinetic characterization [12]. Although different assays were used to screen these collections, we were able to find specific inhibitors for both enzymes, and perhaps more important, mutual inhibitors; suggesting the consistency of inter-assay results. Of note, specific inhibitors for each enzyme were distributed evenly among HAT and CHAGAS boxes with no apparent bias. This fact confirms the importance of not circumscribing the search to just the pathogen-specific box, but instead to widen the search to all the boxes available, as previously observed for T. cruzi cysteine peptidase cruzipain [24].
Due to the limited amount of compound stocks, we decided to implement the screening of chemical boxes in singlet, with primary evaluation of all compounds at a fixed dose and further dose-response analysis of unconfirmed hits in a secondary screening. As expected, given the error-prone nature of the single-well (single dose, single replicate) measurements used in primary screening, significant discrepancies in inhibition were observed for some compounds in comparison to secondary dose-response evaluation. These discrepancies are common and may be due to a variety of factors [25]. Besides intrinsic compound-specific and experimental data variability [26], these factors may include solubility issues (given that in primary and secondary screenings both the final concentration and serial-dilution protocol were different), differential stability of compounds in stock (10 mM) and working (2 mM) solutions [27], unintended absorption of the compounds to different containing materials during storage, moderate dose-dependent quenching effects of compounds on fluorescence readouts, among others [28]. In addition, although we included 0,01% Triton X-100 in assay buffer, compound-specific aggregate formation was not tested and thus, cannot be dismissed.
As mentioned, we identified in this work eight molecules able to inhibit both MCPs. These mutual inhibitors came from both boxes in similar numbers, as previously noted for enzyme-specific compounds. Interestingly, in all cases they were more potent inhibitors of TcMCP-1, for reasons that are as yet unclear. Importantly, four of these compounds proved to be inactive on ACE, a Zinc-dipeptidyl carboxypeptidase involved in various physiological and physiopathological conditions in mammals [29] which shows significant structural similarity to M32 enzymes [22, 30]. This fact strongly suggests that despite the structural resemblance and the small number of compounds tested here, the identification of inhibitors with high selectivity for trypanosomatid M32 MCPs over ACE can be achieved, a point in favor to the specific druggability of these enzymes.
The identified inhibitors display high structural diversity, with many showing only marginal similarity to the other hits, hence representing different structural clusters and presumably, different inhibitory scaffolds. In this regard, the presence of “unpaired” hits is not surprising, considering that no more than two members of the same structural cluster were included per box during collection assembly [16] and that “twin” compounds might well not pass the activity or auto-fluorescence filters included in this work. Among the identified inhibitors, only TCMDC-143265 and TCMDC-143551 share similar core structures, thus probably populating the same cluster and sharing a common active scaffold. A significant part of both molecules is identical and adopts the same spatial conformation (Fig G in S1 Text), with the largest differences located around the benzamide ring. Besides the obvious differences in the length and position of sulfonamide substituents, the chlorine substitution in position 2 imposes a ~90° rotation of the benzamide ring in TCMDC-143265 compared to TCMDC-143551, where all ring systems are almost coplanar. Interestingly, these structural differences seem to dictate the selectivity toward TcMCP-1, as TCMDC-143551 inhibits both enzymes whereas TCMDC-143265 is specific for TbMCP-1. Even for this pair of compounds, there is no evident substructure responsible for M32 MCPs bioactivity; though this is probably a biased observation due to the lack of well-defined structural features for M32 MCPs inhibitors.
Although the crystallographic structure of TcMCP-1 has been determined [8] and subsite specificity have been explored for both enzymes using FRET substrate libraries [12] and mutagenesis [6, 8], little is yet known about how substrates are accommodated into the catalytic groove, which residues are key determinants of subsite specificity and the significance of the hinge-type movement between L and R domains in the stabilization of enzyme-substrate or enzyme-inhibitor complexes. With all these gaps to fill, it seems risky to speculate about the modes of interaction of these new inhibitors with TcMCP-1 and TbMCP-1. However, a presumptive explanation can be put forward. As in the case of many other metallopeptidase inhibitors, it is likely that inhibition of trypanosomatid M32 MCPs occurs throughout the perturbation of the coordination sphere of the catalytic metal ion (presumably Zn2+ in the case of TcMCP-1 and TbMCP-1, by extension from other M32 enzymes [31]). Typically, synthetic metallopeptidase inhibitors achieve preliminary affinity and target selectivity through the formation of stabilizing interactions with specific residues within the active site; while a ZBG is responsible for metal chelation, enhancing binding affinity, modulating selectivity and disrupting catalytic activity [32]. For the majority of the inhibitors presented here, it was possible to identify typical ZBG or at least, heteroatom-containing groups able to establish a coordinative bond with a Zn2+ ion (Fig D in S1 Text). For those compounds, an inhibition mechanism like the one described above is possible. For other molecules not having a Zn-coordinating group, the most plausible explanation is that inhibition occurs as a result of the prevention of substrate binding by the partial occupancy or the deformation of the catalytic cleft by the inhibitor molecule, as previously observed for Non-Zinc-Binding inhibitors of other metallopeptidases [33].
The vast majority of the hits identified here inhibit one or both MCPs in the micromolar range, with only a few of them showing potencies <10 μM. Outstandingly, TCMDC-143620 inhibits TcMCP-1 in the sub-micromolar range (it also inhibits TbMCP-1, but with potency ~7-fold lower). This is the most potent inhibitor described so far for an enzyme of the M32 family and seems a promising candidate for further structure-based optimization. The unusually high flexibility of the M32 MCPs around the active site [31, 34] prevented us to use a docking approach to get insights of the binding mode of this compound within TcMCP-1 and TbMCP-1 catalytic clefts. However, the TCMDC-143620 molecule seems able to form a variety of stabilizing interactions. These may include hydrophobic and electrostatic interactions, hydrogen bonding and the coordination to the metal ion through the pyridine ring. In addition, the presence of a central sulfonamide group and a distal nitrile group add further interaction possibilities to this molecule. For example, the sulfonamide group has been extensively incorporated into metallopeptidase inhibitors due to its ability to improve the enzyme-inhibitor binding by different mechanisms. These mechanisms include: i) direct formation of hydrogen bonds to the enzyme backbone, ii) properly redirection of bulky groups into enzyme pockets by inducing a twist in the structure of the inhibitor molecule and iii) even cooperate with other chelating groups in the coordination of the catalytic metal ion [35]. Similarly, the nitrile group in TCMDC-143620 can establish polar interactions, hydrogen bonds or react with serine or cysteine side chains to form covalent adducts which would greatly stabilize inhibitor binding [36]. Interestingly, the nitrile group is also able to form coordinative bonds with a variety of metal ions including Co2+, Mn2+, Fe3+, Cu2+ and Zn2+ [37]. Thus, a possible role of this group in the direct coordination of the catalytic metal ion cannot be discarded at present. The determination of the crystallographic structure of TcMCP-1 or TbMCP-1 in complex with TCMDC-143620 would provide a definitive answer to these questions as well as important clues to undertake the future lead-optimization of this hit.
A preliminary analysis of the bioactivity profile of TCMDC-143620 (https://pubchem.ncbi.nlm.nih.gov/compound/91800813) indicates that it shows potent activity against T. cruzi in culture and only moderate but measurable activity on T. brucei and L. donovani. Also, this compound exhibits moderate cytotoxicity on mammalian cell NIH 3T3 (IC50 = 13 μM) but resulted inactive on HepG2 (IC50 > 100 μM). Considering target-specific assays; this compound has a single bioactivity report. TCMDC-143620 was found to be a potent inhibitor (IC50 = 79 nM) of T. cruzi sterol 14-α demethylase (CYP51) enzyme, which is involved in the ergosterol biosynthesis pathway and was considered until recent years as a promissory therapeutic target for Chagas disease [38, 39]. The inhibition of this target is probably the cause of its reported anti-T. cruzi activity. This might also explain, at least partially, the moderate cytotoxic and anti-T. brucei and L. donovani activities reported for this compound, considering the global similarities of enzymes within CYP51 family [40, 41]. Although involved in other studies as part of the GSK CHAGAS Box [42], no further information is currently available from the evaluation of TCMDC-143620 against other molecular targets, except for our previous cruzipain study [24] where it was found to be inactive (~7,5% of cruzipain inhibition at 25 μM). A complete profile of the off-target activity of TCMDC-143620 would be critical for future optimization efforts in order to achieve a suitable M32 MCPs probe from this compound.
In summary, 30 micromolar-range inhibitors, presenting both high structural diversity and novelty, have been discovered for TcMCP-1 and/or TbMCP-1 by using continuous, fluorescent-based and HTS-capable enzymatic assays. The best hit shows sub-micromolar affinity for TcMCP-1, inhibits TbMCP-1 in the low micromolar range and, like other potent hits, is inactive on ACE. Considering its potency and specificity, this molecule seems to be a promissory starting point to develop more specific and potent tools to expand our understanding of the biochemistry and biological role(s) of M32 MCPs from trypanosomatid parasites and, hopefully, to assess in a near future their value as drug targets.
Triton X-100, MOPS (3-(N-morpholino)propanesulfonic acid), DMSO, EDTA and captopril were purchased from Sigma-Aldrich. Substrates Abz-RFFK(Dnp)-OH and Abz-LKFK(Dnp)-OH were from GenScript (Piscataway, NJ, USA). Black solid bottom polystyrene Corning NBS 384-well plates were from Sigma-Aldrich (CLS3654-100EA).
TcMCP-1 (MEROPS ID: M32.003) and TbMCP-1 were expressed as GST fusion proteins in E. coli BL21 (DE3) Codon Plus and purified as previously described [6, 8].
The HAT and CHAGAS chemical boxes [16] were provided by GlaxoSmithKline. The collection comprised 404 compounds, prepared as 10 mM stock solutions in DMSO (10 μL each) and dispensed in 96 well plates. For primary screening, a working solution (final concentration of 2 mM) for each compound was prepared by 1/5 dilution in DMSO while 1 μL of the 10 mM stock solution was used for secondary screening of selected compounds, as previously described [24]. The final concentration of compounds tested in primary screening was 25 μM, while the compound concentrations assayed in secondary screening ranged from 7,5 pM to 62,5 μM.
TbMCP-1 and TcMCP-1 activities were assayed fluorometrically with Abz-RFFK(Dnp)-OH and Abz-LKFK(Dnp)-OH substrates, respectively, in 100 mM MOPS pH 7,2 containing 0,01% Triton X-100. Assays were performed in solid black 384-well plates (final reaction volume ~80 μL) and the hydrolysis of the K(Dnp)-OH group was monitored continuously at 30 °C with a Beckman Coulter DTX 880 Multimode Reader (Radnor, Pennsylvania, USA) using standard 320 nm excitation and 420 nm emission filter set.
For each MCP, final substrate concentration was set to a value KM /[S] ~ 1. Optimal enzyme concentration was selected from 2-fold serial dilutions to match three criteria: (i) being linearly proportional to V0, (ii) display robust signal evolution at substrate concentration chosen and (iii) display linear kinetics for enough time to perform several reading cycles (at least 8 cycles, minimum time between cycles: 264 sec) through the 384-wells. In all cases, EDTA (final concentration 31,25 mM) was used as positive inhibition control.
To perform the primary screening, 1 μL of each compound (2 mM in DMSO, final concentration in the assay: 25 μM), EDTA (500 mM, final concentration in the assay: 31,25 mM) were dispensed into 384-well Corning black solid-bottom assay plates. Then, 40 μL of 100 mM MOPS, 0,01% Triton X-100 pH 7,2 containing TbMCP-1 (2,50 nM) or TcMCP-1 (0,34 nM) were added to each well, plates were homogenized (30 seg, orbital, medium intensity) and each well subjected to a single autofluorescence read (ex/em = 320/420 nm). Plates were incubated in darkness for 15 min at 30 °C and then 40 μL of Abz-RFFK(Dnp)-OH (4 μM) or Abz-LKFK(Dnp)-OH (0,8 μM) in assay buffer were added to each well to start the reaction. After homogenization (30 seg, orbital, medium intensity), the fluorescence of the Abz group (ortho-aminobenzoic acid) (ex/em = 320/420 nm) was acquired kinetically for each well (8 read cycles, one cycle every 300 seconds). Considering our previous experiences, the auto-fluorescent cut-off was arbitrarily set at 2x105 RFU to discard highly interfering compounds. All compounds were assayed in singlet (without replicates) due to the limited availability of stocks.
Raw screening measurements were used to determine the slope (dF/dt) of progression curves by linear regression for control and non-interfering compound wells. In the case of control-dependent hit selection criteria, percent inhibition percentage (%Inh) was calculated for each compound according to the following equation:
Inh=100∙[1−(dFdtWELL−μC−)(μC+−μC−)]
(1)
where dF/dtWELL represents the slope of each compound well and μC+ and μC− the average of MCP (no-inhibition) and substrate (no-enzyme) controls, respectively.
Compounds selected from primary screening were re-tested in a dose-response manner (final concentration ranging from 7,5 to 62,5 μM) using identical assay conditions. To avoid any positional and/or association bias, we randomly defined the row position for each compound. One μL of compounds stock (10 mM in DMSO) and EDTA (31,25 mM) were added to the first well of column 1, followed by addition of 40 μL of 100 mM MOPS, 0,01% Triton X-100 pH 7,2 buffer. After addition of 20 μL of the same buffer to subsequent wells of the plate, 22 serial 2-fold dilutions were made horizontally. The last two positions of every row were used, alternatively, for C+ and C− controls to reduce any positional and/or association bias. Then, 20 μL of activity buffer containing TbMCP-1 or TcMCP-1 were added to each well, except for those corresponding to C−; completed with 20 μL of activity buffer. After homogenization, 15 minutes of incubation at 30°C and autofluorescence measurement, the substrate (in activity buffer) was added to the previous mix. Data collection and processing were performed exactly as described above. Percentage of M32 MCPs residual activity was calculated for each condition according to the following equation:
%Res.ActMCP=100∙[(dFdtWELL−μC−)(μC+−μC−)]
(2)
where dF/dtWELL represents the slope of each compound well and μC+ and μC− the average of MCP (no-inhibition) and substrate (no-enzyme) controls, respectively. The IC50 and Hill slope parameters for each compound were estimated by fitting the four-parameter Hill equation to experimental data from dose-response curves using the GraphPad Prism program (version 5.03).
Purified rabbit lung ACE (EC 3.4.15.1) was purchased from Sigma-Aldrich. Enzyme activity was assayed fluorimetrically with Abz-FRK(Dnp)P-OH (ex/em = 320/420 nm) as substrate in buffer 0,1 M Tris-HCl, 50 mM NaCl, 10 mM ZnCl2, pH 7.0 containing 0,01% Triton X-100 as indicated in [23]. Selected compounds were tested in a dose-response manner (final concentration ranging from 7,5 pM to 62,5 μM) using identical assay conditions employed with both MCPs. Captopril (15 pM—125 μM) was used as inhibition control.
Three separate compound clustering routines were used. One of them derived from calculated or predicted molecular features, and the other two directly inferred from different distance metrics between compounds: one using Tanimoto similarity and another one using the overlap score calculated in a MCS (Maximum Common Subgraph) pipeline. The Tanimoto distance compound clustering was performed to rapidly find compound pairs, if available, within the leads. OpenBabel 2.4.1 [43] was used to export molecule MDLs from SMILES format, available from GSK chembox summary.
For Tanimoto clustering, the indexes were calculated using ChemFP 1.3 [44] with ob2fps bindings and simsearch -NxN as parameter. ChemFP results were parsed and analyzed using an ad hoc perl script, setting the distance (D) between compounds as D = 1—Tindex. The distance matrix was built using melt and acast from R Data table package [45].
To assess the MCS clustering, all compounds were imported into a R script using Chemminer [46] and further analyzed using fmcsR [47] for batch MCS calculations.
For the molecular feature clustering, a perl script was built to run XlogP3 v3.2.2 [48] through all lead compounds. Features used to build distance matrix, along with their corresponding values, can be found in Table 4. All clustering plots were achieved using the R base hierarchical clustering tool, hclust.
To find ZBGs among lead compounds, a curated database of such chemotypes was first created (Table B in S1 Text). Structures were drawn using Marvin Sketcher (Chemaxon) and exported to SMILES format. This database was then imported to R and processed similarly to the MCS clustering, though instead of calculating overlapping scores between compounds, the overlapping score was determined for each compound against all ZBGs in the database. Only those compound-ZBG pairs where overlap was complete (score = 1 and, hence, ZBG completely contained in the lead compound) were counted as a match.
|
10.1371/journal.ppat.1007394 | Poly(ADP-ribose) polymerase 1 is necessary for coactivating hypoxia-inducible factor-1-dependent gene expression by Epstein-Barr virus latent membrane protein 1 | Latent membrane protein 1 (LMP1) is the major transforming protein of Epstein-Barr virus (EBV) and is critical for EBV-induced B-cell transformation in vitro. Poly(ADP-ribose) polymerase 1 (PARP1) regulates accessibility of chromatin, alters functions of transcriptional activators and repressors, and has been directly implicated in transcriptional activation. Previously we showed that LMP1 activates PARP1 and increases Poly(ADP-ribos)ylation (PARylation) through PARP1. Therefore, to identify targets of LMP1 that are regulated through PARP1, LMP1 was ectopically expressed in an EBV-negative Burkitt’s lymphoma cell line. These LMP1-expressing cells were then treated with the PARP inhibitor olaparib and prepared for RNA sequencing. The LMP1/PARP targets identified through this RNA-seq experiment are largely involved in metabolism and signaling. Interestingly, Ingenuity Pathway Analysis of RNA-seq data suggests that hypoxia-inducible factor 1-alpha (HIF-1α) is an LMP1 target mediated through PARP1. PARP1 is acting as a coactivator of HIF-1α-dependent gene expression in B cells, and this co-activation is enhanced by LMP1-mediated activation of PARP1. HIF-1α forms a PARylated complex with PARP1 and both HIF-1α and PARP1 are present at promoter regions of HIF-1α downstream targets, leading to accumulation of positive histone marks at these regions. Complex formation, PARylation and binding of PARP1 and HIF-1α at promoter regions of HIF-1α downstream targets can all be attenuated by PARP1 inhibition, subsequently leading to a buildup of repressive histone marks and loss of positive histone marks. In addition, LMP1 switches cells to a glycolytic ‘Warburg’ metabolism, preferentially using aerobic glycolysis over mitochondrial respiration. Finally, LMP1+ cells are more sensitive to PARP1 inhibition and, therefore, targeting PARP1 activity may be an effective treatment for LMP1+ EBV-associated malignancies.
| Epstein-Barr virus (EBV) is one of the most ubiquitous human viruses, with over 90% of adults worldwide harboring lifelong latent EBV infection in a small fraction of their B-lymphocytes. EBV is known to cause lymphoproliferative disorders and is associated with several other types of cancer, including Hodgkin's lymphoma, Burkitt's lymphoma and Nasopharyngeal carcinoma. However, in most cases, the approach to EBV-positive lymphomas does not differ from EBV-negative lymphomas of the same histology. Latent membrane protein 1 (LMP1) is the major transforming protein of EBV and is critical for EBV-induced B-cell transformation in vitro. LMP1 activates several epigenetic regulators to modify host gene expression, including the chromatin-modifying enzyme Poly(ADP-ribose) polymerase 1, or PARP1. In the current study we have determined that LMP1 can activate PARP1 to increase hypoxia-inducible factor 1-alpha (HIF-1α)-dependent gene expression, leading to a change in host cell metabolism indicative of a ‘Warburg effect’ (aerobic glycolysis). This subsequently provides a proliferative advantage to LMP1-expressing cells. The LMP1-induced increase in HIF-1α-dependent gene expression, alteration of cellular metabolism, and accelerated cellular proliferation, can be offset with the PARP inhibitor olaparib. Therefore, targeting PARP1 activity may be an effective treatment for LMP1+ EBV-associated malignancies.
| The Epstein-Barr virus (EBV) is a human gammaherpesvirus that latently infects approximately 95% of the population worldwide [1]. Latent EBV infection causes lymphoproliferative disease in immunosuppressed patients and is associated with Burkitt’s lymphoma and nasopharyngeal carcinoma [2, 3]. Following infection in epithelial cells, EBV often initially establishes a latent type III infection in naive B cells, where it expresses its full repertoire of latency genes. Expression of these genes within infected B cells drives proliferation and differentiation by triggering intracellular signals which mimic antigenic stimulation [4]. Type III latency genes include the six Epstein–Barr nuclear antigens (EBNAs 1, 2, 3A, 3B and 3C and EBNA leader protein (EBNA-LP)), latent membrane proteins LMP1 and LMP2 (which encodes two isoforms, LMP2A and LMP2B) and the non-coding EBV-encoded RNAs (EBER1 and EBER2) and viral microRNA (miRNA) [5].
During various stages of B cell differentiation in vivo, EBV will express either the latency III program, or one of two alternative forms of virus latency (known as latency I and latency II). Expression of the large set of EBV genes in latency III is highly immunogenic and eventually leads to the implementation of a limited gene expression profile (type I latent gene expression program) [3, 6], with only Epstein–Barr nuclear antigen 1 (EBNA1) expressed. EBNA1 is essential for viral episomal maintenance and replication [7] and allows the EBV-infected host cell to evade detection by the immune system [8].
Specific EBV-associated malignancies are associated with different latency types [3, 6]. Therefore, understanding EBV gene regulation during latency and latency switching will provide fundamental new insights into the development of novel, targeted treatments against EBV-associated malignancies. In particular, there is an unmet need for the specific targeting of EBV-positive lymphomas, as in most instances the approach to EBV-positive lymphomas does not differ from EBV-negative lymphomas of the same histology [9].
ADP-ribosylation is a post-translational modification where single units (mono-ADP-ribosylation) or polymeric chains (poly-ADP-ribosylation) of ADP-ribose are conjugated to proteins by ADP-ribosyltransferases [10]. This post-translational modification by the ADP-ribosyltransferases (also known as PARPs) plays a key role in a variety of nuclear processes including transcriptional regulation via epigenetic mechanisms [11–14], and direct histone modification [15, 16]. PARylation of histones reduces their affinity for DNA due to electrostatic repulsion [13], allowing greater accessibility to DNA repair or transcriptional machineries [13, 17, 18]. The host also uses PARylation, specifically through the PARP1 protein, to regulate both the lytic and latent infection of EBV [19–21].
Our group has previously shown that viral gene products can also influence PARylation, and that disruption of PARP regulation is sufficient to alter host gene expression. In that study, the relationship between EBV latency type and PARylation was explored, and type III cells latently infected with EBV were determined to have significantly higher PAR levels than type I latently infected EBV cells [22]. Expression of the type III latency-associated EBV protein Latent membrane protein 1 (LMP1) alone was sufficient to promote PARP1-mediated PARylation [22]. LMP1 is the major transforming protein of EBV and is critical for EBV-induced B-cell transformation in vitro [23, 24].
As LMP1 alone was sufficient to promote PARP1-mediated PARylation, we are reporting here an unbiased approach to identify global targets of LMP1 that are regulated through PARP1. In this approach, LMP1 was ectopically expressed in an EBV-negative Burkitt’s lymphoma cell line DG75. These LMP1-expressing cells were then treated with the PARP inhibitor olaparib and prepared for RNA sequencing. The LMP1/PARP targets identified through this RNA-seq experiment are largely involved in metabolism and signaling. Interestingly, Ingenuity Pathway Analysis, IPA, of RNA-seq data suggests that the transcription factor hypoxia-inducible factor 1-alpha (HIF-1α) is an LMP1 target mediated through PARP1. Dysregulation and overexpression of HIF-1α due to hypoxia or genetic alternations are heavily implicated in oncogenesis, as well as several other pathophysiologies, involving vascularization and angiogenesis, energy metabolism, cell survival, and tumor invasion [25].
Transcriptionally active HIF-1 is a heterodimer composed of α- and β-subunits. The dimer is a member of the basic helix loop helix-PER-ARNT-SIM (bHLH-PAS) family of transcription factors which play a role in cancer development [26]. In normal, non-hypoxic cells, HIF-1α is continually synthesized and degraded, while HIF-1β is constitutively expressed to relatively constant levels within the nucleus. HIF-1α degradation is initiated by hydroxylation of a proline residue (Pro-402 and/or Pro-564) by prolyl hydroxylases (PHD-1, PHD-2, and PHD- 3) using molecular oxygen as a co-substrate [27, 28]. Upon hydroxylation, HIF-1α- OH becomes ubiquitinated by the von Hippel Lindau E3 ubiquitin ligase protein (VHL), and subsequent proteasomal breakdown occurs. In low oxygen, PHDs cannot function, resulting in stabilization of HIF-1α in the cytoplasm and its translocation to the nucleus [29].
Interestingly, several human oncogenic viruses increase levels of the transcription factor HIF-1, including EBV [30]. Specifically, LMP1 was shown to enhance the synthesis of HIF-1α and the expression of HIF-1α-responsive genes in a nasopharyngeal carcinoma (NPC)-derived cell line [31], which could be attributed to enhanced degradation of prolylhydroxylases (PHD) 1 and 3 mediated by SIAH1 [32]. More recent work illustrates that infection of full length EBV increases HIF-1α protein levels and its translocation to the nucleus in comparison to normal cytokine-induced proliferating B cells. EBNA-3 and EBNA-LP were shown to bind directly to PHD-2 and PHD-1, respectively, preventing HIF-1α hydroxylation and consequently allowing it to escape degradation [33]. In addition, PARP1-deficient chronic myelogenous leukemia cells showed reduced HIF-1 transcriptional activation dependent on PARP1 enzymatic activity. PARP1 was found to complex with HIF-1α through direct protein interaction and increased HIF-1α–dependent gene expression [34].
We report here that PARP inhibition offsets LMP1-mediated gene activation. Specifically, we determined that LMP1 can modulate host gene expression by using PARP1 as a coactivator of HIF-1α-dependent gene expression in B cells. PARP1 directly co-activates HIF-1α–dependent gene expression by binding to the promoter regions of HIF-1α targets. Many of these HIF-1α–dependent gene targets are involved in metabolism, and consequently LMP1+ cells are much less dependent on mitochondrial respiration and instead use aerobic glycolysis, conferring a ‘Warburg effect’/aerobic glycolysis (high rate of glycolysis followed by lactic acid fermentation even in the presence of abundant oxygen) [35]. Finally, LMP1+ cells are more sensitive to PARP1 inhibition and therefore targeting PARP1 activity may be an effective treatment for LMP1+ EBV-associated malignancies.
To identify global targets of LMP1 regulated by PARP1, LMP1 was ectopically expressed in the EBV-negative Burkitt’s lymphoma cell line DG75 (S1A Fig). Cells were transduced with retroviral particles containing either pBABE (empty vector) or pBABE-HA-LMP1 vectors. Transduced cells were placed under long-term selection in medium containing 1 μg/ml puromycin and LMP1 expression was confirmed by western blotting, which showed physiological protein levels as observed in latency type III cell lines (S1B Fig). Previously we have demonstrated that expression of the type III latency-associated EBV protein LMP1 alone was sufficient to promote PARP1-mediated PARylation [22], and this was also observed following ectopic expression of LMP1 in DG75 (S1D Fig). LMP1 positive (+) and LMP1 negative (-) cells were incubated for 72 hrs with 1 μM of the PARP inhibitor olaparib or the DMSO vehicle as a control. RNA was then isolated and prepared for RNA sequencing. We observed that the expression of 2504 genes were significantly changed (FDR<0.01) when comparing LMP1- vs LMP1+ cells, with 1578 and 926 genes upregulated and downregulated by LMP1, respectively (S2A and S2B Fig). Ingenuity Pathway Analysis (IPA) predicted HIF-1α as one of the top upstream regulators activated by LMP1 (S2D Fig). Furthermore, gene function analysis identified pathways such as glycolysis I, gluconeogenesis I, Notch signaling and B cell development to be upregulated by LMP1 (S2C Fig). Inspection of regulated genes and IPA analysis showed well-known targets of LMP1 that have been reported in prior literature, confirming that ectopic expression in DG75 could recapitulate the changes in gene expression induced by LMP1.
We then compared untreated LMP1+ cells with LMP1+ cells treated with the PARP inhibitor olaparib. In total, we observed expression of 2435 genes to be significantly changed (FDR<0.01), with balanced up and downregulation following PARP inhibition (1163 and 1272 genes, respectively) (S3A and S3B Fig). In contrast to IPA predicted HIF-1α activation by LMP1, olaparib treatment is predicted to inhibit HIF-1α in LMP1+ cells (S3D Fig). Gene function analysis also identified regulation of pathways such as glycolysis I and gluconeogenesis I by PARP1 (S3C Fig).
We then overlaid the aforementioned two datasets and introduced log2 I1I Fold Change to identify our ‘LMP1/PARP1’ targets, of which there were 292 (Fig 1A). Of these 292 genes, the majority (225) were upregulated by LMP1 and offset by PARP1 inhibition (Fig 1B). We performed unsupervised hierarchical clustering and observed that the LMP1+ samples treated with olaparib and the LMP1- samples clustered together and separately from the LMP1+ untreated samples. We observed that two clusters emerged among the LMP1/PARP1 targets, which were analyzed by IPA gene function analysis. Cluster 1 genes were upregulated by LMP1 and downregulated following PARP1 inhibition, while cluster 2 genes were downregulated by LMP1 and upregulated following PARP1 inhibition (Fig 1C). IPA revealed PARP1/LMP1 targets were largely involved in metabolism and signaling, with two clusters emerging from gene function analysis (Fig 1D). In addition, disease or function analysis identified cancer, proliferation of lymphatic system, and proliferation of lymphocytes as LMP1/PARP1 targets that were decreased following olaparib treatment (Fig 1E).
IPA identified HIF-1α, as well as its dimerization partner ARNT (HIF-1B), as top upstream regulators activated by LMP1/PARP1 and repressed following PARP inhibition (Fig 1F). This was based on increased transcription of HIF-1α-targets by LMP1 and their downregulation following PARP inhibition (Fig 2A). We validated several of these HIF-1α targets by qRT-PCR in both the DG75 cell line (fold change LMP1+/LMP1-) (Fig 2B) as well as EBV infected cells with latency III and I setting (fold change Mutu III/I) (S5E Fig). To establish that the inhibition of HIF-1α targets was due to PARP1 inhibition rather than off-target effects of olaparib, PARP1 was knocked down in LMP1+ and LMP1- DG75 cells (Fig 2C and 2D). Corresponding to PARP1 inhibition with olaparib, HIF-1α targets were upregulated in LMP1 + cells vs LMP1 –cells and this upregulation was diminished by PARP1 knockdown, as shown by qRT-PCR (fold change LMP1+/LMP1-) (Fig 2E), indicating that PARP1 is necessary for activation of these genes by LMP1.
It has been reported in the literature that PARP1 forms a complex with HIF-1α through direct protein interaction and increases HIF-1α–dependent gene expression [34]. To see if this was the case in our B cell lines, we performed an immunoprecipitation assay and found that HIF-1α immunoprecipitated with PARP1. We also observed that the HIF-1α/PARP1 interaction was increased in LMP1+ cells (around 40%) and PARP1 inhibition caused dissociation of the complex (Fig 3A and 3B). Whilst this LMP1-induced global increase in HIF-1α/PARP1 interaction was modest, we observed much greater increases in LMP1-induced PARP/HIF-1α binding at specific HIF-1α-responsive gene promoters (see below).
As there is an increase in PARP1 activity and HIF-1 transcriptional activation in LMP1+ cells, and inhibition of PARP1 catalytic activity reduces HIF-1 transcriptional activation, we wanted to determine if the PARP1/HIF-1α complex was PARylated in LMP1+ cells. As shown in Fig 3C and 3D, following incubation with Poly-ADP-ribose binding macrodomain resin, western blot for HIF-1α and PARP1 confirms that the PARP1/HIF-1α complex is PARylated. Specifically, LMP1+ cells exhibited a two-fold increase in HIF-1α and PARP1 levels, respectively, compared to LMP1- cells following pull down with the Poly-ADP-ribose binding macrodomain resin (Fig 3D). Biological replicates of the IP and PAR resin assays are shown in S6 Fig. This suggests that PARylation of HIF-1α, or proteins bound to HIF-1α in a complex, may play a role in the stability of the complex as well as the increased transcriptional activation of HIF-1α in LMP1+ cells.
To determine if increased PARP activation in LMP1+ cells was augmenting HIF-1 transcriptional activation by influencing HIF-1 binding to its downstream promoters, we performed ChIP-PCR experiments on promoter regions of validated HIF-1α targets. These targets have been validated by RT-qPCR and had demonstrated increased transcription in LMP1+ cells vs LMP1- cells and decreased transcription in LMP1+ cells, following both PARP1 inhibition and PARP1 knockdown. Promoter regions of three such HIF-1α targets were bound by PARP1 and HIF-1α considerably more in LMP1+ cells vs LMP1- cells. Furthermore, binding of HIF-1α and PARP1 was reduced at promoter regions of HIF-1α targets by PARP1 inhibition in LMP1+ cells (Fig 4A and 4B). One exception was at the BNIP3 promoter, where no loss of HIF-1α binding following PARP1 inhibition was observed. Therefore, in the case of BNIP3, it may be that despite HIF-1α binding, the HIF-1α/PARP complex is less active and less stable following PARP inhibition (as shown by IP data and loss of PARP1 binding to BNIP3 promoter), which results in the decreased gene expression observed. This leads to the speculation that the presence of PARP1 at the promoter may be the determining factor for activation of HIF-1-responsive gene expression in a subset of HIF-1-responsive genes. However, after ChIP-PCR experiments with EBV infected cells with latency III and I setting (Mutu III/I) (S5B Fig), we did observe loss of HIF-1α binding at the BNIP promoter following PARP1 inhibition. Thus, it may simply be a cell line specific response.
As shown by the previously discussed ChIP-qPCR experiments, PARP1 is present at the promoters of HIF-1 α–dependent genes. Due to the multiple roles PARP1 can play as a chromatin modifying enzyme [11–14], we wanted to determine if the increased PARP1 binding at the promoter regions of the HIF-1α targets was due to a change in the chromatin landscape of the regions. As shown in Fig 4C and S5C Fig, these targets also had significant accumulation of the positive histone mark H3K27ac. Furthermore, this mark could be lost by PARP1 inhibition, which conversely led to the accumulation of the repressive histone mark H3K27me3 (Fig 4D and S5D Fig). This suggests that the role of PARP1 as a coactivator of HIF-1 α–dependent gene expression could be attributed to its ability to modify histone tails, creating a more permissible environment for gene transcription.
PARP1 and PARylation can affect the ability of proteins to interact with chromatin, therefore we determined whether the activation of PARP1 by LMP1 can influence the association of HIF-1α with chromatin and whether PARP inhibition could reverse this effect. We assessed HIF-1α levels in the cytoplasmic fraction, the nuclear soluble fraction and chromatin-bound fraction by western blot and following subcellular protein fractionation. Western blot for HIF-1α confirms its localization to chromatin in LMP1+ cells, which is reduced after olaparib treatment (Fig 5A). Specifically, we observed a 50% increase in chromatin-bound HIF-1α in LMP1+ cells vs LMP1- cells, which was reduced to 60% of LMP1- levels following PARP inhibition (Fig 5B). This global increase in chromatin bound HIF-1α in LMP1+ cells further suggests LMP1 enhancing HIF-1α transcriptional activation.
Many of the HIF-1α downstream transcriptional targets activated by LMP1 through PARP1 are involved in metabolism, therefore we aimed to determine if LMP1/PARP1interaction lead to any functional metabolic effect at the cellular level. To examine this, we performed mito stress test and glycolytic rate assays using a XF96 Extracellular Flux Analyzer (Seahorse Bioscience) to measure oxygen consumption rate (OCR) and extracellular acidification rate (ECAR). For the mito stress test, OCR and ECAR were detected under basal conditions followed by the sequential addition of oligomycin, fluoro-carbonyl cyanide phenylhydrazone (FCCP) and rotenone + antimycin A. As shown in Fig 6B, mitochondrial respiration is significantly decreased in LMP1+ cells. PARP1 inhibition in these cells subsequently leads to an increase in mitochondrial respiration (Fig 6C). This suggests that LMP-mediated activation of PARP1 leads to decreased reliance on mitochondrial respiration in the cell. PARP1 activation has been shown to damage mitochondrial activity characterized by secondary mitochondrial superoxide production, distorted mitochondrial structure and reduced mitochondrial oxidation and ATP production [36]. This can be seen by the decreased ATP synthase-linked ATP production in LMP1+ cells followed by increase in ATP levels after PARP inhibition (Fig 6D). In the LMP1- cells, we observed an increase in basal respiration upon olaparib treatment, similar to that seen in LMP1+/+ olaparib group. However, olaparib treatment in the LMP1- cells resulted in a decrease in maximal respiration (S10A Fig). We think the differences observed in the maximal respiration was due to the contrast in PARP1 activation states between LMP1- and LMP1+ cells and the resulting disparity in olaparib sensitivity between the two (discussed further below).
Apart from Mitochondrial respiration, the other major cellular energy pathway is glycolysis. Due to the decreased reliance on mitochondrial respiration by LMP1, and knowing that HIFs activate transcription programs which induce glycolysis and inhibit mitochondrial activity [37], we wanted to determine if LMP1 promotes a switch to glycolytic metabolism. To accomplish this, we used the glycolytic rate test procedure to measure the OCR and ECAR. Both were detected under basal conditions followed by the sequential addition of 2μM rotenone + 2 μM antimycin A and 50 mM 2-deoxy-D-glucose. As shown in Fig 7B, 7D and 7E, LMP1 confers a ‘Warburg’ effect, significantly increasing basal and compensatory glycolysis in the cell under aerobic conditions. PARP inhibition subsequently decreased this effect (Fig 7C, 7D and 7E) but had no impact on LMP1- cells (S10B Fig). This suggests that LMP-mediated activation of PARP1 not only leads to diminished reliance on mitochondrial respiration, but also to an increase in aerobic glycolysis. How much of this is mediated distinctly through PARP1, or HIF-1α, or a combination of the two, needs to be elucidated with further experimentation.
Warburg metabolism is thought to enable rapid cell division through the creation of excess carbon obtained from increased glucose consumption, which can subsequently be used to fuel anabolic processes. This excess carbon can then be diverted into the various branching pathways that stem from glycolysis and subsequently used for the production of nucleotides, lipids, and proteins [38]. Activated T cells extensively and rapidly proliferate upon activation and have been shown to engage Warburg metabolism [38, 39]. B cells share certain fundamental metabolic characteristics with T cells, such as increased glucose uptake and induction of glycolysis after activation [40, 41].
As LMP1 appears to be engaging ‘Warburg metabolism’, and our IPA analysis suggested increased proliferation of cells with LMP1 (Fig 8A), we wanted to determine if this conferred a proliferative advantage. To ascertain this, we measured cellular proliferation by staining cells with CFSE (5(6)-Carboxyfluorescein N-hydroxysuccinimidyl ester) staining. CFSE Uptake at time zero was the same for both LMP1+ and LMP1- cells (S8B Fig). We then allowed cells to proliferate for 96 hrs before proceeding with FACS analysis. LMP1 presence led to increased proliferation vs LMP1- cells (Fig 8B), which was markedly curtailed following PARP1 inhibition (Fig 8C). In contrast, proliferation of LMP1- cells was only marginally reduced following PARP inhibition (S4A Fig). This olaparib-induced decrease in proliferation in LMP1+ cells coincided with in an arrest in G2/M (Fig 8D) but appeared to be independent of DNA damage, as we found no evidence of yH2A.x accumulation following intracellular staining and FACS analysis (S1E Fig). Furthermore, we found no evidence of PARP inhibition (1 μM 72 hrs) leading to apoptotic cell death, as determined by Annexin V staining (S1F Fig).
We then used the methylcellulose colony forming cell (CFC) assay to determine the impact of LMP1 and PARP inhibition on the cells’ ability to proliferate and differentiate into colonies. Cells were pre-treated with 2.5 μM olaparib for 96 hrs. Following this pre-treatment, cells were checked for viability using the Annexin V assay (S8A Fig). After confirmation of cell viability, cells were seeded and incubated in CFC media for 14 days. As shown by Fig 8E and 8F, LMP1 enabled cells to form robust colonies. However, colonies were not able to form following olaparib treatment.
We report here that LMP1 can modulate host gene expression by using PARP1 as a coactivator of HIF-1α-dependent gene expression in B cells. In recent decades, research into PARP biology, outside of its classical role in DNA damage detection and repair responses, has led to greater appreciation and understanding to the pivotal role PARP-1 plays in gene regulation. PARP-1 can function as a key regulator of gene expression through a variety of mechanisms, including roles as a chromatin modulator, a coregulator for DNA-binding transcription factors, and a regulator of DNA methylation. The gene regulatory effects of PARP-1 have been linked to the control of inflammation, metabolism, circadian rhythms, and cancer [42]. Previous work by our group has established that expression of the type III latency-associated EBV protein LMP1 alone was able to promote PARP1-mediated PARylation, and disruption of inhibition of PARP activity was sufficient to alter host gene expression. Moreover, the induction of PARylation mediated by LMP1 was also essential for EBV-driven oncogenesis [22]. Building on our previous work, here we are reporting a global approach to identify host gene targets of LMP1 that are regulated through PARP1. Greater understanding of how LMP1 is able to manipulate the host gene regulatory machinery through chromatin-modifying enzymes, such as PARP1, may be exploited by therapeutic intervention to better treat EBV-positive cancers.
Our initial analysis of RNA-seq data suggested that the transcription factor Hypoxia-Inducible Factor 1-alpha (HIF-1α) is an LMP1 target mediated through PARP1. There is strong evidence that activation of HIF-1 is a common pathway affected by human oncogenic viruses [30] and HIF-1's role in the transcriptional upregulation of metabolic, angiogenic and microenvironmental factors is integral for oncogenesis [30]. HIF-1α transcription is continual and several growth factors and their accompanying pathways have been shown to play a role in enhancing HIF-1α signaling in an oxygen-independent manner. However, the majority of work surrounding HIF-1 regulation has been focused on its constitutive normoxic protein breakdown and how this can be subverted in the context of oncogenesis. LMP1 has been shown to increase the synthesis of HIF-1α through the ERK1/2 MAPK signaling pathway [31] and decrease its breakdown through the degradation of PHD 1 and 3, mediated by SIAH1 [32].
Our work does not find any significant evidence of LMP1 increasing HIF-1α protein or mRNA levels. It should be noted however, that our work has taken place in B cells, with the above-mentioned work mainly taking place in epithelial cells and the latter study involved full length EBV infection. Another key difference is our use of a Burkitt’s lymphoma cell line (DG75) which carries a MYC translocation. Overexpression of Myc has been reported to stabilize the α subunit of HIF1 (HIF-1α) under normoxic conditions and enhance HIF-1α accumulation under hypoxic conditions [43]. Therefore, a potentially higher basal level of HIF-1α in our cell lines could have dampened the effects of ectopic LMP1 expression.
Instead, our evidence indicates that PARP1 is acting as a coactivator of HIF-1α-dependent gene expression in B cells, and this co-activation is enhanced by LMP1-mediated activation of PARP1. Outside of EBV, a similar mechanism was reported in PARP1-deficient chronic myelogenous leukemia cells, which showed reduced HIF-1 transcriptional activation dependent on PARP1 enzymatic activity. This agrees with our observations, as inhibition of PARP1 catalytic activity reduced transcriptional activation of HIF-1 targets. The authors of this study also demonstrated PARP1 forming a complex with HIF-1α through direct protein interaction in vitro, as well as endogenously in HeLa cells [34]. Our study adds to the scope of this mechanism by demonstrating that this complex is also PARylated, and that PARP1 inhibition not only leads to loss of PARylation but also destabilization of the complex. Whether only PARP1 is PARyated, or also other factors, such as HIF-1α or histones, has to be further elucidated, which we are planning to achieve in the coming months. What is clear is the requirement of PARP1 enzymatic activity for HIF-1–dependent transcription and the stability of the PARP1/HIF complex, presumably due to the proper scaffolding of PAR polymers by PARP1.
We then demonstrated, through ChIP assays, that PARP1 co-activates HIF-1α–dependent gene expression by binding to the promoter regions of HIF-1α targets, which adds to the PARP1/HIF-dependent gene expression studies assessed by transient transfection of a reporter gene under the control of hypoxia response element [34]. Here we show that promoter regions of HIF-1α targets are bound by HIF-1α and PARP1 considerably more in LMP1+ cells vs LMP1- cells, and the binding of both proteins is significantly reduced following PARP1 inhibition.
Our ChIP experiments also revealed that LMP1 induction led to significant accumulation of the positive histone mark H3K27ac at HIF-1α–dependent genes. This is interesting as one of the key coactivators of HIF-1 is the histone acetyltransferases p300, which can directly associate with the COOH-terminal transactivation domain of HIF-1α [44] and facilitate acetylation of histone H3 at 'lysine 27' (H3K27ac) [45]. Furthermore, this mark was lost following PARP1 inhibition, which conversely led to the accumulation of the repressive histone mark H3K27me3. Previous work by our group has demonstrated that in the absence of DNA damage, both pharmacological inhibition of PARP and knockdown of PARP1 induced the expression of the polycomb repressive complex 2 (PRC2) member EZH2, which mediates the trimethylation of histone H3 at lysine 27 (H3K27me3). This resulted in increased global H3K27me3, with ChIP assays confirming PARP1 inhibition led to H3K27me3 deposition at EZH2 target genes, resulting in gene silencing [12]. Ensuing work found that EZH2 is a direct target of PARP1 upon induction of alkylating and UV-induced DNA damage in cells and in vitro. PARylation of EZH2 inhibits EZH2 histone methyltransferase (H3K27me) enzymatic activity [46, 47]. This lends to the possibility that one of the roles of PARP1, as a coactivator of HIF-1α–dependent gene expression, could be down to its ability to modify histone tails to augment HIF-1α–dependent gene expression. Specifically, the role of PARP1 could be to PARylate EZH2 and inhibit EZH2 histone methyltransferase (H3K27me) enzymatic activity. This may then allow the histone acetyltransferases p300, a key coactivator of HIF-1, to facilitate acetylation of histone H3 at 'lysine 27' (H3K27ac), creating a more permissible environment for gene transcription at HIF-1 transcriptional targets.
PARP1 and HIF-1α occupy prominent positions in mitochondrial homeostasis and metabolism and EBV–transformed B cells have been shown to induce a ‘Warburg effect’ [33, 48, 49]. In addition, LMP1 has been shown to be the key regulator in reprogramming of EBV-mediated glycolysis in NPC cells [50, 51]. Many of the LMP1-induced PARP1/HIF-1α transcriptional targets we identified in our data set are involved in metabolism, and therefore we wanted to determine if PARP1 was required for the LMP1-induced aerobic glycolysis that we observed. We determined that LMP1 significantly increased glycolysis and decreased mitochondrial respiration, and this switch in metabolism appeared to be mediated through PARP1. This isn’t too surprising, as the majority of research points to PARP activation damaging mitochondrial function, while PARP inhibition has the opposite effect [52–54]. For example, we observed significant ATP loss in LMP1+ cells followed by recovery with PARP1 inhibition, as estimated by the mito stress test assay. AMP concentrations can be increased by PAR degradation, and AMP perturbs mitochondrial ADP/ATP exchange [55]. In addition, HIFs are well-known to activate transcription programs that induce glycolysis and inhibit mitochondrial activity [37]. For instance, enzymes catalyzing glucose metabolism—including phosphoglycerate kinase 1 (PGK1) and phosphofructokinase (PFK), are well-established targets of HIF-1 [56] and were identified from our RNA-seq data as being induced by LMP1/PARP1. PDK1 was also identified from our dataset, another recognized HIF-1 target and a key enzyme that contributes to the ‘Warburg effect’ [35].
Previous studies point to the NF-kB signaling pathway and glucose transporter-1 (GLUT1) as being key mediators in the activation of aerobic glycolysis in LMP1+ NPC cell lines and both EBV and spontaneous B-cell lymphomas [48, 49, 51]. While we didn’t find any evidence of increased transcription of GLUT1, it is possible that LMP1 may be inducing glucose transporter-1 (GLUT1) membrane trafficking, as was observed in EBV and spontaneous B-cell lymphomas [49]. Regarding the NF-kB signaling pathway driven aerobic glycolysis, this may be happening upstream of PARP1/HIF-1α -dependent gene expression. Firstly, there is evidence that PARP1 can act as a coactivator of NF-kB in vivo [57], which is supported by our IPA analysis which identified NF-kB as the highest scoring upstream regulator predicted to be activated by LMP1 and inactivated by PARP1, as was seen with HIF-1α (S7 Fig). Secondly, there is evidence of significant crosstalk between the NF-kB and HIF-1α pathways. NF-κB has been shown to be a direct modulator of HIF-1α expression. Specifically, the HIF-1α promoter has been demonstrated to be responsive to selective NF- κB subunits [58].
In summary, our work adds an important branch to the existing model of how LMP1 affects cellular functions through modulation of chromatin modifying enzymes to regulate host gene expression, specifically through PARP1 and PARylation (Fig 9). One remaining question is how LMP1 communicates with PARP1, which we will address in the coming months by determining whether LMP1 regulates PARP1 through direct interaction or via of the signaling pathways that LMP1 activates. Gaining a better insight into the LMP1/PARP1 interaction will reveal new important functions of LMP1 in the context of cellular transformation and EBV infection.
All cells were maintained at 37°C in a humidified 5% CO2 atmosphere in medium supplemented with 1% penicillin/streptomycin antibiotics. Lymphocyte cell lines (EBV-negative Burkitt’s lymphoma cell line DG75 ATCC CRL-2625 (DG75), EBV-positive latency III cell lines Mutu III, Mutu-LCL, KEM III, Raji, GM12878, GM13605 and EBV-positive latency I cell line Mutu I) were cultured in suspension in RPMI 1640 supplemented with fetal bovine serum at a concentration of 15%. 293T ATCC CRL-3216 (HEK 293T) cells were cultured in Dulbecco’s modified Eagle medium (DMEM) supplemented with fetal bovine serum at a concentration of 10%. Olaparib (Selleck Chemical) was dissolved in dimethyl sulfoxide (DMSO), and cells were treated for upon dilution in the appropriate media.
Cellular poly(ADP-ribose) (PAR) levels were quantified using a PARP in vivo pharmacodynamic assay 2nd generation (PDA II) kit (Trevigen) according to the manufacturer’s protocol. Briefly, cells were lysed in the supplied buffer, and protein concentration was determined with a bicinchoninic acid (BCA) protein assay (Pierce). Cell extracts were added to a precoated capture antibody plate, incubated overnight at 4°C, and washed four times with phosphate-buffered saline containing 0.05% Tween 20 (PBST). A polyclonal antibody for the detection of PAR was added, and the plate was incubated at room temperature for 2 h. After washing with PBST, the plate was incubated for 1 h goat anti-rabbit IgG-horseradish peroxidase (HRP). The wells were washed again with PBST before the addition of PARP PeroxyGlow reagent. Luminescence was then measured using a POLARstar Optima microplate reader (BMG Labtech).
Cell lysates were prepared in radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.25% deoxycholic acid, 1% NP-40, 1 mM EDTA) supplemented with 1X protease inhibitor cocktail (Thermo Scientific). Protein extracts were obtained by centrifugation at 3,000×g for 10 minutes at 4°C. For nuclear fractionation, nuclear soluble and chromatin-bound protein fractions were extracted from cells using the Subcellular Protein Fractionation Kit for Cultured Cells kit (Invitrogen) according to manufacturer’s instructions. The bicinchoninic (BCA) protein assay (Pierce) was used to determine protein concentration. Lysates were boiled in 2x SDS-PAGE sample buffer containing 2.5% β-mercaptoethanol, resolved on a 4 to 20% polyacrylamide gradient Mini-Protean TGX precast gel (Bio-Rad), and transferred to an Immobilon-P membrane (Millipore). Membranes were blocked for 1 h at room temperature and incubated overnight with primary antibodies recognizing LMP1 (Abcam ab78113), PARP1 (Active Motif 39559), HIF-1α (Abcam ab1) and Actin (Sigma A2066), as recommended per the manufacturer. Membranes were washed, incubated for 1 h with the appropriate secondary antibody, either goat anti-rabbit IgG-HRP (Santa Cruz sc-2030) or rabbit anti-mouse IgG-HRP (Thermo Scientific 31430). Membranes were then washed and detected by enhanced chemiluminescence.
For immunoprecipitation, 5×106 cells were used per IP. Cells were re-suspended in 1 mL RIPA buffer and the protein extracts were obtained by centrifugation at 3,000×g for 10 minutes at 4°C. The supernatant was then incubated with 5 μg of indicated antibodies overnight at 4°C followed by incubation with 100 μL 50% Protein A/G magnetic beads (ThermoFisher). After 2 hours’ incubation, the beads were washed three times with RIPA Buffer and then re-suspended in Laemmli buffer followed by analysis by SDS-PAGE and western blotting.
For PAR pulldown, 5×106 cells were re-suspended in 1 mL of PAR Lysis buffer [50 mM Tris, pH 8, 200 mM NaCl, 1 mM EDTA, 1% Triton X-100, 10% glycerol, 1 mM DTT, 0.5% deoxycholate, 1X protease inhibitors (Thermo Scientific), 1 μM ADP-HPD (Adenosine 5'-diphosphate (hydroxymethyl) pyrrolidinediol) (EnzoLifesciences)] and put on a rotating device for 2 hours at 4°C. Protein were then extracted by centrifugation at 3000xg for 5 minutes at 4°C. 500 μL of the protein extracts were then incubated with 20 μL (20 μg) of either Poly-ADP-ribose Affinity resin (Tulip BioLabs, 2302) or Poly-ADP-ribose Negative Control Resin (Tulip BioLabs, 2303). PAR Affinity resin is a purified GST-Af1521 macrodomain fusion protein construct. The Af1521 macrodomain has been shown to bind with high affinity to polymeric ADP-ribose modified proteins. The PAR Negative Control resin is identical to the PAR positive except that it contains a mutated Af1521 macrodomain that is unable to bind PAR. After overnight incubation at 4°C on a rotating device, beads were washed three times with PAR Lysis buffer and re-suspended in 80 μL Laemmli buffer, followed by incubation at 65°C for 15 minutes to dissociate the macrodomain fusion protein from the affinity-precipitated proteins. 30 μL of purified PARylated proteins were then analyzed by SDS-PAGE and immunoblotting.
shPARP1 and sh-non-effective scrambled plasmids were bought from Origene (TR315488 and TR30021). Lentiviral particles were generated by transfecting 293T cells with pLKO.1-shPARP1 or scrambled shRNA, the psPAX2 (plasmid number 12260; Addgene) packaging plasmid, and the pMD2.G envelope plasmid (plasmid number 12259; Addgene) according to the Addgene protocol. psPAX2 and pMD2.G plasmids were a gift from Didier Trono. DG75 cells were infected with two separate lentivirus expressing shPARP1 (Origene TR315488A-B), or the sh control vector freshly generated from 293T cells.
Plasmid constructs hemagglutinin (HA)- tagged full-length LMP1, pBABE, pVSV-G, and pGag/Pol were kindly provided by Nancy Raab-Traub (UNC, Chapel Hill, NC) and were described previously [59]. Retroviral particles were generated using the Fugene 6 reagent (Promega) to simultaneously transfect subconfluent monolayers of 293T cells with 1μg pBABE (vector) or HA-LMP1, 250 ng pVSV-G, and 750 ng pGal/Pol according to the manufacturer’s instructions. Supernatant containing lentivirus was collected at 48- and 72-h post-transfection and filtered through a 0.45 μM filter. DG75 cells were transduced by seeding 5x105 cells in 6-well plates in 500 μl medium and adding 500 μl of medium containing retroviral particles. The transduced cells were placed under long-term selection in medium containing 1 μg/ml puromycin.
Chromatin immunoprecipitation (ChIP) assays were performed according to the Upstate Biotechnology Inc. protocol as described previously, with minor modifications [22]. Briefly, cells were fixed in 1% formaldehyde for 15 min, and DNA was sonicated using a sonic dismembrator (Fisher Scientific) to generate 200–500-bp fragments. Chromatin was immunoprecipitated with polyclonal antibodies to PARP1 (Active Motif 39559), HIF-1α (Active Motif 39665), H3K27me3 (Active Motif 39155) and H3K27ac (Active Motif 39135). ChIP-grade protein A/G magnetic beads (Pierce) were used for immunoprecipitation with polyclonal antibody. Realtime PCR was performed with a master mix containing 1X Maxima SYBR Green, 0.25 μM primers and 1/50 of the ChIP DNA per well. Primers are available upon request. Quantitative PCR reactions were carried out in triplicate using the ABI StepOnePlus PCR system. Data were analyzed by the ΔΔCT method relative to DNA input and normalized to the IgG control.
RNA was extracted using a PureLink RNA Mini Kit (ThermoFisher) according to the manufacturer’s protocol. The polyadenylated transcript library used for transcriptome sequencing (RNA-seq) analysis was generated using an Epicentre (Illumina) mRNA-seq kit. Total RNA was depleted of the rRNA component using a RiboZero rRNA removal kit (Epicentre) and then processed with a ScriptSeq (version 2) kit along with ScriptSeq index PCR primers (Epi- centre) to generate a strand-specific library of mRNA. Single reads of 50 bp were obtained using an Illumina genome analyzer II. Sequencing reads were aligned to the human genome rn4 using the TopHat program [60], considering reads encoded across splice junctions (parameters were set to the default). The expression level of all RefSeq transcripts was evaluated using the Cufflinks program [61], and the number of fragments per kilobase of transcript per million fragments mapped (FPKM) was calculated for each transcript (the parameters were set to the default, and the hg19 RefSeq GTF table was used to define the transcripts). Differences in gene expression levels between samples were assessed by use of the Cuffdiff program and calculated as the log2 fold change. RNA-seq data were analyzed using Ingenuity pathway analysis (IPA; Qiagen, Redwood City, CA).
The RNA-seq data are accessible through GEO Series accession number GSE121476. The raw data files can be accessed using the following link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE121476
Following treatment, cells were washed twice with PBS and re-suspended in 500 μl of Annexin V-binding buffer (Abcam) and stained with Annexin V-FITC Apoptosis Detection Reagent (Abcam) and 250μg/mL propidium iodide (ThermoFisher) for five minutes in the dark. Flow cytometric analysis was carried out using a FACS Calibur flow cytometer (Becton Dickinson) and CellQuest software, and the cell population was analyzed using FlowJo software. Double positive Annexin V/PI cells were deemed to be the apoptotic population.
Cells were harvested, fixed and permeabilized in absolute ethanol and then incubated with 1mg/mL propidium iodide (PI) and 10mg/mL RNAse A for 30 mins at 37C. Flow cytometric analysis was then carried out using a FACS Calibur flow cytometer (Becton Dickinson) and CellQuest software, and cell cycle distribution was analyzed using FlowJo software.
500 cells, untreated and pre-treated with 2.5 μM olaparib for 96 hrs, were seeded in 1.4% human methylcellulose media (R and D Systems cat HSC002) and incubated for 14 days at 37°C.
Cells were re-suspended in PBS and incubated with CFSE (5(6)-Carboxyfluorescein N-hydroxysuccinimidyl ester) (ThermoFisher) for 15 mins at 37°C in the dark. Cells were then washed twice in PBS, re-suspended in cell culture media and allowed to proliferate for 96 hrs. Flow cytometric analysis was carried out using the FL-1/FITC channel in a FACS Calibur flow cytometer (Becton Dickinson) and CellQuest software, and the cell population was analyzed using FlowJo software.
Cell-Tak solution (Corning) at a concentration is 22.4 μg/mL (diluted in 0.1 M sodium bicarbonate pH 8.0) was used to coat the XF96 plates (Seahorse Bioscience) to allow for suspension cell adhesion to the plate. 3x105 cells per well were then seeded in the XF96 plates, followed by centrifugation of the cells at room temperature at 200 × g for 5 minutes. The plated cells were then incubated in a 37°C incubator not supplemented with CO2 for 25–30 minutes to ensure that the cells had completely attached. Cells were incubated for a total of 1 hr in a 37°C incubator without CO2 to allow for pre-equilibration with the assay medium. Cells were then analyzed by either the cell mito stress test assay or the glycolytic rate assay (see below).
The XF mito stress test report and glycolytic rate report generator automatically calculates the XF cell mito stress test parameters and glycolytic rate test parameters from Wave (Agilent) data that have been exported to Excel. Respiration and acidification rates are presented as the mean ± SEM of 3 independent experiments in all experiments performed with 4 to 10 replicate wells in the Seahorse XF96 analyzer.
|
10.1371/journal.ppat.0030015 | Newly Synthesized APOBEC3G Is Incorporated into HIV Virions, Inhibited by HIV RNA, and Subsequently Activated by RNase H | APOBEC3G (A3G) is a potent antiretroviral deoxycytidine deaminase that, when incorporated into HIV virions, hypermutates nascent viral DNA formed during reverse transcription. HIV Vif counters the effect of A3G by depleting intracellular stores of the enzyme, thereby blocking its virion incorporation. Through pulse-chase analyses, we demonstrate that virion A3G is mainly recruited from the cellular pool of newly synthesized enzyme compared to older “mature” A3G already residing in high-molecular-mass RNA–protein complexes. Virion-incorporated A3G forms a large complex with viral genomic RNA that is clearly distinct from cellular HMM A3G complexes, as revealed by both gel filtration and biochemical fractionation. Unexpectedly, the enzymatic activity of virion-incorporated A3G is lost upon its stable association with HIV RNA. The activity of the latent A3G enzyme is ultimately restored during reverse transcription by the action of HIV RNase H. Degradation of the viral genomic RNA by RNase H not only generates the minus-strand DNA substrate targeted by A3G for hypermutation but also removes the inhibitory RNA bound to A3G, thereby enabling its function as a deoxycytidine deaminase. These findings highlight an unexpected interplay between host and virus where initiation of antiviral enzymatic activity is dependent on the action of an essential viral enzyme.
| APOBEC3G (A3G) is a cellular enzyme that promotes DNA mutagenesis and can restrict infection by HIV-1. However, HIV counters the antiviral effects of A3G through the action of its Vif protein. In the absence of Vif, A3G is effectively incorporated into virions, where it mutagenizes the first DNA copy (cDNA) generated during reverse transcription of the viral RNA genome. A3G also appears to be able to inhibit HIV via nonenzymatic mechanisms. A3G and related deoxycytidine deaminases can also inhibit the growth of retroviruses other than HIV and protect the cellular genome from endogenous mobile retroelements. In this study, we analyzed the recruitment and enzymatic activity of A3G incorporated into HIVΔVif virions. Unexpectedly, we found that the binding of A3G to viral genomic RNA led to inactivation of the enzyme. However, latent A3G was ultimately activated through the action of HIV RNase H, which degrades the RNA genome during reverse transcription. These findings highlight an unexpected interplay between a host enzyme and HIV, where the antiviral enzymatic activity of the host factor (A3G) is dependent on the action of an essential HIV enzyme (RNase H). The strong interaction with viral RNA also suggests a potential mechanism by which A3G could exert antiviral activity in the absence of enzymatic activity, by physically impeding reverse transcription.
| APOBEC3G (A3G) is a highly active antiretroviral deoxycytidine deaminase that greatly impairs HIV spread in cultures of activated CD4 T cells provided the HIV Vif protein is absent [1]. In these activated cells, the antiviral action of A3G involves its effective incorporation into budding virions and subsequent hypermutation of nascent viral DNA formed during the next round of infection [2–6]. Vif has been proposed to block the incorporation of A3G into HIV virions by targeting this enzyme for accelerated degradation in the 26S proteasome [7–12] and partially blocking its de novo synthesis [7,13]. A different situation occurs in resting CD4 T cells and likely monocytes, which are not permissive for HIV infection. In these cells, a low-molecular-mass (LMM) form of cellular A3G is present, and it functions as a potent postentry restriction factor for HIV by blocking late reverse transcription [14]. This antiviral action of A3G is unchecked by Vif because insufficient quantities of Vif are present in the incoming virions and the virus has not progressed far enough into its life cycle to synthesize new Vif. Thus, the growth of wild-type (WT) HIV is effectively restricted in these cells by LMM A3G.
Incorporation of A3G into virions budding from HIV-infected CD4 T cells has been proposed to involve assembly with the nucleocapsid (NC) component of the Gag polyprotein and/or viral genomic RNA [15–22]. Recent studies with highly divergent Gag proteins [23] or treatment with RNase A [16,18,19,22] suggest that Gag binding may be indirect, involving an RNA intermediate. Following the entry of A3G-containing virions into new target cells, A3G deoxycytidine deaminase activity targets the minus-strand DNA product of reverse transcription, leading to the appearance of deoxyuridines in lieu of deoxycytidines at canonical sites of deamination (5′CC; the residue targeted for A3G-mediated deamination is italicized) [1–6,24]. The nontemplated action of various DNA repair enzymes, including uracil N-glycosylase, may mediate DNA strand cleavage [25], although a recent study suggests that uracil-N-glycosylase 2 is dispensable for the antiviral action of A3G [26]. If plus-strand synthesis proceeds, dA residues are introduced at sites of dC deamination, which results in dG-to-dA hypermutation in the viral coding strand. These mutations may compromise HIV infectivity by altering various viral open reading frames and introducing inappropriate translation termination codons.
In contrast to the LMM form of A3G in resting CD4 T cells, A3G in activated CD4 T cells principally resides in high-molecular-mass (HMM) RNA–protein complexes [14]. These complexes include Staufen RNA transporting granules and Ro/La ribonucleoprotein (RNP) complexes containing Alu and hY retrotransposon RNA [27,28]. These complexes lack detectable deoxycytidine deaminase activity in vitro but interrupt Alu retrotransposition by sequestering the retroelement RNA in the cytoplasm away from the requisite nuclear LINE machinery. Treatment of these complexes with RNase A promotes complex disassembly and generates the LMM, enzymatically active form of A3G. Thus, the cellular forms of A3G in resting and activated CD4 T cells are remarkably different. The recruitment of A3G into HMM RNA–protein complexes during the course of T-cell activation likely explains why cellular A3G fails to function as a postentry restriction factor for HIV in these activated cells.
The purposes of this study were to analyze the form of A3G that is incorporated into HIVΔVif virions and to assess its enzymatic activity. Since virion A3G is readily able to mediate hypermutation of viral DNA formed during reverse transcription, we anticipated that enzymatically active forms of A3G would predominate in virions. We have found that newly synthesized A3G, not preexisting A3G already assembled into the inactive cellular HMM complexes, is encapsidated into budding virions. We also found that A3G recruited into virions assembles with viral RNA to form a large intravirion A3G complex (IVAC) that is enzymatically inactive. Finally, we have demonstrated that the action of viral RNase H during reverse transcription ultimately releases A3G from its state of inhibition, allowing hypermutation of the minus-strand viral DNA. Thus, activation of the enzyme-dependent antiviral action of A3G appears to critically depend on the action of an HIV enzyme, RNase H.
Initially, we sought to identify transfection conditions for the generation of A3G-containing HIV virions that would recapitulate the virion encapsidation of A3G that naturally occurs in T cell lines and primary T cells infected with HIVΔVif viruses. Activated blood-derived primary CD4 T cells or H9 T cells expressing endogenous A3G were spinoculated with HIVΔVif and emergent viruses were harvested 2 d postinfection. In parallel, 293T cells were cotransfected with a fixed dose of proviral expression plasmid DNA and increasing doses of A3G expression plasmid DNA. Virions were similarly collected from the transfected cells 2 d later, after purification of virions by ultracentrifugation through iodixanol cushions. Virion lysates were subjected to immunoblotting to determine the amount of A3G incorporated relative to p24-CA content. These virion preparations were not contaminated with significant cellular material or microvesicles, as determined by immunoblotting with anti–14-3-3γ and anti-CD45 antibodies (Figure S1). Transfection of increasing amounts of A3G expression plasmid resulted in increasing amounts of A3G virion incorporation relative to p24-Capsid (CA) (Figure 1A and 1B). However, when compared to virions produced from infected primary CD4 T cells or H9 T cells, the 293T transfection conditions that best recapitulated the “natural” packaging levels of endogenously expressed A3G were achieved at a plasmid microgram ratio of 2 (HA-A3G) to 60 (pNL4–3ΔVif), which equals a molar ratio of 1 (HA-A3G) to 12.5 (pNL4–3ΔVif). Thus, this condition was used for the production of A3G-containing virions in all the subsequent experiments unless otherwise indicated.
To determine if A3G is packaged into the core of HIV virions and to assess the localization of the additional A3G packaging that occurred at higher transfection doses, virions were subjected to biochemical fractionation. The virion envelope was removed by brief solubilization with Triton X-100, as previously described [29], yielding separable virion cores containing p24-CA, integrase (IN), reverse transcriptase (RT), and NC (Figure 1C). Fractionation of viruses containing A3G at levels comparable to those of virions budding from primary CD4 T cells revealed that A3G is indeed packaged into virion cores (Figure 1C). However, virions derived from cells expressing higher levels of A3G (for example, 20 μg of HA-A3G:60 μg of pNL4–3ΔVif) packaged a similar amount of A3G into virion cores as well as additional A3G that fractionated into the gp41-containing supernatant after Triton X-100 solubilization (Figure 1A–1C). These findings suggest that when A3G is overexpressed in virus-producing cells, considerable amounts of the enzyme reside outside the viral core, likely in the viral matrix region, between the core and outer envelope (for example, 1:3 μg ratio). The appearance of approximately half the virion p24-CA in the Triton X-100 supernatant after solubilization similarly reflects excess Gag packaged into virions that does not contribute to core formation [30]. Thus, A3G may gain access to HIV virion cores through a specific interaction with viral RNA and/or Gag (NC). However, under conditions of overexpression, a lower-affinity interaction, perhaps directly between A3G and Gag, results in the recruitment of additional enzyme into the detergent-sensitive matrix space, into which excess Gag is also packed [30].
Under steady-state conditions in activated CD4 T cells, cellular A3G resides in an HMM RNase A–sensitive complex of 5 to 15 MDa [14,27]. The observation that HA-A3G incorporated into virions packages into the virion core suggested several possible cellular sources of the enzyme. The first possibility, albeit unlikely, is that entire 5- to 15-MDa cellular A3G complexes are recruited into the virion core. Alternatively, HIV RNA and Gag may promote release of A3G from the cellular HMM A3G complex, allowing its recruitment into the virion, with or without a limited subset of cellular cofactors, as has been suggested [28]. Finally, newly synthesized LMM A3G not yet assembled with cellular cofactors or RNA may be recruited into the virion through its association with viral RNA and Gag. To determine if newly synthesized A3G, more “mature” A3G, or both serve as cellular reservoirs for virion recruitment, we performed pulse-chase radiolabeling studies. First, the time course for recruitment of newly synthesized HA-A3G into cellular HMM complexes in the absence of proviral gene expression was determined. Cells were pulsed with radiolabel for 10 min, followed by chases of 30 min to 3 h. Size-fractionation (Figure S2A) of the pulse and chase lysates identified the presence of pulse-labeled HA-A3G initially in low and intermediate mass fractions (Figure 2A, t = 0 fractions 6 and 7) that was chased into HMM complexes within 30 min (Figure 2A, A3G in fractions 4 and 5 at 0.5 and 1 h). The radiolabeled A3G remained stably associated with the cellular HMM complex during longer chase periods (Figures 2A and S2B). Thus, newly synthesized A3G is initially LMM and recruited within 30 min into stable cellular HMM RNA–protein complexes. When HIVΔVif was coexpressed, we observed that the presence of viral RNA and proteins did not alter the ability of newly synthesized LMM A3G to assemble into cellular HMM RNA–protein complexes (Figure 2B).
To assess whether newly synthesized or more “mature” preexisting cellular HMM A3G is recruited into virions, we performed similar pulse-chase radiolabeling studies in cells producing HIV. To enhance radiolabeling and detection of intravirion A3G, the length of the pulse was extended from 10 min to 30 min. The longer pulse time did not affect A3G assembly into cellular HMM complexes, in either the absence (Figure 2C) or presence (Figure 2D) of HIVΔVif expression; however, it did mask the chase of LMM A3G into HMM A3G. Expression of HIVΔVif also did not affect the turnover of radiolabeled A3G (Figure S2D). Both producer cells and their supernatants containing virions were collected and analyzed simultaneously for radiolabeled HA-A3G and p24-CA content. Since all of the virus-containing supernatants were collected at the indicated time points, the radiolabel present in virion p24-CA or HA-A3G reflects labeling events occurring during the discrete intervening time points. In each of the three independent experiments shown, incorporation of radiolabeled p24-CA into virions increased with increasing chase time over the first 1 to 2 h and then declined by 4 h (Figure 2E, lower panels). The incorporation pattern of p24-CA over time is consistent with previous reports [31,32] showing increased accumulation of radiolabeled p24 in virions with long cumulative chase times. In contrast, HA-A3G incorporation into virions displayed a sharp spike between 30 and 60 min after the pulse (Figure 2E, middle panels), even though large cellular pools of radiolabeled A3G were present both before and after this time point (Figure 2E, upper panels). Specifically, despite the presence of radiolabeled cellular HA-A3G at the 2- and 4-h time points, these pools of A3G were not effectively incorporated into virions compared to the 1-h time point. The distinct peak of incorporation of radiolabeled A3G at 1 h after pulse also was not due to a relative loss of radiolabeled cellular A3G available for virion incorporation at the later collection times, since normalization by the available radiolabeled pool of cellular enzyme did not alter the distinct early kinetic pattern for A3G incorporation into virions (Figure S2C). These findings indicate that newly synthesized A3G less than 1.5 h old is incorporated HIV virions and that older, more “mature” A3G in HMM complexes is apparently less available for virion incorporation during the time course examined here. Using a modified version of these experiments (infection with HIVΔVif instead of transfection of a proviral DNA plasmid) and extended chase times, we observed a similar trend of newly synthesized radiolabeled A3G incorporation into virions and low-to-undetectable levels of radiolabeled A3G in virions up to 9 h after the pulse period (Figure S2E).
The finding that newly synthesized A3G is packaged into virion cores (Figure 1C) coupled with the observation that newly synthesized A3G rapidly forms HMM complexes in cells (Figure 2A–2D) led us to next examine whether intravirion A3G resolves as monomers/dimers or instead as a larger complex. We hypothesized that A3G in virions might remain in an enzymatically active LMM form, because it ultimately deaminates the viral minus-strand DNA synthesized during reverse transcription. Lysates derived from virions containing HA-tagged A3G were size-fractionated by fast protein liquid chromatography (FPLC). Each fraction was then analyzed by SDS-PAGE and immunoblotting with anti-HA monoclonal antibodies (Figure 3A). Surprisingly, virion-incorporated HA-A3G was detected almost exclusively in the HMM region, eluting in the void volume of the Superose 6 column (Figure 3A). This result was not due to incomplete lysis of the virion cores since the p24 viral capsid (p24-CA) was detected only in the expected LMM fractions.
To determine if the IVAC contains an RNA component, virion lysates were treated with RNase A (Figure 3B). Under these conditions, HA-A3G shifted to LMM fractions consistent in size with monomers and/or dimers of the enzyme. Therefore, reminiscent of the cellular forms of A3G present in activated CD4 T cells [14] (and in HIV-producing cells, as shown below), A3G incorporated into virion assembles into large RNA–protein complexes that are distinct from the cellular HMM complexes (see below).
As an alternative approach to examine whether A3G in virion cores were indeed freely soluble or associated with other factors, cores purified by solubilization of whole virions (Figure 1C) were biochemically disassembled by exposure to a low pH “STE” buffer at 37 °C, as previously described [33]. This treatment resulted in release of p24-CA into the supernatant of a pelletable viral RNP complex consisting of IN, NC, and viral genomic RNA. Under these conditions, RT is more readily released from the RNP upon biochemical fractionation of the cores (Figure 3C and as previously described [34,35]). Analysis of A3G-containing virion cores revealed that IVAC A3G cofractionated with the viral RNP proteins (Figure 3C), suggesting a continued association with the viral genomic RNA and/or NC protein.
Since virion-derived A3G ultimately exerts deoxycytidine deaminase activity during reverse transcription [1–6], we considered the possibility that A3G might remain enzymatically active even when bound to RNA in the IVAC. However, in an in vitro deoxycytidine deaminase assay, the IVAC HAA3G exhibited no detectable enzymatic activity (Figure 4A). Like the cellular HMM A3G complex (Figure 4D and [14]), deoxycytidine deaminase activity was readily detected when the virion HA-A3G immunoprecipitates were pretreated with RNase A before assessment (Figure 4A). Analysis of whole-virion lysate similarly showed that RNase A treatment was required for detection of A3G enzymatic activity in vitro (Figure 4B). These findings indicate that HA-A3G is incorporated into virions as an enzymatically latent large RNP complex. Of note, previous reports have observed readily detectable enzymatic activity from A3G-containing virions. We believe this may be due to the presence of additional, noncore LMM A3G packaged into the matrix space of virions upon A3G overexpression in cells (Figure 1C). When virions containing increasing amounts of A3G (Figure 1A) were tested for in vitro deaminase activity, substrate was readily deaminated by those virions which contained higher proportions of A3G to p24-CA than is normally packaged by infected CD4 T cells (Figure 4C). In addition, Yu et al. [6] reported that virion-derived A3G was enzymatically active in the absence of RNase treatment. However, in that study, virions were extracted in buffers containing EDTA [6], an agent that disrupts some RNA–protein complexes [36,37]. When we analyzed virion HA-A3G extracted in EDTA-containing buffers, we also detected deoxycytidine deaminase activity, suggesting that this treatment likely activated A3G by promoting its dissociation from inhibitory RNA(s) (Figure S3). Likewise, it has been observed that the addition of salts like magnesium that promote and stabilize RNA tertiary structure enhance the activity of recombinant A3G purified from insect cells while inducing a shift of A3G from large to intermediate-sized complexes [38]. Together these findings support the notion that RNA binding inhibits A3G enzymatic activity, likely by occluding the catalytic site.
The reported recruitment of A3G into virions through viral RNA, the cofractionation of intravirion A3G with viral RNP proteins (Figures 1C and 3C), the assembly of A3G into a large RNase A–sensitive complex within virions (Figure 3A), and the RNase A–dependent in vitro enzymatic activity of intravirion A3G (Figure 4A and 4B) strongly suggested that HIV genomic RNA may be an important constituent, perhaps even a nucleating factor for IVAC assembly. To test this possibility, we immunoprecipitated A3G from IVAC fractions, purified the RNA, and subjected it to RT-PCR with primers that specifically amplify HIV genomic RNA. Genomic HIV RNA was readily detected in the IVAC as well as in A3G immunopre0cipitates prepared from virus-producing cells (Figure 5A).
Next, we compared the FPLC fractionation profile of HIV RNA derived from virions that contained or lacked HA-A3G. In the presence of HA-A3G, HIV RNA was detected in fractions that contained the IVAC, indicative of A3G-dependent assembly of large HIV RNA–protein complexes in virions (Figure 5B). In the absence of HA-A3G, viral RNA was detected in lower fractions 11 through 14. The fractions were determined to contain full-length genome by the production of PCR products using probes that amplify across various regions of the genome. However, these gel filtration experiments were associated with some RNA fragmentation, particularly in the absence of A3G (Figure 5B, TAR/Gag amplicons were observed in fractions 11 through 20). We suspect that such fragmentation also occurred in A3G-containing virions but that the RNA fragments continued to resolve in the IVAC fractions through persistent association with A3G.
Of note, virion NC, which also binds HIV RNA, was detected in the FPLC fractions that contained IVAC (Figure 5C). Conversely, RT, which is more readily released from HIV RNA upon biochemical manipulation/lysis of virions [34,35] (and Figure 3C), did not display a strong shift into the IVAC fractions in the presence of HA-A3G upon gel filtration. As expected, the non–RNP-associated gp41 viral protein resolved independently of A3G. In the absence of A3G, NC resolved in lower FPLC fractions, consistent with pools of protein that either have dissociated from the viral RNP or remain associated with RNA fragments (possibly caused by the gel filtration conditions, as discussed above). In both the absence [14] and presence of HIV gene expression (Figure 5D), cellular A3G resolves as HMM upon gel filtration.
Next our studies focused on how the latent deoxycytidine deaminase activity of A3G present in the IVAC is ultimately activated. As shown in Figure 4A and 4B, the simple addition of single-stranded DNA (ssDNA) substrate was insufficient for triggering its activity. In view of the effects of RNase A treatment in vitro, we were intrigued by the possibility that viral RNase H enzyme might play a role in the activation of A3G enzymatic activity. RNase H resides near the C-terminus of the large subunit of the p66-p51 RT heterodimer [39,40]. The DNA-dependent action of RNase H is required for commencement of second-strand synthesis and concomitantly generates the free, minus-strand ssDNA substrate that is targeted by A3G for deamination [2–6]. We hypothesized that viral RNase H action might not only generate the substrate for A3G-mediated deamination but also reverse the RNA-mediated inhibition of A3G deoxycytidine deaminase activity by degrading the genomic RNA bound to A3G. To examine this possibility, we established in vitro conditions that lead to RNase H activity and assessed the effects of active RNase H on A3G deaminase activity. Both recombinant purified RT and virion-derived RT cleaved an end-labeled RNA oligonucleotide from an RNA–DNA hybrid substrate (Figure 6A and 6B). Importantly, RNase H activity was magnesium dependent [41], was inhibited by a variety of small molecules including Compound I [42,43], and was compromised by specific mutations within its catalytic domain, for example, E478Q ([44] and Figure 6A and 6B). We used these various properties of RNase H and reagents to probe the potential involvement of RNase H as an activator of latent A3G deoxycytidine deaminase activity within virions.
First, we tested whether stimulation of endogenous reverse transcription in virion lysates by the addition of magnesium and deoxynucleotide triphosphates promoted activation of A3G enzymatic activity measured in the deaminase assay. Such treatment effectively induced readily detectable deoxycytidine deaminase activity, suggesting a link between reverse transcription and A3G activation (Figure 6C). Of note, the appearance of deoxycytidine deaminase activity was blocked in a dose-dependent manner by Compound I, an RNase H inhibitor. Importantly, Compound I did not impair the deoxycytidine deaminase activity of A3G induced by prior RNase A treatment (Figure 6D), supporting inhibition of RNase H as the cause of Compound I–mediated inhibition of A3G enzyme activation. Additionally, the introduction of a point mutation in the catalytic core of RNase H (E478Q), which compromised RNase H activity (Figure 6B), also impaired activation of A3G deoxycytidine deaminase activity under conditions permissive for endogenous reverse transcription (Figure 6E). HA-A3G was otherwise equally active upon RNase A treatment of viruses bearing either WT or compromised RNase H domains (Figure 6F). These findings demonstrate that, in addition to generating the substrate for A3G-mediated deamination, HIV-1 RNase H plays a central role in triggering the activity of the latent virion-associated A3G enzyme.
Our observation that A3G packages into HIV virion cores is not unexpected given that RNA and/or the NC region of Gag recruit the enzyme into the virus [15–22]. Further, A3G antiviral activity is ultimately manifested during reverse transcription and therefore proximity to reverse transcription complexes would be anticipated. Indeed, Khan et al. [18] reported core localization of A3G. However, it is clear from the virion fractionations that additional enzyme may gain access to the virion when A3G is overexpressed in virus-producing cells (Figure 1). This additional A3G is not specifically recruited into virion cores. It has been previously demonstrated that roughly twice as much Gag than that present in the virion core is incorporated into immature particles [30]. Based on the RNase A sensitivity of the interaction between Gag (NC) and A3G [16,18,19,22] and the ability of A3G to interact with highly divergent Gag proteins [21,23,45–47], it has been suggested that A3G may recognize an NC–RNA interface that promotes virion incorporation [23,47]. Alternatively, the conflicting reports regarding A3G virion recruitment by Gag/NC in an RNA-dependent [16,18,19,22] or RNA-independent [15,17,20,21] manner could be a consequence of the relative amounts of additional, non–core packaging occurring under the conditions of assay. The RNase-sensitivity of Gag interaction may be observable only at lower (endogenous-like levels) A3G concentrations where affinity is perhaps governed primarily by RNA interactions. At higher expression levels of A3G, a lower affinity but direct interaction with Gag independent of RNA may occur. Whether virion core–incorporated A3G, like NC, coats the viral RNA or whether A3G binding is restricted to certain regions [18] remains to be determined.
Differences in the absolute amount of A3G packaged into virions may also contribute to apparent disparities in prior studies of A3G. Within this study, for example, the additional packaging of A3G under conditions of A3G overexpression masked the RNase-dependent “activation” of virion A3G (Figure 4C), highlighting the importance of establishing conditions that closely recapitulate physiological levels of A3G incorporation into virions.
Several observations in this study support the notion that A3G incorporated into HIV virion cores is assembled into a large RNA–protein complex that we have termed the IVAC. First, A3G was incorporated into virion cores (Figure 1C), which contain viral RNP complexes consisting of viral genomic RNA, NC, IN, and Vpr. IVAC A3G both coimmunoprecipitated viral genomic RNA (Figure 5) and cofractionated with the virion core proteins (Figure 3C). Additionally, the shift in viral genomic RNA from lower to higher mass FPLC fractions upon HA-A3G expression supports the notion that RNA is critical for virion packaging of A3G and suggests its possible central role in nucleation of the IVAC. Importantly, although the resolving power of our fractionation is currently not able to differentiate the sizes of the cellular HMM A3G complexes and IVAC (all resolve at or near the void volume of the Superose 6 column), these complexes are not identical. For instance, the cellular HMM complexes form in the absence of viral genomic RNA in activated but uninfected CD4 T cells [14], while IVAC A3G interacts with HIV RNA (Figure 5).
Second, our preliminary results indicate that many of the protein components of the cellular HMM A3G complex [27,28,48] are not corecruited into HIV virions (unpublished data and [28]). Finally, the activation of IVAC A3G by in vitro endogenous reverse transcription (Figure 6) suggests that viral RNA inhibits IVAC A3G enzymatic activity unless removed by RNase H, a virally encoded enzyme that acts on the RNA component of RNA–DNA hybrids. Of note, the level of A3G activation obtained when endogenous reverse transcription is stimulated (Figure 6E) was consistently less robust than the level of enzymatic activity observed when the IVAC was treated with RNase A (Figure 6F). We suspect this finding reflects a more complete clearance of RNA from IVAC A3G by exogenously added RNase A than occurs with RNase H activation under conditions of endogenous reverse transcription. Alternatively, A3G incorporated into HIV virions may bind both HIV RNA and non-HIV RNA (for example, the tRNA-Lys3 primer); however, since only the viral RNA genome is reverse transcribed, thereby forming a substrate for RNase H activity, only viral RNA-bound A3G may become “activated” during reverse transcription.
The pulse-chase radiolabeling studies of A3G in cells revealed that newly synthesized, LMM A3G is rapidly (within 30 min) recruited into cellular HMM complexes (Figure 2A) and that proviral gene expression has little, if any, effect on A3G assembly into HMM RNA–protein complexes (Figure 2B). Extension of the pulse radiolabeling time did not impede assembly but masked detection of the rapid assembly of newly synthesized LMM A3G into HMM complexes, in both the absence (Figure 2C) and the presence (Figure 2D) of viral gene expression. Notably, these complexes appear to be stable for at least several hours (Figure S2B).
The assembly of intravirion A3G into a large RNP complex could result from recruitment of any of the cellular HMM A3G complexes (Staufen RNA transporting granules, Ro/La RNPs) into virions or by HIV RNA extraction of A3G from cellular HMM A3G complexes. Alternatively, newly synthesized A3G not yet assembled into fully mature cellular HMM complexes could bind to HIV RNA, which in turn targets the enzyme for encapsidation into HIVΔVif virions. As noted, A3G assembled into HMM A3G complexes or A3G assembled with HIV RNA and core proteins sieve near the void volume of the Superose 6 columns and thus these different types of complexes cannot be distinguished by FPLC. To investigate whether newly synthesized A3G or older, “mature” A3G already assembled into HMM complexes is recruited into HIV virions, virus-producing cells were subjected to pulse-chase radiolabeling studies. In each of three experiments, we observed the appearance of a peak of radiolabeled A3G in virions at one discrete time point, occurring between 0.5 h and 1 h after the pulse (Figure 2E, middle panels). Radiolabeled A3G incorporation into virions decreased dramatically after this peak, despite the persistence of substantial pools of radiolabeled A3G in the producer cells from which the virions were derived (Figure 2E, top panel). The loss of radiolabeled A3G in virions after this peak also could not be explained by a sharp decline of radiolabeled A3G in the producer cells (Figure S2C). These findings suggest that once A3G assembles into the cellular HMM A3G complexes [27,28,48], it may no longer serve as a major reservoir of enzyme for virion encapsidation. Pulse radiolabeling of A3G before the peak of viral Gag expression and extension of the collection times up to 9 h after the pulse further confirmed that mature HMM cellular A3G does not form a major pool of enzyme for incorporation into HIVΔVif virions (Figure S2E).
Instead, it appears that newly synthesized A3G is preferentially recruited into HIV virions within 1.5 h after synthesis (Figure 2E, middle panels). Interestingly, the appearance of radiolabeled HA-A3G in virions appeared to be slightly delayed since samples collected (1) immediately after the pulse and (2) in the first 30-min chase contained relatively little radiolabeled A3G compared to virions collected over the second 30-min chase (Figure 2E, 1-h collection time point). This is clearly within the time required for assembly of newly synthesized A3G into cellular HMM complexes (within 30 min). Although virions were budded during the 30-min pulse, newly synthesized Gag did not yet contribute to these virions until the first 30-min chase period (Figure 2E, lower panels). Thus, the recruitment of newly synthesized A3G into virions may be intimately tied to the synthesis and assembly of the viral genome and/or Gag (NC) and the budding of these new virions. Indeed, A3G assembles with viral RNA in producer cells (Figure 5 and [28]) and maintains this association within virion cores (Figures 3 and 5). Since newly synthesized A3G assembles into HMM complexes in cells within 30 min in the presence or absence of HIV RNA (Figure 2A–2D), we cannot determine in these experiments whether the newly synthesized A3G (less than 1.5 h old) recruited into virions represents A3G newly assembled into any one specific cellular complex, a viral-specific HMM complex, or a combination of cellular and viral complexes. However, several other observations in conjunction with these pulse-chase data strongly support a model in which newly synthesized cellular A3G not yet fully assembled into cellular HMM complexes forms the major pool for recruitment into HIVΔVif virions. First, A3G incorporation into virions is mediated by assembly with viral determinants for encapsidation, including the viral RNA genome and/or Gag [15–22], and virion-incorporated A3G ultimately forms a large RNP complex (IVAC) with viral RNA, NC, and IN (Figures 3 and 5). Thus, virion-bound A3G forms a complex distinct from the cellular HMM complexes, at the very least distinguished by the presence of the viral encapsidation determinants (viral RNA genome and/or Gag). Second, since none of the cellular cofactors identified in the cellular HMM A3G complexes are corecruited with A3G into virions in an A3G-specific manner ([28] and unpublished data), viral determinants for A3G virion incorporation would have to extract A3G out of mature multisubunit complexes if they do serve as a reservoir for virion incorporation. If such a mechanism is employed, it is difficult to explain why A3G incorporation into virions is not also readily detected at much later time points in the pulse-chase radiolabeling studies. Finally, overexpression of A3G in cells leads to packaging of additional amounts of A3G into virions that localize outside of the virion core (Figure 3C), and this form of A3G is enzymatically active in vitro in the absence of addition of RNase A (Figure 4C). Thus, this additional extra-core A3G appears to be the LMM monomer/dimer that forms upon RNase A treatment [14] and could possibly arise from (1) newly synthesized LMM A3G not yet assembled into HMM complexes or (2) LMM A3G not assembled into HMM complexes due to saturation of cellular cofactors upon A3G overexpression.
We thus favor a model in which, upon translation, newly synthesized LMM A3G assembles with viral RNA and protein factors to gain access to newly assembling virions (and, in so doing, forms an IVAC-like complex). Since viral genomic RNA is subject to cellular processing that may be common to RNA that nucleates the cellular HMM complexes, a subset of common cellular RNA-binding factors can be predicted to be found in both the cellular HMM complexes [27,28,48] and viral RNP complexes. For example, RNA helicase A, a component of the Staufen-containing HMM A3G complex, has been reported to be packaged into virions [49], but its incorporation occurs independently of A3G and is unaffected by A3G virion incorporation (unpublished data). However, we cannot completely exclude the possibility that A3G is recruited into virions by viral cofactors from very recently assembled HMM complex(es), as recently suggested [28]. However, because we do not observe virion incorporation from older HMM A3G complexes, we do not favor such a model. One limitation of the pulse-chase studies is that we cannot calculate the percentage of radiolabeled (newly synthesized) to unlabeled (mature) A3G that is in virions at any of the given collection times.
The recruitment of A3G into HIV virions is ultimately detrimental to the virus, underscoring the essential function of the HIV Vif protein in blocking encapsidation of the deaminase. The principal mechanism by which Vif abrogates antiviral A3G activity is believed to involve proteasome-mediated degradation of A3G, most of which is resident in HMM A3G complexes. The observation that newly translated A3G (less than 1.5 h old) is preferentially recruited into virions (Figure 2B) implies that Vif must also effectively target this newly synthesized pool of cellular A3G. Recently, it has been reported that more Vif binds to A3G in the presence of RNase that in its absence [28], suggesting that LMM A3G unbound to RNA may be a good target for Vif. Our prior studies have shown that Vif expression promotes polyubiquitinylation of A3G that resolves as HMM [14]. Whether Vif activity leads to A3G ubiquitylation before, during, or after the assembly of this enzyme into cellular HMM complexes remains to be determined. Similarly, it remains to be determined whether ubiquitylated A3G resolving into HMM fractions upon FPLC represents the modification of A3G in the recently identified cellular complexes [27,28,48] or a separate complex of A3G, Vif, Cul5, SCF, and the proteasome. Perhaps Vif targets ribosome-associated A3G, thus destroying newly synthesized A3G and removing the key pool of enzyme that is selectively incorporated into virions. Such a scenario is consistent with the observation that Vif partially inhibits the synthesis of A3G [7,50]. Alternatively, Vif could target A3G bound for virion incorporation by targeting viral RNA–associated A3G. Indeed, Vif has been reported to interact with viral genomic RNA, suggesting a mechanism by which it might preferentially target virion-bound A3G [51–53]. Similarly, the reported association of Vif with the plasma membrane [54,55], the site of virion assembly, could localize this viral protein in proximity to A3G undergoing active encapsidation. While Vif expression ultimately depletes cells of all A3G, others have suggested that such global degradation of the enzyme may not be strictly required for Vif to exert its countereffects on A3G [56,57]. Regardless, our findings suggest that an important target for Vif is the newly synthesized pool of A3G, rather than A3G already assembled into cellular HMM complexes.
Because A3G can hypermutate the nascent minus-strand DNA of HIV, we found it surprising that intravirion A3G is inactive in in vitro deoxycytidine deaminase assays. Indeed, we observed that the binding of HIV RNA to A3G within virions prevented ssDNA binding and/or occluded the A3G catalytic site(s). As shown in Figure 4A, the addition of free ssDNA substrate to virion A3G proved insufficient to compete RNA binding and/or access the catalytic pocket(s). Rather, the inhibitory RNA had to be removed before A3G enzymatic activation was observed. In view of the emerging findings that A3G can also exert antiviral activity independent of deoxycytidine deamination [14,46,58–62], we propose that virion A3G may employ two mechanisms acting sequentially to produce its full antiviral effect. First, the enzymatically latent form of A3G bound to HIV RNA may impair the generation of minus-strand DNA by physically blocking the movement of RT on its viral RNA template. Indeed, short interfering RNA–mediated knockdown of endogenous A3G in resting CD4 T cells enhances the synthesis of late reverse transcription products [14] and the generation of both early and late reverse transcription products is reduced by the presence of A3G in virions [63]. However, because this inhibition is incomplete, minus-strand viral DNA is occasionally generated, setting the stage for the second, enzyme-dependent antiviral action of A3G. During reverse transcription, we now show that RNase H degrades the viral RNA that impairs A3G activity, allowing the enzyme to extensively deaminate the minus-strand DNA. Perhaps incomplete inhibition of reverse transcription by A3G is caused by the occasional ability of RT displace A3G off the RNA template. Although this could result in the generation of enzymatically active A3G, A3G may also be able to rebind the RNA–DNA duplex, reestablishing the inactivated state and dependence upon RNase H for enzymatic activation. Indeed, the lack of A3G activity induced by reverse transcription but under conditions where RNase H activity is inhibited (Figure 6E) suggests that if A3G is displaced by RT, it rapidly rebinds inhibitory nucleic acid. These events of initial inhibition and subsequent activation of A3G enzymatic activity by various components of the virus highlight an unexpectedly complex but interesting interplay between HIV and its cellular host.
Such a dual strategy for A3G inhibition of retroviral replication could account for its potent antiviral activity and explain reports of both enzyme-dependent and -independent antiviral activities. This model could also explain the conflicting results concerning the ability of A3G to inhibit the replication of hepatitis B virus. In some cells, A3G acts independently of deoxycytidine deamination [46,58], while in others, prominent DNA mutation is evident [64–66]. Cell-type differences in the relative effectiveness of these two sequential antiviral actions of A3G could underlie these findings.
The 293T cells were maintained in DMEM supplemented with 10% FBS (Gemini Bio-Products, http://www.gembio.com). H9 cells were maintained in RPMI supplemented with 10% FBS. Primary CD4 T cells were isolated from fresh human peripheral blood mononuclear cells on CD4 magnetic microbeads (Miltenyi Biotec, http://www.miltenyibiotec.com). The isolated CD4 T cells were then activated by 36-h treatment with PHA (5 μg/ml) followed by 36-h IL-2 treatment (20 U/ml; Roche, http://www.roche.com) in RPMI supplemented with 10% FBS, 100 μg/ml streptomycin, and 100 U/ml penicillin. Virions were generated by calcium phosphate–mediated cotransfection of subconfluent 293T in T175 flasks with a proviral plasmid (60 μg), pCMV4-HA-A3G vector (0 to 20 μg), and/or pCMV4-HA (0 to 20 μg). The medium was changed after 16 h, and the supernatant and cells collected were collected after 48 h. The virus-containing supernatant was clarified by low-speed centrifugation, filtered through a 0.22-μm membrane, and sedimented by ultracentrifugation over a 2-ml cushion of 8.4% iodixanol at 20,000 rpm using an SW28 rotor (Beckman Coulter, http://www.beckmancoulter.com) for 2 h at 4 °C. The virus-containing pellet was resuspended in 1 ml of PBS, DNase-treated (RNase-free; Roche), underlaid with a 100-μl cushion of 8.4% iodixanol and ultracentrifuged at 20,000 rpm in an HFA 22.1 rotor (Heraeus, http://www.thermo.com) for 1 h at 4 °C. Unless otherwise indicated, 0.1 U of RNase A inhibitor (RNaseOUT; Invitrogen, http://www.invitrogen.com) was added to virion pellets, which were then immediately lysed or flash-frozen on liquid nitrogen and stored at −80 °C until lysis. The addition of RNaseOUT had no effect on the intrinsic activity of HA-A3G (Figure S4). Cells were washed with PBS, and the pellet was either immediately lysed or flash-frozen on liquid nitrogen and stored at −80 °C until use. To generate VSV-G–pseudotyped ΔVif virions, 293T cells were cotransfected with expression vectors for the ΔVif provirus and the envelope of VSV-G. At 48 h after transfection, supernatants were cleared by low-speed centrifugation and filtration as described above and then used directly on fresh H9 or primary CD4 T cells. The T cells were spinoculated with the pseudotyped virion-containing supernatant as previously described [67]. Briefly, 0.4 × 106 cells/well of a 48-well plate were centrifuged at low speed for 2 h at room temperature with VSV-G–pseudotyped viruses. Cells were then washed five times with cold medium and returned to complete media for an additional 40 h. Supernatants and cells were then collected and processed as described above for the transfected 293T cells.
The proviral clone of pNL4–3ΔVif used to generate the HIV-1ΔVif virions has been previously described [68,69]. pNL4–3ΔVifH−(E478Q) contains a point mutation in the catalytic site of the RNase H domain of RT that compromises RNase H activity. This plasmid was generated by first subcloning the SpeI-EcoR1 Pol-containing restriction fragment of pNL4–3ΔVif into pEF1A. The mutagenesis primer 5′–ACAACAAATCAGAAGACTCAGTTACAAGCAATTCATCTAGC–3′ and its complement (Operon, http://www.operon.eu.com) were used to generate the E478Q mutation in the subclone using the QuikChange site-directed mutagenesis kit (Stratagene, http://www.stratagene.com). The mutation was confirmed by DNA sequencing. The pol region in the subclone was then recloned back into pNL4–3ΔVif. The introduction of the E478Q mutation into NL4–3ΔVifH–(E478Q) was confirmed by sequencing. pCMV4-HA and pCMV4-HA-A3G [7] expression vectors were cotransfected with pNL4–3ΔVif to generate HIV-1ΔVif virions lacking or containing HA-tagged A3G.
Virion cores were obtained using a previously published method [29]. Briefly, virion pellets were resuspended in MOPS Buffer I (200 mM NaCl, 100 mM MOPS [pH 7.0]) and Triton X-100 added to a final concentration of 0.5% for 2 min at room temperature. The cores were then pelleted from the solubilized enveloped by spinning the samples at 14,000g for 8 min at 4 °C. The core pellets were then washed twice with MOPS buffer II (100 mM NaCl, 50 mM MOPS [pH 7.0]). The cores were then either analyzed by immunoblotting or further fractionated to remove the p24-CA shell, as previously described [33]. Briefly, cores were resuspended in STE buffer (10 mM Tris [pH 6.7], 1 M NaCl, 0.5 mM EDTA), incubated at 37 °C for 4 h and subsequently centrifuged at 14,000g to pellet the RNP complex.
Virions present in the supernatants of 293T cells were transiently transfected with HIV proviral plasmids, and the cells themselves were lysed in ice-cold lysis buffer (50 mM HEPES [pH 7.4], 125 mM NaCl, 0.2% NP-40, and 1× EDTA-free protease inhibitor cocktail [Calbiochem/EMD Biosciences, http://www.emdbiosciences.com]). Lysates were clarified by sedimentation, quantified with a protein assay (Bio-Rad, http://www.bio-rad.com), and applied to a calibrated Superose 6 HR 10/30 gel filtration column run by an FPLC apparatus (AKTA; Amersham Biosciences, http://www.amersham.com). One column-volume (24 ml) using FPLC running buffer (50 mM HEPES [pH 7.4], 125 mM NaCl, 0.1% NP-40, 1 mM dithiothreitol, and 10% glycerol) was collected in 1-ml aliquots. Equal volumes of collected fractions were either directly run on SDS-PAGE gels or concentrated with YM-3 Microcon filters with a cutoff of 3,000 Da (Millipore, http://www.millipore.com) before running on SDS-PAGE after normalization for resultant concentrate volume. The size-separated proteins were then transferred to nitrocellulose membranes and immunoblotted. To test nuclease sensitivity, the lysates were pretreated with 50 μg/ml RNase A (DNase-free; Roche) and/or 20 to 200 U/ml DNase (RNase-free; Roche) for 1 h at 37 °C before gel filtration.
Polyclonal antibodies against A3G [7] and Vpr [70] have been previously described. Through the National Institutes of Health AIDS Research and Reference Reagent Program, HIV-1 RT monoclonal antibody (8C4) was obtained from Dr. Dag E. Helland, polyclonal antiserum to HIV-1 IN (757) was obtained from Dr. Duane Grandgenett, and HIV-1 gp41 human antibody (No. 50–69) was obtained from Dr. Susan Zolla-Pazner. Anti–NC-p7 antibody was generously provided by Dr. Robert J. Gorelick (National Cancer Institute, Frederick, Maryland, United States). Mouse monoclonal anti-p24 Gag ascites was generously provided by Beckman Coulter. Other antibodies used include polyclonal anti-HA antibody Y11, monoclonal anti–14-3-3γ antibody C-16, monoclonal anti-CD45 antibody 2D-1, and polyclonal anti-GFP antibody (FL) (all Santa Cruz Biotechnology, http://www.scbt.com) and monoclonal anti-HA antibody HA.11 unlinked or linked to beads (Covance, http://www.covance.com).
Immunoblot analysis of proteins was performed using horseradish-linked secondary antibodies followed by ECL detection (Pierce Biotechnology, http://www.piercenet.com). In Figure 1, A3G and p24-CA were detected and quantified by using fluorescently linked secondary antibodies (LI-COR Biosciences, http://www.licor.com). Blotted proteins were then detected and quantified using the Odyssey infrared imaging system and software (LI-COR).
Four plates of 293T cells were transfected with pCMV4-A3G-HA alone or with pNL4–3ΔVif. After 36 h, the cells were rinsed once and incubated for 1 h with pulse-radiolabeling medium (DMEM without methionine and cysteine; GIBCO, http://www.invitrogen.com) plus 10% dialysed FBS). The cells were pulse labeled for 10 min with 500 μCi/ml EasyTag XPRESS 35S Protein Labeling Mix (Perkin Elmer, http://www.perkinelmer.com) containing radiolabeled methionine and cysteine in fresh pulse-radiolabeling medium. At the end of the pulse-radiolabeling period, the radiolabel was removed and one plate of cells harvested. The remaining radiolabeled samples were incubated with chase medium (DMEM supplemented with 10% FBS, 4.02 mM methionine [20×], and 3 mM cysteine [15×]). Cells were harvested following incubation for 0.5, 1, or 2 h. Cells pellets were lysed in ELB lysis buffer. Each lysate was size-fractionated on gel filtration columns packed with Sepharose CL-6B beads, which crudely separate HMM from LMM proteins (Figure S1). For each sample, ten fractions of 300 μl each were collected, and equal volumes of each fraction were immunoprecipitated with anti-HA antibody. The immunoprecipitates were run on SDS-PAGE, and the signal was detected by autoradiography. The signal from each radiolabeled A3G-HA band was quantitated using Scion Image for Windows software (Version 1.62; Scion Corporation, http://www.scioncorp.com) and divided by the sum of the total signal, in order to assign a relative percent density versus the fraction number for every chase time point sample.
For the pulse-chase analysis of virus-producing cells, 293T cells were cotransfected with pNL4–3ΔVif, pCMV4-HA-A3G, and pEGFP-C1 to generate HA-A3G–containing HIV-1ΔVif virions. After 48 h, the medium was changed, and the cells were rinsed and incubated for 1 h with pulse-radiolabeling medium as described above. The cells were then pulse-radiolabeled for 30 min with 125 μCi/ml EasyTag XPRESS 35S Protein Labeling Mix (Perkin Elmer) in fresh pulse-radiolabeling medium. At the end of the pulse-radiolabeling period, the radiolabel was removed. Supernatant from the initial pulse-labeled samples (t = 0, pulse) was harvested, and radiolabeled cells were incubated with chase medium (DMEM supplemented with 10% FBS, 4.02 mM methionine [20×], and 3 mM cysteine [15×]) for 0.5 h. Again, supernatant was collected (t = 0.5 h), and the cells were incubated with chase medium for a further 30 min to generate the t = 1 h sample. The process was repeated twice more with an incubation of 1 h and 2 h to generate the t = 2 h and t = 4 h samples. At all time points, a fraction of radiolabeled cells were also collected, washed with PBS, pelleted by centrifugation, flash-frozen on liquid nitrogen, and stored at −80 °C. The virus-containing supernatants were filtered through a 0.22-μm membrane, and virions were sedimented by ultracentrifugation over a 2-ml cushion of 8.4% iodixanol at 20,000 rpm in an SW28 rotor (Beckman) at 4 °C. The pellets were resuspended in 1 ml of PBS, underlaid with a 100-μl cushion of 8.4% iodixanol, and ultracentrifuged at 20,000 rpm in an HFA 22.1 rotor (Hereaus) for 1 h at 4 °C. The resultant virion pellets were flash-frozen on liquid nitrogen and stored at −80 °C.
After virion and cell pellets had been obtained for all time points, the samples were lysed in the lysis buffer described above. Lysates were clarified by sedimentation and quantified with a protein assay (Bio-Rad), and immunoprecipitations were set up at equal protein concentration/volume in the presence of monoclonal anti-p24 ascites or monoclonal anti-HA antibody and incubated for 2 h at 4 °C. The immunoprecipitates were washed once with lysis buffer and subjected to SDS-PAGE. The proteins were transferred to nitrocellulose and immunoblotted for GFP or HA with polyclonal antibodies or for p24 with monoclonal antibody.
GFP, HA-A3G, and p24-CA identified by immunoblotting were excised from the membranes and subjected to scintillation analysis. Bands were first identified by immunoblotting since Gag and HA-A3G coimmunoprecipitate with each other [15,17,20,21] and are close in size. Scintillation counts were normalized to the amount of immunoprecipitated material assessed, determined with ImageJ (http://rsb.info.nih.gov/ij). The normalized counts were divided by the sum of the total counts to assign a relative percent density for every sample. No GFP was detected in virions (unpublished data).
In an alternate approach (Figure S2E), 293T cells were first transfected with HA-A3G expression vector DNA using Fugene (Roche) followed by infection of these cells with VSV-G-pseudotyped NL4–3ΔVif for 12 h. The cells were then pulse-radiolabeled with 125 μCi/ml EasyTag, as described above. Also as described above, after the pulse, cells were chased with cold medium and cells and virions were harvested at 1, 3, 5, and 9 h after the pulse-radiolabeling period. In these experiments, samples were subjected to denaturing lysis (50 mM Tris [pH 7.5], 1% SDS, 5 mM dithiothreitol) followed by anti-HA or anti-p24 immunoprecipitations (50 mM Tris [pH 7.5], 250 mM NaCl, 5 mM EDTA, 0.5% NP-40) and immunoblotting or PhosphorImaging (Bio-Rad), as indicated.
Samples for analysis were either (1) whole virion lysates or (2) FPLC fractions from cell or virion lysates. FPLC fractions were immunoprecipitated with monoclonal anti-HA antibody to concentrate HA-A3G. In all cases, the amount of HA-A3G in the input samples was confirmed by immunoblotting before analysis. DNA oligonucleotides (5′-ATTATTATTATTCCCATTTATTTATTTATTTATGGTGTTTGGTGTGGTTG-3′) containing target sites for A3G deamination [italicized] were labeled at the 5′ end with [32P]ATP using T4 polynucleotide kinase (New England Biolabs, http://www.neb.com) or with an FITC fluorophore (Operon). Labeled oligonucleotides and input samples were incubated in 20 μl of 50 mM Tris buffer (pH 7.4), with or without RNase A (1 μg) at 37 °C for 3 h unless otherwise indicated. For incubations under conditions stimulating endogenous reverse transcription, KCl (final concentration, 60 mM), MgCl2 (final concentration, 4 mM), and dNTPs (final concentration, 1 mM) were added. The RNase H inhibitor Compound I, generously provided by Dr. Daria Hazuda (Merck), was used at a final concentration of 0.1, 1, 10, or 100 μM. To terminate the reactions and purify the labeled oligonucleotides, the reactions were subjected to G-25 Mini Quick Spin Columns (Roche). Any uracil bases generated by A3G were converted to abasic sites by treatment of the purified oligonucleotides with 1 U of uracil DNA glycosylase (New England Biolabs) for 30 min at 37 °C. After 10 min of heat inactivation at 95 °C, the reactions were subjected to alkaline hydrolysis by the addition of NaOH (final concentration, 0.2 M) for 10 min at 95 °C. Cleavage products were resolved on 15% PAGE TBE-urea gels (Bio-Rad) and visualized with a Personal FX Imager (Bio-Rad), for radiography or fluorescence.
Samples for analysis were either virion lysates or recombinant protein. Recombinant HIV-1 RT, WT or E478Q, was generously provided by Dr. Matthias Gotte (McGill University) and used at a final concentration of 1 nM. Test substrates included ssDNA, ssRNA, RNA–RNA hybrid, and DNA–RNA hybrid. Unmodified 18-mer PAGE-purified complementary DNA and RNA oligonucleotides were from Operon and are based on the oligonucleotides 18-DAB-DNA and 18-FAM-RNA described by Shaw-Reid et al. [42]. In addition, an unmodified RNA oligonucleotide complementary to 18-FAM-RNA was used to generate the RNA–RNA hybrid (Operon). All oligonucleotides were end-labeled with [32P]ATP using polynucleotide kinase (New England Biolabs). Hybrids were formed by annealing hot oligonucleotide to cold complementary oligonucleotide. Samples were incubated in 20 μl of RNase H buffer (50 mM Tris-Cl [pH 8.0], 60 mM KCl) either with or without 5 mM MgCl2, as indicated, for 10 min at 37 °C. Compound I was added to a final concentration of 0.1, 1, 10, or 100 μM. Radiolabeled substrate (single-stranded or hybrid) was added to a final concentration of 100 nM, and the reactions were allowed to continue at 37 °C for 30 min. After the addition of loading dye to stop the reactions, the cleavage products were resolved on 20% PAGE-TBE-urea gels and were visualized with a Personal FX PhosphorImager (Bio-Rad).
FPLC samples and immunoprecipitates for RNA analysis were first treated with 20 U of DNase (RNase-free; Roche) at 37 °C. The RNA was then extracted with the QiaAmp RNA purification kit (Qiagen) according to the manufacturer's instructions. Viral genomic RNA was detected by reverse transcription with a primer complementary to the gag region of HIV-1 (5′-TGCTATGTCACTTCCCCTTGG-3′, generously provided by Jerry Kropp [Gladstone Institute of Virology and Immunology]) followed by PCR using primers complementary to the R (F496; nucleotides 496–517) and U5 (R573; nucleotides 552–573) or U5 (F592; nucleotides 592–613) and PBS (R666; nucleotides 645–666) regions of HIV-1. All these primers have been described [14]. In addition, reverse transcription was also performed using an antisense primer complementary to the Vpu region (5′-TCATTGCCACTGTCTTCTGCTCT-3′) followed by PCR using the Vpr primer and a primer complementary to Pol (5′-GTAATATGGGGAAAGACTCCT-3′). |
10.1371/journal.pbio.0060263 | Evolution Acts on Enhancer Organization to Fine-Tune Gradient Threshold Readouts | The elucidation of principles governing evolution of gene regulatory sequence is critical to the study of metazoan diversification. We are therefore exploring the structure and organizational constraints of regulatory sequences by studying functionally equivalent cis-regulatory modules (CRMs) that have been evolving in parallel across several loci. Such an independent dataset allows a multi-locus study that is not hampered by nonfunctional or constrained homology. The neurogenic ectoderm enhancers (NEEs) of Drosophila melanogaster are one such class of coordinately regulated CRMs. The NEEs share a common organization of binding sites and as a set would be useful to study the relationship between CRM organization and CRM activity across evolving lineages. We used the D. melanogaster transgenic system to screen for functional adaptations in the NEEs from divergent drosophilid species. We show that the individual NEE modules across a genome in any one lineage have independently evolved adaptations to compensate for lineage-specific developmental and/or genomic changes. Specifically, we show that both the site composition and the site organization of NEEs have been finely tuned by distinct, lineage-specific selection pressures in each of the three divergent species that we have examined: D. melanogaster, D. pseudoobscura, and D. virilis. Furthermore, by precisely altering the organization of NEEs with different morphogen gradient threshold readouts, we show that CRM organizational evolution is sufficient for explaining changes in enhancer activity. Thus, evolution can act on CRM organization to fine-tune morphogen gradient threshold readouts over a wide dynamic range. Our study demonstrates that equivalence classes of CRMs are powerful tools for detecting lineage-specific adaptations by gene regulatory sequences.
| The regulatory control of genes allows an organism to generate a diversity of cell types throughout its body. Gene regulation involves specialized DNA sequences called transcriptional enhancers that increase the expression of genes in specific places and times. Enhancers contain clusters of specific DNA sequences that are uniquely recognized by DNA binding proteins, whose activities are also regulated in space and time. The critical role that DNA enhancers play in generating the diversity of cell types within a single organism suggests that changes in these DNA sequences may also underlie the diversity of organismal forms produced by evolution. However, few examples linking specific changes in enhancer sequences to functional adaptations have been documented. We studied a group of neuro-embryonic enhancers that turn on a certain group of genes in different fruit fly species that have been diverging from each other for ∼50 million years. Each species has experienced unique changes in its protein-coding sequences, gene regulatory sequences, egg morphology, and developmental timing. We found that the organizational spacing between the protein binding sites in these enhancers has evolved in a manner that is consistent with functional adaptations compensating for the dynamic and idiosyncratic evolutionary history of each species.
| The state of a biological cell can be defined by the combined transcriptional status of each gene in a genome. Developmental systems specify cell state by regulating transitions between states. The regulatory logic for these state transitions is encoded in cis-regulatory DNA sequences, which specify the transcriptional activity of each gene [1,2]. Each gene may be controlled by multiple locus-specific, independently acting cis-regulatory modules (CRMs), which function as transcriptional enhancers, silencers, and insulators [3,4]. Such a set of CRMs can function collectively to sculpt a robust, complex spatiotemporal expression pattern [5]. Because of the critical role that CRMs play in specifying the transcriptional states of a cell, they have been proposed to be a primary target of natural selection [6–20]. Nonetheless, the relative importance of cis-regulatory versus protein-coding evolution has been debated because of a relative deficit of specific examples of functional CRM evolution [21].
Some important functional and evolutionary properties of CRMs have been elucidated. For example, enhancers possess switch-like properties that respond to well-defined physiological conditions [22–25], and generally enhancers can drive expression of a heterologous locus when placed almost anywhere into that locus [2]. Each CRM itself is a DNA segment of around 200 to 400 bp long that is composed of clustered binding sites for cooperative and competitive trans-acting factors that interact with the DNA. The elements constituting a CRM can arise rapidly anywhere in a gene locus in response to selection [9,17]. Furthermore, slightly deleterious mutations of binding sites in a CRM can be stabilized by the selection of compensatory sites elsewhere in the same CRM [26]. All of these properties of CRMs clearly establish that the evolutionary histories of such DNA sequences are unlike the evolution of protein-coding sequences. However, little is currently known about how evolutionary forces operate on the internal structure of CRMs simply because the organizational constraints of such sequences have not been fully explored.
To address the role of organizational constraints in CRM evolution, we have used the sequenced genomes for three different Drosophila species [27], which have been diverging for ∼50 million years [28,29]. These lineages have experienced divergent evolutionary pressures related to lineage-specific ecological life histories. Specific morphological differences include egg developmental morphology (e.g., size and shape of egg [30], composition of dorsal respiratory appendages in the egg chamber [31,32]), and embryonic developmental timing. Additionally, each lineage has experienced divergent genomic evolution as a result of differences in mutational processes. For instance, differences have been documented in insertion and deletion rates [33–35], as well as specific chromosomal inversions and transposition events [29]. Thus, Drosophila provides a powerful model system for studying how developmental suites of genes still manage to produce the basic body plan of a fly despite divergent processes affecting embryogenesis and genome composition.
In this study, we show how a class of equivalent developmental CRMs track evolutionary change in different Drosophila lineages. These CRMs act as neurogenic ectoderm enhancers (NEEs) and function to drive gene expression in the early embryonic neuroectoderm before gastrulation has commenced [36]. The NEEs map to unrelated loci: the rhomboid (rho) locus, which encodes a serine protease; the vein (vn) locus, which encodes an epidermal growth factor receptor ligand; the ventral neurons defective (vnd) locus, which encodes an NK-2 class homeobox transcription factor; and the brinker (brk) locus, which encodes a dipteran-specific helix-turn-helix repressor. The NEEs from these loci are located variably in either upstream or intronic positions and do not share sequence homology indicative of a common evolutionary origin.
Each NEE has independently evolved an organized cluster of common binding sites defined by three sequence signatures in D. melanogaster [36]. First, there are one to two pairs of a Dorsal binding sites closely juxtaposed (<20 bp) to a CA-core E-box motif, which is variably bound in different cells by either Twist basic helix-loop-helix (bHLH) complexes or the Snail C2H2 zinc-finger repressor [37]. Synergistic activation by Dorsal and Twist at specific positions along the Dorsal morphogen concentration gradient has been well documented [38–40]. Second, there is a unidirectionally oriented site, the μ motif, which is situated at a relatively fixed distance from the Dorsal–Twist pair, and which resembles the binding site for another co-activator, Dorsal interacting protein-3 (Dip3) [41–43]. Third, there is a unidirectionally oriented site, which is a composition of overlapping binding sites for the Notch signaling effector Suppressor of Hairless [Su(H)] and Dorsal. Other nuclear factors may also operate at NEE motifs in distinct territories along the dorsal/ventral (D/V) axis [44]. These diverse motifs co-occur in a 240–320 bp window defining each NEE [36]. Here, we show that selection acts on the organization of NEE binding sites to fine-tune the threshold readouts along the Dorsal concentration gradient.
We identified NEE-type sequences across the D. melanogaster, D. pseudoobscura, and D. virilis genomes in order to determine how a set of coordinately regulated gene loci co-evolve in a given lineage. We found that the NEE signatures of paired Dorsal–Twist binding sites (5′-SGGAAADYCSS and 5′-CACATGT, respectively) and a Su(H) site overlapping a separate Dorsal site (5′-CGTGGGAAAWDCSM, Su(H) site underlined) were present together in a single CRM across many loci (Figures 1A, 1B, and 2). We refer to such loci as “NEE-bearing” genes. Interestingly, the D. melanogaster NEE signature of an oriented and positioned μ motif (5′-CTGRCCBKSMM) was not discernable in enhancers from either the D. virilis or the D. pseudoobscura genomes.
From these three genomes, we cloned and assayed in transgenic stage 5(2) D. melanogaster embryos all NEE sequences from these species, which comprised five NEEs from D. melanogaster, five NEEs from D. virilis, and four NEEs from D. pseudoobscura, making a total of 14 distinct NEE-like sequences (Figure 1 and Table S1). These sequences include new NEE-like sequences at the short gastrulation (sog) loci (Figures S2 and S3). All of these sequences had interesting lineage-specific properties, described below.
To verify the endogenous expression patterns of NEE-bearing genes in these three species, we performed whole-mount antisense RNA in situ hybridization experiments using species-specific probes. Because the developmental timing of embryogenesis differs in the different species, we focused on one developmental time point corresponding to embryonic stage 5(2), when the embryo is midway through cellularization (Figure S1). At this point, the cell walls are 50% elongated and are easily identifiable under bright field microscopy. For D. melanogaster embryos growing at 25 °C this corresponds to ∼2 h 45 min after egg deposition. For D. virilis embryos growing at 25 °C, stage 5(2) corresponds to ∼5 h after egg deposition (Figure S1). However, in all three systems, NEE-driven reporters show earlier activity in late stage 4. This early pattern of activity in late stage 4 sometimes includes faint staining in the mesoderm that disappears by stage 5(2), at which point the lateral stripes are at their most robust and most reproducible levels.
We found that the expression patterns of orthologous genes across different species were significantly more alike than the expression patterns of NEE-bearing genes within the same species (Figure 1C–1F), despite differences in developmental timing, egg size, and genomic content (Table 1). For example, staining with species-specific rho probes reveals similar lateral stripes of expression in all three species (Figure 1C–1E). The span of expression for all NEE-bearing genes was quantitatively similar in terms of the number of nuclei along the D/V axis at 50% egg-length at the same embryonic stage. Similar observations were obtained at 25% and 75% egg-length (unpublished data). The differences in spans of expression are determined by the dorsal border of expression, as shown by the fact that the equivalent ventral mesodermal region remains unstained in each species. For example, the vnd genes across all three species were expressed in a narrow lateral stripe in the ventral neurogenic ectoderm spanning about six nuclei, whereas the brk genes were expressed more broadly in a domain spanning 10–12 nuclei, including more dorsal nuclei than the vnd expression pattern (Figure 1F).
Despite the similar patterns of endogenous expression between NEE-bearing orthologs, the amount of sequence divergence among these 14 enhancers is such that no two orthologous NEEs are more than ∼60% identical as a result of numerous substitutions, insertions, and deletions. In some cases, the amount of identity between orthologous enhancers is as low as ∼44%. This sequence divergence could represent both neutral drift processes and/or positive selection operating in each lineage.
Although these NEE sequences share certain signatures, some of these are undoubtedly examples of stabilizing selection creating de novo sites that compensate for sites lost by mutation. We find that this has occurred for all types of NEE motifs. For example, overlapping Su(H)/Dorsal motifs are not always present in the same location in some species for the vnd and rho NEEs (details 3 and 12 in Figure 2). Also, entirely new paired Dorsal and CA-core E-box motifs are found in the rho and sog NEEs (details 7–9, 11, 22, and 23 in Figure 2). In the rho example, the spacing has also been adjusted either through substantial deletions in the D. virilis lineage, or else through insertions in the D. melanogaster lineage (detail 8 in Figure 2). Thus, previously reported examples of stabilizing selection [26,45] appear to represent a general property of the NEE equivalence class of enhancers.
Not all NEE-bearing loci in one species are necessarily NEE-bearing in other species. We have found that the D/V patterning gene sog, which encodes a chordin-like inhibitor of the BMP/dpp signaling pathway, is an NEE-bearing locus in the D. melanogaster lineage but not in the D. pseudoobscura and D. virilis lineages (Figures S2 and S3). Previous work identified an intronic lateral stripe enhancer in the D. melanogaster locus (LSE in Figure S2E) [46]. This enhancer was identified by its cluster of multiple Dorsal binding site but lacks Su(H), E-box and μ motifs. We find that this intronic enhancer is not as well conserved across species as the upstream NEE-like sequences. Moreover, the D. melanogaster sog NEE drives the broadest lateral stripe of expression of all the other NEEs we have tested, spanning 15 nuclei across the entire embryo (see Figure 1F). This sog NEE recapitulates the endogenous expression pattern (Figure S2A and S2B). We also tested the orthologous upstream sequence from D. virilis and found that it, too, recapitulates its endogenous expression pattern (Figure S2C and S2D). These upstream NEE-like sequences contain Dorsal-linked TA-core E-boxes (5′-CATATG) and bipartite Su(H)/Dorsal motifs in all three species we studied, but they do not always contain the paired Dorsal and CA-core E-box sites, which are unique to the D. melanogaster sog NEE (Figure S3). In addition, when we tested the poorly conserved D. virilis sequence orthologous to the intronic lateral enhancer in transgenic D. melanogaster embryos, we observed only weak, patchy staining (unpublished data). Overall, these results show dynamic evolutionary history across individual CRMs, gene loci, and genomes.
An unknown portion of the substitution, insertion, and deletion mutations observed in the NEE cis sequences may be lineage-specific adaptations that stabilize changes occurring in trans. To address this question, we assayed individual enhancers from all three species in transgenic stage 5(2) D. melanogaster embryos, with multiple lines per enhancer to ensure reproducibility. This assay effectively decouples lineage-specific changes in trans from changes in cis by testing all enhancers in the same trans-environment of D. melanogaster. If these enhancers have evolved to compensate for lineage-specific changes in the trans regulatory environment, then we should observe similar directional changes for the entire equivalence class from one lineage when tested in transgenic D. melanogaster embryos.
Interestingly, despite similar functional outputs of orthologous NEEs in the context of their native genomes (Figure 1F), the NEEs from each species have unidirectionally modified activities in transgenic D. melanogaster embryos when compared with all the NEEs as a group from other species (Figure 3; in situ detection experiments were conducted in parallel). Specifically, the D. virilis enhancers consistently drive expression of reporters in a significantly more robust and expansive lateral stripe than the D. melanogaster enhancers in transgenic D. melanogaster stage 5(2) embryos, whereas D. pseudoobscura enhancers drive expression in a narrower stripes than the D. melanogaster enhancers (Figure 3). We have also verified this by conducting fluorescent double-label in situ hybridization of NEE-driven lacZ reporter lines using anti-lacZ and anti-snail RNA probes; snail labels the mesoderm (Figures 4 and 5). These experiments reveal that NEE-driven lacZ expression immediately abuts the mesodermal border without overlapping with it, which is consistent with ventral repression of NEEs by Snail.
The vnd enhancers, which produce the narrowest stripes of the NEE modules tested, showed the smallest differences in relative expression patterns when assayed in D. melanogaster (Figure 3D and 3E). It is possible that any extra activation potential or the need for it in vnd NEEs is masked by mechanisms that set the more restricted dorsal limit of expression, such as repression by the Ind and Msh homeodomain proteins or Schnurri-mediated repression via BMP signaling [47–50].
Similar results across NEE orthologs were obtained when we measured lacZ transcript intensity levels along the D/V axis using confocal microscopy. By aligning the sharp border of snail expression we see similar differences in stripe width (Figure 5). These results also show that, in many cases, it is the width of the stripe and not its intensity that changes between enhancers (Figure 5D). This suggests that the changes occurring in cis specifically affect the morphogen concentration thresholds that are being sensed by these enhancers.
As the D/V patterning system is known to be mediated primarily by Dorsal and Twist proteins, we decided to investigate the configuration of their binding sites in all of the enhancers and relate these in turn to their widths of expression across the lateral regions of the embryo. We found, first, with a few exceptions, that the CA-core E-box motif 5′-CACATGT is remarkably constant across the NEE sequences of all three species. Second, the Dorsal site occasionally sustains some point mutations. Third, there appear to have been many insertions and deletions that have adjusted the spacing between these two sites. Thus, the changes from all three types of variables (Twist sites, Dorsal sites, and their spacing) have served to alter the spacing in most cases, and occasionally to alter the number and quality of paired Dorsal and Twist sites (see Figure 2). We therefore suspected that the different widths of expression correlated with just these variables, as predicted by quantitative modeling [51]. In this manner, lineage-specific threshold readouts would be consistent with stabilizing selection in cis for diverse changes occurring in trans.
To test this hypothesis of genome-wide threshold adaptations, we decided to alter specific NEE sequences that differed only slightly in their Dorsal and Twist binding motif configuration relative to another NEE sequence that nonetheless differed greatly in the span of expression along the D/V axis. Of relevance, we also noted that the broadest NEE transgenes had a spacing between the Dorsal site and the adjacent E-box close to 7–12 bp (compare Figure 4 with sequences in Figure 2). Excepting the vnd NEEs, which may be constitutive targets of dorsally expressed repressors [47–50], most NEEs have increasingly narrow stripes the farther they are from this optimal spacer.
For example, the Drosophila brk NEEs have a conserved organization consisting of a central invariant Dorsal site flanked on either side by invariant CA-core E-box motifs (5′-CACATGT) (Figure 2 details 17–19, and Figure 6A). However, the D. virilis Dorsal to E-box spacer is shorter by exactly 3 bp on either side of the Dorsal motif relative to the D. melanogaster NEE (Figures 2 and 6A), in addition to many other substitutions and insertions and deletions (indels) throughout these enhancers. Recall that while the D. melanogaster brk NEE drives a lateral stripe of about eight or nine nuclei wide, the D. virilis brk NEE drives a lateral stripe of ∼13 nuclei in D. melanogaster stage 5(2) embryos (Figure 6B–6D). We therefore reduced the D. melanogaster NEE Dorsal site to E-box spacers by 3 bp on each side, mimicking the D. virilis configuration. This precise adjustment in spacing is sufficient to broaden the expression of the D. melanogaster brk NEE driven transgene to D. virilis brk NEE levels (Figure 6B and 6E). These in situ detection experiments were conducted in parallel to aid comparison. Furthermore, double labeling with probes to the mesodermal marker snail and the lacZ transgene shows that this functional change extends to both intensity of expression as well as expansion of the dorsal border of expression (Figure 7). However, even after normalizing the peak concentrations, a measurable difference in width is still evident (compare Figure 7D and 7E). These in situ detection experiments were also conducted in parallel to aid comparison.
In a similar example, the Drosophila melanogaster vn and sog NEEs possess the same Dorsal motif, which otherwise tends to vary at other loci (Figure 6F). This Dorsal motif (5′-CGGAAATTCCC) in each enhancer is situated 4 bp and 6 bp from the E-box motif (5′-CACATGTG) in the vn and sog NEEs, respectively (Figure 6F). Yet despite this similar NEE configuration in an otherwise nonhomologous DNA sequence, the sog NEE drives a broad lateral stripe of expression (∼15 nuclei; Figure 6G–6I) that is almost twice as broad as the vn enhancer (about eight nuclei; Figure 6G and 6H). Interestingly, the D. virilis vn NEE has an intermediate spacer of 5 bp and drives a lateral stripe of expression of intermediate width (∼11 nuclei; Figure 6G). We then compared a series of modified D. melanogaster vn NEE-driven transgenes possessing spacers adjusted by −1 bp, 0 bp (i.e., wild-type), +1 bp, and up to +2 bp, which mimics the sog NEE spacer, and found a monotonically increasing width in the lateral stripe of expression (Figure 6G–6J; in situ detection experiments conducted in parallel).
Thus, both the natural range of NEE configurations within each genome and across all three genomes, together with our functional manipulation of NEE configurations, confirm that the Dorsal site/E-box configuration controls the precise extent of D/V expression by extending the dorsal border of expression, where concentration of key activators is limiting. Thus, not only is there is an optimal organization for maximum affinity, but there is also a range of affinities that is exploited by natural selection to maintain precise threshold readouts of the concentration gradient of a developmental morphogen.
We next investigated diverse possibilities for lineage-specific selective pressures that might have caused NEEs across each genome to functionally adapt in similar directions via changes in spacing between the binding sites of cooperative activators. Such reasons might help explain why D. pseudoobscura NEEs have the weakest and narrowest stripes of expression in D. melanogaster embryos, while D. virilis NEEs have the strongest and widest stripes of expression when NEEs from all three species are tested in D. melanogaster embryos (Figure S8).
First, amino acid substitutions have occurred in the known NEE transactivators Dorsal and Twist (Figure S4, and unpublished data). Some enhancer evolution could therefore conceivably be due to stabilizing selection for changes in the trans factors themselves. Relative to the D. melanogaster Dorsal peptide sequence, these changes include a few non-synonymous substitutions in the DNA-binding REL homology domain (RHD, underlined sequence in Figure S4), as well as several non-synonymous substitutions and peptide indels in the non-DNA-binding regions. Interestingly, some of these amino acid substitutions in the DNA-binding domain correspond to known mutations that either reduce or augment Dorsal–Twist synergistic activation [52]. A lysine (K) to leucine (L) change in the D. pseudoobscura Dorsal RHD corresponds to a position that augments activation when mutated to alanine (A) in D. melanogaster Dorsal (see M7 in Figure S4). Both are changes of a basic side-chain to an aliphatic one. Such changes in D. pseudoobscura Dorsal might allow the evolution of weaker target NEEs. Remarkably, another mutation in the D. virilis Dorsal RHD corresponds to a position that reduces activation when mutated in D. melanogaster Dorsal (see M23 in Figure S4). Such a change in Dorsal might necessitate the evolution of stronger D. virilis NEEs. Future studies will investigate Dorsal protein and NEE co-evolution.
A second potential reason for genome-wide adaptations could also be changes in the protein expression levels. Staining with polyclonal antibodies made to D. melanogaster Dorsal and Twist factors reveals ventral to dorsal nuclear concentration gradients in all three species, with detectably slightly narrower nuclear Dorsal concentration gradients in D. virilis and broader nuclear Dorsal concentration gradients in D. pseudoobscura relative to the D. melanogaster gradients (Figures S5 and S6). The ratio of intensities for dorsal cytoplasmic levels of Dorsal antigen versus ventral nuclear levels are qualitatively similar across species and indicate that the shapes or profiles of the nuclear concentration gradient are comparable, even though the absolute intensities may not be comparable (Figure S7). Nonetheless, if the Dorsal morphogen gradient really is augmented in the smaller D. pseudoobscura embryos, and reduced in the larger D. virilis embryos, such changes would be consistent with the NEEs adapting to lineage-specific concentration readouts (Figure S8).
We also see a third potential reason for lineage-specific threshold readouts related to genome evolution (Table 1). Flow cytometry analyses of Drosophila genome sizes have confirmed a diverse range of sizes from 130 Mb for D. mojavensis up to 364 Mb for D. virilis [27]. D. melanogaster and D. pseudoobscura have intermediate genome sizes of 200 Mb and 193 Mb, respectively [27]. Of interest, we find that the total number ND = NAG/A of estimated NEE-style Dorsal motifs (5′-SGGAAABYCCH), where NA is the number of motifs found in the unfiltered assembly of size A in a genome of size G, is relatively constant across all three genomes, the total number NE of NEE-style CA-core E-boxes (5′-CACATGT) is 2-fold greater in D. virilis than in D. melanogaster (Table 1). We also note that the relative increase of this motif is a secondary trend related to a simple expansion (D. virilis) or compaction (D. melanogaster) of 5′-CACA repeats occurring primarily in euchromatic regions of the genome (unpublished data). In general, longer microsatellites have been documented in D. virilis than in D. melanogaster [35]. However, a potential effect of a 2-fold greater number of genomic 5′-CACATGT motifs might be to reduce the effective free nuclear concentration gradient of Twist bHLH complexes in D. virilis relative to D. melanogaster via background sequence sequestration [53]. Such an effect would also be consistent with the observed adaptive trend to a lower concentration threshold readout in D. virilis than in D. melanogaster (Figure S8B). Thus there are potentially several lineage-specific changes affecting the activity or profile of the dorsal/ventral morphogen system that would necessitate the observed genome-wide adaptations in downstream target enhancers.
In this study we demonstrated the effectiveness of studying gene regulatory evolution in the context of a CRM equivalence class consisting of all genomic sequences that regulate nearly identical patterns of activity by interacting with a common set of transcription factors.
Equivalence classes of CRMs represent molecular examples of parallelisms, which can be defined as similar, homoplastic patterns of evolutionary innovation constrained by a restrictive set of available regulatory mechanisms [54]. For several important reasons, generalization of CRM logic for a class of functionally and mechanistically equivalent CRMs is difficult to obtain from the phylogenetic study of a single CRM. First, insufficient time for divergence away from an ancestral CRM can leave much superficial similarity among orthologous CRMs. Second, the persistence of initial organizational constraints present in the ancestral CRM can obscure possible alternative configurations of the class-defining cis elements. Third, novel lineage-specific evolutionary adaptations will impede any method that relies purely on phylogenetic conservation.
We used the equivalence class of NEEs from several species to show that selection acts on the organization of enhancers to fine-tune their output. In this case, NEEs have evolved in parallel to adapt to changes in a developmental morphogen gradient, whose activity levels have shifted in different directions in different lineages, necessitating compensatory changes in the cis components of downstream targets (Figure S8). Such stabilizing selection, encoded in the configuration of Dorsal binding sites and Twist-binding CA-core E-boxes, reflects species-specific thresholds of activation (compare θDm, θDp, and θDv in Figure S8). Thus, the D. pseudoobscura enhancers have evolved to respond to higher morphogen concentrations than those in D. melanogaster (θDp > θDm), while the D. virilis enhancers have evolved to respond to lower concentrations than those in D. melanogaster (θDv < θDm). This stabilizing selection is revealed only when finely tuned NEEs are tested in the exogenous concentration gradient of a different species, in this case the reference species D. melanogaster.
It should not be surprising that small changes in the linkage between the Dorsal and Twist complex binding sites could have such a dramatic effect on the D/V range of expression. First, evolutionary changes in the dorsal border of neuroectodermal gene expression are a simple readout of the most limiting amount of nuclear Dorsal, further limited by limiting amounts of Dorsal target proteins working as Dorsal co-factors, such as Twist bHLH complexes [39,40,55]. Second, recent studies on the Bicoid morphogen gradient, which simultaneously patterns the anterior/posterior axis, suggest that its precision is pushed to the physical limits imposed by the stochasticity of molecules [56–58]. Therefore, precision in the morphogen gradients patterning the embryonic axes should indicate a precision in CRM readouts, as we have shown here for one class of D/V enhancers and others have indicated for diverse anterior/posterior enhancers [59,60].
It should also not be surprising that it is the site linkage that is adjusted by natural selection rather than the quality of the binding sites. Quantitative modeling of the classical Dorsal morphogen system has placed heavy emphasis on the quality of binding sites as determinants of differential threshold readouts by target enhancers in the mesoderm, mesectoderm, and neuroectoderm [51]. However, we do not believe this is entirely at odds with our results showing evolutionary modification of NEE activity via organizational linkage because it may indicate that within the neuroectodermal territory, where Dorsal protein is present in limiting amounts, binding sites are likely to have already evolved to be high affinity sites. This would indicate that in this embryonic territory, extra affinity can only be achieved by optimizing positional linkage between these cooperatively binding factors. This in turn implies that these same factors have relatively fixed steric dimensions.
Another potential locus and/or cause of selection lies in the protein-coding sequences of the morphogens themselves, as has been shown for other trans factors [61,62]. Both Dorsal and Twist are used as combinatorial inputs for other regulons in other tissues (e.g., mesoderm and mesectoderm) and the pleiotropic consequences of changes to either protein–protein interaction motifs or DNA-binding domains might limit the number of possibilities for such changes. Additionally, such protein-coding changes may not effectively or precisely target the Dorsal–Twist interactions where their amounts are limiting and/or affect the interaction in the continuously graded fashion that we have documented. Nevertheless, stabilizing selection for changes in the Dorsal peptide sequence itself or other trans factors could be explored by future trans complementation assays. However, in many cases, it may be difficult to disentangle evolutionary cause and effect between co-evolving loci throughout the genome because sequence changes may be either the initiating causes or the products of selection for developmental homeostasis.
In summary, the coordinate changes in both sequence and activity shown by the neuroectodermal enhancers from each species provide strong evidence for functionally adaptive cis-regulatory evolution. The number of fixed nucleotide changes corresponding to these parallel molecular adaptations occurring across a genome is minimal, and corresponds to a few indels between Dorsal and Twist sites across the NEEs in a single genome. Occasionally, new Dorsal and Twist sites with optimal spacing are presumed to have been selected in individual species for the vnd, rho, and sog NEEs while in others only the spacing has changed between existing sites. Our results show that natural selection can act with ease to fine-tune CRM organization and thus calibrate enhancer activity over a wide dynamic range. As such, CRM organization may represent a large and unexplored locus of stored adaptive information.
D. melanogaster strain w1118 was used for P-element transformations of all reporter constructs. D. virilis and D. pseudoobscura were obtained from the Tucson Drosophila Stock Center.
D. melanogaster, D. virilis, and D. pseudoobscura embryos were collected, and subsequently fixed. Hybridization with digoxigenin-labeled antisense RNA probes was conducted as previously described [63]. Fluorescent multiplex in situ hybridization methods were performed as previously described [64]. Briefly, primary antibodies were used to detect fluorescein isothiocyanate- and digoxigenin-labeled antisense RNA probes (used 1:400, Invitrogen), followed by detection of primary antibodies using secondary antibodies labeled with Alexa Fluor dyes (used 1:500, Invitrogen). Images of fluorescently labeled embryos were acquired on a Nikon Eclipse 80i scanning confocal microscope with a 20× objective lens. Sum projections of confocal stacks were assembled and plot profiles of the RNA transcripts were analyzed using the ImageJ software. Anti-sense endogenous probes were created by PCR amplification from genomic DNA with a T7 RNA polymerase promoter included on the reverse primer (see Supplemental methods for all primer pairs). DNA fragments for injection were cloned into the [-42EvelacZ]-pCaSpeR vector and introduced into the D. melanogaster as described previously [3]. Between three and seven independent transgenic lines were obtained for each construct: Dm brk NEE (657 bp), Dp brk NEE (859 bp), Dv brk NEE (744 bp), Dm rho NEE (871 bp), Dp rho NEE (843 bp), Dv rho NEE (726 bp), Dm vn NEE (919 bp), Dp vn NEE (858 bp), Dv vn NEE (836 bp), Dm vnd NEE (1020 bp), Dp vnd NEE (1305 bp), Dv vnd NEE (1093 bp), Dm sog NEE (550 bp), and Dv sog NEE (871 bp).
For protein expression experiments, Drosophila embryos for all species were fixed in 3.7% formaldehyde for 20 min at room temperature. Polyclonal rabbit anti-Dorsal, guinea pig anti-Dorsal, and rabbit anti-Twist antibodies were used for primary detection. Secondary anti-rabbit antibodies conjugated to FITC (Roche Applied Sciences) and anti-guinea pig conjugated to TRITC (Sigma-Aldrich) were used for visualization.
Whole-genome scans for sequences matching enhancer models were conducted using multiple techniques to verify results. These methods included searching fly genomes using scripts written in PYTHON and as well as local-alignment-based methods, primarily the VISTA suite of whole genome alignments. Dialign2 was used for additional sequence alignment and to determine nucleic acid identity.
The enhancer sequences used in this study have been deposited in GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) with accession numbers FJ169871–FJ169884. D. melanogaster and D. virilis DNA fragments containing identified enhancer elements were amplified from genomic DNA with the following primer pairs (lowercase letters denote flanking primer sequence used to introduce the restriction site indicated in brackets): Dm rho: [BsaI] ccgcataattcgagaccCAGTTAAGTGAGTCGCTTTCAGG, ccgcataattcgagaccTAGATAGATATACCCATCCTGGCC; Dv rho: [BsaI] ccgcataattcgagaccCTGGGAAGTTGCACGAGAGACGC, ccgcataattcgagaccGAGAAACTCTTCTGGCACAACGC; Dp rho: [EcoRI] ccgcggaattcACTAGTGGAAGCTGCTCTACAGACGCG, ccgcggaattcACTAGTCACACACAGCGAAGCACTGAGA; Dm vnd: [EcoRI] ccgcggaattcGGAAGATTGGGCGTTGAAAGC, ccgcggaattcCGGCCATTCACACGATTGACACA; Dv vnd: [Mef1] ccgcgcaattgCTGTTTGGCCTGGCTGGC, ccgcgcaattgATGGCCGGAAAGCAACACAATGG; Dp vnd: [BsaI] ccgcataattcgagaccGTATTTCGAAGATGCATTTGTTTGC, ccgcataattcgagaccTGAATGGCCGGAAGGTCCAACAAG; Dm vn: [BsaI] ccgcataattcgagaccCAGTTCTGGATCTTCCGAATCACC, ccgcataattcgagaccAAATTTGTAGCCAGCGGCGACG; Dv vn: [EcoRI] ccgcggaattcCATATGTTGCCCCTTTGCTGTTGC, ccgcataattcgagaccAAATTTGTAGCCAGCGGCGACG; Dp vn: [BsaI] ccgcataattcgagaccTGTGGGGCAATATTTTCTTTTTAGC, ccgcataattcgagaccCTAAAAATGCAACTCCAACTTGTCTG; Dm brk: [EcoRI] ccgcggaattcTTGGTCGGAAAATACCTGCGC, ccgcggaattcATTGTGTGGCGTTAGAAAGATATGG; Dv brk: [EcoRI] ccgcggaattcTGTCCGGGCTTATGGATCG, ccgcggaattcTGCATTATCCGTGCTAAGTTTGGG; Dp brk: ccgcTCTAGAAAAATGCCGAACAGGTACGTCG, ccgcTCTAGAAAATCATATCCTAACCCCATCTGGG; Dm sog NEE: [EcoRI] ccgccgaattcTGTTTATGGCAGCCAATTGATGCCGA, ccgccgaattcgatgatctagaatcgcacggagag. The brk and rho enhancers were mutagenized using overlap extension PCR with the following primers: Dm brk D/T1 (letters flanking deletion underlined): GGCACAGGCACACATGTGTGTTTGTGAACGGGAAAGCCCCATTTT; Dm brk D/T2 (letters flanking deletion underlined): GCCCCATTTTAAAGCTGGCCCAACGGCAACACATGTTCATGTTAG; Dm vn-1 (letters flanking deletion underlined): GGACAGGTAACGGGCCACATGTCTGCCGGAAATTCCCCGTTGACCCCTG; Dm vn+1 (inserted letter underlined): GGACAGGTAACGGGCCACATGTCTGAGCCGGAAATTCCCCGTTGACCCCTG; Dm vn+2 (inserted letters underlined): GGACAGGTAACGGGCCACATGTCTGATGCCGGAAATTCCCCGTTGACCCCTG.
The following primer pairs were used to amplify probes for each of the indicated genes from each species with the T7 promoter sequence indicated): Dm rho: ATCTGGGCTATGCTCTCTACACC, T7-TTAACTGCAAACGGTAACGATAACG; Dv rho: GCCGTCTACACGCAGTACTTCG, T7-CATTTGTTTACACGTTTCGGCCCG; Dp rho: TTGCCATCTTCGCCTACGATCG, T7-GCTTAGGAGACACCCAAGTCG; Dm vn: CTTTGCGGCACCCACCGTTTT, T7-TCCACTCACTATAATTTTCGCTCAC Dv vn: GAGTAGAAGATATATGCGTATGAGC, T7-GTTCACAGCCATTTTAACTGCTTCG; Dp vn: AATTTTGGAGGACGCCATGTATGCG, T7-ACCGATTCCATCACCGAGTGG; Dm brk: GAAATACAACATTCACCGCCGG, T7-TCAGGATTGGCACTTGTATTGGC; Dv brk: ATTTTTGTTACTTCCAAGACGACGG, T7-TATCGAACTTGTTCGCTGTTATCC; Dp brk: CAGCAACCACAGTCCTAACGC, T7-ATCAGGTTGTCGTTGGAGACG; Dm vnd: TCAGTTTGGAATGTGTAGAGTGCGC, T7-GGATAAAAGGCGGCTGGTAGG; Dv vnd: AATGTTTAGAGTGCGGCTGACAACG, T7-TCCAAGGGGGCAGCAATATGG; Dp vnd: TGTCCTACACCTACATCGGTTCC,TCTAGCAGTATTAGGGCCACC; Dm sog: GGAAATGAAGTCCATGTACACCACC, T7-TCTCGTACACCTTGTTGACCACC; Dv sog: GAGGAGATGAAGTCCATGTACACG, T7-CCGTTCTCGTAGATCTTGTTGACG; Dp sog: ATAGTATGTCCCATGCCTCACCG,TTCTGCTCTGGCGAATTTTAAAGC; Dm cg8117: ccgccCTAAAAATTGTGCTTTCCGGTTTCG, T7-ATTTACATATTGTGGAAGCCAACGG; Dm cg8119: ccgccAACAAGTATCCGACCAACAATCTGG, T7-TCACGTCAGCAGCTTCTTCTCC.
|
10.1371/journal.ppat.1006739 | Voltage-dependent calcium channel signaling mediates GABAA receptor-induced migratory activation of dendritic cells infected by Toxoplasma gondii | The obligate intracellular parasite Toxoplasma gondii exploits cells of the immune system to disseminate. Upon T. gondii-infection, γ–aminobutyric acid (GABA)/GABAA receptor signaling triggers a hypermigratory phenotype in dendritic cells (DCs) by unknown signal transduction pathways. Here, we demonstrate that calcium (Ca2+) signaling in DCs is indispensable for T. gondii-induced DC hypermotility and transmigration in vitro. We report that activation of GABAA receptors by GABA induces transient Ca2+ entry in DCs. Murine bone marrow-derived DCs preferentially expressed the L-type voltage-dependent Ca2+ channel (VDCC) subtype Cav1.3. Silencing of Cav1.3 by short hairpin RNA or selective pharmacological antagonism of VDCCs abolished the Toxoplasma-induced hypermigratory phenotype. In a mouse model of toxoplasmosis, VDCC inhibition of adoptively transferred Toxoplasma-infected DCs delayed the appearance of cell-associated parasites in the blood circulation and reduced parasite dissemination to target organs. The present data establish that T. gondii-induced hypermigration of DCs requires signaling via VDCCs and that Ca2+ acts as a second messenger to GABAergic signaling via the VDCC Cav1.3. The findings define a novel motility-related signaling axis in DCs and unveil that interneurons and DCs share common GABAergic motogenic pathways. T. gondii employs GABAergic non-canonical pathways to induce host cell migration and facilitate dissemination.
| Dendritic cells are considered the gatekeepers of the immune system but can, paradoxically, also function as ‘Trojan horses’ to mediate dissemination of the common intracellular parasite Toxoplasma gondii. Previous work has shown that Toxoplasma hijacks the migratory machinery of dendritic cells by inducing secretion of the neurotransmitter GABA and by activating GABAergic signaling pathways, thereby making infected dendritic cells hypermigratory in vitro and in vivo. Here, we show that the signaling molecule calcium plays a central role for this migratory activation and that signal transduction is preferentially mediated through a subtype of voltage-gated calcium channel (Cav1.3). This study functionally implicates Cav1.3 channels in a, hitherto uncharacterized, calcium signaling axis by which dendritic cells are induced to become migratory. The studies show how an obligate intracellular pathogen takes advantage of non-canonical signaling pathways in immune cells to modulate their migratory properties, and thereby facilitate the dissemination of the parasite.
| The obligate intracellular parasite Toxoplasma gondii chronically infects a large portion of the global human population and is capable of infecting any warm-blooded vertebrate [1]. The dissemination of the parasite from the point of entry in the intestinal tract plays a determinant role in the pathogenesis of toxoplasmosis. Although chronic infection is generally considered asymptomatic in otherwise healthy individuals, reactivated infection in the central nervous system (CNS) of immune-compromised individuals may be fatal. Congenital toxoplasmosis occurs by transmission to the fetus from the infected mother and can result in serious disabilities or death of the unborn child [2].
Previous studies have demonstrated that active invasion of dendritic cells (DCs) by T. gondii tachyzoites rapidly (within minutes) induces a hypermigratory phenotype in parasitized DCs [3]. This migratory activation is characterized by cytoskeletal rearrangements and dramatically enhanced cellular locomotion, termed hypermotility [4], and enhanced transmigratory activity in vitro [5]. These phenotypes have been linked to enhanced dissemination and parasitic loads in mice for different species of apicomplexan parasites [5–7]. The initiation of the hypermigratory phenotype in DCs is related to the discharge of secretory organelles during parasite invasion and does not depend on de novo protein synthesis in the host cell [4]. It is mediated through non-canonical GABAergic signaling pathways, and is independent of MyD88-mediated TLR signaling and chemotaxis [3–5, 8].
Within the context of the host-parasite interaction, we have recently shown that DCs possess functional GABAA receptors, and the capability to synthesize and secrete γ–aminobutyric acid (GABA) [8]. Challenge with T. gondii triggered GABA secretion in the invaded DCs and inhibition of GABAA receptors, GABA synthesis or GABA transport abrogated the T. gondii-induced hypermigratory phenotype [8]. Along these lines, mounting evidence shows that GABA, the main inhibitory neurotransmitter in the vertebrate brain, participates outside the CNS in diverse functions including cell migration, immunomodulation and metastasis [9–11].
GABAA receptors are ionotropic chloride channels whose functions are regulated by cation-chloride co-transporters [12]. Membrane depolarization secondary to GABA receptor activation can elicit opening of voltage-dependent Ca2+ channels (VDCCs, also termed voltage-gated Ca2+ channels, VGCCs) that are normally closed at physiologic or resting membrane potential [13–15]. Thus, GABA-mediated Ca2+ influx via VDCCs is a well-established concept in neuronal cells but remains unexplored in immune cells. While various Ca2+ signaling pathways have been implicated in the regulation of multiple DC functions, including activation, maturation and formation of immunological synapses with T cells (reviewed in [16]), knowledge on the role of VDCCs in DCs remains limited [17, 18].
Here, we show how an obligate intracellular pathogen takes advantage of a hitherto uncharacterized Ca2+ signaling axis in DCs to modulate the migration of parasitized host cells. We demonstrate that the hypermigratory phenotype induced in DCs by T. gondii is predominantly dependent on the L-type VDCC subtype Cav1.3, which is activated by GABAergic signaling upon T. gondii invasion.
Shortly after parasite entry, DCs exhibit a dramatic migratory activation [5, 8]. Based on the implication of GABAergic signaling and on the rapid onset of the hypermigratory phenotype minutes after parasite invasion [4], we hypothesized a role for Ca2+ signal transduction for the induction and maintenance of T. gondii-induced migration. When DC motility was assessed in a Ca2+-deprived medium, individual cell track analysis of infected DCs showed a reduction in migrated distances (Fig 1A) and a significant downward shift in the distribution of migrated distances (Fig 1C and 1D). In low Ca2+ medium with 1% FBS, the median velocity of T. gondii-infected DCs was significantly reduced, and addition of Ca2+ (CaCl2) at physiological concentration reconstituted hypermotility in infected DCs (Fig 1B). Similarly, base-line motility of non-challenged DCs was reduced upon Ca2+ deprivation and reconstituted by addition of Ca2+ (Fig 1B and S1 Fig). In line with motility assays, the relative transmigration frequencies of infected DCs, and non-infected DCs, across a transwell porous-membrane were significantly reduced in low extracellular Ca2+ and reconstituted upon addition of Ca2+ at physiological concentration (Fig 1E). Next, we assessed the motility of infected DCs in the presence of NiCl2, which blocks plasma membrane Ca2+ channels. NiCl2 dose-dependently reduced the velocity of hypermotile infected DCs, that reached velocities comparable with non-infected DCs (Fig 1F). Altogether, the present data indicate that the T. gondii-induced hypermigratory phenotype of DCs is dependent on the entry of extracellular Ca2+ through plasma membrane Ca2+ channels.
We have previously established that infection by T. gondii induces motility-related GABAergic signaling pathways in DCs [8]. Because hypermotile Toxoplasma-infected DCs exhibited dependency on Ca2+ and the established links between GABA receptor activation and Ca2+ responses in neuronal cellular systems [13, 14], we tested whether GABAA receptor activation led to Ca2+ responses in DCs. Perfusion of GABA elicited cytosolic Ca2+ elevations in DCs, visualized by fluorescent Ca2+ indicators (Fig 2A and S1 Video). Stimulation of DCs with GABA led to a simultaneous and transient Ca2+ influx (Fig 2B and 2C) in ~ 20% of the tested DC population at a given time point and, for the reference stimulus ATP, in ~ 42% of DCs (S1 Table). Ca2+ transients induced by GABA had relatively similar longevity and relatively lower amplitude than responses to ATP (Fig 2B and 2C), which were in line with ATP responses previously characterized in various types of DCs [19, 20]. Upon repeated stimulations with GABA and at varying GABA concentrations, consecutive Ca2+ responses were observed in individual cells (S2 Fig). Altogether, the data is in line with the previously recorded GABA-induced membrane potential changes by patch-clamping [8] and demonstrate that GABA stimulation of DCs is followed by influx of Ca2+ and transiently increased cytosolic Ca2+ concentration.
Next, we sought to determine if the GABA-induced Ca2+ signaling in Toxoplasma-infected DCs had an impact on hypermotility. First, we determined GABA secretion by infected DCs and the deprivation of GABA upon pre-incubation with GABAergic inhibitors (SC/SNAP) using MALDI mass spectrometry analysis of cell supernatants. In supernatants from Toxoplasma-infected DCs, spectra displayed a distinct peak signal (m/z 104,2; Fig 3A) corresponding to the signal of protonated GABA [M + H]+ chemical grade analytical standard (S3 Fig) [21]. Inhibition of GABA synthesis and secretion (SC, SNAP inhibitors, respectively) selectively reduced the m/z 104,2 peak signal (Fig 3A) and abrogated the hypermotility of T. gondii-infected DCs, which was reconstituted by addition of exogenous GABA (Fig 3B and 3C). This provided further specificity to previously reported elevations of GABA secretion in Toxoplasma-infected DCs, as quantified by GABA-ELISA under the same conditions [8]. Next, to test the impact of Ca2+ influx in DC hypermotility under GABA-deprived conditions, a cell membrane Ca2+ channel/ L-type VDCC agonist (BayK8644) was added to the cells. Importantly, the abrogated hypermotility of infected DCs, generated by GABAergic inhibition, was rescued by addition of BayK8644 (Fig 3B and 3C). A moderate but significant increase in cell motility was also observed in naïve DCs in presence of BayK8644 (Fig 3B and 3C). We conclude that, upon GABAergic inhibition, Ca2+ channel (VDCC) agonism leading to Ca2+ entry in DCs can reconstitute hypermotility in Toxoplasma-infected DCs.
VDCCs are known to respond with Ca2+ permeability to membrane potential changes. Because GABAA receptor activation by GABA elicits membrane potential changes in Toxoplasma-infected DCs [8] and GABA elicited Ca2+ influx (Fig 2), we investigated the putative involvement of VDCCs in DC hypermotility. L-type VDCC inhibition by nifedipine abolished hypermotility (Fig 4A) and significantly reduced transmigration (Fig 4B). In sharp contrast, inhibition of purinergic Ca2+ receptors by PPADS at high concentrations [22] had non-significant effects on hypermotility and transmigration of infected DCs (Fig 4A and 4B), despite that activation of purinergic receptors by ATP caused a significant Ca2+ influx and increased cytosolic Ca2+ levels in DCs (Fig 2). This indicated that VDCC-related effects governed hypermigration. We therefore explored further the function of VDCCs in relation to GABAergic signaling.
We previously reported that inhibition of GABA synthesis (SC) and/or transport (SNAP) significantly reduced GABA secretion and transmigration of T. gondii-infected DCs [8]. Extending these observations, addition of exogenous GABA rescued the hypermotility of infected DCs under GABAergic inhibition (Fig 4C and 4D). In sharp contrast, VDCC inhibition by nifedipine treatment caused a significant decrease in the motility of infected DCs that was not restored by exogenous GABA (Fig 4C and 4D), indicating implication of L-type VDCCs downstream of GABAergic signaling.
At resting membrane potential VDCCs are normally closed and, respond with Ca2+ permeability upon membrane depolarization. To relate the effect of GABAergic signaling to that of membrane depolarization, we treated GABA-deprived infected DCs with the depolarizing agent KCl. Upon blockade of GABA synthesis and secretion, KCl treatment fully restored hypermotility in Toxoplasma-infected DCs (Fig 4E), thus mimicking the effects obtained by addition of exogenous GABA (Figs 4D and 3C). Importantly, hypermotility was not restored by KCl in the presence of the L-type VDCC inhibitor nifedipine (Fig 4E). Taken together with the effects of VDCC agonism (Fig 3), these data demonstrate a link between L-type VDCCs and the hypermigratory phenotype of T. gondii-infected DCs downstream of GABAergic signaling.
In order to determine putative VDCCs mediating the nifedipine-sensitive GABA reconstitution effect, we performed a screen of VDCCs expressed in DCs. RT-PCR analyses indicated transcriptional expression of the L-type VDCC Cav1.3 in DCs, similar to brain homogenate (Fig 5A). A real-time quantitative PCR (qPCR) screen of VDCCs confirmed a consistent high relative expression of Cav1.3 transcripts in 6 mice tested over time, and also less abundant relative expression of Cav2.2 (Fig 5B and 5C). Other VDCC types, e.g. Cav1.1, Cav1.4, Cav2.1, Cav3.1, exhibited low, undetectable or inconsistent relative expression (Fig 5B and 5C). In Toxoplasma-challenged DCs, Cav1.3 remained the predominantly expressed VDCC type over other types (Fig 5D and S4A Fig) and maintained transcriptional expression of Cav1.3 in Toxoplasma-infected DCs related to non-challenged DCs was observed during 24 h infection (S4B Fig). Western blot analyses detected polypeptides (≈ 250 kDa) in DCs, corresponding to Cav1.3 expression as previously characterized in primary astrocytes [23], and with similar relative expression in DCs and Toxoplasma-infected DCs (Fig 5E and S4C Fig). Immunocytochemistry using a mAb to a predicted sub-membranous Cav1.3 epitope yielded a distinct fluorescence signal in non-infected and in infected permeabilized DCs (Fig 5F). Altogether, we conclude that Cav1.3 was the predominantly expressed VDCC in murine bone marrow-derived DCs and that the relative VDCC expression profile varied between mice or varied over time. Upon Toxoplasma-infection, Cav1.3 remains the predominant transcriptionally expressed VDCC.
To functionally assess the relative contribution of Cav1.3 to hypermigration in relation to other putatively expressed VDCCs, we took advantage of a pharmacological antagonist with high specificity for Cav1.3, CPCPT [24], and a broad inhibitor of L, N and T type VDCCs, benidipine [25]. Both inhibitors similarly abolished the hypermotility of infected DCs (Fig 5G and S4D Fig). While CPCPT significantly reduced transmigration of DCs from different mice, benidipine was a more consistent abrogator of transmigration (Fig 5H). Jointly, these data suggest that VDCCs play a significant role in T. gondii-induced hypermotility of DCs. As Cav1.3 appeared to be the most abundantly expressed VDCC, these data suggested that CPCPT and benidipine might act primarily on Cav1.3.
To test the functional implication of Cav1.3 in T. gondii-induced hypermotility, we employed an RNA interference approach. First, transduction efficacy by the recombinant lentiviral vector was optimized in the murine neuroectodermal cell line NE-4C and in primary DCs (S5 Fig). Cav1.3 (shCav1.3) and Cav1.2 (shCav1.2) were successfully targeted in NE-4C cells by this approach (S6A, S6B and S6C Fig). Similarly, in primary DCs, shRNA targeting Cav1.3 (shCav1.3), Cav1.2 (shCav1.2) or control shRNA (shLuc) was delivered and the transduced DCs were challenged with T. gondii tachyzoites (Fig 6A). DCs transduced with shCav1.3 exhibited significantly reduced Cav1.3 mRNA expression, with non-significant effects on Cav1.3 mRNA expression by shCav1.2 and control shRNA (Fig 6B). Western blotting analyses of DCs and NE-4C cells transduced with shCav1.3 showed a reduction in Cav1.3 protein expression (Fig 6C, S6D Fig).
Because primary DCs may become activated by the lentivirus and activation may impact on motility, we assessed expression of IL-12 mRNA in primary DCs and the NE-4C line. While the expression of IL-12 mRNA was relatively unaffected in NE-4C cells, primary DCs exhibited enhanced expression of IL-12 mRNA upon lentiviral transduction, in a similar fashion for shLuc, shCav1.2 and shCav1.3 (S7 Fig). We conclude that Cav1.3 mRNA and protein expression were selectively reduced in DCs exposed to shCav1.3 and that lentiviral transduction generates enhanced IL-12 mRNA expression in primary DCs.
To assess the impact of Cav1.3 silencing on hypermotility, we first optimized the approach using the murine DC line (JAWS II). JAWS II cells and DCs expressed a similar VDCC profile, with Cav1.3 as the most prominently expressed VDCC (S8A Fig) and a similar inhibitory profile by calcium blockers on hypermotility was observed (S8B Fig). JAWS II transduced with shCav1.3 (S8C Fig) exhibited significantly reduced Cav1.3 mRNA expression and enhanced IL-12 mRNA expression (S8D and S8E Fig). Importantly, shCav1.3-tranduced primary DCs (Fig 6D) and JAWS II exhibited reduced motility upon Toxoplasma-challenge. Their velocities reached non-significant differences compared with baseline motility of non-infected DCs (Fig 6E) and JAWS II, respectively (S8F Fig). Significant differences in the reduction of motility were observed for shCav1.3-tranduced DCs compared with shCav1.2-, shLuc- and mock-transduced DCs (Fig 6E). In line with results obtained upon pharmacological L-type VDCC inhibition (Fig 4D), exogenous GABA restored motility in mock-treated GABA-inhibited DCs but failed to restore motility in the shCav1.3-transduced cells (Fig 6F). We conclude that selective silencing of Cav1.3 abolishes T. gondii-induced hypermotility in DCs.
We have previously shown that adoptive transfer of T. gondii-infected DCs in mice leads to rapid dissemination of parasites and to exacerbation of the infection compared to infection with free tachyzoites [5, 26], and that GABAergic inhibition blocks this exacerbated dissemination [8]. To assess if VDCC inhibition impacted on parasite loads, benidipine pre-treated infected DCs were adoptively transferred to mice intraperitoneally. When the infections were monitored by in vivo bioluminescence, photonic emissions indicated dissemination of parasites to spleen and mesenteric lymph nodes (MLN) (Fig 7A and 7B). Plaquing assays of homogenized whole organs (non-perfused) revealed overall reduced mean parasite loads in the spleens of mice challenged with infected DCs (+) benidipine compared to mice challenged with infected DCs (-) benidipine (Fig 7C). To analyze the contribution of parasites in the blood circulation to the total parasite loads in organs, parasite loads were analyzed after blood perfusion. Perfused spleens exhibited overall reduced parasite loads, and reduced or abolished differences in parasite loads between the benidipine-treated and the non-treated conditions (Fig 7C). This showed that both removal of blood and benidipine treatment had a reducing impact on parasite loads in the spleen. In contrast, blood perfusion yielded more discrete relative reductions of parasite loads in MLNs, which are indirectly linked to the blood circulation via the lymphatic system (Fig 7D). In the perfused mice, parasites were consistently detected in all brains from day 4 versus day 3 in non-perfused mice (Fig 7E), indicating a contribution of parasites that were displaceable by blood perfusion to the total parasite loads in non-perfused mice.
Because benidipine-treatment had an impact on splenic parasite loads early during infection and this effect appeared linked to parasites in blood, we analyzed the fate of parasites and DCs within 24 h post-inoculation intraperitoneally. Upon benidipine pre-treatment of infected DCs, significantly reduced parasite numbers were measured in spleen by 24 h (Fig 8A), with a non-significant reduction of parasites in blood and non-significant differences in peritoneum (Fig 8A). Similarly, flow cytometry analyses identified reduced numbers of cell-associated GFP-expressing parasites (GFP+) in the spleen upon benidipine treatment (Fig 8B and S9A Fig), in line with the observed differences by plaquing assays and qPCR (Fig 8A). When adoptively transferred infected DCs were pre-labeled with a cell dye (CMTMR), CMTMR+ GFP+ cells were detected in the spleens and also CMTMR- GFP+ cells (S9B Fig). This indicated direct transport to the spleen by infected DCs and also rapid transfer of parasites to new leukocytes in peritoneum and spleen. Benidipine treatment yielded non-significant effects on DC viability, infection frequencies and parasite viability (S10 Fig). Altogether, the data show that adoptively transferred infected DCs rapidly entered the circulation and that VDCC inhibition led to reduced numbers of parasite-associated cells in spleen during the early phase of infection. VDCC inhibition delayed the appearance of parasites in circulation and, thereby, also the systemic dissemination of T. gondii.
In this study we investigated the molecular signaling mechanisms that govern how T. gondii hijacks the migratory properties of DCs. Building on previous work showing that a hypermigratory phenotype sets in within a few minutes after T. gondii invasion of DCs [4] and depends on GABAergic signaling [8], we addressed the role of Ca2+ signaling in these processes.
Our studies establish that Ca2+ signaling in murine bone marrow-derived DCs is indispensable for T. gondii-induced hypermotility and transmigration in vitro. The observation that the onset of the hypermigratory phenotype was abrogated at sub-physiological extracellular Ca2+ concentrations or by blocking plasma membrane Ca2+ channels underpinned a role for membrane-bound Ca2+ channels. However, Ca2+ mediates signal transduction to multiple cellular pathways. It was therefore crucial to determine its putative interaction with the GABAergic system of DCs. We previously showed that inhibition of GABA synthesis, GABA secretion or GABAA receptor blockade in Toxoplasma-infected DCs abolishes hypermigration [8]. Here, we demonstrate that hypermotility and transmigration are restored in GABA-deprived infected DCs by (i) addition of exogenous GABA, (ii) by cell membrane depolarization with KCl and (iii) by L-type VDCC agonism. Consequently, (iv) L-type VDCC blockade hindered reconstitution of hypermotility by GABA and KCl. This pinpointed a role for VDCCs downstream of GABAergic signaling. Further, inhibition of purinergic Ca2+ channels (P2 receptors) yielded non-significant effects on hypermotility despite a measurable Ca2+ influx in response to ATP. Also, while Ca2+ -deprivation led to a similar proportional reduction of motility of unchallenged DCs (baseline motility) and infected DCs (hypermotility), selective L-type VDCC inhibition abolished hypermigration but had non-significant effects on the baseline motility of DCs. Altogether, this indicated that L-type VDCCs primarily mediated the GABA-evoked motility-related Ca2+ influx and that extracellular Ca2+ influx per se into the cell or increased cytosolic Ca2+ levels per se was not sufficient to induce hypermigration.
To our knowledge, the findings demonstrate for the first time that murine DCs express the L-type VDCC subtype Cav1.3, with a functional implication in motility. Cav1.3 appeared to be the predominant transcriptionally expressed VDCC in primary DCs, a feature also maintained by the DC line JAWS II. Importantly, silencing of Cav1.3 by shRNA or selective pharmacological antagonism of Cav1.3 abrogated the hypermigratory phenotype in Toxoplasma-infected DCs, while baseline motility and morphology of DCs remained intact related to mock-treated and non-infected DC. A caveat of lentiviral transduction in primary DCs is that the lentivirus vector may have activation effects on the DCs [27], yet without reported apparent inhibitory effects on functionality [28], thereof the requirement of appropriate control experiments. We validated and confirmed our results in two additional cell lines. IL-12 mRNA expression indicated activation by the lentiviral vector primarily in DCs, to a lesser extent in JAWS II and, non-significant effects on the NE-4C line. Silencing of Cav1.3 expression abolished the hypermigratory phenotype, in contrast to Cav1.2 silencing. This, together with its apparent predominant expression, attributes a primary role in Toxoplasma-induced hypermotility to the VDCC subtype Cav1.3. However, despite that we did not observe compensatory up-regulation of other VDCCs/Cav1.2 upon Cav1.3 silencing in DCs, the data do not exclude a contributive role for other VDCC subtypes. In fact, the neuronal VDCC family members often display overlapping functions in mediating signal transduction [29]. This may apply to VDCCs in murine DCs too, as relative variations in transcription of several VDCC subtypes were detected in different mice over time, yet conserving a relative predominant expression of the subtype Cav1.3. Altogether, the data at hand defines a role for Cav1.3 in Toxoplasma-induced DC hypermotility and establish Ca2+ as a second messenger to GABAergic signaling in DCs.
We have previously shown that GABA induces GABAA receptor-activated currents in DCs [8]. Here, we demonstrate that DCs can sense membrane voltage changes caused by depolarization (KCl or GABA) and can respond to GABA by a Ca2+ transient. The analogous hypermotility restoration effect of exogenous GABA and depolarization by KCl, together with the opposite effects of the VDCC inhibitors (benidipine, nifedipine, CPCPT) and the agonist BayK8644 (a structural analog of nifedipine with positive inotropic activity) strongly suggests that GABA mediates membrane depolarization-induced Ca2+ release via VDCCs. Also by analogy to findings in neurons [12], it is plausible that GABAA receptors and chloride homeostasis are regulated by cation-chloride co-transporters in DCs. Altogether, our findings provide evidence of a direct link between GABA receptor signaling, Cav1.3 activation and hypermotility.
Although modulated by Toxoplasma infection, functional GABAA receptors appear to be constitutively expressed by murine and human DCs [8]. The effects of GABA on Ca2+ signaling via VDCCs / Cav1.3 has not been previously addressed in immune cells [30]. However, Ca2+ channels mediate some of the most rapid biological processes described and VDCC signaling allows for immediate cellular responses to external stimuli [31]. This is in agreement with the features attributed to the hypermigratory phenotype of Toxoplasma-infected DCs [3]: for example, its rapid onset, cytoskeletal remodeling and switch to amoeboid-type of migration within minutes after T. gondii invasion of the host DC in vitro [4] and is also in line with the observed rapid migration of adoptively transferred DCs in vivo [8, 26].
It has been previously reported that VDCCs may play a role in DC maturation [18] and T cell activation [32]. VDCC-related activity on DCs has been implicated in engulfment of apoptotic bodies, IL-12-production and up-regulation of major histocompatibility complex II [17, 18], all of which are important immune functions of DCs. In line with these observations, the hypermigratory phenotype induced by T. gondii appears to rely on receptors and channels expressed by naïve DCs [8], but additionally requires the active invasion of a T. gondii tachyzoite [4], which is confirmed here by the observation that exogenous GABA per se is not motogenic on naïve DCs [8] (while VDCC agonism is). Altogether, this also advocates that T. gondii primes the host cell for responsiveness to GABA and is consistent with the idea that GABAergic activation occurs in an autocrine fashion with minimal by-stander effect [8]. Notably, the vast majority of GABA-responding DCs also responded to ATP or to consecutive stimuli with GABA ranging from micromolar to millimolar concentrations, indicating that GABA does not render the DCs refractory to other Ca2+-related stimuli and that intracellular Ca2+ homeostasis is rapidly restored. We cannot exclude the involvement of additional mechanisms for Ca2+ entry in DCs [33] acting sequentially or in parallel. However, their possible contribution to the hypermigratory phenotype should be secondary or posterior to Cav1.3 activation, as silencing of Cav1.3 in both primary DCs and the DC cell line JAWS II or selective pharmacological inhibition of Cav1.3 [24] abrogated T. gondii-induced hypermotility. The posterior involvement of intracellular Ca2+ stores is also likely.
Ca2+ also controls a number of critical processes in apicomplexan parasites, including gliding motility, cell invasion and egress [34–36]. It is unlikely that these mechanisms play in the interpretation of our results as inhibitors were added posterior to parasite invasion and non-significant effects were observed on parasite viability, reinvasion after egress or after forced release from treated host cells. On the other hand, our observations suggest that, through activation of the GABAergic system, T. gondii modulates the Ca2+ homeostasis of the infected host cell, albeit transiently and locally. Induction of Ca2+ signaling offers the advantage of bypassing transcriptional regulation in the host cells and thereby accelerating effector functions, i.e. rapid migratory activation of the invaded DC and, thereby, dissemination. We have previously shown that the onset of T. gondii-induced hypermotility precedes chemotactic responses in DCs in vitro and that, after the onset of chemotaxis, GABA/GABAA receptor-mediated hypermotility and CCR7-mediated chemotaxis can cooperatively enhance the migration of infected DCs in vitro [4, 8]. Thus, Ca2+ entry in DCs, secondary to GABAergic activation, could hypothetically also influence Ca2+-dependent chemotaxis, with propagation of the signal to intracellular Ca2+ stores. Future research needs to determine if Cav1.3 is involved in the cytoskeletal rearrangements that accompany the onset of hypermotility, some of which are independent of GABAergic signaling [4], e.g. the dissolution of adhesion-related podosomes [4] and the switch to amoeboid-like high velocity migration [37].
Our data demonstrate that VDCC inhibition in adoptively transferred infected DCs delays the dissemination of T. gondii tachyzoites in mice. VDCC inhibition reduced the parasite numbers in circulation and in the spleen early after inoculation, likely by delaying the outmigration of infected DCs from the peritoneal cavity [38]. The data advocates that the early presence of parasites in blood is important for setting the parasite loads in mice and that VDCC inhibition delayed this process. In line with this, perfusion experiments showed that the circulating pool of parasites contributes to the total parasite loads in organs and to dissemination during acute infection. Also, the high variability in leukocyte-associated parasitemias between mice 24 h post-inoculation is in contrast with the lower variability of parasite loads in the organs later during infection, and may indicate that parasitemia is intermittent early after infection. Because the spleen is an early site of T. gondii replication during acute infection [39], this mobilizable pool of parasites (by blood perfusion) may be important for the systemic dissemination observed at later time points. Also, DCs and monocytic cells are parasitized early during infection [26, 38, 40] and, both leukocyte-associated tachyzoites [41] and extracellular (free) tachyzoites are detected in blood later during acute infection (day 4) [41]. Our data show that adoptively transferred infected DCs reach the circulation and spleen rapidly but also that the transfer of replicating tachyzoites to new leukocytes is rapid and can occur in the peritoneal cavity, in line with previous observations [42, 43]. Toxoplasma tachyzoites replicate in adoptively transferred DCs with lysis of infected DCs occurring within 48 h [5] and VDCC inhibition did not abrogate this process. This, together with the observed absence of parasites in brain parenchyma before day 4, indicates that it is unlikely that the adoptively transferred DCs transported parasites into the brain parenchyma. Rather, the observed delay in penetration to the parenchyma upon benidipine-treatment may be a consequence of delayed or lower parasitemias. Yet, DCs infiltrate the brain parenchyma during toxoplasmic encephalitis [44] and transportation of parasites to the brain by CD11b+ leukocytes has ben shown [40]. However, more recent findings show that replication of tachyzoites in the endothelium is necessary before passage to the brain parenchyma [41]. Our studies contribute to elucidating the role of infected DCs in circulation and their impact on systemic dissemination, which indirectly impacts on parasitic loads in the brain parenchyma, but do not specifically address the mechanisms of passage of T. gondii tachyzoites across the blood-brain barrier [8]. Jointly, mounting evidences show that Toxoplasma utilizes combined strategies for systemic dissemination [6], by hijacking leukocytes [5, 38, 40] and as free parasites [26, 41], and also with significant differences between Toxoplasma genotypes [26, 45]. Additionally, intracellular localization of tachyzoites in migratory leukocytes may offer a safe intracellular niche for replication and delivery to organs and vasculature.
To the best of our knowledge, this constitutes the first report showing that the VDCC signaling axis can be utilized by an intracellular pathogen to modulate host cell migration and potentiate systemic dissemination. Based on the data at hand, we propose a model for the initiation of the hypermigratory phenotype in DCs by T. gondii, mediated by GABAergic signaling and with Ca2+ acting as a second messenger (Fig 9). Initially, tachyzoite invasion triggers activation of the GABAergic system—GABA synthesis, transport and activation of GABAA receptors. Autocrine secretion of GABA by parasitized DCs leads to a membrane depolarization that activates the VDCC Cav1.3, with entry of Ca2+ as a result. Finally, entry of Ca2+ activates downstream signaling pathways that lead to cytoskeletal rearrangements and hypermotility. Mounting evidence indicates that, rather than being passively transported, intracellular microorganisms induce refined molecular orchestrations to manipulate the signaling pathways that modulate the migration of infected immune cells [6, 46, 47]. Continued investigations into how intracellular pathogens manipulate host cell Ca2+ signaling pathways may identify new targets for inhibiting processes associated to pathogenesis.
The Regional Animal Research Ethical Board, Stockholm, Sweden, approved experimental procedures and protocols involving extraction of cells from mice (N135/15, N78/16), following proceedings described in EU legislation (Council Directive 2010/63/EU).
Mouse bone marrow-derived DCs were generated and typified as previously described [8]. Briefly, cells from bone marrow of 6–10 week old C57BL/6 mice (Charles River) were cultivated in RPMI 1640 with 10% fetal bovine serum (FBS), gentamicin (20 μg/ml), glutamine (2 mM) and HEPES (0.01 M), referred to as complete medium (CM; all reagents from Life Technologies), and supplemented with 10 ng/ml recombinant mouse GM-CSF (Peprotech). Medium was replenished on days 2 and 4. Loosely adherent cells were harvested on day 6. The murine DC line JAWS II (CRL-11904) and murine neuroectodermal cell line NE-4C (CRL-2925) were cultured as indicated by the supplier (American Type Culture Collection). Primary astrocytes (ACs) were generated from cortices from 1–3 day-old C57BL/6 mice as previously described [48]. Freshly egressed Toxoplasma gondii tachyzoites of the RFP-expressing PRU-RFP [49] or GFP- and luciferase-expressing PTGluc [39] lines, kept on a 2-day passage cycle in murine fibroblast monolayers (L929, Sigma-Aldrich), were used in assays.
γ–aminobutyric acid (GABA), Adenosine triphosphate (ATP), (S)-SNAP-5114 (SNAP), semicarbazide (SC), nifedipine, Bay K8644, (all from Sigma-Aldrich), pyridoxalphosphate-6-azophenyl-2′,4′-disulfonic acid (PPADS), (4R)-rel-1,4-Dihydro-2,6-dimethyl-4-(3-nitrophenyl)-3,5-pyridinedicarboxylic acid 3-methyl 5-[(3R)-1-(phenylmethyl)-3-piperidinyl] ester hydrochloride (benidipine hydrochloride, all from Tocris) and 1-(3-Chlorophenethyl)-3-cyclopentylpyrimidine-2,4,6-(1H,3H,5H)-trione (CPCPT, Merck Millipore) were used at the indicated concentrations.
Motility assays were performed as previously described [4]. Briefly, 105 DCs were incubated with freshly egressed tachyzoites (MOI 3, 4 h). The cells were mixed with collagen I (0.75 mg/ml, Life Technologies) and transferred to a chamber slide (Nalge Nunc Internat.) or 96-well plate. Imaging was performed for 1 h, 1 frame/min, at 100x magnification (Zeiss AxioImager). Time stacks were stabilized (Image Stabilizer, ImageJ) and motility data obtained by manual tracking of cells (Manual Tracking, ImageJ) of approximately 50–60 cells per condition. In infected samples, only cells where the RFP and DIC signals co-localized were tracked. Transmigration assays were performed as previously described [8]. Briefly, 106 DCs were incubated with freshly egressed tachyzoites (MOI 3, 6 h), transferred into transwell filters (8 μm pore size; BD) in duplicate and incubated over night. Transmigrated DCs were quantified using a Neubauer hemocytometer. Ca2+-free medium was prepared from Ca2+-free DMEM, 1% FBS, gentamicin (20 μg/ml), glutamine (2 mM), 1 mM EGTA and HEPES (0.01 M), all reagents from Life Technologies.
DCs (2x105) were seeded on 5% 3-aminopropyltriethoxysilane coating glass bottom dish and incubated at 37°C with 5% CO2 for 15 min. DCs were then loaded with 2 μM Fluo-8H/AM (AAT Bioquest) in CM at 37°C with 5% CO2 for 15 min, and washed with Krebs-Ringer’s solution (150 mM NaCl, 6 mM KCl, 1.5 mM CaCl2, 1 mM MgCl2, 10 mM HEPES and 10 mM D-glucose) with 5% FBS. Time-lapse imaging was performed 2.5 s/frame, at 37°C with 5% CO2 on 200x magnification (Zeiss LSM 780 microscope equipped with a definite focus function). Cells were perfused with Krebs-Ringer’s solution with 5% FBS via a peristaltic pump (0.5 ml/min), which was also used to deliver pharmacological agents. The signals from individual cells were analyzed with ImageJ (version 1.46r, ROI Multi Measure). Each trace was normalized against the minimum value of all time points and a responding cell was defined as a signal exceeding 20% above baseline.
To determine the expression of the CaV1.3 protein in DCs and ACs, cells were lysed in RIPA buffer (150 mM NaCl, 50 mM Tris, 0.1% Triton, 0.5% deoxycholic acid, 0.1% SDS) with protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific) followed by sonication, addition of 4 x laemmli sample buffer and boiling. Proteins were separated by 8% SDS-PAGE, and blotted onto PVDF membrane (Millipore), blocked in 2.5% BSA followed by Western blotting with monoclonal anti-CaV1.3 C-terminal (Abcam), anti-GAPDH (Millipore) and anti-rabbit HRP (Cell signaling). Proteins were revealed by enhanced chemiluminescence (GE Healthcare) in a BioRad ChemiDoc XRS+.
DCs (105) were plated on poly-L-lysine coated coverslips and incubated with T. gondii tachyzoites (MOI 3, 4 h). Cells were fixed in 4% paraformaldehyde and permeabilized (0.1% Triton X-100), before incubation with mouse monoclonal anti-CaV1.3 biotin (1:100, Abcam) and streptavidin-Alexa555 (1:500, Molecular Probes). Samples were treated with DAPI and imaged by confocal microscopy (Zeiss LSM780).
Total RNA was extracted using TRIzol reagent (Life Technologies). First-strand cDNA was synthesized using Superscript III Reverse Transcriptase (Life Technologies). Real time quantitative PCR (qPCR) was performed in triplicates using SYBR green PCR master mix and a 7900HT Fast Real Time PCR system (Applied Biosystems). Products were analyzed with ABI 7900HT Sequence Detection System (Applied Biosystems) or Rotor gene (Corbett). 2-ΔCt values are used to calculate the relative expression levels of 9 VDCC subtypes, with TATA box binding protein (TBP) as reference gene (S2 Table). For quantification of Cav1.3 knock-down, glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and actin were used as reference genes (S2 Table). For quantification of T. gondii in tissues, the B1 gene was used (S2 Table).
Self-complementary hairpin DNA oligos targeting the CaV1.2 (Cacna1c) mRNA, CaV1.3 (Cacna1d) mRNA, and a non-related sequence (luciferase, Luc) were chemically synthesized (DNA Technology, Denmark), aligned and ligated in a self-inactivating lentiviral vector (pLL3.7) containing a CMV-driven EGFP reporter and a U6 promoter upstream of cloning restriction sites (HpaI and XhoI) [50] (S3 Table). Restriction enzyme analysis and direct DNA sequencing confirmed the correct insertion of short hairpin RNA (shRNA) sequences. Lentivirus production was done using lipofectamine transfection. Briefly, shCaV1.2, shCaV1.3 or shLuc vectors were co-transfected with psPAX2 packaging vector and pCMV-VSVg envelope vector into Lenti-X 293T cells (Clontech) and the resulting supernatant was harvested after 60 h. Recovered lentiviral particles were centrifuged to eliminate cell debris, filtered through 0.45-mm cellulose acetate filters and concentrated by ultracentrifugation. Titers were determined by infecting Lenti-X 293T cells with serial dilutions of concentrated lentivirus. NE-4C cells, JAWS II cells and DCs (day 3) were transduced by spinoculation at 1000 g for 30 min in presence of hexadimethrine bromide (Polybrene, 8 μg/ml; Sigma Aldrich). Three to 5 days post-transduction, EGFP-expression was verified by epifluorescence microscopy before the cells were used in experiments. Transduction frequency was defined as the number of EGFP-expressing cells related to the total numbers of cells in five representative fields of view.
DCs (105) were incubated with freshly egressed T. gondii PRU tachyzoites (MOI 3, 4 h). Cells were washed twice and incubated for 16 h in Krebs-Ringer’s solution supplemented with MEM essential and non-essential amino acids (Life Technologies) and 20 μg/ml gentamicin, referred to as mod. R. Inhibitors were present before and after the washes. 1 μL of cell supernatants were overlaid with 1μl of matrix (2.5 mg α-Cyano-4-hydroxycinnamic acid (HCCA) dissolved in 50% acetonitrile, 47.5% H2O, 2.5% TFA). Samples were analyzed by MALDI TOF mass spectrometry (Microflex LT, Bruker Daltronics) at laser frequency 60 Hz, mass range 0–1000 m/z, delayed ion extraction 100 ns, acceleration voltage 20 kV, lens voltage 6 kV and calibrated using the mass of HCCA matrix ions. Analysis was performed with flexAnalysis (version 3.3, Bruker Daltronics).
Cells were collected from blood, peritoneum and spleen and depleted of red blood cells. Cells were then stained for CD11b (clone M1/70), CD11c (clone N418), CD19 (clone 1D3), NK1.1 (clone PK136), CD3 (clones 145.2C11) and live/dead marker Viability Dye eFluor 780 (eBioscience) or Fixable Yellow Dead Cell Kit (Invitrogen) following blocking of Fc receptors (24G2). All antibodies were from Biolegend (San Diego, CA). After 30 minutes incubation, the cells were washed extensively and then fixed prior to running on FACCyAN ADP LX 9-colour flow cytometer (Beckman Coulter, Pasadena, CA). Data were analyzed using FlowJo software (Tree Star Inc, OR).
Adoptive transfers were performed as previously described [8]. Briefly, DCs were challenged with freshly egressed PTGluc tachyzoites (6 h, MOI 3). Extracellular parasites were removed by centrifugation. Following resuspension in RPMI, tachyzoite-infected DCs or freshly egressed tachyzoites were adoptively transferred intraperitoneally into recipient C57BL/6 mice. Total number of colony-forming units (cfu) injected into animals was confirmed by plaquing assays. Benidipine (40 μM) was added to DCs for the last 3 h of the 6 h challenge with tachyzoites and replenished (40 μM) prior to injection in mice. When indicated, cells were stained with CMTMR following manufacturer´s instructions (Invitrogen).
Eight-10 week old C57BL/6 mice were inoculated i.p. with freshly egressed PTGluc tachyzoites, or with PTGluc-infected DC ± benidipine. 3 mg D-luciferin potassium salt (Caliper Life Sciences, Hopkinton, MA, USA) was injected i.p. and mice were anesthetized with 2.3% isoflurane prior to BLI. Ten min after injection of D-luciferin, biophotonic images were acquired for 180 s (medium binning) with an In Vivo Imaging System (Spectrum CT, Perkin Elmer). For ex vivo imaging, organs are extracted and assessed as above. Analysis of images and assessment of photons emitted from a region of interest (ROI) was performed with Live Imaging Software (version 4.2; Caliper Life Sciences).
Plaquing assays were performed as described [8]. Briefly, organs were extracted and homogenized under conditions that did not affect parasite viability. The number of parasites was determined by plaque formation on fibroblast monolayers. When indicated, tachyzoites where released from infected DCs by repeated passages through a hypodermic needle (gauge 27), previous to plaquing. Transcardial blood perfusion was performed by injection of 25 ml PBS in the left ventricle after incision of the right atrium. Peritoneal lavage was performed by intraperinoneal perfusion and aspiration of 10 ml PBS using a hypodermic needle.
Statistical analyses were performed using R Stats Package version 3.0.2 (R Foundation for Statistical Computing, Vienna, Austria). Normality was tested by the Shapiro-Wilks test. P-values > 0.05 were defined as non-significant.
|
10.1371/journal.pntd.0004408 | A Mycobacterial Perspective on Tuberculosis in West Africa: Significant Geographical Variation of M. africanum and Other M. tuberculosis Complex Lineages | Phylogenetically distinct Mycobacterium tuberculosis lineages differ in their phenotypes and pathogenicity. Consequently, understanding mycobacterial population structures phylogeographically is essential for design, interpretation and generalizability of clinical trials. Comprehensive efforts are lacking to date to establish the West African mycobacterial population structure on a sub-continental scale, which has diagnostic implications and can inform the design of clinical TB trials.
We collated novel and published genotyping (spoligotyping) data and classified spoligotypes into mycobacterial lineages/families using TBLineage and Spotclust, followed by phylogeographic analyses using statistics (logistic regression) and lineage axis plot analysis in GenGIS, in which a phylogenetic tree constructed in MIRU-VNTRplus was analysed. Combining spoligotyping data from 16 previously published studies with novel data from The Gambia, we obtained a total of 3580 isolates from 12 countries and identified 6 lineages comprising 32 families. By using stringent analytical tools we demonstrate for the first time a significant phylogeographic separation between western and eastern West Africa not only of the two M. africanum (West Africa 1 and 2) but also of several major M. tuberculosis sensu stricto families, such as LAM10 and Haarlem 3. Moreover, in a longitudinal logistic regression analysis for grouped data we showed that M. africanum West Africa 2 remains a persistent health concern.
Because of the geographical divide of the mycobacterial populations in West Africa, individual research findings from one country cannot be generalized across the whole region. The unequal geographical family distribution should be considered in placement and design of future clinical trials in West Africa.
| Tuberculosis is caused by bacteria belonging to the Mycobacterium tuberculosis complex (MTBc), which consists of seven major, phylogenetically distinct lineages and their families. West Africa is the only region in the world where, besides the common M. tuberculosis lineages, the two M. africanum lineages are endemic. We demonstrate that the composition of the mycobacterial population in the western part of West Africa significantly differs from the one in the eastern part. This documented variation will impact on generalizability and interpretation of clinical trials outcomes. Therefore future trial designs need to consider the geographical diversity of underlying mycobacterial populations.
| West Africa consists of 15 countries with 245 million inhabitants (S1A Fig), 13 of which belong to the world’s 42 countries with the lowest human development index [1]. Consequently, it faces great challenges in controlling infectious diseases, such as tuberculosis (TB). Clinical trials investigating the local health needs are much needed to understand and tackle the TB epidemic in West Africa.
The composition of the endemic mycobacterial population infecting human study subjects can have a major impact on TB clinical trial outcomes and should ideally be accounted for in the planning phase of any project [2]. Considering bacterial variation between study sites is also essential to estimate to what extent country-specific results can be generalised to the whole of West Africa.
The MTBc can be divided into six major lineages, comprised of the Indo-Oceanic (L1), East-Asian (L2), Central Asian (L3), Euro-American lineages (L4) and the two endemic African lineages M. africanum West Africa 1 (MAF1, L5) and M. africanum West Africa 2 (MAF2, L6) [3]. Although MAF1 seems to be disappearing in some countries, the longitudinal development of MAF2 is not known. Each of these phylogenetically distinct lineages can be further differentiated into mycobacterial families, such as, amongst others, the Latin-American-Mediterranean (LAM) or Haarlem families within the Euro-American lineage [3].
Interestingly and for reasons not understood, West Africa is the only region in the world in which all of the six major human lineages are present. This exceptional diversity necessitates future West African trials to be adjusted for this unique bacterial variability—even more than trials in other parts in the world. Therefore the scope of the present publication was to describe the geographical distribution and spatial variations of mycobacterial families across the region.
We searched Pubmed using terms “spoligotype”, “spoligotyping” with respective country names. Studies on pulmonary TB up to December 2014 were included, in which spoligotypes on all isolates were available. Individual spoligotypes designated as mixed infections were excluded. In case several publications analysed the same dataset, the most comprehensive collection was selected. M. bovis studies, conducted in high risk populations (abattoir staff) were excluded. To assign mycobacterial families to isolates, and to ensure comparability between different datasets, we re-analysed extracted spoligotype information using a standardized approach. Isolates were classified into families using the online platform “Spotclust” at the default settings. For M. africanum isolates, Spotclust identifies, but does not distinguish between MAF1 and 2. Therefore “TBLineage” was further applied to M. africanum isolates previously identified by Spotclust [4]. Both Spotclust and TB Lineage are mathematical algorithms that were shown to reliably identify mycobacterial lineages and families based on respective signature spoligotype patterns. A detailed description of the algorithms and their performance is described elsewhere [4,5]. The lineage/family distribution per country/study site was plotted as chloropleth maps generated using QGIS 2.0.1 (http://qgis.osgeo.org).
To investigate geographical differences in mycobacterial families across West Africa we split West Africa into a Western and an Eastern region. Western countries include Gambia, Guinea-Bissau, Guinea, Sierra Leone, Ivory Coast, Mali, Senegal, while Eastern countries include Benin, Burkina Faso, Ghana, Niger and Nigeria (S1 Fig). With region as response variable, the proportion of each family was tested univariately using logistic regression, with country fitted as a cluster to account for multiple studies per site.
Families found in one region and not in the other cannot be modelled mathematically because the maximum likelihood for these families does not exist. We defined families with complete separation between regions as ‘perfect predictors’.
A two-sided p-value <0.05 was considered statistically significant and a two-sided p-value ≥0.05 & <0.10 was considered of borderline significance. No adjustment was made for multiple testing. All analyses were performed using Stata v12.1 (StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX: StataCorp LP.).
Phylogeographic analysis using linear axis analysis in GenGISvs2.2.2 was conducted [6]. The default GenGIS Africa map was used. A UPGMA phylogenetic tree was constructed from spoligotyping data (S2 Fig) using the publicly available MIRU-VNTRplus software [7] and uploaded into GenGIS allowing for the re-ordering of leaf nodes. A Linear axis plot (10.000 permutations) was run at significance level p = 0.001.
Gambian isolates, collected within a TB Case Contact cohort, in which all cases of the Greater Banjul area are recruited [8] were spoligotyped. Genotyping was approved by the Gambian Government/MRC joint ethics committee. Longitudinal lineage data was modelled using logistic regression for grouped data. The outcome was the number of a particular lineage out of the total number of samples taken in each year. Both lineage and year were fitted as explanatory variables and interactions between the two explored. The multicollinearity between the lineages was avoided by excluding one lineage and fitting the model on the remaining.
Of 20 original research articles, four were excluded (based on above criteria), with the remaining 16 covering 12 of 15 West African countries. In total we collected, extracted and (re)analysed spoligotype information of 3580 isolates, belonging to six major human lineages, of which the Euro-American lineage (L4), together with M. africanum lineages (L5 and 6) were the main causes of pulmonary TB (Table 1). Thirty-two different mycobacterial families were identified, but 84% of all patients are infected by only eight major families (Fig 1).
Common to most of the countries is the “ill-defined” T1 family. We also confirmed the previously described geographical distribution of two M. africanum lineages [24]. While MAF1 (L5) has the highest presence in Nigeria/Benin, MAF2 (L6) is mainly found in Gambia/Guinea-Bissau. Besides MAF2 as a major cause for TB, a variety of Euro-American families (Haarlem 1 and 3, LAM9, amongst others) are prevalent in western West Africa. This is in sharp contrast to eastern West Africa where, besides MAF1, the great majority of TB infections is attributable to only one other dominant family LAM10. A recently introduced family into West Africa is the Beijing family which lead to an outbreak in Cotonou, Benin [25]. The only other place with comparably high numbers of Beijing isolates is Dakar in Senegal. Both cities, Dakar and Cotonou have major international ports.
To evaluate whether identified families are geographically equal, we divided West Africa into a Western and Eastern region (S1B Fig). Univariate logistic regression analysis showed that the proportion of mycobacterial families can serve as predictors for the two regions. 13 out of 32 families were associated with one of the two regions (see S1 Table). Amongst these were four of the eight major families: LAM10 (perfect predictor at proportion ≥0.12) and MAF1 (p = 0.08) as predictors for the East and Haarlem 3 (p = 0.07) and MAF2 (p = 0.09) for the West.
To verify the geographic separation of these four major families, which cause 51% of all TB, we carried out an independent phylogeographic analysis using GenGIS software (Fig 2). We constructed an UPGMA tree based on 279 unique Haarlem 3, MAF1/2 and LAM10 spoligotypes, which was superimposed onto geographic locations and mycobacterial family distributions of the study sites (Fig 2A). In case of geographical separation, one expects significantly less crossings between the phylogenetic tree and the spoligotype distribution in the study sites than by mere chance. A linear axis analysis (p<0.001, 10.000 permutations) identified several orientations of the tree’s geographical axis that resulted in less than the 9759.5 crossings expected by chance. Fig 2B demonstrates that geographical separation occurs at various geographical axis angles, with the least crossings (9144) at 228.1° (Fig 2A). Although spoligotyping might have led to minor misclassifications of MAF1/MAF2 isolates in our phylogenetic analyses (S2 Fig), we expect such misclassification to result in an unbiased underestimation of the observed geographical separation.
1164 consecutive TB patients were recruited between 2002–2010. Logistic regression modelling revealed both a non-significant lineage by time interaction (p = 0.38) and a non-significant time main effect (p = 0.80). Our analysis therefore indicated that the proportions of lineages are stable over time. The overall lineage percentages, in order of magnitude, are Euro-American 57.2% (95%CI 54.4%-60.0%), MAF2 35.4% (95%CI 32.7%-38.2%), Indo-Oceanic 4.3% (95%CI 3.3%-5.6%), East Asian (Beijing) 2.5% (95%CI 1.7%-3.6%), MAF1 1.0% (95%CI 0.4%-2.4%) and East African Indian 0.8% (95%CI 0.2%-3.2%) (Fig 3).
We confirmed that modern Euro-American strains are the predominant lineage followed by the two M. africanum lineages. Although the polyphyletic T1 family [26] is rather equally distributed across the whole region, we find geographical variations of other families. While western West Africa shows a high genetic diversity from a multitude of mycobacterial families, the MTBc of Eastern West Africa is mainly composed of two dominant families (LAM10 and MAF1). Although other West and Central African countries observed a replacement of MAF1 and MAF2 with modern strains [10,11,14,27], our longitudinal analysis from The Gambia did not confirm these findings and MAF2 remains an important cause of TB in the country. The exact mechanism of how MAF2 can maintain a stable prevalence of 35% over the last decade within The Gambia (despite a slower progression to disease when compared to M. tuberculosis [28]) is not fully understood.
Besides the known geographical divide of the two M. africanum lineages, we find for the first time geographical separation of major Euro-American families in West Africa. Due to this spatial variation previous research findings observed in one West African country/region are hardly generalizable to the sub-region. In addition, the unequal distribution has important implications for design of future trials. For instance, western West African countries with their high genetic diversity are appropriate settings for research that aims to test whether novel diagnostics or vaccine candidates work equally well against different MTBc families. In contrast, research on host genetics, benefitting from low diversity, would yield more robust results when conducted in eastern West Africa with predominant LAM10 and MAF1 families. To investigate the spreading of novel TB families, one can follow up on the geographical expansion of LAM10 or on recently introduced Beijing strains into Benin or Senegal. As first studies confirmed that the “ill-defined”T1 is not a monophyletic clade [26], further research using more robust phylogenetic markers could focus on understanding the endemic MTBc composition T1-endemic countries.
The presented phylogeography also has limitations: first, we combined genotypic information, independent from respective collection strategies ranging from convenience to systematic sampling. Therefore data presented are a cross-sectional compilation of genotyping information between 1986–2012. Also, individual patient’s treatment history, whether they presented as new or retreatment cases, was not systematically collected and has not been accounted for. In order to avoid over-interpretation of results, we agree that comparing differing sampling strategies is challenging, and we therefore limited our discussion to proportions of families with larger isolate numbers. Lastly, the families themselves consist of a multitude of strains characterised by specific spoligotypes (shared international types, SITs) and we did not study whether the local expansion of a family was driven by one or several individual proliferating SIT within the family.
Spoligotyping can be successfully used to assign the majority of mycobacterial isolates to one of the major mycobacterial lineages and their families [4]. We appreciate that classification of mycobacteria in West Africa would ideally be based on whole genome sequencing (WGS) data, however, limited bioinformatics capacity combined with financial and infrastructural constraints did not allow high-throughput sequencing in most resource-limited West African countries to date.
By summarizing available and novel data, we showed significant geographical variation of the MTBc, which will impact on the overall outcome of clinical trials in any specific region. With the generated data researchers can consider the demonstrated spatial variation in the planning stage of respective future clinical TB trials.
|
10.1371/journal.pntd.0002695 | Multilocus Sequence Analysis for Leishmania braziliensis Outbreak Investigation | With the emergence of leishmaniasis in new regions around the world, molecular epidemiological methods with adequate discriminatory power, reproducibility, high throughput and inter-laboratory comparability are needed for outbreak investigation of this complex parasitic disease. As multilocus sequence analysis (MLSA) has been projected as the future gold standard technique for Leishmania species characterization, we propose a MLSA panel of six housekeeping gene loci (6pgd, mpi, icd, hsp70, mdhmt, mdhnc) for investigating intraspecific genetic variation of L. (Viannia) braziliensis strains and compare the resulting genetic clusters with several epidemiological factors relevant to outbreak investigation. The recent outbreak of cutaneous leishmaniasis caused by L. (V.) braziliensis in the southern Brazilian state of Santa Catarina is used to demonstrate the applicability of this technique. Sequenced fragments from six genetic markers from 86 L. (V.) braziliensis strains from twelve Brazilian states, including 33 strains from Santa Catarina, were used to determine clonal complexes, genetic structure, and phylogenic networks. Associations between genetic clusters and networks with epidemiological characteristics of patients were investigated. MLSA revealed epidemiological patterns among L. (V.) braziliensis strains, even identifying strains from imported cases among the Santa Catarina strains that presented extensive homogeneity. Evidence presented here has demonstrated MLSA possesses adequate discriminatory power for outbreak investigation, as well as other potential uses in the molecular epidemiology of leishmaniasis.
| Molecular epidemiology of infectious diseases, which uses pathogen genetics to determine risk factors in the human population, is commonly employed to assist in outbreak investigation. While definitive genetic markers and techniques have been developed for several other bacterial, viral, and parasitic pathogens, the scientific community has yet to agree on an international standard for inter- and intra-species differentiation of Leishmania, the parasite that causes the disease leishmaniasis. As leishmaniasis represents one of the highest disease burdens among the neglected tropical diseases, development of molecular techniques, which allow for inter-laboratory comparability through international sequence databases, is imperative for moving forward with disease control. Based on the current standard technique employed for bacteria, the authors propose a panel of six genetic markers for multilocus sequence analysis (MLSA) for intraspecific differentiation of Leishmania braziliensis, the most widely distributed of the Leishmania species in Brazil. Using strains from a recent outbreak in the sub-tropical non-endemic southern Brazil in comparison with strains from eleven other Brazilian states, the authors provide a practical example of how this technique can be applied in a real world outbreak situation.
| Leishmaniasis, a vector-borne disease caused by protozoan parasites of genus Leishmania [1], represents one of the highest disease burdens among the neglected tropical diseases in developing nations [2]. While not often fatal like the visceral form, the cutaneous form of the disease contributes substantially to leishmaniasis disease burden as it requires a lengthy and costly treatment regimen, results in apparent scarring, and can progress to a severely disfiguring mucosal form [1]. In recent years, leishmaniasis outbreaks have been described with increasing frequency [3]–[5], including those in sub-tropical regions or regions not previously endemic across the global [6]–[8]. In Brazil, beginning in 2005, an outbreak of human cutaneous leishmaniasis occurred in the southern Brazilian state of Santa Catarina, where the disease had not been observed previously as endemic. Overtime, cutaneous leishmaniasis has emerged in the region with evidence of a continued transmission cycle [9]. The species responsible for this outbreak has been incriminated as Leishmania (Viannia) braziliensis [9], the most widely distributed Leishmania species in Brazil to date [10], [11]. However, many questions still remain regarding the outbreak, such as: is one main strain or various strains responsible for the outbreak; is the emergence of L. (V.) braziliensis in the region a recent event; and how are Santa Catarina strains related to other strains in Brazil? A wide range of molecular tools are available for the investigation of molecular epidemiology of leishmaniasis, but choosing which method and/or markers to use continues to be a challenge [12]. Particularly for New World species, open access databases based on gold-standard genetic markers have not been developed. Currently, outbreak investigation of leishmaniasis, mainly conducted for visceral leishmaniasis outbreaks caused by L. (Leishmania) donovani species complex [13], [14], commonly employs multilocus microsatellite typing (MLMT). This technique has been proven to discriminate at the intra-species level [15] with high discriminatory power and is useful for determining outbreak strain origin when a database of MLMT strains is available for the Leishmania species of interest [13]. At the present moment, an open access MLMT database for L. (V.) braziliensis, has not been developed. The high discriminatory power of this technique has its drawbacks depending on the type of epidemiological question or analysis. In some cases, almost 20 “different” genotypes can be identified in one focus [13], [16], [17]. Dividing the isolates into many different genotypes reduces the statistical power of analyses involving epidemiological variables, such as clinical and demographic characteristics of the patient. Such reductions in statistical power greatly reduce the ability of researcher to conclude the relationship of factors like clinical form and disease virulence with a particular genotype.
Thus, epidemiological tools with appropriate discriminatory power, increased reliability and inter-laboratory reproducibility and comparability urgently are required. With these characteristics in mind, the method of multilocus sequence analysis (MLSA) provides a promising alternative. Projected as the future gold standard species typing method [12], MLSA involves sequencing a panel of house-keeping gene loci based on the panel of enzymes used in MLEE [18]. Several markers of these conserved regions have already been described, including ten markers for L. (L.) donovani [19], [20], and six markers for New World species [18], [21]. However, for L. (Viannia) species, these studies have mainly focused on interspecies discrimination and phylogenetic/taxonomic analysis and have employed only up to four markers. Given the challenges described above, we propose a panel of six gene loci, including three new markers described here for the first time, as an epidemiological tool for investigation of L. (V.) braziliensis outbreaks. In the present study, the recent outbreak in Santa Catarina is used to demonstrate the applicability of this technique in outbreak settings. The overarching objective of this work will be to generate interest in the community of leishmaniasis investigators to create an international sequence database based on these gene markers, as well as other markers from the original MLEE panel, for a more comprehensive and unified investigation into the distribution and epidemiological characteristics of Leishmania species.
Ethical approval for the use of patient data and their respective sample was received from the UFSC Ethics Committee. CLIOC is a Depository Authority of the Ministry of the Environment [Fiel Depositária pelo Ministério do Meio Ambiente, MMA] (D.O.U. 05.04.2005). Following Resolution 21 (August 31, 2006 – CGEN/MMA), authorization was not required for usage of samples previously deposited in CLIOC since the samples were used for research purposes only and data were analyzed anonymously.
Leishmania (Viannia) braziliensis strains from eleven Brazilian states (n = 53) were obtained from the Leishmania Collection of the Oswaldo Cruz Institute (Coleção de Leishmania do Instituto Oswaldo Cruz- CLIOC) in Rio de Janeiro, Brazil, and strains from Santa Catarina (n = 33) were obtained from the cryobank of the Laboratório de Protozoologia of the Universidade Federal de Santa Catarina (UFSC), Florianópolis, Santa Catarina, Brazil. Patient data from Santa Catarina used in this study were investigated as part of routine reportable disease surveillance and collection procedures have been previously described in [9]. Santa Catarina isolates were deposited in CLIOC and subjected to MLEE characterization, according to routine procedures employed by CLIOC.
Leishmania promastigotes were cultured at 25°C in Schneider's medium supplemented with 20% heat-inactivated fetal bovine serum. DNA extraction was conducted using the Wizard DNA purification Kit (Promega, Madison, USA), according to manufacturer's instructions.
Amplification was performed for a panel of six housekeeping gene loci listed in Table 1. Primers and PCR conditions have been previously described for 6-phosphogluconate dehydrogenase (6pgd), manose-6-phosphate isomerase (mpi), isocitrate dehydrogenase (icd) [18] and for the heat shock protein 70 (hsp70) [22], [23]. Primers for mitochondrial malate dehydrogenase (mdhmt) and nuclear malate dehydrogenase (mdhnc) are described here for the first time. Both follow the reaction condition: for 50 µl, 0,2 mM of each primer, 100 mM Tris–HCl, pH 8.8; 500 mM KCl, 1% Triton X-100; 15 mM MgCl2, 0.25 mM deoxyribonucleotide triphosphate (dNTPs), 0.025 U FideliTaq/GoTaq polymerase and 50 ng DNA. Amplification conditions were 94°C for 2 min, followed by 34 cycles at 94°C for 30 s, 52°C for 30 s and 72°C for 1 min, with a final extension at 72°C for 5 min. PCR products were purified and subsequently sequenced with the same primers used in the PCR.
Consensus sequences were obtained and edited in the software package Phred/Phrap/Consed Version: 0.020425.c (University of Washington, Seattle, WA, USA) and only those with Phred values above 20 were used as contigs. Analyzed sequence fragment lengths for each marker are provided in Table 1. Contigs of all strains were mounted and aligned in MEGA4 (Molecular Evolutionary Genetics Analysis version 4) [24]. Ambiguous sites were divided into two of the possible alleles for all markers using the PHASE algorithm in DnaSP5 [25].
Clonal complexes (CC) were defined through BURST analysis in the software eBURSTv3 [26]. The BURST algorithm identified groups of mutually exclusive genotypes associated with a MLSA population and the founding genotype sequence within each group. Then, the algorithm provided the predicted descent from the founding genotype for all other genotypes [26], [27]. For this analysis, criterion for CC formation was fixed at the most stringent level with at least five identical alleles for the six loci defining a CC. Sequences which were not able to be grouped into a clonal complex remained in the analysis as unique sequences.
Haploid sequences rebuilt from the PHASE algorithm in DNAsp containing homozygous and heterozygous alleles were imported into STRUCTURE 2.3.4 (University of Chicago, Chicago, IL, USA) to investigate the population structure of the 86 samples of L. (V.) braziliensis based on the six MLSA loci. Using a Bayesian statistical approach, STRUCTURE applies a model-based clustering method to infer population structure and assign individuals to clusters based on multilocus genotype data [28]. Genetically distinct clusters (K) are identified based on the frequency of alleles, attributing the fraction of each genotype for each sample. In STRUCTURE, runs were performed using a burn-in period of 200,000 iterations followed by 600,000 running iterations. Runs were repeated three times to obtain data suitable for estimating the value of ΔK (defined as the rate of variation of the log likelihood of data between successive values of K), which provides the most likely K value for the data to be used in STRUCTURE HARVESTER [29]. STRUCTURE HARVESTER generates graphs for the change in the log of k and calculation of ΔK of STRUCTURE results, which were compared for choosing the K that best fit the data. Next, CLUMPP version 1.1.2 [30] was employed to align the multiple replicate analyzes of the same data set. Hierarchical analysis of two to seven K clusters was performed to define the assignment of borderline strains.
Based on clusters found in STRUCTURE, we used Microsatellite Analyser (MSA) [31] to estimate FST values and Genetic Data Analysis (GDA) version 1.1 [32] to calculate expected heterozygosity (He), observed heterozygosity (Ho), and inbreeding coefficient (FIS). Recombination analysis was performed in Recombination Detection Program (RDP) [33].
To view genetic relationships (phylogenetic network) among strains and differentiation provided by the six markers, the median-joining network was mounted in the program SplitsTree 4.0 [34]. The median-joining network was constructed using concatenated character nucleotide sequences with ambiguous sites for all loci and strains. Nodes of the network, representing individual or groups of strains, were labeled by size, color and/or year/location to reflect epidemiological variables associated with the patient from whom the strain was isolated.
Associations between genetic and epidemiological variables were analyzed in Stata SE 13 (StataCorp LP, College Station, TX, USA). Chi-squared test, or Fisher's exact test when appropriate, was used to assess the relationships between categorical variables. Maps were created in ArcGIS 10 (ESRI, Redlands, CA, USA).
BURST analysis identified three clonal complexes (CC) among the 86 strains of L. (V.) braziliensis, with over half (54.7%, 47/86) of the strains not belonging to any of the three CCs and remaining separate as unique sequence types (Supporting Information S1). A total of 76 distinct sequence types were observed among strains. The analysis was heavily weighted by the homogeneity and large number of strains from Santa Catarina included in the analysis, with the large majority (84.8%, 28/33) of Santa Catarina strains being grouped into one nearly exclusive clonal complex (CC1). Five out of six strains from Santa Catarina that did not group with CC1 were registered as imported cases in the epidemiological investigation. No association was found between CC and clinical form (p = 0.660). Figure 1 shows the geographical distribution of the CCs by state, revealing proportionally higher genetic variation in states from the Amazon biome (94.1% (n = 16/17) unique sequence types) (Supporting Information S1).
Through calculation of ΔK in the STRUCTURE analysis, the L. (V.) braziliensis strains included in the present study from 12 Brazilian states were found to best fit into three clusters (POP) (Supporting Information S1). Overall, 41.9% (36/86) of strains belonged to POP1, 40.7% (35/86) to POP2, and 16.3% (14/86) to POP3. As in the BURST analysis, the large majority (87.9%, 29/33) of Santa Catarina strains formed their own cluster (POP2), which also included four strains from Pernambuco, one from Mato Grosso and one from Bahia (Figure 2). The four Santa Catarina strains that did not cluster with POP2 were registered as imported cases in the epidemiological analysis. These four strains were the same strains from imported cases that did not cluster in the BURST analysis. Complete strain information can be found in Supporting Information S2.
As shown in Figure 3, POP1 demonstrated the most extensive geographical distribution, including strains from all states analyzed in this study. A distinction can be made between the genetic variation and genetic structure of coastal states, which contain Atlantic forest, and northern states, which are located in the Amazon basin. States of the Amazon region were predominately comprised of POP1 strains, while strains of POP2 and POP3 were mainly found in coastal states.
A significant association between the genetic cluster designated by STRUCTURE and leishmaniasis clinical form of the patient from which the strain was isolated was observed (p = 0.030) (Table 2). Most strains from cases presenting the mucocutaneous clinical form (4/7) belonged to POP3, including one case from Rio de Janeiro State, one from Pernambuco and two from Bahia.
Based on the scale for the interpretation of FST suggested by Wright (1978), the estimates showed significant genetic differentiation among the STRUCTURE clusters (Table 3). POP1 and POP3 showed moderate genetic differentiation (FST = 0.1087), while POP2 showed great genetic differentiation with POP1 and POP3 (FST = 0.1540 and 0.2028, respectively). POP1 had the highest average number of alleles per locus (23.3), while both clusters POP2 and POP3 were similar in mean number of alleles, being approximately five alleles per locus. Positive values of FIS were found for all clusters. FIS values for POP1 and POP3 were particularly high (Table 4). All loci were polymorphic for POP1 and POP2 and five (83.3%) of the six loci were polymorphic for POP3. The marker 6pgd was not polymorphic for POP3. In general, the new markers hsp70, mdhnc and mdhmt showed the highest number of alleles of 35, 40 and 44, respectively, in comparison to 15–30 alleles for the other three markers.
Results of the BURST and STRUCTURE analysis were found to be significantly associated (p<0.001) (Table 5). The majority (37/48) of unique sequences in the BURST analysis were forced into their own population (POP1) in the STRUCTURE analysis, representing mainly strains from the Amazon regions.
Recombination events were detected by seven algorithms in RDP software (p<0.05). However, neither the beginning nor ending breakpoints could be identified, which may have resulted in recombinant misidentification. Nonetheless, one sample from Santa Catarina (185) and one sample from Bahia (IOC/L 2871) were indicated as potentially parental or recombinant. Thirty-one samples from Santa Catarina had sequences with partial evidence of the same recombination event.
The median-joining network was created from concatenated sequences of the six gene loci for the 86 strains of L. (V.) braziliensis from Brazil. Majority of Santa Catarina strains presented as an evident cluster. Other strains close to the Santa Catarina cluster were from Pernambuco (n = 2), Rio de Janeiro (n = 1), and Pará (n = 1). When the nodes of Santa Catarina strains were highlighted by case origin, all cases not clustered with the principal cluster were imported cases, with the exception of strain 605 (Figure 4). This 605 strain also was grouped within the main Santa Catarina CC and POP in both the BURST and STRUCTURE analyses.
When the strains from cases of mucocutaneous and disseminated clinical form were highlighted, those from Bahia were clustered, while mucocutaneous cases from other Brazilian states appeared closer to the main cluster of Santa Catarina (Figure 5). When the median-joining network was reduced to only strains from Santa Catarina, the resulting network presented three principal branches. Marked by year and city of leishmaniasis case diagnosis, a main cluster can be observed in the center of the network, representing the epicenter of the outbreak which occurred in 2006 in the municipality of Blumenau (Figure 6). From this main epicenter, autochthonous cases branched separately, appearing to evolve over time and space to the neighboring municipality of the capital municipality of Florianópolis. The map in Figure 6 shows this main cluster of related Santa Catarina strains was distributed over a distance of 140 km in four years from Blumenau to Florianópolis.
Multilocus sequencing analysis (MLSA) was successful in detecting epidemiological patterns among L. (V.) braziliensis strains from twelve Brazilian states. Additionally, the technique was able to detect intra-species variation compatible with epidemiological characteristics within a specific outbreak focus, demonstrating the potential of this technique as a molecular tool for outbreak investigation.
In the BURST analysis, strains were found to group into three clonal complexes. Samples from the Amazon region presented largely as unique sequence types, demonstrating a proportionally higher level of heterogeneity in comparison to coastal states. This distinction is particularly apparent when compared to Santa Catarina. Since the BURST analysis permitted samples not to be grouped into a specific CC and remain as unique sequences, the STRUCTURE analysis was observed to force these unique sequences to form a genetic cluster. This was also evident in the significant association between the two analyses (p<0.001). POP1 was comprised of almost entirely unique sequence types. This high heterogeneity is characteristic of strains from the Amazon biome, as previously observed in other studies, and reflects the large variety of vectors and hosts in the region [10], [35], [36]. Furthermore, the emergence of L. (V.) braziliensis in the state of Santa Catarina, Brazil appears to be a recent event, given the high homogeneity observed among the analyzed strains. This conclusion is based on the assumptions of the Hardy-Weinberg equilibrium model of populations, which states if no evolutionary pressure mechanism, such as migration in or out of the population or mutation over a long period of time, is acting upon a given population, then the genetic frequencies will remain unaltered [37], [38]. Therefore, during the period from which the samples were collected during the outbreak, the strains were largely uninfluenced by outside strains, remaining as their own apparently unique population. This could also be caused by a specific transmission cycle in which other Leishmania strains or species were not easily incorporated. Specific vector-parasite relationships would remove the possibility of recombination given the major selective force on Leishmania populations occurs in the vector hosts during the development of the parasite [39].
MLSA utilizing the panel of six markers was able to distinguish epidemiological characteristics among L. (V.) braziliensis strains. In all three analyses (BURST, STRUCTURE, and median-joining), MLSA results were compatible with case origin evaluated in the epidemiological investigation of Santa Catarina strains. These results demonstrate the potential of this method for use in future outbreak investigations and surveillance. Despite being registered as an imported case in the epidemiological investigation, strain 605 was grouped within the main Santa Catarina cluster in all three analyses, pointedly suggesting this patient was most likely an autochthonous case. In such cases, the molecular characterization proved to be a more reliable and precise tool than the epidemiological interview to determine if a case acquired the infection locally or outside of a given region. Results also show the methodology possesses discriminatory power to differentiate imported and autochthonous cases at state macroregion levels. Knowledge on the origin of a case is important for predicting case outcome and treatment course, since several studies have shown a relationship between specific characteristics of the infecting parasite and geographical location with the outcome of the patient [9], [40], [41].
Along the same lines, MLSA showed a significant association between clusters in the STRUCTURE analysis and patient clinical form among the samples analyzed in the present study (p = 0.0296). However, the current study only included seven cases of the mucocutaneous form. A study involving a large representative sample of these cases with controls is necessary to validate these findings. Identification of a genetic marker of Leishmania virulence has not been identified at the present moment [42], and the identification of such a marker would have important clinical and pharmacological significance. Despite the limited number of samples in this study, this methodology could be promising for the identification of a specific L. (V.) braziliensis cluster predisposed to the mucocutaneous form, and therefore, warrants further investigation.
Recombination is often difficult to detect within species because of low inter-strain diversity and/or apparent low diversity due to inappropriate sampling [43]. However, RDP results of the present study were able to reveal recombination occurring between the L. braziliensis strains. This suggests the strains from Santa Catarina may be the result of a clonal expansion from a recombinant event, and the resulting strains then encountered proper conditions to propagate in the state. Previous studies on recombination, including a study on population genetics for inbreeding [44] and a previous MLSA phylogenetic study [18] specifically for L. braziliensis, were able to detect recombination signals as well. In these situations, homologous recombination may have been the responsible mechanism. This phenomenon also may have produced the well-structured clonal complexes in Leishmania in the present study which allowed for the epidemiological inferences to be made.
As no definitive set of markers for MLSA has been defined for the study of populations within a given species of Leishmania, the markers evaluated here could be defined as potential candidates in the panel used for this type of study. Interestingly, the three new markers, hsp70, mdhnc and mdhmt, were the most polymorphic of the six markers, suggesting their addition provided the increase in discriminatory power that allowed for intra-species differentiation. Taken together, these six markers provided adequate discriminatory power to answer epidemiological questions surrounding genetic clusters of a single species. An important benefit of MLSA is the ability to create and store sequences in an international database for global comparison of Leishmania species and strains [15]. The next step will be to determine the viability and discriminatory power of this six loci panel for other species of Leishmania and increase the number of markers and strains sequenced. Four of the markers (6pgd, mpi, icd and hsp70) have already proven to be discriminatory among species of the Leishmania subgenus Viannia, including L. (V.) shawi, L. (V.) lainsoni, L. (V.) naiffi and L. (V.) guyanensis [18], [21].
With the recent increase in development of genetic markers and new statistical methods for analyzing them, the choice of which software is most adequate to your specific analysis is becoming increasingly difficult. No definitive guidelines currently exist [45]. For this reason, we opted to evaluate our MLSA results from three different perspectives, using diverse software (BURST, STRUCTURE and Splitstree) to arrive at our inferences regarding the genetic structure among the L. (V.) braziliensis included in the present study. The BURST analysis, which is commonly used for MLSA of haploid organisms, such as bacteria, permitted a better comprehension of the genetic variability among the samples using conservative parameters for differentiating clonal complexes. As almost all Santa Catarina strains fit into one clonal complex and the remaining strains were mainly unique sequences, we can conclude the cluster in this state is highly homogeneous in comparison to other states. However, with over half of the strains not grouped in a clonal complex, comparison of genotypes with epidemiological factors was not possible. The STRUCTURE analysis forced all strains into a cluster, resulting in the grouping of all of these unique sequences into their own cluster. This phenomenon shows that, despite the high diversity among the samples from the Amazon region, strains from Santa Catarina continue to be genetically distinct from other Brazilian strains analyzed here. In other words, the diverse genetic clusters within POP1 of the Amazon region, as a whole, are still genetically more distinct from Santa Catarina strains than within themselves, as also shown by FST and FIS. Interestingly, our study found high positive FIS values (high inbreeding coefficients) among the populations of L. (V.) braziliensis, which negates the hypothesis of strictly clonal reproduction among Leishmania species. High FIS values have also been observed in various MLMT studies for Leishmania, including a study on L. (V.) braziliensis in Bolivia and Peru [46], a study on L. (L.) infantum in Old World and New World strains [47] and a study on L. (L.) donovani in Ethiopia [48]. In these studies, possible explanations of these high FIS values were the presence of considerable inbreeding and/or sub-structuring of the population, reflecting a possible Wahlund effect.
Despite being too complex for comparing all strains among themselves, the median-joining network was the best visual representation for comparing Santa Catarina strains with all other strains from Brazil. This type of analysis is most applicable in an outbreak situation in which strains from a specific area can be compared to other reference strains, allowing for the distinguishing of imported cases and other epidemiological differences. Overall, until software capable of addressing specific genetic Leishmania characteristics, such as infrequent recombination, is created, use of all three types of genetic analyses can be used as an alternative to provide a robust MLSA analysis.
This information on the genetic variability of circulating strains is important for public health and control efforts. Considering drug resistance and complications in treatment have not been observed in Santa Catarina cases, control of leishmaniasis in Santa Catarina where the parasite strains are genetically homogeneous would be expected to be much more efficient than in regions where the parasite presents genetic heterogeneity and a more complex transmission cycle. This factor emphasizes the need for more urgent and active control methods to prevent further introduction of Leishmania strains and/or species, as well as geographical spread of the disease.
MLSA revealed epidemiological patterns among L. (V.) braziliensis strains from twelve Brazilian states, even within the state of Santa Catarina where the strains presented extensive homogeneity. The addition of three markers, hsp70, mdhnc and mdhmt to the previously described panel of markers increased the discriminatory power of the technique, permitting the identification of three genetic clusters within L. (V.) braziliensis strains. All three analyses (BURST, STRUCTURE and median-joining network) provided a complementary and integral part in the interpretation of the MLSA results. When used in tandem with MLMT, these two methods could provide a more robust approach to the molecular epidemiology of leishmaniasis and increased validity of the population structure model. A prospective study design that seeks to include a representative sample of the patient population and active collection of their Leishmania strains is needed to validate this method as a molecular epidemiology tool. However, the present study has provided sufficient evidence of the effectiveness of this method for pursuing further validation of MLSA for leishmaniasis outbreak investigation.
|
10.1371/journal.pbio.2007050 | Coupling S-adenosylmethionine–dependent methylation to growth: Design and uses | We present a selection design that couples S-adenosylmethionine–dependent methylation to growth. We demonstrate its use in improving the enzyme activities of not only N-type and O-type methyltransferases by 2-fold but also an acetyltransferase of another enzyme category when linked to a methylation pathway in Escherichia coli using adaptive laboratory evolution. We also demonstrate its application for drug discovery using a catechol O-methyltransferase and its inhibitors entacapone and tolcapone. Implementation of this design in Saccharomyces cerevisiae is also demonstrated.
| Many important biological processes require methylation, e.g., DNA methylation and synthesis of flavoring compounds, neurotransmitters, and antibiotics. Most methylation reactions in cells are catalyzed by S-adenosylmethionine (SAM)–dependent methyltransferases (Mtases) using SAM as a methyl donor. Thus, SAM-dependent Mtases have become an important enzyme category of biotechnological interests and as healthcare targets. However, functional implementation and engineering of SAM-dependent Mtases remains difficult and is neither cost effective nor high throughput. Here, we are able to address these challenges by establishing a synthetic biology approach, which links Mtase activity to cell growth such that higher Mtase activity ultimately leads to faster cell growth. We show that better-performing variants of the examined Mtases can be readily obtained by growth selection after repetitive cell passages. We also demonstrate the usefulness of our approach for discovery of Mtase-specific drug candidates. We further show our approach is not only applicable in bacteria, exemplified by Escherichia coli, but also in eurkaryotic organisms such as budding yeast Saccharomyces cerevisiae.
| Methylation is the transfer of a methyl group from one molecule to another. Its importance in gaining bioactivity and acquiring bioavailability of drugs has been recognized by chemists for some time [1]. Chemical methylation ordinarily utilizes noxious reagents and generates toxic waste and often lacks regioselectivity [2]. In contrast, enzymatic methylation is specific, environmentally friendly, and safer to work with.
Most methylation in cells takes place by S-adenosylmethionine–dependent methyltransferases (SAM-dependent Mtases), using SAM as the methyl donor. SAM-dependent methylation is involved in many important biological processes, including epigenetics and synthesis of a wide range of secondary metabolites (e.g., flavonoids, neurotransmitters, antibiotics). In fact, SAM is one of the most commonly used cofactors in cellular metabolism, second only to ATP [3]. SAM-dependent Mtases have become an important enzyme category, used either as biocatalysts, as part of fermentative production pathways in biotechnical and chemical industries [2–4], or as drug targets in the pharmaceutical industry [5,6].
Implementing and engineering functional SAM-dependent Mtases is difficult since all existing assays lack robustness, are not cost effective, and are not generalizable to all types of Mtases or high throughput. Consequently, study and engineering of Mtases or building efficient methylation-dependent pathways is hard to achieve. We designed an in vivo synthetic selection system by coupling SAM-dependent methylation to growth via a homocysteine intermediate of the SAM cycle. This selection system design was first implemented in E. coli by deleting serine acetyltransferase (cysE) to prevent endogenous homocysteine and cysteine synthesis. We diverged homocysteine of the SAM cycle to cysteine using heterologously expressed yeast cystathionine-β-synthase (Cys4) and cystathionine-γ-lyase (Cys3) (Figs 1 and S1). Effectively, this design couples methylation to the conversion of exogenous methionine to the biosynthesis of cysteine, a required amino acid for growth. This growth-coupled design was computationally validated using a genome-scale metabolic model (Fig 1).
We demonstrate the use of our system to improve enzyme properties. Using adaptive laboratory evolution (ALE) with growth selection, we first achieved directed evolution of phenylethanolamine N-methyltransferase (Pnmt) using a non-natural substrate, octopamine (OCT). In recent years, ALE has emerged as a productive approach to address a wide range of biological questions [7–9]. We implemented an ALE-driven workflow to demonstrate our selection system because of the operational simplicity of ALE (serial passages), its ultra-high–throughput screening capability (over 10 million cells per passage), its ability to engage phenotype-driven in vivo evolution, and the affordability of DNA resequencing (Fig 2A). Following ALE, isolated strains were characterized and were subjected to full DNA resequencing. Mutations in E. coli cfa, involved in phospholipid synthesis, were identified in all non-growth–coupled isolates. The cfa gene encodes for a SAM-dependent Mtase, suggesting its role as a competing Mtase during ALE. On the other hand, a Pnmt (F214L) mutation was present in growth-coupled isolates, and a cell-based characterization showed that it led to approximately 2-fold activity improvement on synephrine (SYN) synthesis (Fig 2B).
Most often, Mtase substrates may not be readily available in large quantities or even membrane permeable in order to perform directed enzyme evolution in vivo, and it may then only be feasible to engage an active metabolic pathway. We thus applied our selection system to evolve a methylation-dependent pathway. The chosen candidate pathway was a de novo three-step melatonin biosynthesis pathway from 5-hydroxytryptophan (5HTP). It consisted of three enzymatic steps: decarboxylation (aromatic-amino-acid decarboxylase [Ddc]), acetylation (aralkylamine N-acetyltransferase [Aanat]), and methylation (acetylserotonine O-methyltransferase [Asmt]) (Fig 2C). By using ALE, Asmt was evolved in vivo, and three sequence variants (A258E, G260D, and T272A) were discovered (Fig 2D). All variants showed improved turnover compared to wild-type Asmt under physiological conditions, with the highest improvement observed in A258E (approximately 2.5-fold). It was additionally discovered that high levels of ddc expression from a plasmid caused genetic instability, and mutations in cfa could be seen in non-melatonin–producing cells, affirming its role as an unwanted sink for SAM in E. coli. Upon incorporating Asmt (A258E), a single copy of ddc, and cfa deletion in the background strain, Aanat was further evolved in the next ALE, and the D63G mutation was identified, leading to approximately 2-fold activity improvement (Fig 2D). These results demonstrated the usefulness of this growth selection system for directed evolution of enzymes or metabolic pathways when linked to a methylation reaction.
We next demonstrate the use of our system for drug discovery. SAM-dependent Mtases participate in many important cellular functions and are targeted by a number of drug development programs (such as DNA or histone Mtase inhibitors) [6]. We applied our selection system on catechol O-methyltransferase (Comt), a known drug target for treating Parkinson's disease [5]. Cells bearing human Comt were evolved to grow at high rates using ALE (Fig 2E). All isolates were growth-coupled to Comt activity. Resequencing results showed the comt gene did not acquire any mutations, while many isolates accumulated mutations on RpoC (such as A328P, E1146A, or E1146G), a subunit of E. coli RNA polymerase, suggesting a host factor effect. The suitability of using evolved cells to screen Comt inhibitors by growth was evaluated next by determining Z-factor in a 96-well format [10]. The Z-prime value was calculated to be between 0.87 to 0.97 when cells were grown for 3 h or more, indicating a high-throughput-screening (HTS)–compatible assay with large separation (Fig 2E and S1 Table). We then tested one evolved isolate with two known Comt inhibitors: entacapone and tolcapone, respectively. Both drugs reduced Comt-dependent cell growth at concentrations as low as 200 nM, with a slightly higher potency observed for tolcapone (Fig 2E). Both inhibitors were highly specific to Comt and showed no observable adverse effects on other cellular proteins (such as heterologous Cys3 and Cys4 or the essential E. coli proteins) when homocysteine was additionally supplemented, implying a general suitability of our selection system for in vivo Comt inhibitor screening (Fig 2E).
Lastly, we implemented our design in budding yeast S. cerevisiae. S. cerevisiae is an industrially important production host with growing interest for biobased production of value-added methylated products [11]. It is also a well-studied eukaryotic model organism expressing diverse cellular Mtases [12]. In contrast to E. coli, yeast is capable of synthesizing cysteine through reverse transsulfuration from homocysteine because of the natural appearance of the CYS3 and CYS4 genes (Fig 3A) [13]. Therefore, blockage of homocysteine biosynthesis from aspartate is required to enable the selection, and this was achieved by deleting the genes encoding homoserine O-acetyltransferase (MET2) and O-acetylhomoserine sulfhydrylase (MET17). Additional gene deletion of phosphatidylethanolamine methyltransferase (CHO2) and phospholipid methyltransferase (OPI3) required for phosphatidylcholine biosynthesis was performed to remove potential competing native Mtases for SAM [14]. In this quadruple knock-out strain, the heterologous caffeine synthase I gene (CCS1), encoding an N-Mtase from Coffea arabica acting on theobromine to synthesize caffeine, was introduced. The presence of Ccs1 conferred growth advantage when exogenous theobromine was supplemented compared to nonsupplemented cells, affirming the applicability of the design in yeast (Fig 3B). Control cells without Ccs1 expression showed similar growth regardless of theobromine supplementation (Fig 3C). Acknowledging the large number of native Mtases in yeast [12], this phenotype might be the result of the activity of remaining native Mtases for homocysteine synthesis required for growth.
In summary, we have designed and validated a methylation-dependent growth selection system for Mtases. Not only did this selection system lead to the discovery of causal mutations that improve enzyme properties of heterologous Mtases and acetyltransferase, but it also demonstrated usefulness in drug discovery and the identification of critical host factors. Further applications of this technology can be used to acquire in-depth understanding of Mtases (such as their evolutionary path and governing factors for substrate specificity) or to engineer pathways or host cells of metabolic engineering interests. We have implemented our design in E. coli and S. cerevisiae; however, the conceptual design of this selection is transferable to any organism with an active SAM cycle. Implementing this selection system in higher eukaryotic cells (such as mammalian cells) is particularly valuable for drug development since permeability and stability of promising drug candidates can be determined at early stages. Overall, the described selection is robust, generalizable, compatible with HTS, and widely applicable, and it is likely to become a useful and valuable tool for the chemical biology and metabolic engineering communities.
The E. coli BW25113 strain and derivatives were used throughout this study. Genome engineering of E. coli was facilitated by λ-red recombination, P1 transduction, and/or site-specific Tn7 transposon [15,16]. S. cerevisiae CEN.PK 102-5B (MATa) was used as background yeast strain for the engineered yeast strains. Genome engineering of S. cerevisiae was facilitated by CRISPR/Cas9 [17,18], and transformations were performed with the lithium acetate/single-stranded carrier DNA/PEG method [19]. A list of strains used is shown in S2 Table. Unless stated otherwise, all E. coli strains were maintained at 37°C in LB (Lennox) (Sigma Aldrich, St. Louis, MO, USA), 2xYT, or M9 media containing M9 minimal salts (BD Difco, BD, Franklin Lakes, NJ, USA), 2 mM MgSO4, 100 μM CaCl2, 500-fold diluted trace minerals (10 g/L FeCl3·6H2O, 2 g/L ZnSO4·7H2O, 0.4 g/L CuCl2·2H2O, 1 g/L MnSO4·H2O, 0.6 g/L CoCl2·6H2O, and 1.6 mM EDTA [pH 8.0]), 1× ATCC Vitamin Supplement (ATCC MD-VS), and 0.2% glucose (w/v). Unless stated otherwise, ampicillin and spectinomycin were used at 50 mg/L; chloramphenicol and kanamycin were used at 25 mg/L. Yeast strains were cultured on either rich yeast extract-peptone dextrose (YPD), yeast synthetic drop-out media (lacking the proper amino acids for selection) (Sigma Aldrich), or Delft medium [20] with 2% glucose. Delft medium was sterilized by filtration. All Δcho2- and Δopi3-derived strains were maintained on media supplemented with 1 mM choline chloride (Sigma Aldrich), except for YPD medium.
Plasmid DNA assemblies were performed by Gibson Assembly, USER cloning, or the EasyCloning method [20–22]. E. coli TOP10 (Invitrogen, Carlsbad, CA, USA) and DH5α were used for plasmid propagations. For propagation of yeast plasmids, 100 mg/L ampicillin was used. A list of plasmids used is summarized in S3 Table.
All ALE experiments were performed at 37°C using M9 supplemented with 50 mg/L methionine and substrates, unless stated otherwise. Cell passages were performed automatically according to previously described [23], and six lineages of the initial strain were usually maintained. OCT and PCA at 50 mg/L were supplemented for Pnmt and Comt evolution, respectively. The melatonin pathway was evolved in the presence of 100 mg/L of methionine and 100 mg/L of 5HTP. Antibiotics were not supplied during ALE.
The E. coli ECAH2 strain (ΔcysE), derived from JW3582 of the Keio Collection [24], was used as the host strain to test methylation-dependent growth in the presence of pHM11. The pHM11 plasmid harbored cys3 and cys4 from S. cerevisiae S288C. Homocysteine and Cys3/Cys4-dependent cysteine synthesis was demonstrated by transforming ECAH2 with pHM11. The transformed strain, ECAH3, was allowed to grow on an M9 plate containing 50 mg/L homocysteine at 37°C for 24 h to demonstrate homocysteine-dependent growth. The ECAH6 and ECAH7 strains were used for Pnmt and Comt evolution using ALE. All ΔcysE-derived strains were maintained on LB plates supplemented with 25 mg/L cysteine.
The E. coli HMP236 strain was initially used to evolve the melatonin pathway from 5HTP (S2 Table). It contained two plasmids, pHM11 and pHM12. Its parent strain was HMP221 with the following genome modifications: FolE (T198I), YnbB (V197A), ΔtnaA, ΔcysE, ΔmetE, and ΔmetH. Deletion of tnaA was to prevent 5HTP degradation. Deletion of metE and metH was to prevent a reverse methionine-to-homocysteine synthesis via methionine synthase encoded by both genes, but it was later determined not to be required.
The E. coli HMP579 was used for the subsequent melatonin pathway evolution. It carried two plasmids, pHM70 and pHM79. The pHM70 plasmid was modified from pHM11 with an insertion of Asmt (A258E). The second plasmid, pHM79, was not required in this study because of 5HTP feeding. Its parent strain was HMP553, carrying a single chromosomal copy of ddc and aanat. The ddc and aanat genes were introduced to the attTn7 site using pGRG25.
The S. cerevisiae SCAH124 strain, derived from S. cerevisiae CEN.PK102-5B (MATa), was constructed by sequential CRISPR/Cas9-facilitated full ORF markerfree knock-out of MET17, CHO2, OPI3, and MET2 mediated by homology-directed recombination using guide RNA (gRNA)-expressing plasmids PL_01_A2, PL_01_A3, PL_01_C8, and PL_01_E1 and two linear DNA fragments homologous to the flanking regions up- and downstream of the targeted ORF, as well as PL_01_A9 with CAS9. The following gRNA sequences were used for Cas9-targeting of indicated ORF and identified with the webservice CRISPy adapted for the S. cerevisiae CEN.PK genome sequence [25]: MET17 (5′-GATACTGTTCAACTACACGC-3′), CHO2 (5′-ACCACCTGTAACCCACGATA-3′), OPI3 (5′- GCAGAAACAACCAGCCCCGC-3′) and MET2 (5′- GTAATTTGTCATGCCTTGAC-3′). SCAH134 and SCAH138, derived from SCAH124 and harboring the Mtase gene CCS1 (PL_01_D2) and an empty vector (pRS415U), respectively, were used for demonstration of selection design in S. cerevisiae.
Total DNA was extracted using a PureLink Genomic DNA Kit (Invitrogen) and was processed either commercially by Beckman Coulter Genomics (Danvers, MA, USA) or in house. When prepared in house, DNA libraries were prepared using a Kapa Hyper Prep Library Prep Kit (Roche Molecular Systems, Pleasanton, CA, USA). DNA samples were sequenced using Illumina MiSeq or NextSeq.
Trimmomatic tool (v0.32-v0.35) was used for quality trimming of raw sequencing data with "CROP:145 HEADCROP:15 SLIDINGWINDOW:4:15 MINLEN:30" parameters [26]. Breseq (v0.27.1) was employed for variant calling on processed sequencing data with "-j 4 -b 20" parameters [27]. E. coli BW25113 genome sequence with NCBI accession CP009273 was used as reference along with other relevant parts and plasmids.
All compounds were purchased from Sigma Aldrich. PCA and its methylated product vanillic acid (VIII) were quantified using a Dionex 3000 HPLC system (Dionex, Sunnyvale, CA, USA) equipped with a Cortecs UPLC T3 column from Waters (Milford, MA, USA) and a guard column from Phenomenex (Torrance, CA, USA). The column temperature was set to 30°C, and the mobile phase consisted of a 0.1% formic acid and acetonitrile. Runtime was 11 min, including 4.2 min separation without acetonitrile, 1 min washing with 75% acetonitrile, and an additional 4.5 min run without acetonitrile. The flow rate was constant at 0.3 ml/min, and the injection volume was 1 μl. Both compounds were detected by UV at wavelengths 210, 240, and 300 nm as well as a 3D UV scan. HPLC data were processed using Chromeleon 7.1.3 software (Thermo Fisher Scientific, Waltham, MA, USA), and compound concentrations were calculated using calibration curves.
5HTP, serotonin (HT), acetylserotonin (AcHT), and melatonin were quantified using a Dionex 3000 HPLC system equipped with a Zorbax Eclipse Plus C18 column (Agilent Technologies, Santa Clara, CA, USA) and a guard column from Phenomenex. To achieve separation, the column was heated to 30°C, and the mobile phase consisted of a 0.05% acetate and a variable amount of acetonitrile. Runtime was 12 min, including 10 min of separation, whereas acetonitrile was reduced from 95% to 38.7% in 9.4 min. After holding 0.6 min, acetonitrile concentration was returned to 95% in 1 min and was held till the end of the run. The flow rate was set to 1 ml/min, and the injection volume was 1 μl. Elution of the compounds was detected by UV at wavelengths 210 nm, 240 nm, 280 nm, and 300 nm as well as a 3D UV scan. HPLC data were processed using Chromeleon 7.1.3 software (Thermo Fisher Scientific), and compound concentrations were calculated using calibration curves.
OCT was quantified using a Dionex 3000 HPLC system equipped with a Cortecs UPLC T3 column from Waters and a guard column from Phenomenex. The column temperature was set to 30°C, and the mobile phase consisted of 0.1% formic acid and acetonitrile. Runtime was 9 min, including 2.5 min separation without acetonitrile, 0.5 min washing with 70% acetonitrile, and an additional 5.5-min run without acetonitrile. The flow rate was constant at 0.3 ml/min, and the injection volume was 1 μl. OCT was detected by UV at wavelengths 210, 240, and 300 nm as well as a 3D UV scan. SYN was detected by LC-MS (Fusion, Thermo Fisher Scientific) in the positive full-scan mode using the same separation profile as OCT quantification. SYN was detected as [M + H]+ m/z 168.10191 with a mass accuracy of 2.2 ppm. Data were processed using Chromeleon 7.1.3 and X-calibur 4.1 from Thermo Fisher Scientific, and compound concentrations were calculated using calibration curves.
Growth of E. coli was measured using a Duetz 96-well low well system (Enzyscreen, Heemstede, The Netherlands) coupled to a humidified Innova 44 shaker (5 cm orbit) (New Brunswick Scientific, Edison, NJ, USA) at 37°C and 300 rpm. Seed cells were grown in 400 μl LB in the presence of appropriate antibiotics for 4–5 h in 96-well deep well plates. When transferred to M9, 10 μl of cells were added to 400 μl of M9 with 25 mg/L cysteine and antibiotics. After overnight growth, cells were added to 150 μl of fresh M9 with 50 mg/L of methionine and 50 mg/L methylation substrates to approximately 4% in a 96-well low well plate. Changes in optical density at 600 nm (OD600) were recorded using a SynergyMx microplate reader (BioTek Instruments, Winooski, VT, USA). Growth rates were calculated from an average of four independent biological replicates using KaleidaGraph 4.1.3.
Six biological replicates of S. cerevisiae SCAH134 and SCAH138 were inoculated from seed cultures to similar ODs for sulfur amino acid starvation in Delft medium supplemented with histidine, uracil, and choline chloride and incubated for approximately 24 h. The seed cultures were grown in yeast synthetic complete medium without leucine, supplemented with choline chloride, for approximately 25 h. Both replicate seed cultures and starvation cultures were cultured in a total volume of 500 μl in 96-deep well plates at 30°C/300 rpm. Upon completion of starvation, cells were transferred to Delft medium supplemented with histidine, uracil, choline chloride, 1 mM L-methionine, and with/without 1 mM theobromine (Sigma Aldrich) to similar ODs and to a final volume of 150 μl in a flat-bottomed microtiterplate. Two technical replicates were inoculated for each of the six biological replicates. Growth was recorded in a microtiterplate reader (ELx808 Absorbance Reader, BioTek Instruments) with continuous shaking at strong setting, at 30°C, and with recording of absorbance at 630 nm every 30 min. The cultures had incubated in the plate reader 30 min prior to the first reading at time 0 h. Growth curves were plotted using mean absorbance of the replicates for 30 h. Blank media values were not subtracted.
HL1818 was used to test the inhibition effect of entacapone and tolcapone. Cells were prepared for growth measurement as described above. The test growth medium was M9 with 50 mg/L PCA, 50 mg/L methionine, 1% DMSO, and various amount of inhibitors. The stock concentration of entacapone and tolcapone was 20 g/L dissolved in DMSO. To the positive controls, 50 mg/L homocysteine was additionally included so that effects of the drugs on E. coli cells (such as those other than Comt) could be determined. A drug concentration response curve was plotted using average OD600 after 6 h growth from three independent biological replicates. The EC50 values were calculated using OriginPro 2018b (version b9.5.5.409).
Measurements of SYN production were performed using a Duetz 96-well deep well system (Enzyscreen) coupled to an Innova 44 shaker (5 cm orbit) (New Brunswick Scientific) at 37°C and 300 rpm. HL1815 and HL1816 were used to measure SYN production (S1 Table). Seed cells were grown in LB in the presence of chloramphenicol for 4–5 h and thereafter grew in M9 overnight. Fresh M9 containing 200 mg/L OCT was inoculated with seed culture to 4%. These cells were transferred to a 96-well deep well plate, and each well contained 400 μl. 200 μl samples were withdrawn periodically for exometabolites analysis, while the remaining 200 μl cells were used to determine OD values using a SynergyMx microplate reader (BioTek). In vivo enzyme activity was averaged from four independent biological measurements normalized to dried cell weight. The conversion factor from OD to dried cell weight is 1 (i.e., 1 OD = 1 g/L) for our setup. It was observed that Pnmt activity was biomass dependent.
A Duetz 24-well deep well system (Enzyscreen) coupled to an Innova 44 shaker (5 cm orbit) (New Brunswick Scientific) was used. Physiological Asmt activity was determined by growing HMP231, HMP416, HMP416, HMP417, and HMP418 in 2 ml M9 supplemented with 100 mg/L AcHT at 37°C with shaking at 300 rpm. Samples were withdrawn periodically for exometabolites analysis using HPLC and OD measurements. Physiological Aanat activity was measured using HMP850 and HMP851 in the presence of 100 mg/L HT. In vivo enzyme activity was averaged from three independent biological measurements normalized to dried cell weight. It was observed that Asmt and Aanat activity was biomass independent.
Validation of the selection system design was performed using the most recent E. coli genome-scale metabolic model [28], which computes the flux states of the entire metabolic network. Following established procedures [29], the cysE gene was “knocked out” in silico by setting the upper and lower bounds of the metabolic reaction it catalyzes to 0. Metabolic reactions catalyzed by Cys3 and Cys4 were inserted into the metabolic model. Additionally, a methylation-dependent reaction was inserted into the model (Pnmt, Comt, or Asmt). The metabolic model was then solved for its flux state using linear programming by setting cell growth as the objective. All model simulations were performed using the python package COBRApy 0.7.0 in Python 2.7 [30].
|
10.1371/journal.pgen.1007953 | O-GlcNAcylation of PERIOD regulates its interaction with CLOCK and timing of circadian transcriptional repression | Circadian clocks coordinate time-of-day-specific metabolic and physiological processes to maximize organismal performance and fitness. In addition to light and temperature, which are regarded as strong zeitgebers for circadian clock entrainment, metabolic input has now emerged as an important signal for clock entrainment and modulation. Circadian clock proteins have been identified to be substrates of O-GlcNAcylation, a nutrient sensitive post-translational modification (PTM), and the interplay between clock protein O-GlcNAcylation and other PTMs is now recognized as an important mechanism by which metabolic input regulates circadian physiology. To better understand the role of O-GlcNAcylation in modulating clock protein function within the molecular oscillator, we used mass spectrometry proteomics to identify O-GlcNAcylation sites of PERIOD (PER), a repressor of the circadian transcriptome and a critical biochemical timer of the Drosophila clock. In vivo functional characterization of PER O-GlcNAcylation sites indicates that O-GlcNAcylation at PER(S942) reduces interactions between PER and CLOCK (CLK), the key transcriptional activator of clock-controlled genes. Since we observe a correlation between clock-controlled daytime feeding activity and higher level of PER O-GlcNAcylation, we propose that PER(S942) O-GlcNAcylation during the day functions to prevent premature initiation of circadian repression phase. This is consistent with the period-shortening behavioral phenotype of per(S942A) flies. Taken together, our results support that clock-controlled feeding activity provides metabolic signals to reinforce light entrainment to regulate circadian physiology at the post-translational level. The interplay between O-GlcNAcylation and other PTMs to regulate circadian physiology is expected to be complex and extensive, and reach far beyond the molecular oscillator.
| Circadian clocks are self-sustained, endogenous pacemakers that enable organisms to anticipate daily environmental changes and resource abundance to perform specific time-of-day activities and achieve optimal survival. Multiple time cues are interpreted by circadian clocks to facilitate synchrony between organisms and their environment. A large body of work have identified light and temperature as important zeitgebers. More recent works highlight the significance of metabolic cues as signals to entrain and modulate circadian clocks to drive proper rhythms of physiology and behavior. Metabolic input, primarily through clock-controlled feeding activity, can regulate circadian physiology through multiple pathways. Some of these pathways are unknown while others, such as the O-GlcNAcylation of clock proteins, are just emerging. In this study, we utilized mass spectrometry proteomics to identify O-GlcNAcylation sites of the Drosophila PERIOD (PER) protein, a key regulator of the clock, and performed site-specific functional characterization of PER O-GlcNAcylation. Our results support that PER(Ser942) O-GlcNAcylation, a nutrient-sensitive protein modification that is expected to be more abundant during feeding period, prevents newly synthesized PER from prematurely performing its function during daytime and therefore restricts its activity to nighttime when flies are fasting. This study provides new insights into the mechanisms linking nutrient input and circadian physiology.
| Circadian clocks are endogenous protein machines that integrate external time cues and internal metabolic states to impose temporal organization on physiology, metabolism, and behavior (reviewed in [1–2]). They allow organisms from all kingdoms of life, which experience perpetual 24-hour day-night cycles, to anticipate daily environmental changes and execute biological tasks, from molecular to behavioral levels, at the optimal time of day. Over the years, great progress in elucidating the molecular mechanisms driving circadian rhythms has been made by studying the core circadian oscillator. The molecular oscillator consists of two interlocked transcriptional translational feedback loops (TTFLs) that produce daily oscillations in clock mRNAs and proteins to drive rhythms in diverse cellular processes. During the day and into the early parts of the night, two basic helix-loop-helix (bHLH)-PAS transcription factors, CLOCK (CLK) and CYCLE (CYC; homolog of BMAL1 in mammals), activate transcription of their own repressors, period (per) and timeless (tim) and other clock-controlled output genes [3]. After a time-delay and TIM-assisted entry into the nucleus [4], PER interacts with and inhibits CLK-CYC activity [5]. This repression is relieved upon proteasomal-dependent degradation of PER during late night into early morning, thus initiating another round of CLK-CYC-mediated transcription [6–7]. Among CLK-activated genes are two bZIP transcription factors, vrille (vri) and par domain protein 1ε (Pdp1ε) [8–9]. Due to differential kinetics of VRI and PDP1ε protein accumulation, VRI accumulates first and inhibits Clk expression. As VRI level decreases, PDP1ε accumulates and activates Clk transcription, and the cycle of per/tim expression starts again. More recent evidence however suggested that the main role of VRILLE could be to control clock output by driving rhythms in expression of the neuropeptide PDF (Pigment Dispersing Factor) and neuronal arborization [10]. CLOCKWORK ORANGE (CWO) is another direct CLK target that feedbacks and represses CLK activity by competing with CLK-CYC complexes for E-box binding at circadian promoters [11]. The TTFLs are synchronized to the 24-hour day-night cycle through light-dependent degradation of TIM [12–14], which interacts with the photoreceptor CRYPTOCHROME 1 (CRY1) [15].
To extend the duration of the TTFL to last a 24-hr circadian cycle, post-translational regulation of core clock proteins overlays on the TTFL and has been recognized to be critical in maintaining the functionality of the circadian oscillator (reviewed in [1, 16]). In Drosophila, the phase-specific phosphorylation state of PER is closely linked to its time-of-day specific function and the speed of the oscillator [17–21]. De novo synthesized hypophosphorylated PER goes through a multi-site phosphorylation program that progressively increases its phospho-occupancy until it gets hyperphosphorylated. In particular, phosphorylation at a N-terminal phosphodegron targets PER for degradation in a proteasome-dependent manner [17]. In a study by Robles et al. [22], 25% of the 20,000 phosphosites identified in mouse liver proteins were found to oscillate over the circadian cycle. This suggests that widespread and dynamic oscillations in phosphorylation occur beyond core circadian transcription factors to transition cellular proteins between functional states to regulate circadian physiology.
More recently, O-GlcNAcylation has emerged as another PTM that can regulate the temporal function and activity of circadian transcription factors [23–25]. In contrast to protein phosphorylation, which is mediated by a wide selection of kinases and phosphatases, protein O-GlcNAcylation is regulated by a single pair of enzymes with opposing functions [26]. O-GlcNAc transferase (OGT) and O-GlcNAcase (OGA) facilitate the O-GlcNAcylation and de-O-GlcNAcylation of cellular proteins respectively. In Drosophila, PER and CLK have been identified as substrates of O-GlcNAcylation, and there is evidence from overexpression of ogt to support that O-GlcNAcylation of these clock transcription factors regulates their nuclear translocation, stability, and transcriptional activity [23–24]. In mammalian clocks, BMAL1 and CLOCK have been shown to be rhythmically O-GlcNAcylated over the circadian cycle, and O-GlcNAcylation functions to counteract ubiquitination to stabilize these proteins [25]. In a separate study, PER2 was shown to be modified by O-GlcNAcylation at the S662-S674 region, which is important for regulating clock speed via CK1 phosphorylation [24]. A S662G mutation in humans is known to cause the familial advanced sleep phase syndrome (FASPS) [27]. Interestingly, S662 can also be O-GlcNAcylated, suggesting in vivo interplay between phosphorylation and O-GlcNAcylation in this domain. Increasingly, the interplay between phosphorylation and O-GlcNAcylation is shown to be prevalent in the regulation of diverse cellular processes (reviewed in [1]). As O-GlcNAcylation is a nutrient-sensitive PTM that relies on the availability of UDP-GlcNAc, an end product of the hexamine biosynthetic pathway (HBP), it is expected that levels of cellular protein O-GlcNAcylation may be highly dependent on daily feeding-fasting cycles.
Many unknowns regarding PTM regulation of clock proteins remain, including the mechanisms by which phase-specific phosphorylation and O-GlcNAcylation collaborate to regulate their time-of-day functions. A significant barrier to understanding these mechanisms is the identification of O-GlcNAcylated residues. In fact, although both Drosophila PER and CLK are known to be O-GlcNAcylated, specific O-GlcNAcylated residues have not been identified [23–24]. The effects of O-GlcNAcylation on these clock proteins have only been investigated by global overexpression and knockdown of ogt and oga, which impact all O-GlcNAcylated residues simultaneously. Lessons learned from previous studies on clock kinases [17–21] highlight the likelihood that valuable mechanistic insights may be overlooked by global manipulation in PTM enzyme expression. Other important questions concern the temporal requirement of clock protein O-GlcNAcylation and the relationship between protein O-GlcNAcylation status and feeding-fasting cycles.
In this study, we sort to understand how metabolic input influences the O-GlcNAcylation status of PER and regulates its function. We used Mass Spectrometry (MS) proteomics to identify PER O-GlcNAcylation and phosphorylation sites from adult flies, and characterized the function of site-specific O-GlcNAcylation events in vivo. We focused on PER as its phase-specific function has been shown to be highly dependent on its progressive phosphorylation program over the circadian cycle. Investigating PER O-GlcNAcylation can therefore set the stage for understanding the interactions between phosphorylation and O-GlcNAcylation in regulating its circadian function. We observe that loss of O-GlcNAcylation at multiple residues affect PER repressor function. In particular, loss of O-GlcNAcylation at PER(S942), which is located in the PER-CLK interaction or CLK binding domain (CBD), leads to stronger PER-CLK interaction and premature entry into the circadian repression phase. Conversely, overexpression of OGT in clock neurons weakens PER-CLK interaction, contributing to its period-lengthening phenotype in locomotor activity rhythms. Finally, we report that daily rhythms of PER O-GlcNAcylation in adult head tissues correlate with feeding-fasting cycles. This correlation is expected to be even stronger in peripheral tissues, which are more sensitive to metabolic signals. Specifically, PER O-GlcNAcylation exhibits circadian rhythmicity and is higher during the day when flies are actively feeding. Our results suggest that metabolic input collaborates with other entrainment signals to regulate time-of-day PER function in circadian transcription.
Although O-GlcNAcylation has been shown to influence PER function, specific residues that are modified by O-GlcNAcylation have not been identified [23–24]. This represents a critical barrier to understand the function of site-specific O-GlcNAcylation events and the interplay between O-GlcNAcylation and phosphorylation to regulate the phase-specific functions of PER. We therefore sought to identify PER O-GlcNAcylation sites and obtain temporal data on their occupancy. We purified FLAG-tagged PER from heads of wper0; p{3XFLAG-per(WT)} flies at specific time-points over the circadian cycle using FLAG affinity purification and performed quantitative mass spectrometry (MS) to obtain a circadian profile of PER phosphorylation and O-GlcNAcylation using fly tissues as starting materials. Although we attempted to perform this study using both fly head and body tissues, we were only able to identify PER O-GlcNAcylation sites in head tissues since we were unable to pull down sufficient PER proteins in bodies for comprehensive PTM identification. Nevertheless, we postulate that PER residues that are O-GlcNAcylated in head tissues will also be O-GlcNAcylated in peripheral tissues due to the ubiquitous expression of ogt [28]. To enable quantitation of PTM sites, flies were fed with 15N-labeled (heavy) yeast or 14N (light) yeast for two generations to ensure complete labeling in flies (herein termed 15N and 14N flies) [29]. To ensure that 15N-fed and 14N-fed flies show similar behavioral rhythms, we examined their locomotor activity rhythms using Drosophila activity monitoring [30]. Both types of flies displayed strong behavioral rhythms with periods close to 24-hr in constant darkness (S1A and S1B Fig). Furthermore, we measured PER daily abundance in 15N and 14N flies to confirm that temporal expression of PER was not altered as a result of the diets and observed no difference (S1C Fig). To profile PER PTMs in vivo, 15N and 14N flies were collected at six time-points over the circadian cycle (ZT1, 3, 12, 16, 20, 24), and protein extracts from heads of 15N and 14N flies were separately subjected to FLAG affinity purification (AP) prior to sample preparation for LC-MS/MS analysis (S2A and S2B Fig). Relative PTM quantification was achieved by pooling purified PER from all 14N samples, and aliquoting equal amounts of pooled 14N fractions to each of the six 15N purified PER samples at a 1:1 ratio prior to MS analysis (S2C Fig).
Despite the large amount of fly head tissues we used for our protein extraction and FLAG-PER purification, our 15N/14N quantitative MS did not yield satisfactory temporal resolution of PER PTM cycling. We therefore consolidated our MS data from multiple time-points with the goal of identifying PER O-GlcNAcylation sites qualitatively (Fig 1 and Table 1). We observed that multiple residues located in the CBD and within the CLK:CYC inhibition domain (CCID) [31, 32] are O-GlcNAcylated. These include PER(S942) as well as a potential sites at S951, T952, or T954. Our MS analysis yielded O-GlcNAc-modified peptides in which only PER(S942) is modified, providing unambiguous identification of PER(S942) as an O-GlcNAcylated residue (Fig 1 and Table 1). In addition, O-GlcNAc-modified peptides spanning S942 to T954 were also identified, but we were not able to narrow down the single modified residue within these peptides. It is possible that only S942 is O-GlcNAcylated within this region. Alternatively, either S951, T952, or T954 may represent a second O-GlcNAcylated residue within this region. Taken together, we hypothesize that O-GlcNAcylation in the CBD may be important in modulating PER-CLK interaction and PER repressor activity. In addition to the O-GlcNAcylated residues in the CBD, a number of other potential sites were identified in other parts of the PER protein, but most of them are not located in characterized functional domains.
In addition to identifying PER O-GlcNAcylation sites, we took this opportunity to identify PER phosphorylation sites in fly tissues for the first time and confirm phosphorylation sites that have previously been identified in Drosophila S2 cells [17–19, 21]. Our analysis provides evidence that the majority of phosphorylation sites identified in Drosophila S2 cells are bona fide phosphorylation sites in fly tissues (Table 1). During the process of optimizing PER affinity purification and MS analysis using fly tissues as starting materials, we generated additional qualitative label-free MS datasets by analyzing PER phosphorylation sites at multiple time-points over the circadian cycle (ZT3, 16, 20, 24). S1 Table summarizes the label-free data and compares it to the phosphorylation sites identified in the 15N/14N-labeled MS analysis. PER O-GlcNAcylation site identification was not performed as part of the label-free MS analysis. We observed a high level of congruence between the PER phosphorylation sites identified in our two fly head tissue data sets, as well as those generated using Drosophila S2 cells as starting materials [17–19, 21]. In summary, our MS analysis sets the stage for future studies to understand the functional interplay between PER O-GlcNAcylation and phosphorylation.
Subsequent to the identification of PER O-GlcNAcylation sites, we proceeded to analyze the function of site-specific PER O-GlcNAcylation by mutating one or a cluster of S/T residues to non-O-GlcNAcylatable alanine on the per gene and evaluated PER repressor function on CLK activity using the per-luciferase (per-luc) reporter assay in Drosophila S2 cells [3, 37]. To prioritize, we analyzed PER O-GlcNAcylation residues with a Mascot score >50 (Table 1). We compared per-luc activity in S2 cells expressing per wild type or mutant variants (Fig 2A). In order to take into account the varying expression level of the PER variants (Fig 2B top panel, S3A and S3B Fig), we normalized CLK-activated per-luc activity observed for PER(WT) and PER mutant variants to their respective protein expression levels to more accurately assess the impact of blocking PER O-GlcNAcylation at specific residues on repressor activity (Fig 2C). We observed that all but one PER O-GlcNAcylation site mutant (PER(S174A)) exhibited significant increase in repressor activity. Interestingly, PER(S942A) and PER(T951A-S954A) are the two O-GlcNAc site mutants that exhibited the strongest repressor activity, further strengthening the rationale for testing the hypothesis that O-GlcNAcylation in the CBD may be important in modulating PER-CLK interaction and PER repressor activity. For this reason, this current study will focus on investigating the function of PER O-GlcNAcylation in the CBD, while analysis of other PER O-GlcNAcylation residues will be pursued in the future.
To confirm this finding in whole animals, we generated transgenic flies expressing p{per(WT)-HA10HIS} (herein referred to as per(WT) and mutant variants) in the per0 genetic background [38] so that only transgenic per (WT or mutant) was expressed. First, we performed quantitative RT-PCR to measure clock gene expression (per and tim) in heads of per(S942A) flies to examine the function of PER(S942) O-GlcNAcylation in central clock oscillators. All flies used for molecular analysis are homozygous for the per transgene. As predicted from the elevated PER repressor activity observed in S2 cells, we found that cycling of per and tim mRNAs was significantly dampened in per(S942A) flies as compared to per(WT) flies in both LD cycles and constant darkness (DD) (Fig 2D and 2E). Furthermore, per and tim mRNAs exhibited earlier initiation of repression phase in DD (Fig 2E). Since the per(S942A) mutant exhibited lowered levels of clock gene mRNAs, PER protein level is expected to decrease. As expected, peak PER abundance was significantly reduced in the heads of per(S942A) flies than in per(WT) flies in both LD and DD conditions (Fig 2F and 2G). To ensure that lower level of PER protein is a result of increased PER(S942A) repression on CLK-dependent transcription of per rather than reduced PER stability, we monitored the rate of PER degradation by cycloheximide (CHX) chase assay in S2 cells. The PER(S942A) mutant degraded at a similar rate as PER(WT) in the presence of OGT, demonstrating that the S942A mutation has no significant effect on PER stability (S3C and S3D Fig).
Since peripheral tissues are known to be more sensitive to metabolic fluxes [39, 40] and O-GlcNAcylation is a nutrient-sensitive PTM, we also analyzed the effect of blocking PER(S942) O-GlcNAcylation in oscillators of peripheral tissues, specifically the fat body. The fat body, which is analogous to mammalian liver and adipose tissue, plays an essential role in regulating energy metabolism in insects [39, 40]. We assayed per and tim mRNA levels in the fat body of per(WT) and per(S942A) flies on DD1 after LD entrainment to evaluate PER-dependent repression of CLK activity. Consistent with what we observed in fly heads, per(S942A) flies displayed significantly dampening of per and tim mRNA cycling as compared to per(WT) flies in the fat body (Fig 2H and 2I). This suggests that PER(S942) O-GlcNAcylation normally weakens the activity of PER to repress CLK-dependent transcription in both head and fat body oscillators.
As in PER(S942A) mutant, PER(S951A/T952A/T954A) mutant also exhibited a significant difference in repressor activity as compared to PER(WT) in per-luc reporter assay (Fig 2C). To evaluate the effects of these residues, which are also located in the CBD, we generated wper0; per(S951A/T942A/T954A) transgenic flies to confirm the effects of the PER(S951A/T952A/T954A) mutations in whole animals. We measured temporal cycling of per and tim mRNAs in heads of per(S951A/T952A/T954A) flies and observed dampened expression of per and tim mRNA as compared to per(WT) flies in both LD and DD conditions (S4A and S4B Fig), although not to the extent observed in per(S942A) flies. However, this did not translate into significant differences in PER protein abundance and cycling (S4C and S4D Fig). The differential effects observed when blocking O-GlcNAcylation at PER(S942) and PER(S951/T942/T954) suggest that PER(S942) O-GlcNAcylation is a key event in regulating PER repression within the CBD. It is important to stress that PER(S942) has never been identified as a phosphorylation site in all previous comprehensive mapping studies [17–19, 21], so it is highly unlikely that the effects for S942A mutation is due to disruption of PER(S942) phosphorylation.
Given that several of the non-O-GlcNAcylatable per mutants exhibited elevation in PER repressor activity in S2 cells and in flies, we proceeded to investigate if this molecular phenotype can translate to alterations in output behavioral rhythms. We first evaluated daily locomotor activity rhythms of per(WT) and mutant flies. Activity rhythm is a robust readout that reflects the function and speed of the central oscillators located in fly heads [30]. Flies were entrained for 3 days in LD cycles followed by 7 days in DD. As expected, heterozygous per(WT) flies manifested robust rhythms with ~24-hr periods, indicating full rescue of arrhythmic per0 mutation (Fig 3A). Homozygous per(WT) flies displayed somewhat shorter behavioral rhythms due to the extra copy of per. Interestingly, as compared to per(WT), per(S942A) and the triple per(S951A/T952A/T954A) mutants homozygous for their respective transgenes exhibited shorter periods by 1.4 and 0.7 hrs. The earlier initiation of repression previously observed in per(S942A) mutant likely contributes to its short-period phenotype. In comparison, per(S951A/T952A/T954A) mutants displayed a smaller though significant change in repression activity, which likely accounts for the smaller change in period length.
In addition to monitoring locomotor activity rhythms of per(S942A) flies, we also examined the effects of blocking PER(S942) O-GlcNAcylation in peripheral oscillators by measuring feeding rhythms. Feeding assays were only performed to compare per(WT) and per(S942A) flies as per(S951A/T952A/T954A) mutants displayed minor changes in activity rhythms and clock gene repression. Feeding activity rhythms are governed by oscillators of metabolic tissues (i.e. in the fat body) [40], which are more sensitive to nutrient flux than central oscillators in the brain. Since we demonstrated earlier that clock gene cycling was dampened in the fat body of per(S942A) flies as a result of PER being a stronger repressor, we expected that feeding rhythms will also be disrupted. We entrained per(WT) and per(S942A) flies in 12h:12h LD cycles and assayed their feeding activity rhythms for two consecutive days either in LD or DD conditions using the CAFE assay [41]. The amount of food consumed over a 2-hr period was determined at 4-hr intervals over the circadian cycle. As control, per0 flies were measured in parallel, and as expected they exhibited arrhythmic feeding activity in LD and DD (Fig 3B and 3C) as determined by JTK-cycle (P = 1) [42].
per(WT) flies displayed robust rhythms of feeding behavior (JTK-cycle (P < 0.01)); they feed during daytime and fast at night in both LD and DD conditions (Fig 3B and 3C). Robust feeding rhythms were also observed in per(S942A) flies in LD (JTK-cycle (P < 0.01)), but the peak was phase advanced. This is consistent with the shorter period length in locomotor activity observed in the per(S942A) mutant. The consequence of blocking O-GlcNAcylation at PER(S942) on feeding behavior is even more severe in DD in the absence of light cues as rhythmic feeding was abolished in DD (JTK-cycle (P = 1)). Together, our data demonstrates that O-GlcNAcylation at PER(S942) modulates the function of central and peripheral oscillators by regulating PER activity.
Premature PER nuclear entry during the night could explain why per(S942A) mutant exhibits stronger and/or premature initiation of repression on CLK activity. A previous study showed that PER O-GlcNAcylation is involved in regulating nuclear entry [23]. To examine this possibility, we monitored timing of PER nuclear entry in adult clock neurons in LD cycles. In the fly brain, rhythmic expression and nuclear localization of PER in lateral clock neurons (LNvs) are necessary for proper clock function and maintenance of rhythmic locomotor activity (reviewed in [2]). The lateral neurons (small and large; l-LNvs and s-LNvs) express pigment dispersing factor (PDF), which has been commonly used to label the cytoplasm of the LNvs [43]. Despite having shorter behavioral rhythms, we found that per(S942A) mutants did not exhibit a significant difference in accumulation of nuclear PER as compared to per(WT) flies at time-points when nuclear entry is most prominent, ZT18 to ZT22 [4] (S5A and S5B Fig). Although a previous report demonstrates that global knockdown or overexpression of OGT in clock neurons affects the timing of PER nuclear entry [23], it is not surprising that site-specific non-O-GlcNAcylatable per mutants, in this case per(S942A), may not show the same phenotype. As such, our results suggest that the stronger and premature initiation of CLK repression observed in heads and fat bodies of short-period per(S942A) flies cannot be explained by altered timing of PER nuclear localization in circadian oscillators.
Given that PER does not enter the nucleus prematurely in the LNvs of per(S942A) mutant flies, we speculate that the PER(S942) may have a higher affinity to CLK that results in stronger and/or premature initiation of repression. We therefore examined PER-CLK interactions by performing co-immunoprecipitations (coIP) using head extracts from per(WT) and per(S942A) flies. These experiments were performed in head tissues as IP reactions of clock proteins are more efficient in head tissues and produce more robust results. Indeed, the per(S942A) flies showed significantly higher PER-CLK interaction at ZT16 than per(WT) flies (Figs 4A, 4B and S6A). To validate our results, we performed coIPs using S2 cells coexpressing per(WT)-cmyc or per(S942A)-cmyc with clk-V5, in the presence or absence of pMT-ogt-FLAG. Consistent with the fly data, we observed that PER(S942A) proteins exhibited significantly stronger binding to CLK as compared to PER(WT) (Fig 4C and 4D). Interestingly, when PER(S942A) mutant was coexpressed with CLK in the presence of OGT, the level of PER-CLK binding was lower than the levels when PER(S942A) mutant was coexpressed with CLK in the absence of OGT. This suggests that O-GlcNAcylation of CLK or perhaps at other PER residues may negatively impact PER-CLK interaction. Nevertheless, our results suggest that stronger and/or premature initiation of repression phase in per(S942A) flies is due to higher affinity between CLK and PER(S942A).
In comparison to per(S942A) mutant flies, per(S951A/T952A/T954A) flies only displayed minor changes in activity rhythms and clock gene expression in flies. Based on these observations, the PER(S951A/T952A/T954A) mutant protein is predicted to exhibit minor changes on PER-CLK interactions, despite that these three residues are also localized within the CBD. As expected, no significant changes in PER-CLK interactions were observed between per(WT) and per(S951A/T952A/T954A) flies at the indicated time-points (S7A and S7B Fig). Thus, our results suggest that O-GlcNAcylation events at PER(S951/T952/T954) only have minor modulatory effects on PER-CLK interactions. Alternatively, it is possible that only PER(S942) is O-GlcNAcylated within the CBD (Fig 1).
Since part of PER repressor function is to remove CLK from clock gene promoters, we would expect that CLK may be removed prematurely when PER(S942) O-GlcNAcylation is blocked. We therefore measured CLK occupancy at the E-box elements of per and tim promoters by performing chromatin immunoprecipitation (CLK-ChIP) using extracts from adult fly heads. Consistent with our hypothesis, we found that flies expressing per(S942A) showed significantly reduced CLK occupancy on per and tim promoters at ZT18 and ZT20 as compared to per(WT) flies (Fig 4E). To rule out the possibility that decreased CLK occupancy observed at ZT18 and ZT20 in per(S942A) flies was due to decreased CLK levels, we examined CLK levels after the ChIP reactions and found that CLK was not limiting in flies expressing per(WT) or per(S942A) (S6B Fig).
To further support our hypothesis that dynamic O-GlcNAcylation at PER(S942) is critical for regulating the strength and/or timing of PER-dependent clock gene repression, we overexpressed 3XFLAG-ogt in tim-expressing clock neurons using a tim(UAS)-gal4 driver (referred to as TUG) [44] and examined PER-CLK interactions in head extracts of these flies. First, we verified that ogt overexpressors (TUG>FLAG-ogt) exhibited a significant ~2-hr period-lengthening of behavioral rhythms, as previously observed [23, 24] (Fig 5A). All 4 independent ogt overexpressor fly lines we generated showed similar period-lengthening phenotypes as compared to parental controls. To confirm that the overexpressed OGT enzyme is functional, we optimized O-GlcNAc chemoenzymatic labeling to measure levels of O-GlcNAcylated proteins in fly head extracts.
Currently, detecting protein O-GlcNAcylation remains to be a challenge when using traditional protein analytical techniques because the addition of a sugar group does not influence the migration of a polypeptide in gel electrophoresis (reviewed in [45]). Additionally, commercially available O-GlcNAc-specific antibodies or lectin yield non-specific signals in our hands. Thus, we chose to use a more sensitive and specific chemoenzymatic labeling method to examine in vivo protein O-GlcNAcylation status in this study [46–48].
We first tested the specificity of this approach by detecting O-GlcNAcylated PER and OGT in Drosophila S2 cell culture. S2 cells were transiently transfected with V5 tagged per with or without pMT-ogt-FLAG. Immunoprecipitated PER and OGT were labeled with resolvable mass tags prior to SDS-PAGE and western blotting. Results showed successful labeling of O-GlcNAcylated PER with 20 kD PEG mass tag in the presence of OGT, leading to clear mobility shift of O-GlcNAcylated PER isoforms (S8A Fig). As OGT is known to be auto-O-GlcNAcylated, addition of 10kD PEG mass tag in the reaction resulted in O-GlcNAcylated OGT isoforms that appeared as slower migrating bands as detected by Western blots (S8B Fig). The use of PEG mass tag labeling provides two advantages. First, both modified and non-modified isoforms can be detected, providing information on stoichiometry. Second, epitope-tagged antibodies or target-specific antibodies can be used for detection of all isoforms. Alternatively, O-GlcNAcylated proteins can also be labeled with biotin to facilitate detection with α-streptavidin antibodies [46–48]. Our success in detecting and visualizing in vivo O-GlcNAcylation status of proteins in S2 cells by chemoenzymatic labeling allowed us to proceed and confirmed that PER O-GlcNAcylation was elevated in head extracts of ogt overexpressor flies, an indication that overexpression of ogt led to increase in enzyme activity (Fig 5B and 5C).
Since blocking PER(S942) O-GlcNAcylation promotes PER-CLK interaction, we expect that PER-CLK interaction should be reduced in ogt overexpressors assuming PER(S942) is hyper-O-GlcNAcylated. As anticipated, we observed that control TUG flies exhibited significantly higher PER-CLK interaction at ZT18 as compared to ogt overexpressor flies in coIP assays using head extracts of flies (Fig 5D and 5E). For ogt overexpressor flies, the reduction in PER-CLK interaction observed in fly heads may partially account for the physiological effect of OGT overexpression, i.e. period-lengthening of clock-controlled locomotor activity rhythms (Fig 5A).
We speculate that PER(S942) O-GlcNAcylation occurs either during the daytime or early night to prevent de novo PER from prematurely binding to CLK or at late night to facilitate CLK dissociation from PER after PER-dependent repression. Both of these scenarios would result in stronger clock gene repression in per(S942A) flies. Since we observed that the differences in PER-CLK interactions between per(WT) and per(S942A) flies as well as between WT TUG control and ogt overexpressor flies occurred at around the start of the circadian repression phase (i.e. ZT16 and ZT18 respectively; Fig 4B and 5E), we postulated that the former scenario may be more likely. Temporal data on PER(S942) O-GlcNAcylation status would certainly help to rule out one of the two scenarios. It is therefore unfortunate that we are not able to determine the timing of PER(S942) O-GlcNAcylation by MS. Kim et al. [23] and Kaasik et al. [24] have previously examined global PER O-GlcNAcylation status over the circadian cycle using anti-O-GlcNAc antibodies in combination with Western blotting, but their results were incongruent. Whereas Kim et al. observed that PER O-GlcNAcylation peaks at ZT20 [23], Kaasik et al. observed a peak in PER O-GlcNAcylation at ZT10 [24]. We therefore opted to profile global PER O-GlcNAcylation status by chemoenzymatic labeling to gain insights into the timing and function of PER(S942) O-GlcNAcylation.
We examined the temporal profile of PER O-GlcNAcylation in head extracts of per(WT) flies in LD condition by labeling PER with either 20kD PEG or a biotin tag [46–48]. In both cases, maximal PER O-GlcNAcylation occurred at around ZT4 to ZT8 with subsequent decline from ZT12 to ZT24 (Fig 6A and 6B). Our results were more in line with the temporal PER O-GlcNAcylation profile observed in Kaasik et al. [24]. Daily PER O-GlcNAcylation cycle was found to be significantly rhythmic (JTK cycle; P < 0.0005) (Fig 6C). Our results reveal that PER is more highly O-GlcNAcylated during the day. This suggests that PER(S942) is likely O-GlcNAcylated during the day and may persist into early night to prevent de novo PER from prematurely interacting with CLK to initiate the repression phase.
O-GlcNAcylation of cellular protein is sensitive to nutrient input [25, 49]. Since we observed that PER O-GlcNAcylation is higher during the day and gradually decreases over the circadian cycle, we expect that this temporal pattern may correlate to daily feeding activity. We measured feeding rhythms in per(WT) flies fed ad libitum using the CAFE assay [41]. The flies used for these assays were entrained simultaneously with flies used for PER O-GlcNAc labeling to better assess correlation between feeding activity and PER O-GlcNAcylation. We found that mixed sexes of per(WT) flies displayed rhythmic feeding that peaked during early day in LD condition (S9A Fig). Similar results were obtained when male and female per(WT) flies were housed and tested separately (S9B Fig). Furthermore, a separate experiment comparing feeding activity rhythms of per(WT) and per(S942A) flies also showed higher daytime feeding activity in per(WT) flies (Fig 3B). Taken together, our data suggest that nutrient flux via feeding activity provides time-of-day metabolic signals to the circadian oscillator via temporal O-GlcNAcylation of PER. Specifically, PER(S942) O-GlcNAcylation, which is expected to occur during the day and persist into early night, may prevent interaction of PER and CLK prematurely to regulate timing of circadian repression.
Recent studies reveal that O-GlcNAcylation of circadian transcription factors, PER and CLK in Drosophila and BMAL1 and CLOCK in mice, plays an important role in modulating their function in the circadian oscillator [23–25, 50]. However, to more fully understand the mechanisms by which site-specific O-GlcNAcylation events regulate circadian physiology and to set the stage for investigating the interplay between phosphorylation and O-GlcNAcylation, it is necessary to identify O-GlcNAcylated residues in core clock proteins and other cellular proteins and characterize their site-specific functions. Furthermore, the relationship between feeding-induced nutrient influx and the temporal regulation of clock protein O-GlcNAcylation warrants investigation in whole animals. Since feeding activity is controlled by the circadian clock, we hypothesize that food intake will increase HBP influx, leading to increase in clock protein O-GlcNAcylation during the feeding period or soon after. This could serve as a mechanism by which metabolic input reinforces circadian entrainment by other zeitgebers and regulates oscillator function in a time-of-day specific manner. We therefore set out to (i) identify PER O-GlcNAcylation sites and characterize their site-specific functions in regulating the circadian oscillator; and (ii) determine if there is a correlation between time of feeding and O-GlcNAcylation levels of PER. Although overexpression or knockdown of OGT and OGA has provided insights into the global effects of protein O-GlcNAcylation on circadian clock regulation, we expect that site-specific characterization of O-GlcNAcylation events will alleviate confounding effects resulting from having multiple O-GlcNAcylated residues within a single protein or the involvement of multiple O-GlcNAcylated proteins in clock regulation, leading to new mechanistic insights.
By utilizing MS-based proteomics, we observed that in addition to being heavily phosphorylated, PER is O-GlcNAcylated at at least 6 residues, some of them in the CBD. To understand the role of these O-GlcNAcylation events in regulating PER function and circadian physiology, we analyzed these O-GlcNAcylation sites by replacing serine or threonine to non-GlcNAcylatable alanine either singly or in clusters. Several of these non-O-GlcNAcylatable per mutants exhibit changes in PER repressor function, which consequently result in period-changing phenotypes in their corresponding transgenic fly lines. In particular, we observed that O-GlcNAcylation at PER(S942), which is located in the CBD, reduces PER-CLK interaction (Fig 7A and 7B). Using O-GlcNAc chemoenzymatic labeling, we show that PER O-GlcNAcylation primarily occurs during daytime and correlates with the time period when animals are feeding. We therefore postulate that PER(S942), as in the case for most PER O-GlcNAcylation sites, is O-GlcNAcylated during the day. This ensures that PER does not interfere with CLK activity in the circadian activation phase and its repression of CLK activity does not initiate prematurely when de novo PER starts to translocate into the nucleus. This suggests that OGA may need to actively remove O-GlcNAc from PER residues prior to circadian repression phase. Indeed, OGA level has been shown to oscillate over the circadian day, peaking prior to initiation of circadian repression phase [24]. Moreover, O-GlcNAcylation has been shown to regulate PER nuclear entry [23], suggesting that OGA-dependent removal of O-GlcNAc at unknown PER residue(s) is likely required to facilitate PER nuclear translocation independent of OGA activity on PER(S942). Finally in the evening, since flies are fasting, the level of O-GlcNAcylation at PER(S942) will remain low allowing PER to bind strongly to CLK to repress its activity.
Our findings that PER O-GlcNAcylation at S942 reduces PER repression of CLK activity is not congruent to the observation in [24], where they observed that the repressor activity of PER is enhanced when coexpressed with OGT in S2 cell per-luc reporter assay. However, this apparent incongruence could be explained by the combined activities of other O-GlcNAcylation events on PER, CLK, or other cellular proteins that impact per-luc reporter gene expression. Nevertheless, it is important to point out that none of our non-O-GlcNAcylatable per mutants showed a decrease in repressor activity (Fig 2C).
It is interesting to note that unlike some phosphorylation sites previously identified on PER proteins, residues we identified to be O-GlcNAcylated in Drosophila PER are not conserved in mouse PER proteins. This is perhaps not surprising since nutrient-dependent O-GlcNAcylation on cellular proteins likely depends on when organisms are actively feeding, i.e. whether they are diurnal or nocturnal. For instance, mice are nocturnal, which may restrict protein O-GlcNAcylation of cellular proteins to nighttime. This nighttime peak in O-GlcNAcylation levels corresponds to when mouse PER2 proteins are abundant and active as repressors, suggesting that O-GlcNAcylation may act to promote PER2-dependent repression on circadian transcription. On the other hand, given that flies are diurnal and feed during the day (Fig 3B), nutrient flux promoting O-GlcNAcylation of Drosophila PER inhibits PER function as a transcriptional repressor (Fig 7). Overall, the differences in timing of feeding activity and nutrient flux between diurnal and nocturnal animals will present interesting opportunities for comparative analysis with regard to site-specific and global effects of O-GlcNAcylation on cellular protein function.
In addition to regulating PER-CLK interaction in the Drosophila circadian clock, O-GlcNAcylation is expected to affect oscillator function via other mechanisms [23–25]. Besides regulating clock proteins directly, it is likely that O-GlcNAcylation can modify the activity of clock kinases, just as GSK3β can regulate the activity of OGT [24]. CK2 and GSK3β are two clock kinases that have been shown to be substrates of OGT [51–52], and currently it is unclear how OGT might modulate their activities in clock regulation. Furthermore, O-GlcNAcylation can also impact the activities of chromatin modifiers and transcription machineries, including RNAPII [53]. The interplay between O-GlcNAcylation and other PTMs to regulate circadian physiology is expected to be complex and extensive, and reach far beyond the molecular oscillator and circadian transcription.
In summary, our results support that clock-controlled feeding activity provides metabolic input to reinforce entrainment signals by light-dark cycles to regulate circadian physiology via clock protein O-GlcNAcylation. We expect that circadian rhythms in peripheral systems, where oscillators are more sensitive to metabolic input, to be particularly sensitive to modulation via O-GlcNAcylation of clock proteins. Finally, our results imply that disruptions in daily feeding rhythms, e.g. irregular meal times and late night eating common in modern societies, will likely affect rhythms in protein O-GlcNAcylation and interplay with other PTMs, thereby disrupting circadian rhythms in physiology. Future experiments to manipulate feeding schedules by time-restricted feeding (TRF) [54] can further solidify the causal relationship between feeding-induced nutrient influx and O-GlcNAcylation of cellular proteins. They can also provide mechanistic insights into the benefits of TRF.
To profile PER PTMs, we generated transgenic flies that expressed a 13.2 kb genomic clone of per in w1118 per0 (wper0) background. A previously characterized vector that contains a 13.2kb per genomic fragment tagged with HA and 10X histidine at the carboxyl terminal (pCaSpeR-per(13.2WT)-HA-10HIS) [55] was used as the template for inserting 3XFLAG at the amino terminal before the starting Methionine to facilitate FLAG Affinity Purification. Transformants were generated by P-element transformation (BestGene Inc., Chino Hills, CA), and the 3XFLAG-per(13.2WT)-HA-10HIS transgene was tested for functionality by determining whether it can rescue wper0 flies in behavioral and molecular assays.
To generate transgenic flies carrying wild-type (WT) or O-GlcNAcylation site mutants of per, we opted to use PhiC31 site-directed recombination [56]. The genomic per(13.2WT)-HA-10HIS was excised from pCaSpeR-per(13.2WT)-HA-10HIS using the restriction sites XhoI and BamHI and subcloned into pattB vector (kind gift from Amita Seghal) to yield pattB-per(13.2WT)-HA-10HIS. The pattB vector was modified so that the sites KpnI and XbaI were removed, the BgIII site was replaced by a BamHI site, and the BamHI site was replaced by a BgIII site. For generating flies expressing non-O-GlcNAcylatable mutants on per, a 3.4 kb XbaI-BamHI or 1kb BamHI-KpnI genomic fragment was excised from the pattB-per(13.2WT)-HA-10HIS plasmid and subcloned into a pGem7 vector [20]. The resulting pGem7-per(XbaI-BamHI) or pGem7-per(BamHI-KpnI) plasmid served as a parent template for site-directed PCR mutagenesis (Agilent Technologies, Santa Clara, CA) depending on the location of the O-GlcNAc site (S2 Table for mutagenic primer sequences). After mutagenesis and confirmation by Sanger sequencing (UC Davis Sequencing), the mutant variants of 3.4 kb or 1 kb per subfragments were used to replace the WT fragments in pattB-per(13.2WT)-HA-10HIS. Plasmids were injected into w1118 fly embryos carrying attP sites on chromosome 3 (attP2) (BestGene, Inc.). Transformants were crossed with wper0 flies to remove endogenous copies of per prior to behavioral and molecular analyses.
To generate flies overexpressing ogt, ogt was amplified and subcloned into pUAST-attB vector [57]. 3XFLAG was added to the N-terminus of ogt during the cloning process. Plasmids were injected into w1118 fly embryos carrying attP sites on chromosome 3 (attP2) (BestGene, Inc.). To express 3XFLAG-ogt in clock neurons, transgenic flies carrying the UAS-FLAG-ogt transgene were crossed to w; tim-(UAS)-GAL4 (referred to as TUG) driver line [44].
CAFE assay was performed as described [41] with modifications. Mixed-sex population of five male and five female wper0, wper0; per(13.2WT), and wper0; per(S942A) flies or separately housed male or female (10 per group) flies were fed Bloomington Drosophila Stock Center standard fly food during entrainment in 12 h light/12 h dark day-night cycles and food consumption were measured starting on the third day of LD or the first day of DD. Prior to the day of measurement, grouped flies were transferred to a vial containing 2% agar as the medium with 5% sucrose solution maintained in calibrated glass micropipettes (VWR). After 24 hours of training, old micropipettes were replaced by fresh experimental micropipettes filled with 5% sucrose solution approximately 2 hours before each of the indicated time-points. After 2 hours, the amount of liquid consumed from the experimental micropipette was recorded, and the evaporation effect was evaluated by measuring the change in liquid volume in a micropipette placed in a vial without flies. Food consumption for each group/vial was determined by subtracting the amount of liquid consumed from the experimental micropipette with the amount of evaporated liquid. These values were normalized to the amounts of flies in the vial that survived until the end of the experiment. These experiments were performed in biological triplicates (one group/vial represents one independent experiment). Error bars = SEM. Rhythmicity of feeding was determined by JTK Cycle [42].
Proteins from S2 cells and fly heads were extracted using modified RIPA buffer as previously described [58]. Extracts were quantified and either directly analyzed by immunoblotting or incubated with 15μ α-V5 resin (Sigma) or 20μl of α-HA resin (Sigma) at 4°C for 4 hours. Beads were washed once in M-RIPA and twice with reaction buffer (20mM HEPES pH 7.9, 50mM NaCl, 1μM PUGNAc, 25mM NaF, 0.5mM PMSF, and 5mM MnCl2) supplemented with 1x protease inhibitor (Sigma) [59]. Procedures for chemoenzymatic labeling with biotin or PEG (Polyethylene Glycol) 10kD or 20kD mass tag were performed as described [48] with modifications. Attachment of biotin or PEG mass tag to O-GlcNAc group requires a two-step derivatization process [59]: (1) a mutant galactosyltransferase, GalT1 (Y289L), utilizes UDP-azidogalactose (UDP-GalNAz) as substrate to add an azide onto the O-GlcNAc group; (2) Biotin alkyne or an alkyne-functionalized PEG mass tag indirectly attaches to the O-GlcNAc group via azide-alkyne cycloaddition chemistry. Briefly, after immunoprecipitation followed by washes, immune complexes were resuspended in 20 μl of reaction buffer containing 2 μl of Gal-T1 Y289L (Invitrogen) and 2μl of 0.5mM UDP-GalNAz (Invitrogen) [59] following overnight incubation at 4°C with gentle rotation. Azide-labeled beads were washed twice with reaction buffer and subsequently resuspended in 50μl of labeling buffer (1% SDS and 50mM Tris-HCl pH 8.0). The samples were reacted with biotin alkyne (Invitrogen) or an alkynyl-functionalized poly(ethylene glycol) (10-kDa or 20-kD) (Creative PEGWorks, Chapel Hill, NC) according to manufacturer’s protocol or a previously described protocol by [59] respectively. Samples were eluted in 50μl of 1XSDS sample buffer. For proteins labeled with a mass tag (PEG), PER or OGT was resolved using SDS-PAGE (5% or 8% minigel with a 40:1 acrylamide/Bis-acrylamide solution, Bio-rad, Hercules, CA). For proteins labeled with biotin, PER was resolved using SDS-PAGE (5% Criterion gels, Bio-rad). Antibody dilutions to detect O-GlcNAcylated PER or OGT proteins are as follows: α-V5 (1:5000), α-HA (1:1000), α-FLAG (1:7000), α-streptavidin (Cell Signaling Technologies, Danvers, MA) (1:5000), and α-PER (GP5620; RRID:AB_2747405) (1:2000).
wper0; p{3XFLAG-per(13.2WT)-HA10HIS} flies were fed with an 15N diet (0.2g 15N yeast, 1% Bacto agar, 15% unsulfured molasses, phosphoric and propionic acid mix, and tegosept). Saccharomyces cerevisiae were metabolically 15N-labelled as described [29, 60]. As control, flies were also fed with 14N diet. The adult progenies of 15N- or 14N-fed parental flies were reared in 15N- or 14N diet and entrained for 3 days in 12hr light:12hr dark at 25°C and collected every four hours over a period of 24 hours on the fourth day. Upon collection, flies were immediately frozen on dry ice until protein extraction. For each time-point, 4ml of fly heads were homogenized into fine powder by a chilled mortar and pestle and resuspended in Lysis Buffer (20mM HEPES pH 7.9, 5% glycerol, 350mM NaCl, 0.1% Triton X-100, 1mM DTT, 1mM MgCl2, 0.5mM EDTA, 25mM sodium fluoride, 1x protease inhibitor (Sigma, St. Louis, MO), 1x PhosSTOP (Roche, South San Francisco, CA). Homogenate was dounced for 15 strokes using a 50ml tissue grind tight pestle (Kimble-Chase, Vineland, NJ) and were filtered using a 70μm cell strainer. Samples were spun at 300 rcf for 1 minute and then incubated at 4°C on a nutator for 30 minutes. Additional Lysis Buffer was added to dilute the sample from 350mM to 150mM NaCl before centrifuging at 15,000 rpm for 15 minutes at 4°C. Supernatant was collected and incubated with 200μl α-FLAG resin (Sigma) overnight over a nutator at 4°C. Next day, beads were washed twice for 15 mins with Lysis Buffer without EDTA, DTT, or PhosSTOP. Samples were eluted in 300ul R+A buffer (30% glycerol, 3% SDS, 6mM EDTA, 0.15M Tris-HCl pH 6.8) at 95°C. Eluate was reduced with 15μl 1M DTT for 10 minutes at 65°C and then alkylated with 35μl 1M IAA in room temperature for 20 minutes in the dark. Eluates were then flash frozen using liquid nitrogen immediately. We used the pooled standard approach to enable more accurate comparisons between different time points. 14N eluate from six time-points were pooled together and split evenly to mix with each 15N eluate at a 1:1 ratio on ice. For each time-point, 600μl of cold acetone was added to the 14N/15N eluate mixture and placed in -20°C overnight. Precipitate were spun at 14,000 rpm for 10 minutes at 4°C and the resulting pellet were washed briefly with 1ml cold acetone. Precipitated eluate was resuspended in 80μl R+A sample buffer containing 3μl of 4X SDS sample buffer. The 14N/15N eluate was resolved in 12% SDS-PAGE and the excised PER band was used for protease digestion and analysis by mass spectrometry.
For analysis of 14N/15N-labeled samples, proteins were digested in-gel with trypsin and elastase in separate reactions to result in overlapping peptides, such that individual modified sites can be determined. We have previously used this multi-protease approach [61] to maximize high sequence coverage when mapping PER phosphorylation sites [17]. For in-gel digestion the excised gel bands were destained with 30% ACN, shrunk with 100% ACN, and dried in a Vacuum Concentrator (Concentrator 5301, Eppendorf, Hamburg, Germany). Digests with trypsin and elastase were performed overnight at 37°C in 0.05 M NH4HCO3 (pH 8). About 0.1 μg of protease was used for one gel band. Peptides were extracted from the gel slices with 5% formic acid.
NanoLC-MS/MS analyses were performed on an LTQ-Orbitrap Velos Pro or an Orbitrap Fusion (Thermo Fisher Scientific, Waltham, MA) equipped with an EASY-Spray Ion Source and coupled to an EASY-nLC 1000 (Thermo Fisher Scientific). Peptides were loaded on a trapping column (2 cm x 75 μm ID. PepMap C18, 3 μm particles, 100 Å pore size) and separated on an EASY-Spray column (25 cm x 75 μm ID, PepMap C18, 2 μm particles, 100 Å pore size) with a 90-minute linear gradient from 3% to 30% acetonitrile and 0.1% formic acid. For the Oribtrap Velos MS scans were acquired in the Orbitrap analyzer with a resolution of 30,000 at m/z 400, MS/MS scans were acquired in the Orbitrap analyzer with a resolution of 7,500 at m/z 400 using HCD fragmentation with 30% normalized collision energy. A TOP5 data-dependent MS/MS method was used; dynamic exclusion was applied with a repeat count of 1 and an exclusion duration of 30 seconds; singly charged precursors were excluded from selection. Minimum signal threshold for precursor selection was set to 50,000. Predictive AGC was used with AGC target a value of 1e6 for MS scans and 5e4 for MS/MS scans. Lock mass option was applied for internal calibration in all runs using background ions from protonated decamethylcyclopentasiloxane (m/z 371.10124). For the Orbitrap Fusion, both MS and MS/MS scans were acquired in the Orbitrap analyzer with a resolution of 60,000 for MS scans and 15,000 for MS/MS scans. HCD fragmentation with 35% normalized collision energy was applied. A Top Speed data-dependent MS/MS method applying HCD and ETD fragmentation from the same precursor with a fixed cycle time of 3 seconds was used. Dynamic exclusion was applied with a repeat count of 1 and an exclusion duration of 120 seconds; singly charged precursors were excluded from selection. Minimum signal threshold for precursor selection was set to 50,000. Predictive AGC was used with AGC a target value of 5e5 for MS scans and 5e4 for MS/MS scans. EASY-IC was used for internal calibration.
Mascot Distiller 2.5 was used for raw data processing and for generating peak lists, essentially with standard settings for the Orbitrap (high/high settings). Mascot Server 2.5 was used for database searching with the following parameters: peptide mass tolerance: 7 ppm, MS/MS mass tolerance: 0.02 Da, enzyme: “semi-trypsin” for tryptic digests and “none” for elastase digests; fixed modifications: carbamidomethyl (C); variable modifications: Gln->pyroGlu (N-term Q), oxidation (M), acetyl (protein N-term), phosphorylation (STY), HexNAc (ST). Searches containing both HCD and ETD spectra (Fusion) were searched separately for either b and y ions (HCD) or c and z ions (ETD). For ETD-searches different modifications definitions (without neutral losses) for phosphorylation and HexNAc were applied. Separate Mascot searches were performed for light peptides (quantitation: “none”) and heavy peptides (quantitation: 15N-labeling). Database searching was performed against a small custom database containing 187 of the most abundant proteins identified in these samples before in a first round search (without PTMs) against UniProt Drosophila database. This was necessary to limit search space and processing times.
The results from the different Mascot searches (different time-points, proteases and fragmentation techniques) were merged (separately for light and heavy peptides) and filtered for phosphorylated and HexNAc-modified peptides using a custom software tool (A. Schlosser). A Mascot score cut-off of 15 and a delta score cut-off of 10 [62] were applied, and only “rank 1” peptides were accepted. For one modification site and one type of modification, only the peptide spectrum match (PSM) with the highest score was kept, all other PSMs were filtered out. All remaining spectra were verified manually, e.g. by checking the presence of modification specific marker ions. HexNAc-modified peptides were only accepted when at least one of the HexNAc-specific fragment ions (204, 186 and 168) [63] was present in the corresponding HCD spectra. After manually filtering, all remaining peptides were exported to generate a summary of the results (Table 1). The N14/N15 MS data have been submitted to the Chorus repository (project ID 1424): (https://chorusproject.org/anonymous/download/experiment/e47a30f7f2c749aba438652d7d88ef04) and (https://chorusproject.org/anonymous/download/experiment/e6d6163b31bf40288606f827c6f18371).
All flies were reared on standard Drosophila medium (Bloomington Drosophila Stock Center standard recipe). Entrainment and collection of flies at the appropriate time-points were described as above. Roughly 4ml of fly heads were grinded into fine powder using chilled ceramic mortar and pestle and mixed in 30ml of lysis buffer (20mM HEPES pH 7.5, 1mM DTT, 1x protease inhibitor). Homogenate was dounced and poured over a cell strainer as described above prior to centrifugation at 7000xg for 45 minutes at 4°C to separate nuclear and cytoplasmic lysates, repeated once. Lower layer (pellet) as the nuclear fraction from both spins was resuspended in 10ml Nuclear Extraction buffer (20mM HEPES pH 7.5, 10% Glycerol, 350mM NaCl, 0.1% Triton X-100, 1mM DTT, 1mM MgCl2, 0.5mM EDTA, 1x protease inhibitor, 10mM NaF) with the addition of MG132 (Sigma) and DNAse (Promega). Upper layer (supernatant) as the cytoplasmic fraction was supplemented with Lysis buffer with the addition of MG132 (Sigma) and DNAse (Promega). Nuclear and cytoplasmic fractions were incubated at 4°C for 30 minutes over a nutator. After incubation, nuclear fraction was diluted to 150mM NaCl with Lysis buffer. Nuclear and cytoplasmic fractions were centrifuged at 27,000rpm for 15 minutes at 4°C. Supernatant of nuclear and cytoplasmic samples was recovered before incubation with 200μl α-FLAG M2 resin (Sigma) at 4°C overnight. Beads were washed three times with Lysis buffer for 15 minutes each and subsequently eluted in 200μl of 3XFLAG peptide (Sigma) at a dilution of 250μg/ml at room temperature for 15 minutes. Eluates were resolved on a Tris-Tricine gel and PER bands were excised for protease digestion and mass spectrometry as described in [64]. The label-free MS proteomics data for PER phosphorylation site mapping have been deposited into ProteomeXchange (PXD008281) (ProteomeXchange: http://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD008281), MassIVE repository (MSV000081736) (MassIVE: https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=384c7750b3004b7eac91054935a4e038), and Chorus repository (Project ID 1424) (https://chorusproject.org/anonymous/download/experiment/1e0023a15da84e51bb18c55146104b32).
Locomotor activity rhythms were measured as previously described [30]. 3-4-day old male flies were collected and subjected to 12hr light:12hr dark (LD) cycles at 25°C for four days followed by seven days of constant darkness (DD) to measure free-running period using the Drosophila Activity Monitor System (DAMS) (TriKinetics, Waltham, MA). Data analysis was performed using FaasX as described in [30].
pAc-per-V5, pAc-3XFLAG-per-6Xcmyc, and pAc-clk-V5 were previously described [20, 58]. For generating Serine/Threonine (S/T) to Alanine (A) O-GlcNAc site mutants (S2 Table for primer sequences), pAc-per-V5 or pAc-3XFLAG-per-6Xcmyc served as the template for site-directed PCR mutagenesis using QuikChange site-directed PCR mutagenesis (Agilent Technologies). All O-GlcNAc mutations were verified by Sanger sequencing. ogt cDNA (described above) was subcloned into a pMT-3XFLAG-6XHIS vector described in [20], with the epitope at the C-terminus of the ORF. pCopia-renilla luciferase and per-E-luc constructs were described previously [37]. S2 cells and DES expression medium were obtained from Life technologies (Carlsbad, CA), and transient transfections were performed using Effectene (Qiagen, Valencia, CA) according to manufacturer’s instructions and as previously described [17,20].
Luciferase reporter assays were performed as described [3, 37]. Measurements of luciferase activity were performed using the Dual-Glo luciferase assay system (Promega, Madison, WI) according to the manufacturer’s recommendation on a TriStar LD 941 microplate reader (Berthold Technologies, Oak Ridge, TN).
pAc-per-V5(WT) or mutant variants were transfected into S2 cells with pMT-FLAG-ogt or an empty plasmid. 20 hours after transfection, ogt expression was induced for 16 hours. Cycloheximide was then added to a final concentration of 10 μg/ml. Cells were harvested and lysed with EB2 [20] at the times indicated. Proteins were analyzed by western blotting as detailed below and in [65].
S2 cell and adult fly head protein extractions, western blotting, and image analysis, were performed as previously described [58, 65] with modifications. Primary antibodies α-V5 (Invitrogen, Carlsbad, CA) (1:5000) was used to detect CLK-V5 and PER-V5, α-cmyc (9E10, Sigma, St. Louis, MA) (1:5000) to detect PER-CMYC, α-FLAG (Sigma) (1:7000) to detect FLAG-OGT, α-HA 3F10 (Roche, Indianapolis, IN) (1:1000) to detect PER-HA, α-PER (GP5620; RRID:AB_2747405) [57] (1:3000) to detect PER, and α-HSP70 (Sigma) (1:10,000) was used for normalization. Secondary antibodies conjugated with HRP were added at final dilution as follows: α-mouse IgG at 1:5000 for α-V5 detection, 1:7000 for α-FLAG detection, or 1:10,000 for α-HSP70 detection, α-guinea pig IgG at 1:2000 for α-PER detection, and α-rat IgG (1:1000) for α-HA detection of PER-HA. Membranes were imaged and protein levels were quantified using the ChemiDoc MP system with Image Lab software (Bio-Rad). To calculate PER degradation rate, PER intensity was normalized to HSP70 intensity at each time-point, and was then converted to a fraction of the peak value (peak = 1). For quantifying PER levels from fly heads, PER values were normalized against HSP70 intensity at each time-point, and subsequently expressed as a fraction of the peak PER levels.
Co-IP assays using protein extracts from S2 cells and fly heads were performed as described [58, 65] with modifications. Proteins were extracted using modified RIPA buffer with the addition of 100μM PUGNAc to preserve O-GlcNAcylation of proteins prior to input analysis by western blotting or Co-IP with appropriate antibodies. Samples were pre-cleared using sepharose beads (Sigma) to reduce nonspecific binding. For co-IP in S2 cells, CLK IP samples were incubated with 15μl α-V5 resin (Sigma) and negative non-specific control IP samples were incubated with 15μl α-HA resin (Sigma). For co-IP in fly heads, CLK IP samples were incubated with 4μl α-CLK antibody (Santa Cruz Biotechnology, Dallas, TX) for 3 hours prior to incubation with 20μl gamma sepharose beads (GE, Pittsburgh, PA) for 1 hour; PER IP samples were incubated with 20μl α-HA resin (Sigma). Immune complexes were resolved by SDS-PAGE as described [58, 65]. IP signal intensity was normalized to intensity of the bait protein. These values were then converted as relative to the peak value of the dataset (peak = 1). Representative data shown are averages of normalized PER or CLK interactions from at least three independent experiments.
Total RNA was extracted from fly heads and abdominal fat bodies using TRI-Reagent (Sigma). cDNA synthesis from total RNA and real-time PCR analysis was performed as previously described [58, 65]. For isolating abdominal fat bodies, flies were collected at the indicated time-points on the first day of DD and immediately transferred in TRI-Reagent (Sigma) for 40 minutes at room temperature with agitation following dissection in RNAlater buffer (Thermo Fischer Scientific). At least 16 flies were dissected for each genotype and time-point. After dissection, fat bodies were rinsed with nuclease-free water twice and resuspended in TRI-Reagent prior to RNA isolation.
ChIP assays were essentially performed as described [58]. qPCR of an intergenic region (FBgn0003638) of the Drosophila genome representing background CLK binding was subtracted from input samples. Technical triplicates for the qPCR step were performed for each of the three biological replicates. Two-tailed t-tests were used to determine statistical differences between genotypes at each time-point.
Adult fly brain immunohistochemistry was performed as described previously [65] with modifications. Briefly, adult flies were entrained in LD cycles for 3 days and collected at the appropriate time-point following incubation with fixative solution (4% paraformaldehyde, 0.2% Triton X-100 in PBS) for at least 40 minutes in the dark at room temperature with gentle rotation. Fixative solution was removed, and then wash solution (0.2% Triton X-100 in PBS) was added to transfer flies into an embryo dish. Brains were dissected using #5 Rubis nano tweezers (Electron Microscopy Sciences, Hatfield, PA). Approximately 10 brains were dissected for each time-point. After dissection, brains were incubated in fixative solution at room temperature for 40 minutes with gentle rotation. Brains were rinsed quickly with wash solution three times prior to three slow washes in wash solution for 10 minutes each with gentle rotation. Wash solution was removed, and blocking solution (0.2% Triton X-100, 5% horse serum in PBS) was added to the brains for 40 minutes at 4°C with gentle rotation. Brains were then incubated with new blocking solution at 4°C overnight with primary antibodies at the following dilutions: α-HA 3F10 antibody (Roche) (1:100), and α-PDF antibody (Developmental Studies Hybridoma Bank, Iowa City, IA) (1:100). After ~18 hr, brains were rinsed quickly with wash solution three times prior to four slow washes with wash solution for 10 minutes each with gentle rotation. Brains were then incubated in secondary antibodies in blocking solution overnight. Secondary antibodies used at the following dilutions were Dylight88-conjugated α-rat (1:100) and Alexa647-conjugated α-mouse (1:100) (Thermo Fischer Scientific). After ~18 hr, brains were rinsed quickly with wash solution three times prior to four slow washes with wash solution for 10 minutes each with gentle rotation. Brains were rinsed quickly with PBS following incubation with 85% glycerol for 15 minutes. Brains were mounted on microscope slides in VectaShield mounting medium (Vector Laboratories, Burlingame, CA) under a #1.5 (17-mm) coverslip. Confocal images were obtained with an Olympus FV1000 Confocal Microscope (Olympus Life Science, Center Valley, PA) and processed with the FV1000 software (Olympus Life Science). Scoring of PER subcellular localization was performed as previously described [66, 67]. At least five brains were used for each genotype and time-point. For statistical analysis, scored LNvs from each brain served as one biological replicate. Two-tailed t-tests were used to determine statistical differences between genotypes at each time-point.
|
10.1371/journal.ppat.1006978 | The Cryptococcus neoformans Titan cell is an inducible and regulated morphotype underlying pathogenesis | Fungal cells change shape in response to environmental stimuli, and these morphogenic transitions drive pathogenesis and niche adaptation. For example, dimorphic fungi switch between yeast and hyphae in response to changing temperature. The basidiomycete Cryptococcus neoformans undergoes an unusual morphogenetic transition in the host lung from haploid yeast to large, highly polyploid cells termed Titan cells. Titan cells influence fungal interaction with host cells, including through increased drug resistance, altered cell size, and altered Pathogen Associated Molecular Pattern exposure. Despite the important role these cells play in pathogenesis, understanding the environmental stimuli that drive the morphological transition, and the molecular mechanisms underlying their unique biology, has been hampered by the lack of a reproducible in vitro induction system. Here we demonstrate reproducible in vitro Titan cell induction in response to environmental stimuli consistent with the host lung. In vitro Titan cells exhibit all the properties of in vivo generated Titan cells, the current gold standard, including altered capsule, cell wall, size, high mother cell ploidy, and aneuploid progeny. We identify the bacterial peptidoglycan subunit Muramyl Dipeptide as a serum compound associated with shift in cell size and ploidy, and demonstrate the capacity of bronchial lavage fluid and bacterial co-culture to induce Titanisation. Additionally, we demonstrate the capacity of our assay to identify established (cAMP/PKA) and previously undescribed (USV101) regulators of Titanisation in vitro. Finally, we investigate the Titanisation capacity of clinical isolates and their impact on disease outcome. Together, these findings provide new insight into the environmental stimuli and molecular mechanisms underlying the yeast-to-Titan transition and establish an essential in vitro model for the future characterization of this important morphotype.
| Changes in cell shape underlie fungal pathogenesis by allowing immune evasion and dissemination. Aspergillus and Candida albicans hyphae drive tissue penetration. Histoplasma capsulatum and C. albicans yeast growth allows evasion and dissemination. As major virulence determinants, morphogenic transitions are extensively studied in animal models and in vitro. The pathogenic fungus Cryptococcus neoformans is a budding yeast that, in the host lung, switches to an unusual morphotype termed the Titan cell. Titans are large, highly polyploid, have altered cell wall and capsule, and produce haploid daughters. Their size prevents engulfment by phagocytes, yet they are linked to dissemination and altered immune response. Despite their important influence on disease, replicating the yeast-to-Titan switch in vitro has proved challenging. Here we show that Titans are induced by host-relevant stimuli, including serum and bronchio-alveolar lavage fluid. We identify a bacterial cell wall component as a relevant inducing compound and predict an in vivo Titan defect for a clinical isolate. Genes regulating in vivo Titanisation also influence in vitro formation. Titanisation is a conserved morphogenic switch across the C. neoformans species complex. Together, we show that Titan cells are a regulated morphotype analogous to the yeast-to-hyphal transition and establish new ways to study Titans outside the host lung.
| Fungi change shape in response to environmental stimuli. These morphogenic transitions drive pathogenesis and allow fungi to occupy different environmental niches. Dimorphic fungi undergo a yeast-to-hyphal transition in response to changing temperature, while the pleomorphic gut resident fungus Candida albicans integrates diverse signals depending on its local environment [1, 2]. The basidiomycete Cryptococcus neoformans undergoes an unusual transition in the host lung from haploid yeast-phase growth to apolar expansion and endo-reduplication, producing large, highly polyploid cells termed Titan cells[3, 4]. While there is growing evidence of the important role Titan cells play in disease [5–8], understanding the mechanisms underlying the yeast-to-Titan transition remains challenging due to the lack of an in vitro model.
C. neoformans is an environmental human pathogen that causes cryptococcal meningitis when inhaled yeast and spores disseminate to the central nervous system and brain. The fungus infects an estimated 1 million people worldwide each year and is responsible for between 140,000 and 600,000 deaths, primarily in sub-Saharan Africa [9–11]. Although the majority of patients are immunocompromised, a growing number of infections are seen in immuno-competent individuals [12–14]. Long term azole therapy is associated with relapse due to drug resistance and the emergence of hetero-resistance [14–16]. C. neoformans grows preferentially as an encapsulated budding yeast under physiologically relevant conditions and during culture in standard microbial media, and the vast majority of research has focused on the yeast form. However, there are early clinical reports of Titan cells, in which large encapsulated yeast were isolated from the lung and brain of infected patients [17, 18]. In both cases, cell size was dependent on growth condition, shifting from >40 μm in patient samples to <20 μm during in vitro culture and back to >40 μm in murine infection. Cruickshank et al. also report distinct capsule and cell wall structure of enlarged cells[18]. Despite this clear morphological transition, both early reports concluded that the patient samples represented atypical isolates.
However, far from being unusual outliers, it is now clear that Titan cells represent a unique aspect of cryptococcal biology. Recent work in mouse models of infection have demonstrated that Titan cells comprise 20% of fungal cells in the lung and are associated with dissemination to the brain and a non-protective immune response [5, 7, 8, 19]. Titanisation requires the activity of the Gα protein Gpa1 and the G-protein coupled receptor Gpr5, as well as the mating pheromone receptor Ste3a, likely targeting the cAMP/PKA pathway [20]. Transcription factors that influence cAMP-regulated capsule and melanin also influence Titanisation [20–22]. However, the environmental triggers of Titanisation remain unknown, and reports of in vitro Titanisation have not led to a robust in vitro protocol for their generation [4, 20, 23, 24].
The analysis of Titan cells recovered from infected mice has led to the identification of four defining features: Titans are larger than 10 μm, are polyploid (typically 4-8C, although higher ploidies have been reported), have a tightly compacted capsule, and have a dramatically thicker cell wall [6, 8, 18]. Here we report a simple and robust protocol for the in vitro generation of cells matching this definition. We validate the capacity of our protocol to identify genes required for Titanisation, and predict the capacity of clinical isolates to form Titans. Finally, we identify environmentally relevant ligands that trigger the yeast-to-Titan transition and begin to dissect the underlying molecular mechanisms that drive this novel virulence mechanism.
While investigating the impact of nutrient starvation on virulence factor production, we observed that when C. neoformans cells grown in Yeast Nitrogen Base (YNB) with 2% glucose were transferred to 10% HI-FCS at 5% CO2, 37°C, large cells (up to 50 μm) formed after several days (Fig 1A, 3 days; S1A Fig, 7 days). These cells expressed a compact capsule that was readily distinguished from the more typical yeast capsule by India ink staining (Fig 1A). Similar effects were observed for cells grown in 10% native FCS or heat inactivated (HI)-FCS but not for culture-matched cells transferred to 1xPBS. Given the reported capacity of C. neoformans to form polyploid Titan cells, these large cells were examined for DNA content. Induced cells were passed through an 11 μm filter to enrich for large cells. Both fractions were collected, fixed and stained for DNA content (Fig 1B, S1B Fig gating strategy). While un-induced cells showed distinct 1C and 2C peaks representing progression of haploid yeast through the cell cycle, induced cells additionally showed discrete peaks consistent with populations of higher ploidy cells. The filtered population was enriched for large cells with increased DNA content (Fig 1B), consistent with these large cells being polyploid Titan cells.
Titan cells observed in vivo typically comprise up to 20% of the total cell population, and lower inocula are associated with an increase in the proportion of Titan cells [20]. We likewise observed that a minority of cells fell into the >10 μm category and that the percentage and overall cell size increased at decreasing inoculum concentration (Fig 1C). While large cells were readily observed at OD600 = 0.25, there was an increase in the frequency and size of larger cells at lower optical densities (OD600 0.05, 0.01). Optimization revealed that larger than average cells (10–12 μm) could be observed after 24 hr, but that cells approaching 15 μm were readily observed after 72 hr. Therefore, all subsequent characterizations were performed using an overnight culture of YNB+Glucose grown cells inoculated into 1xPBS + 10% HI-FCS at OD600 = 0.001 and incubated for 72 hr at 37°C, 5% CO2. The induced phenotype was reproducible across labs and users (University of Aberdeen (TD, ERB), University of Birmingham (ERB, LTS, XZ), Clemson University (LK)).
Where yeast cells typically range in size from 5–7 μm, Titan cells have been defined as being >10 μm or >15 μm, and some definitions have included capsule (>30 μm) [20, 23]. When we measured the cell body diameter of H99 induced cells, we observed a size range spanning 3–15 μm (Fig 1D and 1E). Cells >10 μm represented 15.72 ± 4.46% of the population. We also observed that cells with a diploid base ploidy tended to produce a higher proportion of cells >15 μm (Fig 1D and 1E). Cell size and ploidy are proportional, and we tested the impact of base ploidy on induced cell size. When cell body alone was considered, there was a shift in the size and frequency of cells >10 μm between haploid (H99) and diploid (KN994B7#16) base ploidy, although this did not reach significance (Fig 1E; p = 0.0785). When capsule was taken into account, the difference in size became highly significant (Fig 1E; p<0.0001), suggesting that capsule size increases with base ploidy. Additionally, we observed that cell:capsule ratios were not uniform across the entire population as cell body size increased (S1C Fig): The cell:capsule ratio was significantly smaller for cells >10 μm than for cells <10 μm (H99: 1.425 vs. 1.696; p<0.0001).
Titan cells have been reported to have thicker cell walls than yeast cells and to contain a single large vacuole of unknown function. TEM analysis (Fig 2A and 2B) revealed that cells >10 μm had significantly thicker cell walls (314.7 ± 64.0 nm) than those <10 μm (167.3 ± 46.2 nm; p = 0.002). Large cells were mostly devoid of organelles. In rare instances, some cytoplasmic material could be observed along the cell cortex of large cells, consistent with the presence of a large vacuole (S1D Fig). A third population of small (2–4 μm) cells was also observed (Figs 2A, 2B, 3A and 3B). TEM revealed that these small, encapsulated cells resembled yeast in that they appeared metabolically active, with ribosomes, mitochondria, and nucleus readily visualized, and capsule observed extending from the cell wall (Fig 2A). However, where yeast and yeast daughters are round, these tended to be oval and had significantly thinner cell walls (56.14 ± 26.8 nm; p = 0.026) (Fig 2B). These cells appear to be distinct from the previously reported micro-cells, defined as <1 μm with thick cell walls [25]. Because of their association with Titan inducing conditions, we here term these cells Titanides in order to distinguish them from yeast and micro-cells.
Titan cells are uninucleate, highly polyploid, and produce haploid, aneuploid, or diploid daughters [6, 23]. To investigate these features in in vitro induced cells, we used the GFP-Ndc1 mCherry-Cse4 reporter strain CNV111 [26]. GFP-Ndc1 is targeted to the nuclear envelope and mCherry-Cse4 labels a proportion of kinetochores in non-dividing cells. In yeast phase cells, GFP-Ndc1 could be observed surrounding a cluster of mCherry-positive points. Under inducing conditions, large cells likewise contained a single nucleus and were capable of passing DNA to daughter cells (Fig 3A).
Haploid C. neoformans cells have 15 chromosomes [27]. The mCherry-Cse4 reporter has been used as a proxy for nuclear content, where individual points represent chromosomes [26]. In our hands, we never observed more than 9 distinguishable foci in log-phase haploid cultures, and most cells showed 4 foci arrayed within the nuclear membrane, consistent with previous reports [26] (Fig 3A). When Cse4 foci were quantified using z-stack images, un-induced cells grown in YPD had on average 3.96 ±1.363 foci (n = 200) (Fig 3A, S1E Fig). Growth in YNB did not significantly impact the number of resolvable foci (4.26 ±1.342 (n = 200); p = 0.048; S1E Fig). In YNB-FCS induced cells >10 μm it was not possible to fully resolve the densely-packed foci (Fig 3A and 3B). To further investigate, DNA content in Titan mother and budded cells was estimated based on fluorescence intensity (Fig 3A and 3B representative overlay image). We observed a statistically significant overall increase in nuclear fluorescence intensity in cells >10 μm compared to YPD grown cells (mean: 2.281 ± 0.66 vs. 1.172 ± 0.12; p<0.0001; max: 4.462 vs. 1.419; Fig 3C). However, in budding Titans, we observed daughters with resolvable foci and DNA content consistent with haploid cells (Fig 3A and 3C). These observations are consistent with reports that polyploid Titans divide DNA asymmetrically, producing haploid daughter cells. Based on these data, including size, capsule, cell wall, and ploidy, we suggest that YNB-serum induced large cells are in fact bona fide Titan cells.
Having identified robust conditions capable of replicating in vivo Titan induction, we set out to more closely observe changes in DNA content following large cell induction. Induced populations such as those presented in Fig 3B were examined for mCherry-Cse4 intensity, including cells <10 μm, predicted to comprise a mix of yeast and Titan daughters. We observed an overall increase in fluorescence intensity relative to YPD grown yeast (Fig 3C, 1.94 ± 0.3, p<0.0001), suggesting aneuploidy in the population. These cells were also larger on average than YPD-grown cells (5.678 ± 0.74 vs. 5.071 ± 0.60; p = 0.0007, n>50, Fig 3D). Closer examination of this population showed it to be highly heterogeneous, with cell size ranging from 2 μm to 9.9 μm (Fig 3B), and individual cells <10 μm exhibited a wide range in size and relative fluorescence (Fig 3B and 3D). In some instances, cells <10 μm closely resembled cells >10 μm in terms of morphology and nuclear content (Fig 3B, compare T and <). In other instances, yeast sized cells displayed higher than normal relative mCherry-Cse4 fluorescence (Fig 3B, compare Y and *). We also observed cells much smaller than yeast size with mCherry Cse4 fluorescence typical of yeast (Fig 2B, compare Y and t). This heterogeneity is represented graphically in Fig 3D (n = 100 induced cells total). In general, nuclear content was proportional to cell size (Fig 3D).
To further study the impact of induction on cell ploidy, and to rule out condition-dependent artefactual changes in fluorescence, we analysed induced cells incubated on YPD by flow cytometry. Individual Titans (>10μm) were isolated by microdissection and allowed to proliferate on YPD agar at 30°C for 16 hr. The entire colony was picked and immediately fixed for analysis (Fig 3E top). In addition, individual daughters were dissected from Titan-derived colonies and further incubated on YPD agar at 30°C for 24 hr to form colonies. Fig 3E shows representative flow cytometry data measuring DNA content for fourteen daughter cells arising from a single Titan mother. At the time of dissection, these daughters were diploid or aneuploid relative to the H99 haploid parent (Fig 3E, lower right) and showed cell size consistent with diploid DNA content (S2A Fig). Daughters were incubated on YPD agar at 25°C for 1 month and then analysed again. While some daughters resolved back to haploidy, others were stable within this time scale (Fig 3E, lower left).
Our in vitro induction protocol is a two-step process: cells are first incubated under minimal media conditions, and then induced to undergo the yeast-to-Titan switch via exposure to FCS. FCS is commonly used to induce capsule following growth in rich medium [28], suggesting that the pre-growth condition is relevant for Titanisation. YNB-grown cultures reach lower OD than YPD-grown cultures after 16 hours (mean OD600 = 2.733 ± 0.5608 YNB vs 16.67 ± 4.91 YPD). Given the observed impact of subsequent inoculation density on Titanisation (Fig 1C), we tested whether changes in secreted factors dependent on overnight culture cell density might repress Titanisation. YPD-grown cells were washed 6 times in PBS to remove residual exogenous compounds and incubated in 10% HI-FCS at OD600 = 0.5 or 0.001 (Fig 4A). Titan cells were not observed in either YPD or YNB-pre-grown cultures at OD600 = 0.5. At OD600 = 0.001, washed YPD-grown cells produced large cells at rates similar to YNB-grown cells (p>0.99). However, where YNB-grown Titan cells produce disproportionately small daughters, similar in size to yeast daughters, YPD-grown large cells frequently produced large buds, proportional to the large mother cell and not consistent with previous descriptions of in vivo Titan cell behavior [4, 29]. YPD-grown mother-daughter pairs also tended to be dysmorphic, with defects in cytokinesis, atypical of the reported morphology of in vivo Titan cells (Fig 4A)[4, 20].
Having established that Titan cells can be generated from haploid cells in vitro, we next tested whether Titan cells can produce Titan progeny. H99 haploid cells were pre-grown in YNB and then induced to form Titan cells overnight. After 24 hours, cells were passed through an 11 μm filter and the cells >11 μm were collected, stained with calcofluor white (CFW), and returned to fresh inducing conditions at OD600 = 0.001. After 72 hours, the heterogeneous population included Titan cells with robust capsule stained with both high levels of CFW and no CFW (Fig 4B).
These data suggest that nutritional pre-culture and induction cell density influence the generation of Titan cells, and that Titan cells can be stably maintained in vitro.
Titan cells have been identified in the host lung and brain, but have not been observed circulating in the blood or CNS. To test the impact of host-relevant inducing compounds, we asked whether murine Bronchial Alveolar Lavage (BAL) extract could induce Titan cells. When 10% BAL was used in place of FCS, we observed large polyploid cells similar to FCS-induced Titans (Fig 5A). Daughter cells arising from BAL-induced Titans were micro-dissected and cultured as described above for FCS-induced daughters. BAL-induced Titan daughters also exhibited a shift in base ploidy to 2C and 4C, with daughters arising from the same mother showing a range in base ploidy (Fig 5B). Quantification of FCS and BAL-induced Titan cells showed statistically similar populations (Fig 5C). Therefore, BAL fluid and FCS share the same capacity to induce the yeast-to-Titan transition.
BAL extract contains lung-resident bacteria, a normal component of the host microbiome and bacterial cell wall has been identified in FCS as a ligand for C. albicans morphogenesis [30–32]. To model the role of the host microbiome on Titanisation, we tested the impact of co-culture with gram-negative Escherichia coli and gram-positive Streptococcus pneumoniae for Titan induction. Both share a peptidoglycan cell wall, while gram-negative bacteria additionally have a lipopolysaccharide coat. Co-culture of YNB-grown C. neoformans and either live or heat-killed E. coli or live S. pneumoniae was sufficient to induce Titan cells after 24 hr (Fig 5A and 5C).
We tested the in vivo relevance of the host microbiome on Titan cell induction by comparing fungal cell size in the lungs of infected mice to fungal cell size in the lungs of mice pre-treated with antibiotic water for 7 days. There was no difference in fungal CFUs between treated and untreated mice (p>0.085). Whereas bacteria could be cultured on LB at a low level from the homogenized lungs of untreated mice, bacteria in the lungs of treated mice was below the threshold of detection (S3A Fig). We cannot rule out the presence of non-culturable bacteria in the lungs of these mice. We examined lung homogenates (Fig 5D) and histology (S3A Fig) for evidence of Titanisation. Although large cells were observed in homogenates from both treated and untreated mice, there was a significant reduction in median cell size for treated mice (Fig 5D, (untreated = 12.65 ± 5.11 vs. treated = 9.32 ± 4.14; n>500 p<0.0001)) and a 32.9% reduction in cells >10 μm, suggesting that antibiotic treatment reduced the degree of Titanisation in the lungs, possibly through perturbation of the host environment. Exposure of C. neoformans to antibiotic had no impact on Titanisation in vitro (S3B Fig).
Together, these data suggest that host-relevant factors in both FCS and BAL modulate C. neoformans Titanisation and suggest that bacterial factors influence C. neoformans morphogenesis. We therefore aimed to determine the minimum components of FCS necessary to trigger large cells. HI-FCS was fractionated by size exclusion chromatography, and YNB-grown H99 cells were incubated in 10% compositions of each fraction in 1xPBS. Large cells (>10 μm) were observed in cultures incubated with fractions from wells C11-D11, matching a large peak that eluted after 11 min (Fig 5E, S3C Fig). Comparison to size standards suggested that compounds in this peak are in the range of 500 Daltons. We further fractionated the pooled sample by HPLC and tested the fractions for inducing activity. Analysis by 1H-NMR and DOSY suggested a complex mixture of at least 9 different compounds (Fig 5F). NMR data suggested the presence of a metabolite with a sugar component (δH 4.89, 4.32/4.30, 4.27, 3.79 and 3.70 ppm) and an alkyl chain (δH 2.08, 2.03, 0.87ppm). Additionally, 1H NMR and 1H-13C HSQC experiments exhibited a methylene (δH 3.12 ppm, 52.4 ppm) likely to be located in alpha conformation to a carbonyl and an amino group, which suggested the presence of an amino acid substructure in this metabolite. These features are consistent with peptidoglycan structures. Coupled with our observation that bacterial cell wall from both gram-positive and gram-negative cells is capable of triggering Titanisation, we hypothesized that this metabolite might represent a bacterial peptidoglycan.
Muramyl tetrapeptides (MTP) are peptidoglycan subunits common to the cell walls of Gram negative, Gram positive, and myco-bacteria. Muramyl tetrapeptides consist of an ether of N-acetylglucosamine (GlcNAc) and lactic acid (MurNAc), plus a species-specific tetrapeptide. MTPs act as signaling molecules in both mammalian and fungal cells by binding Leucine Rich Repeat (LRR) domains in target proteins, including mammalian NOD receptors, expressed on phagocytes and epithelial cells in the lung, and C. albicans adenylyl cyclase [32, 33]. MTP and its derivatives were identified as potent inducers of the yeast-to-hyphal transition in C. albicans following spectroscopic analysis of serum, which was shown to contain low levels of bacterial cell wall component [32]. The synthetic Muramyl Dipeptide (MDP), N-Acetylmuramyl-L-alanyl-D-isoglutamine (NMAiGn), is structurally similar but not identical to MTP. 1H NMR analysis of NMAiGn was consistent with the peptidoglycan peaks identified in the FCS fractions. Therefore, we tested the capacity of MNAiGn to influence C. neoformans morphogenesis. Titan cells were induced using 2 mM or 4 mM NMAiGn (the concentration sufficient to trigger the yeast-to-hyphal switch in C. albicans). Cells incubated with NMAiGn exhibited limited proliferation; however, cells >10 μm were present at both concentrations, consistent with a yeast-to-Titan switch (Fig 5A and 5C).
Individual large cells were isolated by microdissection and allowed to proliferate for 17 hr at 30°C on YPD agar. Of 6 large cells isolated, all 6 proliferated to form colonies. For four of these colonies, individual daughters, distinguishable through their reduced size relative to the mother, were again isolated and allowed to proliferate for a further 72 hrs. The remaining 2 colonies (T5, T6) from the original large cells were analysed in aggregate. Each of the resulting lineages was analysed by flow cytometry for ploidy. In NMAiGn-induced daughter cells, we observed an overall increase in ploidy, with the majority of colonies arising from individual daughters having a 4C base ploidy. A representative lineage is shown in the right panel of Fig 5B. Aggregate samples (T5, T6) were more heterogeneous and included 2C and 4C cells, consistent with diploid daughter lineages (Fig 5B right). Together, these data demonstrate a role for peptidoglycan such as MDP during in vitro Titanisation and suggest that bacterial components influence Titanisation in vivo.
Muramyl dipeptide is thought to interact directly with the LLR domain of adenylyl cyclase, and the cAMP signal transduction cascade is believed to regulate Titanisation in vivo [20, 22, 32]. However, addition of exogenous cAMP at levels sufficient to induce capsule failed to induce Titan cells in either YPD or YNB-grown cultures. The avirulence of mutants deficient in cAMP signal transduction has precluded direct testing of this model [20, 34, 35]. We therefore examined the influence of GPA1, CAC1, and PKA1, as well as RIC8, a Gα Guanine Nucleotide Exchange Factor (GEF) for Gpa1, on in vitro Titanisation [36]. Strains deficient in each of these genes failed to generate large cells in our assay (Fig 6A and 6B). The G-protein coupled receptor Gpr5 is required for Titanisation, and the gpr4Δgpr5Δ strain exhibits a significant reduction in Titan cell production in vivo [7, 20]. We likewise observed a decrease in the frequency of Titan cells in vitro in the gpr4Δgpr5Δ strain (Fig 6A and 6B). Consistent with the incomplete defect observed in vivo [20], rare Titan cells could be observed in vitro for this strain (Fig 6A and 6B). Although cap59Δ cells were smaller overall, we observed no specific defect in the capacity of the capsule deficient strain to form Titan cells, ruling out that Titan defects in this pathway are related to defects in capsule synthesis (Fig 6A and 6B). Together, these data demonstrate that in vitro-induced Titan cells are regulated via a similar pathway to in vivo Titan cells.
The C2H2 transcription factor Usv101 is a master regulator of C. neoformans pathogenesis that negatively regulates capsule and acts downstream of Swi6, a regulator of cell cycle progression [37–39]. Usv101 is additionally predicted to regulate Gpa1 but is not itself directly influenced by cAMP[37]. We therefore investigated the role of Usv101 in in vitro Titanisation. Consistent with its role as a negative regulator, usv101Δ produced significantly more and larger titan cells in vitro compared to the H99 parent (Fig 6A and 6C; 39.25±6.45%, p<0.0001). No difference in cell size was observed during YNB pre-culture (H99: 5.918±0.8126; usv101Δ: 5.674±0.9084; p = 0.017). Titans evade phagocytosis and are predicted to drive dissemination through the production of daughter cells, but inactivation of USV101 increases phagocytosis of yeast-phase cells [37]. We hypothesized that Titan usv101Δ cells might fail to produce daughter cells required to drive dissemination to the brain. We therefore measured the relative production of daughter cells by purified cultures of Titan cells from H99 vs. usv101Δ inocula. When Titanized cells were taken as the starting culture, no difference in the total number of cells produced over time was observed for the two strains (Fig 6D, proliferation rate, p = 0.116). However, there was a significant difference in the Titanisation rate (proportion of yeast vs. Titan) between the two strains, with usv101Δ Titan daughters 4.5 times more likely than H99 Titan daughters to form new Titan cells (Fig 6E, H99 Y/T- = 0.0777 ± 0.0312; usv1010Δ Y/T = 0.353 ± 0.0467; p = 0.008). These data suggest that the increased capacity of the usv101Δ mutant to form Titan cells over time may contribute to the previously reported reduced dissemination and reduced virulence of this strain in vivo [37].
We examined the capacity of non-H99 strains to produce Titan cells in vitro. Titanisation has not been reported for Cryptococcus gattii, and no increase in cell size was observed for C. gattii isolate R265 (S4A Fig). Next, we screened 62 environmental and clinical C. neoformans isolates representing VNI, VNII, and VNB clades [40]. Strains were classified as Titanising, non-Titanising, or Indeterminate (S4B Fig). A wide variety of cell sizes were observed in response to inducing conditions, and representative isolates Zc1, Zc8, and Zc12 (VNI clade) are shown in Fig 7A. After growth on YPD, these isolates are morphologically similar, and are capsular, thermotolerant, and melanising, comparable to H99, but exhibit distinct Titanisation profiles (Fig 7A, S4C Fig). Across the 62 isolates, we observed a wide range in Titanisation capacity in both clinical and environmental isolates from each clade, including clinical strains with defects in Titan cell production (Zc1, Zc12; Fig 7A), and environmental strains that produced Titan cells (S8963, Ze14 (VNB-B); S4B Fig). We also observed non-H99 clinical strains that Titanised (Zc8; Fig 7A) and environmental strains that did not (Ze18 (VNB-B), S4B Fig). Together, these data suggest that the yeast-to-Titan switch is a conserved morphogenic transition that can occur across the C. neoformans var. grubii species complex, but that individual isolates within each clade exhibit different capacities to form Titans.
Finally, we validated the capacity of our in vitro assay to predict in vivo outcome in a murine inhalation model of infection, the current gold standard for Titanisation analysis, using the type strain H99 and a clinical isolate predicted not to form Titans, Zc1 (Fig 7A and 7B, p = 0.0184). Mice infected with Zc1 or H99 were observed for 7 days and then sacrificed, and the lungs and brain were collected. Notably, there were clear differences in lung pathology at 7 days despite comparable lung CFUs (S4D Fig). Lungs from H99-infected mice exhibited large lesions or granuloma in contrast to lungs from Zc1-infected mice, which exhibited fewer or no apparent lesions or granuloma (S4E Fig). Histology also revealed foci of encapsulated fungi in the lungs of H99-infected mice, with heterogeneous cell size including both Titan (>15 μm) and yeast (<10 μm) cells (Fig 7C and 7D). In contrast, histology of the lungs of Zc1 infected mice revealed disseminated infection, with encapsulated yeast distributed throughout the lung parenchyma. The population was more uniform in size, with the vast majority of cells less than 10 μm (Fig 7C) (mean 13.9±4.6 vs. 7.08±2.38; p<0.0001).
Given the differences in histo-pathology, we measured relative pathogenicity using a long-term survival assay in Balb/C mice. H99 was significantly more virulent than Zc1 (p = 0.007, Fig 7E). No significant difference was observed in lung CFU on day of sacrifice (p = 0.0575, Fig 7F), however H99-infected mice exhibited significantly higher CFUs in the brain (p = 0.0005, Fig 7G). A similar trend was observed when the same analysis was performed in C57Bl/6J mice (S4G Fig). This is consistent with previous observations of differential tropism when Titan cells are present [19].
Titanisation is associated with altered immune response and increased dissemination to the brain [7, 19]. We therefore investigated the impact of the two strains on immune response in the lungs on day 7 post-infection. With the distinct nature of capsules between Titanising and non-Titanising isolates, we speculated that these two groups might express unique PAMPs and thus differ in their interactions with immune cells. For this purpose, we focused our attention on cells of myeloid origins. The total number of CD45 cells was not significantly different (p = 0.158). We observed recruitment of leukocytes into the lungs of both groups of mice, primarily comprised of CD11b+ granulocytes (Fig 8A and 8B). Three distinct subsets characterized by Ly6G and Ly6C expression were observed: mature neutrophils (Ly6Ghi, gate I) and two populations of immature Ly6Gint neutrophils expressing Ly6Glo (gate II) and ly6Chi (gate III) (Fig 8A). While H99-infected mice had more mature neutrophils (12% vs. 3.7%; p = 0.0299, gate I) as well as Ly6Clo immature neutrophils (37.6% vs 5.2%; p = 0.0002, gate II), Zc1-infected mice exhibited significantly higher percentage of the Ly6Chi immature neutrophil pool (Fig 8A, 76.1% vs 19.5%; p = 0.0079, gate III). The remaining CD11b+ non-neutrophil compartment contained, amongst others, eosinophils (SiglecF+) and monocytes (Ly6G- Ly6Chi) (Fig 8B). Eosinophils were found to be higher in H99-infected mice (53.5% vs 28.8%, p = 0.0085; Fig 8B, gate I) and monocytes were elevated in Zc1-infected mice, although the difference did not reach significance (Fig 8B, gate II; p = 0.3095). We also observed differences in the frequency of cells expressing the MHC-II molecule involved in antigen presentation (Fig 8C). Although CD11b+ MHC-IIhi cells (gate I) were present in both groups, mice infected with H99 displayed an increasing trend of this subset, while Zc1-infected mice had a significant increase in CD11b+ MHCIIlo cells (p = 0.0079, gate II) (Fig 8C).
Taken together, these data suggest that our in vitro assay accurately predicts in vivo Titanisation and that isolates that form Titan cells drive quantitatively distinct immune responses from those that do not.
The yeast-to-Titan transition is a host-specific morphogenic switch that can influence disease outcome. Titan daughters have altered stress resistance compared to their mother cells, and the presence of Titans is associated with altered immune status [6–8]. Despite their importance, mechanisms underlying Titanisation have been challenging to dissect due to the lack of a reproducible in vitro induction protocol. Here we present a rapid, robust in vitro induction protocol that generates cells with all the properties of in vivo Titan cells, recapitulates previously identified regulators (Gpr4/Gpr5), directly confirms the role of cAMP pathway elements (Gpa1, Cac1, Pka1) and identifies new regulators (Ric8, Usv101). The assay further accurately predicted an in vivo defect in Titanisation by a clinical isolate, Zc1.
We define in vitro Titans as those >10 μm, and show that in our assay approximately 15% of H99 cells form Titans within three days (Fig 1). Additionally, we show that purified Titan cultures produce heterogeneous cell populations, including new Titan cells (Figs 3B and 4B). Therefore, our growth conditions are sufficient for the induction and maintenance of Titan cell cultures. Whereas budding haploid yeast undergo symmetric DNA division and produce uniform populations of haploid daughters that are proportional in size, in vivo-derived polyploid Titan cells produce small (5μm) haploid, aneuploid, or diploid daughters [3, 6]. Likewise, we show that in vitro Titans are uninucleate and divide DNA asymmetrically (Fig 3). We also distinguish between bona fide in vitro Titan cells and Titan-like cells induced from rich media pre-culture, which tended to form large buds and accumulated chains of large cells that failed to complete cytokinesis (Fig 4A).
Some definitions of Titan cells include capsule when determining Titan cell size (>30 μm) [23]. During yeast phase growth, there is a demonstrated a relationship between capsule size and the length of the G1 phase of the cell cycle [21, 41]. In these reports, capsule size increases with cell body size and defects in cell cycle control influence capsule expression. Consistent with this, we observed that ploidy influences capsule (Fig 1E): Titan cells induced from diploid parents were significantly larger than those from haploid parents when capsule was taken into account, but the difference was not significant when cell body alone was examined. We also show that under Titan inducing conditions the capsule:cell body ratio changes as cells cross the 10 μm threshold (S1C Fig). For example, for the cells shown in Fig 1A, the yeast cell is 4.98 μm with an 18.49 μm capsule (ratio of 3.70), while the Titan cell is 21.9 μm with a 37.17 μm capsule (ratio of 1.69). This is consistent with observations that Titan capsule synthesis is distinct from yeast phase capsule synthesis and suggests a difference in the regulatory control of the two phenotypes [18, 23]. Future work will examine the specific impact of in vitro Titan inducing conditions on capsule regulation and structure.
C. neoformans is a globally distributed environmental fungus, and the mammalian lung is not thought to be a reservoir for C. neoformans[42]. Rather, our data suggest that the host lung may serve as a niche for bacterial-fungal interactions that mediate pathogenesis. Titan cells are observed in the host lung, an environment with a poorly understood but complex microbiome [31], and BAL can replace FCS as the inducing compound. BAL samples from healthy individuals are positive for bacterial 16S RNA (8.25 log copies/ml) and FCS contains 0.1–0.5 μM Muramic acid, a marker of bacterial peptidoglycan[30, 32]. We identified structures consistent with peptidoglycan in serum fractions capable of inducing Titanisation. Antibiotic treatment that reduced culturable bacterial lung burdens reduced Titan induction in a murine inhalation model, and co-culture of live or heat-killed E. coli or live Streptomyces pneumoniae (a dominant genus of the healthy lung) with C. neoformans led to Titanisation (Fig 5A, 5C and 5D)[30]. Exposure of primed cells to NMAiGn, a synthetic version of the bacterial cell wall component MDP, was sufficient to induce Titan cells (Fig 4A and 4B). MDP and its derivatives are known to activate cAMP-mediated morphological transitions in Candida albicans and other ascomycetes[32]. Together, these data point to a conserved mechanism for bacterial-fungal interactions underlying morphological transitions and highlight the importance of polymicrobial interactions for understanding Cryptococcus pathogenesis, adding to increasing importance of the host lung microbiome in health and disease [31, 43].
There are some differences between our findings and the published literature. First, in vivo generated Titan cells achieve extreme cell size within three days (Fig 7C)[3, 19]. Single cell analysis of our in vitro Titan cells did not identify cells exceeding 30 μm after 3 days, however we did observed cells > 60 μm after 7 days continuous culture (S1A Fig). In addition, we identify a subpopulation of cells with very thin cell walls and altered cell shape, which we term Titanides. These cells appear to be distinct from previously described in vivo micro-cells (<1μm, thick cell walls) [25] and typical yeast daughter cells and accumulate during in vitro Titan cell induction (Figs 2A, 2B and 3B). Based on their altered cell wall, Titanides are likely to differ in the exposure of host relevant ligands relative to yeast cells, and their small size may facilitate dissemination, either through phagocytosis or through increased penetration of the lower airways, similar to the dissemination of basidiospores [44]. Despite these differences in our single cell analysis, bulk analysis of total cell cultures (>10,000 cells) revealed a heterogeneous cell size population that also exhibits a wide range in cell ploidy consistent with previously reported in vivo Titan populations (Fig 1B). Future work will further characterize these cells and investigate the role of the various sub-populations in pathogenesis.
Second, while we observed asymmetric division of nuclear content in dividing Titan cells (Fig 3A), microscopic analysis using a mCherry-Cse1 reporter strain for DNA content of individual cells in the heterogeneous population suggested than the majority of cells are diploid or aneuploid. Additionally, FACS analysis of colonies derived from single daughter cells isolated from in vitro Titan mothers were aneuploid, diploid or, in some cases tetraploid, and cell size in these populations was consistent with this increased ploidy (Fig 3C, 3D and 3E; S2 Fig). In contrast, colonies arising from a limited number of in vivo Titan mothers were comprised of primarily haploid or aneuploid cells [6]. However, Gerstein et al. also report that 25% of independent in vivo-derived Titan daughters were diploid. This highlights an important aspect of Titan cell biology that has proved challenging to study. Titan cells are thought to allow phenotypic diversity through the generation of aneuploid daughters, with increased access to the fitness landscape as a result of changing gene dosage[6]. Current models suggest that uninucleate, highly polyploid mothers bud off haploid or aneuploid daughters, requiring asymmetric DNA division via an unknown mechanism. Efforts to understand the molecular mechanisms underlying this unusual process will benefit from an in vitro model, and our data already suggest that the diversity of these daughter cells is greater than previously described.
Our in vitro model offers some initial insights into the underlying molecular mechanisms regulating Titanisation in vivo. First, in vitro Titans form following exposure to low nutrient conditions and dependent on cell density, suggesting that Titanisation occurs in response to a two-part Prime-Induce signal. C. neoformans growth in minimal media is known to alter the expression of secreted proteins relative to YPD and influences stress resistance through transcriptional and post-translational changes [45, 46]. Exposure to serum is a known signal for capsule induction via the cAMP/PKA pathway [28]. Interestingly, Zaragoza et al. have previously reported that co-incubation of C. neoformans in serum + Sabouraud-Dextrose can increase the average cell size (up to 9 μm), yet represses the influence of serum on capsule [28]. Additionally, serum is a potent inducer of the morphogenic switch from yeast to hyphae in C. albicans and Yarrowia lipolytica via cAMP and Ras1 [32, 47, 48]. In the case of C. albicans, bacterial MDP was identified as the essential component of serum driving the activation of Cyr1 and cAMP signaling. In C. neoformans, previous work has strongly suggested a role for the cAMP signal transduction cascade in Titanisation in vivo [20]. Our finding that bacterial MDP similarly induces the Yeast-to-Titan transition suggests a similar signaling cascade may be in place.
Because gpa1Δ, pka1Δ, and cac1Δ strains are rapidly cleared from the host lung, direct testing of their role in Titanisation was not previously possible in vivo [20, 34, 35]. Here, we confirm this model through direct demonstration that cells deficient in adenylyl cyclase activity, but not capsule biosynthesis, fail to form Titans in vitro (Fig 6A and 6B). Despite the requirement for adenylyl cyclase activity, constitutive activation is not sufficient to induce Titans. The addition of exogenous dcAMP does not induce Titanisation nor does it restore Titanisation to cAMP/Pka pathway mutants in our in vitro assay. Additionally, hyper-activation of the pathway using GAL-inducible PKA1 and PKR1 constructs produces a heterogeneous population of both large, polyploid cells and yeast phase cells after 48hr [22]. Similarly, expression of a constitutively active version of the Gα protein Gpa1 doubles the percent Titan cells in vivo [20]. In each of these cases, Titan cells make up a fraction of the total cell population. These data suggest that Titan induction via cAMP/Pka1 interacts with metabolic, transcriptional or post-translational priming of individual cells to determine cell fate upon exposure to inducing conditions such as serum, resulting in a heterogeneous population.
In addition to positive cAMP regulation, we identify the transcription factor Usv101 as a negative regulator of Titanisation (Fig 6). Although USV101 has been shown to be dispensable for virulence, murine infection results in delayed dissemination to the brain and is characterized by pneumonia rather than meningitis [37]. Gish et al. demonstrated that usv101Δ fails to cross an in vitro blood-brain barrier model and usv101Δ yeast are phagocytosed more readily than wild type cells, partially explaining the altered virulence of this strain. This is somewhat surprising, as together with capsule-independent direct crossing, phagocytosis by trafficking macrophages is thought to facilitate transmigration of the BBB [49]. Here, we show that cells lacking USV101 form Titan cells at a high frequency compared to the parental strain (Fig 6A and 6C) and that these usv101Δ Titans themselves form Titans at a higher rate than wild-type cells (Fig 6E). We suggest that increased Titanisation and decreased availability of non-Titan fungal cells inhibits dissemination of this strain outside of the lung.
In addition to its influence on capsule, Usv101 is predicted to act in parallel to the cAMP pathway and downstream of the cell cycle regulator Swi6. The interaction between cell cycle and pathogenicity factor expression is an emerging theme in C. neoformans biology: Recent work has also highlighted cell cycle regulation of pathogenicity factors [39] and the cyclin Cln1 has been shown to regulate capsule and melanin, both of which are regulated by cAMP and negatively regulated by Usv101 [37, 38, 41, 50].
Finally, the clinical isolate Zc1 and the clinical type strain H99 elicited distinct immune responses. While both H99 and Zc1 strains induced leukocyte recruitment into the lungs, granulocytes predominated the response, and these could be clustered into three unique subsets (a mature neutrophil subset (Ly6Ghi) and two distinct immature neutrophil subsets (Ly6Clo and Ly6Chi)). H99 infection was associated with increased frequency of mature neutrophils and the LyC6lo immature neutrophil subset, whereas the Ly6Chi immature neutrophils were dominant during infection with the Zc1 isolate. It is not known whether these cells express different effector functions and what their polarization state is, and thus more work is required to better understand if they mediate protection or susceptibility to infection with C. neoformans. Other notable differences in immune responses involved disparate frequencies of eosinophils and monocytes recruited during infection. Increased frequency of eosinophils and CD11b+MHC-IIhi was observed in mice infected with H99 relative to the Zc1 isolate while the frequency of monocytes and CD11b+MHC-IIlo cells was higher in Zc1-infected mice. Overall, Zc1-infected mice exhibited moderate lung pathology and reduced dissemination to the brain (Fig 7D–7G), suggesting that there might be qualitative differences in the immune responses driven by Zc1 vs. H99.
We note that other labs have recently reported alternate conditions for inducing Titan-like cells in vitro, either through incubation in SabDex+FCS+sodium azide or through exposure to hypoxia, low pH, and low nutrient conditions [51, 52] Together, these data suggest that Titanisation, like filamentation in C. albicans, is a morphogeneic transition that can be initiated in response to a variety of different stress conditions. One intriguing model for the induction of Titan cells suggests a role for the host immune system: in two studies, mouse genotype interacted with C. neoformans cell size [53] [8]. Our findings that Titanisation capacity varies across clinical and environmental isolates, as well as our data demonstrating that Titanisation can be triggered by bacterial MDP, adds additional complexity to these observations. The interaction of the host with fungal and bacterial co-infecting species is a theme of emerging importance in our understanding of fungal pathogenesis, both in the context of increased disease severity and through the inhibition of pathogenicity [43, 54, 55].
Our in vitro system enables ex vivo analysis of the role of specific host factors in Titanisation; in vitro dissection of the molecular mechanisms driving Titanisation; and improved understanding of the interaction between Titanisation and pathogenesis. The complex interaction of Titanisation and pathogenesis is highlighted by the findings that both Zc1, a Titan deficient isolate, and usv101Δ, a hyper-Titanising mutant, cause pneumonia rather than disseminated disease and meningitis [37, 40]. Zc1-infected mice exhibited significant lung pathology, including leukocyte recruitment on day 7 and high lung fungal burden on day of cull in two different models of infection (Fig 7F, S4D, S4E and S4F Fig). Our data suggest that morbidity due to pneumonia might be an important factor to consider during infection with clinical or environmental isolates. While we hypothesize that the Titanisation defect of Zc1 is primarily responsible for its delayed dissemination relative to H99, it is also possible that other differences between the two strains, or in the host response to infection, drive observed differences in pathogenesis. However, Zc1, like H99, is a VN1 clade patient isolate with no defects in the classic pathogenicity factors capsule, thermotolerance, or melaninsation. Dissemination to the brain was similar in female TH2-tilted Balb/C and male TH1-tilted C57Bl/6 mice [56, 57]. Titanisation capacity appears to be the single largest difference between Zc1 and H99, and is representative of the wide variety in Titanisation phenotypes for environmental and clinical isolates. The relative impact of Titanisation on pathogenesis and clinical outcomes is a pressing question, and future work will investigate this further, particularly in the context of immune-altered states such as neutropenia and T and B lymphocytopenia.
Strains used in this study are summarized in S1 Table. C. neoformans H99 [58], gpa1Δ, cac1Δ[34], and pka1Δ[59] were gifts from Andrew Alspaugh, Duke University, NC, USA. The gpr4Δ gpr5Δ [60] was kindly provided by Joseph Heitman, Duke University, NC, USA. Strains ric8Δ and usv101Δ were obtained from the Madhani 2015 collection (NIH funding, R01AI100272) from the Fungal Genetics Stock Centre and were validated by PCR and shown to phenotypically match published strains [36, 37]. C. gattii R265 [61] was provided by Neil Gow, University of Aberdeen, UK. Isolates are summarized in S1 Table. Cells were routinely cultured on YPD (1% yeast extract, 2% bacto-peptone, 2% glucose, 2% bacto-agar) plates stored at 4°C. For routine culture, cells were incubated overnight in 5 mL YPD at 30°C, 150 rpm. For Titan induction, cells were incubated overnight at 30°C, 150 rpm in 5 mL YNB without amino acids (Sigma Y1250) prepared according to the manufacturer’s instructions plus 2% glucose. Fetal Calf Serum (FCS) was obtained from either BioSera (Ringmer, UK) or Sigma, which both induced Titan cells to a similar degree. FCS was routinely stored in 5 ml aliquots at -20 to prevent repeated freeze-thaw cycles. FCS was heat-inactivated by incubation at 56°C for 30 min.
Cells were induced and either fixed with 4% methanol free paraformaldehyde and permeabilised using 0.05% PBS Triton-X and stained for total chitin with calcofluor white (CFW, 10 μg/ml) and DNA with SytoxGreen (Molecular Probes, 5 μg/ml) or stained live using calcofluor white (CFW, 5 μg/ml) and the cell permeable nucleic acid stain SybrGreen (Invitrogen, 0.5X). SybrGreen preferentially stains dsDNA, but has low affinity for ssDNA and ssRNA, so is not appropriate for quantitative DNA analysis. However, it does allow visualization of nucleic acid dynamics within live cells. Cells were imaged using a Zeiss M1 imager for fixed cells and either a Zeiss Axio Imager or a Nikon Eclipse TI live imager, both equipped with temperature and CO2 control chambers for live imaging. To visualize capsule, live cells were counterstained using India Ink (Remel; RMLR21518) or fixed and stained with mAB 18B7 and counterstained with 488-antimouse IgG. Representative images are shown. Cell diameter was measured using FIJI, with frames randomly selected, all cells in a given frame analysed, and at least three images acquired per sample for each of two independent runs, representing experimental duplicates. In all instances unless otherwise stated, n>200 cells. Statistical analyses were performed using Graphpad Prism v7. For pairwise comparisons, the Mann-Whitney test was applied. For multiple comparisons, ANOVA and Kruskal-Wallis were applied. Significance was taken as p<0.01 throughout.
Cells were fixed and stained according to the protocol of Okagaki et al 2010[3]. Briefly, cells were fixed with 4% methanol free paraformaldehyde and permeabilised using 0.05% PBS Triton-X. Cells were washed 3 times with 1x PBS and stained with 300 ng/ml DAPI. Where indicated, samples were enriched for large cells by passing through an 11 μm filter prior to staining. Cells were analysed for DNA content using an LSRII flow cytometer on the Indo-1 Violet channel and 10,000 cells were acquired for each sample. Data were analysed using FlowJo v. 10.1r7. Doublets and clumps were excluded using the recommended gating strategy of SSC-H vs SSC-W followed by FSC-H vs. FSC-w, and cells were then gated to exclude auto-fluorescence using unstained pooled haploid and diploid control cells. The gating strategy is provided in S1B Fig. Gates for 1C, 2C, 4C, and 8C were established using H99 (haploid) and KN994B7#16 (diploid) controls incubated in rich media conditions as described above.
Cells of each type were pre-grown in YNB and then induced to form Titan cells overnight. After 24 hours, cells were passed through an 11 μm filter and the cells >11 μm were collected, normalized to 103/ml, and then returned to inducing conditions. After 48 hours, cells were collected, fractionated by size (>11 μm, <11 μm), and counted. Titanisation rate was expressed as a ratio of cells <11 μm / >11 μm. Data represent three independent biological replicates. Statistical analyses were performed using Graphpad Prism v7, via unpaired t-test. Variance within the two strains was not statistically significantly different (p = 0.6187).
H99 cells were incubated overnight at 30°C, 150 rpm in 5 ml YNB without amino acids + 2% glucose. Cells were inoculated into 1xPBS+ 0.04% glucose (the concentration of glucose present in serum) in the presence or absence of 106 live or heat-killed E. coli (DH5α) or live S. pneumoniae (R6). Co-cultures were incubated for 24 hr at 37μC 5%CO2 and assessed for Titan formation by microscopy as described above. Co-cultures were not sustainable after 24 without supplementation with additional nutrients, and experiments were therefore terminated. Data representative of triplicate independent experiments are shown.
To identify compounds of interest, total HI-FCS (Biosera) was loaded onto an AKTA purifier system from GE Healthcare with an Agilent column: Bio SEC-3, 100A, 4.6x300mm at a flow rate of 0.3 ml/min for size exclusion chromatography. For the initial run, 400 μl serum was run in phosphate buffer (25 mM NaH2PO4, 150 mM NaCl, 0.01% NaN3, 2 mM EDTA, pH 7.2), collecting 100 μl fractions in a 96 well plate. The entire plate was screened for capacity to induce Titans by incubated H99 cells pre-grown in YNB+Glucose at OD600 = 0.01 in 10% fraction+1xPBS in a 96 well plate format. Plates were examined for Titan cells after 48 and 96 hr. The entire assay was run in triplicate and twice independently using distinct lots of FCS. Inducing fractions were pooled and lyophilized. The residue was resuspended in MeOH and desalted. Then, the solution was submitted to HPLC separations, which were carried out using a Phenomenex reversed-phase (C18, 250 × 10 mm, L × i.d.) column connected to an Agilent 1200 series binary pump and monitored using an Agilent photodiode array detector. Detection was carried out at 220, 254, 280 and 350 nm. The entire volume was purified by RP-HPLC using a gradient of MeOH in H2O as eluent (50–100% over 70 min, 100% for 20 min) at a flow rate of 1 ml/min. The main fraction was dried, suspended in a minimal volume of DMSO-d6 and submitted to 1H-NMR, 1H-13C HSQC and 1H-DOSY analyses. NMR data were acquired on a Bruker 500 MHz spectrometer.
All animal experiments were performed under UK Home Office project license PPL 70/9027 which was reviewed and approved by the University of Aberdeen Animal Welfare and Ethical Review Body (AWERB) and the UK Home Office and granted to DMM. Animal experiments adhered to the UK Animals (Scientific Procedures) Act 1986 (ASPA) and European Directive 2010/63/EU on the protection of animals used for scientific purposes. All animal experiments were designed with the 3Rs in mind and were reported using the ARRIVE guidelines.
C57BL/6J mice were bred and maintained in individually ventilated cages (IVCs) at the Medical Research Facility at the University of Aberdeen. Balb/C female mice were obtained from Harlan Laboratories (UK). For each experiment, group size was determined based on previous experiments as the minimum number of mice needed to detect statistical significance (p<0.05) with 90% power (α = 0.05, two-sided). Mice were randomly assigned to groups by an investigator not involved in the analysis and the fungal inocula were randomly allocated to groups. Inocula were delivered in a blinded fashion. Mice were provided with food and water ad libitum. Mice were monitored for signs and symptoms of disease. Weight was recorded daily. Mice showing weight loss of greater than 30% (C57BL/6J) or 20% (Balb/C) and signs of disease progression were immediately culled by a schedule one method (cervical dislocation).
C57BL/6J male mice (n = 5/group (immunology) or 10/group (survival), 8–12 weeks old) or Balb/C female mice (n = 10/group, 6–8 weeks old) were anesthetized using injectable anesthesia and infected intranasally with 20 μl PBS suspension containing 105 C. neoformans (H99 or Zc1) pre-grown in Sabouraud Dextrose medium [8]. For 7 day studies, mice were culled by euthatal injection. Lungs and brains were collected under sterile conditions. Whole brains and one lung lobe were weighed and homogenized for CFU counts. For long term infection studies, mice were observed with daily records of body weights. Mice that reached a predetermined threshold of >25% (C57BL/6J) or >20% (Balb/C) weight loss, or signs and symptoms of neurological disease, were immediately culled. Mice were humanely sacrificed by cervical dislocation following precipitous weight loss (>20%) and the assay was terminated after 28 days. Survival data were assessed by Kaplan Myer and Gehan-Breslow-Wilcoxon test. Lung and brain were sterilely collected, weighed, and homogenized in 1 ml sterile PBS, and 10 μl was plated for CFUs.
Lungs from infected mice were used to generate single-cell suspension using mouse lung dissociation kit and the gentleMACS as per manufactures’ instructions (Miltenyi Biotec). Fungal cells were separated from mammalian cells via a 70%/30% discontinuous Percoll gradient centrifugation. Immune cells were stained with the fixable viability dye eFluor 455UV (eBiosecience) for 30 min at 4°C, washed with 1x PBS then fixed with a 2% paraformaldehyde (PFA) solution for 10 min at room temperature. Cell surface staining with antibody cocktail of mAbs specific to CD45-BV650, MHC-II-PECSF594, CD11c-APC, CD11b-BUV395, Ly6C-PE, Ly6G-FITC (all from BD Biosciences) was performed in FACS buffer containing 2% fetal calf serum, 2 mM sodium azide and anti-CD16/32 for 30 min at 4°C, washed then acquired on the BD Fortessa cell analyser (BD Biosciences). FlowJo software v10 (Tree Star) was used for data analysis.
Data represent percent live CD45+ cells. Statistical analyses were performed using Graph Pad Prism (v 7), and significance was determined using Mann-Whitney U test. Bars represent 95% CI. Variance within the groups was not statistically different (F test to compare variances).
For lung histology, C57BL/6J male mice (n = 5/group, 8–12 weeks old) were anesthetized and infected intranasally with 20 μl PBS suspension containing 105 C. neoformans (H99 or Zc1) pre-grown in Sabouraud Dextrose as above. After 7 days, mice were culled by euthatal injection. Lung sections were preserved in OTC medium and sectioned (2–4 μm) for histology. Fungi were visualized by silver staining and hematoxylin counterstain (Sigma HHS32) using the Sigma-Aldrich Silver Stain modified GMS kit according to the manufacturer’s instructions).
BAL was performed on male C57BL/6 mice (8–12 weeks) culled by CO2 exposure. Lungs were perfused with 1 ml 1xPBS and the collected fluid concentrated by overnight drying on a speedvac. The resulting pellet was weighed, resuspended in sterile PBS, and used at a concentration of 10% w/v in place of FCS in the induction protocol. Animal experiments were performed by ID, ERB, AC, and DMM.
|
10.1371/journal.pbio.1001241 | Rapid Evolution of Enormous, Multichromosomal Genomes in Flowering Plant Mitochondria with Exceptionally High Mutation Rates | Genome size and complexity vary tremendously among eukaryotic species and their organelles. Comparisons across deeply divergent eukaryotic lineages have suggested that variation in mutation rates may explain this diversity, with increased mutational burdens favoring reduced genome size and complexity. The discovery that mitochondrial mutation rates can differ by orders of magnitude among closely related angiosperm species presents a unique opportunity to test this hypothesis. We sequenced the mitochondrial genomes from two species in the angiosperm genus Silene with recent and dramatic accelerations in their mitochondrial mutation rates. Contrary to theoretical predictions, these genomes have experienced a massive proliferation of noncoding content. At 6.7 and 11.3 Mb, they are by far the largest known mitochondrial genomes, larger than most bacterial genomes and even some nuclear genomes. In contrast, two slowly evolving Silene mitochondrial genomes are smaller than average for angiosperms. Consequently, this genus captures approximately 98% of known variation in organelle genome size. The expanded genomes reveal several architectural changes, including the evolution of complex multichromosomal structures (with 59 and 128 circular-mapping chromosomes, ranging in size from 44 to 192 kb). They also exhibit a substantial reduction in recombination and gene conversion activity as measured by the relative frequency of alternative genome conformations and the level of sequence divergence between repeat copies. The evolution of mutation rate, genome size, and chromosome structure can therefore be extremely rapid and interrelated in ways not predicted by current evolutionary theories. Our results raise the hypothesis that changes in recombinational processes, including gene conversion, may be a central force driving the evolution of both mutation rate and genome structure.
| A fundamental challenge in evolutionary biology is to explain why organisms exhibit dramatic variation in genome size and complexity. One hypothesis predicts that high rates of mutation in DNA sequence create selection against large and complex genomes, which are more susceptible to mutational disruption. Species of flowering plants in the genus Silene vary by approximately 100-fold in the rates of mutation in their mitochondrial DNA, providing an excellent opportunity to test the predicted effects of high mutation rates on genome evolution. Contrary to expectation, Silene species with elevated mutation rates have experienced dramatic expansions in mitochondrial genome size compared to their slowly evolving relatives, resulting in the largest known mitochondrial genomes. In addition to the increases in size and mutation rate, these genomes also reveal a history of rapid change in genome structure. They have been fragmented into dozens of chromosomes and appear to have experienced major reductions in recombination activity. All of these changes have occurred in just the past few million years. This mitochondrial genome diversity within the genus Silene provides a striking example of rapid genomic change and raises new hypotheses regarding the relationship between mutation rate and genome evolution.
| Explaining the origins of variation in genome size and complexity has become the defining challenge for the field of molecular evolution in the genomic era. Historically, numerous evolutionary models have been developed, involving mechanisms such as insertion and deletion (indel) bias [1],[2], selfish element proliferation [3],[4], and natural selection on cell size [5], replication rate [6], and evolvability [7]. In recent years, a body of theory known as the mutational burden hypothesis (MBH) has emerged as a potentially unifying explanatory framework rooted in the principles of population genetics and the basic evolutionary processes of mutation and genetic drift [8],[9]. The MBH posits that noncoding elements are generally deleterious but proliferate nonadaptively when small effective population sizes reduce the effectiveness of selection relative to genetic drift, offering an explanation for why noncoding sequences are so abundant in large multicellular eukaryotes. This hypothesis is based on the idea that noncoding elements impose a selective cost associated with the increased chance of mutations disrupting an essential genome function (e.g., alteration of a conserved sequence required for intron splicing) or generating a novel deleterious feature (e.g., an improper transcription-factor binding site in an intergenic region). The MBH has potentially sweeping explanatory power, but some of its tenets are controversial [10]–[13], and its generality as a mechanism of genome evolution remains uncertain [14]–[21].
Mitochondrial genomes display striking diversity in size and complexity [22],[23], reflecting patterns of variation in genome architecture observed more broadly across the tree of life [9],[24]. For example, in contrast to the small (typically 14–20 kb) and streamlined genomes found in most animal mitochondria [25], seed plant mitochondrial genomes are very large (200–2,900 kb), containing introns and abundant intergenic sequences [26]–[28]. Plant mitochondrial genomes are also typically characterized by extremely low point mutation rates, further distinguishing them from their fast-evolving animal counterparts [29]–[31]. The observed disparity in mitochondrial mutation rates across eukaryotes motivated the hypothesis that mutation rates are a major determinant of variation in organelle genome architecture [32]. This argument is a direct extension of the MBH and is based on the premise that the probability of mutational disruption of noncoding elements (which is equivalent to the selective cost associated with maintaining those elements) is directly proportional to the mutation rate. Therefore, genomes with elevated mutation rates are predicted to experience more intense selection for genomic reduction [32].
The discovery that some angiosperms have greatly accelerated mitochondrial mutation rates, sometimes orders of magnitude greater than closely related species [33]–[35], presents an opportunity to test the prediction that high mutation rate environments select for reduced and streamlined genomes. In particular, several species in the genus Silene (Caryophyllaceae) have experienced dramatic increases in mitochondrial mutation rates within just the last 5–10 Myr, while other members of this genus have maintained their ancestrally low rates [35]–[37].
We compared complete mitochondrial genome sequences from four Silene species with very different mutation rates and found that accelerated mutation rates have indeed been associated with dramatic changes in genome size and complexity. However, the direction of these changes is not always consistent with the predictions from existing theory. We discuss the implications of the unprecedented mitochondrial genome diversity found within Silene and possible alternative explanations for the rapid genome evolution in this genus.
Sequencing of purified mitochondrial DNA (mtDNA) from three Silene species generated complete genome assemblies for S. noctiflora and S. vulgaris and a high quality draft assembly for S. conica. We also included the previously published mitochondrial genome of S. latifolia in our analyses [38]. The genomic data extend previous results [35]–[37] by showing that S. noctiflora and S. conica have experienced massive accelerations in nucleotide substitution rates (Figure 1) across all protein genes (Figure 2) with correlated increases in the frequency of both insertions and deletions (Figure 3).
Contrary to the prediction of genomic streamlining in response to high mutation rate, the fast-evolving mitochondrial genomes of S. noctiflora and S. conica have experienced unprecedented expansions, resulting in sizes of 6.7 Mb and 11.3 Mb, respectively. In contrast, the more typical slowly evolving mitochondrial genomes of S. vulgaris (0.43 Mb) and particularly S. latifolia (0.25 Mb) are on the lower end of the angiosperm size range. Thus, Silene mitochondrial genomes have diverged more than 40-fold in size in just the past few million years.
The genomic expansion in S. noctiflora and S. conica does not reflect detectable increases in gene or intron content. Although these genomes contain duplicate copies of some genes (particularly rRNA genes; Table S1), they possess fewer unique genes than other angiosperm mitochondrial genomes (Figures 1 and 4). Notably, the S. conica and S. noctiflora mitochondrial genomes contain only two or three identifiable tRNA genes, which is far fewer than most angiosperms and even less than the already reduced tRNA gene content of S. latifolia and S. vulgaris (Figures 1 and 4) [38]. The four Silene genomes have nearly identical sets of introns (Table 1). With the exception of additional intron copies associated with gene duplications, there were no intron gains among the four Silene species and only one observed intron loss (the third intron of nad4 in S. noctiflora). Interestingly, in contrast to the overall pattern of genome expansion in S. noctiflora and S. conica, average intron lengths in the expanded S. noctiflora and S. conica genomes are actually ∼10%–15% shorter than in their congeners (Figure S1).
Intergenic sequences account for 99% of the bloated mitochondrial genomes in S. noctiflora and S. conica. As in other vascular plants [28],[39], the intergenic regions of all four Silene mitochondrial genomes contain sequences of both nuclear and plastid (chloroplast) origin. Although the expanded mitochondrial genomes of S. noctiflora and S. conica contain more of this “promiscuous” DNA than their smaller Silene counterparts (Table 1), contributions from these sources do not scale proportionally with the increases in genome size and constitute less than 1% of the intergenic content in both species (Table 1). A larger fraction of the intergenic regions in each of these two genomes exhibit similarity to sequences in other plant mitochondrial genomes (Table 1), but most of this sequence (>650 kb) is only shared between S. noctiflora and S. conica and not with any other angiosperms. Overall, >85% of the voluminous intergenic sequence in these two species lacks detectable homology with any of the nuclear, plastid, or mitochondrial sequences available in the GenBank nr/nt database.
Repeated sequences constitute a variable and often large component of seed plant mitochondrial genomes [40], and Silene species are noteworthy in both respects (Figures 5, S2, and S3; Table 1). The S. conica mitochondrial genome contains a remarkable 4.6 Mb of dispersed repeats, which is more than any other sequenced plant mitochondrial genome in both absolute and percentage (40.8%) terms [40]. The largest repeats are >80 kb in size, but the bulk of the repetitive content consists of an enormous number of small, imperfect, and often partially overlapping repeats (Figures 5, S2, and S3). In contrast, repeat sequences make up just 6.7%–18.8% of the other three Silene mitochondrial genomes.
Silene noctiflora and S. conica have also evolved extraordinary mitochondrial genome structures. Although the relationship between genome maps and in vivo physical structure remains uncertain for angiosperm mtDNAs [41], the entire sequence content of the genome typically can be mapped as a single “master circle,” which can be subdivided into a collection of “subgenomic circles” that arise via high-frequency recombination between large direct repeats (Figure S4A) [42],[43]. This model applies to S. latifolia [38], whereas the S. vulgaris genome assembles into four circular-mapping chromosomes, with the largest (394 kb) comprising most (92%) of the genome and containing numerous repeats inferred to undergo active recombination on the basis of their association with alternative rearranged genome conformations (Figure S4). Two of the three smaller mitochondrial chromosomes in S. vulgaris share recombinationally active repeats with the large chromosome, but the majority of sequencing reads support the smaller subgenomic conformations (see Materials and Methods and Figure S4). In contrast, the smallest of the four S. vulgaris chromosomes appears to be almost completely autonomous. It does not share any repeats longer than 100 bp with the rest of the genome, and in the case of all shorter repeats shared between the smallest chromosome and the main chromosome, >99.5% of sequencing read-pairs support the smaller subgenomic conformation. While the presence of this small chromosome is itself unusual for plant mtDNAs, far more extreme are the S. noctiflora and S. conica mitochondrial genomes, each of which assembled into dozens of mostly autonomous and relatively small, circular-mapping chromosomes. The S. noctiflora mitochondrial genome consists of 59 circular-mapping chromosomes ranging from 66 to 192 kb in size (Table S2). Many of these do not share any large (>1 kb) repeats with other chromosomes. Even when S. noctiflora chromosomes do share large repeats (up to 6.3 kb), the clear majority of paired-end sequencing reads (>90% in all cases) support the conformation consisting of two smaller circles rather than a single combined circle. Although the extremely repetitive nature of the S. conica mitochondrial genome precluded complete genome assembly, its structural organization is similar to that of S. noctiflora. The vast majority (98.2%) of sequence content assembled into 128 circular-mapping chromosomes ranging from 44 to 163 kb in size (Table S2). Most of these chromosomes share only short repeats with other parts of the genome.
The number of sequencing reads that cover a given position in a shotgun genome assembly (i.e., the read depth) can be used to estimate the relative abundance of different sequences. The difference in average read depth between the chromosomes with the highest and lowest coverage was only 1.7-fold in S. noctiflora and only 3.1-fold in S. conica (after excluding repetitive regions), indicating that the abundance of the numerous chromosomes was relatively even in both genomes. The different chromosomes also exhibited a high degree of similarity in GC content within each genome (Table S2).
Assembly of repetitive genomes is inherently complicated, and this is particularly relevant to the identification of genomic subcircles because tandem duplications within a larger chromosome can misassemble as subcircles. However, such assembly errors leave clear signatures, including dramatic variation in read depth and conflicting read-pairs associated with the boundary between tandem repeats and flanking regions. The absence of such patterns in our dataset indicates that the assembled circles are not an artifact of tandem repeats within larger chromosomes. Nevertheless, it is possible, particularly in the draft assembly of S. conica mitochondrial genome, that some repeat pairs have been “collapsed” into single sequences, leaving open the possibility that the reported 11.3 Mb genome size for S. conica is a slight underestimate.
Sequencing of the S. latifolia mitochondrial genome showed that it contains a six-copy 1.4-kb repeat that is highly recombinationally active with physical cross-overs between repeat copies generating a suite of rearranged genome conformations [38]. Southern blot analysis confirmed that the many alternative genome conformations occur in roughly equivalent frequencies in S. latifolia [38]. Paired-end sequencing reads can also be used to quantify the relative abundance of alternative genome conformations (see Materials and Methods and Figure S4), and our 454 data suggest a comparably high level of repeat-mediated recombinational activity for the largest repeats in the S. vulgaris mitochondrial genome (Figure 6A). The relative frequency of recombinant genome conformations increases with repeat size, and all surveyed repeats longer than 100 bp exhibit evidence of a history of recombination. The two largest surveyed pairs of repeated sequences (0.9 and 3.0 kb) in the S. vulgaris genome each appear to be at or near a 50∶50 level of alternative genome conformations (Figure 6A).
The rapidly evolving mitochondrial genomes of S. noctiflora and S. conica exhibit reduced frequencies of recombinant genome conformations compared to other Silene genomes (Figure 6B) and all other angiosperm mitochondrial repeats for which recombinational activity has been assessed. Even the largest repeats in the S. noctiflora genome (up to 6.3 kb) are associated with only a small minority of recombinant products (Figure 6B). The largest repeats in the S. conica genome (up to 87 kb) far exceed our paired-end library span and therefore cannot be analyzed for recombinational activity, but analysis of the shorter repeats suggests that the genome has experienced a similar shift in the relationship between repeat length and the frequency of recombinant products (Figure 6B). Recombinational activity (including gene conversion) is expected to homogenize copies of repeated sequences throughout the genome. Therefore, the dramatic increase in the proportion of divergent pairs of repeated sequences within the mitochondrial genomes of S. noctiflora and S. conica (Figures 7 and S5) is consistent with a reduction in recombinational activity in these species, though the existence of divergent repeats could also result from the increased mutation rate in these species or a reduced probability of gene conversion events between physically disparate repeat copies in expanded genomes.
The coexistence of maternally and paternally derived mitochondrial genomes in a heteroplasmic state within the same individual or maternal family would introduce complications for genome sequencing and assembly. Therefore, we looked for evidence of heteroplasmy and nonmaternal inheritance in the families used in this study. S. vulgaris has been the subject of extensive investigation into the patterns of mitochondrial genome inheritance [44]–[47]. These studies have found that mtDNA transmission is predominantly maternal in S. vulgaris, with a low frequency of biparental inheritance or paternal “leakage.” Because of this evidence, the S. vulgaris family used for genome sequencing was chosen, in part, because the maternal source plant had previously been screened with two highly polymorphic mitochondrial markers and revealed no evidence of heteroplasmy [46]. Although similarly intensive investigations of mtDNA inheritance have not been performed in other Silene species, we found evidence of maternal transmission in S. latifolia, S. noctiflora, and S. conica. An analysis of cleaved amplified polymorphic sequences (CAPS) showed that all progeny (16–48 per species) from controlled greenhouse crosses inherited the maternal variant of a SNP. Mitochondrial inheritance therefore appears to be at least predominantly maternal in all four Silene species, making it unlikely that genome assembly complications arising from biparental inheritance and heteroplasmy can explain the observed differences in mitochondrial genome size and complexity among Silene species.
S. noctiflora and S. conica do not show the proportional increases in mitochondrial nucleotide diversity that would be expected on the basis of their accelerated mutation rates (even after accounting for the approximately 2-fold differences in generation times across the four Silene species [48]), suggesting a recent history of lower effective population size (Ne) than their congeners and/or a recent reversion to lower mitochondrial mutation rates as observed in other accelerated angiosperm lineages [33],[34]. In S. conica, there is less than a 10-fold increase in mitochondrial synonymous nucleotide diversity relative to the more slowly evolving Silene species, and S. noctiflora exhibits no sequence variation whatsoever across our sample of mitochondrial, plastid, and nuclear loci (Table S3) (see also [49]).
The dramatic expansion of intergenic content in the mtDNA of S. noctiflora and S. conica has resulted in mitochondrial genomes that are larger than most bacterial genomes (Figure 8) and even some nuclear genomes [50]. These enormous genomes add to the long-standing mystery regarding the origins of intergenic sequences in plant mtDNA [28].
It is possible that a significant portion of this intergenic content is derived from the nuclear genome, for which sequence data are still limited in Silene. However, by comparing the mitochondrial genomes against a large set of cDNA sequences derived from a recent transcriptome project in S. vulgaris [51], we detected similarity for only a trivial amount (<0.1%) of the otherwise uncharacterized mitochondrial sequence in S. noctiflora and S. conica. Therefore, if nuclear DNA is a major contributor to the expanded mitochondrial intergenic regions in these species, it is most likely drawn from the vast repetitive and noncoding fractions of the nuclear genome. That the origin of only a small fraction of the intergenic sequences in S. noctiflora and S. conica can be identified may reflect the rapid rates of sequence and structural divergence in these mitochondrial genomes.
In other plant mitochondrial genomes, the proliferation of “selfish” DNA may have contributed to expansions in intergenic regions. For example, the mtDNA of the gymnosperm Cycas contains numerous copies of repetitive elements known as Bpu sequences [52], and the expanded mitochondrial and plastid genomes of the green alga Volvox share an apparently self-replicating element with the nucleus [53]. The finding of expanded intergenic sequence in S. noctiflora and S. conica mtDNA raises the question of whether some form of selfish element has been involved. This appears possible in S. conica, given the highly repetitive nature of its mitochondrial genome (Figures 5, S2, and S3; Table 1). However, we did not find evidence for any specific sequence or set of sequences that dominate the repetitive content in S. conica. There is even less evidence for a role of mobile, self-replicating elements in S. noctiflora mtDNA given the small amount of repeated sequence in this genome. Interestingly, S. noctiflora harbors a relatively modest proportion of repetitive sequence compared to many other angiosperms' mtDNAs, including the much smaller S. vulgaris genome (Figures 5 and S2; Table 1), indicating that there is no strict relationship between repetitive content and genome size.
It is noteworthy that S. noctiflora and S. conica share a large amount of intergenic sequences with each other (659 kb and 760 kb, respectively) that show little or no homology with any available sequences in the GenBank nr/nt database including all other sequenced plant mitochondrial genomes. These shared intergenic sequences may be the remnants of an ancestral genomic expansion that preceded the divergence of S. noctiflora and S. conica, suggesting a possible sister relationship between these two lineages, an issue that is currently unresolved by molecular phylogeny [37],[54]. If so, this could indicate that the atypical mitochondrial genome size, structure, and substitution rates in S. noctiflora and S. conica represent a single set of evolutionary changes rather than phylogenetically independent events. However, we cannot rule out the possibility that the shared sequences are the result of parallel acquisitions from similar sources, such as the nuclear genomes in each species. Generating sequence data from other genomic compartments, particularly from a large number of unlinked nuclear loci, should provide better insight into the phylogenetic history of these Silene species.
Although the highly multichromosomal genome structures observed in S. noctiflora and S. conica are novel for plant mitochondria, various forms of multicircular organelle genomes have evolved independently in diverse eukaryotic lineages, including in the mitochondria of kinetoplastids [55], diplonemids [56], chytrid fungi [57], and a number of atypical metazoans [58]–[61], as well as in dinoflagellate plastids [62]. In addition, the recent analysis of the cucumber mitochondrial genome showed that a small fraction of that genome can be mapped to two circular chromosomes that appear to be independent from the main chromosome [63].
It should be noted that the maps generated from the assembly of DNA sequence data do not necessarily reflect the structure of the genome in vivo. In particular, linear concatamers and overlapping linear fragments can assemble as circular maps [64]. Efforts to directly observe the molecular structure of angiosperm mitochondrial genomes have identified a complex mixture of linear, circular, and branched molecules [65],[66], indicating that the circular maps produced by genome projects may be abstractions or oversimplifications. Although on the basis of our current data we cannot distinguish between the various structural alternatives capable of producing circular chromosome maps, the sequence assemblies do support the intriguing finding that many of these chromosomes are structurally autonomous, lacking the large, recombinationally active repeats that are characteristic of most angiosperm mitochondrial genomes.
The existence of multichromosomal mitochondrial genomes in Silene raises fundamental questions about the nature of replication and inheritance of these genomes. Notably, we did not detect a single intact gene in many chromosomes, including the smallest chromosome in S. vulgaris, 20 of the 59 chromosomes in S. noctiflora, and 86 of the 128 chromosomes in S. conica (note that these totals do not include chromosomes in S. noctiflora and S. conica that only contain partial gene fragments that require trans-splicing with transcripts originating from other chromosomes to generate complete coding sequences). Therefore, the functional significance (if any) of these “empty” chromosomes and the evolutionary forces that maintain their presence and abundance within the mitochondrion are unclear. While it is possible that these chromosomes contain unidentified genes or noncoding elements that are functionally important and therefore conserved by selection, they may also replicate and proliferate in a nonadaptive or even selfish fashion.
Our analysis was based on mtDNA extracted from predominantly vegetative tissue pooled across multiple individuals from a single maternal family. Therefore, we do not know whether any of the observed structural variation in mtDNA is partitioned within our pooled sample and, if so, at what level it is partitioned (i.e., among individuals, tissue types, cells, or even individual mitochondria). In this light, it would be particularly informative to conduct an analysis of mitochondrial genome sequence and structure in meristematic tissue to compare with our results from vegetative tissue. Any differences between these tissue types would be of interest because the mtDNA in meristematic tissue should better represent the inherited form of the genome.
The co-occurrence of mutational acceleration and genome expansion in the mitochondria of S. noctiflora and S. conica runs counter to patterns in other eukaryotic mitochondrial genomes (e.g., plants versus animals). Although we cannot determine the relative timing of these changes, their co-occurrence in these lineages is at odds with the hypothesis that reduced mutation rates are a major cause of mitochondrial genome expansion in plants [32].
An alternative possibility that would be consistent with the MBH is that these species have a small Ne, which has reduced the efficacy of selection against the proliferation of noncoding elements even if the intensity of that selection has increased with higher mutation rates. There is some evidence to support this possibility, particularly in S. noctiflora, which appears to have a very low Ne based on the striking lack of polymorphism in genes from all three genomes (Table S3) [49]. However, the finding of high levels of mitochondrial polymorphism in S. conica (Table S3) is contrary to the predictions of the MBH. Some caution is warranted in interpreting the nucleotide diversity data because standing levels of polymorphism are very sensitive to recent bottlenecks and do not necessarily represent the long-term average Ne over the entire history of a species or lineage. One alternative proxy for Ne and the relative strength of genetic drift is the ratio of nonsynonymous to synonymous substitutions (dN/dS), with higher ratios indicating a reduced efficacy of selection in purging deleterious changes in amino acid sequence [12]. Based on this alternative measure, there is no indication of a long-term decrease in Ne in either S. noctiflora or S. conica since their divergence from the other Silene species (Table 1). Therefore, with respect to both mutation rate and Ne, the changes in mitochondrial genome size within Silene appear to be inconsistent with any straightforward interpretation of the MBH.
In contrast to the differences in overall genome size in Silene mitochondria, some of the observed changes in these genomes are consistent with predictions of the MBH. Most notably, average intron lengths have decreased in the species with elevated mutation rates, and the only example of an intron loss was observed in a high-rate species. These results could indicate that the consequences of mutational burden vary substantially within a genome. For example, the contrasting patterns observed in introns versus intergenic regions within these lineages might suggest that the burden associated with disruptive mutations in functional noncoding elements such as introns is of far greater evolutionary importance than that associated with gain-of-function mutations creating novel deleterious elements in largely nonfunctional intergenic regions.
The inability of existing theory to fully account for the extreme patterns of divergence in Silene mitochondrial genomes points to a valuable opportunity to expand our understanding of the evolutionary forces that shape genomic complexity. Although this study was restricted to a small number of species from a single genus, it captured enormous variation in genome architecture (e.g., approximately 98% of the known range of organelle genome sizes), indicating that profound and perhaps novel evolutionary mechanisms are acting to shape mitochondrial genome size and complexity in Silene.
The observed differences among Silene species in the frequency of recombinant genome conformations raise the possibility that recombination could be a key factor underlying the extreme patterns of mitochondrial genome evolution in S. noctiflora and S. conica. The mitochondrial genomes in these species differ from those of other angiosperms in numerous respects, including rates of point mutations and indels, presence of duplicated and divergent gene copies, frequency of RNA editing, genome size, and structural organization (Table 1). Many, perhaps all, of these traits are likely affected by the related processes of intragenomic recombination and gene conversion.
Recombinational processes play an important role in plant mitochondrial genome sequence and structural evolution [43],[67]. In addition, recombination between repeated sequences (including very short repeats) has been shown to be an important mechanism for sequence deletion in plant nuclear genomes [68]. Therefore, changes in recombinational activity are expected to affect the evolution of genome size. However, recombinational processes can also have opposing effects on genome size via sequence duplication or integration of new content, so that the relationship between recombination and genome size is likely to be a complex one. Recombination and gene conversion mechanisms have also been implicated in the evolution of other elements of genome architecture. For example, retroprocessing events involving cDNA intermediates are likely responsible for the loss of introns and RNA editing sites [34],[69],[70].
Recombination and gene conversion are key components of DNA repair pathways. Notably, gene conversion mechanisms that are biased against new mutations have been proposed to slow the effective or observed mutation rate in multicopy genomes [71],[72]. Our findings raise the possibility that template-based recombinational repair and biased gene conversion are important factors underlying the typically low rates of nucleotide substitution in plant mitochondrial genomes and that these mechanisms have been altered or disrupted in fast-evolving species such as S. noctiflora and S. conica. The associated increase in the rate of mitochondrial indels in these species (Figure 3) suggests that alterations in replication and repair machinery can have correlated effects on both point mutations and structural changes, which is consistent with the correlation between rates of mitochondrial sequence and structural evolution observed in other lineages [73]–[76].
Our findings highlight the need to characterize Silene nuclear gene families involved in recombination and other aspects of organelle genome maintenance. Unraveling the process of sequence gain and turnover in these rapidly evolving mitochondrial genomes should provide insight into the evolutionary forces underlying the tremendous variation in size and complexity of eukaryotic genomes.
The genus Silene (Caryophyllaceae) consists of approximately 700 predominantly herbaceous species of flowering plants [77], many of which are used as models in ecology and evolution [78]. S. noctiflora L. and S. conica L. both have annual life histories [79], and they are largely hermaphroditic but produce a low frequency of pistillate (female) flowers and can therefore be characterized as gynomonoecious [80]–[82] (DBS, personal observation). S. latifolia Poir. and S. vulgaris (Moench) Garcke are short-lived perennials with an average generation time of approximately 2 y [48] that maintain dioecious and gynodioecious breeding systems, respectively [79],[80].
Details of the Silene latifolia mitochondrial genome project were described previously [38]. For each of the other three species, approximately 200 g of tissue was collected from multiple individuals of a single maternal family. The maternal lineages were derived from seeds originally collected in Abruzzo, Italy (S. conica), Eggleston, VA, US (S. noctiflora), or Stuarts Draft, VA, US (S. vulgaris). Voucher specimens from each of these maternal lineages have been deposited to Massey Herbarium at Virginia Polytechnic and State University: S. conica (L Bergner 003), S. noctiflora (D Sloan 003), S. vulgaris (L Bergner 007).
All aboveground tissue was used for S. vulgaris, including leaves, stems, and flowers, while only leaf tissue was collected for S. noctiflora and S. conica. Mitochondrial DNA was purified from mitochondria from harvested tissue using established protocols based on differential centrifugation, treatment with DNase I, and then either CsCl gradients or phenol∶chloroform extraction [83],[84]. Restriction digests with MspI and HpaII enzymes, which share identical recognition sequences but differ in methylation sensitivity, were performed to confirm the absence of significant nuclear contamination from the purified mtDNA samples prior to sequencing.
For each of the species, 3-kb paired-end libraries were prepared following standard protocols for sequencing on a Roche 454 GS-FLX platform with Titanium reagents. Additional libraries were prepared (also following standard Roche protocols) for the larger S. noctiflora and S. conica mitochondrial genomes, including shotgun libraries for both species and a 12-kb paired-end library for S. noctiflora. The latter was constructed following the standard 8-kb protocol, but the larger 12-kb average fragment size range was selected on the basis of the size distribution of the DNA sample after shearing. Each library was run on a single quarter-plate region except for the S. conica shotgun library and the S. noctiflora 12-kb paired-end library, which were each run on two quarter-plate regions. The shotgun library for S. noctiflora was constructed and sequenced by the Genome Center at Washington University in St. Louis (MO, US). All other 454 library construction and sequencing was performed at the Genomics Core Facility in the University of Virginia's Department of Biology.
To generate sufficient starting material for Illumina library construction, mtDNA samples were amplified with GenomiPhi V2 (GE Healthcare). Paired-end sequencing libraries were generated and tagged with multiplex barcodes using the NEBNext DNA Sample Prep Reagent set 1 (New England Biolabs) in accordance with protocols developed by the University of California Davis Genome Center. In brief, DNA samples were sonicated to a peak fragment size of between 300 and 600 bp. DNA fragments were then end polished and ligated to adaptors carrying a unique 6-bp barcode. The resulting samples were gel-purified and amplified with 14 PCR cycles using paired-end library primers. The three libraries were included in a larger sample pool and sequenced in a single lane of a 2×85 bp paired-end run on an Illumina GAII. Sequencing was performed at the Biomolecular Research Facility in the University of Virginia's School of Medicine.
Each quarter-plate 454 run produced between 32 and 104 Mb of sequence. The total sequencing yield was 270, 210, and 51 Mb for the S. noctiflora, S. conica, and S. vulgaris mtDNA samples, respectively. However, not all sequence data were used in primary genome assembly. For S. noctiflora, only the shotgun and 3-kb paired-end data were analyzed in the initial assembly process. The 12-kb paired-end data were only used to resolve structures associated with large (>3 kb) repeats and to quantify the frequency of alternative genome conformations resulting from recombination among repeat copies (see below). For the smaller, S. vulgaris mitochondrial genome, a single quarter-plate run produced very high coverage (>80×). Preliminary analyses suggested use of the entire dataset increased fragmentation in the assembly. Therefore, a random set of sequence reads totaling 25 Mb was selected for initial assembly. The full S. vulgaris dataset was used for subsequent quantification of alternative genome conformations.
For each genome, the 454 sequence reads were assembled with Roche's GS de novo Assembler v2.3 (“Newbler”) using default settings. The resulting assemblies produced average read depths of 20×, 25×, and 42× for the S. conica, S. noctiflora, and S. vulgaris mitochondrial genomes, respectively. Although the assemblies contained few, if any, gaps or low-coverage regions, they were highly fragmented because of the repetitive and recombinational nature of these genomes (Figures 5 and 6). The assemblies also contained contigs from contaminating nuclear, plastid, and viral DNA. True mitochondrial contigs were distinguished on the basis of read depth and connectivity to other contigs in the assembly, which was inferred from two types of data: (1) paired-end reads that mapped to two different contigs and (2) single reads that were split by the assembler and assigned to the ends of two different contigs. On the basis of these data, contigs were organized into “subgenomes,” each of which represented either a closed circular assembly or a single-copy assembly flanked on either side by recombinationally active repeats. Each of these subgenomic contig groups was then reassembled using a custom set of Perl and BASH scripts that identified all sequencing reads uniquely associated with the corresponding contigs and ran a new assembly using only those reads. The resulting subgenomic assemblies were then manually edited and combined as necessary with the aid of Consed v17.0 [85].
The largest repeats in both the S. conica and S. vulgaris mitochondrial genomes exceed the 3-kb span size of their respective paired-end libraries. Therefore, the relationships between the single-copy regions flanking these large repeats are ambiguous. These ambiguities were tentatively resolved on the basis of the pattern observed in smaller repeats within each genome (Figure 6).
On the basis of the high level of recombinational activity among smaller repeats in S. vulgaris, we assumed that large repeats also have high recombinational activity. Therefore, we assembled the majority of the S. vulgaris genome content into a single chromosome, analogous to the “master circle” typically reported for plant mitochondrial genomes. This large chromosome contains numerous recombinationally active repeats, and, as discussed previously [38], the arrangement of repeats and single-copy regions within this chromosome should be considered only one of many possible alternative representations. We also identified three small circular-mapping structures that were not included in the main assembly. One of these circles (Chromosome 4) shows almost no evidence of recombinational activity with the rest of the genome, while the other two do share repeats that appear to recombine frequently with the main chromosome. However, in both of these cases, the repeats are small (<500 bp), and the clear majority of reads support the closed circle conformations over a single combined circle. For convenience, we refer to these three circles as chromosomes, but their small size and (in the case of Chromosomes 2 and 3) substantial degree of recombinational activity with the rest of the genome distinguish them from the chromosomal structure that characterizes the S. noctiflora and S. conica mitochondrial genomes.
In contrast to S. vulgaris, the bulk of the S. noctiflora and S. conica mitochondrial genomes map to discrete circular chromosomes that exhibit little or no recombinational activity with the rest of the genome. In both species, repeats show much less evidence of recombination than repeats of similar size in S. latifolia and S. vulgaris (Figure 6). Moreover, in cases of recombinationally active repeats, the clear majority of paired-end reads (>90% in all cases in S. noctiflora and the vast majority of cases in S. conica; Figure 6) support minimally sized circular conformations rather than larger combined circles. Therefore, for assembly ambiguities associated with repeats exceeding the 3-kb paired-end library span in S. conica, it was assumed that minimally sized circles predominate over larger combined conformations.
To correct base-calling errors including insertion and deletion errors known to be associated with long single-nucleotide repeats (i.e., homopolymers) in 454 sequence data, we mapped Illumina sequence data onto the completed mitochondrial genome assemblies for each species. After removal of multiplex barcodes and quality trimming, Illumina sequencing yielded average read lengths between 53 and 69 bp with a total of 398, 326, and 168 Mb of sequence data for S. noctiflora, S. conica, and S. vulgaris, respectively. Paired-end read mapping was performed with SOAP v2.20 [86] with the following parameters: m 100, x 900, g 3, r 2. A set of custom Perl scripts were used to call SOAP, parse the resulting output, and modify the genome sequence on the basis of well-supported sequence conflicts. These scripts were run recursively until additional iterations did not produce any further improvement to the sequence.
For both S. vulgaris and S. noctiflora, Illumina mapping provided high-depth (>10×) coverage for essentially the entire genome (>99.9%). This process identified 55 sequence corrections in S. vulgaris and 1,734 corrections in S. noctiflora, the vast majority of which were associated with homopolymer runs. In contrast, because of the larger size and repetitive complexity of the S. conica mitochondrial genome, more than 10% of the sequence had coverage levels below 10×. Furthermore, the recursive mapping approach described above failed to converge for numerous regions in the genome, indicating low confidence in many of the sequence corrections indicated by the Illumina data. To avoid incorporating false sequence changes, we did not use the Illumina data to perform genome-wide corrections in S. conica. Consequently, the reported genome sequence likely contains some errors associated with homopolymer runs. We did, however, use the Illumina data to verify basecalls in S. conica coding genes and introns, including cases of frameshift mutations.
The annotation of protein, rRNA, and tRNA genes was performed using a combination of local BLAST [87] and tRNAscan [88] as described previously [20]. Annotated genome sequences were deposited in GenBank (Table S2).
To identify sequence of plastid origin in the Silene mitochondrial genomes, each genome was searched against a database of seed plant plastid genomes, using NCBI-BLASTN (v2.2.24+) with the following parameter settings: dust no, gapopen 8, gapextend 6, penalty -4, reward 5, word_size 7. Only hits with a raw score of at least 250 were considered. These hits were subsequently filtered to exclude matches involving mitochondrial protein and rRNA genes known to have ancient plastid homologs (e.g., mitochondrial atp1 and plastid atpA [89]). We also excluded hits with very high AT contents (>72%), because we found these to be almost exclusively false positives resulting from the use of sensitive BLAST parameters.
To identify intergenic sequence conserved in other plant mitochondrial genomes, all intergenic regions (excluding those of plastid origin) were searched against a database of all sequenced seed plant mitochondrial genomes using NCBI-BLASTN (v2.2.24+) and the following search parameters: task blastn, dust no, gapopen 5, gapextend 2, reward 2, penalty -3, word_size 9. All hits with a raw score of at least 70 were considered homologous. Note that we included all sequences from “empty” chromosomes in the intergenic category even though such sequences are not technically bounded by genes on either side.
To identify additional conserved sequences (particularly ones of nuclear origin), the remaining intergenic regions (i.e., excluding annotated genes, plastid-derived sequence, and regions conserved with other plant mitochondrial genomes) were searched against the GenBank nr and nt databases (release date 12/15/2010) using NCBI-BLASTX and BLASTN (v2.2.24+). Default settings were used for BLASTX, whereas the BLASTN search parameters were as follows: dust yes, gapopen 5, gapextend 2, reward 2, penalty -3, word_size 9. All BLASTX hits with a raw score of at least 140 and all BLASTN hits with a raw score of 70 or above were considered homologous. Searches with these same parameters were also conducted against a set of assembled cDNA sequences from a recent S. vulgaris transcriptome project [51].
Tandem repeats in each Silene mitochondrial genome were identified with Tandem Repeat Finder v4.04 [90], but these represented a negligible fraction of total repeat content in each genome and are not reported separately. Dispersed repeats were identified by searching each genome against itself with NCBI-BLASTN (v2.2.24+) using default parameter settings. All hits with a raw score of at least 30 were considered repeats. The shortest possible sequence that can satisfy this criterion is a perfect 30-bp repeat, but longer sequences with less than 100% sequence identity can also be identified by this method. Finally, Vmatch (http://www.vmatch.de) was used to precisely define the boundaries of all repeats with 100% sequence identity.
We used paired-end reads from 454 sequencing to quantify the relative abundance of alternative genome conformations associated with repeat-mediated recombination (Figure S4). In the absence of any recombination or alternative genome conformations, 454 read pairs should map to positions in the genome that are consistent with the size span of the sequencing library (∼3 or 12 kb in this case). However, the presence of genomic rearrangements will result in read pairs that are inconsistent with the reported genome conformation (Figure S4). Therefore, for each pair of repeated sequences in a genome, we quantified the number of 454 read pairs that are inconsistent with the reported genome assembly but are consistent with either of the predicted products of recombination between the repeats. This number was then compared against the total number of consistent read pairs in the genome that span one of the two repeat copies to determine the relative abundance of the recombinant products.
To perform this analysis, 454 paired-end reads were mapped on the corresponding genome sequence using Roche's GS Reference Mapper v2.3 software with default parameters. For S. noctiflora, only reads from the 12-kb paired-end library were used. The resulting output was filtered to exclude duplicate read pairs with identical start positions for both the left and right sequences, as these were assumed to have been generated by the PCR amplification step in paired-end library construction, making them nonindependent data points. Inspection of the mapping output suggested that the analysis was too stringent in identifying consistent read pairs. Therefore, any “inconsistent” read pairs that mapped in a proper orientation within a distance of 4–16 kb for a 12-kb library or 1–6 kb for a 3-kb library were reclassified as consistent. These size ranges were determined on the basis of manual inspection of the distribution of mapping spans.
Identified repeats within each genome (see above) were filtered on the basis of multiple criteria prior to inclusion in recombination analyses. First, only repeats of at least 50 bp in length and at least 95% sequence identity were considered. Additional repeat pairs were excluded because their proximity to each other or to other repeats would have led to ambiguity in the interpretation of paired-end mapping results. Specifically, repeats were excluded if the two copies were separated by less than the maximum library span or if there was a “correlated” pair of larger repeats within the maximum library span of each repeat copy. Finally, for S. conica and S. vulgaris (for which only 3-kb paired-end libraries were available), repeat pairs were excluded if one of the repeat copies was within 100 bp of the start of any other repeat >500 bp in size. These cases were excluded because the presence of adjoining repeats would preclude unambiguous mapping of reads to the flanking sequence. Because of the limited physical coverage and short (3 kb) span length in the S. conica paired-end data, there are many repeat pairs (particularly large repeats) in this genome that passed the aforementioned criteria, but have an insufficient number of read pairs to precisely measure the relative frequency of alternative genome conformations. Therefore, frequencies are only reported for repeat pairs that have at least five consistent read pairs spanning each copy. Finally, because of the enormous number of small repeats in the S. conica mitochondrial genome (Figure 5), only a random sample of 5% of repeat pairs shorter than 200 bp was included.
To validate our methodological approach, we ran a set of control analyses that used the same set of repeats except that we reversed the coordinates for one of the copies. Therefore, these analyses assessed rearrangements associated with the same genomic regions but would only detect alternative genome conformations if recombination occurred between two homologous sequences lined up in opposite orientations. The frequency of alternative genome conformations was at or near zero for every one of these control analyses (Figure S6). This suggests that baseline level of genome rearrangement and chimeric artifacts is very low in our dataset and that the alternate genome conformations detected by these methods are the genuine result of repeat-mediated recombination. In addition, the differences in assembly methods across species (see above) should have no effect on the reported estimates of recombinational activity because these differences only pertain to large repeats exceeding the span of our paired-end libraries, which were not assayed for recombination.
Previous analyses based on individual genes have identified massive variation in mitochondrial substitution rates among genes and species within the genus Silene [35]–[37],[91]. To assess these patterns at a genome-wide scale, all protein genes were aligned with MUSCLE v3.7 [92] and levels of synonymous (dS) and nonsynonymous (dN) divergence were estimated using PAML v4.4 [93] as described previously [37]. Analyses were run both on individual genes and on a concatenated dataset of all shared protein genes. Most analyses included six species (Arabidopsis thaliana, Beta vulgaris, and all four Silene species), but a larger dataset of sequenced seed plant mitochondrial genomes was also analyzed. In all cases, the phylogenetic relationships among the four Silene species were left unresolved (i.e., as a four-way polytomy), reflecting the apparently rapid radiation of these four lineages [37],[54]. Because substitutions at RNA editing sites can artificially inflate estimates of dN [94], we excluded all codons that were found to be edited based on genome-wide datasets from four species [70],[95],[96]. To estimate absolute rates of nucleotide substitution in these genomes, dN and dS values were divided by an approximate divergence time of 6 Myr [35],[37],[97]. However, these estimates should be considered only rough approximations because of the uncertainty in divergence time [37] and the potential bias associated with recent polymorphisms [98],[99].
To determine the frequency and size distribution of indels, all protein genes (including cis-spliced introns) from the four Silene species and the outgroup B. vulgaris were aligned with MUSCLE v3.7 and adjusted manually. Unalignable regions at the 5′ and 3′ ends of genes were excluded. The resulting alignments were analyzed to identify all indels that were unique to a single species and did not overlap with any other indels.
A genome-wide analysis of C-to-U RNA editing sites by cDNA sequencing has been reported previously for S. latifolia and S. noctiflora [70]. To estimate the frequency of RNA editing in S. vulgaris and S. conica, protein gene sequences were analyzed with a predictive algorithm (PREP-mt) [100]. Control analyses using Silene sequences with known editing sites suggested that different stringency settings (C-values) are appropriate for species with different rates of sequence evolution. Specifically, the S. conica data were analyzed with C = 0.8 and the S. vulgaris data were analyzed with C = 0.7. PREP-mt does not identify synonymous editing sites, so the reported totals were increased by 10% to approximate the contribution of synonymous edits on the basis of observed rates in other Silene genomes [70]. All intact protein genes were included as well as the following putative pseudogenes: rps13 (S. latifolia), rps3 (S. conica, S. latifolia, and S. noctiflora), and ccmFc (S. conica). For genes with duplicates within the genome, only a single gene copy was included.
To estimate levels of sequence variation within each of the four Silene species in this study, we PCR amplified and Sanger sequenced a sample of five mitochondrial loci as well as a single plastid and nuclear locus for multiple, geographically dispersed populations. Sequencing methods, source populations, and polymorphism data for S. vulgaris and S. latifolia were reported previously [36],[91]. Source populations for S. noctiflora and S. conica are summarized in Table S4. A single individual was sampled from each population. Sequence data from each species were analyzed with DnaSP v5 [101] to calculate nucleotide diversity and the number of segregating sites for each locus. Maximum likelihood estimates of Watterson's Θ and corresponding 95% confidence intervals were calculated as described previously [91]. For the nuclear X4/XY4 locus, a single haplotype was randomly selected from each individual for calculation of polymorphism data. Only X-linked copies were included for S. latifolia males. Haplotypes were inferred from diploid sequence data using the program PHASE v2.1 [102]. Novel sequences were deposited in GenBank (accessions JF722621–JF722652).
We performed a set of greenhouse crosses to test for maternal transmission of mtDNA in S. latifolia, S. noctiflora, and S. conica (S. vulgaris was not included because it has already been the subject of numerous studies examining mitochondrial genome inheritance and heteroplasmy [44]–[47]). Each cross involved an individual from the maternal family used for mitochondrial genome sequencing and an individual from another family in that species known to differ in mtDNA haplotype. For each species, a single pair of reciprocal crosses was performed, and a SNP was used to design a CAPS marker capable of distinguishing the two parental genomes (Table S5) [103]. For each pair of crosses, 16 to 48 progeny were analyzed with the corresponding CAPS marker.
|
10.1371/journal.pntd.0004918 | Symptomatic Dengue Disease in Five Southeast Asian Countries: Epidemiological Evidence from a Dengue Vaccine Trial | Dengue incidence has increased globally, but empirical burden estimates are scarce. Prospective methods are best-able to capture all severities of disease. CYD14 was an observer-blinded dengue vaccine study conducted in children 2–14 years of age in Indonesia, Malaysia, Thailand, the Philippines, and Vietnam. The control group received no vaccine and resembled a prospective, observational study. We calculated the rates of dengue according to different laboratory or clinical criteria to make inferences about dengue burden, and compared with rates reported in the passive surveillance systems to calculate expansion factors which describe under-reporting. Over 6,933 person-years of observation in the control group there were 319 virologically confirmed dengue cases, a crude attack rate of 4.6%/year. Of these, 92 cases (28.8%) were clinically diagnosed as dengue fever or dengue hemorrhagic fever by investigators and 227 were not, indicating that most symptomatic disease fails to satisfy existing case definitions. When examining different case definitions, there was an inverse relationship between clinical severity and observed incidence rates. CYD14’s active surveillance system captured a greater proportion of symptomatic dengue than national passive surveillance systems, giving rise to expansion factors ranging from 0.5 to 31.7. This analysis showed substantial, unpredictable and variable under-reporting of symptomatic dengue, even within a controlled clinical trial environment, and emphasizes that burden estimates are highly sensitive to case definitions. These data will assist in generating disease burden estimates and have important policy implications when considering the introduction and health economics of dengue prevention and control interventions.
| Dengue is a mosquito-borne, viral febrile disease transmitted between humans in most of the tropical and sub-tropical world. In recent years, an increasing number of cases has been widely reported. However, understanding the full disease burden remains a topic of public health research. One reason for under-reporting is that severe episodes are more likely to be captured in routine surveillance statistics, and mild episodes unreported/unrecognized. We re-analyzed data from the control arm of a dengue vaccine clinical trial in five Asian countries. The trial captured dengue incidence rates following active surveillance and virological confirmation, and we compared those with incidence rates from the passive surveillance system. As expected, the active surveillance system captured many more cases of symptomatic dengue than routine systems. Of virologically confirmed dengue in the clinical trial, only ~29% were diagnosed by investigators as dengue, indicating there is a significant disease burden excluded from existing case definitions and diagnostic practices. The analysis confirmed that dengue is under-reported, by different magnitudes, in these Asian countries. Case definition is an important determinant of burden. These findings are important when considering the health economics and public health impacts of new prevention and control tools.
| Dengue is a viral disease transmitted between humans by Aedes mosquitoes throughout the tropical and subtropical world. Infection may be asymptomatic, or can result in a spectrum of clinical disease including self-limiting fever with manifestations of varying severity (classical dengue fever; DF) progressing to life-threatening dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). [1] While this classification remains in clinical use in some countries, a new system was proposed by the World Health Organization (WHO) in 2009 primarily to improve triage and clinical management, and to capture warning signs of potentially severe dengue episodes. [1,2] Disease prevention efforts with mosquito control have been largely unsuccessful and recent decades have witnessed increased disease frequencies and expanded ranges of transmission. [3,4] Dengue is now endemic in over 120 countries worldwide, with almost half of the global population at risk. [3,5]
Approximately 75% of this at-risk population resides within the Asia Pacific region, where the primary vectors (Aedes aegypti and Ae. albopictus) and dengue virus have become widely dispersed over recent decades following a number of social, environmental, and demographic changes. [6,7] Multiple dengue virus serotypes co-circulate and the disease constitutes a leading cause of hospitalization and death in some countries. [8] In the midst of this expansion, and possibly due to it, reliable dengue disease burden estimates are uncommon. [3,9] Passive national dengue surveillance systems are designed to detect outbreak activity rather than describe burden. [10] More reliable estimates are required to guide disease control programs, allow rational allocation of resources, and assess the impact of new interventions such as dengue vaccination. Accordingly, estimating the true disease burden constitutes one of the WHO’s three objectives in the 2012 Global Strategy for Dengue Prevention and Control 2012–2020. [3]
In most scenarios, national surveillance systems underestimate disease burden due to the non-specific clinical presentation of dengue; unavailability and limitations of confirmatory diagnostic tests; and health system issues that result in incomplete reporting. [11] Under-estimation is typically most severe in the milder manifestations of illness, and is a function both of under-ascertainment and under-reporting. [12] In recent years, several methods have been used to improve the accuracy of historical global disease burden estimates of approximately 100 million infections/year. [3,13] These include empirical methods where overlapping data sources enable estimation of cases missed (capture-recapture studies); expert consensus-based approaches; statistical and/or cartographic methods incorporating dengue occurrence data or their covariates; regression methods to estimate unknown variables; and derivations from seroprevalence data. [9,14–16] Notably, a 2013 study by Bhatt et al. used a cartographic modeling approach combining demographic and epidemiological data, adjusted for clinical severity and determinants of dengue incidence, to estimate a global burden of 390 million (95% credible interval: 284–528 million) infections in 2010, of which 96 million (67–136 million) were symptomatic. [9] It has been estimated that 70% of cases and >50% of the economic burden of dengue are in Asia. [9,17]
Prospective cohort studies utilizing active surveillance can yield more accurate estimates of symptomatic disease than passive surveillance systems. [18] Resulting incidence rates (IRs), when compared with data from routine surveillance systems, can describe the extent of under-estimation, expressed as multiplication or expansion factors (EFs). [12,19] In Cambodia, Thailand, and the Philippines, individual studies using these methods calculated EFs for dengue of between 7.2 and 9.1. [20,21] A review using data from all WHO regions found dengue EFs in Asia of up to 126, with significant variation among countries and over time resulting from different underlying epidemiology, surveillance practices, and comparative study design. [19]
Dengue vaccine clinical trials are conducted with a high degree of operational integrity and produce data closely resembling those from active epidemiological studies. Subjects allocated to the control group do not receive dengue vaccine, so incidence data from these individuals can be interpreted as an observational dengue cohort. [22] CYD14 was an observer-blinded dengue vaccine study conducted in 2011–2013 in 10,275 children aged 2–14 years in Indonesia, Malaysia, Thailand, the Philippines, and Vietnam. [23] Each of these countries conducts passive routine dengue surveillance, sometimes using different case definitions and different reporting, laboratory, and diagnostic practices. [10,11] (described in S1 File).
Dengue epidemiological data from CYD14 and its Latin American sister, CYD15, were recently described by L’Azou et al., allowing comparison across countries of data collected using standardized, active methods. [24] Here, we exploit the comprehensive dataset to further explore dengue incidence in the CYD14 control group according to different clinical endpoints (in addition to the primary clinical endpoint of the efficacy trial) to examine the relationship between burden and severity in five Asian countries. We also made comparisons with national surveillance reports to estimate EFs for symptomatic dengue of different clinical severities, from which broader burden estimates can be inferred.
This was a secondary analysis using records from a vaccine clinical trial. The original clinical trial which generated the data (ClinicalTrials.gov number NCT01373281) underwent ethics committee approval of the protocol, amendments, consent, and assent forms. [23] Parents or legal guardians provided informed consent before participation, and written assent was obtained from older children, in compliance with the regulations of each country. All data were analyzed anonymously.
CYD14 (CT.gov identifier NCT01373281) was an observer-masked, randomized, controlled, multicenter, phase 3 dengue vaccine trial in Indonesia (3 study centers), Malaysia (2 study centers), the Philippines (2 study centers), Thailand (2 study centers), and Vietnam (2 study centers) and has been described previously. [23] There was ethics committee approval of the protocol, amendments, consent, and assent forms. Parents or legal guardians provided informed consent before participation, and written assent was obtained from older children, in compliance with the regulations of each country. Briefly, children aged 2–14 years were randomly assigned to receive three injections of a recombinant, live-attenuated, tetravalent dengue vaccine (CYD-TDV), or placebo, at 0, 6, and 12 months. Participants were followed up actively for a total of 25 months and episodes of fever ≥ 38°C on ≥ 2 consecutive days were recorded and clinically diagnosed as DF or DHF based on 1997 WHO guidelines (Fig 1). Concurrent with and irrespective of clinical diagnosis, serum samples were taken for virological confirmation of dengue by detection of NS1 antigen by ELISA and dengue viral RNA by RT-PCR. A positive result for either laboratory test was considered virological confirmation of acute dengue infection. This allowed febrile individuals to be grouped into four case definitions according to their clinical diagnosis and laboratory results: 1) clinically diagnosed dengue (CDD) was diagnosed by the investigator as dengue, irrespective of the laboratory result; 2) virologically confirmed dengue (VCD) was a dengue virological laboratory confirmation, irrespective of the clinical diagnosis; 3) clinical VCD (cVCD) was clinically diagnosed by the investigator as dengue and accompanied by laboratory confirmation of dengue infection; 4) undifferentiated fever VCD (UF-VCD) was laboratory confirmation of dengue infection but was not diagnosed as dengue by the investigator. Detailed case report forms were completed for each febrile episode, including whether subjects required hospitalization. This manuscript describes results of a secondary analysis of anonymous data from this vaccine clinical trial.
Sub-national passive dengue surveillance data from districts, provinces, or cities (hereafter referred to as “geographical units”) encompassing each clinical trial center were retrieved from official government surveillance websites for Thailand [25] and Jakarta, Indonesia, [26] or from personal communications with trial investigators, or sub-national health authorities in Malaysia [27], Indonesia, the Philippines, and Vietnam [28]. Dengue cases of any severity were pooled for the period of time during which CYD14 was active in that country.
Age-specific population data for sites were retrieved from census or other official records for each geographical unit. [29–33] Populations at the mid-point of the study were estimated by applying national-level population growth factors. Where surveillance or census data lacked age-stratifications (Vietnam and Indonesia for census; Malaysia and Thailand for monthly age-specific surveillance data), we assumed age-distributions of populations/cases were proportional to those at the national level. Average annual IRs were calculated for each country by pooling data from all geographical areas participating in the study, expressed as cases/100,000 population/year, as:
Averageannualincidencerate=NcTmPopulationji¯*12
Where Nc is the number of cases reported to the surveillance system over the study observation period; Tm is the duration of the study in each country, in months; and
Populationji¯
is the average population size over the study period.
Site- and age-weighted incidence densities (IDs) were calculated by direct standardization for each country to correct for the fact that the age and geographic distributions of study populations were different from those in the geographic units from which they were drawn. [34] Study populations were divided into three age groups according to their age when they contributed time to the study: < 5 years; 5 −< 10 years; and > 10 years (all were aged <15 years at enrollment). Crude age-specific IRs were calculated for each age group and each center by dividing the number of cases satisfying each case definition by the number of person-years (p/y) of observation. These crude IRs were to match the demographics from CYD14 with those of each geographical area, resulting in age- and site-adjusted IDs aligned with the census populations at the country level. [34] Results were presented as cases/100,000 p/y. Standardized 95% confidence intervals (CIs) were calculated based on the gamma distribution using SAS 9.4 (SAS Institute, Cary, NC).[35]
Expansion factors were calculated by dividing the adjusted ID captured during CYD14 for each case definition by the IRs reported by the national passive surveillance systems at each geographical unit. For calculating 95% CIs, IRs from surveillance systems were considered known data without variability. [34]
Descriptive exploratory statistical analysis was performed on the ability of each case definition to identify symptomatic dengue cases; the proportion hospitalized; and the duration of clinical symptoms, fever, and hospitalization, for each. Using VCD as the gold standard diagnosis of dengue following a febrile episode, positive predictive value, negative predictive value, sensitivity, and specificity were calculated for clinical diagnosis of dengue disease with their 95% CIs according to the efficient-score method [36]. Data used for the analyses described above are provided in S1 and S2 Tables.
Between June and December, 2011, 3,424 children were enrolled in the control arm of the CYD14 study (Table 1). The demographics of the subjects have been reported elsewhere. [23] The period of observation was 6,933 p/y, during which there were 3,099 febrile episodes tested for dengue, of which 319 (10.3%) were VCD. This proportion in each country varied between 6.3% (Malaysia) and 12.3% (Indonesia). The overall crude annual VCD attack rate was 4.6%, varying from 2.2% (Malaysia) to 6.6% (Philippines). A total of 108 cases satisfied the CDD definition and 227 satisfied the UF-VCD definition (underlying data in S1 Table).
Of the 319 VCD cases, 25 (7.8%) were clinically diagnosed as DHF and 67 (21.0%) as DF, giving a total of 92 (28.8%) cases of cVCD. This proportion of VCD correctly diagnosed varied widely among the countries, from 10.3% (Philippines) to 74.5% (Thailand). In addition to the 25 clinical diagnoses of DHF in the VCD group, there were an additional 4 DHF diagnoses which were not virologically confirmed. Only one DHF diagnosis was in a subject aged <5.
There was significant heterogeneity in average IRs observed between countries over the duration of the clinical trial. The IRs for all dengue cases (per 100,000 p/y) reported to national routine surveillance systems for the study locations during the study period varied by country from: 64.7 (Malaysia); 263 (Indonesia); 497 (Thailand); 509 (Vietnam); and 954 (Philippines) (Table 2). Adjusted dengue IDs (per 100,000 p/y) in the CYD14 study were considerably higher than the rates captured by the national systems and varied according to the case definition used. IDs were highest for VCD (range: 2,048 [Malaysia] to 10,960 [Philippines]), and were followed by the IDs for UF-VCD (range: 1,192 [Indonesia] to 10,290 [Philippines]), CDD (range: 701 [Philippines] to 4,383 [Thailand]), and finally cVCD (range: 261 [Vietnam] to 4,262 [Thailand]). Surveillance data and corresponding incidence rates provided in S2 Table.
EFs varied according to case definitions. They were 5.5−31.7 for VCD, 0.5−10.4 for cVCD, and 0.7−12.0 for CDD (Table 3). These factors varied widely but tended to be lowest in Vietnam and highest in Malaysia. The incidence rates of dengue reported to the routine surveillance system appeared to be important determinants of EF: the highest EF (31.7) was observed in Malaysia (with the lowest reported IR) and the Philippines (with low EFs) reported the highest IRs in passive surveillance.
Overall, 126 (4.1%) of the acute febrile episodes in the cohort were hospitalized (Table 4). For the individual countries, this proportion varied between 1.2% (Vietnam) and 8.8% (Thailand). Hospitalization rates varied according to local standard of care (i.e., laboratory and clinical diagnosis): 61 (19.1%) of the 319 VCD episodes; 62 (57.4%) of the 108 CDD cases; and 24 (96.0%) of the 25 VCD cases diagnosed as DHF, were hospitalized. These proportions varied between countries. Clinical dengue diagnosis appeared to be an important determinant of hospitalization. The median duration of clinical symptoms was 5.0 days [min; max: 2.0; 38.0] for cases of UF-VCD; 6.0 [2·0; 38·0] for VCD and 8.0 [2.0; 31.0] for CDD. Median durations of fever were 3.0 [2.0; 11.0], 3·0 [2·0; 11·0] and 4.0 [2.0; 9.0)] and hospitalization 4.0 [3.0; 6.0]; 5·0 [2·0; 9·0] and 5.0 [2.0; 9.0)], respectively.
The positive predictive value (PPV) of clinical diagnosis, using VCD as the gold standard, was 85.2% (95% CI 77.1−91.3) and the negative predictive value (NPV) was 92.4% (91.4−93.3). The sensitivity of clinical diagnosis was 28.8% (95% CI 23.9−34.2) and the specificity was 99.4% (99.1−99.7).
We used the control arm of a large, phase 3 efficacy dengue vaccine trial to describe the symptomatic and virologically-confirmed dengue burden in five Asian countries. This permitted comparison of clinical vs. laboratory dengue diagnosis for different classifications of symptomatic dengue identified through active surveillance. Importantly, data were consistently collected according to standardized case definitions and with high-quality virological confirmation, allowing IDs to be measured for different clinical outcomes and in different countries, within a single study. Rates observed in study participants were adjusted to match the populations from which they were sampled.
Discrepancies between dengue clinical and laboratory diagnosis typically find case definitions which are sensitive but lack specificity, particularly in episodes of mild disease. [37–39] Our results showed that virological confirmation was the most sensitive means of identifying dengue disease, capturing approximately 3.5 times more episodes than clinical diagnosis alone, even in this acutely febrile patient population. Clinical diagnosis alone captured only 28.8% of symptomatic cases. Because most passive disease surveillance systems in Asia rely almost entirely on clinical diagnosis, [10,40] it is reasonable to believe a substantial proportion of symptomatic dengue disease is unrecognized and therefore unreported.
The proportion of VCD clinically diagnosed by investigators as dengue varied substantially between countries (range: 10.3%−74.5%), likely resulting from the multifactorial impacts of local clinical guidelines and case definitions that affect diagnostic practices, and variable clinical presentations. Disease severity and clinical manifestation may be affected by factors including circulating viral genotypes; the order and duration between sequential, heterotypic infections; year/season; and subject age [41–43]. Notably, VCD cases appeared to be younger in the Philippines–where dengue was least-frequently diagnosed–than other countries.
Estimates of dengue burden are, in large part, a function of case definition. [9] The active surveillance methods here allowed calculation of IRs according to different case definitions (VCD, cVCD, and CDD) and thus determine EFs for each. The higher rates of VCD captured gave rise to EFs ranging from 5.5 to 31.7, with lower EFs for more specific case definitions of cVCD and CDD. These figures are notable for their variability and emphasize that study and surveillance system methodology and geography are important determinants of under-reporting estimates, as reported elsewhere. [15,19] However, the finding that dengue is under-reported by factors of >30 in some countries and contexts is consistent between these studies. Notable exceptions are the expansion factors <1, observed in Philippines and Vietnam against specific, clinically-diagnosed case definitions (cVCD and CDD): passive surveillance had captured a higher proportion of cases than the active system. There are two likely possibilities: 1) the passive surveillance reported false-positive cases (ie, episodes of febrile, non-dengue disease, thereby increasing the denominator) or, more likely, 2) the active system excluded febrile cases which failed to satisfy case definitions (thereby reducing the numerator). Both scenarios emphasize the heterogeneity of dengue case definitions, surveillance systems and clinical practices, which challenge the generation of consistent burden estimates. Additional complexity has been observed during outbreaks from both over- and under-reporting due to differing levels of disease awareness and/or reporting practices. [44]
A similar analysis was conducted in slightly older Latin American children, focusing on VCD cases and comparing with dengue reported at different levels of the surveillance system (country; state; local). [45] It found lower rates of VCD (from 2,500 cases/100,000 p/y in Mexico to 3,500 in Brazil), and corresponding EFs which varied widely, from 3.5–45.5 (depending on country and comparator), emphasizing that EFs are a complex outcome of local epidemiology, disease awareness, health system characteristics and other factors. [19] Additional analyses of under-reporting according to indicators of socio-demography or dengue awareness, for example, may be illuminating.
Hospitalization was based on local routine practice and rates were shown to be substantial: over 4% of fevers, and over 50% of dengue diagnoses were hospitalized. A clinical diagnosis–rather than virological confirmation–seemed to determine the decision to hospitalize, demonstrated by the successively decreasing incidence and increasing hospitalization rate of episodes of fever; VCD; CDD; and DHF. Interestingly, four cases of clinically diagnosed DHF (13.8% of the total) could not be virologically confirmed, highlighting a possible over-attribution in endemic areas, of consequence for prospective epidemiological or vaccine effectiveness studies using clinical endpoints. Interpretations of the interplay between severity, case definitions and hospitalization are particularly important from a health economics perspective when we consider that a single hospitalization has been reported to cost between USD 289 (Philippines) and USD 863 (Malaysia). [21,46]
The case definition applied in CYD14 (fever ≥ 38°C on ≥ 2 consecutive days) intentionally captured a broad spectrum of disease, enabling calculation of vaccine efficacy against dengue of any severity. However, a considerable proportion of VCD episodes (7.8%) were assessed as DHF by the investigators, and while empirically-derived global burden estimates of severe dengue/DHF are not available, [47] our data suggest that in Southeast Asia, the burden is substantial. Extrapolations using appropriate baselines and harmonized case definitions associated with clinical severity could theoretically be used to generate estimates of severe disease, and may be a topic of further research.
Most clinical diagnoses of dengue were virologically confirmed, resulting in a PPV of 85.2%. However the sensitivity was 28.8%, reflecting the significant proportion of VCD which was not clinically diagnosed as dengue. This is likely because local DF or DHF reporting case definitions had not been satisfied, even when investigators suspected dengue as the underlying aetiology, and is a finding which should contribute to the understanding of the clinical and economic burden of mild dengue disease. Using VCD as a denominator, the complement to clinical diagnoses were termed undifferentiated fever VCD in our study, and represent symptomatic, febrile, virologically confirmed cases which were not diagnosed. Policymakers sometimes consider milder manifestations of disease unimportant, but a recent Cambodian study found mild dengue cases are significantly more infectious than those with symptoms. [48] Mild cases may thus contribute significantly to transmission and constitute an important viral reservoir. Additional studies will be required to understand the impacts on population-level immunity and transmission dynamics.
A clinical diagnostic exclusion of dengue following a febrile episode was correct in >90% of instances (NPV: 92.4%) and the specificity of clinical diagnosis was 99.4% in these epidemiological settings where >10% of acute fevers were caused by dengue virus infection. The accuracy of diagnosis was much improved when considering only hospitalized episodes, indicating that surveillance reports and burden estimates of more severe disease are likely more reliable than those of mild cases. However this leaves a considerable burden of mild disease which is unaccounted for. We are not aware of health economic or healthcare utilization studies examining the impact of these mild episodes but their frequency implies a significant source of burden. Additional analyses could consider aggregating costs (including indirect costs), disability-adjusted life years or other measures to quantify impacts.
The study has limitations. The ID of cVCD was low in some settings, with only six episodes in Malaysia and nine in Vietnam. This is an unavoidable consequence of examining infrequent disease outcomes using prospective methods. Our approach annualized incidence, which may have introduced some bias, but our comparison with local surveillance data and overlapping timeframes will limit geographical/temporal distortions. As this was a vaccine trial, sites were chosen for their historically high reported dengue burdens, so results from lower-endemic areas may differ. [38] For this reason, incidence rates and other findings could not be combined between countries. However, the socio-environmental determinants of dengue incidence are poorly understood and in many Asian countries burdens are unpredictable throughout urban endemic areas. Where age-stratified incidence data were unavailable, adjustments were made which introduced slight inaccuracies to the data. More substantial variability was caused by the differences in national surveillance systems, with more sensitive surveillance giving rise to lower EFs. This is an inherent study bias but also an interesting result; the use of a stable denominator in expansion factor calculations provides an insight into surveillance system specificities.
WHO 1997 classifications were applied, as assessed by investigators, because at study initiation 2009 guidelines were not in routine use at all sites. Cases were also classified according to a more inclusive definition of severe dengue, integrating criteria from the WHO 1997 and 2009 and South East Asia Regional Office 2011 guidelines, and applied by the study Independent Data Monitoring Committee (IDMC). [49] Of the 25 VCD cases clinically diagnosed as DHF, 20 met these IDMC criteria indicating, in this clinical trial environment at least, a level of concordance between the two.
This analysis was performed to inform policy making and strengthen evidence for public health decisions, including financing for dengue control efforts such as vaccination. It adds to available evidence indicating that passive surveillance systems greatly underestimate dengue burden and emphasizes that burden estimates are highly sensitive to case definitions. The control arms of vaccine clinical trials can provide valuable data to estimate disease burdens.
|
10.1371/journal.pntd.0000311 | Dermal-Type Macrophages Expressing CD209/DC-SIGN Show Inherent Resistance to Dengue Virus Growth | An important question in dengue pathogenesis is the identity of immune cells involved in the control of dengue virus infection at the site of the mosquito bite. There is evidence that infection of immature myeloid dendritic cells plays a crucial role in dengue pathogenesis and that the interaction of the viral envelope E glycoprotein with CD209/DC-SIGN is a key element for their productive infection. Dermal macrophages express CD209, yet little is known about their role in dengue virus infection.
Here, we showed that dermal macrophages bound recombinant envelope E glycoprotein fused to green fluorescent protein. Because dermal macrophages stain for IL-10 in situ, we generated dermal-type macrophages from monocytes in the presence of IL-10 to study their infection by dengue virus. The macrophages were able to internalize the virus, but progeny virus production was undetectable in the infected cells. In addition, no IFN-α was produced in response to the virus. The inability of dengue virus to grow in the macrophages was attributable to accumulation of internalized virus particles into poorly-acidified phagosomes.
Aborting infection by viral sequestration in early phagosomes would present a novel means to curb infection of enveloped virus and may constitute a prime defense system to prevent dengue virus spread shortly after the bite of the infected mosquito.
| Mosquito-transmitted pathogens are a major challenge to humans due to ever-increasing distribution of the vector worldwide. Dengue virus causes morbidity and mortality, and no anti-viral treatment or vaccine are currently available. The virus is injected into the skin when an infected mosquito probes for blood. Among the skin immunocytes, dendritic cells and macrophages are equipped with pathogen-sensing receptors. Our work has shown that dermal macrophages bind the dengue virus envelope protein. We demonstrate that monocyte-derived dermal macrophages are resistant to infection and present evidence that this is due to sequestration of the virus into fusion-incompetent intracellular vesicles. This identifies skin macrophages as the first innate immune cell potentially capable of protecting the human host from infection by dengue virus shortly after a mosquito bite. These findings have important implications for better understanding the early infection events of dengue virus and of other skin-penetrating pathogens.
| Dengue is probably the most important mosquito-transmitted viral disease of humans worldwide. It is caused by dengue virus (DV), which exists as four serotypes (DV1-4) and circulates in an endemic-epidemic mode in most tropical and sub-tropical territories. Transmission of DV to humans occurs when an infected mosquito probes for blood vessels and during a blood meal, through injection of infectious saliva into the human dermis. As a member of the Flaviviridae family, DV infection involves virus uptake into endosomal vesicles that undergo acidification. The low pH induces structural alterations in the envelope (E) protein that lead to membrane fusion and the release of the nucleocapsid into the cytoplasm [1]. After uncoating, the RNA genome is translated to initiate virus replication. It has been proposed that non-neutralizing antibodies raised against one DV serotype may enhance infection by a heterotypic serotype [2]. This may explain why secondary infections are often associated with the more severe forms of dengue fever (hemorrhagic fever with or without shock).
Much research on DV relies on relevant human cell culture models due to the difficulty of establishing appropriate animal models. Progress has been made by showing that DV E protein recognizes the C-type lectin CD209 and its homologue L-SIGN and that expression of either of these lectins is sufficient to render cells permissive to DV grown in mosquito cells [3],[4]. Recently, the mannose receptor (MR) has also been shown to mediate DV binding and infection [5]. Dendritic cells (DC), generated from monocytes in the presence of GM-CSF and IL-4, express CD209, L-SIGN and the MR and are highly susceptible to DV infection [3],[4],[6]. These monocyte-derived DC are thought to be representative of dermal DC (dDC), yet there is increasing evidence that CD209 is not expressed by dDC but primarily by dermal macrophages (dMφ) [7]–[9]. This underscores the importance of dMφ in early infection events and raises the question of whether dMφ are permissive for productive DV infection. Studies of these cells have been hampered by the lack of suitable isolation techniques from human skin and culture methods to generate the cells from monocytic precursors.
Here, we confirmed that human dMφ express CD209 and showed that they bind DV E protein. Based on the finding that dMφ stained for intracellular IL-10, we developed a method to generate the cells from monocytes in the presence of IL-10. The monocyte-derived dMφ bound E protein and acquired DV in intracellular vesicles, but were resistant to viral replication. The inability of DV to grow in these dermal-type Mφ was attributable to accumulation of internalized virus particles into poorly-acidified phagosomes. These findings advance our understanding of the host innate resistance to DV at the early stages of infection and have implications for other pathogens recognizing CD209.
Before blood and tissue samples were collected for the study, all healthy donors and patients gave written informed consent in agreement with the Helsinki Declaration and French legislation. A prospective IRB approval was not obtained since there was no need as specified by French law of the health protection act when employing healthy material destined for disposal or one-time biomedical research. A retrospective IRB approval was given. Fresh skin (about 50 cm2) was obtained from patients undergoing breast reduction surgery or abdominoplasty. The skin was trypsinized to peel off the epidermis and the remaining dermis was processed as described elsewhere [10] with the modification that only collagenase type I (1 mg/ml, Invitrogen) was used for 18 h at 37°C. The resulting cell suspension was pipetted and serially filtered through 100 µm and 70 µm cell strainers (BD Biosciences) to remove undigested tissue fragments and to obtain a homogeneous cell suspension.
A DNA fragment containing the DV3 genomic region (Swiss-Prot accession number P27915) coding for the prM-E protein (1674 nt in total, including all of prM and the E ectodomain, ending at codon 392 of E, at the end of domain III) was amplified by PCR with forward primer 5′TTATGCATATTACTGGCCGTCGTGGCC and reverse primer 5′CTCGCCCGCAGACATGGCCTTATCGTCATCGTCGGGCCCCTTCCTGTACCA-GTTGATTTT and inserted into the plasmid pT352. This is a shuttle vector containing selection markers for yeast and E. coli, as well as a metallotheionein-inducible expression cassette for Drosophila cells. In the construct, called pT352/DV3 sE-GFP, the DV prM-E sequence is in-frame with the Drosophila BiP signal peptide, which directs the recombinant protein to the secretory pathway. Drosophila S2 cells (Invitrogen) were co-transfected with pT352/DV3 sE-GFP and a vector conferring resistance to blasticidine, using the effectene transfection reagent (Qiagen). The selected cells were adapted to serum-free growth medium and grown to high density before induction with CuSO4 at 500 µM. The supernatant was collected 10 days later and concentrated using a flow concentration system with a 10 KDa-cutoff membrane (Vivascience), and DV3 sE-GFP was purified by affinity chromatography using a Steptactin column. The eluate was concentrated and further purified by size-exclusion chromatography, using a Superdex 200 10/300 column (GE Healthcare) with 0.5 M NaCl and 50 mM Tris (pH 8.0). Purified DV3 sE was concentrated to 10 g/liter in Vivaspin ultrafiltration spin columns (Sartorius).
Dermal cells were collected 48 h after culturing in complete medium, RPMI medium supplemented 10% fetal calf serum (FCS) and antibiotics (Invitrogen), and 3×105 cells were incubated with 1, 2, 4 or 8 µg recombinant DV3 sE-eGFP fusion protein in 0.1 ml complete medium at 37°C for 30 min. The cells were then washed twice with complete medium and incubated with anti-CD14-APC, anti-CD1a-PE and anti-HLA-DR-PerCP mAb (BD Biosciences) in PBS/2% FCS for 15 min. Following 3 washes, the cells were fixed in 0.4% formaldehyde and analyzed by flow cytometry (FACS Calibur, BD Biosciences). The relative MFI for 3 donors was determined in triplicate after gating for CD1a+HLA-DR+ or CD14+HLA-DR+ cells using the following formula: (MFI (FL1) protein sE-eGFP – MFI (FL1) no protein sE-eGFP)/MFI (FL1) no protein sE-eGFP. To determine CD209 expression, 3×105 cells were incubated with anti-CD209-PerCPCy5.5 (clone DCN46, BD Biosciences), anti-CD14-APC, anti-CD1a-PE and anti-HLA-DR-PerCP mAb in PBS/2% FCS for 15 min and, after washing, fixed and analyzed by flow cytometry.
Formaldehyde-fixed, paraffin sections were rehydrated and antigen was retrieved in citrate buffer pH 6 at 97°C for 45 min. Biotin was blocked using the avidin-biotin blocking kit (Vector Inc.), and sections were saturated in 5% human serum at room temperature for 40 min. The following primary Abs were used: goat-anti IL-10 (1∶75 dilution, R&D Systems), mouse anti-CD209 (2 µg/ml, R&D Systems), mouse anti-CD1a (Immunotech) and mouse anti-CD14 (1∶40 dilution, Novocastra). The secondary Ab (Jackson) were: biotin-conjugated donkey anti-goat followed by streptavidin-Alexa 488 (Molecular Probes-Invitrogen) and F(ab)'2 rabbit anti-mouse followed by Cy3-conjugated donkey anti-rabbit. Sections were observed by confocal microscopy (LSM510 Zeiss).
Monocytes were isolated from 200 ml of adult human peripheral blood using negative-depletion beads (Dynal-Invitrogen) or by counterflow centrifugal elutriation. To obtain MDdMφ, 3×106 monocytes were cultured for 5 days in 5 ml of complete medium containing 10 ng/ml M-CSF (R&D Systems), 20 ng/ml IL-10 (Immunotools) and 20 ng/ml GM-CSF (Schering-Plough) with refreshment of GM-CSF (10 ng/ml) and IL-10 (10 ng/ml) at day 3. For MDDC, 3×106 monocytes were cultured for 5 days in 5 ml of complete medium containing 50 ng/ml GM-CSF and 10 ng/ml IL-4 (Schering-Plough) with readdition of cytokines at day 3. Non-adherent cells were harvested. Expression of markers was measured by FACS using specific antibodies and their corresponding isotype controls. To assay for DV3 sE protein binding, cells were pre-incubated for 10 min in complete medium in the absence or presence of 5 mM EDTA before adding 3 µg DV3 sE-eGFP protein. After 30 min at 37°C, the cells were washed three times in complete medium and analyzed by flow cytometry.
5×105 MDdMφ and MDDC were exposed to DV serotype 1 (strain FGA/NA d1d) [11], serotype 2 (strain 16681), or serotype 3 (strain PaH 881, isolated in 1988 in Thailand) in RPMI medium supplemented with 0.2% bovine serum albumin for 2 h. Viral growth was determined at 40 h post-infection. Virus titration was performed as previously described [3]. Infectivity titers were expressed as focus forming unit (FFU) on mosquito AP61 cell line (DV1 and DV3) or plaque forming unit (PFU) on mammalian BHK cell line (DV2). Different titering assays were performed to independently confirm our findings, despite the fact both methods may not be equivalent. The limit of titer determination was fixed at 103, below which viral production was considered non-significant. For FACS analysis, infected cells were fixed and labeled for intracellular viral antigens with antiserum raised in mice that had received intracerebral DV injection [3]. IFN-α released from DV1-infected MDdMφ and MDDC was measured by ELISA (R&D Systems).
To observe live DV internalization by MDDC and MDdMφ, the cells were exposed to DV1 at an MOI of 100 at 4°C for 30 min or at 37°C for 1 h and fixed in 2.5% glutaraldehyde. Cells were postfixed in osmium tetroxide, dehydrated in ethanol containing 1% uranyl acetate, treated with propylene oxide and embedded in resin (Durcupan ACM, Fluka). Ultrathin sections were stained with lead citrate and examined by transmission electron microscopy (TEM) (Hitachi H600). Images were acquired using a CCD camera (Hamamatsu). To visualize DV3 sE-eGFP internalization and endosomal acidification, cells were incubated with 10 µM LysoSensor Blue DND-167 (Molecular Probes-Invitrogen) for 30 min at 37°C. Protein sE-eGFP was added at a concentration of 3 µg/ml, and cells were viewed after different incubation times using a confocal microscope (LSM510, Zeiss). The blue color emitted by the LysoSensor dye was digitally converted into red. For TEM, cells were fixed in 2% paraformaldehyde and 0.2% glutaraldehyde. Cells were embedded in 1% agarose, permeabilized with 0.2% saponin and saturated with 2% BSA before incubation with 5 µg/ml polyclonal rabbit anti-GFP antibody (Rockland). The antibody was visualized by pre-embedding labeling using a goat anti-rabbit IgG conjugated to 0.8 nm gold particles, according to manufacturer's instructions (Aurion). Cells were fixed in 1% glutaraldehyde, and gold particles were enhanced using a silver kit (HQ silver, Nanoprobes). Cells were then treated and observed as above.
We wished to determine whether human dMφ are targets of DV infection. To this end, healthy human skin from patients undergoing plastic surgery was processed to obtain a dermal cell suspension. The cells were then cultured without additional cytokines for 48 h to allow re-expression of cell surface markers, such as CD1a and CD209, lost during the collagenase treatment (data not shown). Binding of DV3 E protein to dermal cells was assessed by flow cytometry after staining with CD14 and CD1a-specific antibodies. CD14 is expressed by dMφ and CD1a by dDC [7]–[9]. To detect E protein binding, the soluble form of DV3 E protein (sE) was fused to the reporter protein eGFP and purified from a Drosophila expression system. As shown in Figure 1A, CD1a+ dDC showed only a limited capacity to interact with DV3 sE protein, whereas CD14+ dMφ readily bound the protein. This is corroborated by the distinct expression of CD209 by dMφ (Fig. 1A), whereas dDC expressed little, if any, CD209 (data not shown). Increasing amounts of DV3 sE protein were added to the dermal cell suspension to test if dDC bound the protein at higher concentrations. Figure 1B shows that even at high concentrations, there was little binding of DV3 sE protein to dDC, whereas it bound to dMφ in a dose-dependent fashion. These findings identify dMφ as potential key cellular targets of DV.
To address the question of whether dMφ are infected by DV and whether they are permissive for viral production, we established cell culture conditions to generate dermal-type Mφ from monocytes. We observed on human skin tissue sections that dMφ expressing CD14 or CD209, but not the CD1a+ dDC, stained for IL-10 (Fig. 2A). When purified human monocytes were cultured in M-CSF and increasing concentrations of IL-10, the cells expressed CD14 and CD209 in an IL-10 dose-dependent manner (Fig. 2B). Similar to DC [12], the addition of GM-CSF increased CD209 levels (Fig. 2B), so that a homogeneous CD14+CD209+ cell population could be obtained with CD209 expression nearly identical to that of DC derived from monocytes in the presence of GM-CSF and IL-4 (Figure S1A). Western blotting of cell lysates confirmed the presence of CD209 as a major band of 49 kDa in both cell-types [13] (Figure S1B). The Mφ expressed coagulation factor XIIIa and CD163, two other cell surface markers of dMφ [14] (Fig. 2C). The Mφ and the DC were both able to bind eGFP-tagged DV3 sE protein, which was inhibited by EDTA (Fig. 2C). This distinguishes the monocyte-derived DC from dDC. Upon activation by lipopolysaccharide (LPS), the Mφ rapidly released IL-10, whereas DC or monocytes produced little of this cytokine (Figure S1C).
Monocyte-derived dMφ (MDdMφ) and monocyte-derived DC (MDDC) were analyzed for DV infection using low-passage DV1 and DV3 strains grown in mosquito cells [3] as well as the prototype DV2 strain 16681 [15]. The cells were exposed to DV1 at a multiplicity of infection (MOI) of 1 for 2 h, washed, and then cultured for 40 h. As shown in Figure 3A, intracellular viral antigen was clearly detected in MDDC by flow cytometry, whereas no specific immuno-labeling was observed in MDdMφ. An analysis of DV replication in these cells infected at an MOI of 1 (DV1 and DV3) or 2 (DV2) showed that MDDC were highly permissive to productive infection (∼105 FFU/ml or PFU/ml) (Fig. 3B); in contrast, progeny virus production was undetectable in DV-infected MDdMφ (<103 FFU/ml or PFU/ml). Consistent with this finding, no IFN-α was produced by DV-infected MDdMφ, even at an MOI of 10, whereas MDDC readily released IFN-α when infected with DV at an MOI of 1 or 10 [16] (Fig. 3C). To verify that MDdMφ acquired the virus, both myeloid cell-types were exposed to high DV input (MOI of 100) and electron microscopy analysis was performed after 30 min at 4°C and after 1 h at 37°C (Fig. 3D). Cell surface-bound (at 4°C) and endosomal vesicle-associated virus particles (at 37°C) were clearly detected in both cell-types. Thus, internalization of DV can occur in MDdMφ but does not result in productive infection.
In an effort to define the molecular basis of the inability of DV to grow in MDdMφ. we asked whether internalized DV was sequestered in a manner that hampers productive infection, using DV3 sE-eGFP fusion protein. To monitor DV3 sE protein internalization in MDdMφ and MDDC, the cells were incubated with pH-sensitive LysoSensor dye and analyzed by confocal microscopy (Fig. 4). This dye accumulates in acidic organelles, where its fluorescence emission is highest. After 5 min at 37°C, DV3 sE protein was observed in vesicle-like structures in both cell-types. By 30 min and 60 min, DV3 sE protein dispersed to acidified perinuclear lysosomes in MDDC. In marked contrast, when MDdMφ were examined at these time-points, a large fraction of internalized DV3 sE protein was excluded from the acidic compartment and remained in non-acidic, large endosomes. Electron microscopy analysis using a colloidal gold-conjugated antibody to GFP demonstrated that DV3 sE protein accumulated in large phagosomes in MDdMφ, located close to the plasma membrane (Fig. 5). On the other hand, at 30 min, in MDDC, DV3 sE protein was mostly found in small perinuclear vesicles in the environment of the endoplasmic reticulum. Taken together, these data suggest that the inability of DV to productively infect MDdMφ is due to accumulation of virus particles in immature endosomal vesicles whose pH does not allow efficient viral-cell membrane fusion and subsequent virus uncoating.
In the present study, we demonstrated for the first time the interaction of dMφ with DV3 sE glycoprotein, which correlates with the expression of the DV attachment receptor CD209. Dermal DC displayed only a limited capacity to interact with DV3 sE protein and expressed little CD209. In accordance with these findings, in situ immuno-labeling of human skin section revealed CD209 expression by dMφ but little on DC [7]–[9]. Both cell types carry the MR [7], which also recognizes DV E protein [5]. Due to the nature of our binding assay, the dermal cells with the highest affinity for DV3 sE protein would acquire the most DV3 sE protein, suggesting that dDC may capture the recombinant envelope protein when physically isolated from dMφ. In the skin, the abundance, the location and the co-expression of CD209, L-SIGN and MR are likely to determine the nature of the DV-capturing immune cell.
Based on the observations that dMφ stained for intracellular IL-10 in situ and that IL-10 is produced by dMφ ex vivo [17],[18], we tested the effect of IL-10 on the formation of dMφ from monocytes. By combining IL-10, M-CSF and GM-CSF, a homogenous cell population was obtained which carried CD209 and other markers characteristic of dMφ, rapidly produced IL-10 in response to LPS or other toll-like receptor ligands (data not shown), and bound DV3 sE protein. Like MDDC, the MDdMφ were capable of internalizing live DV but, distinct from MDDC, they displayed an inherent resistance to viral growth. In contrast to DV3 sE protein found in acidified compartments in MDDC, we observed that DV3 sE protein accumulated in non-acidified phagosomes in MDdMφ. The DC vesicles containing DV3 sE protein or live virus were bell-shaped or tubular, whereas they were round, larger and close to the plasma membrane in the Mφ. To our knowledge, this identifies MDdMφ as the first innate immune cell capable of protecting the human host from DV infection and virus propagation. From this data, we propose that dMφ can act to trap infecting virions in a fusion-incompetent endosomal environment and thus to prevent DV spread to dDC at the anatomical site of the mosquito bite. We cannot formally exclude the possibility that downstream delays in the viral life cycle contribute to the inability of DV to replicate in MDdMφ, but the finding that West Nile virus productively infects these cells (data not shown) indicates that they are not generally refractory to flavivirus growth.
IL-10, required for CD209 expression and blockage of endosome acidification, is likely to be produced by the dMφ themselves, constitutively, or in response to stimuli such as UV-light [17]. In this context, a key question is whether mosquito salivary proteins, co-injected with the infectious virus, would also trigger IL-10 production by dMφ or, on the contrary, provoke an inflammatory response. Inflammatory cytokines of the Th2 T-helper cell type, IL-4 and IL-13, may be responsible for the formation of CD209+MR+ DC, which are permissive for DV infection and viral progeny production [3]–[6]. Alternatively, the presence of anti-DV non-neutralizing antibodies raised against a heterotypic DV serotype may render dDC susceptible to DV infection at the site of the mosquito bite.
The abundance and strategic position of the Mφ in the dermis is consistent with their function as first defense barrier against pathogens by isolating and eliminating them and thus avoiding unnecessary immune activation. However, other pathogens that recognize C-type lectins, such as mycobacteria, may exploit these cells to escape immune attack. Accumulating CD209+ Mφ in leprosy skin lesions have been associated with mycobacterial persistence [19]. Important questions to address in future are whether DV is eliminated in MDdMφ, whether infected MDdMφ gradually release DV, as shown for the foot-and-mouth disease virus and pulmonary Mφ [20], and whether rapid DV growth can occur when the Mα convert to DC. Improved knowledge of the molecular mechanisms for suppressing pathogen growth in MDdMφ will provide new insight into the crucial role of dMφ in protective immunity to infectious agents at the skin level.
|
10.1371/journal.pgen.1000650 | Multiple Organ System Defects and Transcriptional Dysregulation in the Nipbl+/− Mouse, a Model of Cornelia de Lange Syndrome | Cornelia de Lange Syndrome (CdLS) is a multi-organ system birth defects disorder linked, in at least half of cases, to heterozygous mutations in the NIPBL gene. In animals and fungi, orthologs of NIPBL regulate cohesin, a complex of proteins that is essential for chromosome cohesion and is also implicated in DNA repair and transcriptional regulation. Mice heterozygous for a gene-trap mutation in Nipbl were produced and exhibited defects characteristic of CdLS, including small size, craniofacial anomalies, microbrachycephaly, heart defects, hearing abnormalities, delayed bone maturation, reduced body fat, behavioral disturbances, and high mortality (75–80%) during the first weeks of life. These phenotypes arose despite a decrease in Nipbl transcript levels of only ∼30%, implying extreme sensitivity of development to small changes in Nipbl activity. Gene expression profiling demonstrated that Nipbl deficiency leads to modest but significant transcriptional dysregulation of many genes. Expression changes at the protocadherin beta (Pcdhb) locus, as well as at other loci, support the view that NIPBL influences long-range chromosomal regulatory interactions. In addition, evidence is presented that reduced expression of genes involved in adipogenic differentiation may underlie the low amounts of body fat observed both in Nipbl+/− mice and in individuals with CdLS.
| Cornelia de Lange Syndrome (CdLS) is a genetic disease marked by growth retardation, cognitive and neurological problems, and structural defects in many organ systems. The majority of CdLS cases are due to mutation of one copy of the Nipped B-like (NIPBL) gene, the product of which regulates a complex of chromosomal proteins called cohesin. How reduction of NIPBL function gives rise to pervasive developmental defects in CdLS is not understood, so a model of CdLS was developed by generating mice that carry one null allele of Nipbl. Developmental defects in these mice show remarkable similarity to those observed in individuals with CdLS, including small stature, craniofacial abnormalities, reduced body fat, behavioral disturbances, and high perinatal mortality. Molecular analysis of tissues and cells from Nipbl mutant mice provide the first evidence that the major role of Nipbl in the etiology of CdLS is to exert modest, but significant, effects on the expression of diverse sets of genes, some of which are located in characteristic arrangements along the DNA. Among affected genes is a set involved in the development of adipocytes, the cells that make and accumulate body fat, potentially explaining reductions in body fat accumulation commonly observed in individuals with CdLS.
| Cornelia de Lange Syndrome (CdLS; OMIM#122470) is characterized by developmental abnormalities of the cardiopulmonary, gastrointestinal, skeletal, craniofacial, neurological, and genitourinary systems [1]–[3]. The clinical presentation ranges from subtle dysmorphology to conditions incompatible with postnatal life. Common structural birth defects observed in CdLS include upper limb reduction (significant in just under half of cases), cardiac abnormalities (especially atrial and ventricular septal defects), and craniofacial dysmorphia (including dental and middle ear abnormalities, occasional clefting of the palate, and highly characteristic facies) [2]–[8]. Other findings include small head size, lean body habitus, hirsutism, ophthalmologic abnormalities, pre- and postnatal growth retardation, and structural abnormalities of the gastrointestinal tract (duodenal atresia, annular pancreas, small bowel duplications) [2], [3], [9]–[11]. Physiological disturbances in CdLS include moderate to severe mental retardation [12] often accompanied by autistic behaviors [13], and severe gastrointestinal reflux [14]. Although prevalence has been estimated at between ∼1/10,000 and 1/50,000 births [8],[15], wide phenotypic variability in the syndrome makes it likely that large numbers of mildly-affected individuals are not being counted.
A genetic basis for CdLS was uncovered in 2004 with the demonstration that many affected individuals carry mutations in Nipped-B-like (NIPBL), so named for its homology to the Drosophila gene, Nipped-B [16],[17]. Heterozygous NIPBL mutations are found in about 50% of individuals with CdLS [18]. As many of these mutations are expected to produce absent or truncated protein, haploinsufficiency is the presumed genetic mechanism [19].
NIPBL/Nipped-B protein is found in the nuclei of all eukaryotic cells, where it interacts with cohesin, the protein complex that mediates sister chromatid cohesion [20],[21]. The NIPBL ortholog in fungi plays a role in loading cohesin onto chromosomes, and a role in unloading has been suggested as well. The fact that a minority of cases of mild CdLS result from mutations in the SMC1L1/SMC1A (∼5%; OMIM 300590) and SMC3 (1 case; OMIM 610579) genes, which encode two of the four cohesin structural components, supports the view that CdLS is caused by abnormal cohesin function [22],[23]. Consistent with the hypothesis that cohesin plays important roles during embryonic development, it was found that mutations in the cohesin regulatory protein ESCO2 cause Roberts'-SC phocomelia syndrome, another multi-organ systems birth defects syndrome [24],[25]. In mice, deletion of the cohesin regulators PDS5A and PDS5B also produces a wide variety of developmental defects, some of which overlap with CdLS [26],[27]. In addition, there has recently been a report of one family showing atypical inheritance of CdLS, in which both affected and unaffected siblings harbor a missense mutation in the PDS5B gene, raising the possibility of some genetic association between PDS5B and CdLS [27].
How alterations in cohesin function give rise to pervasive developmental abnormalities is largely unknown. Cohesin is involved in sister chromatid cohesion and DNA repair in many organisms, but observed alterations in cohesion and repair in individuals with CdLS are mild at best [28],[29]. More recently, observations in model organisms and cultured cells have suggested that cohesin plays important roles in the control of transcription [reviewed in 18]. In Drosophila, for example, changes in levels of Nipped-B or cohesin structural components alter the expression of developmental regulator genes, such as homeodomain transcription factors [30]–[33]. Such effects on gene expression, which have been proposed to reflect the disruption of long-range promoter-enhancer communication, occur with small changes in Nipped-B or cohesin levels that do not produce cohesion defects; they can also occur in postmitotic cells, in which chromosome segregation is presumably not an issue [18].
Studies using Drosophila cell lines have demonstrated that cohesin and Nipped-B binding are concentrated near the promoters of active transcriptional units [34]. In mammalian cells, cohesin often binds, in an NIPBL-dependent manner, to sites occupied by the transcriptional insulator protein CTCF, where it plays a significant role in CTCF function [35]–[37]. Recently, NIPBL has also been shown to bind and recruit histone deacetylases to chromatin [38]. These observations suggest that cohesin and NIPBL may interact in multiple ways with the transcriptional machinery.
As a first step toward understanding the molecular etiology of CdLS, we generated a mouse model of Nipbl haploinsufficiency, which replicates a remarkable number of the pathological features of CdLS. Cellular and molecular analysis of mutant cells and tissues revealed widespread, yet subtle, changes in the expression of genes, some of which are found in genomic locales in which transcription is known to be controlled through long-range chromosomal interactions. We propose that the aggregate effects of many small transcriptional changes are the cause of developmental abnormalities of CdLS, and present evidence that one set of transcriptional changes may explain the notably lean body habitus of many individuals with CdLS.
Two mouse ES cell lines bearing gene-trap insertions into Nipbl were obtained and injected into C57BL/6 blastocysts to produce chimeras (see Materials and Methods). Male chimeras were bred against both inbred (C57BL/6) and outbred (CD-1) mice. For only one cell line (RRS564, which contains a beta-geo insertion in intron1, and is predicted to produce a truncated transcript with no open reading frame; Figure S1) was ES cell contribution to the germline obtained (as scored by coat color; Table 1). Whereas Mendelian inheritance predicts that half the germline progeny of chimeric mice should be heterozygous (Nipbl+/−) for the gene trap insertion, the observed frequency was much lower. When chimeras were bred against CD-1 females, 22 out of 113 germline progeny (19%) carried the mutant allele (Table 1). With C57BL/6 females only one out of 18 germline progeny carried the mutation (5.5%), and this animal, although male, did not produce any progeny when subsequently mated.
In view of these data, it was decided that further analysis of the Nipbl− allele would take place through outcrossing onto the CD-1 background. As shown in Table 1, when the Nipbl+/− offspring of chimera by CD-1 crosses (N0 generation) were bred against wildtype (CD-1) females, 17% of surviving adult progeny carried the mutant allele. When animals of this “N1 generation” were again outcrossed against CD-1, 18% of surviving progeny (N2) carried the mutant allele. Similar survival ratios were observed for subsequent generations of outcrossing.
The data imply that 75–80% of Nipbl+/− mice die prior to genotyping (typically done at 4 weeks of age), a fraction that remains stable as the mutant allele is progressively outcrossed onto the CD-1 background. To determine whether lethality occurs in utero, we examined litters for Nipbl+/− embryos just before birth (gestational days E17.5 and E18.5). With no visible marker available for the ES-cell derived progeny of chimeras, this test was carried out with progeny of the N0 generation, in which the Mendelian expectation for the mutant allele is 50%. Mutants were found to comprise 41% (30 out of 67) of progeny, a frequency not significantly different from expected for this sample size (Table 1). These data imply that most mutants die at or after birth.
To evaluate the extent to which Nipbl+/− mice provide a good model for CdLS, we performed an analysis in which we examined these animals for a number of different structural phenotypes analogous to common clinical findings observed in CdLS (summarized in Table S1). Among the most common clinical features of CdLS are small body size, often evident before birth; heart defects; and upper limb abnormalities ranging from small hands to frank limb truncations [2]–[5],[7],[8],[39]. As shown in Table 2, Nipbl+/− embryos examined shortly before birth (E17.5–E18.5) were 18–19% smaller than wildtype littermates (P<0.001), a reduction not accompanied by decreased placental size. Nipbl+/− embryos at earlier stages were also noted to be slightly smaller than littermates (data not shown).
Nipbl+/− embryos did not display limb or digit truncations, or obvious loss of any other bony elements. However, upon staining embryonic skeletons, we observed delays in ossification of both endochondral and membranous bones of Nipbl+/− embryos. As shown in Figure 1A–1D, delayed ossification of the skull and digits was apparent between E16.5 and E18.5. Measurement of long bones and digits at E17.5 revealed, in addition to a symmetrical reduction in bone length (consistent with smaller body size), a significant decrease in the relative extent of ossification (Figure 1E). Otherwise, the patterning of cartilaginous elements was relatively normal, although some subtle differences in morphology were consistently observed, e.g. the shape of the olecranon process of the ulna was consistently abnormal in Nipbl+/− mouse embryos (Figure 1F–1G). Interestingly, dys- and hypoplastic changes of the ulna are common findings in CdLS [40].
Among the cardiac defects that occur in CdLS, atrial and ventricular septal defects are especially common [2],[5],[7]. Atrial septal defects, which were typically large, were observed in about half of Nipbl+/− mouse embryos, (Figure 1H–1K; Table S1), and could be detected as early as E15.5, shortly after atrial septation normally finishes. A reduction in atrial size was also seen in some mutants, but was not a consistent finding. No defects were detected in the atrioventricular valves or septum, outflow tract, or pulmonary vasculature. However, many mutant embryos displayed subtle abnormalities of the ventricular and interventricular myocardium, including abnormal lacunar structures and disorganization of the compact layer, especially near the apex (data not shown). Significantly, no histological or functional cardiac abnormalities were detected among mutant mice that survived the perinatal period (data not shown). This implies that the cause of perinatal mortality is either cardiac, or correlates strongly with the presence of cardiac structural defects.
Histological examination of other organ systems in late embryonic mutant mice revealed no obvious anatomical abnormalities of the lungs, diaphragm, liver, stomach, spleen, kidney or bladder. Brains of neonatal Nipbl+/− mice displayed relatively normal gross anatomy, although a single mutant was observed to have a large brainstem epidermoid cyst (not shown).
Most Nipbl+/− mice that survived the perinatal period reached adulthood, and appeared to have a normal lifespan. However, marked decrease in the body size of mutant mice was evident at birth and throughout all ages (Figure 2A and 2B). Indeed, the 18–19% weight difference between mutant and wildtype mice observed before birth (Table 2) widens to 40–50% by postnatal weeks 3–4 (Figure 2C–2E; this finding has remained consistent over 6 generations [data not shown]). To investigate early postnatal growth of Nipbl+/− mice in more detail, litters fathered by N1 and N2 generation animals were subjected to daily weighing from shortly after birth until sexual maturity (5–6 weeks of age; Figure 2F). Most mutant mice exhibited failure to thrive during the first weeks of life, with many undergoing several days of wasting followed by death (Figure 2F, inset). By 3 weeks of age, the average weight of surviving mutants was only 40% of wildtype, but after weaning this pattern abruptly changed: mutants (even ones that had already begun to show wasting) underwent rapid catch-up growth (Figure 2F), such that by 9 weeks of age they had reached 65–70% of wildtype weight. These observations suggest that, in addition to being intrinsically small, Nipbl+/− mice may have difficulty with suckling, or may receive inadequate nutrition from milk. Remarkably, the weights of children with CdLS also fall further behind age norms during the first year of life, but show significant catch-up growth later on [11].
The distinctive craniofacial features of CdLS, including microbrachycephaly, synophrys, upturned nose, and down-turned lips, play an important role in clinical diagnosis [3],[6]. Micro-CT analysis was used to assess whether Nipbl+/− mice also display consistent craniofacial changes. Analysis of the skulls of 63 adult mice showed significantly smaller size (microcephaly) among all mutants (N = 23), as well as a variety of significant shape changes (Figure 3). The latter included foreshortening of the anterior-posterior dimensions of the skull (i.e. brachycephaly) and an upward deflection of the tip of the snout (Figure 3B–3E). The upturned nares (Figure 3C and 3E) reflect reduced size of the ethmoid and sphenoid bones, which produces a sunken midface. Together, these shape changes in the basicranium and face are consistent with a greater reduction in the size of chondrocranial, as opposed to dermatocranial, elements within the skull. In addition, an 8% average decrease in bone thickness was also observed (ANOVA, df = 47, F = 18.6, p<0.01).
Neurological abnormalities in CdLS include mental retardation, abnormal sensitivity to pain, and seizures [41]. Although Nipbl+/− mice have not been subjected to intensive long-term neurological or behavioral tests, several distinctive behaviors were observed: Repetitive circling (Videos S1, S2, S3) was noted in 20% (34/173; 15 females and 19 males) of adult Nipbl+/− mice (>5 weeks of age), across all generations examined (N0–N4). Repetitive behaviors—including twirling in place [42]—are common symptoms in children with CdLS. In addition, 30% (4/13; all males) of Nipbl+/− mice were noted to adopt opisthotonic postures in response to administration of a normal anesthetic dose of avertin (see Materials and Methods), strongly suggesting seizure activity. Seizures are also common in individuals with CdLS [43],[44].
We also observed that 15% of Nipbl+/− adult mice (24/158; 11 females and 13 males) displayed reflexive hindlimb clasping when suspended by their tails (Videos S4, S5), whereas only 2% (6/268) littermates showed the same behavior (Table S1). Hindlimb clasping has been observed in several mouse models of neurological disorders, including Rett's syndrome [45]–[47], mucolipidosis type IV [48], infantile neuroaxonal dystrophy and neurodegeneration with brain iron accumulation [49], and Huntington's disease [50]–[53].
Histological examination of mutant brains revealed the presence of all major brain structures, grossly normal lamination of the cerebral and cerebellar cortices, but an overall reduction in brain size, consistent with a 25% reduction in endocranial volume observed with micro-CT (Figure 4A, two-tailed T-test, df = 28, T = 5.7 p<0.01). Absence or reduction in size of the corpus callosum was occasionally observed in Nipbl+/− mice (Figure 4B). Obvious patterning defects were noted only in the midline cerebellum, where lobe IX displayed specific reductions (Figure 4C). Interestingly, midline cerebellar hypoplasia is one of the few consistently-reported changes in brain anatomy in CdLS [54]–[56].
Children with CdLS display a range of ophthalmological abnormalities including ptosis, microcornea, nasolacrymal duct obstruction, strabismus, blepharitis and conjunctivitis [57]–[59]. We noted that 22% of Nipbl+/− mice exhibited one or more gross ophthalmological abnormalities (Table S1). Most frequently observed was ocular opacification, observed in 14% of animals (Figure 4D); opacities were often evident as early as three weeks of age. In several cases, this condition was associated with marked periorbital inflammation, and progressed to permanent closure of the eyelids (not shown). Histological analysis revealed inflammatory and fibrotic changes within the corneal epithelium and stroma (Figure 4E), consistent with repeated abrasion or injury. Such injury might arise from neglect due to abnormalities in corneal sensation, from abnormal production or composition of tear fluid, or secondary to periorbital inflammation or infection (e.g. blepharitis; cf. Table S1).
Some degree of hearing loss is observed in almost all individuals with CdLS, and this may play a role in the marked speech disability often seen in this syndrome [60],[61]. To assess hearing in Nipbl+/− mice, we measured auditory brainstem evoked responses (ABR [62]). Abnormalities were found in the majority of mutant mice examined (Table S1). In a few cases, markedly increased thresholds to stimulation were observed (Figure 4F). More commonly, stimulus thresholds were within normal limits, but the relative intensities of the components of the ABR were altered. In particular, mutant mice displayed a characteristic reduction in the amplitude of the third peak (at about 3 msec following stimulus), a latency consistent with an abnormality in the auditory nerve and/or early brainstem neural pathways (Figure 4G).
The Nipbl564 gene-trap mutation is expected to produce a truncated message lacking all but the first exon (Figure S1). Therefore, the level of full-length Nipbl mRNA in Nipbl+/− mice should provide an indication of the activity of the wildtype allele. To measure this level, we used an RNase protection assay based on hybridization to sequences found in exons 10 and 11. Total RNA was analyzed from two tissues: adult liver and E17.5 brain, using age-matched littermate controls. As shown in Figure 5, Nipbl levels in mutants, as a percentage of wildtype levels, were 72–82% in adult liver, and ∼70% in embryonic brain. When western blotting was used to quantify levels of NIPBL protein in Nipbl+/− embryo fibroblasts (MEFs), a reduction to about 70% of wildtype levels was observed (Figure S2).
The observation that Nipbl+/− mice exhibit only a 25–30% decrease in transcript and protein expression, rather than an expected decrease of 50%, is consistent with Nipbl gene being autoregulatory. An alternative explanation is that the mutant allele is “leaky”, i.e. alternative splicing around the gene trap cassette produces some wildtype message. We favor the former explanation because, in both Drosophila and man, the evidence indicates that null mutation of a single allele of Nipped-B/NIPBL produces only a 25–30% drop in transcript levels, the same decrease we observe in Nipbl+/− mice [31],[63],[64]. Thus, even if the Nipbl allele studied here is not null, it is probably quite close to being so. More importantly, the degree of decrease in Nipbl expression in Nipbl+/− mice is comparable to that which causes CdLS in man. Overall the data from multiple species strongly argue that pervasive developmental abnormalities result from remarkably small changes in NIPBL levels.
There has been one report of precocious sister chromatid separation (PSCS) in cell lines derived from individuals with CdLS [28], which was not seen in a second study [29]. We found no statistically-significant elevation of PSCS in cultured Nipbl+/− MEFs (Figure S3), Nipbl+/− embryonic stem cells (data not shown), or adult B-lymphocytes (Figure S3). These results suggest that cohesion defects in the Nipbl heterozygotes, if present, are very subtle; they are also in accord with findings in Drosophila, where PSCS is seen only when both alleles of Nipped-B are mutated [31].
To investigate whether heterozygous loss of Nipbl leads to alterations in transcription, we turned to expression profiling of tissues and cells from Nipbl+/− mice. Because such mice display pervasive developmental abnormalities, transcriptome data can be expected to reflect not only the direct consequences of reduced Nipbl function, but also a potentially large number of transcriptional effects that are secondary consequences of abnormal morphology and physiology. In an effort to minimize the detection of such secondary effects, we focused on profiling samples in which frank pathology was not seen, or had yet to develop by the time of profiling.
The samples chosen for analysis were embryonic day 13.5 (E13.5) brain, and cultures of fibroblasts derived from E15.5 embryos (mouse embryo fibroblasts; MEFs). Although mature brain appears to be functionally abnormal in Nipbl+/− mice (see above), at E13.5 it at least appears anatomically normal. Cultured MEFs were chosen because they are established with similar efficiency from both mutant and wildtype embryos; exhibit similar morphology and growth characteristics in culture; and by virtue of being maintained ex vivo, are freed of the secondary influences of any systemic metabolic or circulatory derangements within Nipbl+/− embryos. Transcriptome analysis was performed using Affymetrix microarrays. MEF RNA samples were obtained from 10 mutant and 9 wildtype embryos taken from three litters (19 separate microarrays); brain RNA was analyzed from 10 mutant and 11 wildtype embryos from two litters (21 separate microarrays).
Gene expression changes were detected in both comparisons. In the brain (Table S2), 1285 probe sets, corresponding to 978 genes, displayed statistically significant differences in expression between wildtype and mutant mice (per-probe-set false discovery rate of Q<0.05). By and large, the effects were small: 97.5% of changes were within 1.5-fold of wildtype expression values; >99.6% were within 2-fold. The single largest statistically-significant change was 2.5-fold. Genes encoding products of virtually all structural and functional categories could be found among those affected, with no dramatic enrichment of any particular functional sets (by Gene Set Enrichment Analysis [65]; data not shown).
In cultured Nipbl+/− MEFs, 89 probe sets, corresponding to 81 genes (Table S3), displayed statistically-significant (Q<0.05) differences in expression between wildtype and mutant mice. Again, effects were small: 89% of changes were within 1.5-fold of wildtype, and 99% were within 2-fold. The single largest statistically-significant change was 2.1-fold. The lower number of transcriptional changes identified in MEFs versus brain may not be biologically meaningful, as MEFs happened to display a somewhat higher average within-sample variance than E13.5 brain, making it more difficult for small changes to be judged significant.
As with embryonic brain, transcriptional effects in MEFs involved genes that encode a wide variety of proteins. Although automated analyses failed to single out any particular functional class as being highly overrepresented, manual curation revealed significant changes in the expression of a number of genes implicated in adipogenesis (Figure 6A). For example, Cebpb and Ebf1—which encode transcriptional factors central to the process of adipocyte differentiation [66]–[68]—were both down-regulated in Nipbl+/− MEFs, as were Fabp4 and Aqp7, well-known adipocyte markers [69],[70]. Other genes down-regulated in Nipbl+/− MEFs (Table S3) could also be found, through literature searches, to exhibit expression positively correlated with adipocyte differentiation, including Adm, Lpar1, Osmr, and Ptx3 [69],[71],[72]. Several additional genes (Amacr, Avpr1a, Il4ra, Prkcdp, S100b) down-regulated in Nipbl+/− MEFs can be inferred, from publicly-available expression data, to be enriched in pre-adipocytes and/or brown or white adipose tissue [73]–[75]. Conversely, Lmo7, which is normally down-regulated during late adipogenic differentiation [71], was found to be up-regulated in Nipbl+/− MEFs. Furthermore, we noted that genes such as Cebpa and Cebpd (transcriptional activators of adipocyte differentiation [66],[76]), Il6 (a cytokine stimulator of adipocyte differentiation that controls adiposity in man [77],[78]) and Socs3 (an intracellular signaling regulator induced by Il6 [79]), were also down-regulated in the MEF samples, but at false-discovery rates slightly too high to permit their inclusion in Table S3 (Q = 0.065, 0.085, 0.075, and 0.17, respectively).
Together, these data raise the possibility that Nipbl+/− mice are specifically impaired in adipogenesis. Support for this idea was obtained by weighing intrascapular fat dissected from adult mutant and wildtype littermates [80]. As shown in Figure 6B, both brown and white fat are substantially depleted in Nipbl+/− mice. To correct for the fact that mutant mice are generally smaller than their wildtype littermates, we normalized fat measurements to brain weight (which scales with overall body size). As shown in Figure 6C, even by this measure, Nipbl+/− mice displayed a significant, substantial reduction in body fat. As mentioned earlier, lean body habitus is also a characteristic of CdLS.
To investigate whether the reduction in body fat in Nipbl+/− mice reflects an intrinsic defect in the differentiation potential of mutant fibroblasts, we studied adipogenic differentiation in vitro. It is known that embryonic fibroblasts can be converted, in large numbers, to adipocytes by treatment with agents such as glucocorticoids, PPAR-γ agonists, isobutylmethylxanthine and insulin, which stimulate the activity of a core network of pro-adipogenic transcription factors (C/EBPα, C/EBPβ, C/EBPδ, PPARγ; [81],[82]). In response to such agents, we observed no significant difference between Nipbl+/− and wildtype MEFs in terms of the number of adipocytes or adipocyte colonies produced (data not shown). However, when we omitted these pharmacological agents, and measured the (much lower) level of spontaneous adipogenic differentiation that occurs in MEF cultures [83], we observed a substantially-lower level in mutant cultures (Figure 6D–6F). The observation that Nipbl+/− MEFs are impaired in spontaneous, but not induced, adipogenesis implies that their primary defect does not lie downstream of the targets of pharmacological inducers.
Of the 80 genes (not counting Nipbl itself) with significant differential expression in Nipbl+/− MEFs (Table S3), 20% (16/80) are also found among the 978 genes whose expression was altered in Nipbl+/− embryonic brain (Table S2). Using a more stringent false discovery rate cutoff of Q<0.02 for both samples, we find that 23% (9/40) of differentially expressed MEF genes are among the 560 that are differentially expressed in brain. These data suggest that common transcriptional targets exist in the two tissues. Further support for this idea is obtained by correlating fold-increase or -decrease of affected transcripts. In this case a less conservative approach to false discovery is justified (the goal is to estimate overall correlation between samples, not implicate individual genes), so the log-fold changes for all probe sets that exhibited differential expression exceeding an arbitrary t-statistic threshold (t>2) in both tissues were plotted against each other (shown in Figure 7). The data are clearly strongly correlated (R = 0.77), suggesting that at least some of the transcriptional effects of Nipbl deficiency are shared across tissues. Among the genes in which expression changes contributed substantially to the correlation are four members of the protocadherin β cluster (Pcdh17, Pcdh20, Pcdh21, Pcdh22; all down-regulated), Lpar1 (also down-regulated; encoding the lysophosphatidic acid receptor), Vldlr (down-regulated; encoding a receptor involved in both lipid metabolism and cerebral cortical development), and Stag1 (up-regulated; encoding SA1, a cohesin component). Interestingly, in Drosophila, inhibition of Nipped-B expression also leads to up-regulation of the ortholog of Stag1 [31]. Recently, STAG1 up-regulation has also been seen in lymphoblastoid cell lines of individuals with CdLS [64]; Table S5.
Among the most significant changes common to mutant MEF and brain samples were decreases in expression of transcripts from the 22-gene Pcdhb (protocadherin beta) cluster on chromosome 18 (Table S2 and Table S3, Figure 7). As shown in Figure 8A, affected transcripts included Pcdhb7,16,17,19,20,21 and 22, which lie predominantly at the 3′ end of the cluster. This observation raised the possibility that the transcriptional effects of Nipbl might be related to the physical locations of genes. However, as genes at the 5′ end of the Pcdhb cluster tend to be expressed at lower levels than those at the 3′ end, lower signal-to-noise ratios might have made small changes in expression at the 5′ end more difficult to detect.
To resolve this issue, and to provide independent confirmation of microarray data, quantitative RT-PCR was used to measure transcripts levels at multiple locations throughout the Pcdhb cluster (Figure 8B). For these experiments, brain mRNA was prepared at a later developmental stage (E17.5, when most Pcdhb transcripts are more highly expressed) from 13 independent samples (7 mutant and 6 wildtype embryos). Robust RT-PCR signals were obtained for 14 of 15 transcripts tested (Pcdhb2,3,4,5,7,8,9,10,13,14,16,17,19, and 22; but not Pcdhb1). As shown in Figure 8B, the data support the microarray results from the earlier embryonic stage, and indicate that most transcriptional changes in Nipbl+/− brain indeed occur preferentially at the 3′ end of the cluster (Pchdb13,14,15,16,17,19,22). Additionally, they suggest that at least one 5′ gene, Pcdhb2, may also be affected.
A more revealing analysis of the data can be obtained by correlating Pcdhb transcript levels in each tissue sample, regardless of genotype, against Nipbl transcript levels within that sample (i.e. treating Nipbl expression as a quantitative trait; Figure 8C, Figure S4). This approach offers greater discriminatory power because Nipbl expression in individual samples varies significantly, even within mutant and wildtype groups, and occasionally overlaps between the two groups. Indeed, the results of the analysis indicate that Pcdhb expression correlates strongly with Nipbl transcript level, lending support to the view that Pcdhb transcription is directly affected by the amount of NIPBL present in cells. In Figure 8C, the results of such correlations for all 13 tested Pcdhb transcripts are summarized by plotting the slopes of regression lines (the sensitivity of each transcript's expression to Nipbl level) against gene location, with error bars reflecting the strength of correlation for each gene. The results strongly suggest a continuum of sensitivity to Nipbl across the entire Pcdhb cluster, with genes at both the 5′ and 3′ ends being the most sensitive, and those in the middle being least affected.
We show here that mice heterozygous for a gene-trap mutation upstream of the first coding exon of Nipbl displayed many features of human CdLS, including pre- and postnatal growth retardation, cardiac septal defects, delayed bone development, lean body habitus, microbrachycephaly with characteristic craniofacial changes, behavioral disturbances, ophthalmological abnormalities, cerebellar hypoplasia, and hearing deficits (Figures 1–4, Table S1, Videos S1, S2, S3, S4, S5). These phenotypes remained stable through many generations of outcrossing, and occurred in the context of modest (25–35%) reductions in levels of Nipbl mRNA in every tissue measured (embryonic brain, MEFs, adult liver). Similarly modest reductions have recently been reported in cell lines derived from individuals with CdLS [63],[64].
In some cases, quantitative agreement between the mouse model and CdLS is remarkable, e.g. fall-off and catch-up in growth rates during early postnatal life, the upturned nose. Yet some common features of CdLS are not observed in the mouse model, such as reduction and fusion abnormalities of the upper limb, which is seen in up to 30–50% of children with CdLS (depending on criteria used). The mutant mouse heart also displays only atrial and not ventricular septal defects, whereas both occur at similar frequency in CdLS. Mutant mice also display some pathological features, such as corneal opacities, that are atypical of CdLS (corneal scarring has been noted, however [58]). Furthermore, the frequency of perinatal mortality in CdLS is estimated at about 10% [8], not as high as in the mutant mouse (although this may simply reflect better postnatal care).
Despite these differences, it is clear that the Nipbl+/− mouse is an excellent animal model for many features of CdLS, and provides the first experimental verification that Nipbl mutations cause the syndrome. Interestingly, wide variation in the penetrance or severity of phenotypes, a distinctive feature of CdLS, was also observed in mutant mice. Because the mouse line was maintained on an outbred (CD-1) background (given high mortality of heterozygotes, it was not practical to maintain the line on an inbred background), genetic heterogeneity could have accounted for some of the variability.
It is fascinating that the diverse and severe pathology observed in this study is caused by only a 25–35% decrease in the level of Nipbl transcripts (Figure 5; also see Table S2 and Table S3). A recent study of a rare familial case of CdLS involving a mutation in the 5′-untranslated region of NIPBL suggests that a mere 15% decrease in transcript levels is associated with a clinically significant phenotype [63]. Given the extraordinary sensitivity of development to the level of expression of this one gene, it would not be surprising if an unusually high proportion of disease-causing mutations in Nipbl occur in regulatory DNA, where they would be difficult to detect. This could help explain why such a large proportion of CdLS mutations (∼40%) have yet to be identified [23],[84],[85].
Results of the present study support the view that changes in Nipbl level have significant, yet modest, effects on transcription throughout the genome. At present, it is impossible to know how many observed gene expression changes (Table S2 and Table S3) are primary—due to direct transcriptional actions of NIBPL—and how many are downstream consequences of gene misregulation. Among the affected MEF transcripts that were noted to participate in adipogenic differentiation, for example, many are transcriptional targets of each other (Figure 6A), raising the possibility that direct actions of NIPBL may be confined to a subset of these.
Among the most likely candidates for direct NIPBL “targets” are those genes that displayed similar expression changes in both cultured MEFs and E13.5 brain (Figure 7). Prominent among these were genes of the protocadherin beta (Pcdhb) cluster. Measurements in later-stage brain confirmed that alterations in gene expression occur throughout the Pcdhb locus in Nipbl+/− mice, but in a manner that is positionally graded across the cluster (Figure 8). Such effects are consistent with a role for NIPBL in the long-range, coordinated regulation of sets of genes.
Additional evidence for this hypothesis can be found in the E13.5 brain expression data: As shown in Table S4, there are at least 13 other examples of small clusters (usually 2–4 genes) of related (paralogous) genes, in which Nipbl+/− mice display similar expression changes in more than one paralog. Two of these are the well-studied β- and α-globin loci [86], in which long-range cis-regulatory elements (locus control regions) are known to control and coordinate expression of different transcripts (the globin transcripts in brain RNA presumably come from fetal erythrocytes in the tissue). Interestingly, whereas decreases in expression were seen at all four β-globin genes in Nipbl+/− samples, the magnitudes varied greatly among the genes within each cluster (arguing against the trivial possibility that Nipbl+/− brains simply have less blood in them). In fact, the single greatest gene expression change in the entire study (∼2.5 fold decrease) involved one of the transcripts (Hbb-bh1) of the β-globin locus.
It is known that the transcriptional insulator protein CTCF plays an important role in establishing chromatin boundary elements at the β-globin locus [87]. The Pcdhb locus is also flanked, at least in man, by sites occupied by CTCF [88], the functional significance of which has yet to be studied. CTCF insulation is also involved in control of the myc-locus [89], which is highly significantly down-regulated in E13.5 Nipbl+/− brain (Table S2). Even the Igf2/H19 locus, at which long-range, CTCF-dependent regulation of gene silencing has been shown to occur [90],[91], displayed evidence of H19 down-regulation (by ∼20%) in the Nipbl+/− brain, albeit at lower statistical significance (Q = 0.14).
In view of recent work showing that cohesin and CTCF binding sites extensively co-localize in the mammalian genome (including at the β-globin, Igf2/H19 and myc loci [35]–[37]), and that cohesin contributes to CTCF function [35],[37], it is reasonable to speculate that at least some of the transcriptional effects in Nipbl+/− tissues arise from impaired CTCF function. It should be noted, however, that CTCF sites are far more common (>13,000 per mammalian genome) than Nipbl-sensitive genes, and we find no clear correlation between the locations of Nipbl-sensitive genes (or the magnitudes of transcriptional effects in those genes) and known or predicted CTCF-sites (X. Xie and N. Infante, personal communication). It remains possible that only a subset of NIPBL transcriptional effects is related to CTCF function. Indeed, it is possible that NIPBL acts primarily by influencing other aspects of long-range cis-regulatory interaction (e.g. histone methylation, DNA looping), which simply take place frequently at CTCF-regulated loci.
Despite the extensive overlap between the phenotypes of Nipbl+/− mice and CdLS, it is interesting to note that the gene expression changes recently reported in lymphoblastoid cell lines of CdLS individuals [64] exhibit only limited overlap (6–8%; cf. Table S5) with those we observed in mouse embryo fibroblasts and embryonic brain (Table S2 and Table S3). So far it is unclear whether this stems from a high degree of tissue-specificity in the expression of genes that are directly affected by NIPBL; a high proportion of indirect NIPBL targets (which might be more likely to vary from tissue to tissue); or the effects of differences in genomic organization between mouse and man.
Among the gene expression changes detected in Nipbl+/− MEFs and brain (Table S2 and Table S3) one can find many genes that, when mutated in mice or man, produce phenotypes that overlap with CdLS. These include skeletal and craniofacial abnormalities (Lpar1, Pitx2, Satb2, Tcof1, Trps1); heart defects (Adm, Cited2, Cxcl12, Gja1, Hey2, Pitx2, Mef2c); reduced body size (Ebf1, Lpar1, Hsd3b7, Mef2c); decreased adiposity (Cebpb, Ebf1, Lpar1, Npy, Vldlr); behavioral abnormalities (Avpr1a, Ctnnd2, Lpar1, Vldlr); seizures (Cdk5r1, Gabrb1, Gabrb2, Neto1, Nr4a3, Plcb1, S100b, Sv2b); and hearing deficits (Cldn11, Eya, Gjb2, Gjb6) [66], [92]–[124].
For most of the genes mentioned above, however, phenotypes are observed only with complete loss of gene function. In the few cases in which significant heterozygous phenotypes are seen (e.g. Satb2, Pitx2, Trps1, Tcof1, Cited2), expression changes in the same genes in Nipbl+/− samples are not even as great as would be expected for heterozygous loss. Of course, it is possible that greater expression changes occur at other stages, or in other tissues, than those sampled here. However, the alternative interpretation is that phenotypes in Nipbl+/− mice, and in individuals with CdLS, arise from the collective effects of small changes in the expression of many genes. Although we are not yet in a position to distinguish between these hypotheses, we recognize that this issue is closely related to a major unanswered question in human genetics: whether most common disease phenotypes arise from large effects at a few loci, or from very many loci of small effect. The results of the present study suggest that further study of CdLS, and related “cohesinopathies” [125],[126], could shed light on a fundamental question of widespread importance.
Ethics Statement: All animals were handled in strict accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all animal work was approved by the University of California Irvine Institutional Animal Care and Use Committee (protocol 1998-1656).
A search for Nipbl sequences in mouse gene-trap databases (http://www.genetrap.org/) initially identified two targeted ES cell lines (generated using the E14 parental cell line, which has a 129/Ola background). One of these (RRS564) contains a gene-trap in the intron between exon1 and exon 2; the other (RRJ102) in intron 25 (the exon numbering of [16] for the human gene is used here). Gene-trap constructs are designed to terminate transcription and translation, producing a truncated or absent protein product. Both cell lines were injected into blastocysts of C57BL/6 mice (for the RRJ102 cell line, 83 blastocysts were injected; for RRS564, 324 blastocysts were injected). Multiple male chimeras were obtained and bred against outbred (CD-1; Charles River) females. Germ line progeny (distinguishable by chinchilla coat color) were obtained only from RRS564-derived chimeras, and further work on RRJ102 was suspended. The RRS564 allele is hereafter referred to as Nipbl564 and mice heterozygous for this allele as Nipbl+/− for simplicity. Nipbl+/− mice were maintained under normal laboratory conditions, and the line propagated by successive rounds of outcrossing to CD-1 mice. Offspring were genotyped using LacZ-(Forward 5′-TGATGAAAGCTGGCTACAG-3′ and Reverse 5′-ACCACCGCACGATAGAGATT-3′) primers.
Anatomical and histological evaluations were performed using fresh-frozen or paraformaldehyde fixed tissues. In some cases, fixation was carried out by cardiac perfusion. Alcian Blue/Alizarin Red staining was carried out as described [127]. Hematoxylin-eosin and cresyl violet staining was carried out using standard techniques. Micro-CT analysis of adult (>90 days) skulls was performed using a Scanco VivaCT as described [128],[129]. Craniofacial shape was assessed using geometric morphometric techniques, and cranial vault thickness was assessed in 3D as described [130],[131]. Scapular fat pads were dissected and measured as described [80]. Auditory brainstem response recordings were generated as described [62]. Briefly, a cohort of young adult mutant and littermate control animals were anesthetized with avertin (2.5% solution of tribromoethanol in tert-amyl alcohol; 20 µl/g body weight administered by i.p. injection), and subcutaneous electrodes inserted at the level of the brainstem to record neural potentials evoked by a variety of clicks and tones introduced into one ear.
Embryo fibroblasts (MEFs) were cultured from E15.5 Nipbl+/− and wildtype littermate embryos as described [127]. Metaphase spreads of MEFs were prepared from cells cultured for 12 hrs in medium supplemented with 0.1 µg/ml colchicine. Trypsinized cells were pelleted, incubated in 75 mM KCl for 20 min at 37°C, then re-pelleted and fixed in 3∶1 methanol/acetic acid. Cells were dropped onto glass slides and stained with 4′-6-Diamidino-2-phenylindole or Giemsa. Microscopic assessment was carried out for each slide by three independent observers who were blinded to the genotypes of the sample.
B cells were isolated from mouse spleens by immunomagnetic depletion with anti-CD43 beads (Miltenyi Biotech), cultured in RPMI1640 with 10% fetal bovine serum, and stimulated with lipopolysaccharide (25 µg/ml; Sigma) and IL4 (5 ng/ml; Sigma) for 3 days. Cells were arrested at mitosis by treatment with 0.1 µg/ml colcemid (Roche) for 1 hour, and metaphase chromosome spreads prepared following standard procedures. Cells were stained with 4′-6-Diamidino-2-phenylindole and images of metaphases acquired with an Axioplan2 upright microscope (Zeiss), using Metamorph software.
Measurements of Nipbl levels by RNase protection were made according to standard methods. The Nipbl probe contained 39 bases of exon 10, all 183 bases of exon 11, and 4 bases of exon 12. There is no expressed sequence tag evidence supporting alternative splicing of exons 10–11, and in situ hybridization studies in mouse embryos indicated they are ubiquitously expressed, so it was felt that this probe would provide a good indication of overall levels of expression. Briefly, for each reaction, 20 µg of total RNA was hybridized with 32P-labeled probes for Nipbl (90,000 cpm) and Gapdh (glyceraldehyde 3-phosphate dehydrogenase; containing 116 bases of exon 4 and 15 bases of exon 5; 20,000 cpm) and processed according to manufacturer's instructions (Ambion RPA III kit). Samples were run on a 5% polyacrylamide/8M urea gel, dried, and bands quantified by phosphorimager.
For measurement of NIPBL protein levels, MEFs were lysed in cold buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, and protease inhibitors [1 mM phenylmethylsulfonyl fluoride (PMSF), 1 mM EDTA, 2 µg/ml aprotinin, 2 µg/ml leupeptin, and 2 µg/ml pepstatin]. Total protein (20 µg) from 3 wildtype MEF homogenates and 3 Nipbl+/− MEF homogenates was separated, in duplicate, on 7.5% SDS-PAGE gels and transferred onto Immobilon-P membranes (Millipore, Bedford,MA). The membranes were blocked with 2% BSA and sequentially incubated with the anti-GAPDH (1∶200,000, 6C5; Ambion, Austin,TX) and anti-NIPBL-N (1∶60000, anti-NIPBL antibody was produced in rabbit from a GST fusion protein containing amino acids 1–380 of human NIPBL, and was affinity purified using original antigen). The membranes were then incubated with the horseradish peroxidase-conjugated anti-rabbit antibody (1∶10000) and detected using chemiluminescence. Images were scanned and densitometry performed.
To evaluate spontaneous adipogenic differentiation, MEFs (passage 2) were seeded into 96-well plates at 7,500 cells/well, maintained in Dulbecco's modified Eagle's Medium with 10% fetal bovine serum, 100 U/ml penicillin and 100 µg/ml streptomycin at 37°C, cultured to confluence, and maintained for 7 additional days. Lipid accumulation was visualized by staining with Oil Red O. Briefly, cells were washed with PBS and fixed with 10% formaldehyde (30 minutes at room temperature), rinsed, and permeabilized in 60% isopropanol for 2–5 minutes. Isopropanol was removed and Oil Red O solution (Chemicon, Temecula, CA) was added for 30 minutes. Wells were washed several times with PBS then counterstained with Hoechst 33258 (0.01 mg/ml). Four wildtype and five Nipbl+/− MEF lines were analyzed. Each was plated in triplicate wells and 7 fields (at 10× magnification) within each well were photographed and lipid-containing cells (stained lipids appear red under phase contrast) and nuclei (appear blue with fluorescent microscopy) were counted using Image J software (NIH) to detect nuclei (nucleus counter plug-in: approximately 1000–1200 nuclei were detected per field) and lipid-containing cells (point picker plug-in).
Analysis of gene expression was performed on total RNA isolated using the trizol method from MEFs, adult liver, or manually dissected E13.5 brain. Hybridization and data collection were carried out by the Broad Institute (Cambridge, MA). RNA was labeled and hybridized to Affymetrix Murine 430A 2.0 (for MEFs) or 430 2.0 (for brain) array chips using the protocol described at http://www.broad.mit.edu/mpr/publications/projects/Leukemia/protocol.html, and data were analyzed using GenePattern software (http://www.broad.mit.edu/cancer/software/genepattern). Expression data sets were assembled from individual CEL files using the RMA algorithm with quantile normalization. Data were log2-transformed and transcripts with near-background expression filtered. Measures of statistical significance were obtained by permutation testing [132], using the Comparative Marker Selection module. Significance is presented in terms of per-sample false discovery rates, or Q-values [133]. Data were also analyzed using D-chip software (http://www.biostat.harvard.edu/complab/dchip), which yielded similar enrichment sets (not shown). Probe sets were annotated, and gene locations obtained, according to the NCBI m37 mouse assembly, and Affymetrix annotation files ver. 28 (March 2009).
For measurements of transcript abundance by quantitative PCR, RNA (5 µg) from E17.5 mouse brains was reverse transcribed with Superscript II, oligo dT, and random hexamers according to manufacturer's instructions (Invitrogen, Carlsbad, CA). Reactions were assembled using iQ SYBR Green Supermix (BioRad, Hercules, CA) and processed in 20 µl volumes with 1 µl of cDNA (diluted 1∶25) and primers at a final concentration of 100 nM. Specificity of amplification was verified for each reaction by examination of the corresponding melt curve. Normalization was carried out using beta-2 microglobulin as a standard, and genomic amplification controlled for using samples prepared without reverse transcription. All PCR reactions were performed on an iQ5 iCycler (BioRad). Cycling conditions were 95°C for 4 min and then 40 cycles of 95°C 10 sec, 61°C 30 sec and 72°C 30 sec. Primers were:
Pcdhb2: agcccacctggtagatgttg and attggggatgattggtttca; Pcdhb3: cctggaaatacaccgcagaa and cctagacatggacccagcaa; Pcdhb4: cagtcagtcccaacctcca and tgaactgtggtcatcccagac; Pcdhb5: cagaggggaaatcaggaaca and gggcttaaactggcaatgaa; Pcdhb7: accccacacaggaagttgag and ctttatccccacgaaaagca; Pcdhb8: gccttggcttctgtgtcttc and caccactgacatccaccaag; Pcdhb9: atgcctggtgaacactttcc and gcagtggggactttccataa; Pcdhb10: gctgaccctcacctctcttg and accaccacgagtaccaaagc; Pcdhb13: ggcttctctcagccctacc and cagcaccacagacaagagga; Pcdhb14: cattgcacataggcaccatc and tgatggagatgagcgagttg; Pcdhb16: tggcttctctcagccctacc and aacagcagcacagacaccag; Pcdhb17: gcaagtcctggctttctttg and ggatatctctgccaggtcca; Pcdhb19: gacaaggcaagtcctgcttc and ccccaggtcctttaccaaat; Pcdhb22: tatcatcgctcaccaatcca and cagagctccatctgtcacca, beta-2-microglobulin :, atgggaagccgaacatactg and cagtctcagtgggggtgaat, and Nipbl (exons 6–7): agtccatatgccccacagag and accggcaacaataggacttg. PCR product sizes were between 107 and 182 bp.
|
10.1371/journal.pntd.0002157 | Indirect Immunofluorescence Assay for the Simultaneous Detection of Antibodies against Clinically Important Old and New World Hantaviruses | In order to detect serum antibodies against clinically important Old and New World hantaviruses simultaneously, multiparametric indirect immunofluorescence assays (IFAs) based on biochip mosaics were developed. Each of the mosaic substrates consisted of cells infected with one of the virus types Hantaan (HTNV), Puumala (PUUV), Seoul (SEOV), Saaremaa (SAAV), Dobrava (DOBV), Sin Nombre (SNV) or Andes (ANDV). For assay evaluation, serum IgG and IgM antibodies were analyzed using 184 laboratory-confirmed hantavirus-positive sera collected at six diagnostic centers from patients actively or previously infected with the following hantavirus serotypes: PUUV (Finland, n = 97); SEOV (China, n = 5); DOBV (Romania, n = 7); SNV (Canada, n = 23); ANDV (Argentina and Chile, n = 52). The control panel comprised 89 sera from healthy blood donors. According to the reference tests, all 184 patient samples were seropositive for hantavirus-specific IgG (n = 177; 96%) and/or IgM (n = 131; 72%), while all control samples were tested negative. In the multiparametric IFA applied in this study, 183 (99%) of the patient sera were IgG and 131 (71%) IgM positive (accordance with the reference tests: IgG, 96%; IgM, 93%). Overall IFA sensitivity for combined IgG and IgM analysis amounted to 100% for all serotypes, except for SNV (96%). Of the 89 control sera, 2 (2%) showed IgG reactivity against the HTNV substrate, but not against any other hantavirus. Due to the high cross-reactivity of hantaviral nucleocapsid proteins, endpoint titrations were conducted, allowing serotype determination in >90% of PUUV- and ANDV-infected patients. Thus, multiparametric IFA enables highly sensitive and specific serological diagnosis of hantavirus infections and can be used to differentiate PUUV and ANDV infection from infections with Murinae-borne hantaviruses (e.g. DOBV and SEOV).
| Hantaviruses are the causative agents of hemorrhagic fever with renal syndrome (HFRS) and hantavirus cardiopulmonary syndrome (HCPS) — serious emerging diseases, with case-fatality rates of up to 15% and about 35%, respectively. So far, over 21 human pathogenic serotypes have been described, which are classified into New World (circulating in the Americas) and Old World (Asia and Europe) hantaviruses. The prodromal phase of hantavirus infections — fever, myalgia, headache and gastrointestinal symptoms — is indistinguishable from those of many other viral infections. The cardiopulmonary phase of HFRS and diuretic phase of HFRS mimic the acute respiratory distress syndrome and renal failure, respectively. In this context, clinical diagnosis has to be confirmed by laboratory testing, which is predominantly based on serology. Although there is an increasing awareness of hantaviruses, infections are still underdiagnosed, in part due to a lack of available standardized serological assays. This study evaluated a commercial multiparametric indirect immunofluorescence assay for the simultaneous detection of antibodies against clinically important Old World (Hantaan, Puumala, Seoul, Saaremaa and Dobrava) and New World (Sin Nombre and Andes) hantaviruses. Test performance was found to be comparable to established highly sensitive and specific in-house assays.
| Hantaviruses are enveloped and negative-sense single-stranded RNA viruses of the Bunyaviridae family. The hantavirus genome consists of three segments (L, M, and S), coding for the viral RNA polymerase (L protein), glycoproteins (Gn and Gc) and the nucleocapsid (N) protein, respectively [1]–[5]. The majority of hantaviruses are etiologic agents of either hemorrhagic fever with renal syndrome (HFRS) or hantavirus cardiopulmonary syndrome (HCPS). The number of hantavirus infections is increasing, as reflected by a very recent outbreak at Yosemite National Park (USA; June–August 2012), which put an estimated 10,000 persons at risk of infection and caused several fatal cases [6]. Transmission to humans occurs through the respiratory tract by inhalation of dust and aerosols containing virus-contaminated particles shed by persistently infected viral reservoir species (primarily mice, voles and rats).
So far, over 21 human pathogenic hantavirus serotypes have been described [7]–[9], which are classified into New and Old World hantaviruses according to their worldwide distribution and genetic relatedness. New World hantaviruses include, amongst various others, Andes virus (ANDV) [10] and Sin Nombre virus (SNV) [11], the main causative agents of HCPS in South and North America, respectively, with case-fatality rates of about 35%, mainly due to pulmonary complications and cardiogenic shock [12]. Clinically relevant Old World hantaviruses, predominately distributed in the Eastern Hemisphere, include Dobrava (DOBV) [13], Hantaan (HTNV) [14], Puumala (PUUV) [15], Seoul (SEOV) [16] and Saaremaa (SAAV) virus [17], [18]. The mildest form of HFRS, designated nephropathia epidemica, is caused by PUUV and is associated with a mortality rate of less than 0.1%. SAAV also causes fairly mild HFRS, whereas SEOV, DOBV and HTNV cause moderate to severe HFRS with fatality rates of 1–15% [19].
Due to the rather unspecific symptoms such as headache, backache, myalgia, shivering, abdominal pain and nausea in a high proportion of infected patients, hantavirus syndromes are often clinically misdiagnosed as influenza-like infections, renal failure or idiopathic acute respiratory distress. In this context, implementation of at least one laboratory test is mandatory to support clinical diagnosis. Hantaviruses can be detected either directly by virus isolation or reverse transcriptase polymerase chain reaction (RT-PCR)-based amplification of hantaviral RNA or indirectly by serology [20]. With respect to direct detection, it has to be noted that the level of plasma-associated hantaviral RNA rapidly decreases after the onset of initial symptoms and is suggested to be associated with disease severity (highest RNA load in patients with severe/critical disease) [21]–[23]. RT-PCR in peripheral blood mononuclear cells (PBMC) yields a higher sensitivity, and in most ANDV-infected patients it is successful for up to 3 months after hospitalization (P.A. Vial, unpublished results). Because of the short-termed viremia, detection of hantavirus-specific serum antibodies of class IgM and IgG is most reliable and, thus, widely used for confirmation of hantavirus infection.
For the simultaneous detection of specific serum IgG and IgM against the clinically important hantaviruses, multiparametric indirect immunofluorescence assays (IFAs) were developed based on mosaics of biochips coated with hantavirus-infected cells (positive for serotypes HTNV, PUUV, SEOV, SAAV, DOBV, SNV and ANDV). For assay evaluation, IgG and IgM antibodies were determined in serum panels from healthy blood donors (controls) and confirmed hantavirus-infected patients provided by diagnostic laboratories in six different geographic regions. Previous diagnostic data from these laboratories using mainly in-house or sometimes commercial assays served as reference data.
Samples tested were derived from already-existing collections at the indicated institutes and were numerically coded, with the identity of the persons available only to clinicians interacting with the patients or blood donors. Furthermore, patient samples used in this study were routine diagnostic samples taken with patient consent and sent to the respective laboratory to be tested for anti-hantavirus antibodies. Stored aliquots of these anonymized samples were used in this study (conducted in the years 2006–2009) for a comparison of diagnostic methodologies with each sample being tested for the same parameter for which it had initially been screened. The results were not accessible to any outside body and were not reported to the donor or the patient. Therefore written informed consent was not required. Samples were included in this study on the basis of a statement from the Central Ethics Committee of Germany on the use of human samples for research studies [24].
Assay evaluation was based on 5 serum panels comprising 184 samples from patients in the acute or convalescent phase of hantavirus infection or with past hantavirus infection. Diagnosis of infection was based on clinical (HFRS or HCPS) and serological findings. Characteristics of infected patients are summarized in Table 1. Panel V was a subset of the serum panel previously analyzed by Schmidt and colleagues [25], including 23 sera from Argentina (original sample IDs: 1A, 2A to 5A, 6A, 6B, 7B, 8B to 9B, 11 to 16, 18 to 20) and 29 sera from Chile (original sample IDs: 1CH-A to 2CH-B, 3CH-A to 5CH-B, 6CH-A to 11CH, 13CH, 14CH, 16CH to 22CH). These sera were obtained from 38 patients (7 female, 31 male; median age 30 years; age range 8–77 years), including 13 follow-up patients. Data for gender and age of the other individuals included in this study were not available. The control group comprised 89 serum samples from apparently healthy Canadian (n = 25) and German (n = 64) blood donors (Table 1). All sera were stored at −80°C until analysis.
Laboratory confirmation of hantavirus infection was performed at the indicated diagnostic centers shown in Table 1. Regarding panel V, laboratory diagnosis had been performed at the WHO Collaborating Centre for Arbovirus and Haemorrhagic Fever Reference and Research (Hamburg, Germany). Reference tests were characterized as follows (Table 1): panel I, IgG IFA based on PUUV-infected Vero E6 cells, IgM μ-capture ELISA based on lysates of Sf9 insect cells expressing recombinant nucleocapsid (rN) protein of PUUV [26]; panel II, IFA based on HTNV- or SEOV-infected Vero E6 cells; panel III, IgM enzyme immunoassay (EIA) based on rN protein (Dobrava-Hantaan IgM EIA from Reagena, Toivala, Finland, and/or Hantavirus IgM EIA from Focus Diagnostics, Cypress, USA) and rN-protein-based IgG EIA and/or IgG ELISA (Hantavirus IgG EIA from Focus and/or Hantavirus IgG ELISA from Progen, Heidelberg, Germany) or IgG Western blot (recomBlot Bunyavirus IgG from Mikrogen, Neuried, Germany); panel IV, IgG ELISA based on lysates from Black Creek Canal virus (BCCV)-infected Vero E6 cells and IgM capture ELISA based on antigen prepared from BCCV-infected Vero E6 cells [27]; panel V, anti-ANDV ELISA based on yeast-expressed rN protein of ANDV [25]. All patient samples were positive for hantavirus-specific IgG and/or IgM in the corresponding reference tests. In panel III, one follow-up serum from a patient with past DOBV infection was previously tested for IgG only, whereas in the present study both IgG and IgM were determined by means of IFA. Reference tests used for examining the control panel were: Anti-Hantavirus Pool 1 “Eurasia” ELISA (IgG, IgM) based on recombinant nucleocapsid antigens of HTNV, DOBV and PUUV; Anti-Hantavirus Pool 2 “America” ELISA (IgG, IgM) based on recombinant nucleocapsid antigens of ANDV and SNV; anti-BCCV ELISA (IgG, IgM) as described above.
For multiparametric IFA-based detection of hantavirus-specific IgG and IgM, the Hantavirus Mosaic 1 and 3 (Euroimmun, Luebeck, Germany; Fig. 1) were used. These assays are CE-marked and validated according to Directive 98/79/EC on in vitro diagnostic medical devices, fulfilling the requirements for standardized and reproducible analyses. The IFAs are based on millimeter-sized fragments of glass slides (biochips) glued side by side on the reaction fields of a microscope slide. Biochips were coated with hantavirus-infected EU14 cells, followed by acetone fixation and gamma irradiation.
For standardized testing, the TITERPLANE Technique (Euroimmun) was applied at room temperature according to the manufacturer's instructions. In brief, serial tenfold dilutions (1∶10 to 1∶10,000) of blinded sera were prepared in sample buffer (Euroimmun). For class IgM antibody determination, serum IgG and rheumatoid factors were first preabsorbed by diluting sera 1∶10 in EUROSORB (Euroimmun), mixing thoroughly and incubating for 15 min. After centrifugation (5 min, 2,000 rpm) the supernatant was diluted serially as described above. Samples were applied to the reaction fields of a reagent tray. Mosaic-containing slides were placed into the corresponding recesses of the reagent tray, where all substrates came into contact with the fluids, and the individual reactions commenced simultaneously. After incubation for 30 min, slides were rinsed with a flush of PBS-Tween (PBS containing 0.2% Tween-20) and immersed in PBS-Tween for 5 min. For detection of bound antibodies, slides were placed on reagent trays prepared with fluorescein isothiocyanate (FITC)-conjugated goat anti-human immunoglobulin. To test for IgG antibodies, the respective reaction fields were loaded with the anti-human-IgG FITC conjugate. Accordingly, the anti-human-IgM FITC conjugate was applied to those reaction fields intended for the detection of IgM. Following a 30-min incubation, slides were washed as described above, embedded with mounting medium, coverslipped and evaluated by fluorescence microscopy. Evaluation was performed independently by at least two experienced laboratory experts without reference to the clinical diagnosis and serological precharacterization data. Positive reactions were characterized by a fine- to coarse-granular immunofluorescence (IF) in the cytoplasm of infected cells. Intensities of specific IF were compared to those of hantavirus-negative and -positive reference sera and scored as negative, weak, moderate or strong. Antibody titers were determined based on the 10-fold dilution series, allowing for assumed intermediate titers (corresponding to a theoretical dilution factor of 3.2). Samples with at least a weak specific IF at a dilution of 1∶100 (cut-off) were considered positive. The reciprocal endpoint titer was defined as the highest sample dilution factor for which a weak specific IF was detected. For example, if a serum showed a strong IF at a dilution of 1∶10 and 1∶100, a moderate IF at 1∶1,000 and a negative IF at 1∶10,000, it was assigned a reciprocal endpoint titer of 3,200. If another serum showed a strong IF at 1∶10 and 1∶100, but only a weak IF at 1∶1,000 and negative IF at 1∶10,000, the reciprocal endpoint titer was 1,000. The groups' reciprocal geometric mean titers (rGMT) were determined using Excel (Microsoft, Redmond, USA). In a group of n samples, the rGMT was calculated as the nth root of the product of the samples' reciprocal endpoint titers.
The overall qualitative performance of multiparametric IFAs in detecting anti-hantavirus antibodies was analyzed by considering those samples as seropositive that showed specific reactivity (cut-off 1∶100) against at least one of the different hantavirus serotypes contained in the biochip mosaics.
As shown in Table 2, the overall agreement between the reference tests and multiparametric IFAs was 96% for IgG and 93% for IgM analysis in the patient sera. Multiparametric IFA-based combined IgG and IgM analyses revealed 100% sensitivity for all serum panels, except for panel IV (96%). With respect to the control cohort, none of the healthy blood donors was antibody positive by the reference tests, whereas 2/89 (2%) were IgG positive for HTNV by IFA (98% specificity).
In detail, among a total of 184 hantavirus-positive sera, 177 (96%) and 183 (99%) were IgG positive in the reference tests and IFA, respectively. All samples of panels I, II and III were anti-hantavirus IgG positive by precharacterization and multiparametric IFA. In panel IV, 20/23 (87%) Canadian HCPS patients had hantavirus-specific IgG according to the ELISA-based precharacterization, while the multiparametric IFA revealed a higher positivity rate of 96% (22/23). Discrepant IgG results were found in 4/23 (17%) samples of this panel, including 3 ELISA IgG negative/IFA IgG positive sera that were obtained during acute SNV infection and were confirmed by RT-PCR (Table 3; sera #CA-22, #CA-23 and #CA-24). Another serum (#CA-12) was ELISA IgG positive/IFA IgG negative. In panel V, the IgG IFA achieved a higher seropositivity rate (52/52, 100%) than the reference test (48/52, 92%). The 4 sera of this panel that were ELISA IgG negative/IFA IgG positive had been drawn during acute infection, i.e. within 3 to 5 days after onset of initial symptoms (Table 3; sera #2B, #13, #18 and #19).
According to the reference tests, 131 (72%) patient sera were IgM positive, referring to a total of 183 sera with IgM precharacterization. Using IFA, 131 (71%) out of all 184 patient samples tested IgM positive. Comparing the performance within each serum panel, IgM positivity rates were equal in panel II (reference test/IFA, 100%/100%), but different in panel I (54%/56%), panel III (83%/71%), panel IV (74%/83%) and panel V (100%/92%). Discordant results were obtained for 13 (7%) of the patient sera, for 7 of which IgM-positivity by precharacterization contrasted with IgM-negativity by IFA. However, 5 of them had been drawn in the convalescent phase of the disease (Table 3; panel III, #RO-7; panel V, #1A, #3A, #3B and #6B). Serum #6B derived from a follow-up patient whose first serum sample (#6A), drawn 16 days after onset of symptoms, was IgM positive in the IFA. The remaining 6 sera with discordant IgM results were negative in the reference tests but positive in the multiparametric IFA (Table 3).
When restricting the evaluation to only the (endemic) serotype-specific substrate, a subset of IFA positive sera (5/184, 3%) was either IgM or IgG negative, indicating the possibility of misdiagnosis when serological screening is limited to the suspected (endemic) serotype. Among these 5 cases, 2 sera from PUUV-infected patients (panel I) showed IgM reactivity on the HTNV substrate only, 2 sera from Canadian SNV-infected patients (panel IV) showed IgG/IgM reactivity on the PUUV substrate only, while the remaining serum from a SEOV-infected Chinese patient (panel II) showed IgM reactivity to HTNV only.
Depending on the causative hantavirus, cross-reactivity rates of up to 100% were observed when comparing reactivity rates between the seven serotypes used as IFA substrates (Fig. 2 A and C). In general, reciprocal geometric mean titers (rGMTs) of anti-hantavirus IgM were lower than IgG rGMTs (Fig. 2 B and D).
In panel I, all (100%) IFA IgG positive and 96% of the IFA IgM positive sera reacted on PUUV-infected cells (IC), with rGMTs of 2,530 (IgG) and 347 (IgM). Positivity rates of serum IgG/IgM from PUUV-infected patients were also high on SNV-IC (86%/75%) and HTNV-IC (79%/49%), but only moderate (IgG, 53–57%) or low (IgM, 6–15%) on SEOV-, SAAV- and DOBV-IC.
Sera from SEOV-infected patients (panel II) were IgG positive on all tested substrates, while serum IgM revealed highest reactivity (100%) on HTNV-IC. On SEOV- and HTNV-IC, rGMTs of IgG were identical (10,000) or higher compared to those on the other substrates (<4,000); rGMTs of IgM were highest on HTNV-IC (251), followed by SEOV- and SAAV-IC (158).
In accordance with the phylogenetic relatedness of hantaviruses, the highest positivity rates of sera from DOBV-positive patients (panel III) were found on DOBV-, SAAV-, HTNV- and SEOV-IC (IgG/IgM, 100%/60–100%), and markedly lower on PUUV- and SNV-IC (≤43%). The rGMTs of IgG and IgM were highest on HTNV-IC (22,952) and DOBV-IC (1,005), respectively.
Among all hantavirus-infected patients, SNV infections (panel IV) were associated with the lowest serotype-specific rGMTs of 198 (IgG) and 138 (IgM). Serum IgG/IgM from SNV-infected patients reacted on SNV-IC (95%/95%), PUUV-IC (55%/42%) and HTNV-IC (5%/5%), whereas SEOV-, SAAV- and DOBV-IC were negative. These samples were not available for testing on ANDV-IC, but considering the antigenic relatedness of SNV and ANDV, a significant degree of cross-reactivity can be expected, similar to the results obtained for ANDV-infected patients' sera on SNV-IC.
For ANDV-positive sera (panel V), the highest reactivity rates (100%) and rGMTs (IgG/IgM, 13,999/5,389) were detected on ANDV-IC. Slightly less reactivity (98%/83%) and lower rGMTs (1,166/976) were observed on SNV-IC.
Accordingly, serotyping by endpoint titration was successful in the majority of ANDV-infected patients (Fig. 3, panel V), when IgG titers were evaluated separately (77%) or in conjunction with IgM (96%). Regarding panel I, PUUV could be serotyped in 87% and 91% of patients by IgG antibody titration and by combined IgG and IgM analysis, respectively. In panel II and III, a clear serotype could be determined in only a minority of cases due to the high cross-reactivity rates. In panel IV, IgG plus IgM analysis revealed a clear serotype in 58% of the patients; the remaining sera reacted equally on SNV- and PUUV-IC, but SNV could be assigned as the causative agent due to its distribution in North America and the absence of PUUV on the American continent.
Due to the nonspecific clinical symptoms associated with the majority of hantavirus infections, confirmatory laboratory diagnosis is crucial. Generally, acute hantavirus infections are diagnosed serologically by determination of an at least four-fold increase in the IgG titer in consecutive serum samples and/or detection of specific IgM. In-house enzyme-linked immunosorbent assays (ELISAs) and indirect immunofluorescence assays (IFAs) based on antigen from a single or two hantavirus serotypes are widely used for this purpose [28], although these assays may not allow standardized testing of consistently high quality. Furthermore, considering the fact that hantavirus serotypes co-circulate in parts of Europe (PUUV, DOBV and SAAV; [29], [30]), Russia (SAAV, PUUV, DOBV and HTNV; [31]) and Asia (HTNV, SEOV and PUUV; [32], [33]), monospecific tests may fail to detect the causative agent of hantavirus infection, despite the high cross-reactivity rates between closely related hantaviruses [34]. Between the different hantaviruses there are extensive antigenic/serological cross-reactivities that closely follow the phylogenic tree [20]. Thus, the cross-reactivities are especially strong within each group (Murinae-borne, Arvicolinae-borne, Sigmodontinae-borne, Neotominae-borne).
In this study the diagnostic performance of commercial multiparametric hantavirus IFAs based on mosaics of biochips coated with seven different hantavirus-infected cell substrates was evaluated and compared with results from previous laboratory testing using in-house or commercial assays. Among the reference tests were ELISAs using recombinant hantavirus nucleoprotein (rN) as antigen substrate. The N protein represents the major hantavirus antigen and induces an early, strong and long-lasting antibody response [35]–[38]. Recombinant N protein-based assays have been reported to show high sensitivity for hantavirus-specific IgG and IgM [39], [40].
To evaluate the sensitivity and specificity of multiparametric hantavirus IFAs, five laboratory-confirmed hantavirus-positive serum panels obtained from six diagnostic centers located in different geographic regions were used. Multiparametric IFA-based determination of hantavirus-specific IgM and IgG yielded an excellent diagnostic sensitivity of 100% for all panels, except for panel IV (96%). The IFA total seropositivity rate for IgG detection (99%) exceeded that of the reference tests (96%), with an overall agreement of 96%. With respect to the control group, none of the 25 Canadian but 2/64 German healthy blood donors tested IgG positive on the HTNV-IC at the cut-off dilution of 1∶100. These two samples tested negative on the SEOV-, DOBV- and SAAV-IC IFA substrates. Both an anti-hantavirus ELISA and an anti-hantavirus lineblot based on nucleocapsid antigen from PUUV, DOBV, HTNV, SEOV, SNV and ANDV (Euroimmun) were negative, too. Therefore unspecific (false-positive) reactions cannot be ruled out. In the absence of clinical symptoms and without travel history to HTNV endemic regions, such borderline and isolated reactivities on HTNV-IC should be considered as unspecific.
The seropositivity rate of IgM detection by IFA (71%) was almost the same as by the reference methods (72%). However, there was only 93% agreement between the methods, and 13 sera showed discordant IgM results. Among these discordant cases, 7 serum samples had IgM positive reference data, but tested IgM negative by IFA. Five of these sera were drawn in the convalescent phase of a hantavirus infection, namely 1, 7, 11 and 13 months (ANDV-infected patients) and more than 2 months (DOBV-infected patient) after the onset of initial symptoms. Consequently, four of these probably contained persisting IgM antibodies against hantaviral rN protein, since hantavirus-specific IgM usually disappears two to three months after the onset of symptoms [41]. Persistence of IgM antibodies against hantaviral rN protein for as long as two to three years after hospitalization has been reported previously in DOBV-infected patients [42].
With respect to hantavirus-specific IgG, persistence over many years or even life-long may occur, and the IgG response can be delayed in some patients. In the multiparametric IFA, none of the 184 samples was isolated IgM-positive, whereas ELISA-based analysis of sera from SNV-infected patients revealed 3/23 (13%) isolated IgM-positive results. This finding corresponds with recent studies, demonstrating that SNV-specific IgM occurs early after infection, whereas anti-SNV IgG is not detectable in a sizable proportion of sera drawn in the early acute phase [34], [43]. Furthermore, in four sera obtained from ANDV-infected patients 3 to 5 days after onset of initial symptoms, isolated IgM was detected by the ANDV rN protein-based reference ELISA [25]. These discordant IFA/ELISA IgG results could be explained by an earlier appearance of IgG antibodies against hantavirus glycoprotein Gn (formerly termed G1) compared to anti-N protein antibodies as observed previously in acute phase sera [36]. Our data suggest that IgG seroconversion from negative to positive as well as IgM seroconversion from positive to negative is detected earlier by whole native antigen (presented in the IFA) than by recombinant N protein (presented in the ELISA). Considering the need for IgM confirmation by IgG seropositivity, isolated IgM results involve sampling and analysis of at least one consecutive serum sample. Regarding the higher IFA IgG sensitivity, heterogeneous antigen seems to be at least as suited as homogenous antigen to screening for anti-hantavirus IgG in patients suspected of having hantavirus infection.
Notably, multiparametric IFA analysis improved the diagnostic sensitivity, since three samples precharacterized as positive for anti-SEOV IgM and anti-SNV IgM or IgG were found to be only positive on cells infected with closely related hantaviruses (HTNV and PUUV, respectively). This reflects the cross-reacting ability of hantavirus-specific antibodies, which is particularly strong for antibodies against the highly conserved N protein [44]–[49]. As a consequence, serological identification of the causative hantavirus in areas with co-existing serotypes is difficult. Reliable serotyping is particularly important, because severity of syndromes depends on the causative hantavirus serotype [9]. The gold standard for hantavirus serotyping is the neutralization test [50]–[52], which is most reliable but laborious, time-consuming and expensive and has to be performed in a containment laboratory (BSL-3). Serotyping ELISA based on truncated N-proteins have been developed [53]–[55], but can be used as second line diagnostics only, due to a reduced sensitivity. In our study, serotyping by IgG in conjunction with IgM IFA analysis was successful in the majority of HCPS patients infected with ANDV (96%) and HFRS patients infected with PUUV (91%), representing a fast and simple alternative to more elaborate methods. Serotyping failed in patients infected with murinae-borne Old World hantaviruses (DOBV and SEOV), because of their close phylogenetic relatedness with HTNV and SAAV: DOBV N protein has an amino acid sequence identity of 99%, 83% and 80% with the N protein of SAAV, HTNV and SEOV, respectively. Here only neutralization tests or serotyping ELISA can reliably distinguish antibodies raised against these serotypes. However, with respect to the different geographical distribution of HTNV (predominantly South Korea/China/Russia) and DOBV (Balkans) [7], multiparametric IFA-based serotyping in combination with the patient's travel history and clinical characteristics is possible in many cases. For example, in the Romanian hantavirus-infected patients with severe HFRS and without travel history, an infection with HTNV could be excluded, revealing DOBV or SAAV as the causative agent. Infection with SAAV, circulating in Estonia, Finland, Germany, Hungary, Lithuania, Russia, Slovenia and Slovakia [56], could be further excluded because it is associated with milder symptoms.
The worldwide increasing number of hantavirus infection demonstrates the need for reliable serological tests which are simple to perform and allow detection of all clinically relevant hantaviruses. Many hantavirus-infected patients are still misdiagnosed [57]–[61], often due to the lack of generally available standardized assays and of epidemiological data. In line with the most recent European external quality assurance study for hantavirus diagnosis [62], the present study revealed similar performance of IFA and ELISA/EIA. In contrast to the homogenous antigen-presenting ELISA/EIA, the mosaic-based IFA evaluated in this study provides multiparametric testing by combining different substrates of cells infected with clinically relevant Old and New World hantaviruses. Furthermore, unlike in-house IFAs, this commercial assay does not depend on cell culture (establishment of infected cells) and propagation of hantaviruses, since large batches of identical infected cells were created and stored in liquid nitrogen, allowing standardization of immunological analyses. This makes hantavirus testing more widely available to all laboratories familiar with IFA-based diagnostics. In conclusion, analysis of hantavirus-specific IgG and IgM by indirect immunofluorescence on substrate mosaics consisting of cells infected with different hantaviruses is a globally applicable and reliable diagnostic tool for screening of patients suspected of having hantavirus infection, and can be useful for serotyping in areas where hantaviruses of different serogroups are endemic.
|
10.1371/journal.pcbi.1000795 | Optimal Workloop Energetics of Muscle-Actuated Systems: An Impedance Matching View | Integrative approaches to studying the coupled dynamics of skeletal muscles with their loads while under neural control have focused largely on questions pertaining to the postural and dynamical stability of animals and humans. Prior studies have focused on how the central nervous system actively modulates muscle mechanical impedance to generate and stabilize motion and posture. However, the question of whether muscle impedance properties can be neurally modulated to create favorable mechanical energetics, particularly in the context of periodic tasks, remains open. Through muscle stiffness tuning, we hypothesize that a pair of antagonist muscles acting against a common load may produce significantly more power synergistically than individually when impedance matching conditions are met between muscle and load. Since neurally modulated muscle stiffness contributes to the coupled muscle-load stiffness, we further anticipate that power-optimal oscillation frequencies will occur at frequencies greater than the natural frequency of the load. These hypotheses were evaluated computationally by applying optimal control methods to a bilinear muscle model, and also evaluated through in vitro measurements on frog Plantaris longus muscles acting individually and in pairs upon a mass-spring-damper load. We find a 7-fold increase in mechanical power when antagonist muscles act synergistically compared to individually at a frequency higher than the load natural frequency. These observed behaviors are interpreted in the context of resonance tuning and the engineering notion of impedance matching. These findings suggest that the central nervous system can adopt strategies to harness inherent muscle impedance in relation to external loads to attain favorable mechanical energetics.
| Movement in organisms is a result of the interplay between biomechanics, neural control, and the influence of external environmental loads. Understanding the interaction between these factors is important not only for scientific reasons but also for engineering robotic systems and prostheses that strive to match biological performance. Muscle mechanical impedance is key in defining the mechanical interaction between muscles and their loads. It is well known that neural activation modulates muscle impedance, particularly stiffness, and that such modulation can be used advantageously to stabilize the posture and motion in organisms. Here, we show computationally and experimentally that stiffness modulation can also be used to enhance the capability of muscle to generate mechanical power, which is key in determining how fast animals can run, fly, swim, or jump. When muscles are activated optimally in relation to their external loads, they can create resonance conditions at optimal frequencies that significantly enhance their mechanical energetics by up to 7-fold. These findings can be interpreted in the context of the engineering notions of impedance matching and resonance tuning, which are commonly used as guiding principles in the design of diverse power optimal systems, such as communication circuits and robotic systems.
| The capability of skeletal muscles to deliver mechanical power is key in determining the neuromechanical performance envelope of organisms. How fast and how far animals run, fly, swim, or jump is clearly limited by the mechanical power delivered by the muscle-tendon units to skeletal and environmental loads. Therefore, estimating the mechanical energetics of muscles (henceforth simply called energetics) has been of interest in diverse fields such as organismal biomechanics, biomimetic robotics and prosthetics [1]–[3].
Many factors influence the neuromechanical performance of organisms, including i) the dynamics and mechanical properties of muscle actuators, ii) skeletal mechanics, iii) neural control and iv) influence of loads external to the organism. Integrative approaches have been proposed to capture the interaction of all, or subsets of these factors. For example, the connection between muscle impedance (particularly stiffness) and neural control has been studied in depth with respect to postural and dynamic stability [4], [5], locomotory functions [6]–[9], manipulation [10], [11], and other biomechanical tasks [12]. In this work, we adhere to the definition of muscle mechanical impedance as the “static and dynamic relation between muscle force and imposed stretch” [4]. Muscle impedance encompasses muscle stiffness, which is the static relation between muscle force stretch only.
In the context of muscle energetics, most investigations focused on experimentally measuring the power output of individual muscles at a range of frequencies, phases and electrical stimulation parameters, and finding maximal power generating capability of muscles under prescribed motion trajectories. However, the role of muscle-load interaction on output energetics has not been formalized. The central premise of this work is that the mechanical energetics of a muscle-actuated system cannot be determined in a meaningful manner without considering the coupling of muscle properties, load dynamics and neural activation. By considering this coupling explicitly, we arrive at phenomena that cannot be captured using standard workloop testing methodologies, including the opportunity to harness muscle-load interaction in an energetically advantageous manner.
Muscle energetics have been characterized under dynamic conditions, both in vitro [13] and in vivo [9], [14], [15]. In vitro measurements relied almost invariably on the workloop technique [16]. In this approach, isolated muscles are subjected to predetermined periodic length variations in time (typically sinusoidal, but not always [17]) by means of an external motion source. At a given phase of the imposed oscillation, an electrical stimulus is delivered synchronously, resulting in periodic muscle contractions. A plot of muscle contractile force versus displacement results in a cyclic workloop, with the integrated area within the loop being a measure of the net muscle work done. These and similar measurements have been reproduced in the muscle physiology literature for various muscle groups within various organisms [18]–[21], and connections between the muscle function and its mechanical energetics have been made [22]–[24]. While such measurements provide useful energetic connections with muscle function, the experimental conditions do not capture representative in vivo conditions because motion profiles are imposed on single isolated muscles with no muscle-load interactions [25], and without incorporating the effects of antagonist activity. In vivo measurements, on the other hand, capture all of the above effects in principle, but lack the experimental flexibility of varying load conditions in an unambiguous manner.
Capturing the effect of muscle-load interaction on muscle energetics is critical. This interaction can be captured by considering the impedance of the muscles in relation to the impedance of the load. When a group of muscles acts on a common load, as exemplified by an antagonist pair acting on a common load, each muscle forms part of the load borne by the other muscles in its group. Because muscle impedance is activation dependent, neural control can be used to modulate the effective load observed by each muscle by modulating the impedance of the opposing muscles, thereby offering the opportunity to create favorable impedance conditions that maximize power transfer to the external environmental load. This is akin to the notion of impedance matching in engineering systems, where the driving source and the load are “matched” to provide optimal power transfer. In the context of neuromuscular control, impedance matching can enable groups of muscles to work synergistically to provide significantly higher energetics than the sum of individual muscles.
Consequently, in this investigation we studied the influence of muscle-load interaction on muscle workloop energetics both computationally and experimentally. We set up a model problem consisting of a mass-spring-damper system actuated by either a single muscle (Figure 1B), or a pair of symmetric, antagonist muscles (Figure 1D). The input to the system (either neural control or electrical stimulation) can modulate the net force exerted by the two muscles as well as the net impedance. In the context of this problem, we investigated two hypothesis. Hypothesis 1 states that the power optimal oscillation frequency of a muscle actuated system is greater than the resonance frequency of the load. This is in direct contrast to an impedance-free actuator (such as an ideal electric motor) where the optimal oscillation frequency occurs exactly at the resonance of the load. Hypothesis 2 states that a pair of antagonist muscles can work together to produce more power synergistically than individually by margins that cannot be predicted without explicit incorporation of muscle impedance. We tested these hypotheses both computationally and experimentally. Our computational approach relied on optimal control solutions to the workloop maximization problem, which was based on a mathematical model of the problem. The experimental approach relied on in vitro measurements of workloop energetics of electrically-stimulated, frog muscle acting against emulated mass-spring-damper loads.
To investigate the role of muscle-load interaction and muscle impedance on output energetics, a mathematical model of the problem was developed. This model formed the basis for the ensuing optimization of workloop energetics. We modeled the case of Figure 1D. Note that the case of Figure 1B is a special case of the problem considered with the coefficients of the antagonist muscle set to zero. The key ingredient is a muscle model that captures activation and impedance characteristics of the muscle.
The model of Equation (6) was treated as the basis for our analysis. Since our objective is to analyze optimal muscle workloop energetics, we maximize the average power transfer from the muscles to the load integrated over one periodic cycle. The instantaneous power delivered to the load is given by . The cyclic work done by the muscles on the load is the integral of the power over one complete cycle. Therefore the control inputs, , that characterize power-optimal oscillations are given by the solution of the following optimization problem:(7)where is defined in Equation (6) and is the control input vector. In this formulation, we assumed that the terminal time was given and defined by the objective task. Therefore, to optimize power at oscillations of frequency [Hz], we set the solution time horizon [sec].
To derive necessary conditions for the optimal solution of Problem (7), we applied the Pontryagin Minimum Principle [29]. We followed the following procedure:
Details of this derivation, and the numerical methods employed therein are described as follows. The integrand of the Lagrangian cost function is given byWe augment the dynamical constraints to the cost function, and define the Hamiltonian scalar functionFrom the Pontryagin principle [29], the evolution of the optimal co-state variables at the optimal solutions are governed by:
The optimal control is given bywhere the last equality follows since is not a function of in this particular context. Substituting in Equation (6), we getwhich implieswhere and are upper and lower bounds, respectively, on the control inputs. Depending on the signs of the switching functions and , the control assumes either the values or . This is a bang-bang control solution, and is an expected outcome in such power-optimal (or maximum acceleration) problems [30]. Mathematically, such solutions appear when the Hamiltonian is a linear function in the control , as is the case in this problem. In the absence of limits on the control, the optimization problem would be unbounded, implying that the muscles that can generate unbounded forces will add infinite power to the load. Therefore, for the optimization problem to be mathematically well-posed, upper and lower bounds on the control inputs and are necessary.
In summary, the first order necessary conditions for power-optimal solutions are given by:(8)(9)(10)with cyclic boundary conditions:(11)(12)
Equations (8) and (9) define a two-point boundary value problem (2-point BVP) that is subject to the cyclic boundary conditions (11) and (12) and control constraints (10). This 2-point BVP was solved to give the optimal state trajectory (), the optimal control inputs , and the multipliers () associated with the power optimal solution. Methods for solving this problem numerically are detailed in the supporting material Text S1.
The optimal control problem (Problem (7)) was solved for various values of the time horizon that characterized the oscillation frequencies of interest. An example solution is shown in Figure 2 for an oscillation frequency (5 Hz) that is greater than the load resonance frequency ( Hz).
To investigate Hypothesis 1 computationally, successive optimizations similar to those of Figure 2 were conducted as the oscillation frequency was swept across the range of interest, and comparisons between optimal power generated by the bilinear muscle model and the optimal power generated by an impedance free actuator were drawn. As shown in Figure 3A, in the case of the system with = 2 Hz, the peak power was generated at = 2.4 Hz. In Figure 3B, in the case with = 4 Hz, the peak power was at = 4.8 Hz. This result is in direct contrast to the case when the load is driven by impedance-free actuators, where the optimal driving frequency is exactly equal to the resonance frequency of the load. The increase in optimal stimulation frequency is attributed to the contribution of active muscle stiffness to the net stiffness of the system (shown in the stiffness sub-plots of Figure 2), and thereby tuning the resonance of the combined muscle-load system.
To investigate Hypothesis 2 computationally, we compared the power output of the optimal solutions of the single-muscle case against the optimal solutions of the case of a muscle pair in Figure 4 across the frequency range of interest. The computed power-optimal responses show that synergistic activation of antagonist muscles may produce more cyclic work than individual muscle activation by a factor of more than two (Figure 4B). This is captured by the synergistic ratio , and is in direct contrast to constant impedance actuators where the ratio is exactly two. This model prediction implies that the energetics of individual muscles (obtained by zero-admittance workloop tests) cannot simply be summed to draw conclusions regarding the workloop energetics of the entire system.
Figures 3C and 3D show the results of experimental workloops with single muscles acting on mass-spring-damper loads. To test Hypothesis 1 experimentally, that the peak normalized power output was indeed at , measurements were conducted on two load cases with different natural frequencies (Hz and Hz). For both loads, we found that the normalized power measures and , with ( for all measurements). We attribute this increase in the optimal oscillation frequency over to the stiffness contribution of the muscles. This increase in optimal frequency over cannot be achieved via an impedance free force source, and can therefore be directly attributed to the increase in muscle stiffness due to the activation profile over the course of a full cycle.
Figure 5 shows the power output measurements of a pair of antagonist muscles acting synergistically compared to their power output acting individually. When the oscillation frequency was set to 3 Hz, the value of the energetic ratio was not statistically different from . However, when the oscillation frequency was set to 4 Hz, we found to be . The ratio was significantly greater than 2 (), showing that the energetics of the muscle pairs are greater than the sum of the energetics attained by individual activation. This is qualitatively compatible with the model predictions plotted in Figure 4 and is in support of Hypothesis 2. This implies the possibility that energetic synergies may be achieved by a muscle-actuated system to enhance their energetic performance at particular frequency ranges.
In the experimental measurements above, the absolute power value of the muscles, normalized by muscle mass, ranged between 17 [W/kg] and 81 [W/kg] at the optimal conditions.
The role of active and passive muscle impedance, particularly stiffness properties, has been studied intensively in the neuromechanics and motor control literature from the perspective of stability of posture and movement. The main focus of this work is to extend this literature to include the study of muscle mechanical energetics, particularly in the context of periodic motions. We focused on the representative problem of driving a mass-spring-damper by either a single muscle or a pair of antagonist muscles. This setup can be considered as an idealization of a single degree-of-freedom joint.
One consequence of explicitly accounting for muscle-load interaction is the increase in the optimal stimulation frequency of the coupled system relative to the natural frequency of the uncoupled load. This is captured by Figure 3 where the maximal power was generated at a frequency higher than the uncoupled natural frequency of the load, which directly supports Hypothesis 1. This is shown computationally (Figure 3A & 3B) where it is possible to scan the range of oscillation frequencies systematically to search for the frequency of peak power generations, and also experimentally (Figures 3C & 3D) where it is possible to do so only at select frequencies chosen to show the location of peak power. The increase in optimal power generation frequency is not an unexpected result since the stiffness contributions of the muscles should couple in with the overall frequency of the load. What this enables, however, is that resonance conditions can be tuned relative to the desired frequency of oscillation via an appropriate muscle activation pattern.
Taken to the limit of zero load stiffness, we conjecture that this feature potentially enables creating resonance conditions out of non-resonant loads. The biomechanics of natural loads in many biological systems are non-resonating. Consider, as an example, the motion of a swimming fish. The external restoring force on a fish's body is negligible, therefore the sideways bending dynamics can be considered non-resonant. In the presence of muscle activation, however, significant activation modulate stiffness is added to the system, which can be tuned to the desired oscillatory frequency of the undulating motion. The importance of body bending stiffness in relation to the undulating frequency and speed of swimming fish has been reported in [32], [33].
Another consequence of the coupling between muscle impedance and load dynamics pertains to energetic synergies that are observed in systems driven by multiple muscle systems. When multiple muscles act jointly on a common load, each muscle contributes to the effective load observed by the other muscles acting on that load. This contribution can be strongly modulated by the neural input to the muscles.
Taking the simplest case of two antagonist muscles acting in parallel on a common load, Figure 5 shows that a pair of muscles can generate more power on a common load than the sum of them acting individually. The margins of collaboration were much higher than those theoretically predicted with impedance-free actuators. For a pair of identical impedance-free actuators, the ratio is exactly 2 at all frequencies of oscillation. When one impedance-free actuator is capable of producing more force than the other, the ratio ranges between 1 and 2, but never exceeds 2. The maximal value of 2 is achieved if the two muscles provide equal forces, and the minimal value of 1 is approached as the relative contributions of the two muscles vary widely. Ratios greater than 2, as demonstrated in the 4 Hz oscillation case (shown in Figure 5C), and as demonstrated in the maximal values of Figure 4B, are in direct support of Hypothesis 2, and can only be achieved if additional muscle properties are introduced, such as activation dependent impedance.
Our findings may be interpreted in the context of the engineering notion of impedance matching. In engineering systems, impedance matching plays an essential role when it is desired to maximize power transfer between two dynamical systems. When a power source is connected in series with a load (in a Thevenin equivalent connection), maximal power transfer occurs when the internal impedance of the source is equal to the complex conjugate of the load impedance [34]. In a similar manner, neural activation of muscle modulates its stiffness to allow matching of muscle mechanical impedance to that of the load. When such a condition occurs, the power transfer is maximized. This implies that the mechanical work achieved by a single muscle is highly affected by the activation pattern of antagonist muscles, because such antagonist muscles form part of the load on the agonist muscle, and therefore the energetics of muscle-actuated systems must be considered holistically.
The impedance of a linear mass-spring-damper load () is the transfer function relating the velocity () and force () applied on the load, and can be expressed aswhere , , and are the mass, damper and spring coefficients of the load. Assuming that the source is primarily dominated by stiffness terms, as is the case of a bilinear muscle model, the impedance of the source () is:Therefore, for this source impedance, which is purely reactive, we do not have the ability to arbitrarily change the phase. To maximize the power transfer from the source to the load, impedance matching conditions require that the reactive part of the source impedance is negative the reactive part of the load impedance [34]. ThereforeUnder such conditions, the total system natural frequency becomeswhich implies that the source stiffness is chosen so that the natural frequency of the system matches the desired oscillation frequency .
Therefore, as the muscle pair modulates net stiffness , to a value that matches the desired load impedance, energetic advantages can be attained. Clearly there are limitations to the efficacy of impedance matching in helping maximize workloop energetics. For a pair of antagonist muscles to tune their stiffness to match the reactive impedance component of the load, certain amounts of co-contraction may be required. This was observed computationally with the time overlap of the control signals ( and ). While co-contraction may attain the desired frequency tuning, it will decrease the peak-to-peak net forces produced by the muscle pair. Beyond a certain break-even point, the peak-to-peak forces will be greatly diminished to the point that impedance matching becomes non-optimal.
Research in organismal motor control and biomechanics has reported extensively on the modulation of stiffness in limbs to enhance postural and dynamic stability. Our findings here provide further motivation to hypothesize that the central nervous system may utilize impedance matching as a means to enhance energetics against external loads. Prior studies support the notion that muscle stiffness is modulated to attain resonance tuning, though none have made an explicit energetic connection. Most of these investigations have focused on arm movements. In the context of rhythmic movements, perhaps the clearest evidence was provided in [35], where forearm stiffness was found to increase quadratically with oscillation frequency, and that the stiffness was minimal at the resonance of the load. It was shown that by increasing the oscillation frequency above the load resonance, the arm stiffness increased in a manner that created resonance of the arm-load system. In other studies [36]–[38], surface EMG measurements in horizontal arm reaching movements have shown that the overall co-contraction levels increase with increasing frequency of oscillation, and that co-activation increases with the square of frequency. Furthermore, in [39], neuromuscular models of the forearm that predict qualitative resonance tuning behavior in rhythmic oscillations were proposed. These arguments have also been extended to the context of of non-rhythmic movements by comparing the average forearm stiffness during reaching tasks with the fundamental frequency content of these movements [40].
The degree to which impedance matching is utilized by organisms specifically for energetic purposes remains to be addressed in future studies. Using antagonist activation of variable impedance actuators can enable the central nervous systems to learn optimal impedances that, when coupled with external loads, can provide higher energetics. Viewed from this perspective, activation dependent muscle impedance may be regarded as a favorable biomechanical property. Furthermore, this postulates that the mechanical energetics of individual muscles cannot be directly summed to estimate the total energetics of a multiple-muscle system.
|
10.1371/journal.pcbi.0040041 | Control of Cation Permeation through the Nicotinic Receptor Channel | We used molecular dynamics (MD) simulations to explore the transport of single cations through the channel of the muscle nicotinic acetylcholine receptor (nAChR). Four MD simulations of 16 ns were performed at physiological and hyperpolarized membrane potentials, with and without restraints of the structure, but all without bound agonist. With the structure unrestrained and a potential of −100 mV, one cation traversed the channel during a transient period of channel hydration; at −200 mV, the channel was continuously hydrated and two cations traversed the channel. With the structure restrained, however, cations did not traverse the channel at either membrane potential, even though the channel was continuously hydrated. The overall results show that cation selective transport through the nAChR channel is governed by electrostatic interactions to achieve charge selectivity, but ion translocation relies on channel hydration, facilitated by a trans-membrane field, coupled with dynamic fluctuations of the channel structure.
| Communication between a cell and its environment relies on channel-forming proteins to provide a low energy pathway for ions to move in and out. Although channel-forming proteins are essential to all life forms, the atomic-scale mechanisms that enable ions to pass through the channel remain elusive due to the lack of experimental approaches to monitor the protein and ion in real time and at atomic resolution. A powerful alternative approach is molecular dynamics (MD) simulation based on the laws of physics applied to the increasing body of protein structures resolved at atomic resolution. Here we present all-atom MD simulations applied to the nicotinic acetylcholine receptor (nAChR) that initiates voluntary movement in skeletal muscle. By focusing on individual permeant cations, we find that selective cation translocation occurs in stages: cations are first selected through a series of oppositely charged residues within the protein vestibule leading to a narrow hydrophobic constriction, but then hydration of the narrow region and dynamic fluctuations of the protein enable the cation to pass through. The findings provide a general framework for understanding how ions are selected for transport based on charge, and how the dynamic interplay between water, the ion, and the channel protein enable rapid ion translocation through the broad class of channel-forming proteins with hydrophobic barriers.
| Channel proteins circumvent the enormous energetic barrier to ion transport imposed by the cell membrane and are essential to all life forms. When a channel activates, permeant ions flow passively down their electrochemical gradients, changing the membrane potential and allowing communication between extra- and intra-cellular environments. The selectivity of ion transport depends on the type of channel, which can be non-selective, charge-selective or ion-selective, suggesting a diversity of mechanisms underlying transport. Present day atomic structures of ion channels allow unprecedented studies of ion transport using computational approaches. Here we use all-atom molecular dynamics (MD) simulations to study transport of single cations through the charge-selective channel of the nicotinic acetylcholine receptor (nAChR) from the motor endplate.
The nAChR is a hetero-pentamer of 250 kD, and contains an intrinsic channel pore that triggers flow of cations in response to nerve-released ACh. The channel selects for cations, mainly according to size, and is formed by α-helices from the second transmembrane domains (TMD2) of each of the five subunits. Selectivity for cations is achieved by anionic residues located on either side of the channel mouth [1] and in fenestrated structures in the cytoplasmic domain [2], which stabilize cations and concentrate them relative to bulk solution. Hydrophobic residues extending from TMD2 line the narrow region of the nAChR channel [3], suggesting that the hydration shell around the ion is maintained as the ion passes through. The availability of a 4 Å resolution structural model of the Torpedo nAChR [4] has brought the atomic-scale mechanism of ion transport under intensive investigation [5,6].
MD simulations of a simplified channel, composed of only TMD2 domains embedded in a bilayer-mimetic slab, revealed that although water filled the channel along its entire length, ions did not enter [7]. In simulations containing the entire pore domain in an explicit lipid bilayer, water and ions were excluded from the narrow region of the channel [8]. However, manually widening the pore radius by 1.5 Å allowed penetration of both water and ions, giving an ion transport rate approaching that expected from the single channel current amplitude. Although neither of these studies included the effect of membrane potential, they concluded that the cryo-electron microscopic structure of the Torpedo nAChR is in the non-conducting, inactive state.
Analogous MD simulations have been conducted on the mechano-sensitive channel of small conductance (MscS), which like the nAChR, also contains a hydrophobic pore. The initial simulations suggested that the x-ray structure of MscS was in the non-conducting state because water was excluded from the narrow hydrophobic region during the majority of the simulation [9]. However, application of a membrane potential promoted water entry, increased the channel radius, and produced ion transport at rates approaching those measured experimentally [10–12]. These later simulations suggest that membrane potential facilitates entry of water, promoting ion transport, and that the x-ray structure of MscS is closer to a conducting than a non-conducting state.
Here, we generate a homology model of the nAChR found at the human adult motor endplate, embed it in explicit lipids and solvate it with water and ions. We apply a trans-membrane electric field to mimic the cell membrane potential, and perform all-atom MD simulations with and without restraints of the protein structure. Our results reveal transport of single cations, and show that the transport depends on hydration of the channel, facilitated by a trans-membrane field, coupled with dynamic fluctuations of the channel structure.
We first subjected the nAChR ensemble, including one copy of the receptor protein, membrane lipids, water, Na+ and Cl− ions, to a 10 ns equilibration step (see Methods). Then we applied a voltage bias equivalent to a physiological trans-membrane potential of −100 mV, and generated a further MD simulation of 16 ns. Starting at 8 ns following application of the voltage bias, a series of snapshots reveal a single cation traversing the hydrophobic TMD2 region of the channel (Figure 1). The snapshots depict the cation and the water-filled volume within TMD2, and show temporal oscillations of the water volume that thins and widens at varying positions along the channel axis, suggesting cation transport is associated with water translocation.
With the simulation trajectory in hand, we tracked the position of the cation along an axis through the center of the channel and plotted position as a function of time (Figure 2). The cation exhibits step-wise position changes, with dwells at each position varying in duration. In the extracellular region bordering TMD2, dwells at each step are prolonged, whereas dwells within TMD2 are transient. The position of the prolonged dwell time immediately apical to TMD2 (upper gray arrow) corresponds to the “outer” ring of negatively charged residues previously shown to affect unitary conductance of the channel [1]. Further in the apical direction, two more positions of stability are detected (black arrows), corresponding to rings of polar or negatively charged residues within the lumen of the extracellular domain; the contributions of these more apical residues to unitary channel conductance have not been examined experimentally. Within TMD2, several positions are occupied transiently, corresponding to rings of hydrophobic side chains that line the channel and form the barrier to ion transport.
Color-encoded changes in the effective pore radius along TMD2 are superimposed upon the time course of cation transport (Figure 2). Four horizontal stripes enriched in yellow and red colors indicate narrow regions corresponding to rings of hydrophobic side chains at the standard consensus positions 1′, 5′, 9′, and 13′ of TMD2 [13]. The pore radius corresponding to each of these rings oscillates during the simulation trajectory, but the radius at position 9′ shows much smaller oscillations. Until around 8 ns, the radius at the upper 13′ position is narrow, but it widens just before the cation approaches and it stays wide until the cation passes by. Concurrent with the widening at position 13′, the radius at position 1′ constricts even though the cation is far from this position. Next, the cation surpasses the central 9′ position with little change in the local pore radius, but it hops back and forth across position 5′ until the radius at position 1′ widens enough to allow complete translocation of the cation. Simultaneous with widening at position 1′, the more extracellular 13′ position constricts. Thus when the cation enters, the extracellular portion of TMD2 widens while the intracellular portion constricts, and when the cation exits the sequence of motions is reversed. The changes in channel radius reveal a back and forth tilting of TMD2 about an axis centered at the 9′ position that accounts for the peristaltic changes in water volume in Figure 1.
Next, we examined the cycles of emptying and filling of water within the channel by counting the number of water molecules in three contiguous zones, 6 Å thick, centered at positions 5′, 9′, and 13′ (Figure 3). At the beginning of the simulation, water occupancy is high throughout the channel, but 2 ns after application of the trans-membrane potential, water occupancy in the zones corresponding to positions 13′ and 9′ falls precipitously, creating a water-free volume known as a vacuum plug. However at 8 ns, just before the cation enters, water transiently refills these two zones for a 2 ns period that coincides with cation transport. Afterward, water diminishes in these two zones and no further cation transport occurs. Thus cation transport is associated with periods of channel hydration.
To examine the contribution of the trans-membrane potential to cation transport, we started with the same equilibrated system and generated another 16 ns simulation with a hyperpolarized potential of −200 mV. The resulting trajectory shows that three cations achieve stable positions in the extracellular vestibule bordering TMD2, and that two of them traverse the channel (Figure 4). As observed at −100 mV, each cation moves in discrete steps, exhibiting both prolonged and transient occupancies in the course of transport. Four positions of prolonged occupancies are observed in the extracellular vestibule, three of which were seen in the simulation at −100 mV. The fourth stable position corresponds to yet another ring of negatively charged residues that lines the extracellular lumen in the vicinity of the ligand-binding domain (black arrow). A fifth stable position on the intracellular side of TMD2 is also evident in this trajectory (gray arrows), corresponding to the intracellular rings of charged residues previously shown to contribute to unitary channel conductance [1]. Although the hyperpolarized potential leads to an overall widening along the length of the channel (less yellow and red color in Figure 4), the radius again shows reciprocal changes at positions 1′ and 13′, consistent with tilting of the TMD2 helices about an intervening axis.
The simulation at −200 mV also captures transport of two cations close together in time. While the first cation is traversing the channel, the second cation remains outside of TMD2, oscillating between two stable positions within the central vestibule of the ligand binding domain. Once the first cation exits the channel, the second cation enters. Meanwhile a third cation reaches the border of TMD2 only after the second cation enters the channel. Although the increase in transmembrane potential decreases the time between entry of successive cations, only a single cation at a time occupies the hydrophobic region between positions 1′ and 13′.
We noted an increase of 0.5 Å in the effective channel radius at −200 mV, corresponding to widening of the two zones centered at positions 13′ and 9′ (Figure 5). This relatively small change in radius is associated with substantial increases in water occupancy within these zones, and loss of the cycles of emptying and filling seen at −100 mV. In MD simulations of the MscS channel, similar increases in channel radius and water occupancy were observed when increasing transmembrane potentials were applied [10]. Our results for the nAChR and those for MscS suggest that water occupancy of the channel is the principal requirement for cation transport through a hydrophobic channel. However, the continuous hydration of the channel observed at −200 mV did not promote a commensurate increase in cation transport during the 16 ns simulation. Thus in addition to changes in channel radius and hydration, an additional kinetic limitation for cation translocation appears to be involved.
The changes in water volume and channel radius shown in Figures 1 and 2 suggest that dynamics of the protein may be the additional kinetic limitation for cation transport. To examine the contribution of protein dynamics, we performed two additional MD simulations at each of the two trans-membrane potentials, but with restraints of the protein structure. No cation translocation was observed at either potential. At −100 mV, one cation entered the channel and approached the narrowest constriction, but it retraced its path by the end of the simulation (Figure 6). Parallel measurements of water occupancy reveal full hydration of the channel in all three hydrophobic zones, and show no differences in hydration between the two potentials (Figure 7). Thus, in addition to increases in channel radius and hydration, dynamic fluctuations of the protein structure are required for cation transport. Our overall simulations show that cation selectivity of the nAChR is governed by multiple electrostatic interactions between the cation and charged residues flanking the channel, but that subsequent translocation relies on channel hydration, facilitated by the trans-membrane potential, and dynamic fluctuations of the channel structure.
Our simulations reveal single cation translocation events through the channel of the human muscle nicotinic AChR. The findings shed new light on the nature of cation selectivity and on how single cations are transported one at a time across the central hydrophobic barrier. Four rings of polar or negatively charged residues line the central vestibule of the extracellular domain and likely contribute to cation selectivity. Although the narrowest hydrophobic pore is wide enough to accommodate a sodium ion with a single hydration shell, ion transport requires both hydration of the channel and dynamic motions of the protein. Continuous hydration of the channel is achieved by increasing the trans-membrane potential [14], which leads to widening of the channel and loss of the cycles of water emptying and filling of the narrow hydrophobic region. However, restraining the protein structure prevents ion transport even though the channel is continuously hydrated and the narrowest opening remains large enough for a sodium ion with its hydration shell. Dynamic motions during ion transport are associated with a back and forth tilting of TMD2 about the 9′ position and the peristaltic increases in the water-filled volume, some of which are associated with cation translocation across the central hydrophobic barrier. Thus ion transport through the nAChR requires a coordinated interplay between dynamic fluctuations of the protein, water and the permeating cation.
Our findings suggest that rings of negatively charged residues on both sides of TMD2 stabilize permeant cations. The functional significance of the rings immediately flanking TMD2, αGlu262, αGlu241 and αAsp 238, was originally uncovered by monitoring changes in unitary channel conductance following site directed mutations [1], and give confidence in the capability of our simulations to define structural counterparts of cation selectivity. Our findings suggest three more rings of functionally analogous polar or negatively charged residues within the central vestibule of the ligand binding domain, corresponding to αAsn 47, αGlu 83 and αAsp 97, together with residues at equivalent positions in the non-α-subunits (Figure 8). A series of rings along the external vestibule seems well suited to selecting cations for transport, as once a cation passes the first ring, it is further stabilized by additional rings within the lumen, thus increasing the cation concentration compared to bulk solution. This interpretation can be readily tested by mutagenesis coupled with electrophysiological measurements of unitary channel conductance and ion selectivity, and by x-ray crystallography of isolated ligand binding domains in the presence of suitable ions.
Once the cation reaches TMD2, hydrophobic interactions dominate the transport process. The rings of non-polar side chains lining TMD2 form a narrow constriction analogous to the hydrophobic channel of MscS [10–12] and hydrophobic nanotubes [15–18]. As we observe in the nAChR, these systems exhibit increases in hydration with only slight increases in channel radius, and the cycles of water filling and emptying diminish with increasing trans-membrane potential, likely through increasing the degree of order of the water molecules. The collective observations suggest that the switch from an inactive to an active channel could occur with increases in channel radius of one Å or less. The non-polar side chains along TMD2 provide little stabilization for a hydrated cation, minimally slowing transport, and suggest a steric interplay between the rings of non-polar side chains and the cation may limit transport in this region. The non-polar side chains also create an environment of low electric permittivity, extending the distance range of inter-cation Coulombic repulsion, perhaps explaining why only a single cation at a time occupied TMD2.
Our findings indicate that dynamic fluctuations of the narrow hydrophobic region of the nAChR channel are crucial for cation transport. The effects of dynamic fluctuations likely originate from multiple sources. First is the effect of dynamics on hydration of the channel. We find that although the restrained protein allows continuous hydration of the channel and maintains an opening wide enough for a hydrated sodium ion, transport is prevented. Similarly in model hydrophobic pores, rigid walls promote water filling and flexible ones favor water vapor states [17]. Second, computational studies of gramicidin suggest protein dynamics affects mobility of water and ions in a narrow pore where the magnitude of thermal fluctuations is significant relative to pore size [19]. Third, protein dynamics can affect the free energy barrier to ion translocation [31]. The potential of mean force (PMF) for a pore composed of rigid TMD2 helices from the Torpedo nAChR exhibited a peak for a sodium ion of 9 kcal/mol [7]. A much smaller peak of 3 kcal/mol was obtained in PMF calculations based on dynamically fluctuating TMDs from a homology model of the α7 nAChR [20]. Although the two reports employed different models of the nAChR channel and different methods for calculating the PMF, the overall findings suggest protein dynamics reduces the energetic barrier for ion translocation. The effect of protein dynamics on the PMF profile can be further addressed by calculations applied to the same nAChR channel under restrained and unrestrained conditions.
A natural question arises of why cation translocation is observed at all in our unrestrained simulations. Our nAChR structural model is expected to be in the inactive state because the structure of the modeling template, the Torpedo nAChR at a resolution of 4 Å, was obtained in the absence of agonist [4]. However, the inactive state may have a low but non-zero cation permeability. Single channel recording techniques register step changes in current between inactive and active states, but cannot resolve single ion translocation events due to limitations in bandwidth and background noise. Thus a low frequency of single ion translocation events may be inherent to channels with hydrophobic pores. Because a unitary current amplitude of 5 pico-amperes amounts to 4 or 5 cations transferred per 100 ns, simulations well beyond 16 ns are required to establish the functional state of the channel. Future MD simulations spanning much longer times are thus of high importance.
By focusing on the permeating cation, the present simulations show that selective ion transport arises from electrostatic interactions between the cation and multiple charged side chains in the protein, combined with dynamic interactions between the hydrophobic constriction, the cation and water. Although the functional state of the nAChR in our simulations is undetermined, the combination of dynamic fluctuations of the protein and cycles of water emptying and filling, facilitated by the transmembrane potential, likely govern cation permeation in both active and inactive receptor states.
We used the comparative protein structural modeling program, MODELLER [21], to generate a homology model of the adult human nAChR using the Torpedo structural model [4] as the template. The modeling procedures are described in previous papers [6,22,23]. The nAChR model was imbedded in a fully solvated lipid (POPC) bi-layer (120 Å × 120 Å) using the VMD membrane plug-in [24]. Lipids within 0.8 Å of the protein were removed. The total number of lipids was 298, with 141 on the extracellular side and 157 on the intracellular side. Next, the membrane-protein complex was solvated in TIP3P water using the VMD solvate plug-in. Ions were added to neutralize the net charge of the protein using the VMD autoionize plug-in, and amounted to 84 sodium and 26 chloride atoms, achieving a salt concentration of 100 mM. The resulting system comprises 247,568 atoms, which includes 1886 protein residues and 59,080 TIP3P water molecules.
We used the highly scalable molecular dynamics simulation program, NAMD [25], and the CHARMM27 force field [26]. Once the protein ensemble was built, the following four rounds of equilibrations were completed. (i) 2,000 steps of energy minimization for the non-backbone atoms. (ii) five cycles of a 500-step energy minimization with decreasing position restraints on the protein Cα atoms. (iii) a gradual increase in the temperature from 50 °K to 310 °K in 10,000 steps of constant volume (NVT ensemble ) simulation with restraints (with a force constant of 3 kcal · mol−1 · Å−2) applied to the protein Cα atoms. (iv) a 2 ns constant surface area ensemble MD equilibration with decreasing positional restraints on the Cα atoms. A short cutoff of 9 Å was used for non-bonded interactions, and long-range electrostatic interactions were treated using the Particle Mesh Ewald method [27]. Langevin dynamics and a Langevin piston algorithm were used to maintain the temperature at 310 °K and a pressure of 1 atm. The r-RESPA multiple time step method [28] was employed with a 2 fs time step for bonded atoms, a 2 fs step for short-range non-bonded atoms, and a 4 fs step for long-range electrostatic forces. The bonds between hydrogen and heavy atoms were constrained with the SHAKE algorithm. An external electric field was applied uniformly to all atoms of the system along the z-direction perpendicular to the membrane plane [29]. The voltage difference V across the simulated cell is determined by the product of the length of the simulated cell Lz and the uniform electrostatic field Ez, both in the z-direction. All atoms of the nAChR protein were fixed during restrained simulations. All MD simulations were performed on a 64-processor Linux-cluster in the Receptor Biology Laboratory at Mayo Clinic. |
10.1371/journal.pgen.1008221 | Two-By-One model of cytoplasmic incompatibility: Synthetic recapitulation by transgenic expression of cifA and cifB in Drosophila | Wolbachia are maternally inherited bacteria that infect arthropod species worldwide and are deployed in vector control to curb arboviral spread using cytoplasmic incompatibility (CI). CI kills embryos when an infected male mates with an uninfected female, but the lethality is rescued if the female and her embryos are likewise infected. Two phage WO genes, cifAwMel and cifBwMel from the wMel Wolbachia deployed in vector control, transgenically recapitulate variably penetrant CI, and one of the same genes, cifAwMel, rescues wild type CI. The proposed Two-by-One genetic model predicts that CI and rescue can be recapitulated by transgenic expression alone and that dual cifAwMel and cifBwMel expression can recapitulate strong CI. Here, we use hatch rate and gene expression analyses in transgenic Drosophila melanogaster to demonstrate that CI and rescue can be synthetically recapitulated in full, and strong, transgenic CI comparable to wild type CI is achievable. These data explicitly validate the Two-by-One model in wMel-infected D. melanogaster, establish a robust system for transgenic studies of CI in a model system, and represent the first case of completely engineering male and female animal reproduction to depend upon bacteriophage gene products.
| Releases of Wolbachia-infected mosquitos are underway worldwide because Wolbachia block replication of Zika and Dengue viruses and spread themselves maternally through arthropod populations via cytoplasmic incompatibility (CI). The CI drive system depends on a Wolbachia-induced sperm modification that results in embryonic lethality when an infected male mates with an uninfected female, but this lethality is rescued when the female and her embryos are likewise infected. We recently reported that the phage WO genes, cifA and cifB, cause the sperm modification and cifA rescues the embryonic lethality caused by the wMel Wolbachia strain deployed in vector control. These reports motivated proposal of the Two-by-One model of CI whereby two genes cause lethality and one gene rescues it. Here we provide unequivocal support for the model in the Wolbachia strain used in vector control via synthetic methods that recapitulate CI and rescue in the absence of a Wolbachia infections. Our results reveal the set of phage WO genes responsible for this powerful genetic drive system, act as a proof-of-concept that these genes alone can induce gene drive like crossing patterns, and establish methodologies and hypotheses for future studies of CI in Drosophila. We discuss the implications of the Two-by-One model towards functional mechanisms of CI, the emergence of incompatibility between Wolbachia strains, vector control applications, and CI gene nomenclature.
| Wolbachia are the most widespread endosymbiotic bacteria on the planet and are estimated to infect half of all arthropod species [1,2] and half of the Onchocercidae family of filarial nematodes [3]. They specialize in infecting the cells of reproductive tissues, are primarily inherited maternally from ova to offspring, and often act in arthropods as reproductive parasites that enhance their maternal transmission by distorting host sex ratios and reproduction [4,5]. The most common type of reproductive parasitism is cytoplasmic incompatibility (CI), which manifests as a sperm modification in infected males that causes embryonic lethality or haploidization in matings with uninfected females upon fertilization [6–8]. This embryonic lethality is rescued if the female is infected with the same Wolbachia strain. As such, CI selfishly drives CI-inducing Wolbachia into host populations [9–13], and the incompatibilities between host populations cause reproductive isolation between recently diverged or incipient species [14–18].
In the last decade, Wolbachia and CI have garnered significant interest for their utility in combatting vector borne diseases worldwide. Two strategies are currently deployed: population suppression and population replacement. The population suppression strategy markedly crashes vector population sizes through the release of only infected males that induce CI upon mating with wild uninfected females [19–22]. In contrast, the population replacement strategy converts uninfected to infected populations through the release of both infected males and females that aid the spread Wolbachia via CI and rescue [23,24]. Replacing a vector competent, uninfected population with infected individuals can notably reduce the spread of arthropod borne diseases such as Zika and dengue [25,26] because Wolbachia appear to inhibit various stages of viral replication within arthropods based on diverse manipulations of the host cellular environment [27–33]. The combination of Wolbachia’s abilities to suppress arthropod populations, drive into host populations, and block the spread of viral pathogens have established Wolbachia in the vanguard of vector control efforts to curb arboviral transmission [22–25,34–36].
An unbiased, multi-omic analysis of CI-inducing and CI-incapable Wolbachia strains revealed two adjacent genes, cifA and cifB, in the eukaryotic association module of prophage WO [37] that strictly associate with CI induction [38]. Fragments of the CifA protein were found in the fertilized spermathecae of wPip infected Culex pipiens mosquitoes [39], and these genes are frequently missing or degraded in diverse CI-incapable strains [40,41]. Dual transgenic expression of cifA and cifB from either of the CI-inducing strains wMel or wPip in uninfected male flies causes a decrease in embryonic hatching corresponding to an increase in CI-associated cytological abnormalities including chromatin bridging and regional mitotic failures [38,42]. Single transgenic expression of either cifAwMel or cifBwMel in an uninfected male was insufficient to recapitulate CI, but single transgenic expression of either gene in an infected male enhances wMel-induced CI in a dose-dependent manner [38]. Importantly, dual transgenic CI induced by cifAwMel and cifBwMel expressing males was rescued when they were mated with wMel-infected females [38]. Moreover, transgenic expression of cifAwMel alone in uninfected females rescues embryonic lethality and nullifies cytological defects associated with wild type CI caused by a wMel infection [43].
As such, we recently proposed the Two-by-One genetic model of CI wherein dual expression of cifAwMel and cifBwMel causes CI when expressed in males and expression of cifAwMel rescues CI when expressed in females [43]. However, confirmation of the model’s central prediction requires the complete synthetic replication of CI-induced lethality and rescue in the absence of any Wolbachia infections since it remains possible that other Wolbachia or phage WO genes besides cifA and cifB contribute to wild type CI and rescue by wMel Wolbachia. Moreover, CI induced by dual cifAwMel and cifBwMel expression previously yielded variable offspring lethality with a median survival of 26.5% of embryos relative to survival of 0.0% of embryos from CI induced by a wild type infection under controlled conditions [38]. The inability to recapitulate strong wild type CI suggests other CI genes are required, other environmental factors need to be controlled, or the transgenic system requires optimization.
Here, we utilize transgenic expression, hatch rates, and gene expression assays in Drosophila melanogaster to test if an optimized expression system can generate strong transgenic CI and whether bacteriophage genes cifAwMel and cifBwMel can fully control fly reproduction by inducing and rescuing CI in the complete absence of Wolbachia (Fig 1). We further assess if both cifwMel genes are required for CI induction in the optimized system and whether cifAwMel in females can rescue transgenic CI. Results provide strong evidence for the Two-by-One model in wMel-infected D. melanogaster, offer context for conceptualizing CI mechanisms and the evolution of bidirectional incompatibilities between different Wolbachia strains, raise points for CI gene nomenclature, and motivate further research in developing these genes into a tool that combats vector borne diseases. To the best of our knowledge, they also represent the first case of completely engineering animal sexual reproduction to depend upon bacteriophage gene products.
Dual transgenic expression of cifAwMel and cifBwMel was previously reported to induce highly variable and incomplete CI relative to CI caused by an age-controlled wMel infection [38], indicating either the presence of other genes necessary for strong CI, environmental factors uncontrolled in the study, or inefficiency of the transgenic system. Here, we test the latter hypothesis by dually expressing cifAwMel and cifBwMel in uninfected D. melanogaster males under two distinct GAL4 driver lines that express in reproductive tissues: nos-GAL4-tubulin and nos-GAL4:VP16 [44]. Both driver lines contain a nos promoter region, but differ in that nos-GAL4-tubulin produces a transcription factor with both the DNA binding and transcriptional activating region of the GAL4 protein, and nos-GAL4:VP16 produces a fusion protein of the GAL4 DNA binding domain and the virion protein 16 (VP16) activating region [45,46]. The GAL4:VP16 transcription factor is a particularly potent transcriptional activator because of its binding efficiency to transcription factors [47,48]. Additionally, the nos-GAL4-tubulin driver has a tubulin 3’ UTR, and nos-GAL4:VP16 has a nos 3’ UTR that may contribute to differences in localization within cells or between tissues [44–46]. As such, we predict that differences in the expression level or profile of these two driver lines will lead to differences in the penetrance of transgenic CI.
Since CI manifests as embryonic lethality, we measure hatching of D. melanogaster embryos into larvae to quantify the strength of CI. We confirm previous findings [38] that dual transgenic expression of cifAwMel and cifBwMel under nos-GAL4-tubulin in uninfected males yields low but variable embryonic hatching in crosses with uninfected females (Mdn = 26.3%, IQR = 10.4–38.1%) that can be rescued in crosses with wMel-infected females (Mdn = 97.5%; IQR = 94.2–100%) (Fig 2A). However, dual cifAwMel and cifBwMel expression under nos-GAL4:VP16 in uninfected males yields significantly reduced embryonic hatching relative to nos-GAL4-tubulin (p = 0.0002) with less variability (Mdn = 0%; IQR = 0.0–0.75%) and can be comparably rescued (Mdn = 98.65%; IQR = 95.93–100%; p > 0.99) (Fig 2A). Together, these results support that dual cifAwMel and cifBwMel expression under nos-GAL4:VP16 induces the strongest CI and that the transgenic system, not the absence of necessary CI factors, contributed to the prior inability to recapitulate strong wild type CI.
Next, we tested the hypothesis that differences in the penetrance of transgenic CI between the two drivers are due to differences in the strength of expression. To assess this, we used qPCR to measure the gene expression of cifAwMel and cifBwMel under the two drivers relative to a Drosophila housekeeping gene (rp49) in male abdomens (Fig 2B and 2C). Fold differences in RNA transcripts of cifAwMel relative to rp49 reveal nos-GAL4-tubulin (Mdn = 0.0098; IQR = 0.0082–0.122) drives significantly stronger and more variable cifAwMel expression relative to nos-GAL4:VP16 (Mdn = 0.0075; IQR = 0.0064–0.0090) (p = 0.016, MWU, Fig 2B). The same is true for cifBwMel expression where nos-GAL4-tubulin (Mdn = 0.022; IQR = 0.0165–0.0265) drives significantly stronger cifBwMel expression than nos-GAL4:VP16 (Mdn = 0.0168; IQR = 0.0135–0.0179) (p = 0.02, MWU, Fig 2C). Moreover, while cifAwMel and cifBwMel expression significantly correlate with each other under both nos-GAL4-tubulin (R2 = 0.85; p <0.0001) and nos-GAL4:VP16 (R2 = 0.75; p <0.0001; S1A Fig), neither cifAwMel (R2 = 0.02; p = 0.62; S1B Fig) nor cifBwMel (R2 = 0.04; p = 0.48; S1C Fig) expression levels under the nos-GAL4-tubulin driver correlate with the strength of CI measured via hatch rates. Notably, cifBwMel is consistently more highly expressed than cifAwMel within the same line (S1A Fig). We predict that expression differences are due to either differences in transgenic insertion sites or more rapid degradation of cifAwMel relative to cifBwMel. Taken together, these results suggest that an increase in CI penetrance in these crosses is not positively associated with higher transgene transcript abundance from different drivers.
cifAwMel expression under the maternal triple driver (MTD) in uninfected females can rescue CI induced by a wild type infection [43]. MTD is comprised of three drivers in the same line: nos-GAL4-tubulin, nos-GAL4:VP16, and otu-GAL4:VP16 [44]. We previously reported that cifAwMel expression under the nos-GAL4-tubulin driver alone is rescue-incapable [43]. Here, we test if cifAwMel expression under either of the other components of the MTD driver independently recapitulate rescue of wMel CI. Hatch rate experiments indicate that CI is strong and expectedly not rescued when an infected male mates with a non-transgenic female whose genotype is otherwise nos-GAL4:VP16 (Mdn = 0.0%; IQR = 0.0–0.0%) or otu-GAL4:VP16 (Mdn = 0.0%; IQR = 0.0–0.0%) (Fig 3A). Transgenic expression of cifAwMel in uninfected females under either of the two drivers rescues CI induced by wMel. However, rescue is significantly weaker under cifAwMel expression with the otu-GAL4:VP16 driver (Mdn = 70.4%; IQR = 0.0–90.45%) as compared to the nos-GAL4:VP16 driver (Mdn = 94.2%; IQR = 83.3–97.1%; p = 0.0491) which produced strong transgenic rescue (Fig 3A). Gene expression analysis of cifAwMel relative to rp49 in the abdomens of uninfected females reveals that nos-GAL4:VP16 expresses cifAwMel significantly higher (Mdn = 1.08; p < 0.0001) than otu-GAL4:VP16 (Mdn = 0.03) (Fig 3B), suggesting that high expression in females may underpin the ability to rescue. Alternatively, nos-GAL4:VP16 and otu-GAL4:VP16 are known to express GAL4 at different times in oogenesis, with the former in all egg chambers and the latter in late stage egg chambers [44].
With the transgenic expression system optimized for both transgenic CI and rescue, we then tested the hypothesis that the Two-by-One model can be synthetically recapitulated by dual cifAwMel and cifBwMel expression in uninfected males to cause CI and single cifAwMel expression in uninfected females to rescue that transgenic CI. Indeed, dual cifAwMel and cifBwMel expression in uninfected males causes hatch rates comparable to wild type CI (Mdn = 0.0%; IQR = 0.0%-2.55; p > 0.99) (Fig 4). Transgenic CI cannot be rescued by single cifBwMel expression in uninfected females (Mdn = 1.25%; IQR = 0.0–3.35%). Transgenic CI can be rescued by single cifAwMel expression (Mdn = 98.6%; IQR = 97.35–100%; p = 0.41) or dual cifAwMel and cifBwMel expression (Mdn = 96.7%; IQR = 88.3–98.2%; p > 0.99) to levels comparable to rescue from a wild type infection (Mdn = 95.6%; IQR = 92.5–97.4%). In addition, cifAwMel rescues a wild type infection at comparable levels to wild type rescue (Mdn = 96.6%; IQR = 93.5–98.85%; p > 0.99). These data provide strong evidence for the Two-by-One model in wMel-infected D. melanogaster, namely that CI induced by transgenic dual cifAwMel and cifBwMel expression is sufficient to induce strong CI, and that cifAwMel alone is sufficient to rescue it.
Next we reevaluated if single cifAwMel or cifBwMel expression under the more potent nos-GAL4:VP16 driver in uninfected males can recapitulate CI. Hatch rates indicate that dual cifAwMel and cifBwMel expression induces strong transgenic CI (Mdn = 0.0%; IQR = 0.0–1.15%) that can be rescued by a wild type infection (Mdn = 93.8%; IQR = 88.2–97.4%), whereas single expression of cifAwMel (Mdn = 96.1%; IQR = 97.78–98.55%; p < 0.0001) or cifBwMel (Mdn = 92.85%; IQR = 84.28–96.4%; p < 0.0001) failed once again to produce embryonic hatching comparable to expressing both genes together (Fig 5). In one replicate experiment, we note a statistically insignificant (p = 0.182) decrease in hatching under cifBwMel expression relative to wild type rescue cross (S1 Data file). Thus, both cifAwMel and cifBwMel are required for strong CI. Together, these and earlier results validate the Two-by-One model of CI in wMel whereby cifAwMel and cifBwMel expression are required and sufficient for strong CI, while cifAwMel expression is sufficient to rescue it.
CI is the most common form of Wolbachia-induced reproductive parasitism and is currently at the forefront of vector control efforts to curb transmission of dengue, Zika, and other arthropod-borne human pathogens [22–25,34,35]. Two prophage WO genes from wMel Wolbachia cause CI (cifAwMel and cifBwMel) and one rescues wild type CI (cifAwMel) [38,43], supporting the proposal of a Two-by-One model for the genetic basis of CI [43]. However, dual transgenic expression of cifAwMel and cifBwMel recapitulates only weak and highly variable CI as compared to CI induced by a wild type infection [38]. In addition, the Two-by-One model predicts that both CI and rescue can be synthetically recapitulated by dual cifAwMel and cifBwMel expression in uninfected males and cifAwMel expression in uninfected females. Here we optimized the transgenic system for CI and rescue by these genes, further validated the necessity of expressing both cifAwMel and cifBwMel for CI, and synthetically recapitulated the Two-by-One model for CI with transgenics in the absence of Wolbachia.
CI induced by wMel Wolbachia can be highly variable and correlates with numerous factors including Wolbachia density [49], cifAwMel and cifBwMel expression levels [38], host age [50–52], mating rate [50], rearing density [53], development time [53], and host genetic factors [52,54–56]. Some of these factors, such as age, are known to also correlate with the level of cifwMel gene expression [38]. As such, we hypothesized that prior reports of weakened transgenic CI could be explained by low levels of transgenic cifAwMel and cifBwMel expression in male testes [38].
Indeed, strong CI with a median of 0% embryonic hatching was induced when both cifAwMel and cifBwMel were expressed under the nos-GAL4:VP16 driver. However, contrary to our expectations, nos-GAL4:VP16 generates significantly weaker cifAwMel and cifBwMel expression than the nos-GAL4-tubulin driver previously used to recapitulate weak CI [38]. Thus, the expression data conflict with previous reports in mammalian cells wherein the GAL4:VP16 fusion protein is a more potent transcriptional activator than GAL4 [48]. Other differences between the two driver constructs may explain phenotypic differences, including the presence of different 3’ UTRs that may contribute to differences in transcript localization [44]. While it remains possible, though unlikely, that other Wolbachia or phage WO genes may contribute to CI, the induction of near complete embryonic lethality confirms that cifAwMel and cifBwMel are sufficient to transgenically induce strong CI and do not require other Wolbachia or phage WO genes to do so. Moreover, comparative multi-omics demonstrated that cifA and cifB are the only two genes strictly associated with CI capability [38].
We previously recapitulated transgenic rescue of wMel-induced CI by expression of cifAwMel under the Maternal Triple Driver (MTD) [43], which is comprised of three independent drivers [44]. Expression of cifAwMel using one of the MTD drivers in flies was previously shown to be rescue-incapable [43]; the other drivers had not been evaluated. Here, we tested the hypothesis that expression of cifAwMel using either of the two remaining drivers is sufficient to rescue CI, and we found that cifAwMel expression under both driver lines recapitulates rescue, but at different strengths. Indeed, rescue is strongest when cifAwMel transgene expression is highest. These data are consistent with reports that cifAwMel is a highly expressed gene in transcriptomes of wMel-infected females [57] and the hypothesis that rescue capability is largely determined by the strength of cifAwMel expression in ovaries [43]. These results combined with those for transgenic expression of CI now establish a robust set of methods for future studies of transgene-induced CI and rescue in the D. melanogaster model.
The central prediction of the Two-by-One model is that transgenic CI can be synthetically rescued in the absence of Wolbachia through dual cifA and cifB expression in uninfected males and cifA expression in uninfected females. Here, we explicitly validate the model that two genes are required in males to cause CI, and one in females is required to rescue it using wMel cif gene variants. However, to confirm that the optimized expression system does not influence the ability of cifAwMel or cifBwMel alone to induce CI, we singly expressed them with the improved driver and found that embryonic hatching does not statistically differ from compatible crosses. Coupled with prior data in wMel [38,43], these results strongly support the Two-by-One genetic model whereby dual cifAwMel and cifBwMel expression is required in the testes to cause a sperm modification that can then be rescued by cifAwMel expression in the ovaries (Fig 6A).
While the genetic basis of unidirectional CI appears resolved, it remains unclear how cifAwMel and cifBwMel functionally operate to generate these phenotypes. Numerous mechanistic models have been proposed over the last two decades [58–64]. We can broadly summarize these models into either host-modification (HM) [59] or toxin-antidote (TA) [58] models. HM models suggest that CI-inducing factors modify host products in such a way that would be lethal unless they are later reversed by rescue factors [59–64]. Conversely, TA models state that the CI-inducing factor is toxic to the developing embryo unless it is crucially bound to a cognate antidote provided by the female [42,58,59]. There are numerous lines of evidence in support of both sets of hypotheses and while the Two-by-One genetic model does not explicitly support or favor one set of models over the other, it can be used to generate hypotheses related to the mechanism of CI.
HM models [59] predict that CI factors directly interact with host products in the testes, modify them, and are displaced. These modifications travel with the sperm, in the absence of Wolbachia and Cif products, and would induce the canonical cytological embryonic defects including delayed paternal nuclear envelope breakdown, slowed Cdk1 activation, a failure of maternal histones to deposit onto the paternal genome, stalled or failed replication of the paternal DNA, a failure of paternal chromosomes to segregate, and later stage regional mitotic failures [7,38,60,61,64–67], or they are reversed by female-derived rescue factors. Leading HM models are the Mistiming [60,61] and Goalkeeper [63] models that leverage findings that male pronuclei are delayed in the first mitosis during embryonic development in CI crosses [61,65,67]. Since the first mitosis is initiated when the female pronucleus has developed, the delay of the male pronuclei leads to cytological defects [60]. It is thus proposed that rescue occurs through resynchronization of the first mitosis by comparably delaying the female pronucleus [60,61]. The Goalkeeper model expands the mistiming model to propose that the strength of the delay is what drives incompatibility between different Wolbachia strains [63]. There are numerous hypotheses to explain the role of the Cif products in these kinds of models. One such hypothesis would be that CifA is responsible for pronuclear delay, thus capable of delaying both the male and female pronuclei, but it requires CifB to properly interact with testis-associated targets. This hypothesis may predict that CifB acts to either protect CifA from ubiquitin tagging and degradation, localize it to a host target, or bind CifA to elicit a conformational change required for interacting with male-specific targets. Alternatively, CI-affected embryos express defective paternal histone deposition, protamine development, delayed nuclear breakdown, and delays in replication machinery [7,60,61,64–67]. Any of these factors could be explained by modifications occurring from HM-type interactions between Cif and host products.
TA models [58] contrast to HM models and require that the CI toxin transfers with or in the sperm and directly binds to a female-derived antidote in the embryo. If the antidote is absent, the CI toxin would induce cytological embryonic defects [7,38,60,61,64–67]. There is mixed evidence in support of this model. First, mass spectometry and SDS-PAGE analyses in Culex pipiens reveal that CifAwPip peptides are present in female spermatheca after mating, suggesting CifAwPip is transferred with or in the sperm [39]. CifBwPip was not detected in these analyses, curiously suggesting that the CifB toxin was not transferred [39]. These results are inconsistent with the TA model, but the lack of transferred CifB may occur because cifB gene expression is up to nine-fold lower than that of cifA [57], and the concentration may have been too low to be observed via these methods. Second, CifA and CifB bind in vitro [42]. However, it remains unclear if CifA-CifB binding enables rescue since this binding has no impact on known enzymatic activities of CifB [42]. While the Two-by-One model does not explicitly support or reject the TA model, it does further inform it. Most intriguing is to understand how CifA acts as a contributor to CI when expressed in testes and as a rescue factor when expressed in ovaries. One hypothesis is that CifA and CifB bind to form a toxin complex that is later directly inhibited by female derived CifA [43,59]. The difference in function between these two environments could be explained by post-translational modification and/or differential localization of CifA in testes and embryos [43,59]. Alternatively, CifB may be the primary toxin, but is incapable of inducing CI unless a CifA antidote is present in both the testes and the ovaries [58]. This hypothesis predicts that male-derived CifA rapidly degrades, leaving CifB with or in the sperm. On its own, CifB would induce lethal cytological embryonic defects [60–62,64] unless provided with a fresh supply of CifA from the embryo.
It has been suggested that divergence in CI and rescue factors causes the incipient evolution of reciprocal incompatibility, or bidirectional CI, between different Wolbachia strains [38,43,68,69]. Here, we review a non-exhaustive set of hypotheses that we previously proposed to explain the emergence of bidirectional CI and are consistent with the Two-by-One model [43]. First, the simplest explanation for CifA’s role in both CI and rescue is that it has similar functional effects in both testes/sperm and ovaries/embryos. Thus, instead of requiring a separate mutation for CI and another for rescue [69], bidirectional CI may emerge from a single CifA mutation that causes incompatibility against the ancestral strain while maintaining self-compatibility. Second, CifA in testes and ovaries may also have different functions, localizations, or posttranslational modifications that contribute to CI and rescue. If this occurs, or if CifB is also an incompatibility factor, the evolution of bidirectional CI may require two or more mutations, and the strain may pass through an intermediate phenotype wherein it becomes unidirectionally incompatible with the ancestral variant or loses the capability to induce either CI or rescue before becoming bidirectionally incompatible with the ancestral variant. In fact, some Wolbachia strains are incapable of inducing CI but capable of rescuing CI induced by other strains [70], and some can induce CI but cannot be rescued [71]. Furthermore, sequence variation in both cifA and cifB from Wolbachia strains in Drosophila [38] and in small regions among strains of wPip Wolbachia [68] have been correlated to incompatibility, suggesting that variation in both genes influence incompatibility.
Additionally, it remains possible that significant divergence in cifA, cifB, or both may be necessary to generate new phenotypes. Indeed, comparative genomic analyses reveal high levels of amino acid divergence in CifA and CifB that correlates with incompatibility between strains [38,40]. Moreover, some Wolbachia strains harbor numerous phage WO variants, each with their own, often divergent, cif genes, and the presence of multiple variants likewise correlates with incompatibility [38,40,68]. Thus, horizontal transfer of phage WO [37,72–76] can in theory rapidly introduce new compatibility relationships, and duplication of phage WO regions, or specifically cif genes, in the same Wolbachia genome may relax the selective pressure on the cif genes and enable their divergence. Determining which of the aforementioned models best explains the evolution of incompatibilities between Wolbachia strains will be assisted by additional sequencing studies to identify incompatible strains with closely related cif variants.
The genetic bases of numerous gene drives have been elucidated in plants [77], fungi [78–81], and nematodes [82,83]. Some gene drives have also been artificially replicated with transgenic constructs [84–86]. However, to our knowledge, the synthetic replication of the Two-by-One model of CI represents the first instance that a gene drive has been constructed by engineering eukaryotic reproduction to depend on phage proteins. Additionally, vector control programs using Wolbachia rely on their ability to suppress pathogens such as Zika and dengue viruses, reduce the size of vector populations, and spread Wolbachia into a host population via CI and rescue. However, there are limitations to these approaches. Most critically, not all pathogens are inhibited by Wolbachia infection and some are enhanced, such as West Nile Virus in Culex tarsalis infected with wAlbB Wolbachia [87]. Additionally, it requires substantial effort to establish a Wolbachia transinfection in a target non-native species [88] that could be obviated in genetically tractable vectors utilizing transgenic gene drives.
The complete synthetic replication of CI and rescue via the Two-by-One model represents a step towards transgenically using the cif genes in vector control efforts. The separation of CI mechanism from Wolbachia infection could theoretically expand CI’s utility to spread ‘payload’ genes that reduce the vectoral capacity of their hosts [89] into a vector population by, for instance, expressing the CI genes and the payload gene polycistronically under the same promoter in the vector’s nuclear or mitochondrial genomes. Moreover, these synthetic constructs have potential to increase the efficiency of Wolbachia-induced CI if they are transformed directly into Wolbachia genomes. For these efforts to be successful, considerable work is necessary to (i) generate a constitutively expressing cif gene drive that does not require GAL4 to operate, (ii) understand the spread dynamics of transgenic CI, (iii) characterize the impact of cif transgenic expression on insect fitness relative to wild vectors, (iv) generate and test effective payload genes in combination with cif drive, (v) explore and optimize the efficacy of cif drive in vector competent hosts such as mosquitoes, (vi) assess the impact of host factors on cif drive across age and development, (vii) compare the efficacy of a cif gene drive to other comparable technologies (CRISPR, homing drive, Medea, etc), in addition to numerous other lines of study. For example, while a substantial body of literature exists to describe the spread dynamics of CI [10,12,13,36,90,91], none yet describe how the Two-by-One model would translate into nuclear or mitochondrial spread dynamics in the absence of Wolbachia. As such, this study represents an early proof of concept that these genes alone are capable of biasing offspring survival in favor of flies expressing these genes under strictly controlled conditions, and should motivate additional study towards its application in vector control.
The generality of the Two-by-One model remains to be tested because it may be specific to certain strains of Wolbachia and/or phage haplotypes. For instance, transgenic expression of cifBwPip from C. pipiens in yeast yields temperature sensitive lethality that can be rescued by dual-expression of cifAwPip and cifBwPip [42]. Moreover, attempts to generate a cifBwPip transgenic line failed, possibly due to generalized toxicity from leaky expression [42]. Therefore, cifBwPip alone could in theory cause CI. However, this model has not been explicitly tested, it has not been explained how cifAwPip and cifBwPip dual-expression induces CI in transgenic Drosophila but prevents CI in yeast, and transgenic wPip CI has not been rescued in an insect. As such, it remains possible that cifBwPip lethality could be explained by artefactual toxicity of overexpression or toxic expression in a heterologous system. Thus, confirmation of an alternative model for CI in wPip is precluded by lack of evidence that cifBwPip alone can induce rescuable lethality in an insect. Since cifBwPip transgenic UAS constructs have not been generated due to toxicity from leaky expression, alternative PhiC31 landing sites or expression systems (i.e., the Q System) could prove valuable in addressing these questions.
Finally, these results further validate the importance of cifAwMel as an essential component of CI and underscore a community need to unify the nomenclature of the CI genes. When the CI genes were first reported, they were described as both CI factors (cif) and as CI deubiquitilases (cid), both of which are actively utilized in the literature. The cif nomenclature was proposed as a cautious naming strategy agnostic to the varied biochemical functions to be discovered, whereas the cid nomenclature was proposed based on the finding that the B protein is in part an in vitro deubiquitilase that, when ablated, inhibits CI-like induction [38,42]. A recent nomenclature proposal suggested that the cif gene family name be used as an umbrella label to describe all CI-associated factors whereas cidA and cidB would be used to describe the specific genes [58]. However, we do not agree with this nomenclature revision despite the appeal of combining the two nomenclatures. CifA protein is not a putative deubiquitilase [40], does not influence deubiquitilase activity of CifB [42], functions independently to rescue CI [43] and, as emphasized by the work in this study, is necessary for CI induction and rescue. The competing nomenclature presumes that it is appropriate to name the A protein cid because it could be expressed in an operon with the B protein. However, the evidence for the operon status of the genes is weak, and more work is needed to describe the regulatory control of these genes before they can be categorized as an operon [59]. Moreover, distant homologs that cluster into distinct phylogenetic groups are proposed to be named CI nucleases (cin) [42] yet the merger of these two groups into one name lacks phylogenetic rationality as the two lineages are as markedly divergent from each other as they are from cid [59]. In addition, none of these distant homologs have been functionally characterized as CI genes [38,40]. As such, it is more appropriate to call these genes “cif-like” to reflect their homology and unknown phenotypes. Thus, the holistic and conservative cif nomenclature with Types (e.g., I-IV) used to delineate phylogenetic clades is appropriately warranted in utilizing and unifying CI gene naming.
In conclusion, the results presented here support that both cifAwMel and cifBwMel phage genes are necessary and sufficient to induce strong CI. In addition, cifAwMel is the only gene necessary for rescue of either transgenic or wild type wMel CI. These results confirm the Two-by-One model of CI in wMel Wolbachia and phage WO with implications for the mechanism of CI and for the diversity of incompatibility between strains, and they provide additional context for understanding CI currently deployed in vector control efforts. The synthetic replication of CI in the absence of Wolbachia marks an early step in developing CI as a tool for genetic and mechanistic studies in D. melanogaster and for vector control efforts that may drive payload genes into vector competent populations.
D. melanogaster stocks y1w* (BDSC 1495), nos-GAL4-tubulin (BDSC 4442), nos-GAL4:VP16 (BDSC 4937), otu-GAL4:VP16 (BDSC 58424), and UAS transgenic lines homozygous for cifA, cifB, and cifA;B [38] were maintained at 12:12 light:dark at 25o C and 70% relative humidity (RH) on 50 ml of a standard media. cifA insertion was performed with y1 M{vas-int.Dm}ZH-2A w*; P{CaryP}attP40 and cifB insertion was performed with y1 w67c23; P{CaryP}attP2, as previously described [38]. UAS transgenic lines and nos-GAL4:VP16 were uninfected whereas nos-GAL4-tubulin and otu-GAL4:VP16 lines were infected with wMel Wolbachia. Uninfected versions of infected lines were produced through tetracycline treatment as previously described [38]. WolbF and WolbR3 primers were regularly used to confirm infection status [38]. Stocks for virgin collections were stored at 18o C overnight to slow eclosion rate, and virgin flies were kept at room temperature.
To test for CI, hatch rate assays were used as previously described [38,43]. Briefly, GAL4 adult females were aged 9–11 days post eclosion and mated with UAS males. Age controlled GAL4-UAS males and females were paired in 8 oz bottles affixed with a grape-juice agar plate smeared with yeast affixed to the opening with tape. 0–48 hour old males were used since CI strength rapidly declines with male age [50,52]. The flies and bottles were stored at 25o C for 24 h at which time the plates were replaced with freshly smeared plates and again stored for 24 h. Plates were then removed and the number of embryos on each plate were counted and stored at 25o C. After 30 h the remaining unhatched embryos were counted. The percent of embryos hatched into larvae was calculated by dividing the number of hatched embryos by the initial embryo count and multiplying by 100.
To assay transgenic RNA expression levels under the various gene drive systems, transgene expressing flies from hatch rates were immediately collected and frozen at -80°C for downstream application as previously described [43]. In brief, abdomens were dissected, RNA was extracted using the Direct-zol RNA MiniPrep Kit (Zymo), the DNA-free kit (Ambion, Life Technologies) was then used to remove DNA contamination, and cDNA was generated with SuperScript VILO (Invitrogen). Quantitative PCR was performed on a Bio-Rad CFX-96 Real-Time System in duplicate using iTaq Universal SYBR Green Supermix (Bio-Rad) using the cifA_opt and rp49 forward and reverse primers as previously described [43]. Samples with a standard deviation >0.3 between duplicates were excluded from analysis. Fold expression of cifA relative to rp49 was determined with 2−ΔΔCt. Each expression study was conducted once.
All statistical analyses were conducted in GraphPad Prism (Prism 8). Hatch rate statistical comparisons were made using Kruskal-Wallis followed by a Dunn’s multiple comparison test. A Mann-Whitney-U was used for statistical comparison of RNA fold expression. A linear regression was used to assess correlations between hatch rate and expression. All p-values are reported in S1 Table.
|
10.1371/journal.ppat.1002110 | CD39/Adenosine Pathway Is Involved in AIDS Progression | HIV-1 infection is characterized by a chronic activation of the immune system and suppressed function of T lymphocytes. Regulatory CD4+ CD25high FoxP3+CD127low T cells (Treg) play a key role in both conditions. Here, we show that HIV-1 positive patients have a significant increase of Treg-associated expression of CD39/ENTPD1, an ectoenzyme which in concert with CD73 generates adenosine. We show in vitro that the CD39/adenosine axis is involved in Treg suppression in HIV infection. Treg inhibitory effects are relieved by CD39 down modulation and are reproduced by an adenosine-agonist in accordance with a higher expression of the adenosine A2A receptor on patients' T cells. Notably, the expansion of the Treg CD39+ correlates with the level of immune activation and lower CD4+ counts in HIV-1 infected patients. Finally, in a genetic association study performed in three different cohorts, we identified a CD39 gene polymorphism that was associated with down-modulated CD39 expression and a slower progression to AIDS.
| HIV-1 infection is characterized by a chronic activation of the immune system. Regulatory T cells (Treg) represent a population of lymphocytes that controls inappropriate or exaggerated immune activation induced by pathogens, thereby influencing the outcome of various infections. Several studies have shown that Treg are expanded in HIV infected patients. However, the mechanisms of Treg immune-modulator functions are not clearly known. CD39 is an ectonucleotidase which converts the proinflammatory ATP signal into AMP and the immunosuppressive adenosine in concert with CD73. A critical role of CD39 has been described for Treg in general but few studies have analyzed its role in HIV infection. We report here an expansion of Treg expressing CD39 in a cohort of HIV-infected patients. In vitro these cells exerted a strong suppressive effect on the effector CD8 T cells. Treg inhibitory effects were relieved by CD39 down-modulation using an anti-CD39 monoclonal antibody. Treg suppressive effects were reproduced by an adenosine agonist in accordance with a higher expression of the adenosine A2A receptor on patients' T cells. From a clinical stand point, we show also a correlation between Treg CD39+ expansion and both immune activation and CD4+ T cell depletion in patients. Finally, by genetic analysis of three different cohorts of patients, we found that a CD39 gene polymorphism associated with a lower CD39 expression correlated with a slower progression to AIDS. Thus, our results contribute to elucidate the mechanisms by which Treg suppression occurs during HIV infection.
| HIV-1 infection is characterized by chronic immune activation which, in combination with the progressive depletion of CD4+ T cells, profoundly perturbs antigen-specific T cell responses [1]. The population of CD4+CD25high FoxP3+ regulatory T cells (Treg) suppresses antigen-specific T cell responses and controls inappropriate or exaggerated immune activation induced by pathogens, thereby influencing the outcome of various infections [2], [3]. In particular, these cells suppress in vitro HIV-1-specific CD4+ and CD8+ effector T-cell responses [2], [4]. We, and others, have reported an HIV-1-driven expansion of Treg expression in chronic and acute HIV-1 infection [5], [6], including a relationship between the expansion of Treg, the level of cellular immune activation and the depletion of CD4+ T cells in acute HIV infection [5].
The molecular mechanisms by which Treg mediate their suppressive activity remain poorly understood. In humans, the Treg population exhibits considerable diversity. Phenotypically and functionally distinct subsets of Treg can mediate suppression through distinct mechanisms from secretion of IL-10, TGF-ß, IL-35, Granzyme B, perforin, to CTLA-4 and GITR interactions [7], [8], [9]. Recently, it has been reported that CD39 is expressed on human and murine Treg, while CD73 is found only on the surface of murine Treg [10], [11], [12]. CD39, a member of the ectonucleotidase triphosphate diphosphohydrolase family (ENTPD), also referred to as ENTPD-1 (EC 3.6.1.5), is the dominant immune system ectonucleotidase that hydrolyses extracellular ATP and adenosine diphosphate (ADP) into adenosine monophosphate (AMP) at the sites of immune activation. CD73 is an ecto-5′-nucleotidase (5′NT) that exists in a soluble or membrane-bound form and catalyzes the dephosphorylation of AMP to adenosine [13], [14], [15]. Adenosine is a critical regulator of innate and adaptive immune responses [16], [17], inhibiting T lymphocyte proliferation and the secretion of inflammatory cytokines including IL-2, TNFa, and IFN-γ [13], [14], [15]. These effects are mediated through A2A receptors stimulating the generation of cAMP, and are mimicked by adenosine agonists [18]. CD39 has also been described as an activation marker of lymphoid cells [19]. Therefore, the CD39/Adenosine pathway may be important to the balance between activation and regulation of effector immune responses.
Here we tested the hypothesis that the CD39/adenosine pathway is involved in the pathogenesis of HIV-1 disease. First, we investigated the phenotype and the function of Treg-expressing CD39 molecules in a cohort of chronically HIV-positive patients and determined whether these characteristics are associated with clinical outcomes. Second, to assess our hypothesis in an in vivo context, we investigated whether CD39 genetic polymorphisms were associated with rates of HIV-1 disease progression in three independent cohorts.
In order to discriminate between Treg and activated T cells, we further characterized Treg population as gated T cells expressing CD4+CD25high FoxP3+high and CD127low (gating strategy is shown in Fig. S1). These cells are designated thereafter as Treg cells while CD4+CD25lowCD127high T cells are designated as activated CD4+CD25low T cells (T act). First, we confirmed a significant increase in the percentages of Treg cells in a cohort of HIV-positive individuals, receiving either a combination of antiretroviral drugs (c-ART+, n = 39) or not (c-ART−, n = 39), as compared to healthy controls (n = 25) (mean 5.8% and 6.2% respectively vs 2.4%, P<0.0001) (Fig. 1a). As shown in Fig. 1b and 1c, percentages of Treg expressing CD39+ (Treg CD39+) were significantly higher in both c-ART+ and c-ART− patients, as compared to healthy controls (mean 2.79% and 2.26% vs 0.97%, P<0.001, Fig. 1b). Moreover, Treg from both c-ART− and c-ART+ subjects expressed a higher density of CD39 molecules as compared to those from HIV-1 negative controls (mean fluorescence intensity (MFI) 1327 and 1203, respectively, vs. 652, P<0.001 and P<0.01) (Fig. 1c).
Phenotypic analyses were performed in 16 HIV-1 positive patients before and 12 months following c-ART initiation. Among them, 9 patients experienced a good response to c-ART (group A; undetectable plasma viral load at month 12), while in 7 patients (group B) viral replication remained detectable (above 50 copies/ml). No significant decrease of CD39 expression was observed in group A: % Treg CD39+ (mean ± SD): 2.4±1.2 vs.1.8±1.0 at baseline; TregCD39+ MFI (mean ± SD): 1557±360 vs. 1261±656 at baseline, (P>0.05 for both). Moreover, in patients with on-going viral replication %Treg CD39+ increased significantly in spite of ART (6.1±2.4 versus 3.4±2.3 at baseline; P = 0.043).
CD39 has also been described as an activation marker of lymphoid cells [19]. Therefore, we looked at the percentages of Tact in HIV-1 positive patients and controls. As expected, the frequency of activated CD4+CD25low T cells was significantly higher in both populations of patients as compared to controls (Fig.S1b). Consequently, percentages of CD4+CD25lowCD39+ were significantly higher in HIV-1 positive patients as compared to controls (Fig.S1c). In contrast to Treg, CD4+CD25− T cells from both HIV-positive subjects and controls did not express CD39 (not shown). Thus, an expansion of CD39+CD4+ T cells in both Treg and T act T cell populations, which persist in patients with controlled viral load under c-ART, is observed in HIV-1 positive patients. In HIV-positive subjects and in HIV-negative controls, Treg cells were mostly of CD45RA−CD28+ memory phenotype (mean 75%). CD45RA−CD28+ Treg contained a higher percentage of CD39+ cells as compared to CD45RA+CD28+ Treg cells (mean 65% vs. 28%, respectively, P<0.05) (Fig. S2).
We next investigated whether down-modulation of the CD39 enzyme can impact Treg function. First, by exposing cells to a blocking anti-CD39 (BY40) mAb, we induced a down-modulation of CD39 expression at the surface of the YT2C2 NK line cells (Fig. S3a). Next, BY40 mAb down-modulated the expression of CD39 on ex-vivo purified peripheral blood Treg from HIV-negative controls as compared to untreated cells or cells treated with an IgG1 control mAb (% of positive cells (mean ± SD): 32±11% vs 44±13%, and 42±14%, respectively) (Fig. 2a,b). In these experiments, CD39 expression following in vitro incubation with BY40 mAb was assessed using a commercial PE anti-CD39 (clone TU66) which has been previously checked to be non-competitive with BY40 (Fig. S4). Finally, we found that this down modulation effect of BY40 was associated with decreased CD39 ATPase activity on primary monocytes (Fig. S3b).
The functional consequences of CD39 down-modulation were investigated in co-culture assays developed to evaluate the suppressive effects of Treg on T cell proliferation [5], [6], [20]. As shown in Fig. 3a and b (for one representative experiment and pooled data from 6 HIV-positive subjects), the Treg-mediated inhibition of anti-CD3 induced CD8 T cell proliferation was significantly higher in HIV-positive subjects (n = 6) as compared to HIV-negative controls (n = 6), (mean inhibition 56% vs 22.5%; P<0.01) (Fig. 3b). Pre-incubation with anti-CD39 BY40 mAb reversed by ∼50% the suppressive effect of Treg from HIV-positive subjects (average suppression rate of 28% in the presence of Treg pre-treated with BY40 as compared to 56% and 57% for Treg pre-treated or not with IgG1 control mAb, (P = 0.01; one-way ANOVA and paired T-test P = 0.01 for group by group comparisons). Interestingly, although the suppression mediated by Treg from HIV-negative controls was less significant, a similar effect of anti-CD39 BY40 mAb was noted (average inhibition 12.3% as compared to 22.5%, one-way ANOVA P<0.01 and paired T-test P<0.01). These results are in accordance with the higher density of CD39 molecules expressed by Treg from HIV-positive subjects and indicate that this enzyme is involved, at least in part, in the Treg-mediated inhibition of CD8+ T cell proliferation.
Next, we evaluated the effects of Treg on the cytokine production of CD8 T cells in response to HIV-1 antigens. Cytokine production (IFN-γ, TNFα and IL-2) of CD8-gated T cells was analyzed by intra cytoplamic staining and flow cytometry after overnight stimulation with a pool of whole Gag 15mer peptides (2 µg/ml). As shown in Fig. 4, the percentages (mean ± SD) of CD8+ Cytokines+ T cells were 2.1+/−0.7% vs. 3.3%+/−1% (n = 5) in the presence of Treg and CD4+CD25− respectively (P = 0.05). Pre-incubation of Treg with anti-CD39 mAbs, but not with isotype control, relieved this suppressive effect: 3.2+/−0.8%, (P = 0.05).
Together, these results indicate that CD39 enzyme participates in the Treg-mediated suppression on CD8 T cell proliferation and responses to HIV peptides.
To further investigate the involvement of CD39/adenosine in the Treg-mediated inhibition of CD8+ T cell proliferation in HIV-1 positive subjects, we studied the effects of the A2AR agonist CGS21680 on proliferation of anti-CD3 stimulated T cells. The mean (±SD) inhibition of CD4+ T cells was 47% (±11) and 57% (±8.3) in the presence of 0.1 and 1 mM of CGS, respectively in c-ART− HIV positive patients. Similarly, the same doses of CGS inhibited by 47% and 65% the proliferation of anti-CD3 activated CD8+ T cells from c-ART− HIV-positive subjects (P<0.05) (Fig. 5a,b). In contrast, the proliferation of CD4+ and CD8+ T cells from HIV-negative controls and c-ART+ HIV-positive subjects was much lower and below 20% at the highest dose of CGS21680 (1 mM) (Fig. 5a,b) (P = 0.015 and P = 0.027 respectively; one-way ANOVA and P<0.05 unpaired T-test for comparison between c-ART−HIV-positive patients and the two other groups (Fig. 5a,b).
In accordance with this, we found that both CD4+ and CD8+ purified T cells from c-ART− HIV-positive subjects (n = 7) expressed a significantly higher level of A2AR mRNA than c-ART+ subjects (n = 5) or HIV-negative controls (n = 6) (Fig. 5c).
Since the HIV-positive subjects we studied were heterogeneous in terms of disease duration and clinical stage, we assessed whether CD39 expression correlated with established markers of disease progression. The frequency of the Treg CD39+ subset correlated directly with plasma HIV-1 viral load in the group of c-ART− subjects (P<0.05, R = 0.45) (Fig. 6a). Moreover, the percentage of Treg CD39+ subset correlated directly with the activation of CD4+ T cells in c-ART− subjects, assessed by the percentage of CD4+HLA-DR+ (P<0.05, R = 0.66) (Fig. 6b). Finally, the percentage of Treg CD39+ cells and CD39 MFI correlated inversely with absolute CD4+ T cell count in c-ART− subjects (P<0.001, R = −0.51 and P<0.001, R = −0.57, respectively) (Fig. 6 c,d) as well as in c-ART+ subjects (P<0.001 , R = −0.57 and P<0.01, R = −0.43) (Fig. 6 e,f).
The independent prognostication value of CD39 expression on Treg for CD4 T cell counts was studied in c-ART− and in c-ART+ patients, using multiple linear regression models (SPSS v.17.0). The frequencies of Treg, Treg CD39+, Tact, Tact CD39+, and viral load (for c-ART− group only) were included as predictors of CD4 absolute count. For c-ART+HIV-positive patients, in a full model (R2 = 0,398, ANOVA P = 0.02) the percentage of Treg CD39+ had the most important partial predictive effect (partial correlation coefficient −0.479), confirmed by sequential multiple regression analysis of the same set of variables (partial correlation coefficient −0.612 vs. 0.360 for Tact, ANOVA P = 0.001).
For c-ART−HIV-positive patients, in a full model (R2 = 0,392, ANOVA sig. = 0.045), again the percentage of TregCD39+ had the most important contribution as predictor for CD4 absolute count, followed by CD39Tact (partial correlation coefficients −0.375, and 0. 265 respectively).
These results indicate that the frequency of Treg CD39+ is an independent predictive factor for CD4 cell count variability.
Our results highly suggest that the frequency of Treg CD39+ cells, as well as the density of the enzyme molecule at the surface of those cells, predict disease progression. Recently, CD39 gene polymorphisms associated with the level of enzyme expression have been shown to be associated with susceptibility to Crohn's disease [21]. In order to assess the role of CD39 on HIV-1 disease progression, we investigated whether CD39 gene polymorphisms could be associated with clinical outcomes. For that, we exploited the GRIV cohort, comprising subjects exhibiting extreme profiles of AIDS progression (LTNP, long-term non-progressors and RP, rapid progressors) [20], [22], [23]. We thus performed a genetic case-control association study on the candidate gene CD39 using the genotype data collected from our previous genome-wide association studies [22], [23] (see Methods).
Fourteen SNPs were identified in the CD39 gene. No polymorphism was significantly associated with rapid progression, whereas four SNPs were significantly associated with LTNP: rs10882665 (P = 1.33×10−2), rs3181123 (P = 1.38×10−2), rs1933166 (P = 1.76×10−2), and rs11188513 (P = 3.60×10−2) (Fig. S5). Of note, rs10882665 and rs3181123 are in full linkage disequilibrium (r2 = 1). To eliminate a potential association with HIV-1 infection rather than with LTNP, we compared the allelic frequency of each of these SNPs in the RP population. The frequency observed in the RP group was similar to the frequency observed in the control group, confirming that this was an association with LTNP.
To confirm these results, we used two additional independent Caucasian cohorts that examined AIDS progression phenotype: the ACS and the MACS cohorts (see Methods). The rs11188513 SNP (whose frequency in LTNP and control groups were, respectively, 39% and 34%, P = 3.60×10−2, (Fig. 7a) was the only polymorphism also associated with disease progression both in ACS (P = 2.64×10−2) and MACS (P = 2.07×10−2) (Fig. 7b,c and Table S1). The P values compute the probability that an association is due to chance and the combined P value for rs11188513 over the three cohorts was significant after Bonferroni corrections, P = 6.11×10−3. Importantly, as shown in Fig. 7, the rs11188513-C allele favoured slower progression of HIV infection in all three cohorts. This association was independent from the CCR5 polymorphisms (P1 and Delta32) also located in chromosome 3, since the p value was not modified by using the CCR5 variants as covariates.
To further explore this association, we examined the Genevar [24] and the Dixon [25] mRNA expression databases, and found a correlation (P = 3.26×10−5 and P = 1.9×10−14, respectively) between the rs11188513-C allele and lower expression of the CD39 gene. Thus, the genetic association study combined with the mRNA expression database information demonstrate that the rs11188513-C allele is associated both with a slower progression to AIDS and with a lower expression of CD39 gene.
We show here the involvement of the CD39/adenosine pathway in the Treg-mediated suppressive effect on HIV-1-infected subjects' T cell functions. We demonstrate that HIV-positive subjects exhibit both a higher frequency of Treg CD39+ and a higher in vitro sensitivity of effector T cells to the suppressive effect of adenosine, due to a higher expression of its predominant A2A receptor. Expansion of Treg CD39+ correlates inversely with CD4 T cell counts in HIV infection independently of plasma viral loads and T cell activation. Finally, in a genetic association study conducted in three different HIV-positive cohorts we show that the level of CD39 gene expression can indeed impact the course of disease progression.
Recent data have shown that mouse Treg constitutively express CD39 [26], while the proportion of Treg CD39+ cells appears highly variable in healthy human controls [10]. Therefore, in contrast to mice, CD39 expression might delineate a subpopulation of human Treg [10], [27]. However, studies on human Treg CD39+ cells are scarce. Few studies have analyzed the expression of CD39 in HIV disease [28]. Leal et al. have shown an increased nucleotidase activity related to enhanced CD39 expression on lymphocytes of HIV-positive subjects [28]. More recently, and in accordance with results presented here, an increase in the frequency of Treg expressing CD39 has been shown in different cohorts of HIV infected patients [29]. However, these observations warrant further investigations on the role of CD39 and the clinical relevance of these findings. Our results reinforce these observations and provide new insights about the biological mechanisms involving the CD39/adenosine axis. The demonstration that blocking of CD39 with BY40 mAb relieved, although not completely, the suppressive effect of Treg on effector T cells opens the way to new therapeutic interventions aimed to modulate Treg functions [29]. Moreover, we found that Treg CD39+ inhibit cytokine production by HIV-specific CD8 T cells, an effect partially relieved by pre-incubation of Treg CD39+ with anti-CD39 mAb. These results demonstrate that CD39 enzymatic pathway is responsible, at least in part, for the inefficiency of CD8 T cells responses in chronic HIV-1 infection. In contrast, the CD39 pathway seemed to be less predominant in coculture studies performed with cells purified from HIV negative controls. However, we cannot rule out that down-modulation of CD39 enzymatic activity may also interfere with other suppressive pathways.
Our results are similar to those reported in cancer and HIV patients in whom the purified Treg CD39+ subset mediated a higher suppression as compared to control patients [27]. From a clinical stand-point, it is interesting to note the persistence of a higher frequency of Treg CD39+ cells in HIV-positive subjects with controlled viral load, as compared to HIV-negative controls. Likely, this may reflect ongoing chronic immune activation. We show here that the frequency of TregCD39+ is correlated positively to the percentages of activated CD4+ T cells expressing HLA-DR (Fig. 6b) and a higher frequency of conventional T cells (CD4+CD25−) expressing CCR5 (not shown) which may partly explain CD4+ T cell depletion. Alternatively, since the Treg CD39+ subset is mostly confined to the memory CD4 T cell compartment, this population may represent HIV-inducible Treg, as previously reported [5], [6]. Recently, an expansion of suppressive FoxP3+CD39+ CD8 regulatory T cells associated with poor antiviral response has been reported in HIV-infected patients [30]. In our study, we have checked that expression of CD39 molecule on other blood subsets (B, NK and monocytes) did not vary significantly between patients' groups (Fig S6).
Altogether these results support the conclusion that the Treg subset expressing a high density of both CD25 and CD39 molecules represents a highly-enriched population of suppressor T cells in HIV-1 infected patients.
Adenosine is formed in tissue microenvironments under inflammatory insult [16], [31], [32], [33]. Several studies have shown that adenosine plays an important non-redundant role in the regulation of T cell activation [18], [34], [35]. Using the dose-dependant inhibitory effect of the adenosine receptor agonist CGS21680 [18], we confirmed the involvement of CD39/adenosine pathway in the Treg-mediated inhibition of T cell proliferation in HIV-1 infected patients. It is noteworthy that CD39/adenosine inhibition affected both CD8 and CD4 T cells, and was significantly more important in c-ART-naïve HIV positive subjects. This latter difference was due to a significantly higher level of A2AR expression. We found that CGS21680 did not inhibit the proliferation of T cells from c-ART treated patients. However, as we did not evaluate CGS21680 effects on other T cell functions, we cannot rule out that A2AR agonists may also impair T cell cytotoxicity and production of cytokines such as IL-2 and IFN-g rather than cell proliferation, as recently demonstrated [36], [37].
Our data provide clues to the suppressive mechanisms of Treg in the context of chronic immune activation. CD39 expression by Treg is important for the extracellular removal of ATP and allows Treg infiltration of inflamed tissues, resulting in an increase of local extracellular adenosine concentration by ATP catabolism [11], [38]. Extracellular ATP depletion may also increase Treg survival and favour the local accumulation of Treg, since high levels of ATP have been shown to be a pro-apoptotic factor [39], [40]. On the other hand, this microenvironment represents a self-protective mechanism against immune attacks [16], [41] by inducing a rapid tolerization of activated cells, as demonstrated in cancer models [42]. Recent data in a mice model has shown that tissue-derived adenosine promotes peripheral tolerance by inducing T cell anergy and Treg differentiation [37]. Altogether, these studies show that initiation of T cell activation in inflamed tissue and/or tumour microenvironments might result in the induction of T cell unresponsiveness by an A2AR-dependent mechanism. These observations may explain the reports of HIV infection in which Treg coexist in tissues infiltrated with HIV-specific T cells that are poorly capable of controlling local HIV replication [43], [44]. Of note, our study was limited to peripheral blood. Whether, the involvement of CD39/adenosine pathway plays also a key role in secondary lymphoid organs or in mucosa deserves further studies.
Treg CD39+ expansion may help establish the relationship between immune activation and Treg-mediated suppression in HIV-1 infection. Increased ATP and adenine nucleotides in inflamed sites may serve as substrates for Treg-expressed nucleotidases but also may exert direct Treg-activating effects [45]. Thus, the ATP-Treg balance might be crucial for the regulation of inflammation. However, in the long term, CD39-mediated inhibition of T cell proliferation might exert an adverse effect not only on the immediate generation of T-cell immune responses, but also on the maintenance and restoration of the T-cell pool, thus contributing to disease progression. We also showed that despite efficient c-ART, the percentage of Treg CD39+ remains higher in c-ART+ HIV-1 subjects as compared to controls. Although T cells from these individuals express low levels of A2AR, we found that Treg still exert a significant inhibitory effect that was relieved by anti-CD39 blocking antibodies. This observation corroborates the observation of an inverse relationship between the frequency of Treg CD39+ and CD4+ T cell counts in patients (Fig. 6). Although the role of Treg in HIV-1 infection remains unclear, the identification of a novel Treg subset participating in Treg suppression may be useful to discriminate between a “friend or foe” role of Treg in HIV-1 infection.
Through a candidate gene association study, we identified a CD39 gene variant associated with down-modulation of CD39 expression that impacts the course of disease progression, a finding that was replicated in three different cohorts. Such high P values for the association of this variant and CD39 expression in both Genevar and Dixon databases are extremely rare. Since the SNP identified is in high linkage disequilibrium (r2>0.9) with several other SNPs within the CD39 gene, further studies are warranted to determine which of them is a causal variant. It is important to note that, according to the HapMap database, this SNP exists at a allelic frequency of ∼30% in the African population and at ∼70% in the Asian population, suggesting that this genetic variant may be an important determinant of disease progression in both populations. Overall, the genetic association study confirms in vivo the hypotheses put forward by our experimental work: subjects carrying the CD39-C allele are likely to exhibit a lower CD39 expression, which could impact the control of T cell immune responses, and in turn slow down HIV-1 disease progression.
Our data show that the CD39/adenosine axis might be a novel pathway involved in the Treg-mediated suppression in HIV infection through both an expansion of Treg strongly expressing the ectonucleotidase CD39, and an increased sensitivity of patients' T cells to adenosine. In this context, the possibility to revert Treg-mediated inhibition using CD39-blocking mAb or by modifying the adenosine turnover with specific drugs seems an attractive approach for the design of novel treatments to enhance T lymphocyte restoration and effector T cell responses.
Blood samples were collected from HIV-1-positive subjects either naive from treatment (c-ART–, n = 39, CD4+ T cells counts (mean ± SD): 387±242 cells/µl; viral load (mean ± SD): 4,2±1,1 log HIV RNA copies /ml or stable under c-ART for more than 6 months (c-ART+, n = 39, CD4+ T cells counts (mean ± SD): 485±440 cells/µl ; viral load <1,6 log copies /ml), at the Hospital of Infectious Diseases, Sofia, Bulgaria and Henri Mondor Hospital, Créteil, France. Blood from 25 HIV-negative donors was obtained at the Regional Blood Transfusion Centre, Creteil, France. CD8+ and CD4+ T cells were purified using RosetteSep enrichment antibody cocktails (StemCell Technologies, Vancouver, BC, Canada) according to the manufacturer's instructions. CD4+CD25hi cells were further isolated with CD25 magnetic beads and two passages on MS columns (Miltenyi Biotec, Bergisch-Gladbach, Germany). The positive fraction contained >80% Treg expressing high levels of FoxP3 transcription factor as verified by flow cytometry (data not shown).
CD8+ T cells were stained with 0.5 mM CFSE (Molecular probes, Eugene OR, US) as previously described [46]. CFSE-labelled CD8+ T cells were cultivated in 96-well U-bottom plates, coated with 5 mg/mL anti-CD3 mAb (UCHT1; Beckman Coulter, Villepinte, France) in the presence or absence of Treg (total cell concentration 1.25×105/ml and final volume 200 ml and the Treg/Effector ratio was 1/4 as determined in previous studies [43], [44]). In some experiments, Treg were pre-incubated with 10 µg/ml of anti-CD39 (BY40, IgG1) or isotype control mAb for 15 min at 37°C, and added to CD8+ T cells without a washing step.
The effects of BY40 mAb on CD39 expression and inhibition of ATPase activity were evaluated using YT2C2 NK cell line (flow cytometry) and fresh monocytes using malachite green phosphate detection kit (R&D System, Minneapolis, USA), according to manufacturer's instruction (See methods in the legend of Fig. S3). To assess the effect of adenosine analogue CGS 21680, PBMC were pre-incubated for 1 h with different concentrations of either CGS 21680 (Sigma-Aldrich, Lyon, France) or DMSO as control. Cells were then stimulated with anti-CD3 for 5 days as described above. At day 2 of culture, DMSO and CGS 21680 were added in identical concentrations.
For intracellular staining (ICS), CD8+ T cells were stimulated in the presence or absence of Treg (Treg/effector ratio:1/4) overnight with a pool of whole Gag 15-mer peptides (2 µg/ml) supplemented with anti-CD28 and anti-CD49d antibodies (1 µg/ml of each). Brefeldine A (10 µg/ml) was added 1 h after the peptide stimulation. Cells were surface stained with anti-CD8 mAb and ICS was performed with PE-Cy7-conjugated IFN-γ, TNFα and IL-2 antibodies. When indicated, Treg were pre-incubated with 10 µg/ml of anti-CD39 mAb or isotype control for 15 min at 37°C, and added to CD8+ T cells without a washing step.
Total RNA was isolated from purified CD4+ and CD8+ T cells and RT-PCR was performed by the ABI Prism 7500 Sequence Detection System (Applied Biosystems, Courtaboeuf, France) in 50 µL reaction with Platinum SYBR Green qPCR SuperMix-UDG w/ROX (Invitrogen) and 0.2 µM of each primer. S14 mRNA which expression was found to be stable among the different group of patients was used as control to normalize each sample. Sequences of the A2AR- and S14-specific primers were forward: CGAGGGCTAAGGGCATCATTG, reverse: CTCCTTTGGCTGACCGCAGTT) and forward: GGCAGACCGAGATGAATCCTCA, reverse: CAGGTCCAGGGGTCTTGG TCC. The relative levels of A2AR mRNA were calculated using the 2−ΔΔCT method.
Anti-CD39-PE (clone TU66), anti-CD25-PC7, anti-CD4-FITC or Pac.blue, anti-CD8-PerCP, anti-CD3-APC, and CD28-PerCP-Cy5.5, were products of BD Biosciences (Le Pont de Claix, France),CD45RA-ECD from Beckman Coulter (Villepinte, France), and CD127-Biot/ strepta-APCCy5.5, FoxP3-Alexa 488, CCR7-APC-Alexa 750 from ebiosciences (Montrouge, France). Blocking anti-CD39 mAb (BY40) was produced in one of our laboratories (A.B) by immunizing mice with the YT2C2 NK cell line. BY40 is IgG1 monoclonal antibody, which is with BY12 mAb unique regarding its epitope mapping as we previously reported [47]. BY40 is not cytotoxic and it inhibits directly ATPase activities mediated by cell membrane anchored CD39 (AB personal data and this paper Fig. S3) Cells were analysed by LSR II (BD Immunocytometry systems). At least 20 000 CD4 or CD8-gated events were collected for cell surface studies.
Statistically significant differences were assessed by one-way ANOVA, followed by paired t-samples T-test, or by unpaired T-test assuming independent samples where appropriate. Correlations were assessed using Spearman's rank order test (GraphPad° Prism 5.0 statistical software). The independent prognostication value of CD39 expression on Treg was evaluated in multiple linear regression models (SPSS v.17.0).
(For more details, see previously published works [23], [50], [51]). For the GRIV (cases and controls) and ACS analyses, the CD39 genotyping data were obtained using the Illumina Infinium II HumanHap300 BeadChips, when for the MACS analysis, they were obtained using the Affymetrix GeneChip Human Mapping 500K Array. In each study, quality control filters (e.g. missingness, low minor allele frequency, Hardy-Weinberg equilibrium deviation) were applied to ensure reliable genotyping data as previously described [23], [50], [51]. In each cohort, potential population stratification was also considered using the Eigenstrat software [52]. First, to confirm continental ancestries, the genotypes of each participants group were combined with the genotypes from the three HapMap reference populations. Among the initial ACS group, 13 subjects were thus excluded from further analyses (n = 404) to avoid spurious associations resulting from a non-European ancestry. Then, in each study group of European descent, the top ten most significant principal components were identified and included as covariates in the regression models described below. The rs11188513 SNP untyped in the MACS group was imputed using Impute software [53] and the HapMap release 21 phased data for the population of European descent (CEU) as the reference panel.
We first performed a genetic case-control association analysis in the GRIV cohort using a logistic regression and an additive model, including as covariates the 10 principal components identified by Eigenstrat. All SNPs found to be significant in the GRIV cohort were tested for replication in ACS and MACS cohorts. The SNP rs11188513 was the only polymorphism exhibiting a significant p-value both in ACS and MACS. For the replication in the ACS and MACS groups, we performed Kaplan-Meier survival analysis and regression -Cox proportional regression and linear regression for ACS and MACS respectively- in an additive model including as covariates the 10 principal components identified by Eigenstrat. The significant associations (P<0.05) were also retested using age, sex, and CCR5-P1 and D32 polymorphisms as covariates and yielded identical results.
To evaluate the combined p-value obtained over the 3 cohorts for each SNP, we used the classical Fisher method [54].
Approval and written informed consent from all subjects were obtained before study initiation. The study was approved by the following ethical committees : Hospital of Infectious Diseases, Sofia, Bulgaria and CCP IX Ile de France - Henri Mondor Hospital, Créteil, France. Ethic statements for GRIV, MACS ACS cohorts have been already reported [23], [50], [51].
|
10.1371/journal.pbio.1001933 | The Cytoplasmic Capping Complex Assembles on Adapter Protein Nck1 Bound to the Proline-Rich C-Terminus of Mammalian Capping Enzyme | Cytoplasmic capping is catalyzed by a complex that contains capping enzyme (CE) and a kinase that converts RNA with a 5′-monophosphate end to a 5′ diphosphate for subsequent addition of guanylic acid (GMP). We identify the proline-rich C-terminus as a new domain of CE that is required for its participation in cytoplasmic capping, and show the cytoplasmic capping complex assembles on Nck1, an adapter protein with functions in translation and tyrosine kinase signaling. Binding is specific to Nck1 and is independent of RNA. We show by sedimentation and gel filtration that Nck1 and CE are together in a larger complex, that the complex can assemble in vitro on recombinant Nck1, and Nck1 knockdown disrupts the integrity of the complex. CE and the 5′ kinase are juxtaposed by binding to the adjacent domains of Nck1, and cap homeostasis is inhibited by Nck1 with inactivating mutations in each of these domains. These results identify a new domain of CE that is specific to its function in cytoplasmic capping, and a new role for Nck1 in regulating gene expression through its role as the scaffold for assembly of the cytoplasmic capping complex.
| We previously described a cyclical process of mRNA decapping and recapping termed “cap homeostasis.” Recapping is catalyzed by a complex of cytoplasmic proteins that includes the enzyme known to catalyze nuclear capping, and a kinase that converts RNA with a 5′-monophosphate end to a 5′-diphosphate capping substrate. The current study shows these two enzymatic activities are brought together in the cytoplasmic capping complex as both bind to adjacent domains of the adapter protein Nck1. Nck1 is a cytoplasmic protein best known for transducing receptor tyrosine kinase signaling. We identify a proline-rich sequence at the C-terminus of a human capping enzyme that is required for binding to Nck1, and we show that this interaction is required for integrity of the cytoplasmic capping complex. Depletion of Nck1 causes the cytoplasmic capping complex to dissociate. The inhibition of cytoplasmic capping by Nck1 with mutations in either the 5′-kinase or capping enzyme binding sites identified a functional role for Nck1 in cap homeostasis and a previously unknown function for Nck1 in cell biology.
| The 7-methylguanosine “cap” is a defining feature of all eukaryotic mRNAs, and the cap plays a role in almost every step of mRNA metabolism. In the nucleus, the cap is bound by a heterodimer of CBP80-CBP20, and its interaction with other proteins coordinates many of the subsequent steps in pre-mRNA processing and mRNA surveillance [1]. mRNAs are exported to the cytoplasm cap-end first, where the CBP80-CBP20 heterodimer is replaced by eIF4E, leading to translation initiation through the eIF4F complex. Translation and mRNA decay are interconnected processes, and for many transcripts loss of the cap is thought to be an irreversible step leading to mRNA decay [2].
Nuclear capping is catalyzed by the sequential actions of capping enzyme (RNGTT, RNA guanylyltransferase, and 5′-phosphatase, CE) and RNA cap methyltransferase (RNMT), both of which are positioned at the transcription start site by their binding to the C-terminal domain of RNA polymerase II [3]. A number of approaches use the cap to map transcription start sites. These include paired end analysis of transcription start sites (PEAT) [4], Capped analysis of gene expression (CAGE) [5], and RNA annotation and mapping of promoters for the analysis of gene expression (RAMPAGE) [6]. In human cells ∼72% of transcription start sites have matching CAGE data [7]. However, a significant number of CAGE tags do not correspond to transcription start sites [8], mapping instead to locations within the body of the transcript. Intriguingly, there is no evidence for downstream CAGE tags in the Drosophila transcriptome [6], suggesting that the presence of capped ends located downstream within the transcript body is unique to higher metazoans.
The decay of nonsense-containing human β-globin mRNA in erythroid cells results in the accumulation of a reproducible pattern of metastable decay intermediates that are missing sequences of their 5′ ends [9]. These were previously characterized as having a cap or cap-like structure on their 5′ ends [10], and it was in the course of re-examining this observation that we discovered cytoplasmic capping [11]. In nuclear capping, the diphosphate substrate for guanylic acid (GMP) addition is generated by hydrolysis of the β-γ phosphate bond on the 5′ ends of newly transcribed pre-mRNA, followed by the transfer of GMP bound covalently at lysine 294 to generate GpppX, where X is the 5′-most nucleotide. In cytoplasmic capping the proximal substrate for cytoplasmic capping is also a 5′-diphosphate, but this is generated by a 5′-monophosphate kinase that sediments with CE in a ∼140 kDa complex [11].
Our first in vivo experiments looked at the impact of inhibiting cytoplasmic capping on cellular recovery from stress. Stress was selected for study because it results in a generalized inhibition of translation, with non-translating mRNPs accumulating in P bodies and stress granules [12]. We reasoned that some transcripts might be stored in an uncapped state, and cytoplasmic capping might be required to restore these to the translating pool. Support for this hypothesis was seen in Otsuka and colleagues [11], where the ability of cells to recover from a brief arsenite stress was reduced by expression of an inactive form of capping enzyme (termed K294A) that is restricted to the cytoplasm by deletion of the nuclear localization sequence and addition of the HIV Rev nuclear export sequence. Proof that K294A expression inhibits cytoplasmic capping came from work in Mukherjee and colleagues [13]. The original purpose of that study was to identify mRNAs that are regulated by cytoplasmic capping, and in the course of doing so we discovered a cyclical process of decapping and recapping that we termed “cap homeostasis.” Cytoplasmic capping targets can be grouped into three categories on the basis of their cap status and stability in cells that are inhibited for cytoplasmic capping. One group of natively uncapped transcripts is destabilized when cytoplasmic capping is inhibited. In Mukherjee and colleagues [13] these are referred to as the “uninduced” pool. Another group has natively uncapped transcripts that are not destabilized. Instead, inhibition of cytoplasmic capping results in an increase in the uncapped population of each of the “common” transcripts. The third group accumulates uncapped forms only when cytoplasmic capping is inhibited. There is no change in the steady-state level of these “capping inhibited” mRNAs, and their uncapped forms accumulate in non-translating mRNPs. A number of these transcripts encode proteins that are involved in the mitotic cycle, which may explain the reduced survival of K294A-expressing cells after arsenite stress.
Ultimately, progress in understanding the function of cytoplasmic capping depends on identifying the components of the cytoplasmic capping complex and determining how the CE, the 5′-kinase that generates the diphosphate substrate, and a cap methyltransferase are brought together in a single complex. We noticed that modifications to the C-terminus reduced the relative amount of kinase and capping activity recovered from cytoplasmic extracts with CE. Because these modifications had no impact on covalent binding of GMP (i.e., guanylylation activity), this suggested the C-terminus might play a role in assembling the cytoplasmic capping complex. A search for functional domains identified a proline-rich SH3 binding site close to the C-terminus of vertebrate CE, but not in CE of lower metazoans. Drosophila CE has a run of three prolines, but there is an additional 34 amino acids that separates these from the C-terminus. The current study identifies this region as a third domain of CE that is bound by adapter protein Nck1, which in turn brings CE together with the 5′-monophosphate kinase to form the core of the cytoplasmic capping complex.
Murine and human CE have proline-rich sequences immediately upstream of their C-termini (Figure 1A), whereas the three prolines in Drosophila CE are separated from the C-terminus by 34 amino acids (Figure S1). To determine if differences here are relevant for cytoplasmic capping we examined the impact of modifying this portion of the protein (Figure 1C) on the in vitro activity of the cytoplasmic capping complex recovered from cells that were transfected with the constructs shown in Figure 1B. The proteins analyzed here included wild-type enzyme (CE), the same protein missing 25 amino acids from the C-terminus (CEΔ25C), the cytoplasmically restricted form of CE described in [11] (CEΔNLS+NES), which has an N-terminal Myc tag and a C-terminal FLAG tag, a similar construct without FLAG that has an added N-terminal sequence that becomes biotinylated in vivo (bio-cCE), the same construct missing the C-terminal 25 amino acids (bio-cCEΔ25C), and a construct similar to CEΔNLS+NES in which the C-terminal FLAG tag was replaced with one that is biotinylated (cCE-bio). These plasmids or a control expressing Myc-GFP were transfected into 293 cells and cytoplasmic forms of each of the epitope-tagged proteins and their associated partners was recovered using anti-Myc or streptavidin paramagnetic beads (Figure 1C, upper panel).
The first experiments examined the impact of the C-terminal modifications on covalent binding of GMP (guanylylation) to the active site lysine at 294 [14]. Proteins recovered from transfected cells were incubated with α-[32P]GTP and analyzed by SDS-PAGE and autoradiography (Figure 1C, middle panel). For the most part the amount of [32P]GMP bound covalently to each of these proteins matched the relative amount of protein determined by Western blotting (Figure 1C, upper panel), indicating the C-terminal modifications had little or no impact on the ability of these proteins to bind GMP. In Otsuka and colleagues [11] we described a functional in vitro capping assay that measures the labeling of a 23 nt long 5′-monophosphate RNA with α-[32P]GTP in a reaction containing unlabeled ATP. While the C-terminal modifications had relatively little impact on guanylylation activity they each reduced in vitro capping activity of the recovered proteins (Figure 1C, bottom panel).
The in vitro capping assay depends on the activity of a 5′-kinase to generate a diphosphate substrate for transfer of GMP [11]. The recovery of this activity with the different forms of CE was examined by incubating the recovered proteins and 5′-monophosphate RNA with γ-[32P]ATP (Figures 1D and S2). Again the Δ25C deletion had no impact on recovery of CE or guanylylation of the recovered protein; however, it resulted in the parallel loss of kinase and capping activities. The experiment in Figure S2 also included a [32P]labeled, capped human β-globin transcript that was added to each reaction to control for contaminating ribonuclease activity. The similar recovery of this RNA from each of the reactions confirmed that the differences seen with each of the C-terminal modifications were due to differences in activity of the complex.
MIT ScanSite [15] identified adapter protein Nck1 (NP_006144.1) as a potential binding partner for the proline-rich C-terminus. Nck1 has 3 SH3 domains and a single C-terminal SH2 domain, and it has roles in transducing tyrosine kinase signaling [16], in translation [17] and in development [18]. It is classified as a cytoplasmic protein, and this was confirmed by Western blotting of nuclear and cytoplasmic extracts and by indirect immunofluorescence (Figure S3). To determine if Nck1 binds the proline-rich C-terminus we examined its recovery with bio-cCE, bio-cCE missing the C-terminal 25 amino acids (bio-cCEΔ25C), or bio-cCE missing the five C-terminal proline residues (bio-cCEΔpro). Cytoplasmic extracts from cells that were co-transfected with plasmids expressing each of these forms of CE and HA-Nck1 were recovered on streptavidin beads and assayed for Nck1, guanylylation, and capping activities (Figure 2A). As in the preceding experiment loss of the proline-rich C-terminus sequence had no impact on guanylylation activity. However, each of the deletions affected the recovery of both Nck1 and in vitro capping activity. The recovery of Nck1 with cCE was unaffected by prior treatment with micrococcal nuclease (Figure 2B), indicating RNA is not required for the interaction between these proteins. The binding of HA-Nck1 to CE is also independent of GMP binding as changing the active site lysine to alanine (K294A) had no effect on its recovery (Figure S4).
We next examined if endogenous Nck1 also binds to cytoplasmic CE. In Figure 2C cells were transfected with plasmids expressing bio-cCE or a protein with two copies of MS2 binding protein fused to the same biotinylated peptide sequence [19]. Selective binding of cCE by endogenous Nck1 was confirmed by Western blotting of protein recovered on streptavidin beads with anti-Nck1. Most cells also express Nck2 (NCK adapter protein 2, NP_001004720.1) a structural and functional paralog with 68% sequence identity to Nck1, and Grb2 (CAG46740.1), which is similar except that it has only two SH3 domains. Even with prolonged exposures there was no evidence for recovery of Nck2 or Grb2 with cytoplasmic CE (Figure 2D). Lastly, the interaction of endogenous Nck1 with endogenous CE was confirmed by a guanylylation assay performed on complexes recovered by immunoprecipitation of cytoplasmic extract from non-transfected cells with anti-Nck1 antibody (Figure 2E).
Previous work showed that CE and the 5′-kinase activity co-sediment on glycerol gradients in a ∼140 kDa cytoplasmic complex [11]. To determine if Nck1 is part of the cytoplasmic capping complex, extract from bio-cCE-expressing cells was separated on a 10%–50% glycerol gradient and Western blotting was used to determine the sedimentation of each of these proteins in the input fractions (Figure 3A, upper panel), and in protein recovered on streptavidin beads (lower panel). A portion of Nck1 overlapped in the input fractions with bio-cCE, and the recovery of Nck1 with bio-cCE on streptavidin beads confirmed its presence in the cytoplasmic capping complex.
We next looked for the evidence of a native complex containing CE bound to Nck1. In the experiment in Figure 3B cytoplasmic extract from non-transfected cells was separated on a calibrated Sephacryl S-200 column, and individual fractions were analyzed by Western blotting for CE and Nck1. Both proteins eluted in the same fractions, at a size estimated from standards to be larger than that seen on glycerol gradients. The difference in size determinations may be due to the shape of the complex or the dissociation of one or more proteins during prolonged sedimentation. Nck1 was also present in later fractions, a result that is consistent with an excess of Nck1 over CE (see Discussion). To determine if CE was bound to Nck1 in the co-eluted fractions these were pooled, immunoprecipitated with control immunoglobulin G (IgG) or anti-Nck1, and the recovered proteins were analyzed by Western blotting with antibodies to both proteins (Figure 3C). The selective recovery of CE with Nck1 confirmed that the native proteins were indeed bound to each other in this complex.
The relative amount of Nck1 bound to CE was estimated by immunoprecipitating cytoplasmic extract from non-transfected cells with anti-CE antibody followed by Western blotting with anti-Nck1 antibody (Figure 3D). On the basis of signal intensity and the amount of protein loaded onto the gel, we estimate that 1% of Nck1 is bound to cytoplasmic CE.
Because Nck1 lacks catalytic activity its presence in the cytoplasmic capping complex suggested it might act as a scaffold to bring CE together with the 5′-kinase and perhaps other proteins. To test this concept we first asked if a functional complex could assemble in vitro on recombinant Nck1. Gst and Gst-Nck1 were expressed in Escherichia coli (Figure 4A, left panel) and bound to glutathione beads that were added to pre-cleared extracts from cells expressing bio-cCE or MS2-bio (Figure 4A, middle panel). Selective in vitro binding of bio-cCE to Nck1 was demonstrated by Western blotting with HRP-streptavidin (Figure 4A, right panel). Perhaps of greater importance, guanylylation assay showed that CE present in each of the extracts also bound selectively to Nck1 (Figure 4A, right middle panel, lanes 4 and 6). Finally, capping assay was performed on the bead-bound proteins to determine if all of the activities (i.e., CE plus the 5′-kinase) can assemble in vitro on Nck1. Bead-bound proteins were incubated with 5′-monophosphate RNA, ATP, and α-[32P]GTP, and the products were separated on a denaturing gel. The GMP labeling of 5′-monophosphate RNA by proteins recovered with Gst-Nck1 but not with Gst alone (Figure 4A right, bottom panel, lanes 4 and 6) supports the hypothesis that Nck1 functions as a scaffold for assembly of the cytoplasmic capping complex.
The preceding data also suggest that CE and the 5′-monophosphate kinase each bind to Nck1 but not to one another. If so, Nck1 knockdown should reduce the amount of kinase and capping activity recovered with bio-cCE. In the experiment in Figure 4B cells were transfected with bio-cCE and Nck1 siRNA or a scrambled control, and protein recovered on streptavidin beads was assayed by Western blotting, and for kinase activity and capping activity. Nck1 knockdown had no impact on the amount of bio-cCE or its recovery on streptavidin beads. However, significantly less kinase and capping activity were recovered in cells knocked down for Nck1 compared to the scrambled control (lower panels). Together with results in Figure 4A these data point to Nck1 as the scaffold that brings cytoplasmic CE together with the 5′-kinase to form a functional capping complex.
The CE-binding domain on Nck1 was identified by co-transfecting cells with bio-cCE and a panel of Nck1 constructs with inactivating mutations in each of the functional domains (Figure 5A) [20]. The almost complete loss of Nck1 mutated in the third SH3 domain from protein recovered on streptavidin beads identified this as the CE binding site (Figure 5B, M3, lane 4, 3SH3M, lane 6). The functional impact of this mutation was determined by assaying the recovery of kinase and capping activity with bio-cCE from cells expressing wild-type Nck1 or Nck1 with the CE-binding domain mutation (Figure 5C). In both cases the loss of Nck1 binding was matched by a similar loss in recovery of kinase activity and capping activity, thus confirming that Nck1 acts as a scaffold to bring the 5′-kinase together with cytoplasmic CE to form the cytoplasmic capping complex.
To determine which of the other SH3 domains binds the 5′-kinase activity cells we transfected cells with plasmids expressing HA-tagged wild-type Nck1, or Nck1 mutated in the first SH3 (M1) or second (M2) SH3 domain (Figure 5D, upper panel). Proteins were recovered on anti-HA beads and assayed for kinase activity by incubation with γ-[32P]ATP and a 23 nt 5′-monophosphate RNA, followed by denaturing gel electrophoresis of recovered RNA. 5′ kinase activity was recovered with wild-type Nck1 and Nck1 mutated in the first SH3 domain (M1), but not with Nck1 mutated in the second SH3 domain (M2, Figure 5D, lower panel). Thus, the core of the cytoplasmic capping complex consists of CE and the 5′-kinase bound to adjacent sites on Nck1.
We next sought to build on our success in knocking down Nck1 (Figure 4B) to demonstrate a functional role for Nck1 in cap homeostasis. However, Nck1 knockdown resulted in a general decrease in the steady-state level of every transcript examined, regardless of classification with respect to cytoplasmic capping (Figure S5). The reason for this is not known, but it does not appear to be related to cell viability, as this was unaffected by Nck1 knockdown (Figure S6). As an alternative we asked whether cap homeostasis could be disrupted by overexpression of Nck1 with inactivating mutations in the CE and 5′-kinase binding domains.
As noted in the Introduction, cytoplasmic capping targets can be categorized by differences in their behavior when cytoplasmic capping is inhibited. The “capping inhibited” pool make up the most obvious targets because stable uncapped forms of these transcripts appear when cytoplasmic capping is inhibited. Triplicate cultures of U2OS cells were transfected with plasmids expressing wild-type Nck1, or the M2 and M3 forms of Nck1, and their overexpression with respect to endogenous Nck1 was confirmed by Western blotting (Figure S7). We also confirmed that their overexpression did not have an inhibitory impact on their steady-state levels as seen with Nck1 knockdown. The appearance of uncapped transcripts was determined using an assay from our previous study in which these are ligated to an RNA adapter, hybridized to a biotinylated antisense DNA oligonucleotide, and recovered on streptavidin beads [21]. Each preparation included an internal control of uncapped β-globin RNA. In agreement with a central role for Nck1 in cap homeostasis, overexpression of Nck1 mutated in the CE (Figure 6A, M3) or 5′-kinase binding domain (Figure 6B, M2) resulted in the appearance of uncapped forms of each of four “capping inhibited” targets (DNAJB1, NM_006145; ILF2, NM_004515.3; MAPK1, NM_002745.4; and RAB1A, NM_004161.4).
Inhibition of cytoplasmic capping also results in the Xrn1-mediated degradation of natively uncapped transcripts in the “uninduced” pool. The impact of overexpressing the M3 (Figure 6C) and the M2 (Figure 6D) forms of Nck1 was examined for three of these transcripts (TLR1, NM_003263.3; NME9, NM_178130.2; S100Z, NM_130772.3). The steady-state level of each target RNA was reduced compared to wild-type control, again confirming cap homeostasis is inhibited by overexpression of Nck1 with mutations in the CE or 5′-kinase binding domain. The mutant forms of Nck1 had little impact on the steady-state level of MAPK1, a result that is consistent with the stability of the uncapped forms of this class of transcripts. We also looked at the impact of each of these proteins on steady-state levels of BOP1 (NM_015201.3), one of the control transcripts whose cap status is unaffected by changes in cytoplasmic capping. M3 overexpression had no impact on the level of BOP1 mRNA; however, BOP1 was unexpectedly increased in cells that overexpress the M2 form of Nck1. The reason for this is not known, but we suspect it may be a consequence of interfering with one of the other pathways in which Nck1 participates.
Prior to this study the functional domains of mammalian capping enzyme were limited to the N-terminal triphosphatase domain and the C-terminal guanylyltransferase domain. Our results add a third domain at the C-terminus of the mammalian protein whose binding to the third SH3 domain of Nck1 functions in the assembly of the cytoplasmic capping complex. In the course of this work we also discovered that C-terminal extensions (such as the FLAG tag on CEΔNLS+NES, Figure 1C and K294A) interfere with CE binding to Nck1. Because CE and the 5′-kinase activity bind to Nck1 rather than to each other this may explain why the cytoplasmic capping targets identified in Mukherjee and colleagues [13] have 5′-monophosphate ends rather than 5′-diphosphate ends as might be expected if these proteins interacted directly.
We realized early on that knowing the targets of cytoplasmic capping was a necessary first step toward validating the identity of proteins in the cytoplasmic capping complex. Results in [13] grouped the targets of cytoplasmic capping into three broad classes on the basis of the relative stability of their uncapped forms. Overexpression of Nck1 with inactivating mutations in the CE (Figure 6A) or the 5′-kinase-binding domains (Figure 6B) resulted in the appearance of uncapped forms of transcripts that also appear when cytoplasmic capping was blocked by induction of the inactive K294A form of cytoplasmic CE. Other cytoplasmic capping targets have natively uncapped forms that are degraded when cytoplasmic capping is inhibited. Together with the appearance of stable uncapped forms of the “capping inhibited” transcripts, the lower steady-state levels of these RNAs in cells expressing Nck1 with inactivating mutations in the CE (Figure 6C) or 5′-kinase (Figure 6D) binding sites provided in vivo confirmation of an essential role for Nck1 in cytoplasmic capping. Together, they confirm that Nck1 is essential for the assembly of the cytoplasmic capping complex and for cap homeostasis.
Nck1 is a ubiquitously expressed cytoplasmic protein that is best known for its role in transducing tyrosine kinase signaling [22],[23]. It also functions in the resolution of endoplasmic reticulum stress [24],[25] and it stimulates translation by binding to eIF2β [17]. On the organismal level Nck1 is required for proper mesoderm development [18] and in establishing neuronal circuitry [26]. Nck1 and its paralog Nck2 are elevated in many cancers, and their overexpression promotes malignant transformation [27]. Figure 7 presents a model of our current understanding of the cytoplasmic capping complex. By binding to adjacent SH3 domains, Nck1 juxtaposes CE and the 5′-kinase activity in a manner that is likely to facilitate the generation of a diphosphate capping substrate and the transfer of GMP onto this. Although Nck1 and Nck2 share 68% sequence identity CE only binds to Nck1 (Figure 2C). We know from results in [11] that the products of cytoplasmic capping are properly methylated, but have yet to confirm the identities of the 5′-kinase activity or the cap methyltransferase.
Approximately 25% of the 5′ ends identified by CAGE analysis of mammalian transcriptomes map within downstream exons rather than transcription start sites [7],[28]. Those findings are consistent with the recent identification of unique protein products translated from small ORFs located downstream of canonical start sites [29]–[32]. Our findings suggest that the absence of downstream CAGE tags in the Drosophila transcriptome [6] results from the inability of Drosophila CE to bind Nck1 and participate in the cytoplasmic capping complex. The presence of Nck1 at the core of the cytoplasmic capping complex also suggests that cytoplasmic capping, and perhaps the proteins translated from recapped transcripts, may vary in response to different stimuli, for example by activation of a particular receptor tyrosine kinase. Adding to this complexity, the amount of Nck1 is also regulated by the ubiquitin/proteasome pathway [33]. Ubquitination by the c-Cbl E3 ubiquitin ligase targets Nck1 for degradation by the proteasome, which in turn may impact cytoplasmic capping by reducing the amount of Nck1 that is available for bringing cytoplasmic CE together with the 5′-kinase. Nck1 ubiquitination is inhibited by the binding of synaptopodin, a proline-rich actin binding protein, to the same site on Nck1 as the 5′-kinase activity. These findings raise the possibility of competition by these proteins, and of a link between cytoplasmic capping and the cytoskeleton.
pcDNA3 myc-mCE [11] was used as a template to amplify myc-mCEΔ25C using the primers T7 and YO125 containing Kpn1 and Apa1 sites, respectively. The amplified PCR product was digested with Kpn1 and Apa1 and then ligated into similarly digested pcDNA3. The C-terminal FLAG tag was removed from pcDNA4/TO-myc-NES-mCEΔNLS-Flag [11] by amplifying the region containing Myc-cCE with primers Tevbio-cCE-F and Tevbio-cCE-R. This was further modified by addition of a C-terminal tag containing a site for cleavage by Tev protease and a peptide that is biotinylated in vivo [34]. The plasmid pFA6a-HTB-hphMX4 containing this sequence was provided by Peter Kaiser (University of California, Irvine). The biotinylation tag sequence was amplified from this plasmid using cCE-TevBio-F and cCE-TevBio-R. The PCR products from the preceding two reactions were mixed together and PCR amplified with primers cCE-F and R-Tev Bio to create Myc-cCE-Tev-Biotin. This was digested by Kpn I and Apa I and cloned in similarly digested pcDNA3/TO vector yielding pcDNA3/TO myc-cCE-bio (cCE-bio). To generate CE with an N-terminal biotinylation tag (bio-cCE), the sequence corresponding to the tag only (without the Tev cleavage site) was amplified from the above template with primers Bio RP and Bio FP. This tag was introduced into pcDNA3 myc-NES-mCEΔNLS, pcDNA3 myc-NES-mCE (K294A) ΔNLS [11] and pcDNA3 myc-mCEΔ25C using the In-Fusion HD cloning kit (Clontech). The NES from construct used in our previous work was added to pcDNA3 bio-myc-mCEΔ25C to generate pcDNA3 bio-myc-NES-mCEΔ25C (bio-cCEΔ25C). Bio-cCEΔpro was generated from bio-cCE by site directed mutagenesis using the QuikChange Site-Directed Mutagenesis kit (Stratagene) with oligos Amp-R (Stratagene) and cCE-ΔPPP. The plasmid for expression of GFP (pcDNA4/TO myc-GFP-Flag) was described previously [11]. All constructs were verified by sequencing. Sequences of oligos used in the study are listed in Table S1. The human Nck1 constructs were described in [20] and kindly provided by Wei Li and Louise Larose [17]. In these the SH2 domain in SH2M was inactivated by an arginine-to-lysine mutation in the sequence FLVRES, and the SH3 domains in M1, M2, M3, and 3SH3M were each inactivated by changing the first tryptophan in the WW motif to lysine. The MS2-Biotin construct was provided by Marion Waterman, University of California, Irvine [19].
U2OS and HEK-293 cells were grown in McCoy's 5A medium (Invitrogen) containing 10% fetal bovine serum. 1×106 HEK293 or 2×106 U2OS cells in log phase growth were transfected with 8 µg (total) of plasmid DNA using FuGENE 6 (Promega) following the manufacturer's protocol. Cells were harvested 36 h post-transfection. Transfection efficiency (typically 95% for 293 cells and 70% for U2OS cells) was determined by parallel transfection with a GFP-expressing plasmid.
Cells were lysed using 1× lysis buffer (20 mM Tris-HCl [pH 7.5], 10 mM NaCl, 10 mM MgCl2, 10 mM KCl, 0.2% NP40, 1 mM PMSF [Sigma], 1× protease inhibitor [Sigma], 1× phosphatase inhibitor cocktail II and III [Sigma], and 80 units/ml RNase out [Invitrogen]). These were placed on ice for 5 min, the tubes were gently flicked and incubated on ice for an additional 5 min. The lysates were centrifuged at 5,000 g at 4°C for 10 min to pellet the nuclei, and the supernatant (cytoplasmic) fractions were transferred to chilled microcentrifuge tubes. Cell lysate was used directly for Immunoprecipitation and Gst pull-down assays except for experiments where extracts were treated with micrococcal nuclease to remove nucleic acid. RNA was recovered from cytoplasmic fractions with Trizol reagent (Invitrogen) as per the manufacturer's instructions. The recovered RNA was treated with DNase I and poly(A) RNA was selected using Dynabeads mRNA DIRECT kit (Invitrogen) according to the manufacturer's instructions. Cells were fractionated into cytoplasmic and nuclear fractions by NE-PER kit using manufacturer's (Pierce) protocol for gel filtration and subcellular localization of Nck1.
Cytoplasmic extracts from transiently transfected HEK293 were layered onto freshly prepared 10%–50% linear glycerol gradients and centrifuged at 200,000 g for 22 h at 4°C. Each experiment included a gradient containing gel filtration standards as described in [11] for use as reference points for determining size as a function of position within the gradient. 500 µl fractions were collected from the top of the gradient and stored at −80°C. Protein present in 100 µl of each fraction was first recovered by precipitation with ten volumes of ethanol prior to Western blot analysis, and remaining fractions were used to recover cytoplasmic capping enzyme on streptavidin beads.
Two mg of HEK293 cell cytoplasmic extract was filtered through a 0.2 µ low protein binding filter (Millipore) and then loaded at 4°C into a calibrated HiPrep 16/60 Sephacryl S-200 High Resolution column (GE) in 15 mM Tris-HCl (pH 7.5), 150 mM NaCl. The column was developed in the same buffer at a flow rate of 0.5 ml/min, and starting from the void volume (elution volume 37 ml) 0.5 ml fractions were collected up to an elution volume of 62 ml in order to separate monomeric proteins as small as 29 kDa. 250 µl from each fraction was TCA precipitated and analyzed by Western blotting with anti-CE and anti-Nck1 antibodies. The remaining 250 µl of the four fractions indicated with a box in Figure 3B were pooled and immunoprecipitated with control IgG or anti-Nck1 antibody and Dynabeads protein G. The recovered proteins were analyzed by Western blotting with anti-CE and anti-Nck1 antibodies.
E.coli BL21(DE3)pLysS cells (Promega) were transformed with plasmid pGEX-2TK-Nck1or pGEX-2TK. These were grown in Luria Bertani broth containing 100 µg/ml ampicillin and 50 µg/ml chloramphenicol. Cells were induced at an OD600 of 0.5 with 0.5 mM IPTG at 18°C and cultured overnight. Cells were lysed in GST-lysis buffer (20 mM Tris-HCl [pH 7.5], 20 mM NaCl, 1 mM DTT, 1 mM PMSF, 1% Triton-X). Gst or Gst-Nck1 expressed in E. coli were bound with gentle rocking to glutathione Sepharose for 2 h at 4°C. The beads were washed 5× with Gst wash buffer (25 mM Tris-HCl [pH 7.5], 150 mM NaCl, 0.1% NP40) to remove unbound proteins and stored at 4°C in the same buffer containing 20% glycerol and 1 mM phenylmethylsulfonyl fluoride. Cytoplasmic extracts were prepared 36 h after transfection of HEK293 with plasmids expressing bio-cCE or MS2-bio. Nonspecific proteins were removed by incubating each extract with glutathione-Sepharose, and 5 mg of pre-cleared extract was incubated with rocking for 2 h at 4°C with Gst- or Gst-Nck1-bound beads. These were recovered by centrifugation and unbound proteins were removed by five washes with Gst wash buffer. The bead-bound proteins were analyzed by Western blotting, and for guanylylation and capping activity [11].
For immunoprecipitation reactions, cytoplasmic extract was pre-cleared with mouse IgG antibody coupled beads (Cell Signaling). The pre-cleared lysate was incubated with anti-Myc or HA monoclonal antibody coupled beads for 2 h at 4°C in a rocking platform. Biotinylated proteins were recovered by incubating extracts for the same time as above with Streptavidin Dynabeads (T1, Invitrogen). For immunoprecipitation with anti-Nck1 or anti-CE antibodies cytoplasmic extract was first pre-cleared with Protein G Dynabeads (Invitrogen), followed by addition of control rabbit IgG or antibody IgG bound Protein G Dynabeads. The reactions were incubated overnight at 4°C in a rocking platform before washing and recovery of antibody-bound complexes. In each experiment bead bound proteins were washed and processed for western blotting or in vitro guanylylation, kinase and capping reactions [11].
Mouse anti-Myc monoclonal antibody, rabbit IgG, anti-Myc coupled agarose beads and rabbit anti-HA antibodies were obtained from Santa Cruz Biotechnology. Anti-HA antibody coupled magnetic beads were purchased from Pierce. Mouse monoclonal antibodies to β-tubulin and GAPDH were obtained from Sigma. Mouse anti-HA monoclonal antibody was purchased from Roche. Monoclonal Rabbit anti-Nck1 was obtained from Cell Signaling and polyclonal rabbit anti-Nck1 antibody used for immunoprecipitation was obtained from Millipore. Rabbit polyclonal antibody to Nck2 was obtained from Upstate Biotechnology and rabbit polyclonal antibody to CE was purchased from Novus. Rabbit polyclonal antibody to Grb2 was provided by Ramesh K Ganju, Ohio State University. Conformation specific mouse anti-rabbit antibody was obtained from Cell Signaling. Alexafluor coupled goat anti-rabbit IgG (680), goat anti-mouse (800), goat anti-rabbit IgG (488), goat anti-mouse IgG (594) and streptavidin (800) were purchased from Molecular Probes (Invitrogen). HRP coupled goat anti- rabbit antibody and HRP-streptavidin were obtained from Santa Cruz Biotechnology.
5% input (otherwise indicated) and 70% of each immunoprecipitated sample were denatured in 2× Laemmli buffer (Bio-Rad Laboratories) containing β-mercaptoethanol and incubated at 95°C for 5 min prior to electrophoresis on 10% Mini-PROTEAN TGX precast gels (Bio-Rad Laboratories). Proteins were transferred onto Immobilon-P PVDF membrane (EMD Millipore), which were blocked using 3% bovine serum albumin in Tris-buffered saline (TBS). This was followed by incubation with primary antibody for 2 h in blocking solution containing 0.05% Tween-20 (TBS-T). These were then washed with TBS-T, incubated for 1 h with HRP or Alexafluor-coupled secondary antibody (1∶10,000 dilutions), and visualized on X-ray film (GeneMate) after detection with ECL-plus detection system (GE Healthcare), or with a Licor Odyssey imager. To prevent detection of antibody denatured chains in immunoblots, conformation specific mouse anti-rabbit antibody was used following incubation with primary antibody. The antibody dilutions used for Western blots are as follows: NCK1, 1∶2,000; NCK2, 1∶2,000; Grb2, 1∶1,000; HA, 1∶2,000; Myc, 1∶1,000; GAPDH, 1∶5,000; β-tubulin, 1∶5,000; Streptavidin HRP, 1∶10,000; conformation specific mouse anti-rabbit antibody, 1∶2,000.
U2OS cells grown on coverslips were fixed with methanol at −20°C for 10 min and probed with a 1∶50 dilution of rabbit anti-Nck1 and a 1∶200 dilution of mouse anti-α tubulin. Secondary antibodies consisting of Alexafluor 488 goat anti-rabbit IgG or Alexafluor 594 anti-mouse IgG were used at 1∶1,000 dilution. All images were acquired at ambient temperature using an Olympus IX-81 microscope, with 100× Plan Apo oil immersion objective (1.4 numerical aperture) and a QCAM Retiga Exi FAST 1394 camera, and analyzed using the Slidebook software package (Intelligent Imaging Innovations).
A total of 50 ng of cytoplasmic poly (A) RNA was used to synthesize cDNA with Superscript III reverse transcriptase (Invitrogen) according to the manufacturer's instructions. q-PCR was performed with 2×Sensi-FAST Sybr No Rox Mix (Bioline) using an Illumina Eco system. Uncapped RNAs were recovered by a ligation based approach as described in [21] and analyzed by qRT-PCR as described in [13].
HEK293 or U2OS cells were transfected with 10 nM of Nck1 siRNA (J006354-09, Dharmacon) and a scramble control (Dharmacon) using Lipofectamine RNAimax (Invitrogen) according to the manufacturer's protocol. Cells were harvested 72 h after transfection. Knockdown efficiency was monitored by Western blotting with rabbit anti-Nck1 antibody. The impact of Nck1 knockdown on transcript levels was determined by qRT-PCR using primers listed in [13]. The impact of Nck1 knockdown on assembly of the cytoplasmic capping complex was determined by recovery with bio-cCE that was introduced into cells 48 h after transfection with Nck1 siRNA and 12 h before recovering complexes on streptavidin beads.
Data are shown as the representative result or as mean of at least three independent experiments ± standard deviation. Statistical analyses were performed using Student's unpaired two-tailed t test. Differences were considered significant at p<0.05. Graphs were generated using GraphPad Prism 5 (GraphPad Software, Inc.) and bars represent standard deviation.
|
10.1371/journal.pntd.0006318 | Dynamic spatiotemporal analysis of indigenous dengue fever at street-level in Guangzhou city, China | This study aimed to investigate the spatiotemporal clustering and socio-environmental factors associated with dengue fever (DF) incidence rates at street level in Guangzhou city, China.
Spatiotemporal scan technique was applied to identify the high risk region of DF. Multiple regression model was used to identify the socio-environmental factors associated with DF infection. A Poisson regression model was employed to examine the spatiotemporal patterns in the spread of DF.
Spatial clusters of DF were primarily concentrated at the southwest part of Guangzhou city. Age group (65+ years) (Odd Ratio (OR) = 1.49, 95% Confidence Interval (CI) = 1.13 to 2.03), floating population (OR = 1.09, 95% CI = 1.05 to 1.15), low-education (OR = 1.08, 95% CI = 1.01 to 1.16) and non-agriculture (OR = 1.07, 95% CI = 1.03 to 1.11) were associated with DF transmission. Poisson regression results indicated that changes in DF incidence rates were significantly associated with longitude (β = -5.08, P<0.01) and latitude (β = -1.99, P<0.01).
The study demonstrated that social-environmental factors may play an important role in DF transmission in Guangzhou. As geographic range of notified DF has significantly expanded over recent years, an early warning systems based on spatiotemporal model with socio-environmental is urgently needed to improve the effectiveness and efficiency of dengue control and prevention.
| Dengue fever (DF) as a mosquito-borne viral disease remains a challenge for the prevention and control caused by the increased population, global development, human movement, and urbanization in the last five decades. The largest DF outbreak occurred with more than 40,000 cases in Guangdong in 2014 since DF re-emerged in China. The accurately spatiotemporal identification of DF transmission and the related socio-environmental factors are considered to be important for the strategy decision-making of the official government. This study first identified the spatiotemporal pattern and socio-environmental factors associated with DF occurrence at street and daily level in Guangzhou, China from 2006 to 2014, using spatiotemporal scan statistical methods. The results suggested that DF control should be targeted in the southwest of Guangzhou during autumn, particularly 75 high risk streets. We found that the aged population, floating population, low-education population and the non-agricultural population significantly contributed to the DF clustering risk at street level. Finally, a spread trend of DF toward southwest part of Guangzhou was noticed. These results could be implemented towards prevention and control measures of DF in high-risk areas in Guangzhou.
| Dengue fever (DF) is a widespread vector-borne viral infectious disease which has a rapidly increase in infections, geographic distribution, and the severity cases[1]. The rapidly expanding global footprint of DF has evolved to a major public health problem due to increased geographical extension, climate changes, population growth and global travel in the last 50 years [2]. DF is endemic and has been reported in more than 100 countries including the southeast Asia, the Americas, the western Pacific, Africa [3]. 3.9 billion people are at the potential risk of DF in these endemic regions [4]. The high economic burden brought could not been neglected [5].
Historically, DF has re-emerged in China in 1978, from its first appearance in Foshan city of Guangdong province and then subsequently it has been reported in other areas such as Guangdong, Guangxi province and Hainan island after 32 years [6]. Since then, DF outbreak and epidemics were reported every year affecting several thousands of people, predominantly in the southeast coastal regions including Hainan, Guangxi, Fujian, Zhejiang and Yunnan provinces [7]. It was assumed that the large-scale epidemics occurred before 1990s was due to the imported dengue virus [8]. During the period 1978–2008, a total of 655,324 cases including 610 deaths were recorded by Guangdong province Health Department. Vector-borne scientists have predicted that DF could potentially become an endemic disease in China [9]. For example, in 2014, a large outbreak with more than 37,000 cases has occurred in Guangzhou city [10].
Due to the lack of effective vaccine and antiviral treatment, vector control is considered as a useful measure towards prevention of dengue disease [11]. DF epidemics in the different districts appeared not homogenous, due to the change of the transmission pattern of spatial and time [11]. However, the spatial clusters, socio-environmental factors at the new and smallest administrative unit (street level) and the temporal cluster at daily level in Guangzhou have not been explored in this epidemic regions. To help decision-makers or policy-makers in targeting the prevention and control areas and reduce the economic burden, vector control techniques could be selectively applied at high-risk areas or clusters of DF. Hence, this study aimed to examine the spatiotemporal pattern of DF using spatiotemporal scan technique at street-level[10,12,13], to identify the socio-environmental risk factors of DF and to explore the spread of DF over the study period for improving prevention and control of DF and guiding to future study.
Ethical approval for this project was approved by Sun Yat-Sen University Ethical Review Committee (Approval No: 2015024) and all of the data analyzed were anonymized.
Guangzhou, as the third-largest city in China and the world-famous trade port, located at the Pearl River Delta Region of Guangdong province and spanned from 112° 57' to 114° 03' E longitude and 22° 26' to 23° 56' N latitude [14] (Fig 1). The total area under the city's administration is 7,434.4 square kilometers and the permanent resident population is 12,700,800 (2010) [15]. Guangzhou city has 12 districts and 166 streets. The permanent resident population of each street ranged from 3397 to 391287 (2010) [16]. Monthly averages range from 13.6°C in January to 28.6°C in July, while the annual mean is 22.6°C [14], the relative humidity is approximately 68%, whereas annual rainfall in the metropolitan area is over 1,700 mm [14].
Daily data on indigenous DF cases were collected from China Notifiable Disease Surveillance System and Guangzhou Center for Disease Control and Prevention (CDC) for the years 2006 to 2014. There were 240 cases with unknown street-level address in 2014. These cases were excluded in this study. DF cases were diagnosed according to the national diagnostic criteria of DF, including the epidemiological exposure history, clinical manifestations and laboratory confirmation [10].
The street-level geographic vector polygon map of Guangzhou city was obtained from Guangzhou CDC and the latitude and longitude of the centroid of each street were calculated directly in the ArcGIS 10.0 software. The counts number of the indigenous DF cases were aggregated to counts at the street-level. Street-wise socio-demographic data was retrieved from the demographic bulletin of the 6th National Population Census [17]. Data on the urban-rural structure of communities was collected from the National Bureau of Statistics of People's Republic of China [18]. The location of all cases were matched to the street-level vector map based on their home addresses.
The annual occurrence of street-wise first indigenous DF cases were mapped along with the date of onset.
A retrospective spatiotemporal scan test was implemented using SaTScan (Version 9.4.1) software. Firstly, the spatiotemporal cluster analysis of DF in Guangzhou from 2006 to 2014 was conducted annually. In brief, DF case, population and coordinates data were used as inputs in SaTScan.
Scanning window for the spatiotemporal scanning method is the spatial scan combining with temporal scan. The scan window is a cylinder. The base of the cylinder is circle which represents the spatial dimension, and the height of the cylinder represents the temporal dimension. The radius of the circle varied from zero to the maximum spatial cluster size of 50% of the population at risk which could avoid pre-selection bias. In this study, the heights of the cylinder were varied daily from zero to 1 year. The results with the statistical significance of p-value were reported by Monte Carlo simulation replication at 9999. The maximum log likelihood ratio (LLR) calculated in Poisson distribution is considered as the most likely cluster.
The secondary clusters are defined as the second maximum LLR estimated by poisson model [19].
In this study, a holistic purely spatial cluster analysis from 2006 to 2014 was implemented with the same upper limits in the spatial window. ArcGIS (Version 10.3.1) were used to convert the outputs of scan analysis into maps and visualize the spatial and temporal clusters.
Univariate logistic regression and a stepwise logistic regression model were conducted to explore the relationship between the socio-environmental risk factors and the street with DF cases at high risk and low risk. Dichotomous dependent variable was set based on relative risks (RRs) of each street from the purely spatial cluster analysis result. The streets with RRs ≥1 were assigned “1” and those with RRs <1 were assigned “0”. The potential socio-environmental risk factors included at street-level were as following: percentage of people in each age-group; floating population; non-agriculture population; percentage of people with lower education (lower than undergraduate); percentage of different type communities (urban communities, urban-rural communities and rural communities) in all of the communities in each street. The floating population is defined as the people living in the street currently whose census registers were recorded in other street of the district in Guangdong province. There are two type of the census registers including agriculture and non-agriculture in China. The non-agriculture population was defined as the people whose census registers were recorded in the urban, not in the rural.
The variations in the distribution of DF along the latitude and longitude of streets centroids were detected using Poisson model during the study period [20]. To explore the difference of DF distribution in the last three years and the first six years, we divided the study period into two periods: period 1 is from 2006 to 2011 and period 2 was from 2012 to 2014. The dependent variable in this modeling was the differences of DF annual mean incidence rates of the all the streets which occurred DF epidemic between the period 1 and period 2 in Guangzhou.
The epidemic pattern of daily indigenous DF cases fluctuated during 2006 to 2014 with three major outbreaks in 2006, 2013 and 2014 (Fig 2). The number of DF cases ranged from 0 to 1,627 cases daily (mean = 52.9, SD = 182.48). Interestingly, outbreaks showed an increasing trend after 2010. Fig 2 also displayed the daily variability of the number of streets with infected cases from 2006 to 2014. The peaks in DF cases generally coincided with streets of high DF cases.
The spread of indigenous DF incidence rates in each high-risk street was displayed in Fig 3. All streets in Yuexiu, Liwan and Haizhu district, several streets in Baiyun, Panyu and Tianhe districts and streets in Huangpu, Luogang and Nansha district had relatively high DF spread. The streets with highest increase in DF were located in Baiyun, Panyu and Huangpu district. Baiyun districts included the streets with highest spread.
Fig 4A showed the spatial distribution of high-risk areas or clusters of DF at street-wise. There were 75 high risk streets (RRs ≥ 1) in the southwest of Guangzhou city. These streets were located mostly in Yuexiu, Liwan and Haizhu district, the southern part of Baiyun, the northern part of Panyu, Tianhe and Huangpu district. Fig 4B depicts the sum of daily indigenous DF cases of the streets with RRs <1 and RRs ≥ 1 during the study period.
Spatial and temporal clusters of indigenous DF cases were showed in Fig 5A and 5B, respectively. The most likely clusters (n = 9) were detected each year during 2006 to 2014 (P<0.01) and the secondary clusters (n = 2) were identified in 2006 and 2013 (P<0.01) (Table 1). The most likely clusters were concentrated in streets of Yuexiu, Liwan and Haizhu districts. In 2006, the most likely cluster included the southern Panyu district and part of the southern Nansha district whereas the secondary cluster included the northern Conghua district. In 2014, the most likely clusters included the farther northern Baiyun district with the secondary clusters in the northern Zengcheng district (Table 1).
The significant temporal clusters were found in autumn season, i.e., late August to early November during 2006 to 2014, except in 2008 and 2009. Fig 5C shows the streets with the occurrence of first indigenous DF cases each year. The first indigenous DF cases occurred within or close to the spatial cluster circles yearly, except in 2014, where it occurred in the distant Nansha district.
The results of univariate and step-wise logistic regression model analyses were presented in Table 2. In the univariate analysis, the age-groups, the percentage of non-agricultural population and the urban-rural population per street had significant association with DF risk: 0–14 years (OR = 0.84, 95% CI = 0.75 to 0.94), 15–64 years (OR = 0.94, 95% CI = 0.88 to 0.99), urban-rural communities (OR = 0.97, 95%CI = 0.95 to 0.98) and rural communities (OR = 0.95, 95% CI = 0.93 to 0.97) had negative association with DF risk whereas 65+ years (OR = 1.26, 95% CI = 1.15 to 1.39), nonagricultural population (OR = 1.05, 95%CI = 1.04 to 1.07) and urban communities (OR = 1.03, 95% CI = 1.02 to 1.05) had positive association with DF risk. After the stepwise variable selection, four variables were entered into the multivariate logistic regression model. The results demonstrated that DF was statistically significantly associated with population belonging to 65+ years (OR = 1.49, 95% CI = 1.13 to 2.03), floating population (OR = 1.09, 95% CI = 1.05 to 1.15), non-agricultural population (OR = 1.07, 95% CI = 1.03 to 1.11) and low-education population (OR = 1.08, 95% CI = 1.01 to 1.16).
A statistically significant and negative association was obtained between the spread of DF incidence rates and longitudes (β = -5.08, P < 0.01) and latitudes of the streets (β = -1.99, P < 0.01) (Table 3). The results indicated that DF incidence rates increased with the areas geographically variation which may provide with the information of target streets for DF prevention and control in the future.
The results of this study suggested that DF incidence rates in the different districts appeared to be heterogeneous which was due to the changes in the transmission pattern of DF spatially and temporarily. A previous study has indicated that the prevention and control strategies towards DF will depend on high-risk and low-risk clusters [21]. Understanding and identifying the potential spatial and temporal clusters of DF transmission is the fundamental measure for surveillance and control [22]. A couple of studies have conducted cluster analysis of DF in Guangdong [23,24]. Previous research identified six risk factors for DF infection in Pearl River Delta [25] based on 2013 dengue surveillance data, which may improve our comprehension of the differences and socio-environmental factors on DF incidence rates. But in addition, few other studies have demonstrated that socio-demographic factors, such as population growth, levels of education, demographic structure and urbanization could influence the DF spread [26–30]. However, our research used a dynamic spatial and temporal analysis based on long term data (ie., January 2006 and December 2014) to detect the spatial clusters of DF and identify associated socio-environmental factors at a street level in Guangzhou. Moreover, Guangzhou was struck by an exceptionally severe outbreak in 2014, resulting in almost 40,000 laboratory-confirmed DF cases. This outbreak is the largest and most severe epidemic of dengue fever ever documented in China, with incidence rates exceeded the combined total of all previous years [31,32].
This study detected spatial clusters of DF high risk regions in Guangzhou city and suggested the geographic range of notified dengue cases has significantly expanded over recent years. Relative importance of risk factors may vary across space and time. This finding will provide useful information for developing dynamic early warning system for DF transmission. We have performed stepwise logistic regression model as this technique was applied in the vector-borne diseases research. Our results demonstrated that old aged population (65+ years), floating population, low-education people and non-agriculture people were the potential determinants for the spread of DF.
DF transmission has been reported in both rural and urban areas, and the dengue viruses have fully adapted to a human-Aedes aegypti-human transmission cycle, previous studies showed that the urbanization was linked to the DF incidence rates [33,34]. Guangzhou, as a large urban center of the tropics, where crowded human populations, especially nonagricultural population, live in intimate association with equally large mosquito populations. This setting provides the ideal home for maintenance of the viruses and the periodic generation of epidemic strains. In this longitudinal study, the result indicated nonagricultural population was positively related with DF risk, the central urban area and the old city area were the high-risk areas, where most aged (65+ years) Guangzhou residents lived. The streets with high nonagricultural population in Guangzhou normally have higher population density and poor housing conditions and less environmental management.
Previous studies have suggested that the accumulation of a susceptible population was essential to trigger DF epidemics [35]. In this study, a large number of floating population may be more susceptible for DF transmission. Residents, especially the aged, have the habit of planting flowers or hydrophyte in flowerpots or in household courtyards in Guangzhou. Several studies have identified the vegetation and breeding mosquitoes to DF that “vegetation can provide resting or feeding sites for mosquitoes or can serve as a proxy for the presence of breeding sites." Water storage, containers with an abundance of organic matter (e.g. those used for striking plant cuttings) or those amongst foliage or under trees (e.g. discarded plastic). As such progeny have been linked to a greater risk [36]. These containers with water provide a suitable breeding condition for mosquitoes. The water landscape and afforest landscape around the houses were also a perfect breeding habitat for mosquitoes. In addition, the movement of aged population may be limited to house surroundings and nearby areas, thus, increasing the chances of exposing themselves to mosquitoes. People with low-education generally have lack of knowledge and practices on the prevention measures of DF. These people usually work as laborers and spend most of their time outside, this in turn, may have given the possibility of being bitten by the mosquitoes. Another possible reason could be that these people live in rented apartments where the sanitary conditions are sub-optimal, thus this may have increased the chances of mosquitoes breeding and exposure.
The results from temporal cluster analysis indicated that the DF clusters occurred mainly in autumn, particularly, in late August to early November. Indigenous DF cases peaked seasonally despite limited intra-annual climatic variability and seasonal fluctuations. In addition, the availability of immature densities of Aedes albopictus (primary vector in Guangzhou) was consistent with the dengue seasonality [37,38] as the vector biology and viral replication are temperature and moisture dependent [39,40]. These results could be used in planning future prevention and control measures towards DF, particularly, during the high-risk season.
The consistent occurrence of first indigenous DF case within or close to the spatiotemporal clusters during the study period, except in 2014 requires further investigation. Over all, in the high risk streets, there were more indigenous DF cases than in the low risk streets: The cases in high risk streets occurred earlier and accelerated faster than those in the low risk streets as well. Without considering the number of cases, similar waves and crests were found in 2 sorts of streets. This could be due to the daily movements of working people from their living areas to working areas, i.e., the high-risk areas.
We observed an interesting result in the epidemic patterns of DF incidence rates during the study period. If the first case occurred in early summer, i.e., June or July, large outbreaks often occurred. For example, large epidemics in 2006, 2012, 2013 and 2014 were initiated with the occurrence of first case in June, July, July and June respectively. Although there were not many DF cases in 2012, the longest cluster period of DF was observed. On the contrary, if the first case occurred too early and too late, the large outbreaks often could not be triggered. In 2007 and 2010 epidemics, the first case occurred in April whereas in 2008, 2009 and 2011 epidemics, the first case occurred in November, August and September. If the first case occurs too early, the local department of health may plan to provide early warnings of DF outbreaks and implement prevention and control measures, whereas if it occurred too late, the reduced density of mosquito and the capacity of virus loading could help to decrease the risk of a large DF outbreak. Although imported cases was considered as an important trigger for the DF outbreak in Guangzhou, scientists could not confirm whether or not the dengue outbreaks in Guangzhou were initially triggered by the imported cases [39]. So other uncertainties of DF outbreak are still unknown and needs further studies.
In recent years, the impact of climate change on the transmission of mosquito-borne diseases has been studied in China [40]. Our results showed significant variation in the spatial distribution of DF in Guangzhou and that the geographic range of notified cases has expanded in this city (from south towards north and concentrate on the southwestern Guangzhou city) over the study period. Previous study reveal the movement tracks of the centre of mass for annual incidence rate of DF at municipality level in China, showing that the geographic expansion of dengue epidemics, such as gradually shifting from southern China (Guangdong, Guangxi, and Hainan) to northeastern China (Fujian and Zhejiang) and southwestern China (Yunnan) [41]. The associations between the spread of DF incidence rates and longitude and latitude were observed in this work, also demonstrated that DF has spread towards the southwestern Guangzhou city during the study period. Dengue is a complex disease and the spatiotemporal distribution involves socio-environmental factors, such as climate change, population movement, mosquito density and urbanization. Hence, future studies should include the impact of climatic and entomological factors on the transmission of DF in Guangzhou city.
To our knowledge, this is the first study to investigate the spatiotemporal clusters of DF and assess the socio-environmental factors in Guangzhou city using the spatial techniques at street-level. The study provides readily accessible information on DF spread and GIS maps on high-risk areas which can be used by the local Department of Health towards prevention and control of DF in Guangzhou. There are two limitations in this study: 1) Model included few variables on socio-environmental factors, as it was difficult to obtain all other street-level data. 2) As this study is an ecological study, measurement and information biases are possible. For example, the data on the socio-demographic factors were only obtained from the 6th Nation Population Census (collected in 2010) as the national demographic census in China was only conducted once 10 years. The socio-demographic data varied by time in Guangzhou and may have little impact on our results. However, we believe that the relative changes by different street level is unlikely to change dramatically in Guangzhou. We obtained the floating population in Guangzhou between January 1st 2006 and December 31st 2014 by accessing the registers at the online Guangzhou Statistics Bureau website (http://www.gzstats.gov.cn/). In addition, under-reporting is most likely possible as people with sub-clinical symptoms usually do not seek medical attention. The biases and drawbacks of stepwise multiple regression are well established within the statistical literature, including bias in parameter estimation, inconsistencies among model selection algorithms, etc. Whittingham et, al. discussed these issue and showed that stepwise regression allows models containing significant predictors to be obtained from each year's data [42]. In this study, we conducted stepwise logistic regression model as this technique was applied in the vector-borne diseases research, so as to select the main risk factors and develop predictive model. The spatial-temporal analysis presented in this paper differs from the one by explaining the observed distribution and perhaps ultimately permitting prediction.
In conclusion, this study has detected spatiotemporal clusters and variation of DF epidemics, and assessed socio-environmental risk factors for DF in Guangzhou city. These results could be implemented towards prevention and control measures of DF in high-risk areas in Guangzhou.
|
10.1371/journal.pgen.1007743 | Loss of the Mia40a oxidoreductase leads to hepato-pancreatic insufficiency in zebrafish | Development and function of tissues and organs are powered by the activity of mitochondria. In humans, inherited genetic mutations that lead to progressive mitochondrial pathology often manifest during infancy and can lead to death, reflecting the indispensable nature of mitochondrial biogenesis and function. Here, we describe a zebrafish mutant for the gene mia40a (chchd4a), the life-essential homologue of the evolutionarily conserved Mia40 oxidoreductase which drives the biogenesis of cysteine-rich mitochondrial proteins. We report that mia40a mutant animals undergo progressive cellular respiration defects and develop enlarged mitochondria in skeletal muscles before their ultimate death at the larval stage. We generated a deep transcriptomic and proteomic resource that allowed us to identify abnormalities in the development and physiology of endodermal organs, in particular the liver and pancreas. We identify the acinar cells of the exocrine pancreas to be severely affected by mutations in the MIA pathway. Our data contribute to a better understanding of the molecular, cellular and organismal effects of mitochondrial deficiency, important for the accurate diagnosis and future treatment strategies of mitochondrial diseases.
| Mitochondrial pathologies which result from mutations in the nuclear DNA remain incurable and often lead to death. As mitochondria play various roles in cellular and tissue-specific contexts, the symptoms of mitochondrial pathologies can differ between patients. Thus, diagnosis and treatment of mitochondrial disorders remain challenging. To enhance this, the generation of new models that explore and define the consequences of mitochondria insufficiencies is of central importance. Here, we present a mia40a zebrafish mutant as a model for mitochondrial dysfunction, caused by an imbalance in mitochondrial protein biogenesis. This mutant shares characteristics with existing reports on mitochondria dysfunction, and has led us to identify novel phenotypes such as enlarged mitochondrial clusters in skeletal muscles. In addition, our transcriptomics and proteomics data contribute important findings to the existing knowledge on how faulty mitochondria impinge on vertebrate development in molecular, tissue and organ specific contexts.
| Mitochondria are important organelles with multiple cellular functions, serving as energy- and biosynthetic centres, participating in Ca2+ signalling and cell stress responses, and executing cell death. Thus, compromised mitochondrial activity is associated with numerous human pathologies. A group of genetically inherited diseases that are relatively rare but present devastating symptoms include encephalomyopathies, Leber’s Hereditary Optic Neuropathy (LHON), Leigh Syndrome and Barth Syndrome [1–3]. Another group of pathologies relate to more common, age-related and genetically predisposed conditions. These are often associated with neurodegeneration, such as Parkinson’s and Alzheimer’s disease, or Amyotrophic Lateral Sclerosis (ALS) [4–6]. Finally, given that mitochondria orchestrate cellular metabolism, pathological conditions with symptoms of altered metabolism, including diabetes, obesity, cancer, as well as aging processes, have also been associated with mitochondrial dysfunction [7, 8]. Tissues with high energy demands such as the heart, brain, muscles and liver are typically most affected in patients with mitochondrial disease [1–3]. The fact that treatment of mitochondrial disorders is mostly directed at mitigating their symptoms reflects our relatively poor understanding of the biology of these disorders. It also underlines the necessity to generate new tools and models to study the molecular, cellular and organ related consequences of mitochondrial malfunction in complex organisms.
The mitochondrial proteome is of dual origin. While a handful of proteins are encoded by mitochondrial DNA (mtDNA), the vast majority are encoded by nuclear DNA. Thus, genetically determined mitochondrial diseases are heterogeneous and can be the consequence of mutations in either genome [1, 2, 7–9]. Nuclear-encoded mitochondrial proteins are synthesised on cytosolic ribosomes. Efficient import of such precursor proteins from the cytosol into their destined mitochondrial sub-compartments is critical to maintain the biogenesis of the complex organelle and attributed to the interplay of mitochondrial translocases [10–13]. The vast majority of our knowledge on the mechanism and function of the evolutionarily conserved import machineries stems from studies in unicellular organisms, such as the yeast S. cerevisiae [11]. However, the effect of impaired mitochondrial protein biogenesis on the development of vertebrates remains largely unexplored.
Mitochondrial translocases sort precursor proteins according to signals harboured within the polypeptide sequence. The import and maturation of cysteine-rich mitochondrial precursor proteins relies on the Mitochondrial Intermembrane space and Assembly (MIA) pathway. Mia40 (Chchd4), a central oxidoreductase of the mitochondrial intermembrane space, drives these final steps of the biogenesis of cysteine-rich mitochondrial proteins [11, 14–17]. The oxidation-coupled import of MIA substrates leaves Mia40 at a reduced state which is regenerated by its partner Augmenter of Liver Regeneration (Erv1/ALR) [18, 19]. Importantly, the majority of Mia40 substrates are directly or indirectly involved in the assembly of the respiratory complexes, which are often disturbed in pathological conditions [14, 20–22].
Animal models have historically been a source of knowledge on the mechanisms of a plethora of human pathologies. In zebrafish, mitochondrial dysfunction has been predominantly modelled by anti-sense approaches, transiently mimicking a deficiency of mitochondrial protein associated with a human disease [23–29]. In addition, defects in mitochondrial proteins were found to underlie specific phenotypes identified in forward genetic screens. Studies on the oliver mutant identified Tomm22, a component of the translocase of the outer mitochondrial membrane, to be essential for hepatocyte survival and consequently for the development of the liver in zebrafish [30]. Xavier and dark xavier zebrafish mutants carry mutations in genes encoding the electron transfer flavoprotein dehydrogenase (etfdh) [31] and the electron transfer flavoprotein a (etfa) [32], respectively. These mutations trigger phenotypic abnormalities similar to those observed in the human Multiple Acyl-CoA Dehydrogenase Deficiency syndrome (MADD), proving the zebrafish to be a valid model to study the highly evolutionarily conserved processes that operate in mitochondria. A study on mtDNA deficiency, modelled in zebrafish by introducing mutations in the gene that encodes the mtDNA polymerase (DNA polymerase gamma, polg), shows that mutant larvae were able to survive to the juvenile stage despite a progressive decrease in the level of mtDNA [33]. The biochemical and morphological phenotype of early onset Parkinson disease was recapitulated in zebrafish by introducing mutations in the PTEN-induced kinase-encoding gene (pink1) [34, 35], further supporting the use of zebrafish to model human pathologies.
The MIA pathway has been previously modulated in zebrafish by targeting the Mia40 partner Erv1/ALR. Using anti-sense technology, it was reported that Erv1/ALR plays an important role in the development of the liver and the pancreas [36]. Another approach using a chemical inhibitor of Erv1/ALR and a translation-blocking morpholino suggested a role during heart development [37]. However, with the recent advancement in genome engineering techniques, phenotypic discrepancies between knockdown and knockout are frequently reported in zebrafish [38, 39]. To date, no study investigated how mutations in genes encoding MIA pathway components impinge on the development of vertebrates. In this study, we used Transcription Activator-Like Effector Nucleases (TALENs) to generate mia40 mutants in zebrafish to model mitochondrial dysfunction caused by an imbalance in mitochondrial protein biogenesis. We first characterised the two paralogues of mia40 in zebrafish, mia40a and mia40b. We showed that both paralogues exhibit Mia40 activity, as they both rescued the lethal phenotype of S. cerevisiae devoid of endogenous MIA40. By generating mutations in both paralogues, we also found that mia40a, but not mia40b, is essential for survival in zebrafish. At the cellular level, genetic disruption of the MIA pathway led to abnormal, enlarged mitochondrial structures in skeletal muscles. At the organismal level, the resulting mitochondrial deficiency triggered a glycolytic phenotype and ultimately starvation. Global transcriptomics and proteomics analyses indicated abnormal development of the liver and the pancreas in mia40a mutants. Taken together, our findings contribute to the understanding of the consequences of compromised mitochondrial biogenesis at the organismal level.
We investigated the role of mitochondrial biogenesis in vertebrate development by compromising the MIA pathway in Danio rerio. First, we focused on characterising the zebrafish Mia40 protein. As a consequence of a teleost-specific genome duplication [40], the zebrafish genome contains two copies of the mia40 (chchd4) gene, referred to as mia40a (chchd4a) and mia40b (chchd4b). When comparing the protein sequences of Mia40a and Mia40b with the human, mouse, frog and yeast orthologues (S1A Fig) it is apparent that both zebrafish protein paralogues retained the conserved cysteine motifs, the redox-active cysteine-proline-cysteine (CPC) and the twin CX9C motifs (Fig 1A). These motifs underlie the enzymatic activity of Mia40 and its intermembrane-space import and retention, respectively (Fig 1A and S1A Fig). Nonetheless, the protein sequence alignment (S1A Fig) and the phylogenetic analysis (S1B Fig) reveal significant differences between the two zebrafish paralogues, including amino-acid residue identity of less than 50%.
To determine whether the two proteins are enzymatically active, we performed a yeast complementation assay. The two zebrafish paralogues were expressed separately in yeast S. cerevisiae deleted for MIA40, under the endogenous promoter of yeast MIA40. We observed that expression of mia40a or mia40b partially rescued the lethal phenotype of the loss of Mia40 in yeast (Fig 1B, compare right panel lane 1 with 3 and 5). In contrast to its homologues in higher eukaryotes, yeast Mia40 is not a soluble protein in the intermembrane space. Due to an N-terminally located presequence and transmembrane domain, it is instead anchored in the inner mitochondrial membrane (S1A Fig) [41]. To faithfully mimic the import and maturation of yeast Mia40, we fused the zebrafish mia40a or mia40b with the cytochrome b2 bipartite signal sequence and expressed it in Mia40-deficient yeast [41]. We observed that both fusion proteins were able to sustain yeast viability at a similar level to that of the endogenous protein (Fig 1B, compare right panel lane 2 with 4 and 6). Thus, we conclude that both zebrafish paralogues, mia40a and mia40b, exhibit Mia40 activity.
To characterise the function of the mia40 genes in zebrafish, we first examined their spatiotemporal expression pattern. We performed RT-PCR analysis on RNA isolated from zebrafish embryos at various developmental stages, from the zygote to 3 days post fertilization (dpf). We found that mRNA for both paralogues was maternally contributed and expression persisted until at least 3 dpf (Fig 1C). Next, we compared the expression of mia40a and mia40b mRNA to the level of a known, abundant mRNA of the housekeeping gene encoding the 60S ribosomal protein L13a (rpl13a) by quantitative PCR (RT-qPCR). While the expression of both zebrafish paralogues was lower compared to the reference rpl13a, mia40a transcript was more abundant compared to mia40b (Fig 1D).
We next sought to dissect the spatiotemporal distribution of mia40a and mia40b transcripts in zebrafish larvae by in situ hybridization with transcript-specific RNA probes. We observed that mia40a was strongly expressed in organs of high-energy demand including somites, brain and the eye at 24 hours post fertilization (hpf) (Fig 1E). In contrast, mia40b transcripts were expressed in the head and eye, but absent from developing somites at 24 hpf (Fig 1E). At 3 dpf, both transcripts were expressed in the liver and eye. Although mia40a expression was strongly reduced at later time points, we conclude that its expression was generally broader compared to that of mia40b. At 5 dpf, mia40b expression was detected in the head, branchial arches, epidermis, and lateral line primordia (Fig 1E).
To evaluate the loss of Mia40 function in vivo, we used TALEN gene editing to generate zebrafish lines with mutations in the two mia40 paralogues. The third exon of mia40a was targeted for mutagenesis (Fig 2A). A TALEN was designed to target the sequence encoding the conserved phenylalanine 68, located on the hydrophobic concave surface that supports Mia40 interaction with its substrates (Fig 2A) [42–44]. We recovered a delta 8 (bns292) indel in mia40a. The predicted protein product in mia40abns292 mutants (also referred to as mia40a-/- or mia40a mutants) contains a frameshift mutation in the 67th codon leading to a premature stop codon (p. Gln67Gly*9) (Fig 2A). In parallel, the third exon of mia40b paralogue was TALEN- targeted. A delta10 (bns293) indel was recovered, preceding the CPC-coding region of mia40b. The predicted protein product in mia40bbns293 mutants (also referred to as mia40b-/- or mia40b mutants) contains a frameshift mutation and yields a truncated protein product (p.Glu48Gly*32) (Fig 2B).
By in-crossing the mia40a+/bns292 or mia40b+/bns293 zebrafish, we observed no severe developmental abnormalities among siblings up to 5 dpf (mia40a: Fig 2C). However, while mia40b-/- larvae survive to adulthood and are fertile, no individuals homozygous for the mutation in mia40a are recovered. To monitor in detail the survival of homozygous mutants of mia40a or mia40b, we genotyped the larvae at 72 hpf by fin-fold amputation, separated larvae according to genotype and monitored their growth. We noticed that in contrast to mia40b homozygous mutants, mia40a-/- larvae succumb to death starting at 10 dpf, reaching 50% lethality by 2 weeks of age (Fig 2D). The observation that mia40b-/- show no growth retardation and that wild-type Mia40b protein is not able to compensate for the loss of Mia40a in mia40a mutants provides evidence for non-redundant roles of these paralogues in zebrafish.
The skeletal muscles of zebrafish are rich in mitochondria [45, 46]. We used the zebrafish Tg(Xla.Eef1a1:mlsEGFP) line [47] to study mitochondrial morphology in the skeletal muscles of the mia40a mutants. This transgenic line expresses a mitochondrial matrix-targeted GFP under a ubiquitous promoter. We observe prominent, GFP-positive inclusions in skeletal muscle cells of the mia40a-/- but not in mia40a+/+ or mia40a+/- siblings (Fig 3A). Inclusion-positive mutant larvae do not show any abnormalities in the organisation of the actin filaments in the muscles, as visualized by phalloidin staining (Fig 3A). Interestingly we were able to rescue the inclusion phenotype at 3 dpf by injection of wild-type mia40a or mia40b mRNA at the 1-cell stage (S2A and S2B Fig, left panels). In contrast, the injection of mRNA encoding the mutant alleles had no effect on the presence of inclusions (S2A and S2B Fig, right panels).
The import of mitochondrial matrix-targeted precursor proteins requires an electrochemical membrane potential across the inner mitochondrial membrane. This potential is generated by functional electron transport chain (ETC). Therefore, we asked whether the GFP-positive inclusions are mitochondrial or cytosolic pools of GFP that fail to be imported into the mitochondrial matrix due to ETC insufficiency. Using an antibody against Tomm20, a component of the translocase of the outer membrane (TOM) complex, we observed the GFP signal to be surrounded by the outer membrane in commonly occurring mitochondria as well as GFP-positive, enlarged structures (Fig 3B, bottom panel, arrowhead). From these data, we conclude that mia40a deficiency results in large inclusions that represent malformed mitochondria.
Transmission electron microscopy (TEM) of 5 dpf old larvae revealed osmiophilic inclusions in the mitochondria of the skeletal muscles in the tail of mia40a mutant larvae (Fig 3C). The inclusions have a resemblance of myelin-like bodies (Fig 3C, inset). Apart from these inclusions, mutant mitochondria show well-conserved morphology in general, with well-defined cristae. Some mutant mitochondria present less well-compacted outer membrane compared to the wild-type siblings. Skeletal sarcomeres are well-structured and well-defined in mia40a-/- larvae (Fig 3C), suggesting that muscle structure is intact, in line with conclusions derived from phalloidin staining experiments (Fig 3A).
We next investigated mitochondrial function in mia40a mutants and measured respiration that reflects mitochondrial energetics in vivo. Interestingly, we found that MIA pathway insufficiency does not trigger changes in the basal oxygen consumption rate at 5 dpf (Fig 4A). Mitochondrial spare respiratory capacity (SRC) represents a measure for the ability to respond to an increased energy demand and is established by subtracting the basal respiration from maximal respiration. Maximal respiration can be measured upon uncoupling of the mitochondrial membrane potential using the FCCP ionophore. Importantly, unlike wild-type and heterozygous siblings, exposure to FCCP did not trigger an increase in respiration rate in mia40a mutants (Fig 4A). Taken together, these data suggest that mia40a mutants fully utilize mitochondrial capacity to meet respiratory demands of 5 dpf larvae with little or no spare capacity. Unexpectedly, we found that the relative level of mtDNA was increased in the mia40a-/- larvae at this stage (S3A Fig).
The yolk of zebrafish larvae is nearly depleted at 5 dpf, which coincides with a peak in whole body glucose content in larvae that have never been fed. After 5 dpf, the larvae start fasting and glucose levels decrease unless exotrophic nutrition is introduced [48]. We analysed the whole-body glucose levels in 5 dpf larvae and found that mia40a mutants present a significant decrease in glucose content compared to wild-type and heterozygous siblings (Fig 4B). As the yolk size was similar between the genotypes at all time points analysed, we conclude that glucose is consumed as an alternative energy source via the glycolytic pathway. Indeed, at 5 dpf, the level of lactate, a by-product of glycolysis, is significantly increased in mia40-/- larvae (Fig 4C), further supporting this conclusion.
We next investigated the metabolic state of mutant larvae at later time points. At 10 dpf, a drop in mia40a mutant survival was observed (Fig 2D). Measuring basal respiration in living larvae at 10 dpf, we found that mia40a mutants respire at levels similar to those of the starved wild-type and heterozygous siblings (S3B Fig). Upon introduction of exotrophic nutrition from 5 dpf onwards, mia40a+/+ and mia40a+/- larvae significantly increase oxygen consumption (S3B Fig). In contrast, homozygous mutants retain respiration at an equal level to starved animals (S3B Fig). We thus conclude that post 5 dpf, when yolk-derived nutrition is exhausted, phenotypes in mia40a mutant larvae are likely consequences of starvation. As predicted from this conclusion, glucose levels are barely detected in the mia40a homozygous mutants at 10 dpf (S3C Fig).
Based on the observations that mia40a mutants show challenged cellular respiration already at 5 dpf and its progressive decrease later in the development, we conclude that they represent a valid model of mitochondrial insufficiency caused by compromised mitochondrial protein biogenesis. In support, we have observed that subunits of respiratory complex I (Ndufa9) and complex IV (Cox4i1) are reduced in mia40a mutants at 5 dpf (S4 Fig) compared to other mitochondrial proteins (Atp5a) and cytosolic controls (Rpl7, β Actin) in line with previous reports [20].
To gain more insights into molecular processes and pathways affected, we determined global changes at the transcriptomic and proteomic levels by isolating total RNA and proteins at 5 and 8 dpf from mia40a mutants (Mut) and wild-type (WT) siblings. We used RNA sequencing (RNA-Seq) and quantitative proteomic analysis to identify differentially expressed genes and proteins between mia40a mutants and corresponding wild-type siblings. In total, the differential analysis showed expression changes of 236 genes and 213 proteins (203 unique genes) at 5 dpf and 277 genes and 320 proteins at 8 dpf (FDR 5%; log2FC = < -0.9 or > = 0.9) (Fig 5A, S1 Table). Generally, we observed that the proteomics approach yielded a higher proportion of differentially regulated and predominantly decreased features (in red, right bars, Fig 5A), hinting at a post-translational origin of the differences. We proceeded to identify what proportion of the differentially expressed genes is targeted to mitochondria. Similarly, proteomics yielded more mitochondrial features (up to 27% for 5 dpf) compared to transcriptome profiling (less than 10% for both time points), suggesting that the response triggered by the depletion of Mia40a is more pronounced at the proteomic rather than transcriptomic level (Fig 5B, S2 Table). These observations further strengthen the hypothesis that post-translational processes are likely involved in the mitochondrial phenotype. As the cellular milieu for Mia40a activity are mitochondria and Mia40a deficiency affects post-translational protein processing, proteomics yielded a higher percentage of mitochondrial proteins among those strongly affected in the mutant (Fig 5B, compare bottom panels to top panels). Due to the possible bias from starvation on the expression of genes and proteins in 8 dpf mia40a mutant larvae, further analysis focused on 5 dpf samples.
Next, we carefully investigated the expression changes for MIA pathway substrates [15, 21, 49–51] and noticed a more severe response at the proteomic level (Fig 5C). The proteins identified as substantially depleted, included several classical CX3C motif-containing chaperones of the intermembrane space (Timms), as well as important components of the respiratory chain, represented by such proteins as Cox6b, Ndufa8 or Ndufs5 (Fig 5C). Interestingly, the transcriptomic response reveals mild, but global, upregulation of the MIA pathway substrates, with the mia40a transcript showing the strongest upregulation (3-fold) (Fig 5C).
To further determine the function of differentially expressed genes and proteins we performed a Gene Ontology (GO) enrichment analysis. The upregulated genes were largely involved in amino-acid activation and tRNA metabolic processes (Fig 5D), recapitulating recently reported observations using transcriptome profiling in tissue culture models of MIA dysfunction [52]. Interestingly, proteolysis was identified as the top-rated GO term for downregulated genes (Fig 5D). Enriched proteins were mainly associated with lipid transport and other metabolic processes (Fig 5E). In contrast, downregulated proteins were notably enriched in mitochondrial terms including electron transport chain, cellular respiration and ATP metabolic process (Fig 5E). This observation was supported by steady state level analysis of protein lysates obtained from zebrafish larvae at 5 dpf using antibodies with confirmed specificity in zebrafish (S4 Fig).
We next performed an analysis which aimed at detecting tissue and organ-specific phenotypic effects in mia40a mutant larvae by comparing differentially expressed features (FDR 5%; log2FC = < -0.9 or > = 0.9) for 5 and 8 dpf samples in the transcriptomic and proteomic responses (Fig 6A). The cellular-retinol binding protein 2b (Rbp2b) was identified to be downregulated in all datasets (Fig 6A). Rbp2b plays a key role in retinol homeostasis and is expressed predominantly in the liver [53]. In situ hybridization confirmed a substantial decrease in abundance of rbp2b mRNA in the mia40a mutant liver compared to heterozygous and wild-type siblings (Fig 6B). Following this finding, we observed that the transcript for ceruloplasmin (cp), a gene encoding a liver-specific marker [54], is also decreased in mia40a mutant larvae at 5 dpf (Fig 6B), as were other liver-specific transcripts (S5 Fig). These findings suggest that the changes observed at the global proteomic level could be a consequence of tissue-specific downregulation of the transcript level as exemplified by rbp2b.
To further explore biological pathways directly affected by the loss of Mia40a, we performed an enrichment analysis using the Kyoto Encyclopaedia of Genes and Genomes (KEGG). This analysis revealed no enriched pathways for upregulated proteins in 5 dpf samples, whereas interesting results were obtained for depleted proteins (Fig 6C). Oxidative phosphorylation (OXPHOS) was the most affected metabolic pathway, followed by several neurodegenerative diseases that are known to coincide with defects in OXPHOS (Fig 6C). KEGG enrichment analysis also revealed an interesting link between mitochondrial and pancreatic dysfunction. In fact, pancreatic secretion, as well as protein digestion and absorption were also identified among pathways enriched in downregulated genes (in bold Fig 6C, S6 Fig) of 5 dpf samples.
We further explored the resulting hypothesis of pancreatic insufficiency by cross-referencing our results to published transcriptome profiles for pancreatic cell types in zebrafish [55], which mainly include genes encoding various digestive enzymes and their precursors including trypsin (try), chymotrypsinogen B1 (ctrb1), elastases (ela2, ela2l, ela3l) and pancreatic alpha-amylase (amy2a). Interestingly, this analysis yielded 23 acinar-specific genes as downregulated (Fig 6D), 13 (56%) of which are known to be evolutionary conserved across zebrafish, human and mouse (Fig 6D, in bold) [55]. Using a probe against trypsin to visualize the exocrine pancreas [56], we validated the omics-derived hypothesis and observed the pancreas to be smaller in mia40a mutant larvae as compared to heterozygous and homozygous siblings (Fig 6E). In separate experiments, we measured the pancreas size and normalized it to the larvae body length. We observed a nearly 40% decrease in the pancreas to body length ratio specifically in mia40a mutant larvae (Fig 6F), which suggests the difference in pancreatic size does not arise from a general delay in development. In conclusion, we found that Mia40a-induced mitochondrial insufficiency influences the development of endodermal organs such as pancreas.
In this study, we provide evidence for the requirement of a functional MIA pathway for zebrafish development and survival. We have shown that both paralogues of zebrafish mia40 appear to retain their activity as they sustain growth of yeast strains which lack endogenous MIA40 (Fig 1B) [14], providing evidence for MIA conservation during evolution. We found that expression of mia40a suffice for zebrafish development and survival in conditions when mia40b is mutated, but not vice versa, suggesting a non-redundant role for the paralogues. We think that the lethal phenotype of mia40a mutants can be explained by the fact that the mia40a transcript is expressed more abundantly and in more tissue types in zebrafish (Fig 1C–1E). Nonetheless, by overexpressing wild-type mia40b we were able to rescue the appearance of GFP-positive inclusions in the mia40a mutants in the transgenic background (S2B Fig).
Mia40 deletion in mice causes respiratory chain complex CI-20 defects and developmental arrest at gastrulation [20], whereas our zebrafish mia40a mutant does not show any strong retardation until much later in development (Fig 2B and 2C). The phenotypic differences between mice and zebrafish, observed also in mtDNA polymerase gamma (polg) mutants [33, 57], are interesting and difficult to explain. One can hypothesize that zebrafish embryos, through the adaptation to the water environment, can better tolerate low oxygen levels [58, 59] and as such do not rely on cellular respiration during early development to the extent that the mice do. Additionally, the zebrafish zygote is preloaded with maternally-contributed mitochondria and mia40a transcript (Fig 1C). When combined, these effects possibly suffice for the initial steps of embryogenesis. This view is supported by the observation that mtDNA copy number drastically decreases during the first 24 hpf and genes responsible for mtDNA replication are expressed later in development [60]. Another possibility is the ability of the developing zebrafish embryo to retrieve energy from other sources, including glycolysis. In support of this hypothesis, we observe that mia40a mutants respire at levels comparable to the heterozygous and wild-type siblings at 5 dpf (Fig 4A), yet the mutants have to take advantage of the entire respiratory capacity of mitochondria to achieve that (Fig 4A) and already utilize glucose as an energy source (Fig 4B and 4C). Surprisingly, we observed an increase in the mtDNA content at this time point (S3A Fig). It is intriguing to hypothesize that by upregulating mtDNA synthesis, mia40a mutants attempt to compensate for reduced mitochondrial respiration capacity, compromised by insufficient import of MIA substrates. Whether the enlarged mitochondria structures in the skeletal muscle of the zebrafish mutant represent a further compensatory mechanism is an interesting possibility that may gain support from previous studies, which showed that mitochondrial fusion has a protective function in sustaining mtDNA stability in the skeletal muscle [61].
Our transcriptomics and proteomics data point to a stronger mitochondrial-specific response at the protein level (Fig 5B), in line with a recent omics report of hearts from mouse mitochondrial mutants [62] and studies which modelled compromised mitochondria using various stressors in mammalian cell culture [52]. We provide an elegant validation of the cellular respiration phenotype by our proteomics data enrichment analysis (Fig 5E) and provide a more specific analysis of Mia40 substrates, which are strongly downregulated at the protein level in our zebrafish mutant (Fig 5C). Interestingly, the transcripts presented the opposite tendency, with mutated mia40a transcript being most significantly increased (Fig 5C). This may suggest yet another level of attempted compensation for the loss of Mia40a activity.
We were surprised to not observe acute molecular hallmarks of programmed cell death in the 8 dpf enrichment analysis in the mia40a mutants, suggesting that the previously reported interaction between Mia40 and AIF plays only a role in mitochondrial bioenergetics [20]. Lack of apoptotic signalling may also be explained by the fact that whole larvae were analysed, which hampers the detection of processes in specific or more strongly affected tissues. Based on metabolite analysis (S3B and S3C Fig), we expected the 8 dpf larvae to undergo starvation and thus focused our interpretation on the 5 dpf data. From our omics approaches it is evident that endodermal organs such as the liver and pancreas are affected in the mia40a-/- larvae already at that stage. It has previously been reported that mutations in the gene tomm20, which encodes a component of the protein translocase of the outer mitochondrial membrane, plays a role in hepatocyte survival in zebrafish [30]. Moreover, studies that aimed at transient knock-down of Erv1/ALR, the partner of Mia40, reported a role in liver and pancreas development in zebrafish [36]. Here, we expand on these reports and show that the acinar cells specifically are the cell type most strongly affected in the mia40a mutant (Fig 6D).
The question remains why acinar cells specifically suffer most prominently from MIA dysfunction. Mitochondria are grouped in three functionally distinct subpopulations in these highly polarized cells, forming perigranular, perinuclear and subplasmalemmal mitochondria [63, 64]. This mitochondrial compartmentalization is of physiological importance and enhances optimal Ca2+ signalling and ATP supply, which are instrumental in governing the basic function of the exocrine gland [63, 65]. We thus hypothesize that MIA dysfunction can contribute on both the functional and structural level to the acinar cell-specific phenotype in our mutant. First, we have shown that mitochondrial function is compromised in our mutant (Fig 4 and S3 Fig) and we expect that cells and tissues that strongly rely on mitochondrial biogenesis to be most affected by functional shortage. In support, it has been reported that genes responsible for mtDNA replication are expressed in the exocrine pancreas from 4 dpf to meet the increased demand for mitochondrial biogenesis in this organ at this developmental stage [60]. The authors suggest an interesting idea that this boost in expression of mtDNA replication genes can enhance mitochondrial biogenesis to support the excessive adaptation of the exocrine organ for the hydrolysis of external food [60]. Second, Mia40 has been shown to control the biogenesis of MICU1, a regulator of the mitochondrial Ca2+ uniporter (MCU) responsible for mitochondrial calcium intake [66]. Thus, the mia40a mutation could trigger functional abnormalities especially in Ca2+ signalling-dependent acinar cells. Finally, several mutants identified in ENU mutagenesis screens showed a smaller pancreas phenotype [67–69]. Interestingly, the mitomess (mms) zebrafish mutant, which presents dileted mitochondria in acinar cells, was additionally characterised with decreased levels of acinar-specific markers, as is observed in our mia40a-/- model [67]. The gene underlying the mitomess phenotype has not yet been identified and it is intriguing to speculate it may be a component of the MIA pathway [67].
Dysfunction of acinar cells and the exocrine pancreas as observed in this study have been previously reported as resulting from mutations in mitochondrial-encoded proteins, including in such pathologies as Pearson’s syndrome [70–72]. The model of mitochondrial insufficiency presented here shares additional symptoms with those syndromes including lactic acid accumulation and nutrient malabsorption [70, 72]. To our knowledge, this is the first time these clinical features are linked to a mutation in a nuclear-encoded mitochondrial gene. Our data may thus contribute to a better understanding of the organismal consequences of mitochondrial pathologies, and enhance a more accurate diagnosis and advancement of treatment strategies in the future.
There are two paralogues of the mia40 (chchd4) gene in the zebrafish genome, mia40a (chchd4a, ENSDARG00000033376) and mia40b (chchd4b, ENSDARG00000040707).
Zebrafish husbandry was performed in accordance with institutional and national ethical and animal welfare guidelines. Procedures involving animals were approved by the First Warsaw Local Ethics Committee for Animal Experimentation (357/2012 and 197/2016), the Veterinary Department of the Regional Council of Darmstadt (B2/1017) and the Medical University of South Carolina Institutional Animal Care and Use Committee (AR #2850).
Mutant lines were generated in the AB background. The Tg(Xla.Eef1a1:mlsEGFP)cms1 zebrafish line was used in this study [47].
Whole-mount in situ hybridization was performed essentially as described [78] with the following changes: 5 dpf larvae were permeabilized in 40 μg/ml proteinase K for 20 min, and probes were used at a concentration of 1 μg/ml in hybridization solution. Images were acquired using a Nikon SMZ25. Contrast and brightness were adjusted using Adobe Photoshop. In situ primers are listed in S3 Table.
Full-length wild-type and mutated CDS of mia40a a mia40b were amplified from mia40a+/bns292 and mia40b+/bns293 cDNA using specific primers (S3 Table). Amplicons were cloned into pCS2+ plasmids linearized with BamH1 using Cold Fusion (System Biosciences) and sequenced to confirm correct inserts. Plasmids were linearized with Not1 and mRNA was transcribed using the mMESSAGE mMACHINE SP6 synthesis kit (Ambion), followed by TURBO DNase treatment. mRNA was purified using the RNA Clean-Up and Concentrator kit (Zymo Research) and mRNA quality was assessed using the Nanodrop 2000c spectrophotometer (Thermo Fisher Scientific) and agarose gel electrophoresis. Different amount of wild-type versions, ranging from 50 to 500ng, were injected at 1-cell stage into embryos from a heterozygous in-cross of mia40a+/bns292 larvae in the Tg(Xla.Eef1a1:mlsEGFP) transgenic background. The highest dose of mRNA that did not trigger a severe phenotype in larvae was chosen for the experiments. At 3 dpf, larvae were embedded in 1% low-melting agarose containing 0.04% Tricaine solution in a glass-bottom dish (MatTek Corporation, Ashland, MA, USA) and analysed using the LSM800 confocal laser scanning microscope (Zeiss).
Oxygen consumption rate (OCR) measurements from zebrafish at 5 and 10 dpf were performed using the XF-24 Extracellular Flux Analyzer (Seahorse Bioscience) with slight modifications of the standard method [80]. Briefly, fish were placed individually in 20 wells of a 24-well islet plate (Seahorse Bioscience) and kept in place with an islet capture screen. Basal respiration was calculated for each animal from the two consecutive OCR readings prior to injection of 1.5 μM carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP). Maximal FCCP-uncoupled respiration was determined by averaging the maximum OCR after injection for each larva. Each individual was harvested after respirometry and genotyped as above. Nonparametric comparisons for each genotype pair were conducted for basal respiration, maximal respiration, and spare respiratory capacity (SRC, the result of subtracting basal respiration from maximal respiration), with the Wilcoxon method using JMP software.
Measurements of total body glucose and lactate were performed on a pool of 5 larvae per genetic condition using a fluorescence-based enzymatic detection kit (Biovision Inc.) as described before [81]. Briefly, larvae were collected in 1.5 ml microcentrifuge tubes, the excess of water was removed and samples were frozen stored at -80°C until further analysis. Upon defrosting, 15 μl PBS per larvae was added and tissues were disrupted with a sterile pestle (Axygen), on ice. Samples were spun at 14,000 x g for 10 min, at 4°C. The supernatant was immediately used for metabolite measurements according to manufacturers’ protocol. Fluorescence readings (Ex/Em 535/590) were taken with the FLUOstar Omega instrument (BMG Labtech). Results were read from a linear regression curve based on the dilutions of the standard. Statistical analysis between indicated groups was performed using the unpaired t-test and considered significant at P < 0.05 (*), P < 0.01 (**) and P < 0.001 (***).
Zebrafish larvae were genotyped by fin-fold amputation at 3 dpf. Whole DNA extraction was performed from individual larva of indicated genotype using the DNeasy Blood and Tissue kit according to manufacturers’ protocol (Qiagen). The concentration and quality of isolated DNA was assessed using the Nanodrop 2000c spectrophotometer (Thermo Fisher Scientific). 5 ng of obtained DNA was used in qPCR reactions according to previously described protocol (refer to RT-qPCR subchapter). Primers specific to mitochondrial-encoded NADH-ubiquinone oxidoreductase chain 1 (mt-nd1) and nuclear-encoded DNA polymerase gamma (polg) for control were used (S3 Table). Statistical analysis between indicated groups was performed using the unpaired t-test and considered significant at P < 0.05 (*) and P < 0.01 (**).
Protein was extracted from pools of at least 20 larvae at 5 dpf. Larvae were lysed in buffer containing 20 mM Tris-HCl pH 7.4, 200 mM NaCl, 0.5% Triton X-100 and 2 mM PMSF using a sterile pestle to disrupt the tissue (Axygen). The samples were centrifuged at 3000 x g for 5 min at 4°C and lysates were mixed with same volume of chloroform/methanol (2:1) solution. Upon vortexing, one volume of 10% TCA diluted in cold 100% acetone was added and samples were precipitated on ice for 5 min. Upon centrifugation at 10,000 x g at 4°C for 1 min the upper, aqueous phase was discarded and 2 volumes of 10% TCA in cold dH2O was added to the samples and vortexed. The 10% TCA wash was repeated thrice. Next, 5 volumes of cold dH2O was added followed by centrifugation at 10,000 x g at 4°C for 1 min and addition of 1 ml of ice cold 100% acetone. Upon another centrifugation step, samples were washed with 80% acetone. Pellets were dissolved in 75 μl of sample buffer containing 30 mM Tris-HCl pH 6.5 and 4% SDS and heated to 70°C. For equal loading, protein concentration was established using Direct Detect spectrometer (Millipore). Samples were mixed with Laemmli buffer containing 50 mM DTT and incubated for 15 min at 65°C. 50 μg protein sample was subjected to SDS-PAGE and western blotting in conjunction with specific antibodies: anti-Ndufa9 (ab14713, 1:300 dilution), anti-Cox4i1 (ab14744, 1:1000 dilution), anti-Atp5a (ab14748, 1:4000 dilution), anti-Rpl7 (A300-741A, 1:000 dilution), anti-β Actin (A1978 Sigma Aldrich, 1:1000 dilution).
For RNA-Seq, RNA was isolated from a pool of 5 larvae at 5 and 8 dpf using the miRNeasy micro Kit (Qiagen) combined with on-column DNase digestion (DNase-Free DNase Set, Qiagen) to avoid contamination by genomic DNA. RNA and library preparation integrity were verified with a BioAnalyzer 2100 (Agilent) or LabChip Gx Touch 24 (Perkin Elmer). 1–2 μg of total RNA was used as input for Truseq Stranded mRNA Library preparation following the low sample protocol (Illumina). Sequencing was performed on the NextSeq500 instrument (Illumina) using v2 chemistry, resulting in average of 20 M reads per library with 2 x 75 bp paired end setup. The resulting raw reads were assessed for quality, adapter content and duplication rates with FastQC [82]. Trimmomatic version 0.33 was employed to trim reads after a quality drop below a mean of Q18 in a window of 5 nucleotides [83]. Only reads above 30 nucleotides were cleared for further analyses. Trimmed and filtered reads were aligned versus the Ensembl Zebrafish genome version DanRer10 (GRCz10.87) using STAR 2.4.0a with the parameter “—outFilterMismatchNoverLmax 0.1” to increase the maximum ratio of mismatches to mapped length to 10% [84]. The number of reads aligning to genes was counted with featureCounts 1.4.5-p1 tool from the Subread package [85]. Only reads mapping at least partially inside exons were admitted and aggregated per gene. Reads overlapping multiple genes or aligning to multiple regions were excluded. 8 dpf samples were collected and processed on different time-points and a significant batch effect was noticed for these samples. No such effect was noticed for 5 dpf samples, collected and processed simultaneously. The batch effect correction for 8 dpf samples was done using removeBatchEffect function from edgeR package version 3.16.5 on log-transformed expression matrix (S4 Table) with a minimum combined mean of 5 reads. The result of batch removal is presented on the S7 Fig and the expression matrix with no batch effect is available in S5 Table. Pearson correlation analysis revealed high reproducibility of biological replicates with correlation coefficient values > 97% for transcriptomic data (S8 Fig). Differentially expressed genes were identified using DESeq2 version 1.62 [86]. Only genes with a minimum fold change of log2 = < -0.9 and > = 0.9, a maximum Benjamini-Hochberg corrected P-value of 0.05, and a minimum combined mean of 5 reads were deemed to be significantly differentially expressed (S1 Table). The Ensemble annotation was enriched with UniProt data (release 06.06.2014) based on Ensembl gene identifiers (IDs) [87]. Raw data were deposited to GEO repository with the accession number GSE113272.
Fractionated peptides were reconstituted in 10 μl of solvent A (0.1% formic acid) and subjected to mass spectrometric analysis in line to reversed phase capillary chromatography. Peptides were separated using an UHPLC system (EASY-nLC 1000, ThermoFisher Scientific) and 20 cm in-house packed C18 silica columns (1.9 μm C18 beads, Dr. Maisch GmbH) coupled to a QExactive HF orbitrap mass spectrometer (ThermoFisher Scientific) using an electrospray ionization source. The gradient employed consisted of linearly increasing concentrations of solvent B (90% acetonitrile, 1% formic acid) over solvent A (5% acetonitrile, 1% formic acid) from 5% to 30% over 215 min, from 30% to 60%, from 60% to 95% and from 95% to 5% for 5 min each, followed by re-equilibration with 5% of solvent B. The constant flow rate was set to 400 nl/min.
Full MS spectra were collected for a mass range of 300 to 1750 m/z with a resolution of 60,000 at 200 m/z. The ion injection target was set to 3 x 106 and the maximum injection time limited to 20 ms. Ions were fragmented by higher energy collision dissociation (HCD) using a normalized collision energy of 27, an isolation window width of 2.2 m/z and an ion injection target of 5 x 105 with a maximum injection time of 20 ms. Precursors characterised with unassigned charge state and a charge state of 1 were excluded from selection for fragmentation. The duration of dynamic exclusion was 20 s. Resulting tandem mass spectra (MS/MS) were acquired with a resolution of 15,000 at 200 m/z using data dependent mode with a top 15 loop count.
MS raw data were processed by MaxQuant (v. 1.6.0.1) [90] using the Uniprot zebrafish database (as of 17.08.2017) containing 59064 entries and the following parameters: a maximum of two missed cleavages, mass tolerance of 4.5 ppm for the main search, trypsin as the digesting enzyme, carbamidomethylation of cysteines as a fixed modification and oxidation of methionine as well as acetylation of the protein N-terminus as variable modifications. For the dimethyl-labeled protein quantification, isotope labels were configured for peptide N-termini and lysine residues with a monoisotopic mass increase of 28.0313 and 36.0757 Da for the light and heavy labels, respectively. Peptides with a minimum of seven amino acids and at least one unique peptide were included in the analysis. MaxQuant was set to filter for 1% false discovery rate at the peptide and protein levels, both. Only proteins with at least two peptides and one unique peptide were considered identified and included in further data analysis (S6 Table). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD009594 [91, 92].
The obtained data was subjected to differential expression analysis using the in-house R package autonomics (https://github.com/bhagwataditya/autonomics; version 1.0.21), which makes use of functionality provided by limma (S1 Table) [93].
Gene Ontology (GO) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) enrichment analysis was performed for RNA sequencing and quantitative proteomics data using KEGG.db (v. 3.2.3) and GO.db (v. 3.4.1) R/Bioconductor packages. The KEGG enrichment analysis was done using human orthologue genes from package org.Hs.eg.db (v 3.4.0), while GO enrichment analysis using zebrafish data from package org.Dr.eg.db (v 3.4.1). Pathways and genes selected in each developmental stage were filtered after Benjamini-Hochberg correction for an adjusted P-value < 0.05.
The list of expressed proteins (S6 Table) was combined with transcriptomic data. The first gene name in a protein group was used to merge the proteomic data with Ensembl gene IDs for zebrafish. Next, this set was combined with the transcriptomic data by merging on Ensembl gene IDs (S7 Table). The identification of genes encoding mitochondrial proteins using the MitoCarta 2.0 gene set [94] was based on zebrafish Ensembl gene IDs as well (S2 Table).
PTU-treated larvae were fixed with 4% paraformaldehyde (PFA, Sigma) overnight (O/N) at 4°C. Next, samples were washed thrice with phosphate-buffered saline (PBS, Sigma). Larvae were embedded laterally, in 1% low-melting agarose in a glass-bottom dish (MatTek Corporation, Ashland, MA, USA). Images were acquired using a Nikon SMZ25. Measurements were done blindly and manually using the NIS Elements software. To obtain the ratio, the pancreas area [μm2] was divided by larvae length [μm].
|
10.1371/journal.pntd.0006990 | Epidemiological characteristics and determinants of dengue transmission during epidemic and non-epidemic years in Fortaleza, Brazil: 2011-2015 | After being eliminated during the 1950s, dengue reemerged in Brazil in the 1980s. Since then, incidence of the disease has increased, as serotypes move within and between cities. The co-circulation of multiple serotypes contributes to cycles of epidemic and interepidemic years, and a seasonal pattern of transmission is observed annually. Little is known regarding possible differences in the epidemiology of dengue under epidemic and interepidemic scenarios. This study addresses this gap and aims to assess the epidemiological characteristics and determinants of epidemic and interepidemic dengue transmission, utilizing data from the 5th largest city in Brazil (Fortaleza), at fine spatial and temporal scales.
Longitudinal models of monthly rates of confirmed dengue cases were used to estimate the differential contribution of contextual factors to dengue transmission in Fortaleza between 2011 and 2015. Models were stratified by annual climatological schedules and periods of interepidemic and epidemic transmission, controlling for social, economic, structural, entomological, and environmental factors. Results revealed distinct seasonal patterns between interepidemic and epidemic years, with persistent transmission after June in interepidemic years. Dengue was strongly associated with violence across strata, and with poverty and irregular garbage collection during periods of low transmission, but not with other indicators of public service provision or structural deprivation. Scrapyards and sites associated with tire storage were linked to incidence differentially between seasons, with the strongest associations during transitional precipitation periods. Hierarchical clustering analysis suggests that the dengue burden concentrates in the southern periphery of the city, particularly during periods of minimal transmission.
Our findings have direct programmatic implications. Vector control operations must be sustained after June even in non-epidemic years. More specifically, scrapyards and sites associated with tires (strongly associated with incidence during periods of minimal transmission), require sustained entomological surveillance, particularly during interepidemic intervals and in the urban periphery. Intersectoral collaborations that address urban violence are critical for facilitating the regular activities of vector control agents.
| Almost half of the world population is at risk of a dengue infection. Although dengue often presents annual transmission seasonality (peaking during the summer) and features epidemics occuring between multi-year interepidemic intervals, little is known about how the epidemiology and determinants of dengue vary seasonally and during different stages of the epidemic cycle. As a result, control efforts likely neglect factors that sustain transmission between epidemics, or that set epidemics in motion. Using five years of monthly data at fine spatial scale from the 5th largest city in Brazil, we observed that dengue epidemiology in interepidemic years deviates from the common pattern, showing a much longer and sustained transmission season. Occurrence of violence was a strong and consistent predictor of high dengue incidence, while specific areas in the city (e.g. with a high concentration of scrapyards and locations used for tire storage) were critical during transitional precipitation periods. Indicators of social deprivation were only correlated with dengue incidence during periods of low transmission, suggesting that practices typifying these areas–such as water storage–may extend viremic circulation during dry seasons. These findings can be translated into revised and better strategies of control, but also call for the need to better understand how variations in the epidemiology of dengue may affect the success of current control recommendations.
| Dengue virus (DENV, genus Flavivirus, family Flaviviridae) is an arbovirus with at least four distinct serotypes (DENV1, DENV2, DENV3, and DENV4) that confer homologous immunity [1]. Manifestations of DENV infection range from completely asymptomatic in as many as three quarters of cases, to dengue hemorrhagic fever (DHF) and dengue shock-syndrome (DSS), with an intervening continuum of febrile symptoms and nonspecific signs of infection such as headache and rash [2]. Dengue occurs primarily in tropical and sub-tropical latitudes with an estimated burden of 390 million cases annually, of which 96 million cases manifest symptomatically [3]. These figures represent a thirty-fold increase in disease globally over the last fifty years that exposes nearly three billion people to the risk of infection [4]. The heaviest burden is in Asia, with an estimated 67 million symptomatic cases annually, followed by the Americas, with 13 million, of which more than half occur in Brazil and Mexico [3].
Dengue is transmitted primarily by Aedes aegypti and secondarily by Ae. albopictus [5], in predominantly urban human transmission cycles [6]. Ae. aegypti prefers human blood [7–9]; is more likely to oviposit in water storage containers than natural depressions [10, 11]; and its eggs withstand desiccation for more than four months, on average, in high humidity settings [12]. Common breeding habitats include containers for water storage; domestic and cookware receptacles; flower pots; and discarded objects, including tires and trash [13]. As a result, Ae. aegypti flourishes in crowded human settlements lacking access to piped water, waste collection, and adequate health and vector control systems. Such areas typify expansive and unplanned urban peripheries in tropical and subtropical latitudes [14]. In those settings, even when access to piped water is prevalent, intermittent and unreliable provision may compel populations to store water in containers, introducing Aedes oviposition sites [15]. High population density (and thus easy availability of a blood meal or breeding site) and anthropogenic constraints on urban low-level flight (such as highways, railroads, and buildings), likely limit mean Ae. aegypti flight range to < 200m in urban settings [10, 16, 17]. Notwithstanding average and maximum flight distances, adult Ae. aegypti remain in close proximity to the site of their larval/pupal development [18], predominantly concentrating and transmitting DENV indoors [19], such that a substantial share of viral circulation is believed attributable to the socially structured movement of human hosts through areas where they are exposed to DENV-infected mosquitoes [20, 21].
A variety of vector control strategies are available, targeting different stages of the mosquito [22], as well as educational campaigns aimed at promoting behaviors that minimize the proliferation of breeding habitats [23–25]. The World Health Organization (WHO) endorses an integrated approach to vector control [26], including an ecological, biological, social (‘Eco-bio-social’) dengue transmission model emphasizing community participation, as well as identification of local sources of dengue exposure [23, 27]. A primary challenge to interventions, however, is implementation at scale in a manner that can be sustained across periods of epidemic and interepidemic transmission [28, 29]. Insecticide treated bednets are inadequate to rebuff diurnal, endophilic Aedes mosquitoes, and large-scale adulticide fogging has been characterized as a cosmetic measure [30]. There is evidence that treatment of curtains and screens with insecticide reduces Aedes density [31, 32] with communal spillover effects [33], and that indoor residual [17, 34] and space [35] spraying reduce both vector density and dengue incidence.
Determining the time, place, and combination of control interventions requires proper knowledge of dengue epidemiology, which is influenced not only by the characteristics of the vector, but also by many biological (e.g., serotype), social (e.g., education), behavioral (e.g., storage of water), economic (e.g., income), and environmental (e.g., climate) factors [15, 36–43]. Temporal trends in dengue incidence within an endemic area are commonly characterized by typical inter-annual and seasonal variability. First, a cyclical inter-annual pattern of epidemic and lower-level interepidemic transmission exists, with as much as a tenfold difference in observed cases year-to-year [35, 44]. Second, regardless of the degree of endemicity, within each year dengue incidence exhibits seasonality, with peaks usually observed during warmer and wetter months [37, 45]. The appearance of DENV genetic variants with greater epidemic potential is linked to sustained urban interepidemic transmission [44, 46], increasing the importance of identifying factors contributing to transmission during these intervals.
This cyclical, seasonal, and context-dependent nature of urban dengue transmission present challenges for targeting control, and a better understanding of possible differences in drivers of transmission within and between interepidemic and epidemic years is needed. This study addresses this need, and aims to assess the role of structural, environmental, and human factors in the spatial and temporal distribution of reported dengue cases in Fortaleza, the 5th largest city in Brazil. The analysis considers five consecutive years (2011 to 2015), capturing both epidemic and interepidemic transmission. Observing the ubiquity of reported dengue infections in Fortaleza during epidemic months of peak transmission, we hypothesize that ecological correlates of infection differ within and between years according to the scale of transmission (epidemic vs. interepidemic) and the season (low vs. high precipitation). If factors associated with dengue incidence in Fortaleza during periods of low transmission are different than those that characterize incidence during epidemic peaks, vector control efforts may inadvertently neglect issues that sustain local viremic circulation between epidemics. Since suppression of interepidemic transmission may be a means to forestall or avert epidemics, identifying conditions that are associated with incidence during each stage of the epidemic cycle represents much needed evidence to inform dengue prevention and control.
Located in the Northeast (NE) region of Brazil, Fortaleza is the capital of Ceará state and the 5th most populous city in Brazil (estimated at 2.6 million inhabitants in 2018) [47]. Fortaleza is divided into six regionais (administrative districts) and 119 bairros (administrative sub-districts, or neighborhoods) (Fig 1). The United Nations income-based urban inequality index classifies Fortaleza as the 9th most unequal city in the world [48]. The city has the highest homicide rate in Brazil, and the 12th highest homicide rate in the world, with approximately 61 homicides per 100,000 people in 2015 [49].
The Köppen climate classification describes Ceará as equatorial savannah, with dry winters and humid summers [50]. The seasonality of annual precipitation cycles in Ceará is modified by the Atlantic Multidecadal Oscillation (AMO) climate cycle, which causes anomalous wet-dry phases lasting for multiple decades [51].
After the elimination of Ae. aegypti from most of the Americas in the 1950s –except for the United States, Suriname, Venezuela, and several Caribbean states [52]–the mosquito was reintroduced in Brazil in the late 1970s, at a time when entomological surveillance activities and vector control teams were, for the most part, non-existent [53]. Dengue reemerged in Brazil during the early 1980s, with epidemics of DENV1 in Northern Brazil and Rio de Janeiro state in 1981 [54]. These outbreaks were followed by a series of epidemics in coastal cities of the Southeast (SE) and NE regions, and nearly 300,000 cases were reported nationally between 1986 and 1993 [55]. Between 2000 and 2010, national incidence varied widely by year, ranging from 74,000 cases in 2004 to more than 1 million in 2010 [56].
Following an absence of over fifty years, cases of DENV1 were first recorded in Fortaleza in 1986. DENV1 was the only serotype identified in the city until 1994, when DENV2 was associated with an epidemic of nearly 30,000 cases. DENV1 and DENV2 predominated until 2002, when DENV3 was first observed in the city. While reported annual incidence between 1994 and 2008 did not exceed 20,000 cases, recent epidemics of DENV2 (2008), DENV1 (2011) and allochthonous DENV4 (2012) all exceeded 34,000 reported cases [57].
Within the Fortaleza Municipal Health Secretariat, the Vector Control Program coordinates activities to prevent and respond to dengue epidemics in accordance with national protocols. The municipality employs approximately 1,260 individuals to conduct Ae. aegypti monitoring and control activities [58]. Epidemiological surveillance and vector control departments carry out three to four sampled entomological surveys annually to estimate property infestation, and visit targeted surveillance sites (called strategic points) every fifteen days. In parallel with epidemiological surveillance, Vector Control Program agents conduct community canvassing operations throughout the year to identify and destroy breeding sites, chemically treat potential breeding sites that cannot be destroyed (e.g. elevated water tanks and cisterns), and implement public information campaigns to promote community-based vector control [59].
This study assembled data from multiple sources. Dengue cases recorded in Fortaleza between 2011 and 2015 were acquired from Brazil’s Notifiable Diseases Information System (Sistema de Informação de Agravos de Notificação, SINAN) [60]. By law, suspected dengue cases are reported by clinicians and health professionals to SINAN within 24 hours of initial diagnosis [61]. Suspected dengue cases include all episodes of fever lasting two to seven days which are accompanied by at least two symptoms including nausea, vomiting, positive tourniquet test (i.e. capillary fragility test), petechiae, leukopenia, headache, retro-orbital pain, myalgia, or rash; and exposure to areas with active dengue transmission or presence of Ae. aegypti during the prior fourteen days [62]. Within 60 days of reporting, suspected cases are coded by the Municipal Health Secretariat as discarded/inconclusive (when not all criteria for a suspect case, as described above, was met), or coded as confirmed dengue according to: (i) laboratory analysis–based on results of IgM serology (ELISA), NS1, viral isolation, RT-PCR, or postmortem immunohistochemistry; or (ii) clinical/epidemiological criteria–when a laboratory test was not performed, but all criteria were met [63]. The MoH recommends that as many cases as possible should be confirmed by laboratory analysis, except in epidemic periods, when about 10% of laboratory confirmation is consired to be sufficient to characterize the epidemiological situation [63]. In this study we considered only confirmed dengue cases.
We used data from years 2011 to 2015 to capture both the cyclical and seasonal patterns of dengue transmission: 2011, 2012, and 2015 were considered epidemic years, and 2013, and 2014 were interepidemic years. Case home addresses were geocoded and then aggregated to the bairro-level. Sufficient information to geocode cases to bairros was available for 97.5% of confirmed cases in 2011, 77.5% in 2012, 98.1% in 2013, 95.4% in 2014, and 77.7% in 2015. Temporally, cases had the exact date of first symptoms, which allowed aggregation to epidemiological weeks and to months. All data analyzed were anonymized.
We obtained annual population data by bairro from the 2010 Demographic Census [64] and estimated bairro population growth from 2011–2015 using estimates of population growth for the city of Fortaleza from the Brazilian Institute of Geography and Statistics (IBGE). Using dengue cases and population, we calculated incidence rates per 100,000 people for each bairro, and created a binary variable to indicate low and high transmission months in each year: (i) high corresponds to months with more than 9% of annual transmission (or more than one would expect if cases were uniformly distributed over the year), or more than 1,000 total reported cases citywide, and (ii) low corresponds to the remaining months.
We obtained bairro-level socio-ecological data from the 2010 Demographic Census [64]. These include: mean number of inhabitants per household; percent of households connected to electricity, piped water, sewage, and regular garbage collection; population density; average income per household in Reais (R$); and literacy rates for men and women separately. Two new bairros were created after 2010 (both originated from larger bairros split into two). In those cases, socio-ecological data for the original bairro in 2010 was extended to the new bairros.
We estimated the degree of structural deprivation in a bairro using two data sources. First, we obtained information for subnormal agglomerations (AS, aglomerados subnormais) by census tract from the 2010 Demographic Census. AS is a habitation class characterized by tenuous legal claim and poor structural development. An AS is constituted by, at minimum, 51 inhabited units on illegally occupied land or land obtained within the prior ten years; constructed haphazardly or outside preexisting structural standards; and lacking access to essential public services such as electricity, garbage collection, water grids, and sewage systems [65]. Second, from the Planning Institute of Fortaleza (IPLANFOR, Instituto de Planejamento de Fortaleza) [66], we obtained data on precarious settlements (AP, assentamentos precários). AP is an exclusively urban classification introduced by the National Housing Secretariat (Secretaria Nacional de Habitação) that extends the AS definition to include settlement types that are structurally insecure, neglected, or unplanned, even if legally occupied [67]. The spatial boundaries of AP and AS were merged, then overlaid on the bairro boundaries to estimate the intersecting areal proportion for each bairro. We refer to this aggregate class as subnormal settlements (SS).
Annual homicide counts by bairro were collected from the Mortality Information System (Sistema de Informações de Mortalidade, SIM), and used to calculate a homicide rate per 10,000 people. It is established that interpersonal violence can deter and disrupt provision of human services by introducing risk for individuals providing and receiving health-related services [68]. As such, we hypothesized that bairro-level homicide rates are associated with spatial heterogeneity in access to health and vector control services, and with higher dengue transmission.
Entomological surveillance data were acquired from Ae. aegypti infestation surveys, conducted by Brazilian Municipal Health Secretariats at a local level. The Aedes aegypti Infestation Rapid Survey (Levantamento Rápido de Índice para Aedes aegypti, LIRAa) is a sample survey that randomly selects properties for inspection for immature forms (mosquito larvae and pupae). Municipalities are divided into areal groups of between 8,100–12,000 properties, from which a sample of 450 properties is drawn. LIRAa was introduced in Fortaleza in October 2011 to replace the Ae. aegypti Infestation Survey (Levantamento de índice de infestação amostral, LIA), which covered a much larger sample of properties in the municipality (about 10%), and required considerably more time to complete. LIRAa is repeated three to four times per year, with a national survey conducted every October. The Brazilian Ministry of Health uses three categories of infestation intensity to classify risk: (i) satisfactory, <1% infestation; (ii) alert, 1–3.9%; and (iii) risk of outbreak, >3.9% [62]. We obtained data on all surveys available for LIA in 2011 and for LIRAa from 2011 to 2015. For the purposes of this analysis we used data from the January surveillance of both programs–immediately preceding the early-spring rainy season characteristic of NE Brazil–that is considered the most valuable for forecasting regional epidemic risk. In 2014, the January LIRAa survey was delayed until February, resulting in elevated bairro infestation indexes relative to the other years; as a result, we designated 2014 LIRAa results as missing, and used a missingness indicator variable [69]. Under the assumption of representative sampling, we used total houses sampled by bairro as a denominator to create a bairro-level household infestation variable for each year.
We also used data from surveillance of strategic points (SP)–locations identified by the municipality as higher risk for harboring Aedes breeding sites–to assess the epidemiological relevance of Aedes infestation at targeted non-residential locations. In Brazil, SP are a central component of national directives for vector control to prevent dengue. Municipal vector control operations agents are tasked with creating a roster of sites likely to be susceptible to mosquito oviposition and inspecting those sites every 15 days. These operations are coordinated locally, such that each municipality is responsible for its own roster of strategic points [62]. Vector control teams inspect SPs in 15-day cycles and code the visit as positive or negative for the presence of Aedes larvae or pupae. Between 2011 and 2015, surveillance agents identified 82 types of SPs in Fortaleza; we aggregated those into six categories: construction, industrial fabrication, recyclable processing, scrapyards, sites associated with tire storage–such as tire repair shops and garages, and an “other” category, which included sites associated with vacant lots, public and private facilities (e.g. hospitals and schools), husbandry (e.g. cattle pens, vacarias), private residences, outdoor locations (e.g. cemeteries), and general commercial properties.
We created two bairro-level variables for each SP category: (i) positivity, defined as the proportion of successfully visited sites that were registered as positive for larvae/pupae; and (ii) proximity, defined as the areal proportion of the bairro that falls within a 150-meter buffer area around a positive strategic point. We selected the buffer size using two criteria: (i) expected Aedes mosquito flight distance in an urban setting [10, 16, 17]; and (ii) buffer size used by the Brazilian Ministry of Health to define the transmission block area around identified vector oviposition sites [62]. For the positivity variable, bairros without an SP visit during an inspection cycle were coded as missing with a missingness indicator variable. The proximity variable was calculated in ArcGIS. To properly associate the presence of mosquito larvae/pupae with monthly dengue virus transmission, SP-related covariates were lagged by a 15-day cycle, as shown in Fig 2.
We acquired rainfall data from the Climate Hazards Infrared Precipitation with Stations (CHIRPS) dataset [70]. CHIRPS has a spatial resolution of 0.05°, thus Fortaleza was covered by 15 pixels. Daily precipitation estimates were aggregated to weeks, and spatially joined to bairros. Efforts to associate precipitation volume with dengue incidence must incorporate a time lag to adjust for vector development. While lags of zero [45], one [71, 72], and two months [37, 73] were reported, the mechanistic role of precipitation in dengue transmission can be confounded by water storage behavior [43], and with the potential for heavy rains to wash out exposed oviposition sites [74]. For the purposes of our analysis, precipitation was included as a continuous variable (monthly sum in millimeters), and lagged by two weeks, consistent with the lag used with the SP data. To assess potential effects of rainfall anomalies [75] during the period, we calculated the average monthly rainfall by bairro over the five years of study and computed the deviation of each bairro-month from its five year average. This variable was included in all models, except those stratified by monthly precipitation volume.
Lastly, daily temperature values (high, low, and daily average) recorded at Fortaleza’s Pinto Martins International Airport were obtained from the Weather Underground data archive [76] and used to generate two measures of temperature variability over the course of a month [77]: monthly average of daily temperature ranges (high minus low daily temperature), and monthly standard deviation of daily temperature averages.
We defined the outcome variable as the monthly case incidence rate (dengue cases per 100,000 people) by bairro, observed from January 2011 to December 2015, and used a zero-inflated negative binomial longitudinal regression model (hereafter, longitudinal model). The zero-inflated negative binomial model accounts for extra-variation (overdispersion) in the data [78, 79], and was considered as an alternative to the negative binomial hurdle model given the likelihood that seasonally inflated zeros are attributable to both true changes in ecological dynamics that depress transmission, as well as reduced surveillance or misclassification during low seasons and interepidemic years [80–82]. The final model was selected based on comparison of Akaike Information Criterion (AIC) values for alternative distributions, including negative binomial, poisson, and zero-inflated poisson. We chose longitudinal models because our data are longitudinal at the bairro level, and used Huber-White standard errors [83]. Climatological and ecological independent variables–selected based on previously identified associations with Aedes-borne disease transmission and summarized by bairro-month–included: mean household size; percentage of households with access to electricity; percentage of households with access to piped water; percentage of households with access to regular garbage collection; percentage of households connected to sewage network; male literacy rate; female literacy rate; household income; population density; proportion of a bairro classified as a subnormal settlement; homicide rate; separate lagged strategic point proximity variables for each of the following types: construction, material fabrication, recyclables, scrapyard, tires, and others; lagged total infestation index for all SP categories combined; categories of infestation (LIA and LIRAa); lagged total precipitation (mm) and deviation from average precipitation (mm); and standard deviation of daily temperature averages (degrees Celsius). The AIC was used to select the final temperature covariate, and likelihood ratio test statistics (chi-square) were used to compare the full model with naïve intercept-only models, indicating that the full model was a better fit (p<0.001) for all analyses.
Considering the cyclical and seasonal patterns of dengue transmission, we run eight temporally stratified models. Model 1 contains all 7,140 bairro-months for which data were collected. Models 2 and 3 distinguish between epidemic (2011, 2012, and 2015) and interepidemic (2013, 2014) years. Models 4 and 5 stratify the analysis by transmission intensity, while Models 6, 7, and 8 are stratified by the intensity of precipitation (Fig 3). We addressed the family-wise error rate in all models using the false discovery rate (FDR) [84]. All data preparation and regression analysis was done in Stata v.14.2 (Stata Corp., College Station, TX, USA).
Lastly, to characterize the spatial and temporal patterns of epidemic intensity between years we used average linkage hierarchical clustering with an Euclidean distance dissimilarity measure, based on eight metrics, standardized as z-scores, reflecting the duration and burden of dengue incidence in a bairro. The number of clusters selected for each year was determined with reference to the gap statistic [85], reflecting the difference of within-cluster variability (as the within-cluster sum of squares around cluster means) at each number of k clusters in the observed data from its expectation in a null reference distribution computed from repeated sampling using the package factoextra [86]. Values of the gap statistic indicate the strength of clustering at each quantity of clusters considered (i.e. sequential values k). Maxima of the gap statistic may be local (in comparison to their immediate neighboring values k) or global (in comparison to all possible or considered levels k), and a plateau in the statistic indicates the negligible added value of additional partitions between clustering groups [85]. We considered between 3–7 clusters for each year to preserve interpretative value, prioritizing global maxima of the gap statistic (for years 2012, 2013, 2015 –S1 Fig), or local maxima that distinguish between well-separated groups, where further divisions between clusters in the considered range would exclusively partition single-bairro clusters (2011). Though the gap statistic first plateaued at three clusters in 2014, we partitioned the middle cluster–with high within-cluster variability along the first principal component of our clustering parameters–to distinguish two clusters, each with low internal variability along that axis (see S2 Fig, clusters II and III). In addition to plots of the first two principal components of clustering parameters (S2 Fig), grouping decisions considered potential informational value to surveillance operations. After grouping, clusters were ordered by the average full-year case rate of clustered units. Clustering analysis and mapping were conducted with R version 3.3.2 [87]. Data used in this study are available in S1 Dataset.
Between 2011 and 2015, 130,430 dengue cases were reported in Fortaleza. In total, 98,339 cases were confirmed by clinical/epidemiological or lab criteria and had sufficient information to be geocoded to bairros and included in this analysis. Annual citywide case rates per 100,000 varied from a low of 192 in 2014, to a high of 1,360 in 2011. Monthly rates within a single bairro exceeded 3,500 cases per 100,000 people in January 2011, April 2011, and June 2012. In total 26.5% (1,891 of 7,140) of bairro-months did not register a case; 75.2% of bairro-months with zero cases occurred between August and January, reflecting dengue seasonality. However, the pattern differed between epidemic and interepidemic years, as shown by the percent of annual reported cases that occurred between July and December each year (Table 1 and Fig 4). Incidence during epidemic years was higher, but more concentrated seasonally, while in interepidemic years transmission continued into seasons that were less climatologically hospitable to mosquitoes.
Hierarchical cluster analysis grouped bairros into ascending patterns according to the scale and duration of their dengue burden during annual transmission cycles (Fig 5, S2 Table). Spatially, results suggested that the epidemics of 2011 and 2012 were distinguishable by the increased prevalence of cases in northern, coastal bairros–such as the northern coast and northeastern peninsula near Mucuripe–after which incidence concentrated farther from the coastal urban core. Small, outlying clusters (typically composed of only one or two bairros) with elevated incidence rates and shorter periods of transmission were present in all years, such that the distance between observations in patterns 1 and 2 was smaller than between other clustering levels, and clustering was strongest during years when a large majority of bairros reported minimal dengue transmission (such as 2014 and 2015) (S2 Fig and S3 Table).
In 2011, clustering isolated 15 bairros (patterns 2–5), spatially dispersed throughout the city, with substantially elevated or protracted transmission (Fig 5); five of these bairros (patterns 4 and 5) exhibited outlying incidence rates (greater than 6,000 per 100,000) over shorter intervals (S2 Table). During the 2011 epidemic, transmission was present or elevated in all regionais: two bairros in the lowest clustering level experienced more than 3,000 cases per 100,000, over 44 and 48 weeks with confirmed incidence.
The 2012 epidemic and 2013 interepidemic years featured lower levels of transmission overall and fewer outlying bairros, decreasing clustering distances between high and low patterns. As a result, clustering distinguished additional patterns at both high and low scales of transmission. Similar to the 2011 epidemic, transmission during the 2012 epidemic was spatially diffuse, with coastal, affluent bairros grouped amongst peripheral communities in pattern 2. In contrast, in 2013 northeastern coastal bairros were almost exclusively grouped in pattern 1 (Fig 5).
This trend continued in subsequent years, when outlying transmission was either largely isolated (2014) or concentrated (2015) in bairros of the southern periphery. Confirmed incidence was negligible in 112 bairros of pattern 1 in 2014, yet low level residual dengue circulation is evident in six bairros of patterns 2 and 3 (Messejana, Jangurussu, Barroso, Maraponga, Bom Jardim, and Mondubim), where cases were confirmed for 38.5 and 44.5/52 weeks, on average. The scale of annual transmission in these six bairros is comparable to pattern 1 bairros in 2011, 2012, and 2015, but occurred over much longer intervals (S2 Table). Confirmed dengue incidence in three of these bairros (Messejana, Jangurussu, and Barroso) increased to epidemic levels in 2015 (identifiable in 2015 patterns 4 and 5), inflating case rates for the city as a whole.
Table 2 presents descriptive statistics. Mean household size in Fortaleza was 3.42 persons. Access to infrastructural and municipal services was high (except for sewage), though unequally distributed. Reported literacy exceeded 86% for both men and women in all bairros. Minimum reported average annual household income by bairro was R$ 800 reais (approximately US$450, 2010 exchange rate). Population density exceeded 100 persons per km2 in all bairros ranging from 148 in Sabiaguaba to more than 34,000 persons per km2 in Pirambu. Ten bairros included less than 1% SS in their area, and 13 bairros had more than 50% SS (highest values of 74%, 87%, and 92% observed in Genibau, Pirambu, and Curio). Homicide rates by bairro ranged from zero to 585.6 per 100,000 persons per year; rates for the whole city ranged from a low of 48.33 per 100,000 in 2011 to 76.93 per 100,000 in 2014, exceeding regional averages.
Temperature in Fortaleza did not vary substantially over the course of our five year study period, but daily temperature averages were marginally higher earlier in the year; the lowest daily temperature average was 22.2 degrees on January 10, 2011 and the highest daily average was 29.4, on March 3, 2013 (Fig 6). In contrast, precipitation differed seasonally, annually, and between bairros. At least one bairro registered no precipitation during 76% of our study period (197 weeks), and in 37% (96 weeks) no precipitation occurred anywhere in the city. The highest weekly sum precipitation (226mm) was recorded in the second week of March 2015, and precipitation exceeding 83mm was recorded during 10% of all bairro-weeks. The largest bairro differential in weekly precipitation was 124.2mm, with an average weekly differential of 16.2mm. Average deviation from monthly averages, measured at the bairro level, was highest in February 2011, when average bairro precipitation exceeded five-year averages for that month by nearly 240mm. This trend continued for the following three months (with positive deviations exceeding 100mm in March, April, and May), amidst the 2011 epidemic. In comparison, citywide averages during spring months of interepidemic years 2013 and 2014 were starkly lower than five-year averages (April 2013 = -137mm, and March 2014 = -101mm).
Entomological surveillance data is summarized in Table 3. On average, bairro-level infestation was higher during the LIA (2011) survey than for LIRAa surveys (2012, 2013, and 2015), though the maximum bairro-level infestation value (6.45% of visited sites positive for vector immature forms) was recorded in Vila Ellery in January, 2015. The highest average January bairro infestation over the course of the four-years (excluding 2014) was in Cambeba, 3.4%.
The spatial distribution of SP types varied across the city (Fig 7 shows their location in 2015), and around 3,000 SP sites were inspected annually, on average, during the study period. The largest number of SP inspected by surveillance agents was 3,522 in 2014, the majority being tire repair shops and garages (Tires category in Table 3). Considering all types of SPs, the highest infestation indices were registered in 2011 (8.38%) and 2015 (8.91%) (Table 3). There were large differences between infestation by site types: on average, sites in the “Others” category registered the highest monthly infestation indices, while those under the tire category had the lowest (Table 3). SP infestation exceeded household infestation (LIRAa/LIA), on average, and was substantially more variable at the bairro level. The monthly areal proportion of a bairro within 150 meters of a positive SP varied by type of site, reflecting differences in SP prevalence and infestation rates by class. On average, about 5% of bairro area was within 150-meter radius of a positive SP citywide, and more than half of this exposure was attributable to sites in the construction and tire categories.
In the analysis of all 7,140 bairro-months (Model 1, Table 4), amongst socio-ecological variables only bairro homicide rates was a statistically significant correlate of dengue incidence rates: on average, increases in annual household income by R$1,000 were accompanied by an 8% decrease in dengue incidence rates, while an additional ten homicides per 100,000 people was associated with a 6% increase. Over the entire study period, a 1% increase in the areal proportion of a bairro within 150 meters of a SP in the tires class was associated with 3% increases in bairro-month dengue incidence rates. Indicator variables of epidemic year, peak transmission season, and of precipitation intensity were statistically significant, supporting stratified analyses.
Tables 5 and 6 show the models stratified by transmission intensity inter-annually (Models 2 and 3) and seasonally (Models 4 and 5). Results indicate that some covariates were strongly associated with dengue incidence rates during periods of low (Models 3 and 5), but not high (Models 2 and 4) transmission. Though income was not a statistically significant correlate of dengue incidence during epidemic years (Model 2), during years of interepidemic transmission a R$1,000 increase in average annual bairro household income is associated with reduced dengue incidence by more than 10% (Model 3). Further, while exhibiting a negative, protective effect in all strata, the percentage of properties with regular garbage collection was only statistically significant during interepidemic years (Model 3).
With respect to SPs, a 1% increase in areal exposure to tires SP sites was associated with 3.3% increased dengue incidence during epidemic years (Model 2). Conversely, though unit increases in exposure to construction, recycling, scrapyard, and tire sites were all positively correlated with dengue rates during interepidemic years, none were statistically significant when corrected for multiple testing (Model 3). Unlike every other model, the continuous variable for monthly sum precipitation (mm) was not positively associated with incidence over the course of interepidemic years in our sample, reflecting the absence of typical dengue seasonality during those years.
Models 4 and 5 (Table 6) present results stratified by monthly transmission intensity (Fig 3). Though homicides were associated with dengue rates in all models, an additional 10 homicides per 100,000 was associated with a 7.3% higher incidence rate during high transmission months (Model 4), and 7.6% higher rate during epidemic years (Model 2), exceeding increases of 4.6% and 4.4% estimated for periods of lower transmission (Models 3 and 5). Income was a statistically significant protective correlate during both high and low transmission months (Models 4 and 5), and–though not statistically significant when corrected for multiple testing–models estimated stronger associations between regular garbage collection and dengue incidence during late-season months of residual transmission (Model 5). Finally, literacy was a statistically significant bairro-level correlate of dengue incidence rates during low months (Model 5), but with divergent effects: male literacy was associated with increased dengue incidence rates, while female literacy was correlated with lower rates. The strongest associations between proximity to infested SPs and dengue incidence related to scrapyards and tire sites within both seasonal strata (Models 4 and 5); while the estimated effects were larger during residual months (Model 5), the corrected associations were not statistically significant. Movement from satisfactory (<1% property infestation) to alert (1–3.9%) levels of larval infestation, as measured by cross-sectional LIRA and LIA surveys conducted in January, was a statistically significant correlate of incidence in bairros across epidemic years (Model 2), but not when the sample was isolated to peak months (Model 4). Temperatures were slightly lower and more variable, on average, during epidemic years (Table 2); nevertheless, a negative correlation between variability and incidence during months of seasonal transmission (Model 4) suggests lower temperature variance during peak months of epidemic years relative to interepidemic years.
Models 6, 7, and 8 are stratified by precipitation intensity and presented in Table 7. Consistent with results for low-transmission months (Model 5), literacy and garbage collection were correlates of dengue incidence only during months with minimal rainfall (Model 6). Correlation between dengue incidence and income peaked during “transition” precipitation strata (Model 7)–where R$ 1,000 increases in income were associated with an 11% decrease in dengue incidence rates–but the effect was nearly halved during periods of high precipitation (Model 8). Relative to dry months (Model 6), during wetter months (Models 7 & 8) the magnitude of the association with homicides doubled and a negative association with population density increased in magnitude. Though a relationship between tires SP sites and dengue incidence was observed during low precipitation months (Model 6), associations between dengue incidence and SP proximity were strongest amidst intermediate precipitation (Model 7), when marginal increases in the areal proportion for tire and scrapyard sites were associated with over 4% increased incidence rates. The estimated effect for these SP types declined to 2.9% and 1.1%, respectively, during high precipitation periods (Model 8). While temperature variability was a strong predictor of monthly dengue incidence rates during periods with minimal precipitation (Model 6), it registered a null association during wetter months (Model 8).
This study analyzed five years of dengue incidence at fine spatial and temporal scales. One of the most important findings was the distinct seasonal pattern between interepidemic and epidemic years. Our results indicate that dengue epidemiological curves differed according to the intensity of annual transmission: while the pattern of transmission during epidemic years conforms to widely documented seasonality, during non-epidemic years low-level transmission persisted into climatologically inhospitable conditions. Sustained transmission after June could result from relaxed control activity following a high season with minimal transmission. Such lapses during interepidemic periods are likely to compound entomological and epidemiological challenges for the next annual cycle, and call for sustained vector control activities regardless of transmission intensity. Bairros in the southern periphery of the city experienced a larger dengue burden than the coastal city center, particularly during interepidemic years, suggesting that sustained, non-seasonal transmission is linked to conditions in those peripheral areas. Socio-ecological indicators of poverty and deprivation were correlated with higher bairro-level dengue incidence rates during seasonal and non-seasonal months–highest during non-epidemic years–but not across epidemics years. The most pronounced associations between Ae. aegypti surveillance at targeted sites and dengue incidence occurred during transitional precipitation seasons, suggesting that they may be one factor linking stages of residual and epidemic transmission. In contrast to the SP class covariates (which quantify the proportion of a bairro spatially proximate to an infested site in bi-monthly intervals), measures of the SP infestation index and entomological cross-sectional surveys (LIA and LIRAa)–ostensibly conducted to identify areas with heightened vulnerability to dengue transmission–did not reliably capture the variability in risk between seasons or bairros, corroborating other studies [88, 89]. These results are not surprising, given that the relationship between larval indices and adult densities is diminished by adult flight [90], and variable survival rates of immature forms and productivity by container type [91]. In fact, the association between cross-sectional entomological surveys (LIRAa/LIA) and incidence during epidemic years was dependent upon the inclusion of the precipitation deviation covariate in the model, reinforcing the need for closer scrutiny of those surveys. These findings have direct programmatic implications, identifing places and stages of the epidemic cycle when targeted community interventions can be prioritized by the Fortaleza Municipal Health Secretariat.
While the contribution of non-residential sites (such as the SPs) to vector propagation has been discussed [92], minimal attention is given to their epidemiological relevance. In some cases, non-residential surveillance has been limited to natural sites and niches uncharacteristic of Ae. aegypti oviposition [93] or a restricted class of non-residential sites, such as schools [94, 95]. Consideration of non-residential structures and spaces that sustain oviposition during dry seasons–such as vacant lots [96], septic tanks [97], and drains [98, 99]–is sparse. Although special surveillance of SPs is mandatory in Brazil, the absence of rigorous examination of the importance of SP infestation for dengue virus transmission limits the capacity of municipal actors to take evidence-based steps to improve their routine control activities. To the best of our knowledge, this is the first study to use fine scale information on SP inspection.
Infestation of SPs was spatially and temporally associated with dengue incidence in Fortaleza, and the magnitude of the associations differed by SP type and according to the scale of transmission and precipitation. Specifically, scrapyards and sites associated with tire collection and storage–such as repair shops and garages–showed associations with dengue incidence during periods of both interepidemic and epidemic transmission. The proportion of a bairro proximate to sites known for storing tires was a strong and reliable correlate of dengue incidence within nearly all strata, but the magnitude of the coefficient was largest during transitional precipitation regimes (when rainfall was neither sparse nor extreme) and dry seasons. Further, after adjustment for FDR, scrapyards registered statistically significant correlations with dengue incidence rates during the same intermediate rainfall strata. These results, combined with the new findings regarding the seasonal pattern of dengue transmission in interepidemic years, strongly suggest that enhanced surveillance should be sustained during low transmission periods, in both epidemic and interepidemic years. The Municipal Health Secretariat could characterize scrapyards and tire repair shops as “high risk” strategic points; such a classification could entail more frequent and thorough surveillance proportional to the physical size of the site, and to the number of potential oviposition habitats. An enhanced program would also impose strict and transparent consequences for sustained infestation at these locations (such as revocation of licenses or more effective fine enforcement), and where operators exhibit disregard for vector control protocol during successive visits.
Further, many of the socioeconomic, structural, and environmental factors expected to be associated with dengue transmission showed varied significance according to transmission and precipitation intensity. Among the factors related to access to public services, regular garbage collection was the most consistent correlate of dengue incidence, consistent with previous studies [100, 101]. However, we also show that the link is most pronounced during low transmission and low precipitation periods. Empty lots filled with abandoned trash–as well as discarded materials such as cans, plastic bottles, and debris commonly observed in yards of houses and along sidewalks [15]–could preserve desiccation-resistant eggs, which hatch following intermittent late-season rains. Thus, closer property surveillance and scrutiny of empty lots may be an important component of targeted vector control during interepidemic periods to prevent the escalation to epidemic-scale transmission.
With regard to income, negative [102] and null [100] associations between poverty and dengue incidence have been reported for Fortaleza. In contrast, with the exception of the epidemic year (Model 2) and high precipitation (Model 8) models, when incidence was shown to be more widely dispersed throughout the city, our results indicate a persistent concentration of dengue cases in lower income bairros. The association is strongest during periods of intermediate precipitation (Model 7), where a unit increase in bairro average household income (equal to one half standard deviation) is associated with 11% decrease in dengue incidence. The associations with income were highly statistically significant during months of high and low transmission (Models 4 & 5), but not in the combined dataset (Model 1), demonstrating the importance of stratifying models by transmission scale. Nonetheless, these results need to be interpreted with caution: health care provided by the private sector is largely underreported in SINAN, despite the fact that notification is mandatory [103]. In Fortaleza, approximately 30% of the population uses private health services, and in more affluent bairros this proportion reaches 85%. While the results observed may reflect underreporting of care provided to higher income populations, when stratified annually (Models 2 & 3) correlations with poverty were isolated to non-epidemic years, suggesting that changing dengue transmission dynamics at different stages of the interannual epidemic cycle also underly this association.
Consistent and large correlations between dengue incidence and interpersonal violence, which was also observed for tuberculosis [104], exemplify the difficulty of implementing effective population health measures in expansive cities such as Fortaleza, as well as the importance of involving different governmental sectors to address these challenges. In contrast to many other exposures, homicides were more strongly associated with dengue rates during epidemics and high transmission months. When violence deters actors tasked with providing municipal services it limits access to health services and implementation of responsive vector control [105]. Vector control agents–who are often tasked with canvassing bairros and entering properties–may neglect areas that are rife with violence, replicating challenges to effective control of yellow fever that were encountered by Oswaldo Cruz in the early 20th century [106]. In conjunction, wary property owners may refuse entry to vector control and health service officers out of fear for their personal safety [107, 108]. In the context of Aedes control, while achieving total coverage is rare, the larger the areas of the city that remain uninspected (and thus untreated), the higher the threat to effective vector control [109].
With regard to bairro structural factors, subnormal settlements were not associated with dengue transmission. This may reflect data limitations. Classification of a census tract as AS does not characterize the degree of subnormality, such as the proportion of houses that are subnormal, or the nature of the AS that predominates in a tract. Persistent negative correlations between population density and dengue incidence may reflect the dengue burden in Fortaleza’s peripheral bairros–where large expanses without dense development and disconnected from proper urban planning are common (e.g. Prefeito José Walter)–and lower incidence rates in densely populated coastal bairros, whether affluent (e.g. Meireles) or not (e.g. Pirambu).
Though we did not analyze serological data, the Fortaleza Municipal Health Secretariat conducts limited sampling to monitor the introduction of allochthonous serotypes. All four serotypes circulated in Fortaleza between 2011–2015, but DENV1 and DENV4 predominated. DENV1 was reintroduced in 2011 after 10 years, and was the primary serotype in 2011, 2014 and 2015. DENV4 was first introduced in 2012, and was responsible for the majority of cases in 2012 and 2013. In all years, the predominant serotype was present in more than 90% of serological samples, indicating that spatial transmission dynamics may be driven by the degree of population susceptibility throughout Fortaleza’s neighborhoods. As a result of diminished population susceptibility, it is unlikely that DENV1 (which caused an epidemic in 2011) was solely responsible for the estimated disease burden in 2015. The Fortaleza Health Secretariat has found that a significant number of Zika virus (ZIKV) cases were misclassified as dengue in 2015, and that DENV1 and ZIKV co-circulated. Though imported cases of chikungunya (CHIKV) were detected by the surveillance of the City of Fortaleza in 2014, autochthonous cases were not confirmed until December 2015. Therefore, the circulation of ZIKV in 2015 offers one possible explanation for the epidemiological curve observed for that year. The epidemic peak in 2015 was lower than in 2011 and 2012, and incidence in 2015 extended into non-seasonal months at a scale that is more characteristic of interepidemic years (Table 1). It is possible that the peak of 2015 transmission is attributable to seasonal misdiagnosis of ZIKV and CHIKV cases, in which protracted incidence would be characteristic of interepidemic dengue virus transmission in Fortaleza. Alternatively, it is possible that 2015 did have seasonal epidemic dengue transmission, and that protracted non-seasonal transmission is attributable to Zika misdiagnosis.
This study has many strengths. First, data are drawn from multiple years, permitting analysis of interepidemic and epidemic transmission patterns, with and without typical patterns of dengue seasonality. Second, by using temperature and rainfall information, seasonality by year could be properly characterized, accounting for possible weather anomalies. Third, the analysis combined entomological surveys with socio-ecological and climatological data to provide a comprehensive assessment of transmission patterns and correlates. Lastly, the study incorporates routine cross-sectional surveillance surveys of immature forms (considered to be a proxy of the concentration of dengue cases in an urban landscape).
This study has some limitations. First, as any analysis that uses administrative records, data refer to passive surveillance, and do not include asymptomatic infections and/or mild cases that do not trigger search for care [110]. This issue may be qualitatively dependent on the intensity of transmission [56], and communities with access to private healthcare providers may underreport at greater rates than low-income bairros [111]. Second, given the circulation of ZIKV in Fortaleza in 2015, the fact that about 20% of cases were lab confirmed in 2015, and the lack of serological tests with high specificity, we expect that some dengue cases reported that year were, in fact, ZIKV infections. Yet, since both diseases are Aedes-borne, they are subject to the same socio-ecological factors, and thus their co-circulation should not detract from our conclusions. Third, some variables may not properly capture access and behavior. For example, while we might have expected associations between dengue and access to piped water during dry seasons (as a result of increases in water storage behavior), an ethnographic study in Fortaleza suggested that potable water storage is pervasive [15], such that bairros-scale statistics on access to piped water may not be adequate to capture the phenomenon. Similarly, including socio-demographic covariates such as income, access to municipal services, interpersonal violence, and prevalence of subnormal agglomerations at the bairro level may not satisfactorily control for socio-economic factors expected to be associated with dengue incidence. As a result, in addition to their role as Aedes breeding sites, the prevalence of strategic points–such as scrapyards and tire shops–in a bairro may be a proxy for unobserved differences that are relevant to the transmission cycle. Fourth, we only have access to one weather station for temperature data; yet, considering the low variability in temperature observed in Fortaleza, we do not expect this to be a major limitation. Lastly, the entomological surveillance data may only be a partial depiction of the scale of infestation as a result of barriers that vector surveillance agents face when conducting their inspection. Since we are using data from a targeted surveillance program, we assume that some areas may be prioritized because of their known high risk for mosquito breeding.
|
10.1371/journal.pbio.1002220 | Beyond the Whole-Genome Duplication: Phylogenetic Evidence for an Ancient Interspecies Hybridization in the Baker's Yeast Lineage | Whole-genome duplications have shaped the genomes of several vertebrate, plant, and fungal lineages. Earlier studies have focused on establishing when these events occurred and on elucidating their functional and evolutionary consequences, but we still lack sufficient understanding of how genome duplications first originated. We used phylogenomics to study the ancient genome duplication occurred in the yeast Saccharomyces cerevisiae lineage and found compelling evidence for the existence of a contemporaneous interspecies hybridization. We propose that the genome doubling was a direct consequence of this hybridization and that it served to provide stability to the recently formed allopolyploid. This scenario provides a mechanism for the origin of this ancient duplication and the lineage that originated from it and brings a new perspective to the interpretation of the origin and consequences of whole-genome duplications.
| Genome duplication is a major evolutionary process that has shaped the genomes of several eukaryotic lineages including vertebrates, plants, and fungi. The sequencing of the baker's yeast Saccharomyces cerevisiae in the 1990s revealed the presence of conserved blocks of duplicated genes, indicating an ancestral duplication of the entire genome. Subsequent work has clarified when this event occurred and what genomic rearrangements followed, but the underlying mechanistic origin of such a large-scale event remains poorly understood. Here we used a large-scale phylogenetic approach to examine the individual evolutionary histories of all yeast genes and assessed the time at which each duplication occurred. This survey revealed evidence for an ancient hybridization event between two ancestral species in the lineage in which the whole-genome duplication had occurred. We further characterize this hybridization event and the properties of the putative parental species. We propose that the whole-genome duplication was a direct consequence of this hybridization, providing a means by which the initially sterile hybrid could regain fertility. This scenario provides a mechanistic understanding of the origin of the ancient yeast whole-genome duplication and brings a radically different perspective on the interpretation of the origin and evolutionary consequences of whole-genome duplications in eukaryotic lineages.
| Ancient whole-genome duplications (WGDs) are major evolutionary events that have impacted several eukaryotic lineages, including plants, animals, and fungi [1]. Among plants, ancestral WGDs have been identified in monocots and core eudicots [2], and more recent events are apparent in many lineages such as Arabidopsis, maize, and soybean [3–5]. In vertebrates, the existence of two ancestral WGDs (but also more recent ones in teleost fishes and frogs) has been proposed [2]. Earlier work has focused on establishing the periods at which these events occurred [6,7] and on assessing the functional and evolutionary aftermath of the doubling of the entire genetic complement [8]. However, we still do not fully understand what initially triggered these events. Perhaps the best-studied WGD is the one affecting an ancestor of the baker's yeast Saccharomyces cerevisiae, an event supported by the finding of numerous blocks of paralogs with conserved synteny [7,9]. It is now established that this event occurred just before the separation of Vanderwaltozyma polyspora from the S. cerevisiae lineage, originating a clade of post-WGD species (Fig 1A) [10]. In addition, it has been shown that the genome doubling was followed by extensive genome rearrangements and rampant gene loss that have since shaped these species' genomes, resulting in only a minor fraction of the WGD-derived paralogs (ohnologs) being retained [11,12]. Based on the high level of synteny found between reconstructed ancestrally duplicated gene blocks, it has been proposed that the yeast WGD has its origin in an autopolyploidization event [11]. This proposition has important implications with respect to the possible initial selective advantages that played a role after the polyploidization event. Polyploidy has been considered to promote evolutionary innovation because it facilitates neo- and subfunctionalization and buffers deleterious mutations. However, these mechanisms only provide an advantage after some time has passed and a number of mutations have accumulated. Conversely, simple increase in ploidy has been considered to put barriers to fast adaptation, as it masks beneficial recessive mutations and avoids rapid purging of deleterious mutations. Furthermore, most experimental work comparing populations of different ploidy generally provides support for the superiority of the normal ploidy versus increased ploidies in a given species [13]. Thus, the nature of the initial evolutionary advantage of the yeast WGD remains an open question.
WGDs leave a footprint in the form of cohorts of homologous genes that duplicated in the same period. Phylogenetic analysis of gene families informs on the relative age of duplications [15,16] and hence is a powerful tool to study WGDs. When ancestral duplications are inferred from the genes encoded in a genome and their relative dates are mapped to a reference species tree, ancient WGDs are expected to lead to an accumulation of duplications mapped to the lineage in which the event occurred. Earlier analyses have used such approach to detect ancient duplications in vertebrates [17,18] and plants [19]. However, despite extensive phylogenetic work [20–22], no study has assessed the global phylogenetic congruence of gene duplications and the WGD that occurred in the lineage leading to S. cerevisiae. Here, we set out to investigate patterns of past duplications in S. cerevisiae by analysing genome-wide sets of gene phylogenies (i.e., phylomes).
We based our analyses on a set of 26 completely sequenced genomes, for which we reconstructed a reference species phylogeny based on the alignment concatenation of 516 widespread, single-copy orthologs (See Fig 1A, Materials and Methods). Subsequently, we used the phylomeDB pipeline [23] to reconstruct the evolutionary history of every protein encoded in the S. cerevisiae genome. These gene family trees were used to detect and date well-supported duplication events, using a phylogeny-based method described elsewhere [16]. In brief, the method exploits the temporal information provided by the branching patterns in a given gene tree: a duplication must be older than the lineages diverging subsequent to it and younger than lineages branching earlier. Using this information, we can map duplications to the reference species tree and compute duplication densities per gene and branch (S1 Fig). Unexpectedly, our analyses revealed the largest duplication peak (0.28 duplications per gene) at the branch preceding the divergence between Saccharomyces and a clade containing the genera Kluyveromyces, Lachancea, and Eremothecium (Ashbya gossypii) [24], hereafter referred to as KLE (Fig 1B). To assess whether this peak was indeed related to the WGD event, we limited our analysis to those duplications leading to conserved pairs of WGD-ohnologs as defined in the Yeast Gene Order Browser (YGOB) database [25]. Note that YGOB uses a synteny criterion which is independent of the specific gene phylogeny. We found that the pre-KLE duplication peak was more apparent in the subset of duplications leading to conserved pairs of ohnologs, which indicates that this ancestral duplication peak is indeed related to the observed WGD paralogous blocks (Fig 1A and 1B). Of note, not all duplications resulting in pairs of conserved, syntenic ohnologs mapped to the pre-KLE peak (n3). A second accumulation of duplications appeared at the branch preceding the divergence of a clade formed by Zygosaccharomyces rouxii and Torulaspora delbrueckii (referred to as ZT hereafter) with the post-WGD species (n4). A smaller fraction of duplications mapped to the expected WGD location (n5) or subsequent branches. We assessed the degree of divergence between syntenic ohnologs derived from duplications at the pre-KLE peak and those from duplications at the WGD node, as the two more divergent points of interest, and found that the former had significantly larger divergences (Fig 1C). This supports that gene pairs whose duplications are predicted to be more ancestral by a topological approach are also more divergent at the sequence level. It also indicates that the genes in paralogous blocks may be composed of distinct sets of genes, diverged at different times.
To discard the possibility that our unexpected result was artifactual and to understand what may have caused the dispersion of the duplication mappings outside the WGD node, we carefully assessed possible methodological and interpretation pitfalls. First of all, given that the pre-KLE branch is among the longest in our species phylogeny, the ancestral peak could simply indicate a higher number of duplications accumulated over a longer period of time. We thus measured the correlation between duplication densities and branch lengths for the whole phylogeny. While a high correlation was indeed observed when considering all the duplications (r2 = 0.92 Pearson), this was not the case when the analysis was restricted to only those duplications leading to syntenic ohnologs (r2 = 0.00). We next assessed the effect of using alternative yeast species as a seed in the phylome reconstruction and observed that the use of Candida glabrata or V. polyspora phylomes resulted in similar patterns of duplication densities (see S2 Fig). Another, always contentious point is the use of a reference phylogeny. Although the reconstructed species tree was highly supported and congruent with earlier reconstructions [24], an alternative branching order for the KLE species had been previously presented [26]. This alternative topology suggested that the Lachancea, Kluyveromyces, and Eremothecium are not monophyletic but rather stem out sequentially from the lineage leading to S. cerevisiae (see S3 Fig). Such organization could potentially affect our results if, for instance, the pre-KLE duplications were found to be partitioned among the new internodes (i.e., branches) created by this topology. To test this, we repeated the analysis using the alternative topology as a reference. Our results show that the underlying topology does not affect the central finding that an apparent duplication peak existed before the divergence of KLE species (S3 Fig). Finally, we tried an alternative method to map the duplication events of ohnologs by using the reconciliation-based algorithm implemented in Notung [27], which rendered similar results (S4 Fig). Thus, a different species topology and a different duplication detection method do not alter the main result that the majority of ohnologs have apparently diverged before the expected WGD.
We next tried to assess the possible effect of stochastic errors or artifacts in the gene trees. We did so by focusing on the trees that contained pairs of conserved ohnologs in S. cerevisiae. Short sequences tend to be less reliable and more prone to stochastic errors. First, we examined the signal present in subsets of sequences of varying lengths (<500 aa, 500 to 1,000 aa, and >1,000 aa). As seen in (S5 Fig), the three groups of genes consistently provide a very low duplication signal at the WGD, while signals at the two previous branches (pre-ZT and pre-KLE) are much larger.
Secondly, we assessed the robustness of our main result across a range of different methodological approaches for gene tree reconstruction. We tested three different maximum likelihood programs: PhyML [28], RAxML [29], and Fasttree [30]; and one program based on Bayesian inference (BI): Phylobayes [31]. In addition to the best-fitting evolutionary model used in our standard analyses, we used PhyML to test the effect of using two different, more complex models (C20-CAT [32] and Covarion [33]) and a different search heuristic, subtree pruning and regrafting (SPR), instead of the default nearest-neighbour interchange (NNI). Finally, we tested two different support methods in RAxML (rapid bootstrapping and Shimodaira–Hasegawa (SH) support) and in PhyML (approximate likelihood ratio test [aLRT] and bootstrapping). A summary of the different methods can be found in S1 Table. Results of different methods are not directly comparable because different subsets of trees pass the filters for a given procedure (see S6 Fig). However, when a tree passed the filters for any given two methods, the result was highly consistent in most cases (86% overall agreement). Overall, our main result that duplications are apparently older than the expected WGD remained consistent (see S7 Fig). The fraction of ohnolog duplications mapped to the expected WGD node is minimal (<15% in all datasets), while more ancestral duplications are prominent with >50% of the duplications being mapped to the two nodes preceding the WGD (pre-ZT and pre-KLE), although the balance between these two prominent peaks differed between the methods. These differences notwithstanding, the main conclusion of the duplication density analysis is consistent across methods: the majority of ohnologs have inferred duplication ages that predate the expected time of the WGD.
Finally, phylogenetic artifacts such as long-branch attraction (LBA) can produce wrong topologies with high support [34]. It is possible that trees containing paralogs diverging at very unequal rates may have been affected by LBA, misplacing duplications closer to the root. In fact, differential rates among paralogs are expected when processes of neofunctionalization are acting. One way to ascertain whether LBA is affecting the topology is reconstructing the tree with and without the out-groups. In the absence of LBA, the in-group topology is expected to remain stable [34]. We applied this test to the trees containing ohnologs and found that the majority of trees (85%) gave consistent mappings of the duplication of the ohnologs, indicating that the effect of LBA is not widespread and does not significantly affect the duplication mapping. We performed a second test to see whether LBA could explain the observed patterns. For this, we devised sequence simulations in which one of the ohnologs was made to evolve 20 times faster than its paralog. Despite the use of such extreme values, the duplication peak at simulations was detectable at the expected location, and artifactual peaks were significantly smaller and not apparent at the pre-KLE lineage (see Materials and Methods, S8 Fig). Gene conversion among duplicates may result in underestimation of duplication ages, possibly accounting for part of the disappearance of the WGD peak, but not for the presence of the pre-KLE peak. Thus, LBA and gene conversion may have blurred the signal of the WGD peak but cannot account for the prominent pre-KLE peak.
Our results show compelling evidence that a majority of yeast genes defined as ohnologs have diverged before the expected period of the WGD. This overall result holds even though the exact mapping from individual gene trees may vary across methodologies and datasets. The event under study is very ancient, and genes contain a limited amount of information; thus, degradation of the signal is expected. However, stochastic noise would explain a diffusion of the signal but not the existence of a stronger, more ancient duplication peak. We have also shown that distorting processes such as LBA cannot account for the observed patterns. We thus turned to assess other possible biological explanations for our observation.
We further considered possible evolutionary scenarios that could result in the observed patterns of ancestral duplications seen for the ohnologs. We reasoned that an interspecies hybridization would result in phylogenetic patterns reminiscent of duplications that would be mapped to the common ancestor of the two hybridizing species (Fig 2A), providing a possible scenario to explain our puzzling results. The process of hybridization originates a new lineage by bringing together two diverged genomes. Orthologous genes coming from each of the parental species would appear as paralogs in standard analyses, since they are homologous genes encoded in the same genome [35]. A phylogenetic analysis, however, would map the apparent duplication to the time of divergence of the two parental species (Fig 2B). This necessarily predates the time of the formation of the hybrid: that is, the hybridization point does not coincide with the point at which the apparent duplications would be mapped. As we will see below, our hypothesis is that the hybridization may have shortly predated the actual WGD point (i.e., occurred at node n5 in Fig 1).
In support of this, we calculated the duplication densities on the well-studied yeast interspecies hybrid S. pastorianus [36]. This species is the result of a recent hybridization between S. cerevisiae and S. eubayanus [36]. The sequenced genome of S. bayanus is the closest related genome to S. eubayanus, and therefore we expect the highest duplication peak to appear at the common ancestor between S. cerevisiae and S. bayanus. The duplication density analysis, as predicted, yielded an apparent duplication peak at the common ancestor S. cerevisiae and S. bayanus, but not at the lineage where the hybridization and the doubling of the genome is known to have occurred (Fig 2C).
The results found for the S. cerevisiae lineage could thus be readily explained by a past hybridization between lineages diverging just after the observed peak and before the post-WGD species. Considering this and the current genomic sampling, species close to, but not necessarily within the KLE and ZT clades, would be the prime suspects of potential partners in the proposed ancestral hybridization (Fig 2D). To explore this possibility further, we inferred properties of the two putative parental lineages from the current genomic sampling. We did so by inspecting individual S. cerevisiae gene phylogenies in the above-mentioned phylomes (see Fig 3 as an example) and by measuring phylogenetic affiliations using phylomes reconstructed with reduced taxonomic sets (see S2 and S3 Tables). Phylogenetic affiliations were measured by scanning the gene tree topologies to examine the species contained in the sister groups (i.e., neighbouring clades) of the sequences from post-WGD species (see Materials and Methods). We categorized them according to one of the two lineages that diverged after the pre-KLE peak and the origin of the post-WGD species: the KLE clade and the ZT clade. From now on, we consider the ZT cluster as the extant clade closest to one of the parents (parent A), while the KLE cluster will be considered as the closest to the other parent (parent B). Although, for simplicity, we refer to ZT and KLE clades as parental lineages, it must be clearly stated that it is our understanding that the actual parents may have been close to, but not necessarily within, these clades. Accordingly, three possible topologies can be considered: two in which the S. cerevisiae seed sequence groups with either parental species (A or B, respectively) and a third one in which the S. cerevisiae sequence has the two parental lineages as a sister group (C) (see Fig 4A). Our results (Fig 4A) indicated that a large majority (60%–82%, depending on the choice of species used in the reduced phylome; see S9 Fig) of the trees showed a topology congruent with the currently accepted phylogeny, i.e., the post-WGD species grouping with the ZT clade. When only the trees that contain S. cerevisiae proteins with a conserved ohnolog are considered, the results remain very similar (see Fig 4A) (54%–78%, depending on the choice of species used in the reduced phylome; see S10 Fig). This suggests that this or a related lineage would have been involved in the hybridization (parent A) and that genes derived from this parental species constitute a majority of the genome in extant post-WGD species. In contrast, a remarkably low fraction of genes showed an affiliation only to the KLE lineage (4%–14%), whereas a larger percentage (14%–28%) of genes had as a sister group a combination of the two putative parental clades (C). This would suggest that one of the actual parental lineages did not belong to the KLE but rather diverged before. The analysis repeated using different phylogenetic methodologies confirmed these results (S11 Fig).
The high percentage of trees supporting the A topology could be the result of total or partial gene conversion, which is common in recent hybrids [37]. We can only clarify this matter by analysing gene trees that contain pairs of conserved ohnologs. Depending on the distribution of the two ohnologous genes when compared to KLE and ZT, we can distinguish between nine different topologies (see S12 Fig). Forty percent of the trees contained a topology in which the two yeast ohnologs grouped together (topologies A–A 1, B–B 1, and C–C 1). This could be due to total or partial gene conversion from one of the parents to the other. The gene conversion events seem to favour genes from parent A, since in 30% of the mentioned cases both retained genes are more closely related to this parent. We performed sequence evolution simulations including different degrees of gene conversion to estimate what levels would be necessary to alter the tree topology (see Materials and Methods). In our settings (S13 Fig), conversion of 25% of the gene sequence was sufficient to lead to a higher probability of the duplication being mapped to a younger node. Thus, gene conversion, which renders duplications to appear younger, has a much larger effect than LBA.
These analyses underscore the difficulty of correctly determining the position of parent B. There is a strong signal for the parent B to have diverged just before the KLE clade (shown by topology A–A 2 and B–B 2), which is present in 33% of the trees. As we will discuss below, we consider that recombination between the two parental subgenomes, including total or partial gene conversion, must have been common in the period following the hybridization, explaining not only the bias in descent among ohnologs and singletons but also the widespread mixture of phylogenetic signals in gene trees that is typical for this clade [20].
The availability of genomes from fungal species in which recent hybridizations or WGDs have been described allows us to assess the patterns of phylogenetic affiliations and compare them with the patterns observed for S. cerevisiae. On the one hand, Rhizopus delemar [38] and Hortaea werneckii [39] are thought to have undergone a recent WGD. On the other hand, S. pastorianus [36] and the wine strain S. cerevisiae x S. kudriavzevii VIN7 [40] are recognized as recent hybrids for which the putative parental species are known. It is important to remark that some of the described WGD species may indeed be as well the result of hybridizations, as it is proposed here for the post-WGD clade, but that the current sampling of species prevents the detection of the alternative parental signals. We reconstructed the phylomes of these four species (see S2 Table) and computed phylogenetic affiliations as explained above, but adjusting A and B to the known parents or the corresponding neighbouring clades. Putative WGD species showed a clear dominance of the immediate preceding clade (Fig 4B). The recent hybrids, on the other hand, presented a split topology distribution, with roughly half of the trees supporting the A topology and another half supporting the B topology (Fig 4C). This clearly provides evidence of the dual origin of these species. As negative controls, we examined the phylomes of two species without anomalous ploidy, Candida albicans and Penicillium digitatum (Fig 4D) [41], and the above-mentioned simulated yeast phylome in which one of the ohnologs was evolving at a faster rate (Fig 4E). This analysis shows that hybrids present a clear dual pattern of phylogenetic affiliations when the gene phylogenies are examined in the presence of the two parental lineages. This pattern is clearly distinct from what is in genomes with normal ploidy or in recent WGDs. This dual pattern is also present in the analysis of the yeast genome. Of note, in this case the two alternative phylogenetic affiliations are not equally represented. This difference with respect to recent hybrids can be attributable to the larger period of time since the hybridization and the preferential loss or conversion of genes coming from one of the parental lineages, which necessarily altered the balance between the two phylogenetic affiliations.
As mentioned above, inferred ancestral collinearity has been used to favour simpler WGD scenarios involving autopolyploidization [11]. However, such studies indistinctly used KLE and ZT clades to infer ancestral gene arrangements and thus could not inform about differences between the putative parents. Although the position of the parental species cannot be ascertained with confidence, we can take KLE and ZT clades as the two extremes of their possible divergence. We therefore assessed the level of micro- and macrosynteny conservation among the KLE, ZT, and post-WGD clades, by considering them separately. To do this, we reanalysed the information of orthology and syntenic blocks provided by YGOB [42]. We first assessed the differences between ZT and KLE by searching for gene arrangements conserved within ZT and KLE, but different between the two groups. These differences can be considered ancestral to the two groups and thus likely present at the time of the proposed hybridization. Only 32 cases of broken synteny and 11 translocations of a single gene were noted (S4 Table). When searching for these synteny breaks in post-WGD species, we found that they had inherited the arrangement present in either KLE or ZT in similar amounts (15 and 17, respectively) (S5 Table). Of note, the patterns shared by KLE and post-WGD species could result from lineage-specific rearrangements in the lineage leading ZT clade so we cannot unequivocally impute them to the hybridization. This result is consistent with the absence of disagreements between syntenic ohnologous blocks noted earlier [11].
However, an autopolyploidization scenario would predict a larger number of shared syntenic arrangements between the post-WGD and its closest clade (ZT). Furthermore, the absence of disagreements in such a small number of blocks can be explained by other factors, including gene conversion, so it cannot be considered a definitive proof of autopolyploidization. In addition, we found that the number of conserved pairs of adjacent orthologs between KLE and ZT clade was high, as was the number of conserved pairs between post-WGD species and any of the KLE and ZT clades (S6 and S7 Tables). Finally, we found no differences in terms of the minimal amount of rearrangements [43] between each S. cerevisiae syntenic block [11] and those in either ZT or KLE species (Fig 5). These results speak for the high collinearity of the two putative parental clades at the proposed time of hybridization (see S4 and S5 Tables), which is congruent with the small divergence time estimated between the two clades at the time when the post-WGD clade originated (S14 Fig). In addition, considering the high level of collinearity between the ZT and KLE clades, and the lack of differences in terms of synteny conservation when compared to S. cerevisiae, the proposed hybridization is as compatible with the observed level of conserved synteny between duplicated blocks as a simpler autopolyploidization scenario. Recent yeast hybrids have been shown to present extensive recombination between parental genomes, including total or partial gene conversion [37,44,45], which breaks the initial correlation between phylogenetic origins of neighbouring genes and removes sequence and structural differences between homologous chromosomes. This and extensive differential gene loss and genome rearrangements that have occurred within the post-WGD clade have presumably eroded the few initial differences between the two parents that we can reconstruct. We conclude that, given the similar levels of collinearity implied by both scenarios and the confounding effects of extensive gene loss, homologous recombination, gene conversion, and genome rearrangements, synteny cannot be used in this case to disentangle whether the WGD was triggered by an autopolyploidization or a hybridization event.
The proposed hybridization is a very ancient event, and thus, the remaining signal must be necessarily weak. We have shown that gene order differences between the putative parental species involved in the hybridization were extremely low, and we consider that this signal may have been completely eroded, which explains why the hybridization was not evident from earlier analyses based on synteny. Our phylogenetic results, however, do provide clear support for the existence of an ancient interspecies hybridization and are not compatible with a simple autopolyploidization scenario. The observed phylogenetic affiliations in ohnologs and singletons, biased towards one of the putative parental lineages, as well as the absence of synteny disagreements in ohnologous blocks, can be reconciled with the assumption that the proposed hybridization was followed by widespread recombination events between the two parents subgenomes, some of which would have led to partial or total gene conversion. As noted before, this process is common in recent yeast hybrids [37,44,45], and it is natural to expect that this would have occurred in an ancient hybridization. Notably, hybridization followed by recombination between parental subgenomes also explains another long-held observation of the post-WGD clade: that there is a variable mixture of disparate phylogenetic signals present across different gene trees [20,22].
Our results also indicate that an apparently more ancestral duplication peak occurred in addition to duplications around the expected WGD point. We hypothesize that the occurrence of these two rare events in the same lineage is not the result of coincidence. We propose two possible scenarios that naturally link the two events and explain the observed patterns (Fig 6). In the simplest scenario, two diploid cells from distinct species form an allotetraploid. Subsequent recombination and massive gene loss would render a lineage in which the number of chromosomes has effectively doubled. In this case, hybridization directly results in the observed WGD, because a fraction of the final gene set is retained as “ohnologous” pairs, either from the same or from different parental species. Alternatively, two haploid cells from different species form an allodiploid. Such hybrids are largely unstable and cannot undergo the sexual cycle, but they can propagate clonally [46]. An additional duplication by autopolyploidization would stabilize the hybrid by enabling meiotic recombination. This mechanism, which also prevents backcross with the parental lineages, has been proposed as a necessary step to stabilize some interspecies hybrids [47] and is a scenario commonly considered in recent plant hybrids [48]. Both scenarios cannot be distinguished with the data at hand, but the ability of haploid cells to fuse through mating provides a possible mechanism for the latter. Further investigation of how these two mechanisms participate in the formation of natural hybrids is necessary [46]. Importantly, some of the steps proposed by our model were also contemplated in models considering autopolyploidization scenarios [10]
Our results provide compelling evidence for an ancient hybridization in the yeast lineage and bring about novel implications in our understanding of the evolution of eukaryotic genomes and the origin of functional divergence after WGDs. Remarkably, besides the pattern of ancient duplications, the proposed model provides plausible explanations to other common observations in the post-WGD clade. The phylogenetic relationships within and around the post-WGD clade have always been difficult to resolve, and a great diversity of phylogenetic histories among different genes has been noted [20,22]. A chimeric origin of the clade, combined with events of recombination between genes from different parents—as observed in current hybrids [44,49]—would readily explain an increased variability in phylogenetic signals recovered from different genes. Such intragenic recombinations, together with full gene conversion and differential gene loss, may as well partially explain the observed dispersion of the phylogenetic mapping of duplications from syntenic ohnologs around the expected WGD point. Furthermore, ohnologs have been shown to present selection pressures intermediate of singleton genes and those from small-scale duplications of a similar age [50]. Finally, notable exceptions to expectations from the gene balance hypothesis, which posits that WGD would favour duplications of entire complexes rather than single subunits, have been noted [51]. Most of these observations have been interpreted in the light of an assumed rapid sequence and functional divergence after duplication. However, under a hybridization scenario, a fraction of the predicted ohnologs originate from distinct species, and thus, sequence and functional differences are expected from the start. In contrast, an autopolyploidization scenario poses the problem of how reproductive isolation was achieved and faces the lack of a clear selective advantage before neo- or subfunctionalization occurs. Interspecies hybridization brings together different physiological properties and isolates sexually the newly formed lineage, hence providing an initial selective advantage to explain observed WGDs in eukaryotes. Considering the widespread presence of hybrids among current species, this scenario should also be considered when interpreting ancient polyploidies. The proposed approach and an increased genome sampling around the relevant lineages will enable testing the possible implication of interspecies hybridization in other eukaryotic WGDs.
Proteomes were downloaded from their original databases (S8 and S9 Tables). The proteomes of S. pastorianus and H. werneckii were not available. We thus downloaded the genomes and predicted their proteomes using Augustus [52]. The final S. pastorianus [36] and H. werneckii [39] proteomes comprised 11,460 and 20,509 proteins, respectively.
Phylomes—complete collections of phylogenetic trees for each gene encoded in a given genome—were reconstructed using the automatic pipeline described in Huerta-Cepas et al. [23]. Briefly, the pipeline starts with a seed genome and proceeds as follows: for each protein encoded in the seed genome, a Smith-Waterman similarity search was performed against a database containing the proteomes listed above. Results were then filtered based on e-value (<1e-05) and sequence overlap (>50% coverage over the query sequence). The query and the selected hits (homologous sequences) were then aligned using a sophisticated multiple sequence alignment strategy in which three different alignment programs were used (Muscle v3.8 [53], Mafft v6.712b [54], and Kalign v2.04 [55]) to align the sequences in forward and reverse orientation. The resulting six alignments were combined into a consensus alignment using M-coffee [56]. This alignment was then trimmed to remove poorly aligned columns with trimAl v1.3 [57] using a consistency-score cutoff of 0.1667 and a gap-score cutoff of 0.9. Trees were reconstructed using the best-fitting evolutionary model. The selection of the model best fitting each alignment was performed as follows: a neighbour joining (NJ) tree was reconstructed as implemented in BioNJ [58]; the likelihood of this topology was computed, allowing branch-length optimization, using seven different models (JTT, LG, WAG, Blosum62, MtREV, VT, and Dayhoff), as implemented in PhyML v3.0 [28]; the two models best fitting the data, as determined by the AIC criterion [59], were used to derive maximum likelihood (ML) trees. Four rate categories were used, and invariant positions were inferred from the data. Branch supports were computed using an aLRT based on a chi-square distribution, as implemented in PhyML [60].
S2 Table lists the complete phylomes reconstructed for this project. Seven complete phylomes were reconstructed using S. cerevisiae, C. glabrata, V. polyspora, S. pastorianus, H. werneckii, the yeast S. cerevisiae VIN7, and R. delemar as seed species. These phylomes have been deposited in phylomeDB (http://phylomedb.org [61]). A simulated phylome using S. cerevisiae as seed was also reconstructed (see below). In addition, a total of 18 reduced phylomes were reconstructed (see S3 Table; http://genome.crg.es/~mmarcet/yeast_hybrids/phylome_table.htm). In these reduced phylomes, for the seed species, only one sequence was present in the tree; all paralogs for this species were removed to ensure that a clear phylogenetic position could be established. Finally, two previously reconstructed phylomes, stored in phylomeDB, were used for comparative purposes: C. albicans (phylomeID: 205) and P. digitatum (phylomeID: 150) [41]. Phylomes were scanned using ETE v2.2 [62], which implements all the algorithms described here.
The reference species tree shown in Fig 1 was reconstructed using a multigene concatenation method. From the S. cerevisiae phylome, we selected 516 protein-coding genes found in single copy across the 26 species considered. Their protein alignments were then concatenated, resulting in a combined alignment of 285,507 positions. An ML phylogenetic tree was then reconstructed using PhyML v3.0 [28] using the LG model. Four rate categories were used, and invariant positions were inferred from the data. Bootstrap support was calculated based on 100 replicas. All nodes were fully supported (100% bootstrap). The species tree presented in S3 Fig was reconstructed using the same data as the previous tree, but enforcing the desired topology when reconstructing the tree. For the S. pastorianus tree (Fig 2C), 215 genes were selected from the phylome, and the final alignment contained 117,408 amino acids. The same methodology was used to reconstruct the tree.
Each tree in a phylome was scanned to detect and date duplications using a phylogeny-based algorithm described earlier [16]. In brief, this algorithm traverses the tree and uses a so-called species-overlap algorithm to detect duplication nodes. Duplication nodes are defined as those nodes where the two daughter branches share at least one species. The relative age of this duplication is assumed to be at the last common ancestor of the species diverged after the duplication (i.e., those contained in the two daughter branches). Each duplication was then mapped onto the corresponding ancestral lineage in the species tree. The total number of duplications was divided by the total number of trees that were rooted at a deeper branch in the species tree (i.e., those that are informative for the evaluated lineage). For instance, to estimate the duplication density at the WGD branch, only trees that contain at least one pre-WGD species were considered. S1 Fig shows a schematic representation of the duplication mapping process. This analysis was performed using three different phylomes, in which S. cerevisiae, V. polyspora, and C. glabrata were used as seed, respectively. For each phylome, two different datasets were used. In the first one, all the trees in the phylome were used (see Fig 1B, green dot, and S2 Fig, lighter dots), the second was based on trees in which a pair of retained ohnologs was present, and only the duplication node leading to the two seed ohnologs was used (see Fig 1B, yellow dot, and S2 Fig, darker dots). Ohnologs were obtained from YGOB [42], which uses a synteny criterion combined with sequence similarity but is not phylogenetically informed. Only trees that contained both ohnologs were considered. This second set ensured that the duplication density was not affected by duplications not related to the WGD event. We plotted the correlation between duplication densities and branch lengths.
We mapped the duplication event of the two ohnologs to the species tree and only kept those S. cerevisiae sequences whose duplication point mapped to the WGD node or to the pre-KLE node (see Fig 1). Only trees that contained at least one ZT sequence, one KLE sequence, and one out-group sequence were considered. Blast scores were normalized by dividing the blast score obtained when searching from a seed yeast protein to the ohnolog pair by the blast score obtained from searching the seed yeast protein to itself. In a separate analysis, pairwise alignments of the conserved ohnologs were reconstructed using Muscle v3.8 [53]. The Kimura distance between the two sequences was calculated using protdist as implemented in the phylip package [63]. The frequency of distances of the two different distributions and blast score frequencies were plotted with R [64]. Significance of the difference in distributions was assessed using a two-sample Kolmogorov-Smirnov test (see Fig 1C). The two populations were significantly different, with a p-value for the blast scores of 2e-04 and for the Kimura distance of 2.9e-05.
PL-R8s [14] was used to assess the divergence times in the concatenated species tree (S14 Fig). Smoothing parameter was estimated using cross validation. The divergence between S. cerevisiae and C. albicans (235 MyA as estimated by Douzery et al. [65]) was used as calibration point. The same protocol was used in individual trees that contained two ohnologous pairs. Trees were pruned so that they only contained the closest sequence belonging to each ZT-KLE group. The frequencies of ages (see Fig 1C) were plotted using R [64]. The two populations were significantly different, with a p-value of 4.5e-07.
Notung v2.6 [27] was used to reconcile the same set of trees used above to the species tree obtained from the concatenation of 516 proteins (see above). Once the two trees were reconciled, we used the option to estimate upper and lower bounds to obtain the time when the duplication of the two S. cerevisiae ohnologs had taken place. Only estimates that had a definite upper and lower bound that could be mapped to a single branch of the species tree were considered. The number of trees that mapped the duplication onto a given branch was divided by the total number of trees in order to obtain the duplication density.
A set of 846 trees were selected from the S. cerevisiae phylome where pairs of conserved ohnologs were found, as predicted by YGOB [42]. The alignments were taken from the phylome reconstruction done previously. Then, for each tree, several additional phylogenetic reconstruction methods were used. Fasttree [30] was used with default values. PhyML [28] was run again three times; in all cases, four rate categories were applied and invariant positions were calculated from the data. The first time the CAT model C20 was used [32], the second time the Covarion model [33] was used (—cov_free –cov_ncats = 3), and finally, the same models as in the phylome were used, but instead of using NNI to estimate the tree topologies, SPR was used. For the three methods, the aLRT support was calculated. A fourth run with PhyML was performed using the same method as during the phylome reconstruction, but instead of calculating aLRT support values, bootstrap values based on 100 replicates were computed. RAxML [29] was applied using the PROTGAMMALG model and rapid bootstrapping to obtain the branch support. The SH support as implemented in RAxML was calculated over the same set of trees. A Bayesian approach was also used. Phylobayes [31] was used to reconstruct the trees; for each tree, two chains were run for a minimum of 500 cycles; every 100 cycles, the two chains were automatically compared; and if the discrepancies were lower or equal to 0.3 and the effective sizes were larger than 50, the process was stopped. The majority rule consensus, annotated with posterior probabilities, was obtained for each tree.
For each set of trees, the duplication density for the duplication point that led to the diversification of the two S. cerevisiae ohnologs was calculated. Results can be found in S7 Fig. Only nodes in which the support value at the common ancestor of the two ohnologous sequences has an aLRT > 0.95 or a bootstrap > 95 or a posterior probability > 95 were considered.
The same set of 846 trees was reconstructed with no out-group sequences using the same methodology used for phylome reconstruction (see above). The trees included only the post-WGD sequences and the ZT and KLE sequences. Trees were then checked to see whether the two S. cerevisiae ohnologs had a common ancestor that contained no sequences of the ZT and KLE groups, therefore giving support to the WGD, or if they had sequences of either group in between. Only trees in which the common ancestor of the two S. cerevisiae sequences has a support over 0.5 were considered. The same procedure was performed in the same set of trees taken from the phylome. Out-groups in this case were used to root the tree, and then the same analysis was performed. Fifteen percent of the trees gave a different prediction when the two methodologies were performed.
For each sequence encoded in the yeast genome that had one-to-one orthologs in all the species considered, alignments obtained during the phylome reconstruction were trimmed to remove all positions with gaps. The number of species considered was reduced to 12, including S. cerevisiae, all the species belonging to the ZT and KLE clades (T. delbrueckii, Z. rouxii, Kluyveromyces lactis, A. gossypii, Lachancea kluyveri, L. thermotolerans, and L. waltii) and four outgroups (Schizosaccharomyces pombe, Yarrowia lipolytica, C. albicans, and Wickerhamomyces anomalus). The species tree (see Fig 1) was pruned to match this set of species. The existing tree branch that contained S. cerevisiae was bifurcated to create two new branches containing simulated yeast paralogs. The first branch contained the original S. cerevisiae leaf, but its branch length was cut in half. The second branch contained a new S. cerevisiae leaf with a branch length ten times longer than the original. Each protein was then made to evolve along this tree using Rose [66]. Tree-puzzle [67] was used to obtain the mutation frequency observed at each site of the alignment. Tree-puzzle was run with the JTT model; the gamma distribution was estimated from the data using 16 rate categories. These mutation frequencies produced very conserved, unrealistic alignments with few mutations; therefore, the frequencies were multiplied by 20, resulting in more realistic alignments. Indel frequency was set at 0.0003. The resulting sequences were then treated as a newly simulated phylome, which was run through the phylome pipeline. Duplication densities were then mapped onto the species tree (see S8 Fig).
Reduced phylomes were reconstructed in such a way that they contained only one species for the post-WGD (seed species), one species for the ZT clade, and one for the KLE clade, in addition to three outgroups (C. albicans, Y. lipolytica, and S. pombe). In addition, for the seed species, only the seed sequence was included; other paralogs in this organism were excluded from the tree. A reduced phylome was reconstructed for each pair of ZT-KLE species. Three post-WGD species were used as seed (S. cerevisiae, C. glabrata, and V. polyspora) (see S3 Table). For each seed sequence in the reduced phylomes, the sister branch was analysed. First, trees were excluded if they did not have any homologs in ZT, in KLE, or in any of the out-group species. Then, the support of the clade containing the seed sequence and its most immediate neighbouring clade was evaluated using aLRT values. Only clades with support higher than 0.95 were considered. The phylogenetic affiliation of the seed sequence was classified into one of the following groups, according to the species that were present in its neighbouring clade (i.e., sister branch): A, the species located in the sister branch belonged to the ZT clade formed by Z. rouxii and T. delbrueckii (putative parent A); B, they belonged to the clade formed by A. gossypii, K. lactis, L. thermotolerans, L. waltii, and S. kluyveri (KLE clade, putative parent B); and C, they contained a mix of both clades. This was done for the whole phylome (S9 Fig) and for the trees in which the seed sequence was part of a conserved ohnologous pair (S10 Fig). Analysis was repeated across several phylogenetic methods (see above) (S11 Fig).
Topologies of pairs of ohnologs were assessed by reconstructing the trees including the ohnologous pair to those trees that already contained a sequence with a conserved ohnolog. Depending on the relation between the two ohnologs and the chosen KLE and ZT parent sequences, we distinguish between nine possible topologies: A–A 1, A–A 2, B–B, B–B 2, C–C, C–C 2, A–B, A–C, and B–C (see S12 Fig).
For the complete phylomes used (C. albicans phylome, H. werneckii phylome, S. pastorianus phylome, R. delemar phylome, and S. cerevisiae x S. kudriavzevii VIN7 phylome), the two groups of species situated closest to the seed species according to the species tree were used as parental species unless the parental species were known (see S2 Table). Trees were then pruned so that only the seed, the two parents, and out-groups were kept. ETE v2.2. [62] was then used to analyse the sister branch (i.e., neighbouring clade) to the seed sequenced. Sequences were classified as explained above.
For the same set of sequences used in the LBA simulation (see above), we used ROSE [66] to make the sequences evolve along a species tree that contained two S. cerevisiae sequences. The branch lengths of the tree were inferred by selecting those genes that had an A topology and a C topology and were consistent across different phylogenetic methods. Two species trees were derived from these two sets of genes, and branch lengths were mapped onto our simulated species tree. Once sequences were reconstructed, sets of genes affected by different levels of gene conversion were reconstructed. For each percentage of gene conversion, one yeast sequence was taken for each set of sequences, and a given percentage of its sequence was replaced by the same fragment of the second yeast sequence. Phylogenetic trees were then inferred in the same way used in the phylome (see above), and duplication densities were calculated (see S13 Fig)
Orthologous relationships between species and gene order data were obtained from the YGOB. Blocks of conserved synteny between the S. cerevisiae genome and the ancestral genome as predicted by Gordon et al. [11] were considered as conserved syntenic blocks.
The genome of L. waltii was not used because of the high fragmentation of the assembly. Genes in the genomes were arranged using Z. rouxii as reference (see S5 Table). Genomes were scanned for the presence of breaks in gene order that were common in the KLE clade and not found in either ZT species. Orthologs of the genes surrounding the breaks were searched in five post-WGD species (S. cerevisiae, Tetrapisispora blattae, Kazachstania naganishii, Naumovozyma castellii, and C. glabrata) in order to assess whether they followed the ZT or the KLE clade in their gene order (see S5 Table).
For each pair of genes located next to each other in the S. cerevisiae genome, we checked whether the orthologs in each of the ZT-KLE species were also contiguous. The same procedure was repeated in order to compare the ZT and KLE species.
For each syntenic block, the orthologs were obtained for each of the seven species in the ZT and KLE clades. MGR [43] was used to compute the number of rearrangements that occurred between each ZT/KLE species and S. cerevisiae.
|
10.1371/journal.pgen.1001064 | Ancient Protostome Origin of Chemosensory Ionotropic Glutamate Receptors and the Evolution of Insect Taste and Olfaction | Ionotropic glutamate receptors (iGluRs) are a highly conserved family of ligand-gated ion channels present in animals, plants, and bacteria, which are best characterized for their roles in synaptic communication in vertebrate nervous systems. A variant subfamily of iGluRs, the Ionotropic Receptors (IRs), was recently identified as a new class of olfactory receptors in the fruit fly, Drosophila melanogaster, hinting at a broader function of this ion channel family in detection of environmental, as well as intercellular, chemical signals. Here, we investigate the origin and evolution of IRs by comprehensive evolutionary genomics and in situ expression analysis. In marked contrast to the insect-specific Odorant Receptor family, we show that IRs are expressed in olfactory organs across Protostomia—a major branch of the animal kingdom that encompasses arthropods, nematodes, and molluscs—indicating that they represent an ancestral protostome chemosensory receptor family. Two subfamilies of IRs are distinguished: conserved “antennal IRs,” which likely define the first olfactory receptor family of insects, and species-specific “divergent IRs,” which are expressed in peripheral and internal gustatory neurons, implicating this family in taste and food assessment. Comparative analysis of drosophilid IRs reveals the selective forces that have shaped the repertoires in flies with distinct chemosensory preferences. Examination of IR gene structure and genomic distribution suggests both non-allelic homologous recombination and retroposition contributed to the expansion of this multigene family. Together, these findings lay a foundation for functional analysis of these receptors in both neurobiological and evolutionary studies. Furthermore, this work identifies novel targets for manipulating chemosensory-driven behaviours of agricultural pests and disease vectors.
| Ionotropic glutamate receptors (iGluRs) are a family of cell surface proteins best known for their role in allowing neurons to communicate with each other in the brain. We recently discovered a variant class of iGluRs in the fruit fly (Drosophila melanogaster), named Ionotropic Receptors (IRs), which function as olfactory receptors in its “nose,” prompting us to ask whether iGluR/IRs might have a more general function in detection of environmental chemicals. Here, we have identified families of IRs in olfactory and taste sensory organs throughout protostomes, one of the principal branches of animal life that includes snails, worms, crustaceans, and insects. Our findings suggest that this receptor family has an evolutionary ancient function in detecting odors and tastants in the external world. By comparing the repertoires of these chemosensory IRs among both closely- and distantly-related species, we have observed dynamic patterns of expansion and divergence of these receptor families in organisms occupying very different ecological niches. Notably, many of the receptors we have identified are in insects that are of significant harm to human health, such as the malaria mosquito. These proteins represent attractive targets for novel types of insect repellents to control the host-seeking behaviors of such pest species.
| Ionotropic glutamate receptors (iGluRs) are a conserved family of ligand-gated ion channels present in both eukaryotes and prokaryotes. By regulating cation flow across the plasma membrane in response to binding of extracellular glutamate and related ligands, iGluRs represent an important signalling mechanism by which cells modify their internal physiology in response to external chemical signals.
iGluRs have originated by combination of protein domains originally encoded by distinct genes (Figure 1A) [1]–[2]. An extracellular amino-terminal domain (ATD) is involved in assembly of iGluR subunits into heteromeric complexes [3]. This precedes the ligand-binding domain (LBD), whose two half-domains (S1 and S2) form a “Venus flytrap” structure that closes around glutamate and related agonists [4]. Separating S1 and S2 in the primary structure is the ion channel pore, formed by two transmembrane segments and a re-entrant pore loop [5]. S2 is followed by a third transmembrane domain of unknown function and a cytosolic carboxy-terminal tail.
Animal iGluRs have been best characterised for their essential roles in synaptic transmission as receptors for the excitatory neurotransmitter glutamate [1], [6]. Three pharmacologically and molecularly distinct subfamilies exist, named after their main agonist: α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA), kainate and N-methyl-D-aspartate (NMDA). AMPA receptors mediate the vast majority of fast excitatory synaptic transmission in the vertebrate brain, while Kainate receptors have a subtler modulatory role in this process. NMDA receptors require two agonists for activation, glutamate and glycine, and function in synaptic and neuronal plasticity. Representatives of these iGluR subfamilies have been identified across vertebrates [7], as well as invertebrates, such as the fruit fly Drosophila melanogaster, the nematode worm Caenorhabditis elegans and the sea slug Aplysia californica [8]–[10].
While most iGluRs have exquisitely tuned synaptic functions, identification of iGluR-related genes in prokaryotic and plant genomes provided initial indication of more diverse roles for this class of ion channel. A bacterial glutamate receptor, GluR0, was first characterised in the cyanobacterium, Synechocystis PCC6803 [11]. GluR0 conducts ions in response to binding of glutamate and other amino acids in vitro, suggesting a potential function in extracellular amino acid sensing in vivo. The flowering plant Arabidopsis thaliana has 20 iGluR-related genes, named GLRs [12]–[13]. Genetic analysis of one receptor, GLR3.3, has implicated it in mediating external amino acid-stimulated calcium increases in roots [14].
We recently described a family of iGluR-related proteins in D. melanogaster, named the Ionotropic Receptors (IRs) [15]. Several lines of evidence demonstrated that the IRs define a new family of olfactory receptors. First, the IR LBDs are highly divergent and lack one or more residues that directly contact the glutamate ligand in iGluRs. Second, several IRs are expressed in sensory neurons in the principal D. melanogaster olfactory organ, the antenna, that do not express members of the other D. melanogaster chemosensory receptor families, the Odorant Receptors (ORs) and Gustatory Receptors (GRs) [16]. Third, IR proteins localise to the ciliated endings of these sensory neurons and not to synapses [15]. Finally, mis-expression of an IR in an ectopic neuron is sufficient to confer novel odour-evoked neuronal responses, providing direct genetic evidence for a role in odour sensing [15].
The identification of the IRs as a novel family of olfactory receptors in D. melanogaster provides a potential link between the well-characterised signalling activity of iGluRs in glutamate neurotransmitter-evoked neuronal depolarisation and a potentially more ancient function of this family in environmental chemosensation. In this work, we have combined comparative genomics, molecular evolutionary analysis and expression studies to examine the evolution of the IRs. Four principal issues are addressed: first, when did olfactory IRs first appear? Are they a recent acquisition as environmental chemosensors in D. melanogaster, or do they have earlier origins in insect or deeper animal lineages? Second, what is the most recent common ancestor of IR genes? Do they derive from AMPA, Kainate or NMDA receptors, or do they represent a distinct subfamily that evolved from the ancestral animal iGluR? Third, what mechanisms underlie the expansion and diversification of this multigene family? Finally, do IRs function only as olfactory receptors or are they also involved in other sensory modalities? Through answers to these questions, we sought insights into IR evolution in the context of the origins of iGluRs, the appearance and evolution of other chemosensory receptor repertoires and the changing selective pressures during animal diversification and exploitation of new ecological niches.
iGluRs and IRs are characterised by the presence of a conserved ligand-gated ion channel domain (the combined Pfam domains PF10613 and PF00060 [17]) (Figure 1A). All iGluRs additionally contain an ATD (Pfam domain PF01094), which is discernible, but more divergent, in only two D. melanogaster IRs, IR8a and IR25a. Most IRs have only relatively short N-terminal regions preceding the LBD S1 domain (Figure 1A). To identify novel iGluR/IR-related genes, we therefore constructed a Hidden Markov Model (HMM) from an alignment of the conserved iGluR/IR C-terminal region, which is specific to this protein family. In combination with exhaustive BLAST searches, we used this HMM to screen raw genomic sequences and available annotated protein databases of 32 diverse eukaryotic species and 971 prokaryotic genomes (see Materials and Methods and Table S2 in Supporting Information). These screens identified all previously described eukaryotic iGluRs and all D. melanogaster IRs, as well as 23 prokaryotic iGluRs. Novel sequences were manually reannotated and classified by sequence similarity, phylogenetic analysis and domain structure as either non-NMDA (i.e. AMPA and Kainate) or NMDA subfamily iGluRs, or IRs (Figure 1B, Table S3, and Datasets S1 and S2). Like D. melanogaster IRs, newly annotated IRs have divergent LBDs that lack some or all known glutamate-interacting residues, supporting their distinct classification from iGluRs.
iGluRs are widespread in eukaryotes, present in all analysed Metazoa (except the sponge, Amphimedon queenslandica [18]) and Plantae, but absent in unicellular eukaryotes (Figure 1B, Table S3, and Datasets S1 and S2). Analysis of iGluR subfamilies on the eukaryotic phylogeny suggests that NMDA receptors may have appeared after non-NMDA receptors, as we identified them in Eumetazoa but not in the placozoan Trichoplax adhaerens. Further support for this conclusion will require additional genome sequences. One member of the Eumetazoa, the sea urchin Strongylocentrotus purpuratus, may have secondarily lost NMDA receptors. Different species contain distinct numbers of each iGluR subfamily: vertebrates, for example, have more NMDA receptor subunits than invertebrates.
Notably, IRs were identified throughout Protostomia, encompassing both Ecdysozoa (e.g. nematodes and arthropods) and Lophotrochozoa (e.g. molluscs and annelids) (Figure 1B, Table S3, and Datasets S1 and S2). There is substantial variation in the size of the IR repertoire, from three in C. elegans to eighty-five in the crustacean Daphnia pulex. Amongst insects, Diptera (i.e. flies and mosquitoes) generally had a larger number of IRs than other species. We did not identify IRs in Deuterostomia, Cnidaria or Placozoa.
To explore the evolutionary origin of the IRs, we examined phylogenetic relationships of the identified protostome IRs. Reciprocal best-hit analysis using D. melanogaster sequences as queries revealed that a subset of this species' IRs was conserved in several distant lineages, allowing us to define putative orthologous groups. These include one group containing representatives of all protostome species (IR25a), one represented by all arthropods (IR93a), nine by most or all insects, and three by dipteran insects (Figure 2A and 2B). For most orthologous groups, a single gene for each species was identified. In a few cases, for example the IR75 group, certain species were represented by several closely related in-paralogues, some of which appeared to be pseudogenes (Figure 2A and 2B, Table S3, and Datasets S1 and S2).
Consistent with its conservation in Protostomia, IR25a is the IR with the most similar primary sequence to iGluRs, suggesting that it is the IR gene most similar to the ancestral IR. Analysis of the phylogenetic relationship of IR25a and eukaryotic iGluRs locates it clearly together with the animal iGluR family, in the non-NMDA receptor clade (Figure 2C). To substantiate this conclusion, we asked whether the IR25a gene structure resembles more closely that of NMDA or non-NMDA receptors. Intron positions and numbers are extremely variable across IR25a orthologues, with multiple cases of intron loss, gain and putative intron sliding events by a few nucleotides (Figure 2D). Nevertheless, we identified eight intron positions that are conserved between at least subsets of IR25a orthologues and D. melanogaster non-NMDA receptor genes, some of which may represent intron positions present in a common ancestral gene. By contrast, only a single intron that was conserved in position (but not in phase) was identified between DmelIR25a (but not other IR25a orthologues) and DmelNMDAR1 (Figure 2D). A phylogram of intron positions in IR25a, non-NMDA and NMDA sequences reveals greater similarity of IR25a intron positions to those of non-NMDA receptors than NMDA receptors (Figure 2D). Together, these observations support a model in which IR25a evolved from a bilaterian non-NMDA receptor gene.
The conserved D. melanogaster IRs encompass the entire subset of its IR repertoire that is expressed in the antenna [15]. Moreover, evidence for antennal expression of the three additional genes, DmelIR41a, DmelIR60a and DmelIR68a, has been obtained by reverse transcription (RT)-PCR analysis, although we have not yet been able to corroborate this by RNA in situ hybridisation (data not shown). These combined phylogenetic and expression properties led us to designate this subfamily of receptors the “antennal IRs”.
We examined whether antennal expression of this subfamily of IRs is conserved outside D. melanogaster by performing a series of RT-PCR experiments on the honey bee, Apis mellifera, for all six putative antennal IR orthologues: IR8a, IR25a, IR68a, IR75u, IR76b and IR93a (see Materials and Methods for the nomenclature of newly-identified IRs). As in D. melanogaster, we could reproducibly amplify all of these bee genes from antennal RNA preparations but not in control brain RNA, except for AmelIR68a and AmelIR75u, which are also detected in the brain (Figure 2E). Thus, antennal expression of this subgroup of IRs is conserved across the 350 million years separating dipteran and hymenopteran insect orders [19], and therefore potentially in all insects.
To investigate whether IRs are likely to have an olfactory function beyond insects, we examined expression of the IR repertoire from a representative of a distantly related protostome lineage, Aplysia molluscs, whose last common ancestor with D. melanogaster probably existed 550–850 million years ago [20]. We first used RT-PCR to analyse the expression of the ten Aplysia IR genes in a variety of sensory, nervous and reproductive tissues (Figure 3A). Notably, the Aplysia IR25a orthologue is predominantly expressed in the olfactory organs, the rhinophore and oral tentacle [21]. Two other Aplysia-specific IR genes, IR214 and IR217, are expressed in the rhinophore and oral tentacle, respectively, and not detected in other tissues, except for the large hermaphroditic duct (IR214) and skin (IR217). Five additional IRs are also expressed in the oral tentacle, but displayed broader tissue expression in skin and the central nervous system; both of these tissues are likely to contain other types of chemosensory cells [22]–[23]. Expression of two IR genes, IR209 and IR213, was not detected in this analysis (data not shown).
To further characterise Aplysia IR25a, we analysed its spatial expression in the mature A. dactylomela rhinophore by RNA in situ hybridisation. An antisense probe for AdacIR25a labels a small number of cells in rhinophore cryosections. Their size and morphology is typical of neurons, although we lack an unambiguous neuronal marker to confirm this identification (Figure 3B–3D). These cells are found either singly or in small clusters adjacent or close to the sensory epithelial surface in the rhinophore groove, in a similar position to cells expressing other types of chemosensory receptors [21]. A control sense riboprobe showed no specific staining (Figure 3E). Together, these results are consistent with at least some of these molluscan IRs having a chemosensory function.
The expression of putative IR25a orthologues has previously been reported in two other Protostomia. An IR25a-related gene from the American lobster, Homarus americanus, named OET-07, is specifically expressed in mature olfactory sensory neurons [24]–[25]. In C. elegans, a promoter reporter of the IR25a orthologue, GLR-7, revealed expression in a number of pharyngeal neurons [9], which might have a role in food sensing [26]. While both crustacean and nematode genes were classified in these studies as iGluRs, there is no evidence that they act as canonical glutamate receptors, and we suggest that they fulfil instead a chemosensory function.
The antennal IR subfamily accounts for only a small fraction of the IR repertoire in most analysed insects and only 1–2 genes in other Protostomia. The remaining majority of IR sequences are - amongst the genomes currently available - largely species-specific, with low amino acid sequence identity (as little as 8.5%) with other IR genes in either the same or different species. We refer to this group of genes here as the “divergent IRs”. Dipteran insects have particularly large expansions of divergent IRs (Figure 1B). Phylogenetic analysis revealed no obvious orthologous relationships of these genes either between D. melanogaster and mosquitoes or amongst the three mosquito species (Aedes aegypti, Culex quinquefasciatus and Anopheles gambiae) (Figure 4). Instead, this subfamily of IRs displays a number of species-specific clades, perhaps reflective of the distinct ecological niches of these insects.
By contrast to antennal IRs, divergent IR expression has not been detected in D. melanogaster olfactory organs [15], leading us to test whether these genes are expressed in other types of chemosensory tissue. As endogenous transcripts of non-olfactory chemosensory genes, such as GRs, are difficult to detect [27]–[28], we employed a sensitive transgenic approach to investigate divergent IR expression. We transformed flies with constructs containing putative promoter regions for these genes upstream of the yeast transcription factor GAL4 and used these “driver” transgenes to induce expression of a GAL4-responsive UAS-mCD8:GFP fluorescent reporter [29]. We sampled divergent IRs from several distinct clades, including IR7a, IR11a, IR52b, IR56a and IR100a (Figure 4). All IR promoter-GAL4 constructs were inserted in the same genomic location using the phiC31 integrase system [30], eliminating transgene-specific position effects on expression resulting from their site of integration.
Expression of three of these divergent IR reporters was observed in highly selective populations of neurons in distinct gustatory organs (Figure 5A). In the adult, IR7a is expressed in at least eleven neurons in the labellum, a sense organ involved in peripheral taste detection (Figure 5B) [31]. Two reporters labelled neurons in internal sense organs in the pharynx: IR11a is expressed in one neuron in the ventral cibarial sense organ and IR100a is expressed in two neurons in the dorsal cibarial sense organ (Figure 5C and 5D). These internal pharyngeal neurons are thought to play a role in assessment of ingested food prior to entry into the main digestive system [16]. Expression was not detected in any other neurons or other cell types in the adult head (data not shown), although we cannot exclude expression in other regions of the body. IR52b and IR56a reporters were not detected in these experiments.
We also examined expression of these reporters at an earlier stage in the D. melanogaster life cycle, third instar larvae, which display robust gustatory responses [16]. The same three IR reporters were exclusively detected in unique bilaterally-symmetric larval gustatory organs: IR7a was expressed in two neurons in the terminal organ at the periphery, IR11a in a single neuron in the ventral pharyngeal sense organ and IR100a in two neurons in the posterior pharyngeal sense organ (Figure 5E–5H). Notably, all of these neurons in both adult and larval tissues (except for a single IR7a-expressing cell in the terminal organ) co-express IR25a, as revealed by a specific antibody against this receptor (Figure 5) [15]. IR25a is also expressed in several other cells in each of the gustatory organs, which may express other divergent IRs not examined here. Together these results support a role for divergent IRs as taste receptors in distinct taste organs and stages of the D. melanogaster life cycle.
To obtain more detailed insights into the processes underlying the expansion and diversification of IR repertoires, we investigated their evolution over a shorter timescale by comparative analysis of D. melanogaster with 11 additional sequenced drosophilid species [32]–[33]. The last common ancestor of these drosophilids is estimated to have existed 40 million years ago [34], by contrast to the ∼250 million years since the last common ancestor of D. melanogaster and the mosquito A. gambiae [35]. Certain species may have diverged much more recently, such as D. simulans and D. sechellia, whose last common ancestor may have existed only 250,000 years ago [36].
We used D. melanogaster sequences as queries in exhaustive BLAST searches of the drosophilid genomes. Retrieved sequences were manually reannotated to unify gene structure predictions across species and, in some cases, genes were partially resequenced to close sequence gaps or verify them as pseudogenes (see Materials and Methods, Table S3, and Datasets S1 and S2). Although predicted full-length gene sequences could be annotated for most genes, 28 sequences remain incomplete - but assumed in further analysis to be functional - because of a lack of sequence data or difficulty in precise annotation of exons in divergent regions of these genes. Of the 926 drosophilid sequences identified (including those of D. melanogaster), 49 genes were classified as pseudogenes because they consisted of only short gene fragments or contained frameshift mutations and/or premature stop codons. We clustered all genes into orthologous groups by examining their sequence similarity, phylogenetic relationships and, in the case of IR47a, IR47b, IR47c, IR56e and IR60f, their micro-syntenic relationships (Table S1 and Figure 6). For drosophilid species that are most distant from D. melanogaster, definition of precise orthologous relationships was not always possible, particularly for groups of closely related IR genes (e.g. IR52a–f, IR60b–f) (Table S1). Orthologous groups were named after their D. melanogaster representatives or a logical variant in groups where no D. melanogaster gene was identified (see Materials and Methods).
This analysis identified 14 iGluR and 58–69 IR genes in each of the twelve drosophilid species (Figure 6A and Table S1). iGluRs are highly conserved, with a mean amino acid sequence identity of 89±1% s.e.m., and a single representative for each species in every orthologous group. Antennal IRs are also well conserved (mean sequence identity = 76±2%) and amongst these genes we identified only a single pseudogenisation event, in D. sechellia IR75a, and a single gene duplication event, of D. mojavensis IR75d. By contrast, divergent IRs, though also largely classifiable into monophyletic groups, display a more dynamic pattern of evolution (mean sequence identity = 61±2%), with multiple cases of gene loss, pseudogenisation or duplication (Figure 6 and Table S1).
We reconciled the gene phylogeny with the drosophilid species phylogeny to estimate the number of IR gene gain and loss events. While this analysis is necessarily constrained by our ability to accurately define gene orthology, we estimated across the entire phylogeny there to be sixteen gene gain events (gene birth rate, B = 0.0006/gene/million years) and 76 gene loss events (gene death rate, D = 0.0030/gene/million years) (Figure 7A, see Materials and Methods). Most (46/76) gene losses are pseudogenisation events, which indicates that many of these events must have occurred relatively recently, as drosophilid species appear to eliminate pseudogenes rapidly from their genomes [37]–[38]. Notably, 13 gene loss events – 12 of which reflect the presence of just one or a small number of premature stop codons or frameshift mutations – occur on the branch leading to the specialist D. sechellia. Consequently, the gene loss rate on this branch is remarkably high compared with its generalist sister species D. simulans (Figure 7A and 7B).
We studied the selective forces acting on drosophilid iGluRs and IRs by calculating the ratio of nonsynonymous to synonymous nucleotide substitution rates (dN/dS, ω1) in these genes from all 12 species. All tested iGluR, antennal IR and divergent IR genes are evolving under strong purifying selection (ω1<<1) (Figure 7C, left and Table S4), suggesting that they all encode functional receptors. iGluRs have the lowest estimated dN/dS ratio (median ω1 = 0.060), consistent with a conserved role in synaptic communication. Antennal IRs have an intermediate dN/dS ratio (median ω1 = 0.107) and divergent IRs the highest (median ω1 = 0.149), suggesting that divergent IRs have evolved under weaker purifying selection and/or contain more sites that have been shaped by positive selection. Amongst the IRs, IR25a has the lowest dN/dS ratio (ω1 = 0.028), consistent with its high sequence conservation in and beyond drosophilids (Figure 2).
To compare these properties with those of other insect chemosensory receptor families (ORs and GRs) [39], we also calculated dN/dS ratios for IR genes from only the five sequenced species of the melanogaster subgroup (D. melanogaster, D. sechellia, D. simulans, D. erecta and D. yakuba). For this subset of sequences, the relative differences between median dN/dS ratios (ω2) for the iGluR and IR gene subfamilies observed with all twelve species was reproduced (Figure 7C, right). The GR gene family has previously been noted to evolve under weaker purifying selection than ORs [39]. Notably, we found that the median dN/dS ratios for antennal IRs (ω2 = 0.120) is statistically indistinguishable from that of ORs (ω2 = 0.137) (p>0.4, Wilcoxon rank-sum test), and that the median dN/dS ratio of divergent IRs (ω2 = 0.176) is statistically indistinguishable from that of GRs (ω2 = 0.217) (p>0.5, Wilcoxon rank-sum test). Thus, the selective forces acting on the IR receptor gene subfamilies parallel those on the ORs and GRs and appear to correlate with their putative distinct chemosensory functions in olfaction and gustation (Figure 7C, right). The reason for this difference is unknown, but might reflect reduced evolutionary constraints on co-expressed and partially redundant taste receptor genes or selection for higher diversity in taste receptor sequences to recognise more variable non-volatile chemosensory ligands in the environment.
Most residues of IR proteins can be expected to have evolved under purifying selection to maintain conserved structural and signalling properties, which may mask detection of positive selection (ω>1) at a small number of sites that contribute to their functional diversity. To obtain evidence for site-specific selection we applied site class models M7 and M8 in PAML to analyse 49 sets of orthologous IR genes of the six species of the melanogaster group. This test did not identify any sites significantly under positive selection after Bonferroni correction (Table S4), a result consistent with orthologous IR genes having the same function across drosophilids.
Site-specific positive selection may be more easily detectable in relatively recent IR gene duplicates potentially undergoing functional divergence. We therefore analysed the sole duplication of an antennal IR, IR75d.1 and IR75d.2 in D. mojavensis. Assuming an estimated divergence time of 35 My between D. virilis and D. mojavensis [40], and based on analysis of dS of IR75d genes in these species (see Materials and Methods), we estimated this duplication to have occurred relatively recently, approximately 2.6–5.1 My ago. Using a branch-site test we identified two sites (p<0.05) that have evolved under positive selective pressure, where DmojIR75d.1 and DmojIR75d.2 appear to contain the ancestral and derived residues, respectively: DmojIR75d.2-S670 maps to the third transmembrane domain and DmojIR75d.2-Q365 maps to the putative ligand binding domain. Functional characterisation of these variant receptors will be required to determine their significance.
From potentially one ancestral IR, what genetic processes underlay the generation of large repertoires of IR genes? We initially sought evidence for these mechanisms through analysis of the D. melanogaster IR family. Several monophyletic groups of IR genes exist in clusters in the genome suggesting an important role of gene duplication by non-allelic homologous recombination. For example, eight divergent IRs of the IR94 orthologous groups are located in three close, but separate, tandem arrays on chromosome arm 3R (Figure 8A). Other genes in the same clade are also found scattered on other chromosome arms (X, 2R, 3L) (Figure 6 and Figure 8A), indicating that interchromosomal translocation has also occurred frequently, most likely both during and after formation of the tandem arrays. Similar patterns are observed in the orthologous/paralogous sequences of these IRs in other drosophilid species (Figure 8A), as well as for other IR clades (data not shown). These features are also observed in IR repertoires in other insects, although incomplete genome assembly prevented a more precise analysis. For example, in Aedes aegypti the 23 IR7 clade members are found in arrays of 1, 1, 2, 5, 7 and 7 genes on 6 different supercontigs (data not shown).
We also noticed an unusual pattern in D. melanogaster IR gene structures, in which antennal IRs (as well as iGluRs) contain many (4–15) introns, while the vast majority of divergent IRs are single exon genes (Figure 8B). Drastic intron loss in multigene families is a hallmark of retroposition, where reverse-transcription of spliced mRNAs from parental, intron-containing genes and reinsertion of the resulting cDNA at a new genomic location may give rise to a functional, intronless retrogene [41]. The few introns that are present in these IRs in D. melanogaster have a highly biased distribution towards the 5′ end of the gene (19/25 introns in the first 50% of IR gene sequences) (Figure 8C), which is characteristic of recombination of partially reverse-transcribed cDNAs (a process which initiates at the 3′ end) with parental genes [42]. Sequence divergence of IRs prevented us from identifying parental gene-retrogene relationships. Nevertheless, these observations together suggest that divergent IRs arose by at least one, and possibly several, retroposition events of ancestral antennal IRs. Once “born”, single exon IRs could presumably readily further duplicate by non-allelic homologous recombination.
Our comprehensive survey and phylogenetic analysis of iGluR/IR-like genes permits development of a model for their evolution (Figure 9). The shared, unusual “S1-ion channel-S2” domain organisation of prokaryotic GluR0 and eukaryotic iGluRs is suggestive of a common ancestor of this family by fusion of genes encoding the separate domains that were present in very early life forms (Figure 9) [11]. However, we have found prokaryotic glutamate receptors in only a very small number of bacterial species. Thus, if an iGluR evolved in the common ancestor of prokaryotes and eukaryotes, it must have subsequently been lost in a large number of prokaryotic lineages. It is possible, therefore, that iGluRs only originated in eukaryotes and were acquired by certain prokaryotic species by horizontal gene transfer [43]. If the latter hypothesis is true, the presence of closely related iGluRs in both plants and animals implies their early evolution within eukaryotes, potentially in the last common eukaryotic ancestor [44]. However, the absence of iGluRs in sponges and all examined unicellular eukaryotes raises the alternative possibility that animal and plant receptors evolved independently, or were acquired by horizontal transmission, perhaps from prokaryotic sources. Whatever the precise origin of iGluRs in animals, their subsequent divergence into AMPA, Kainate and NMDA subfamilies also occurred early, although variation in the size of these subfamilies suggests continuous adaptation of the synaptic communication mechanisms they serve to nervous systems of vastly different complexities.
Several outstanding issues regarding IR evolution can now be addressed. First, we have shown that the IRs were very likely to have been present in the last common ancestor of Protostomia, an estimated 550–850 million years ago [20]. IR25a represents the probable oldest member of this repertoire and conservation of chemosensory organ expression of IR25a orthologues in molluscs, nematodes, crustaceans and insects strongly suggests that this receptor may have fulfilled a chemosensing function in the protostome ancestor.
Second, the apparent absence of IRs in Deuterostomia suggests the parsimonious model that IRs evolved from an animal iGluR ancestor rather than representing a family of chemosensing receptors that was present in a common ancestor of Animalia and lost in non-protostomes. Our phylogenetic and gene structure analysis suggests that IR25a may have derived from a non-NMDA receptor gene. The transition from an iGluR to an IR may not have involved drastic functional modifications: both receptor types localise to specialised distal membrane domains of neuronal dendrites (post-synaptic membranes and cilia, respectively) and, in response to binding of extracellular ligands, depolarise these domains by permitting transmembrane ion conduction which in turn induces action potentials [45]. Thus, it is conceivable that IRs arose simply by a change in expression of an iGluR from an interneuron (where it detected amino acid signals from a pre-synaptic partner) to a sensory neuron (where it could now detect chemical signals from the external environment).
Third, our analyses of IR repertoires across both divergent and relatively closely related species provide insights into the mechanistic basis for the expansion and functional diversification of the IR repertoire. Gene duplication by non-allelic homologous recombination is a widespread mechanism for growth of most multigene families in chemosensory systems [46], and this is also true for the IRs. Our implication of retroposition as a second mechanism in the evolution of IR repertoires offers two advantages for functional diversification. First, by arising from random re-insertion of reverse transcribed copies of parental genes, retrogenes normally lack endogenous promoter sequences, and can therefore potentially acquire novel expression patterns from genomic sequences flanking their insertion site that are distinct from their parental ancestor [41]. Indeed, in D. melanogaster, retrogene or retrogene-derived IRs - the divergent IRs - are apparently no longer expressed in antennal neurons like their ancestors, but instead in gustatory (and perhaps other) tissues. Second, release from the evolutionary constraints of the preservation of splicing signals near exon boundaries may have contributed to the more rapid divergence of the protein sequences of these intronless IRs [47].
Analysis of IR repertoires across the well-defined drosophilid phylogeny provides clear evidence for a birth-and-death model of evolution, in which, following gene duplication, individual family members progressively diverge in sequence and, in some cases, are lost by pseudogenisation and/or deletion [48]–[49]. Differential rates of these processes will ultimately shape the precise IR repertoire of an individual species (discussed below).
Our molecular evolutionary analysis has distinguished two subfamilies in the IR repertoire: conserved, antennal IRs and the species-specific, divergent IRs. Their distinct evolutionary properties may correspond to fundamental functional differences, as we provide here the first evidence, to our knowledge, for expression of divergent IR subfamily members in subsets of neurons in both peripheral and internal gustatory organs at both adult and larval stages of D. melanogaster. The selective and non-overlapping expression patterns observed in the small sample of IR genes examined indicate that a large fraction of the divergent IR repertoire may be expressed in gustatory neurons. It is also possible that some of these IRs may be expressed in non-chemosensory tissues. Although subsets of GR genes have been implicated in the detection of sweet or bitter compounds in peripheral taste bristles in D. melanogaster [31], a comprehensive understanding of the physiological breadth and molecular logic of taste detection is lacking. Our results introduce further complexity into the molecular mechanisms of taste detection and demand comprehensive and comparative expression and functional analysis of divergent IRs and GRs in this sensory system.
Although many gustatory-expressed divergent IRs in D. melanogaster are recently derived in drosophilids, the ancestral chemosensory function of IRs is likely to be not in the detection of airborne volatiles but rather water-soluble, non-volatile compounds, as the last common ancestor of Protostomia was probably aquatic. Indeed, the strikingly similar expression of IR genes in internal pharyngeal neurons in D. melanogaster and C. elegans suggests a conserved role for these receptors in sensing chemical signals from ingested food. In this light, the derivation of IRs from receptors detecting amino acid-related neurotransmitters invites the attractive hypothesis that ligands for these gustatory IRs (as well as species-specific IRs in other protostomes) are also amino acids. Almost nothing is known about sensory responses to this class of chemical signals in D. melanogaster, despite their vital importance for normal insect physiology and metabolism [50], but amino acids are chemosensory stimulants in other insects, lobsters and molluscs [51]–[53].
Our evolutionary and expression studies have highlighted IR25a as an atypical member of the repertoire, displaying deep conservation and broad expression in many olfactory and gustatory neurons. While we cannot exclude the possibility that IR25a recognises a specific chemical ligand, co-expression of this receptor with other cell-type specific IRs favours a model in which this acts as a co-receptor, analogous both to the heteromeric assembly of iGluR subunits into functional complexes [1], as well as to the pairing of ligand-specific ORs with the common OR83b co-receptor [54]–[55]. An insect- and antennal-specific homologue of IR25a, IR8a, may play a similar role specifically for olfactory IRs.
In addition to IR25a and IR8a, many other D. melanogaster antennal IRs are highly conserved in insects, both in sequence and expression pattern. These properties contrast starkly with the insect OR repertoires, which probably evolved only in terrestrial insects [56], and which contain only one member displaying orthology across multiple orders, the atypical OR83b co-receptor [57]. ORs are an expanded lineage of the ancestral GR repertoire whose evolutionary origins are unknown [56]. Homologues of GR genes exist in D. pulex and C. elegans [56], [58], but in the latter species these receptors may not be involved in chemosensation [59]–[60]. These observations suggest that, in insects, the IRs represent the first olfactory receptor family, whose members were fixed functionally early in their evolution to detect olfactory stimuli that are important for all species of this animal class. Consistent with this, the antenna of the mayfly Rhithrogena semicolorata – an insect belonging to the Paleoptera and not the Neoptera that encompasses all species described here – bears coeloconic sensilla (potentially housing IR-expressing neurons) but not trichoid or basiconic sensilla (which house OR-expressing neurons in all other insects examined) [61]. Available data on ligands for IR sensory neurons - and the role of specific IRs within these neurons - are limited, but include stimuli such as carboxylic acids, water and ammonia, which are known to be physiologically and behaviourally important in many insect species [62]. ORs, by contrast, may be primarily dedicated to detection of species-specific odour cues. In this light, the IRs are attractive molecular targets for novel, broad-spectrum chemical regulators of insect odour-driven behaviours, with applications in the control of disease vectors, such as mosquitoes, and agricultural pests.
Given the general conservation of the antennal IRs, what is the significance of the more recently evolved, species-specific variation in this family of chemosensory receptors? It is particularly informative to consider this question in the evolutionarily closely related drosophilid species. These display prominent differences in their global geographical distribution and chemosensory-driven behaviours [63]–[64], and include both generalists, which feed and breed on a wide range of substrates, and specialists, which have highly restricted ecological niches. The chemical ecology is best-understood for D. sechellia, a species endemic to the Seychelles that utilises the acid-rich fruit of Morinda citrifolia as its sole food source and oviposition site, a remarkable specialisation as this fruit is repulsive and toxic for other drosophilids [64]–[65]. Genetic hybrids between D. sechellia and D. simulans indicate that host specialisiation is due to loss-of-function mutations, rather than gain of new chemosensory perception abilities [65]. The accelerated rate of IR gene loss in D. sechellia compared to its sibling D. simulans (and other drosophilids) bears the hallmark of genetic adaptation of this chemosensory repertoire to the restricted host fruit. Notably, one of the D. sechellia pseudogenes is IR75a, an antennal IR expressed in a neuron responsive to several acids [62]. Thus, DsecIR75a represents an interesting gene whose mutation may be directly linked to host specialisation of this species. Future study of this receptor, and other species-specific IRs, may offer novel models to link genetic changes with phenotypic adaptation during animal evolution.
Finally, our results may shed light into the outstanding question of the evolutionary origin of animal olfactory systems. Common neuroanatomical features have long been appreciated in animal olfactory circuitry, notably glomeruli, which represent sites of synaptic connection of OSNs of identical molecular and physiological specificity with second order neurons [66]. Whether these represent homologous or analogous structures across phyla is unclear. Revelations of fundamental distinctions in the structure, function and regulation of mammalian and insect ORs support a theory of convergent evolution of the neuronal circuits in which these receptors act [67]–[68].
Our demonstration that most, if not all, insect olfactory systems comprise two molecularly distinct receptor families, the ORs and IRs, indicates that the evolution of receptor repertoires can be uncoupled from a presumed common origin of the OR and IR neuronal circuits within the insect ancestor. Thus, during a significantly greater timescale across animal phyla, profound molecular differences between olfactory receptor genes do not necessarily imply distinct evolutionary origins of the neuronal circuitry in which they are expressed. Our discovery of IRs in mollusc olfactory organs reveals this to be an interesting potential “hybrid” organism in olfactory system evolution. The A. californica rhinophore and oral tentacle also express a large family of GPCR-family candidate chemosensory receptors, belonging to the same Rhodopsin superfamily as vertebrate ORs [21]. The co-existence of both insect-like and vertebrate-like olfactory receptors in this species provides evidence for the occurrence of an evolutionary transition between these distinct olfactory receptor families. Thus, while extant animal olfactory systems display an enormous diversity in their receptor repertoires, there may remain - perhaps unexpectedly - a sufficient genetic trace within receptor gene families themselves to open the possibility of a common evolutionary origin of this sensory system.
IR genes were named according to a unified nomenclature system based upon a foundation of the cytologically derived D. melanogaster IR gene names [15]. Receptor names are preceded by a four-letter species abbreviation consisting of an uppercase initial letter of the genus name and three lower case initial letters of the species name (e.g. Anopheles gambiae = Agam; Daphnia pulex = Dpul). Orthologues of D. melanogaster sequences are given the same name (e.g. CquiIR25a, AcalIR25a). If multiple copies of an orthologue of a D. melanogaster gene exist for a species (based on sequence, not function), they are given the same name followed by a point and a number (e.g. ApisIR75d.1, ApisIR75d.2). If several in-paralogues exist both in D. melanogaster and other species, these are all given the same number (indicating their grouping within a common clade), but different final letterings. For novel, species-specific IRs, we defined new names numbering from 101 upwards to avoid confusion with D. melanogaster gene names, which number up to IR100a. For species-specific IRs that form monophyletic clades and had high (>60%) amino acid identity, we gave these the same name with an additional number suffix after a point (e.g. AaegIR75e.1, AaegIR75e.2). We did not rename genes with previously published names (e.g. C. elegans GLR-7 and GLR-8 [9]).
For vertebrate iGluRs, we used the NC-IUPHAR nomenclature [81]: each species name is followed by “Glu”, a letter representing the subtype of the receptor (K for Kainate, A for AMPA and N for NMDA), and a number, reflecting predicted orthology with mammalian iGluRs. We did not name (or rename) invertebrate iGluRs in this study, except for newly predicted gene sequences (Table S3), where logical variants of NC-IUPHAR nomenclature were assigned.
Genomic DNA was extracted from the sequenced drosophilid genome strains (obtained from the Drosophila Species Stock Center, University of California-San Diego) using a standard DNA extraction protocol. PCR primers were designed to amplify ∼500 bp regions covering putative nonsense or missense mutations or spanning gaps in the genome sequence (oligonucleotide sequences are listed in Table S5). PCR amplifications were performed using Taq DNA Polymerase (PEQLAB Biotechnologie GmbH) in a MasterCycler Gradient Thermocycler (Eppendorf) with the following programme: 95°C for 3 min, 35 cycles of (95°C for 30 sec, 55°C for 1 min, 72°C for 1 min) and 72°C for 10 min, with minor modifications of annealing temperature and elongation times for different primer pairs and amplicon sizes. Products were gel purified (Machery-Nagel) and sequenced with BigDye Terminator v3.1 according to the manufacturers' protocols.
Insects: total RNA was extracted from hand-dissected tissues of wildtype A. mellifera and D. melanogaster (w1118 strain) using the RNeasy Mini Kit (Qiagen), and reverse-transcribed using oligo-dT primers and the SuperScript III First-Strand Synthesis System (Invitrogen). Genomic DNA was extracted using standard procedures. Primers were designed to amplify short regions overlapping an intron, if possible at the 3′ end of the coding sequence (Table S5). PCR product amplification and purification were performed as described above and sequenced to verify their identity. Multiple independent cDNA preparations were analysed for each primer pair.
Primers were designed to amplify putative promoter regions from Oregon-R D. melanogaster genomic DNA with flanking restriction sites, extending from immediately upstream of the predicted start codon to the following 5′ extents: IR7a (2318 bp), IR11a (2099 bp), IR52b (446 bp), IR56a (2400 bp) and IR100a (512 bp) (Table S5). Gel purified PCR products were T:A cloned into pGEM-T Easy (Promega), end-sequenced, and sub-cloned into a pGAL4-attB vector, comprising the GAL4 ORF-hsp70-3′UTR in the pattB vector [30]. These constructs were integrated into the attP2 landing site [88], by standard transformation procedures (Genetic Services, Inc.). IR-GAL4 transgenic flies were double-balanced and crossed with flies bearing a UAS-mCD8:GFP transgene [89] to visualise driver expression.
|
10.1371/journal.pcbi.1003249 | Ligand Clouds around Protein Clouds: A Scenario of Ligand Binding with Intrinsically Disordered Proteins | Intrinsically disordered proteins (IDPs) were found to be widely associated with human diseases and may serve as potential drug design targets. However, drug design targeting IDPs is still in the very early stages. Progress in drug design is usually achieved using experimental screening; however, the structural disorder of IDPs makes it difficult to characterize their interaction with ligands using experiments alone. To better understand the structure of IDPs and their interactions with small molecule ligands, we performed extensive simulations on the c-Myc370–409 peptide and its binding to a reported small molecule inhibitor, ligand 10074-A4. We found that the conformational space of the apo c-Myc370–409 peptide was rather dispersed and that the conformations of the peptide were stabilized mainly by charge interactions and hydrogen bonds. Under the binding of the ligand, c-Myc370–409 remained disordered. The ligand was found to bind to c-Myc370–409 at different sites along the chain and behaved like a ‘ligand cloud’. In contrast to ligand binding to more rigid target proteins that usually results in a dominant bound structure, ligand binding to IDPs may better be described as ligand clouds around protein clouds. Nevertheless, the binding of the ligand and a non-ligand to the c-Myc370–409 target could be clearly distinguished. The present study provides insights that will help improve rational drug design that targets IDPs.
| Intrinsically disordered proteins (IDPs) exist as conformational ensembles that change rapidly. They are an important and common class of proteins in all kingdoms of life. IDPs are widely associated with human diseases and may serve as potential drug design targets. However, drug design targeting IDPs is difficult and only limited examples have been reported. One example is the oncoprotein, c-Myc, for which seven inhibitors were discovered by experimental screening. Understanding how small inhibitor molecules bind to c-Myc may help in understanding the binding mechanism of IDPs with ligands. In the present study, we conducted extensive molecular dynamics simulations to explore the binding mechanism for the c-Myc peptide with an inhibitor 10074-A4. We found that 10074-A4 could bind to c-Myc370–409 at different sites along the peptide chain and its binding behavior could be described as a ‘ligand cloud’. Even in the bound state, the structure of the c-Myc370–409 peptide remained a dynamic ensemble. Compared to c-Myc peptides that do not bind to 10074-A4, c-Myc370–409 binds selectively with 10074-A4, but the specificity of binding was not high. The interactions of IDPs with ligands can perhaps be described as a scenario in which ligand clouds around protein clouds.
| Intrinsically disordered proteins (IDPs), discovered in the 1990s, are proteins that lack a stable three-dimensional native structure under physiological conditions [1]–[5]. IDPs are sometimes described as “protein clouds” because of their structural flexibility and dynamic conformation ensemble [6]. Various bioinformatics methods have been developed to predict IDPs based on their sequences [7], [8]. It was revealed that IDPs are abundant in all kingdoms of life; for example, more than 40% of the proteins in eukaryotic cells possess disordered regions longer than 50 residues [9], [10]. Because of the flexibility of the chain and the resulting advantages in protein-protein interactions [1], [11], [12], IDPs play important roles in various critical physiological processes such as the regulation of transcription and translation [2], cellular signal transmission, protein phosphorylation and molecular assemblies [3], [13], [14]. On the other hand, IDPs also have some adverse effects. It was revealed that many IDPs are associated with human diseases such as cancer, cardiovascular disease, amyloidosis, neurodegenerative diseases, and diabetes [15]. It was also reported that the Swiss-Prot keywords for eleven severe diseases are strongly correlated with IDPs [16]. Given their abundance and their biological importance, IDPs are regarded as promising and potential drug targets [15], [17]–[19].
Compared with rational drug design targeting ordered proteins [20]–[22], drug design targeting IDPs is still in its infancy. Though some general strategies have been proposed [23], most of the studies [24]–[30] have been limited to only a few systems, namely, p53-MDM2, EWS-FLI1 and c-Myc-Max. Among them, the oncoprotein c-Myc is an encouraging example. C-Myc is a transcription factor with a basic helix-loop-helix leucine zipper (bHLHZip) domain which becomes active by forming a dimer with its partner protein Max [31]. In their unbound forms, both c-Myc and Max are disordered. However, in the dimerized forms, they undergo coupled folding and binding. In most cancers cells, c-Myc protein is expressed persistently by a mutated Myc gene, causing its unregulated expression in cell proliferation and signal transmission. Therefore, inhibiting either the overexpression of c-Myc and/or its dimerization with Max may provide a therapy for cancer. Yin et al. [30] have used high-throughput experimental screening to successfully identify seven compounds that inhibit dimerization between c-Myc and Max. Further biophysical studies using nuclear magnetic resonance (NMR), circular dichroism (CD) and fluorescence assays have verified three different binding sites (residues 366–375, 374–385, and 402–409) in the bHLHZip domain of c-Myc [28]. These binding sites contain several successive residues that can independently bind different small molecules [28]–[30]. It should be noted that, after binding with the small molecule inhibitors, the c-Myc sequence remains disordered, making the detailed experimental characterization of the molecular interactions almost impossible. Therefore, the inhibition mechanism is still unclear. For example, a recent study using drift-time ion mobility mass spectrometry suggested that the binding between c-Myc and these inhibitors is not as specific as previously thought [32]. The lack of conformation data also hampers the application of the well-developed structure-based drug design approach to optimize the inhibition.
Molecular simulations are useful in understanding the characteristics of IDPs because they can provide an atomic description of molecular interactions. Coarse-grained models [11], [33]–[35] and all-atom simulation [36]–[42] have both been used to investigate IDPs. Recently, Knott and Best [40] used large-scale replica exchange molecular dynamics (REMD) simulations with a well-parameterized force field to obtain a conformational ensemble of the nuclear coactivator binding domain of the transcriptional coactivator CBP. Their simulation results were in good agreement with NMR and small-angle X-ray scattering measurements, validating the efficacy of all-atom simulations in exploring the highly dynamic conformations of IDPs. For the c-Myc/inhibitor complex described above, Michel and Cuchillo [43] built a structural ensemble using all-atom simulations for c-Myc402–412 with and without an inhibitor (10058-F4) and found that 10058-F4 bound to multiple distinct binding sites and interacted with c-Myc402–412. However, because the c-Myc segment used in their simulation contained only the 11 residues that covered the binding sites of 10058-F4 (residues 402–409), it is unclear how the inhibitors would interact with longer segments of c-Myc and how specific the interaction would be.
In the present study, we conducted extensive all-atom molecular dynamic (MD) simulations to investigate the c-Myc370–409 conformational ensemble and its interactions with a small-molecule inhibitor (10074-A4). First, we performed implicit-solvent REMD simulations to clarify the conformational features of the unbound c-Myc370–409. Next, we performed MD simulations with an explicit water model to explore in detail the interactions between c-Myc370–409 and 10074-A4. Finally, a negative control using a different peptide segment (c-Myc410–437) was simulated to address the issue of interaction specificity. The conformational ensemble that we obtained will be useful not only in clarifying the structural features of c-Myc and the binding mechanism with inhibitors, but also in providing reference structures for drug design targeting c-Myc via structure-based approaches.
Conformational sampling of IDPs for molecular modeling is challenging because the energy landscapes of IDPs are relatively flat [44], [45]. In the present study, extensive REMD simulations using an implicit solvent model were performed to explore the conformational characteristics of c-Myc370–409. The accumulative total of simulation time reached 34.5 µs (see Methods). C-Myc370–409 is a 40-residue truncated construct of a full-length c-Myc. The conformational properties of c-Myc370–409 in its bound state (with 10074-A4) and more dynamic unbound state, have been studied experimentally using CD and NMR spectroscopy, and a likely average conformation was built based on chemical shift data which is not meant to (and cannot) define detailed structural features [28]. We compared our simulation results with the available experimental results.
To assess the sampling quality of the REMD simulations, we computed 1H and 13C chemical shifts from the simulated conformational ensemble using SHIFTS [46] and compared the computed values with the experiment values (Figure 1). The agreement is reasonable, though not excellent. Deviations between the average chemical shift values for a simulated ensemble and experimental values have been observed previously in several studies on IDPs [40], [47], [48]. The chemical shift calculation performed using several other software (SHIFTX [49], CamShift [50], SPARTA+ [51]) also showed deviations between the computed and experimental values (Figure S1). A possible reason is that chemical shifts are difficult to calculate accurately and the underlying parameterizations applied in current software for the calculation of chemical shifts have been optimized for ordered proteins but not for IDPs [47]. Interestingly, when we back-calculated chemical shifts from the NMR-refined structure using either the SHIFTS [46] or SHIFTX [49] software, the resulting values also deviated from the experimental ones (Figure S2). In addition, the ensemble nature of IDP conformations suggests that the chemical shifts of IDPs should be described as a distribution, and not merely as average values. The calculated distributions of the Hα chemical shifts obtained from our simulations are summarized in Figure 2. All the Hα chemical shifts are distributed over a broad range. The experimental values, indicated by arrows in Figure 2, are located close to the centers of the distributions, indicating the validity of the conformational sampling. Data for the HN, Cα and Cβ chemical shifts are given in Figure S3, showing similar behaviors as the Hα chemical shifts. We also computed the distribution of the backbone dihedral angles (Ramachandran (φ,ψ) plot) for the simulations and the dihedral angles of the NMR-refined apo structure lie well within the simulation distributions (Figure S4).
The secondary structure content of the simulated structures was also calculated [43], [52]–[54] and compared with that estimated from the experimental chemical shifts (Figure 3). The helix and polyproline II content of the simulated structure were consistent with the experimental structures (Figure 3A). However, the sheet content of the simulated structures was much lower than the sheet content of the experimental structures. In a previous study [43] on a shorter c-Myc segment, c-Myc402–412, a similar underestimation of sheet content was observed in the simulated structures. The deficiency of sheet content in the simulated structures might be caused by a bias in the force fields. Although c-Myc370–409 is intrinsically disordered, it possesses a high content of residual helical structure (>25%). The simulated helix propensity (Figure 3B) showed three helical regions separated by proline residues, Pro382 and Pro391.
To clarify the conformational features of c-Myc370–409, backbone-RMSD clustering with a cutoff of 2.0 Å of the conformations was performed. Representative structures (the central structure of each group) of the first eight groups were depicted in Figure 4. They are all somewhat collapsed compared to the fully extended structure and possess a rich residual helical structure. These states with considerable population will be useful references for rational drug design targeting c-Myc. The existence of residual structure may be related to the functional misfolding that prevents IDPs from unwanted interactions with non-native partners [55]. A quantitative analysis on the distributions of dimension and helix content was provided in Figure S5. The mean radius of gyration is around 10.3±0.6 Å, which is much smaller than the expected value of random coils (18.5 Å) under the same chain length. The mean helix content of the conformational ensemble is 27.7±11.1%, showing a broad distribution. These results indicated that c-Myc370–409 is disordered in nature and interconversions between dispersed structures occur.
To reveal how the conformations of apo c-Myc370–409 were stabilized, we analyzed the Lennard-Jones and electrostatic residue-residue interactions among all the residues (Figure S6). The Lennard-Jones interaction matrix was rather weak (Figure S6A), indicating that the conformations were disordered and that the packing in the collapsed structures was poor. This finding is consistent with the contact map, which showed that residue-residue contacts were dispersed and low in magnitude (Figure S6B). The electrostatic interactions, on the other hand, were comparatively strong (Figure S6C), probably because nearly one-third of the residues in c-Myc370–409 (12 out of the 40) are charged residues. The favorable electrostatic interactions of the Arg372, Arg378, Lys389, Lys392 and Lys398 residues with the Asp379, Glu383, Glu385 and Glu409 residues (Figure S6C) are the result of the electrostatic attraction between residues with opposite charges. Residues like Ser373 and Gln380 also contributed to the electrostatic interactions by forming hydrogen bonds (Figure S6D). Therefore, charge-pair interactions and hydrogen bonds were the main stabilized factors for the c-Myc370–409 conformations.
We conducted MD simulations with an explicit solvent model to investigate the interactions between c-Myc370–409 and the inhibitor 10074-A4. 10074-A4 is the only inhibitor (among seven inhibitors of c-Myc) that binds to the 375–385 sites in loop region of the bHLHZip domain of c-Myc and we wanted to see whether or not stable local structures were induced when 10074-A4 interacted with the flexible loop region. In the experimental study, 10074-A4 is a mixture of two chiral forms, the S and R forms (Figure 5). In the simulations, both chiral forms were tested. For comparison, the apo c-Myc370–409 was simulated with the same explicit solvent model. The accumulative simulation time for each group was 7 µs (see Methods). We calculated and compared the simulated chemical shifts with experimental chemical shifts for both implicit solvent REMD and explicit solvent simulations (Figures S7, S8, S9, S10). Reasonable agreements were found. For example, the average discrepancy between the simulated and experimental chemical shifts for Hα atoms of apo c-Myc370–409 is 0.14 and 0.16 in the MD simulations with explicit solvent model and REMD simulations, respectively (see Table S1).
The relative binding free energy of c-Myc370–409 with the two chiral 10074-A4 forms was analyzed from the MD trajectories using the Molecular Mechanic/Poisson-Boltzmann Surface Area (MM/PBSA) method [56]. The results of this analysis, together with the average non-bonded interactions Unon-bonded (Lennard-Jones and electrostatic potentials) between c-Myc370–409 and 10074-A4, are given in Table 1. We found that the interaction between c-Myc370–409 and the S form of 10074-A4 was much stronger than the interaction with the R form. The difference of Unon-bonded between the S and R forms (−3.7 kcal/mol) was close to the difference of ΔH from MM/PBSA (−3.2 kcal/mol). The difference of binding free energy between the S and R forms was −2.2 kcal/mol, resulting in a binding-affinity ratio of for the S and R forms. Therefore, compared with the binding of the S form to c-Myc370–409, the binding of the R form can be ignored. Thus, only the holo system with the S form of 10074-A4 is discussed further.
Hammoudeh et al. [28] reported an induced circular dichroism (ICD) effect on c-Myc370–409 by the binding of a racemate (1∶1 mixture of the S and R forms) of 10074-A4. There were two possible reasons for the observed ICD effect [28]; either the chiral surroundings affected the absorption transition of the compound, or the enantiomer-specific effect (the different binding affinity of the S and R forms) led to the ICD effect. We have shown above that the S form of 10074-A4 bound much stronger with c-Myc370–409 than the R form. Therefore, we suggest that it was the enantiomer-specific effect that was responsible for the observed ICD effect. Further experiments using single chiral forms of 10074-A4 would be helpful in clarifying this observation.
We clustered the conformations from MD simulations with the explicit solvent model for both the apo and holo c-Myc370–409 peptide based on RMSD of the backbone atoms. Figure 6 and 7 showed the representative conformations for the top eight clusters of the apo and holo peptides. It is clear that both the apo and the holo peptides have a rather broad conformation distribution, which is typical of disordered proteins. Upon binding to the ligand 10074-A4, the conformational distribution became more condensed. The top eight conformation clusters of the holo peptide were more highly populated compared to that of the apo peptide, with a total of about 77% occupancy compared to 50%. Similar to the apo c-Myc370–409 structure, the holo c-Myc370–409 structure is rich in helical structures. A quantitative analysis indicated that the helix and polyproline II content was almost unaffected by the binding of 10074-A4 (Figure S11), while the sheet content was enhanced (see also in Figure 7). The electrostatic interactions (from both charged residues and hydrogen bonding) dominated the intramolecular stabilizing force for holo c-Myc370–409 (Figure S12).
The residue-specific binding of c-Myc370–409 with 10074-A4 was tracked by calculating differences in the solvent accessible surface area (ΔSASA) between 10074-A4 and each residue of c-Myc370–409. The binding sites were determined as a function of time and representative conformations are shown in Figure 8. Binding of the 10074-A4 ligand was not restricted to a single site in c-Myc370–409, instead, it spread across almost the whole chain of c-Myc370–409. 10074-A4 usually binds simultaneously to two or more regions that are flanked by several residues. The binding was highly dynamic and could switch between different modes within a trajectory.
The time percentage of binding for each residue was calculated and is shown in Figure 9. Three binding sites were detected, which included site I (residues 372 to 384), site II (387 to 395), and site III (398 to 408). Site I was near the N-terminal and showed stronger potency than that of the other two sites. This result was supported by the intermolecular interaction analysis (Figure 10), which showed that both the electrostatic and Lennard-Jones interactions for site I were much stronger than those of the other two sites. In fact, in the latter cases, hydrogen bonds hardly formed and the electrostatic interactions were weak. Site I was similar to the experimentally determined binding site of 10074-A4 on c-Myc at residues 374–385 [28]. Binding at all the other sites generated in our simulations was much weaker, which would make them difficult to be observed experimentally. The low residue interaction specificity that we observed in the simulations is consistent with a recent simulation on an 11 residue peptide of c-Myc402–412 that suggested that ligand binding was driven by weak and nonspecific interactions [43]. The mass spectrometry experiment on c-Myc reported by Harvey et al. [32] also supported this conclusion.
To further investigate the inherent specificity features of IDPs, we conducted a negative control study in which we chose another segment of c-Myc (residues 410–437) that does not bind with 10074-A4 [28]. The simulated binding between c-Myc410–437 and 10074-A4 is shown in Figure 11. Unexpectedly, c-Myc410–437 “bound” with 10074-A4 in most simulation durations. Comparing with the binding of 10074-A4 with c-Myc370–409, its binding with c-Myc410–437 was less lasting and switched more frequently among different modes. The longest continuous binding time at one binding region within a trajectory is about 800 ns for c-Myc370–409 (see lower part of Figure 8), while it is about 200 ns for c-Myc410–437 (Figure 11).
The observed “binding” in the c-Myc410–437 negative control was different from what is found in negative controls for conventional ordered proteins where binding is usually not observed. To clarify the nature of this unexpected finding, we calculated the relative binding free energy using the MM/PBSA method and the results are provided in Table 1. We found that the binding of 10074-A4 with c-Myc410–437 was much weaker than with c-Myc370–409; the difference in binding free energy was about 3.4 kcal/mol. Therefore, the binding in c-Myc410–437 could not compete with that in c-Myc370–409. Although 10074-A4 scattered around the c-Myc370–409 and c-Myc410–437 peptides (Figures 8 and 11), its interaction with c-Myc370–409 was stronger and more selective than with c-Myc410–437. The sites at which 10074-A4 “bound” with the c-Myc410–437 peptide were much more disperse than the sites at which it bound with c-Myc370–409. Therefore, though the binding of 10074-A4 and c-Myc370–409 was not strong (the experimentally determined dissociate constant was 21±2 µM), it showed selectivity and thus specificity.
The specificities of IDPs in molecular recognition are complicated [57]. Our simulation results showed that the specificity of c-Myc in binding the small-molecule ligand 10074-A4 was not high. C-Myc is a typical example of IDPs. It is sticky and binds the ligands at different regions with different interaction strengths. Because of the lack of coupled folding and binding, after binding, c-Myc is still in an ensemble with diverse conformations and the distinct conformations are all capable of binding the ligand. Furthermore, for a given c-Myc structure, the binding of ligand occurred at disperse sites (Figure 12). We named this phenomenon ligand clouds. Ligand clouds are remarkably different from the type of binding that is found in ordered proteins where a dominant binding structure is formed. We expect that ligand clouds may be a general feature for IDPs binding with small-molecule ligands. For IDPs binding with macromolecule partners, it was reported that some IDPs remain disordered in the holo state [57]; for example, β-catenin/Tcf4, β-catenin/APC peptide, β-catenin/APC phosphorylated, Vif/EloB/EloC, and ERRγLBD/PGC-1α. These IDP complexes assume dynamic structures upon binding, suggesting that IDPs may interact with their partners in a similar manner to the ligand clouds. The ligand clouds concept supports the idea that there is no definite binding mode in the interactions between IDPs and small-molecule inhibitor [43]. It suggests that the interactions could be described as protein clouds interacting with ligand clouds.
The ligand cloud concept describes a scenario for the interactions between IDPs and small-molecule ligands and may provide a basis for drug design targeting IDPs. A straightforward strategy for rational drug design on IDPs is to extract metastable structures from simulations and then to conduct a virtual screen on them to identify potential inhibitors. A similar strategy was applied successfully in designing an inhibitor for Aβ fibrillation [58]. However, the ligand clouds concept for small molecules binding with IDPs implies that different strategies from those used for ordered proteins should be developed for better rational drug design on IDPs. For example, because ligand binding on IDPs occurs in disperse locations and in different orientations, multimode interactions should be considered in the scoring functions instead of the single-mode interaction that is commonly used for other proteins. Therefore, schemes that can consider binding energy landscapes [59] might be expected to perform better when designing small molecule ligands for IDPs. On the other hand, in contrast to the conventional ordered proteins that are in either “binding” or “non-binding” states with small molecules, IDPs are “sticky” and would be either in “strong binding” or “weak binding” with small molecules. So more cares should be paid to the problem of specificity in drug design targeting IDPs.
For conventional ordered proteins, the binding conformation is unique which could be selected from pre-existing conformations (the conformational selection mechanism) or be induced (the induced fit mechanism) by particular ligands. The scenario of ligand clouds around protein clouds for IDPs indicates that multiple protein conformations are selected and/or induced by the binding of a ligand on IDPs. This may extend the conformational selection-induced fit continuum in a new dimension.
In conclusion, we conducted extensive simulations to explore the conformational ensemble of c-Myc370–409 and its complex with a small-molecule inhibitor 10074-A4. The conformational space was found to be rather dispersed. In contrast to conventional structured proteins, the conformations of c-Myc370–409 were mainly stabilized by charge interactions and hydrogen bonds. Upon binding to 10074-A4, c-Myc370–409 remained disordered. The 10074-A4 ligand bound at different sites throughout the c-Myc370–409 chain with different strength. Accordingly, a ligand cloud concept was proposed, that is, the interactions between small molecule ligands and IDPs were like ligand clouds around protein clouds. The different binding probabilities between the protein clouds and ligand clouds indicated that the ligand could be selective and thus specific. Though the specificity of the binding was not high, the binding of ligand and non-ligand to the target IDP could be clearly distinguished.
Hammoudeh et al. [28] measured chemical shifts and several NOE signals of c-Myc370–409 and predicted dihedral angle distributions and atomic contacts. To build the c-Myc370–409 peptide, we first built a completely extended conformation with the following sequence: 370LKRSFFALRDQIPELENNEKAPKVVILKKATAYILSVQAE409 (Accession number: P01106). We then built the initial structures from the reported dihedral angles [28] using PyMOL [60]. The apo and holo structures for c-Myc370–409 were refined further using the GROMACS 4.5.4 software package [61] and the AMBER99SB force field, with the NMR data [28] as the dihedral angle and distance restraints in the simulation. Each initial structure was minimized in vacuum. Then, it was solvated, minimized, and equilibrated as described below. The time step was set to 0.5 fs. Finally, a 5 ns production simulation was performed and the final structure was adopted as the refined structure.
The conformations of the c-Myc370–409 peptide were sampled by REMD simulations with a Generalized Born/Surface Area (GB/SA) implicit solvent model. The AMBER molecular simulations package was used with AMBER99SB force fields [62]. A total of 30 replicas were adopted with temperatures ranging between 284.6 K and 608.8 K. All adjacent replicas attempted to exchange temperature every 10 ps with the average exchange rate between 35% and 40%. To produce the 30 starting conformations for an REMD simulation, an initial structure (described below) was minimized using steepest descent for 500 steps and then switched to conjugate gradient for another 500 steps. The minimized conformation was then heated to the defined temperature over a time of 200 ps for each replica. The obtained conformations were adopted as starting conformations in the REMD simulations, which were run with a time step of 2 fs. Replica temperature was controlled with a coupling time constant of 2 ps. Bonds involving hydrogen atoms were constrained with SHAKE. Chirality restraints on the backbone were employed to prevent non-physical chiralities. Ionic strength was set to 0.2 M. The cutoff for non-bonded interactions and for the GB pairwise summations involved in calculating Born radii was 999 Å to consider all probable interactions entirely. Snapshots from each trajectory were stored every 10 ps.
We conducted four groups of REMD simulations with different initial structures: (a) the extended structure of the peptide; (b) apo NMR refined structure; (c) the structure after a 80-ns MD simulation at 300 K starting from the extended conformation; and (d) the most occupied representative conformation generated previously from the REMD simulations of the extended structure in (a). The simulation time for the four groups of REMDs was 150 ns, 270 ns, 210 ns and 520 ns, respectively. The total simulation time was 34.5 µs (1.15 µs per replica).
The trajectories of 292.2 K, 300 K and 308 K were used in the further analyses except that only the trajectory of 300 K was used in the chemical shifts calculations.
To investigate the interactions between c-Myc370–409 and 10074-A4, MD simulations for the complex structure were carried out with an explicit solvent model [63]. The apo c-Myc370–409 was also simulated with the same explicit solvent model for comparison. Three groups of simulations were performed, one for the apo and two for the holo (with the two chiral 10074-A4 forms (see Figure 5)). Each group contained seven trajectories of 1 µs, therefore, the total simulation time was 21 µs. One of the seven initial structures was the NMR refined structures (apo and holo); the other six initial structures were adopted from representative conformations generated previously in the 150-ns REMD simulations (for the holo structures, the 10074-A4 isomers were docked using the AutoDock 4.2 program [64]).
MD simulations with the explicit solvent model were performed with the GROMACS 4.5.4 software package [61] and AMBER99SB force field under particle mesh Ewald periodic boundary conditions. The TIP4P-EW water model [63] was used with AMBER99SB force field because of its previously reported good performance in other simulations of IDPs [36], [47], [65]. In the holo simulations, the small molecule 10074-A4 ligand involved was parameterized using a general amber force field (gaff) with ACPYPE software [66]. An AM1-BCC charge model [67] was used to assign charges to the ligand.
Each initial structure was immersed in an explicit TIP4P-EW truncated octahedral water box. The dimensions of the box, defined as the distance between the farthest atoms of the peptide and the edge of the box, was set to 10 Å. The system was neutralized by adding ions, and extra NaCl was added to represent a solution with an ionic strength of 0.15 M. The system was minimized using the steepest descent minimization approach. After the minimization, the system was equilibrated in the NVT ensemble with all-heavy atom restrained with a force constant of 239 kcal/mol. The temperature was maintained at 300 K using a V-rescale thermostat with a coupling constant of 0.1 ps. Further equilibration was carried on in the NPT ensemble without strains, and where the pressure was maintained at 1 atmosphere using a Parrinello-Rahamn barostat with the coupling constant set to 2.0 ps. Both equilibrations were performed for 200 ps with a time step of 1 fs. For the production run, the thermostat and barostat settings were the same as for the NPT run. To enable 2 fs time steps, bonds involving hydrogen atoms were constrained to equilibration length using the LINCS algorithm [68]. A real-space cutoff of 10 Å was used for the electrostatic and Lennard-Jones forces. Snapshots from each trajectory were stored every 20 ps.
To further investigate the inherent specificity features of IDPs, we conducted a negative control study using the c-Myc410–437 truncated peptide (410EQKLISEEDLLRKRREQLKHKLEQLRNS437), which did not bind to 10074-A4. The extended structure of the peptide was used as the initial structure in an 80 ns implicit solvent MD simulation and the final structure that was generated was applied in all-atom explicit simulations. Two groups of simulations were performed for each of the two chiral 10074-A4 isomers. Each group contained one trajectory of 1 µs; the other parameters were the same as the parameters used for the holo c-Myc370–409 simulations described above.
All the simulations were analyzed using the GROMACS utilities [61] with either PyMol [60] or in-house scripts. ΔSASA was used in determinations of the binding sites. Upon small molecule binding, for each residue in the peptide there would be a clear decrease of SASA related to the difference between the SASA of the bound and unbound states. Backbone RMSD clustering of peptide conformations was performed to identify distinct structural clusters and to estimate their populations. The relative binding free energy was calculated every 200 ps using MM/PBSA [56] methods.
|
10.1371/journal.pntd.0006392 | Zika virus infection and microcephaly: Evidence regarding geospatial associations | Although the Zika virus (ZIKV) epidemic ceased to be a public health emergency by the end of 2016, studies to improve knowledge about this emerging disease are still needed, especially those investigating a causal relationship between ZIKV in pregnant women and microcephaly in neonates. However, there are still many challenges in describing the relationship between ZIKV and microcephaly. The few studies focusing on the epidemiological profile of ZIKV and its changes over time are largely limited to systematic reviews of case reports and dispersal mapping of ZIKV spread over time without quantitative methods to analyze patterns and their covariates. Since Brazil has been at the epicenter of the ZIKV epidemic, this study examines the geospatial association between ZIKV and microcephaly in Brazil.
Our study is categorized as a retrospective, ecological study based on secondary databases. Data were obtained from January to December 2016, from the following data sources: Brazilian System for Epidemiological Surveillance, Disease Notification System, System for Specialized Management Support, and Brazilian Institute of Geography and Statistics. Data were aggregated by municipality. Incidence rates were estimated per 100,000 inhabitants. Analyses consisted of mapping the aggregated incidence rates of ZIKV and microcephaly, followed by a Getis-Ord-Gi spatial cluster analysis and a Bivariate Local Moran’s I analysis.
The incidence of ZIKV cases is changing the virus’s spatial pattern, shifting from Brazil’s Northeast region to the Midwest and North regions. The number of municipalities in clusters of microcephaly incidence is also shifting from the Northeast region to the Midwest and North, after a time lag is considered. Our findings suggest an increase in microcephaly incidence in the Midwest and North regions, associated with high levels of ZIKV infection months before.
The greatest burden of microcephaly shifted from the Northeast to other Brazilian regions at the beginning of 2016. Brazil’s Midwest region experienced an increase in microcephaly incidence associated with ZIKV incidence. This finding highlights an association between an increase in ZIKV infection with a rise in microcephaly cases after approximately three months.
| The increasing evidence of a relationship between ZIKV in pregnant women and fetal congenital ZIKV syndrome with microcephaly has been reported in the literature over the last two years. Our findings suggest a spatial dependency between the diseases. Therefore, using the spatial pattern of ZIKV incidence to better understand risk areas for microcephaly may help the design of surveillance policies. Brazil had a large epidemic of ZIKV, leading to several important studies of the ZIKV outbreak and its association with microcephaly. This study used a geospatial analysis approach to examine the association between ZIKV and microcephaly in Brazilian regions. It was possible to highlight a spatial association between ZIKV and microcephaly considering a time lag between diseases. Brazilian regions with the highest incidences of microcephaly were the regions where the highest incidence of ZIKV occurred months before. This finding can help the organization and planning of health services to offer better screening actions dedicated to pregnant women in high-risk areas.
| On February 1, 2016, the World Health Organization (WHO) declared that the Zika virus (ZIKV) epidemic was an international public health emergency [1]. The increasing evidence of a causal relationship between ZIKV in pregnant women and an unpredicted rise in the incidence of microcephaly, later characterized as fetal congenital ZIKV syndrome [2–4], prompted this designation. Findings suggest that ZIKV affects neurogenesis during human brain development, leading to neurological syndromes as observed in Guillain-Barré or microcephaly [4]. As of the latest ZIKV status report in March 2017, 48 of 50 countries and territories in the Americas have confirmed autochthonous cases of ZIKV [5]. Half of these countries and territories (24) have confirmed cases of congenital ZIKV syndrome [5].
Brazil remains at the epicenter of the ZIKV epidemic with reports of 130,000 cases in 2016 [6]. In October 2015, early warning signs of a link between ZIKV in pregnant women and microcephaly in neonates surfaced when the number of infants born with microcephaly in the Northeastern state of Pernambuco rose [7]. From 2015 to 2016, 2,229 cases of microcephaly in infants were confirmed [8], over a 10-fold increase from the yearly average of 157 cases between 2000 and 2014 [9]. Consequently, a body of literature has emerged supporting a causal association between ZIKV infection during pregnancy and infant microcephaly [2,10–12]. Existing literature focused on the Brazilian ZIKV epidemic consists heavily of clinical management guidelines and longitudinal and case-control studies of mothers diagnosed with ZIKV and their infants to assess risks of congenital ZIKV syndrome [13,14].
Despite the ongoing research, challenges related to preventing ZIKV and its consequences, such as microcephaly, are still staggering. First, the committed countries have a limited epidemiological surveillance capacity. Second, the time delay between the onset of the ZIKV epidemic and the microcephaly reports means public policy is still defining the epidemic and not yet able to prevent its consequences. The rise in microcephaly incidence was documented only after infants were born, mostly due to limited ZIKV testing during intrapartum infection, leading to delays in timely epidemiologic and geographic surveillance of both diseases.
Using mapping techniques to study vector-borne disease epidemiology has proven crucial, as seen with previous research on dengue virus [15] and chikungunya [16]. These health geography studies can identify disease propagation patterns and high-risk areas, then model forecasts allowing inferences for the determinants of these outcomes [17]. To date, however, few studies have utilized geospatial techniques to investigate the ZIKV epidemic in Brazil, and the spatial-temporal association between ZIKV and microcephaly remains uninvestigated. The only available works [18,19] rely on systematic reviews of case reports or dispersal mapping of ZIKV spread over time without quantitative methods to analyze patterns and their covariates. By using a framework of health geography, we believe we can provide insights into disease spread patterns, high-risk areas, and forecast disease models that allow for inferences regarding the determinants of these outcomes [17].
This study examines the geospatial association between ZIKV and microcephaly January—December 2016. Specifically, we aim to 1) spatially represent diffusion patterns for both ZIKV and microcephaly incidence; 2) identify hot and cold spots of high and low incidence clusters for both diseases and any changes in their distribution across time; and 3) measure the spatial-temporal association between ZIKV and microcephaly spread. We hypothesize that areas with higher ZIKV incidence will be positively associated with an increase in microcephaly incidence after a time lapse of at least 16 weeks [9].
This ecological, retrospective study utilizes secondary data analysis of national health data systems during the ZIKV epidemic from January to December 2016 in Brazil. The largest country in Latin America in both size and population, Brazil spans approximately 3.2 million square miles with an estimated 190.7 million inhabitants [20]. An upper middle—income country and member of BRIC, Brazil ranks ninth among global economies [21] and has a high human development index level of 0.754 [22].
Brazil achieved universal health care coverage in the 1990s with the implementation of and reforms to the Unified Health System (SUS) [23]. Driven by national policies favoring decentralization and community-based models of health services delivery, the structure of the SUS is conducive to ecological studies of health outcomes [24]. The SUS maintains over 15 national-level health informatics and epidemiological databases to guide population health surveillance [24]. Data of infrastructure to outcome indicators are available, comprising information at individuals, municipalities or states levels. [6]. Marked inequality among Brazil’s regions, namely lower development levels and widespread poverty in the Northeast, results in disparities in health services coverage and population health indicators [25], which are significant when addressing diseases with natural and built environmental determinants (Fig 1). The availability of publicly accessible government databases at the national level, coupled with the socio-geographic landscape of the country and manifestations of the ZIKV epidemic, make Brazil an optimal setting in which to investigate the spatial-temporal association between ZIKV infection and microcephaly spread.
Data on confirmed cases of ZIKV were obtained from the Disease Notification System [26]. ZIKV infection was included in the Brazilian Ministry of Health compulsory notification disease list on February 17, 2016. From this date on, every health system unit in Brazil was obligated to report any confirmed or suspected case of ZIKV to the Ministry of Health [27]. ZIKVnotification is performed on a weekly basis, and deaths related to ZIKV must be reported within a maximum of 24 hours of death. Notification information is uploaded to the SINAN NET system (acronym in Portuguese—Disease Notification Information System) [28]. During our study, we included only confirmed cases of ZIKV. A suspected case was considered confirmed if one of the following characteristics was observed: positive of viral isolation test result, RNA viral detection by reaction of reverse transcriptase, or IgM serology. After confirmation of autochthonous circulation, the cases of Zika should be confirmed by clinical-epidemiological criteria. Despite that, suspected cases in pregnant women, neurological manifestations, and death still need to be confirmed using a serology test [27].
Data on confirmed cases of microcephaly were retrieved online from the System for Specialized Management Support [8]. Microcephaly was defined as an infant with 37 or more weeks of gestation with a head circumference equal to or less than 31.9 cm for male infants, or equal to or less than 31.5 cm for female infants, in concurrence with WHO standards [1]. For babies less than 37 weeks gestation at birth, the InterGrowth curve was used since the cephalic perimeter varies according to an infant’s gestational age [1]. Monthly case reports of microcephaly must be sent to the RESP-Microcephaly system (acronym in Portuguese—Register of Events in Public Health for Microcephaly). Additionally, population data were obtained from the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística) [20]. Data from these three sources were merged for all 5570 Brazilian municipalities. All secondary data extracted correspond to 2016.
Raw values of confirmed cases of ZIKV and microcephaly were used to compute incidence rates. Incidence rates were expressed continuously per 1,000 inhabitants for ZIKV and 100,000 for microcephaly, at the municipal level. We opted to use different scales because the prevalence of ZIKV and microcephaly in the general population occur on different scales. Ideally, we would have used better exposure controls such as pregnant women and newborns, such as the results reported by De Oliveira et al [29]. Thus, we decided to present the results in indices by population, which would give us a robust metric.
Data analyses were carried out in three steps. First, we conducted a descriptive analysis for the aggregated incidence rates of ZIKV and microcephaly for six 2-month time periods between January and December 2016 at the regional level. As such, the first bi-monthly period comprised January and February, the second period March and April, and so on. Next, we conducted a Getis-Ord-Gi [30] spatial cluster analysis to identify the presence of clustering according to incidence rates of both diseases throughout 2016. The Getis-Ord-Gi analysis produced two types of spatial clusters: hotspots with high values of incidence of both diseases, and coldspots highlighting low incidence areas. Lastly, a Bivariate Local Moran’s I analysis was carried out to evaluate the temporal-spatial association between ZIKV and microcephaly incidence rates over time [31].
The Bivariate Local Moran’s I is a statistic that evaluates the spatial correlation between two variables [31]. It verifies whether the value of the first variable in a reference municipality is related to the average value of the second variable in neighboring municipalities. Therefore, if the two variables are measured in different time periods, and a long enough time lapse is taken into account, this technique can provide insights regarding whether previous incidence of ZIKV infection in any reference municipality is associated with microcephaly cases in the neighboring region. This analytical strategy relies on the assumption that ZIKV has a causal role in microcephaly when pregnant women are infected [4,13]. Although there is not a consensus of the exact time in pregnancy that a ZIKV infection will cause microcephaly, there is a high volume of evidence supporting the association [14,32]. Thus, a time lag between ZIKV infection and microcephaly incidence can be approximated to the gestational period (in our case, 3 to 4 bi-monthly time periods) [9]. Therefore, considering the importance of a time lag between ZIKV and the emergence of microcephaly cases we opted to test multiple scenarios. For each scenario, a minimum difference of one bi-monthly period was considered. Incidence of ZIKV infection during the first and second bi-monthly periods were compared to microcephaly incidence rates of the third to sixth bi-monthly periods. This time arrangement was applied to all 2016 bi-monthly periods for both diseases, for all possible combinations that respect a minimum time lag of two bi-monthly periods. We chose this time lag period considering previous findings by the Centers for Disease Control [9].
The categorization provided by Bivariate Local Moran’s I technique can identify clusters based on ZIKV incidence considering, simultaneously, the microcephaly levels months later. In this scenario, a High-High cluster, for example, would represent a group of municipalities with elevated rates of microcephaly surrounded by municipalities with high values of ZIKV incidence a given number of months beforehand.
Incidence rate mapping and the Getis-Ord-Gi cluster analysis were performed in ARCGIS 10.3 [33]. The Bivariate Local Moran’s I analysis was conducted using the software GEODA [34].
From January—December 2016, Brazilian incidence rates of ZIKV per 100,000 inhabitants varied from 13.01 to 0.21. During the year, ZIKV incidence substantially decreased in all regions. Though this reduction was observed in all regions, it was more pronounced in the Midwest and Northeast regions. A high number of Midwest region municipalities showed incidence rates above 20 cases per 100,000 inhabitants during the first and second bi-monthly periods. At these time points, the Midwest contained the greatest number of confirmed ZIKV cases, with mean incidence rates of 82.06 for the first bi-monthly period and an annual average of 21.36 cases per 100,000 inhabitants. ZIKV incidence decreased in the following bi-monthly periods of the year. During the third bi-monthly period, the greatest mean ZIKV incidence rate was seen in the Northeast (7.56), which also had the greatest mean (2.81) for the fourth bi-monthly period. The fifth and sixth bi-monthly periods were marked by a continued reduction of high- and medium-incidence municipality clusters. By the fifth and sixth bi-monthly periods, mean ZIKV incidence rates had declined, with the highest for the fifth bi-monthly period in the Northeast (0.61) and the highest for the sixth bi-monthly period in Midwest (0.42).
For both the first and second bi-monthly periods, the Northeast had its highest mean microcephaly incidence rates of 0.66 and 0.71, respectively. During the third and fourth bi-monthly periods, the density of microcephaly incidence clusters in the Northeast diminished. During the fifth and sixth bi-monthly periods, the Northeast had the highest mean microcephaly incidence rates of 0.20 and 0.18, respectively, remaining the region most affected across the observed period (Fig 2).
The geospatial distribution becomes more diffuse over time, with scattered groups of municipalities with high incidence in the Midwest, North, Northeast, and Southeast regions. Patterns of microcephaly geospatial distribution, distinct from that of ZIKV infection, tended to be concentrated in the Northeast during the first, second, and third bi-monthly periods (Figs 3 and 4). The results of the cluster analysis (Getis-Ord-Gi) highlighted ZIKV hotspots in the Midwest, Northeast, and Southeast regions during the first two bi-monthly periods; hotspots then shifted to the Northeast for the third and fourth bi-monthly periods. The fifth and sixth bi-monthly periods are marked by persisting hotspots in the Northeast, in addition to appearance of hotspots in the North, and the reemergence of those in the Midwest. In contrast to the varied locations of ZIKV incidence hotspots, those for microcephaly incidence varied less between the regions across all bi-monthly periods. From the third until sixth bi-monthly period, hotspots also appeared in the Midwest and North. The South and Southeast regions also both consistently remained coldspots of confirmed microcephaly across all bi-monthly periods (Figs 3 and 4).
Bivariate Local Moran’s I analysis was performed, focusing on evidence of a possible spatial relation between the spread pattern of ZIKV and microcephaly (Fig 5). Considering the multiple time lag intervals adopted in the analysis, it was possible to identify an increasing wave in High ZIKV cluster areas becoming microcephaly High cluster areas across time. We are more interested in the High-Low/Low-High clusters in this representation. High-Low clusters (light red) represent areas with a high incidence of ZIKV surrounded by areas with low incidence of microcephaly, while Low-High areas are the inverse. From the results depicted in Fig 5, we noticed High-Low clusters mostly in the Midwest region. These clusters highlight regions with high incidence of ZIKV and low values of microcephaly, considering the different time lags observed. Analyzing the simultaneous presence of High-High clusters in the Midwest region, the High-Low clusters (light red) have the potential to become High-High clusters.
This finding highlights a relationship between an increase in ZIKV infection with a growth in microcephaly cases after two bi-monthly periods. Early in 2016, the ZIKV epidemic was already decreasing in the Northeast region, but was followed up with an increase in microcephaly (Low-High clusters). However, as we entered the 2016 epidemic year, the High clusters of ZIKV transitioned to the Midwest and North regions. The maps showing the association between the first bi-monthly period of ZIKV and the fifth and sixth bi-monthly periods of microcephaly demonstrate an increase of Low-High clusters in the Midwest and North regions, indicating that the microcephaly epidemic followed the distribution of the ZIKV infection. Additional analysis considering other bi-monthly periods as a starting point highlighted a similar growth pattern since a time lag of at least two months was observed (Fig 5).
From 2015–2016, Brazil experienced an unprecedented epidemic of microcephaly that carried devastating social and economic costs. A better understanding of the association between ZIKV and microcephaly was necessary to prevent a pandemic. Despite the importance and relevance of health geography, there is a lack of literature employing geospatial methods to analyze ZIKV and microcephaly. This study is the first to conduct a spatial-temporal evaluation of the association between ZIKV and microcephaly. Through this new approach, it was possible to identify evidence of an increase in microcephaly incidence associated with ZIKV incidence in the Midwest region of Brazil.
A potential link between ZIKV and microcephaly was first examined following reports of an abnormal rise in microcephaly incidence in Brazil’s Northeast region. This unexplained rise in microcephaly rates led public health authorities to begin epidemiological investigations. It was not until mid-2015 that suspicions regarding a link to ZIKV surfaced. This late identification of a possible cause carried implications for epidemiological surveillance. For most of 2015, there was no attention given to the causal link between ZIKV and microcephaly and, as a result, no reliable registry of ZIKV incidence rates was maintained. ZIKV was only designated as a disease of compulsory notification on February 17, 2016 [27]. Scientific evidence in support of the causal link later emerged in the beginning of 2016 [2,35,36]. In this context, it was not possible to analyze the spatial relationship between ZIKV and microcephaly in the early stages of the outbreak in the Northeast region. There were no data about ZIKV incidence before the 2016 microcephaly epidemic. After ZIKV was classified as a compulsory notification disease, more resources were invested toward more thorough and reliable reporting. Thus, data for all of 2016 is available at the municipal level. These improvements in disease surveillance facilitated research on the association between ZIKV and microcephaly spread patterns to identify those areas that are disproportionately affected and remain at an elevated risk. Our study contributes to these efforts and found a significant spatial pattern of association between both diseases.
The spread of ZIKV showed higher rates of infections in the Midwest in early 2016 that diminished by the end of the year. Confirmed ZIKV patterns in the Northeast and Southeast are consistent with a previous study of the spatial distribution of dengue fever in Brazil from 2014, one year before the Zika outbreak [37]. The Southeast, North, and Midwest regions experienced an increase in microcephaly incidence across 2016. This trend is explained by the high incidence of ZIKV previously in those regions. Our findings reveal that ZIKV incidence is positively associated with an increase in microcephaly incidence in the same location. Spread patterns of ZIKV and microcephaly cases in Brazil in 2016 suggest that a high number of cases of ZIKV in the Midwest are associated with a high number of cases of microcephaly in the region after a certain time lag. Measuring this association assists in probabilistic forecasting; monitoring the incidence of ZIKV may help predict where there will be increased incidence of microcephaly.
Our data support that the greatest burden of microcephaly could shift from the Northeast to other regions that reported a high volume of ZIKV during the beginning of 2016. A similar finding was reported by De Oliveira et al [29]. Our findings are of importance to health care providers and managers in these regions who should anticipate a greater need for prenatal care and adjust protocols in light of new systematic public health data about both diseases. The establishment of a regular monitoring system informed by the methodologies defined in the present study is needed to further confirm if observed relationships are maintained over time.
Best practices for pregnancy management during the ZIKV epidemic detail clinical manifestations of infection, endorse serological testing [38] depending on symptom presentation and timing of acute infection, and recommend routine ultrasounds before 24 weeks gestation [35]. Laboratory confirmation of ZIKV facilitates systematic efforts to estimate its prevalence and risk [39]. These practices need to be considered in the Midwest region for pregnancy management, as our findings suggest that this region will face an increase in microcephaly cases. Therefore, training actions for primary care professionals are recommended, as well as a revision of protocols related to pregnant women in areas at risk.
The transmission of ZIKV and other arbovirus diseases through genus Aedes mosquitos places every region with a tropical climate in a position of risk. The European Centre for Disease Prevention and Control mapped and categorized patterns of ZIKV transmission globally [40]. Several countries in South and Central America, Africa, and portions of Oceania were categorized by the World Health Organization as regions with active circulation of ZIKV. Tracking the relationship and behavior of ZIKV and microcephaly geographically is essential to design and implement response strategies to avoid outbreaks of microcephaly and other neurological complications [41].
The observed reduction in ZIKV incidence in the country as a whole, in fact, requires additional explanation. The incidence pattern of ZIKV followed the same tendency of dengue and chikungunya in 2016; there were peaks in incidence during the first months of the year followed by a decrease [8]. This trend might be due to the rainy season in Brazil that lasted from November until the end of March for most of the country. The increase in the rainfall index contributes to the growth of Aedes mosquito breeding sites, producing a rise in diseases transmitted through this vector. Brazil has continental dimensions with different climatic characteristics between its regions, as well as historical regional inequalities related to access to basic sanitation services. Simultaneous access to water supply by general network, sanitary sewage by general or rainwater network, and direct or indirect collection of garbage are still unequal among the five Brazilian regions. Municipalities lacking adequate sanitation are subject to a higher risk of infestation by Aedes and are consequently exposed to a higher risk of dengue, chikungunya, and ZIKV [42]. Even with this hypothesis, there is not yet longitudinal data on ZIKV to attribute the decrease in incidence rates to seasonal events like the rainy season. Additional hypotheses are being tested to explain the decline in ZIKV cases, including a massive infection perspective leading to a lack of susceptible individuals [43].
As limitations of the present study, we can highlight the lack of laboratory confirmation for part of the cases considered. However, they meet the epidemiological case criteria. Another limitation was the impossibility of estimating the incidence rates of ZIKV infection only in women due to the absence of the gender variable in the database. Thus, the estimates refer to the overall rates including men and women. The presence of gender information, as well as other details, such as pregnancy status or week of pregnancy, could increase the surveillance capabilities of the present information. Additionally, this information carries the potential to better support the relationship among ZIKV and microcephaly. However, as the current evidence supports sexual transmission [38,44–47] and salivary transmission [48], a high incidence in men increases the chances of infection in women. Thus, the overall incidence is a good estimator of the disease in the population. The fact that compulsory notification was only instituted in Brazilian health services in February 2016 may have led to an underreporting of both events (ZIKV and microcephaly), but especially of the first. The detection of abnormal levels of microcephaly cases was the trigger event responsible for raising additional investigations. Only after several months was a clear relation between microcephaly and ZIKV established. Thus, ZIKV was not on the surveillance radar of Brazilian epidemiological authorities when the microcephaly cases peaked. Therefore, there is no solid information about ZIKV incidence before the first rise in microcephaly cases, limiting the possibility of additional investigations regarding the first outbreak of microcephaly. As a consequence, there was potential bias towards the null hypothesis in association estimates, i.e., if all cases of ZIKV infection had been effectively reported, the associations found would have been even stronger.
Our study helps clarify the spatial association of microcephaly incidence in neonates whose mothers were previously infected with ZIKV. However, doubts remain about a possible relationship between the time of infection in pregnancy and the severity of sequelae in the fetus, or whether the symptoms of microcephaly depend on virus titers in fluids but not at the time of infection [49]. We don’t know if co-infections like dengue and chikungunya play any role in the severity of microcephaly [50]. Little is known [51,52] about the consequences of co-infection events [50,53]. What are the mechanisms used to break placenta barriers? What cells are involved in the pathogenesis of severe disease? [54] It is imperative to establish Aedes aegypti control in the Americas and the rest of the world to prevent the spread of ZIKV to new areas [55]. Understanding these and other issues may contribute to plans to control new outbreaks of this or other variations of the virus.
|
10.1371/journal.pcbi.1000712 | Tumor Growth Rate Determines the Timing of Optimal Chronomodulated Treatment Schedules | In host and cancer tissues, drug metabolism and susceptibility to drugs vary in a circadian (24 h) manner. In particular, the efficacy of a cell cycle specific (CCS) cytotoxic agent is affected by the daily modulation of cell cycle activity in the target tissues. Anti-cancer chronotherapy, in which treatments are administered at a particular time each day, aims at exploiting these biological rhythms to reduce toxicity and improve efficacy of the treatment. The circadian status, which is the timing of physiological and behavioral activity relative to daily environmental cues, largely determines the best timing of treatments. However, the influence of variations in tumor kinetics has not been considered in determining appropriate treatment schedules. We used a simple model for cell populations under chronomodulated treatment to identify which biological parameters are important for the successful design of a chronotherapy strategy. We show that the duration of the phase of the cell cycle targeted by the treatment and the cell proliferation rate are crucial in determining the best times to administer CCS drugs. Thus, optimal treatment times depend not only on the circadian status of the patient but also on the cell cycle kinetics of the tumor. Then, we developed a theoretical analysis of treatment outcome (TATO) to relate the circadian status and cell cycle kinetic parameters to the treatment outcomes. We show that the best and the worst CCS drug administration schedules are those with 24 h intervals, implying that 24 h chronomodulated treatments can be ineffective or even harmful if administered at wrong circadian times. We show that for certain tumors, administration times at intervals different from 24 h may reduce these risks without compromising overall efficacy.
| Chronotherapy of cancers aims at exploiting daily physiological rhythms to improve anti-cancer efficacy and tolerance to drugs by administering treatments at a specific time of the day. Recent clinical trials have shown that chronotherapy can be beneficial in improving quality of life and median life span in patients, but that it can also have negative effects if the timing is wrong. A theoretical basis for the rational development of individualized therapy schedules is still lacking. Here, we use a simple cell population model to show how biological rhythms and the cell cycle interact to modulate the response to cancer therapy. In particular, we show that the proliferation rate of cancer cells determines when treatments are most effective. We provide a simple formulation of the problem that can be used to compute an objective response function based on the drug sensitivity and the proliferation rate of tumor cells. Finally, we show that in some cases, treating at a different time every day may be more appropriate than standard daily chronotherapy. These results constitute an important step in designing individualized chronotherapy treatments, and point out to ways to design better clinical trials.
| Neurons located in the suprachiasmatic nuclei (SCN) of the hypothalamus form a dominant circadian pacemaker that controls timing of many physiological processes, including cell cycle. The pacemaker integrates environmental cues and communicates timing information to peripheral organs, which respond appropriately to optimize their functions [1]. In host and cancer tissues, drug metabolism and susceptibility to the drug vary throughout the day. The characterization of daily rhythms in drug toxicity and efficacy was a foundation for the chronotherapy of cancer [2].
The main aim of anti-cancer chronomodulated treatment is to achieve an optimal balance between chronotolerance and chronoefficacy (drug tolerance and efficacy as a function of time of administration). However, because many circadian-dependent factors influence the outcome of a treatment, determining the optimal schedule has been difficult to implement in clinics [3]. Cytotoxic chemotherapy suppresses the hematopoietic system, and neutropenia is a major limitation to the doses of drug that can be tolerated. Therapeutic advantages of chronomodulated treatments are seen mainly in the tolerance to higher drug doses, along with a decreased severity of side-effects, rather than in the prolonged survival of the patients [4],[5].
The efficacy of a cytotoxic drug, at a given concentration, is given by the product between the fraction of cells sensitive to the drug and the fraction of sensitive cells killed by the drug. For cell cycle phase specific (CCS) drugs used in chronotherapy, the fraction of sensitive cells is defined by their cell cycle status (e.g. fraction of cells in S or M phase) [6]. The entry to S phase is induced by c-MYC and cyclin D1, and the entry to M phase is gated (blocked) by WEE1 [7],[8]. Since those genes are controlled by the circadian clock, the cell cycle status is determined by the time of the day as well. Thus, drugs like cisplatin or 5-fluorouracil (5-FU) (S phase specific), docetaxel (M phase specific) and selicilib (G1 phase specific) would each be expected to have maximal efficacy and minimal toxicity at different times of the day.
Synchronization properties of the cell cycle to signals from the circadian pacemaker, namely phases and amplitudes, are tissue-specific. Blood cell progenitors [9], tongue epithelium [10], and cancer tissues [11] show tissue-specific daily variation in their DNA synthesis activity. In tumors, the response is perturbed and advanced-stage cancer cells can escape or even disrupt circadian control [12],[13]. Therefore, we would expect that the development of a cell cycle phase specific cancer chronotherapy strategy would depend on at least three circadian-dependent factors.
Here, we use a simple model of cell populations under circadian clock control and chronomodulated treatment to identify which biological parameters are important for the successful design of a chronotherapy strategy. We show that optimal CCS drug administration schedules, which minimize the sensitive fraction of the host cells and maximize the sensitive fraction of the tumor cells, are separated by 24 h intervals. However, if timing is wrong, a daily chronomodulated treatment schedule can lead to the worst therapeutic outcome as well. Using a theoretical analysis of treatment outcome (TATO), we show that clinically measurable cell cycle kinetics parameters are crucial in determining the response to CCS drugs. We show that chronomodulated treatments can be beneficial if tailored for individual patients, but can also be ineffective or even harmful if administered at wrong circadian times. We show that for fast growing tumors, administration times at intervals longer than 24 h may reduce these risks while maintaining a good overall efficacy.
Renewing tissues have daily peaks in the fraction of cells in S phase [9]–[11]. To explore the influence of daily modulations of cell cycle kinetics on cell proliferation, we used a simple cell population model [17]–[19] (Figure 1). The cell population is divided into four phases: G0/G1, S, G2 and M. G1 phase has a variable duration controlled by the transition rate and S, G2 and M phases have a fixed duration . The circadian clock controls the G1-S phase transition and the G2 phase duration: the G1-S phase transition rate and the G2 phase duration are 24 h periodic functions (see Methods for a more detailed description).
We simulated time courses over 48 h for cell populations with different cell cycle phenotypes: host cells, tumor cells with a short S phase duration (fast growing tumors), and tumor cells with a long S phase duration (slow growing tumors). Because G1 phase has a variable duration (represented by an exponential distribution of times with parameter ), cells tend to desynchronize when there are no synchronization factors present. Even when cells are initially synchronized, once the clock control is off (), the fractions in each phase of the cell cycle reach a steady state within a few division cycles (asynchronous cell growth). While the clock control is on (), all populations, irrespective of their cell cycle length, show a circadian variation in the fraction of cells G1, S, G2 and M phases (Figure 2). The fraction of host cells in S phase varies from 20% to 30%, and peaks around 12:00 every day (Figure 2A, solid line). The fractions of tumor cells in S phase vary between 15% and 30% for fast growing tumors and between 42% and 47% for slow growing tumors, and they peak at different times (Figure 2A, dashed and dashed-dotted lines respectively). The fractions of cells in G1 and G2/M phases also peak at different times of the day and their amplitudes are different for each phase (Figure 2B, C). These results indicate that the fractions in each cell cycle phase match the circadian period but the time at which they peak is influenced by the cell cycle status (tumor and host cells respond with different strength to the external cues).
S phase fractions in the host and tumor populations peak at different times, a feature that could be exploited by a well-timed administration of an S phase specific drug. We simulated the effect of one course of treatment based on a standard protocol (see Methods). We compared two tumor cell phenotypes: a fast growing tumor (Figure 3A,B) and a slow growing tumor (Figure 3C,D). Cell cycle kinetic parameters for the host and tumor cells were estimated from experimental data in patients when available; otherwise, data from mice were used. We assumed that the circadian clock acts at the same time of the day in the host and tumor cells, albeit more strongly on the host cells. To determine the optimal treatment time, we defined an outcome function that measures the trade-off between anti-tumor efficacy and toxicity. We calculated the outcome of treatments given at different circadian times. The optimal treatment time for the fast and slow growing tumors is during night. However, the worst times of treatments are different: 17:30 for the fast growing tumor and 5:00 for the slow growing tumor (Figure 3B,D). This shows that the S phase duration alone can strongly affect the outcome of a chronomodulated treatment.
The fraction of cells in each cell cycle phase determines how sensitive to treatment tissues are. Therefore, it would be useful to predict the best time of treatment based on kinetic data without having to run full simulations. We developed a theoretical method, TATO, to predict the influence of cell kinetics on CCS drug toxicity and efficacy. If the G1-S phase transition rate (due to circadian entrainment) and the surviving fraction (due to the treatment) are 24 h-periodic, we can solve the periodic treatment problem by calculating the average host and tumor population growth rates under 24 h period perturbations. The contribution of the rhythmic entrainment of the cell cycle to the growth rate can be approximated by(1)where the subscript denotes the tumor and , the host. (See Methods for a mathematical analysis). The value is the periodic component of the survival fraction of the cells that divide at time , when treated at time . The value is the periodic component of the G1-S transition rate at time . The integral, which is the average of the product between the two terms, is the net contribution of the periodic component to the rate of viable newborn cells over 24 h. As a function of , the sign of the integral determines the effect (positive or negative) of the clock and the treatment on the growth rate. We found that the integrals and are good approximations of the response values and computed by numerical simulation (Figure 4).
The functions and , as approximations of response functions and , are useful to study the dependence of the treatment outcomes on the cell cycle kinetic parameters. For drugs targeting the S phase, three cell cycle parameters affect the periodic part of the growth rate: (1) the duration of the S phase , (2) the timing of the peak of the G1-S phase transition rate , and (3) the timing of the cell death rate, given by the timing of the drug administration . These parameters appear, explicitly or implicitly, in Eq. 1. The extrema of Eq. 1, which represent the largest and the smallest growth rates of the cell population, can be located when and are known. As a first approximation, when the death rate and the G1-S phase transition rate are sinusoidal and are largest at times and , the location of the extrema can be calculated explicitly. The maximum of occurs whenand the minimum of occurs when(Figure 4A, white lines). Therefore, to kill the largest fraction of cells, i.e. to minimize , treatments should be applied halfway the S phase duration after the daily peak in G1-S phase transition. To spare the largest fraction of cells, the treatment should be applied 12 h later (detailed analysis in Methods). Based only on and , TATO predicts that the extrema of are 12 h apart. This approximation is good for durations between 7 h and 24 h (Figure 4B). When is larger than 24 h, the extrema are shifted by 12 h (Figure 4). When h, the timing of the treatment has no effect.
Anticancer drugs interfering with DNA synthesis (S phase) are widely used, but other phases of the cell cycle can be targeted as well. Therefore, in addition to the simulations for drug specific to S phase, we ran full model simulations for drugs acting on G1 or G2/M phase and compared the outcome to prediction from TATO (Table 1). The treatment protocol was the same as for the S phase drug, which is also included in Table 1. Optimal times of treatment in G1, S and G2/M phases vary by as much as 9 h between fast and slow growing tumors (formulas for optimal times are given in Methods). The worst times of treatment also show large differences between fast and slow growing tumors. Despite this, TATO predicts the optimal time within 2.5 h.
Taken together, these results indicate that TATO, using only a reduced set of kinetic parameters, can reliably predict the outcome of full simulations.
Previous computational studies have found that the fraction of cells killed with a constant drug infusion is higher (more toxic) than that killed with a chronomodulated infusion, for the same average killing rate [20]–[23]. Our model is consistent with these findings, and indicates that higher total doses of chronomodulated drug can be tolerated and are needed to achieve the same anti-tumor efficacy. These theoretical results are in agreement with clinical trials that showed consistent higher tolerance for chronomodulated compared to constant infusion [4], even when given at non-optimal times [24]. Lesser toxicity is independent from the circadian rhythms, i.e. chronomodulated treatments are less toxic even in absence of circadian rhythms . Thus, clinical and theoretical evidence shows that the shape of the infusion profile alone affects the treatment outcome significantly. For that reason, a direct comparison between constant and chronomodulated treatment is not really possible. Instead, we asked whether the same drug concentration profile administered at intervals different from 24 h could improve efficacy.
We simulated the chronomodulated administration protocol with intervals ranging from to h, starting on the first day at a time between 0:00 and 24:00. The total quantity and the infusion profile of the drug administered was the same for all intervals tested. Therefore, the resulting difference between outcomes depends only on the initial timing and the period . As expected, the largest amplitude of outcomes as a function of , and the best outcomes globally, are at intervals h (Figure 5A–D, solid lines). Likewise, the worst treatment outcomes also occur at intervals of 24 h.
To avoid the worst outcomes, it may be safer to seek treatment intervals that minimize outcome amplitudes, while optimizing the average outcome (maximizing ). When is close to 24 h, the treatment times can be averaged over the treatment course and TATO predicts an outcome given by(2)where is the number of drug administrations during one course of treatment, and is the phase of a 24 h interval treatment. If is larger than 24 h, the starting time of treatment needs to be advanced to produce an outcome equivalent to the one obtained at . Here, using , each hour increment in leads to a 2 h-advance in the starting treatment time. When is much different from 24 h, i.e. , , the average treatment phase is undefined, and TATO predicts an outcome independent from . In both fast and slow growing tumors, at these values h and h, the outcome depends little on . These two intervals offer circadian-independent treatment controls for the chronomodulated treatment (Figure 5B,D dashed and dotted lines). For a 24 h interval treatment to be safe to use, the time window during which the treatment is better than control should be large. TATO predicts that the outcome at h and h depends significantly on the duration of the sensitive phase (Eq. 24 in Methods). Treatment intervals longer than 24 h are predicted to spare the most host and slow growing tumor cells while shorter intervals are expected to spare the most fast tumor cells. Numerical simulations confirmed that the outcomes depend on the intervals in a way that is specific to the tumor. Fast growing tumors showed the best response at intervals h except for a small time window around midnight (Figure 5B), while the slow growing tumors showed a better response at h (Figure 5D).
Differences in the cell cycle lengths between the tumor and host cells could be exploited by adapting the interval between drug administrations [25],[26]. Cell cycle length effects were also observed in the model in the presence of the circadian clock. Overall, a long interval tended to improve anti-tumor efficacy in fast growing tumors, while a short interval was detrimental (Figure 5A). The opposite was observed for slow growing tumors, where shorter treatment intervals had a better outcome (Figure 5C). This indicates that the cell cycle kinetics interacts with the timing of the drug administration to modulate outcomes, even in the presence of a circadian clock.
Several randomized clinical trials have demonstrated significant improvements in tolerability and antitumor efficacy of chemotherapy with standardized chronomodulated administrations in comparison with a constant rate infusion of chemotherapy [27] or a chronomodulated delivery with an opposite timing [28],[29]. However, these studies did not show any survival benefit. In a recent large trial involving colorectal cancer patients, standardized chronotherapy achieved significantly better survival as compared to conventional treatment in men, but not in women [30]. This indicates that the response of patients to standardized chronotherapy can be heterogeneous, and that there is a need for tailoring delivery pattern to an individual patient or to subgroups of patients with distinct chronotherapeutic determinants.
These determinants are structured in different levels: whole body/systemic, target tissues, and cellular levels. A combination of these three factors contributes to the therapeutic advantage of chronomodulated delivery in an individual patient, and to the best delivery time. Systemic level includes the main behavioral and physiological characteristics like sleep/wake and eating patterns. The phase difference in peak expression of clock genes of each chronotype indicates that the optimal treatment time could vary at least by 2 h [15]. For example, the efficacy and toxicity of 5-FU are dependent on thymidylate synthase (TS) activity, its molecular target [31],[32], and dihydropyrimidine dehydrogenase activity (DPD), the enzyme responsible for the elimination of 5-FU [33]. Circadian rhythms in both TS and DPD activity have been detected [34],[35]. TS activity is higher during S, G2 and M phases, therefore the rhythms might be due to cell cycle synchronization [36],[37], or to direct circadian clock control. Also, circadian rhythms in DPD activity modulate 5-FU concentration during the day, regardless of whether 5-FU delivery is constant or chronomodulated.
In this study, we showed how cell cycle kinetics, i.e. cell cycle length and duration of the susceptible phase, can affect the timing of the optimal chronomodulated treatment. We used a mathematical model for normal cell and tumor growth under circadian regulation to investigate: (i) how we can use differences of cell cycle dynamics between host and tumor cells to establish an optimal treatment schedule, and (ii) how timing of the best and the worst treatment outcomes depends on individual chronotype and the growth rate of the tumor.
Optimization of treatment schedules based on cell cycle kinetics of target tissues has been explored before [26],[38]. These experimental and theoretical studies were based on the concept of resonance therapy, where treating at integer multiples of the cell cycle length leads to a reduction of killing of normal cells. This could be exploited in cancers where tumors cells have a cell cycle time distinct from normal cells, or where there is a large variability in tumor cell cycle times. It was noted, however, that heterogeneity in normal cell cycle times reduces the benefits of resonance therapy [25]. These alternative schedules have so far received little attention in the context of chronotherapy.
Recently, Altinok et al. [39] used a computational approach based on cellular automata to explore the effect of the variability in the cell cycle length on chronotolerance and chronoefficacy of 5-FU and oxaliplatin. Their model accounted for the observation that the toxicity profiles of 5-FU and oxaliplatin are antiphase, and showed how variability in cell cycle lengths reduces the benefits of chronomodulated treatments. Cell populations with cell cycle times just below 24 h are most likely to benefit from chronotherapy, a result that could be explained by a synergy between cell cycle times, circadian rhythms and periodic treatments.
We have developed an analytic method, TATO, that allows us to identify the optimal treatment time based on the circadian status and on the cell cycle kinetics of the host and tumor tissues. TATO measures the average differential growth rate of host and tumor cells that is caused by the circadian modulation of the cell cycle. Three parameters are essential to calculate the differential growth rate: the G1/S phase transition rate, the duration of the drug susceptibility phase, and the death rate. Our model indicates that the cell cycle length, which can vary from 18 h to over 100 h in colorectal cancers [40], is important to determine the best treatment times and intervals.
24 h interval treatments at the right time provided the best efficacy. Yet, the worse time of treatment can be as near as few hours from the optimal time [41], making it risky to treat at 24 h intervals. A previous study has found a significant correlation between S phase duration and 5-FU sensitivity [36]. Here we showed that for fast growing tumor (short S phase duration), administering a drug that targets the S phase of the cell cycle at 28.8 h intervals may be safer than treating at 24 h intervals. However, we found that for slow growing tumor (long S phase duration), treating at 24 h intervals was indeed the best option, even when deviating from the optimal time. So far, schedules different from 24 h have not been tested in the context of circadian chronotherapy, but in this paper, we show that for fast growing tumors they might be a safer strategy.
Drugs and the active drug metabolites used in chronotherapy are rapidly eliminated after delivery, which causes large modulations in their concentrations during the day. For that reason, patients with decreased 5-FU clearance rate due to a partial or complete loss of DPD activity might not benefit from chronomodulated treatments. An observed lower mean and amplitude of DPD activity in women is a possible explanation for the lower survival time with chronotherapy [5].
Here, we suggest how to individualize chronomodulated treatment schedules. First, patients with no overt circadian rhythm perturbations need to be selected, and their tumor kinetics assesed by measuring the S phase duration () and potential doubling time (). If the S phase duration of the tumor cells is short, a non-24 h schedule may be preferable. If the S phase duration of the tumor cells is long, a 24 h schedule could be more effective.
Second, the best treatment time could be determined using TATO. Constant infusion is not the best control for 24 h schedules since the shape of the infusion profile is likely to have a significant effect on outcomes [3]. Chronomodulated treatments with intervals spanning the whole day equally allows minimizing circadian effects, thus they could make suitable controls. Unlike for 24 h schedules, a constant infusion control group could be used to assess the efficacy of non 24 h interval treatments.
Third, once the optimal treatment time is determined, reverse pharmacokinetics could be used to retrieve the corresponding dose delivery schedule. Given a fixed dose delivered to a tissue at time , the fraction of surviving cells depends on the fraction of sensitive cells and the killing rate. If the killing rate varies in a predictable way during the day due to metabolism or elimination, it is possible to find a normalization dosage profile to make the killing rate time-independent. Thus, by knowing the quantity of drug needed to achieve a given killing rate, the fraction of surviving cells can be determined by the fraction of sensitive cells given by the model presented here.
The accepted administration time for 5-FU, 4:00, is based on the observation that in mice, the maximal tolerance is reached 5 h after light onset, corresponding to 5 h after beginning sleeping at 23:00 in humans [4]. In a recent study [28], 8 groups of patients received chronomodulated 5-FU-LV with peak times staggered every 3 h. Toxicity showed a marked circadian dependency of timing of chronomodulated 5-FU with leucovorin and oxaliplatin or carboplatin in cancer patients, with optimal time of 5-FU in cancer patients near 4:00 with 90% confidence limits. This study also showed more toxicity and large variability in women. Chronomodulated drug infusion differs in two respects from constant rate infusion: modulated concentration profile and timing. Chronotherapy is based on adapting the timing of treatment regimens to the circadian rhythms [27]. Thus, for the chronotherapy principle to work once the effect of concentration profile is discounted, there should be a 12 h time window during which the therapeutic outcome improves. This means that only 6 h would separate the optimal treatment time and a no-effect treatment time. We conclude that for chronotherapy clinical trials, patients need to be grouped according to the chronotype, tumor growth kinetics and pharmacokinetics/pharmacodynamics characteristics.
The cell population is divided into four phases: G0/G1, S, G2 and M. The G0/G1 phase includes cells that are actively dividing, but are in the pre-DNA synthesis or growth phase (G1) and cells that are quiescent but can be recruited to the cell cycle (G0). The S phase includes cells in DNA synthesis. The G2 and M phases include cells that have synthesized DNA and are progressing through mitosis. We used a population model of cell proliferation [17]–[19] in which we introduced a circadian control (Figure 1). Each stage of the cell cycle and its relationship to the circadian clock is modeled. The input to the model is a treatment course and the output is the population size in each cell cycle phase at any given time of the day.
We consider two cell types, host and tumor cells. Cell kinetic parameters for the host correspond to blood cell progenitors and for the tumor, to colorectal cancer cells. The model tracks the total cell number and fraction of cells in each phase for host and tumor during a course of chemotherapy, allowing estimates of efficacy and toxicity. The equations for the cell populations are(3)(4)(5)(6)
Each equation represents the balance between fluxes of cells (cells/hours) entering ( terms) and leaving ( terms) a cell cycle phase (see Figure 1 for details about the model). (Eq. 3) is the G0/G1 phase cell number, (Eq. 4) the S phase cell number, (Eq. 5) the G2 phase cell number, and (Eq. 6) is the M phase cell number. The total cell number is denoted .
The term , , is the fraction of cells surviving the cell cycle (S/G2/M phases) at time . It is the product of phase specific survival rates,(7)Time delays () account for the finite time required for cells to progress through each phase. The survival rates for the S, G2 and M phases are determined by integrating the phase-specific death rates over the duration of each phase,(8)where is one of , , . The duration is the total length of S, G2, and M phases of cell dividing at time ,(9)
The phase and amplitude of are given by and . Similarly, the phase and amplitude of are given by and ( and are relative to and ). A sinusoidal circadian input with a specific phase and amplitude is assumed for and ,(10)(11)where the circadian function is(12)The coefficient and phase-shift are set for all simulations to 0.2 and 14 h respectively. The function mimics the typical expression profile of circadian genes in many tissues, for a given individual. Note that circadian rhythm variability among individuals affect these parameters.
Kinetic parameters for bone marrow (host) and colorectal cancer (tumor) are derived from experimental data or were adjusted using this model. For the bone marrow [25], [25], [25],[40], , , h [25], h, h, [9], [8], h [9], h [8]. For the tumors, parameters are identical except , (fast), (slow), (fast), (slow), .
The population model is linear and simulations of host and tumor cell growth show that their cell numbers grow exponentially with a circadian modulation. Here we neglect nonlinear terms that would eventually cause the cell number to stabilize. We assume that with the treatment, the cell number is far from equilibrium. For a small-size tumor, this is a reasonable assumption. We also neglect the systemic feedback mechanisms of normal tissue homeostasis, which are more relevant to study between courses of chemotherapy when patients are recovering. Therefore, a linear model is also considered for the host tissues under cytotoxic stress.
We simulate a colorectal cancer treatment with 5-FU [42],[43]. 5-FU is an S phase specific drug that inhibits thymidylate synthase activity required for DNA synthesis, and consequently induces cell death. Chemotherapy schedules used clinically are either chronomodulated at 24 h intervals, or a constant infusion of 5-FU for a few consecutive days. The treatment is repeated every two to three weeks [4].
For simplicity, we simulate only one course of chemotherapy. We consider three different schedules: chronomodulated with 24 h intervals, flat infusion, and chronomodulated with intervals different from 24 h. One course of treatment lasts 5 days or 5 chronomodulated administrations. To isolate the effect of chronomodulation of treatment, we ignore the pharmacodynamics/pharmacokinetics aspects and we assume that chemotherapy acts on tumor and host cells in the same way. Because cytotoxic chemotherapy affects the hematopoietic system, and neutropenia is a major limitation to drug tolerance, we simulate the effect of 5-FU with blood cells as the host tissue.
The effect of 5-FU is simulated by adding a drug-induced death rate to the basal apoptosis rate of S-phase cells,(13)The chronomodulated drug-induced death rate, , takes the form of a truncated Gaussian function centered at circadian time , the treatment time (between 0 and 24 h),(14)Drug administration is repeated at intervals of hours. The duration of drug infusion is h [4]. The coefficient is the maximal drug-induced cell death rate. The equivalent flat rate infusion (normalized so that it kills the same fraction of cells than the chronomodulated infusion, in one day) is the constant(15)The normalization factor is .
For all simulations, the initial conditions were set to , , , and (total number initialized to ). With the parameters chosen, the relativepopulation is quickly synchronized by the circadian rhythm.
Numerical simulations were performed with the Volterra solver of the package XPPAUT. Analysis was done with Matlab 7.0. Codes (XPPAUT and Matlab) are available as supplementary text (Texts S1, S2, S3,S4).
The treatment outcome measure is defined as(16)where the functions and measure the cytotoxicity in tumor (C) and host (H) cells. The parameter is the circadian time of drug administration in case of a 24 h treatment interval. For non-24 h intervals, it is the time of administration on the first day of treatment. and , obtained from numerical simulations, are the normalized cell numbers 7 days after the first day of treatment , where is the total cell number as a function of . The outcome function E must increase with (high tolerance) and decreases with (high killing rate). For the flat infusions, E is constant. Close to zero, a Taylor expansion gives(17)The outcome measures the difference between responses and , and penalizes both excessive toxicity and poor anti-tumor efficacy. An optimal treatment maximizing tumor cell kill and minimizing host cell loss is found by maximizing the outcome function .
Equation 3 does not depend on other dynamical variables, so its stability analysis is simplified. Assuming a exponential growth, , where is a = 24 h-periodic function and is the growth rate, we have from Eq. 3,(18)Taking the average over a period, we obtain(19)For cell death occurring in the S, G2 or M phase, the death rate is chronomodulated. By making the simplifying assumption that the function , (20)The angle brackets denote the average over a period and the tildes the remaining, oscillatory part with a zero average. Thus, periodic parameters act only on through the integral term,(21)The integral can be either positive or negative, modulating the growth rate accordingly. As a consequence, the growth rate (tolerance) is maximal when the integral is maximal and the death rate (toxicity) maximal when the integral is minimal.
We consider and a drug specific to the S phase . Then,
The values and are shifted 12 h when 24 h. If the drug acts on the G2/M phases, with then
For cell death occurring in the G1 phase, the death rate is chronomodulated. We assume that is constant and therefore, the integral term becomes(22)If peaks at , meaning many cells in G1 are lost, the periodic solution will reach a minimum value at . Thus the ratio will have a maximum at and a minimum at . Assuming that peaks at and is minimum at ,
When treatment intervals are different from 24 h, the outcome will depend on the administration times over the whole course of treatment. If is the time of the -th administration, the effect on the growth rate isThe average effect of successive administrations at times , is(23)When , it is justified to replace the term with , whereTherefore, the outcomes will be equivalent when , with the phase of the 24 h interval treatment. The starting treatment time must then be(24)When , with , administration times are distributed equally around the circadian period and has little effect on the outcome. Neglecting the circadian clock allows computing the treatment intervals that minimize the growth rate of the equation , with a -periodic survival fraction if and 0 otherwise. This means that all cells in the sensitive phase are killed at intervals . The minimal growth rates occurs at values , since not a single cell would come out of the sensitive phase alive. The maximal growth rate occurs when and the fraction of cells in the sensitive phase is minimal. Let be the cell number in sensitive phase, given by . Right after administration, . The sensitive fraction reaches a minimum when . This occurs a time after the last administration, where(25) is the Lambert W function, and satisfies .
|
10.1371/journal.pbio.0050294 | Semantic Associations between Signs and Numerical Categories in the Prefrontal Cortex | The utilization of symbols such as words and numbers as mental tools endows humans with unrivalled cognitive flexibility. In the number domain, a fundamental first step for the acquisition of numerical symbols is the semantic association of signs with cardinalities. We explored the primitives of such a semantic mapping process by recording single-cell activity in the monkey prefrontal and parietal cortices, brain structures critically involved in numerical cognition. Monkeys were trained to associate visual shapes with varying numbers of items in a matching task. After this long-term learning process, we found that the responses of many prefrontal neurons to the visual shapes reflected the associated numerical value in a behaviorally relevant way. In contrast, such association neurons were rarely found in the parietal lobe. These findings suggest a cardinal role of the prefrontal cortex in establishing semantic associations between signs and abstract categories, a cognitive precursor that may ultimately give rise to symbolic thinking in linguistic humans.
| We use symbols, such as numbers, as mental tools for abstract and precise representations. Humans share with animals a language-independent system for representing numerical quantity, but number symbols are learned during childhood. A first step in the acquisition of number symbols constitutes an association of signs with specific numerical values of sets. To investigate the single-neuron mechanisms of semantic association, we simulated such a mapping process in rhesus monkeys by training them to associate the visual shapes of Arabic numerals with the numerosity of multiple-dot displays. We found that many individual neurons in the prefrontal cortex, but only a few in the posterior parietal cortex, responded in a tuned fashion to the same numerical values of dot sets and associated shapes. We called these neurons association neurons since they establish an associational link between shapes and numerical categories. The distribution of these association neurons across prefrontal and parietal areas resembles activation patterns in children and suggests a precursor of our symbol system in monkeys.
| Humans and animals share an evolutionarily old quantity representation system that allows the estimation of set size or number of events [1]. The assessment of numerical information is advantageous for the individual's fitness. This is particularly evident in social interactions (fight or flight decisions in contests) [2], foraging (exploiting the richer food source) [3], and parenting (discrimination of offspring) [4]. Quantity representations arise spontaneously without training as has been shown numerous times in monkeys [3] and human infants [5,6], supporting the idea that numerical competence is an ontogenetically and phylogenetically early faculty. Nonverbal numerical cognition, however, is limited to approximate quantity representations [1,7] and rudimentary arithmetic operations [5,6,8,9]; precise number representations and exact calculation are beyond its reach.
In contrast, humans familiar with number symbols are able to grasp exact cardinalities and to execute even the most abstract calculations. Humans learn to use number symbols as mental tools during childhood. Prior to the utilization of signs as numerical symbols [10], long-term associations between initially meaningless shapes (that become numerals) and inherently semantic numerical categories must inevitably be established [11,12]. Associations between shapes and quantities, a necessary first step towards the utilization of number symbols in linguistic humans, can even be mastered by animals [13–16].
Several studies in humans point to the prefrontal cortex (PFC) and the intraparietal sulcus (IPS) as key structures for both non-symbolic [17,18] and symbolic quantity information [18–20]. In monkeys, it has been shown that potentially homolog brain areas are involved in processing non-symbolic numerosity [21–26]. These studies support the hypothesis of a phylogenetic precursor system in monkeys on which higher, verbal-based numerical abilities in adult humans build up [27]. If the precursor hypothesis holds true, the same network that is involved in quantity estimation in nonhuman primates should also be engaged in the association of visual shapes with numerical categories. Here, we test this prediction by investigating whether single cells associate approximate numerosity representations with symbolic-like representations, and if so, what the respective contributions of the prefrontal and parietal cortices in this mapping process could be.
To that aim, we trained monkeys to assign visual shapes to numerical categories and recorded from single cells in both candidate regions. We report that many neurons in the PFC encoded the learned numerical value of a visual shape. In contrast, such association neurons were rarely found in the parietal lobe. Overall, the results suggest that the PFC is the prime source for the linking of signs to numerical categories in monkeys and may serve as a neuronal precursor for number symbol encoding.
We trained two rhesus monkeys in a delayed match-to-sample protocol to discriminate small numerosities (one to four) in multiple-dot patterns (Figure 1A; dot protocol). The monkeys had to judge whether two successive task periods (first sample, then test) separated by a 1-s delay showed the same numerosity. If so, the animals had to release a lever. In a second step, the monkeys learned over months to associate visual shapes (Arabic numerals) with the numerosity in multiple-dot displays, i.e., Arabic numeral 1 was associated with one dot, numeral 2 was associated with two dots, and so on (Figure 1B; shape protocol). Finally, both protocols were presented in a randomly alternating fashion within a given session.
We ensured that non-numerical parameters in the dot protocol could not be used by the monkeys to solve the task by varying and controlling low-level visual features. For each session, 100 different images per numerosity were generated with pseudo-randomly varied visual properties. Sample and test stimuli were never identical. All four quantities were presented in each session with one standard and one control condition. Different control conditions were applied day by day. Controls in the dot protocol included dot displays with constant circumference, linear configuration, and constant density across all presented quantities (see Figure 1C). To force the monkeys to generalize to the overall sign characteristics in the shape protocol, the numeral shapes were varied in size, position, and font. The font “Arial” was used for the standard condition; fonts “Times New Roman,” “Souvenir BT,” and “Lithograph Light,” were used in control conditions (see Figure 1D). The test stimulus for the shape protocol consisted of sets of black dots, equivalent to the dot protocol. Trials of the standard and control conditions as well as the dot and shape protocols were pseudo-randomly intermingled and appeared with equal probabilities in each session.
Both monkeys learned reliably to associate numerical values with the visual shape of numerals. Average performance in the dot protocol (Figure 2A and 2B) and the shape protocol (Figure 2C and 2D) was comparable (87% and 88%, respectively) and significantly better than chance for all tested quantities (p < 0.0001, binomial test). The numerical size and distance effect [22] could be observed in both protocols, irrespective of whether the standard or control condition was applied (see Figure S1). This suggests that the monkeys were indeed judging the direct and associated numerical values.
We recorded 692 randomly selected neurons from the lateral PFC of the monkeys while they performed the tasks. Intermingled presentation of both protocols during each session allowed us to investigate individual neurons' responses to both dot and shape protocols. Many neurons were selective to numerical category and discharged strongest to specific (direct or associated) numerical values, irrespective of the protocol. Neuron 1, in Figure 3A–3E, for example, showed a maximum response to numerosity two (the neuron's preferred numerosity) in the early sample phase, and a progressive drop-off with increasing numerical distance from the preferred numerosity in the dot protocol (Figure 3A). The same neuron preferred the same (associated) numerical value (i.e., two) in the shape protocol (Figure 3B) and had an equivalent tuning function (Figure 3C). Neuron 2, in Figure 3F–3J, preferred numerosity four in both the sample and delay phase in the dot protocol (Figure 3F). The same neuron exhibited a remarkably similar temporal discharge pattern to the signs associated to specific numerical values in the shape protocol (Figure 3G). For both protocols, the neuron showed monotonically increasing tuning functions (Figure 3H). Neuron 3, in Figure 3K–3O, showed strikingly similar responses during the memory period in both the dot (Figure 3K) and shape protocol (Figure 3L), with a preferred numerical value two; the tuning functions obtained with the dot and the shape protocols were almost identical (Figure 3M).
For a quantitative analysis of the neurons' selectivity to numerical values, we first calculated a two-way analysis of variance (ANOVA) (with factors numerical value [i.e., 1, 2, 3, 4] × stimulus condition [i.e., standard versus control], p < 0.05) separately for the dot and shape protocols. During the sample period, 263 (263/692, or 38%) neurons were selective for shapes and 229 (229/692, or 33%) for the number of dots irrespective of whether standard or control conditions were used (significance only for factor “numerical value”; no other significant effects). During the delay period, 297 (297/692, or 43%) and 300 (300/692, or 43%) neurons were significantly tuned to shapes and the number of dots, respectively. We found 210 neurons during the sample and/or delay phase that were selective only to factor “numerical value” in both protocols, irrespective of the displays' visuospatial properties. For all quantities from one to four, we found neurons with the same preferred numerosities and associated numerical values. The observed frequency of those neurons was significantly higher compared to chance occurrence (p < 0.001, binomial test; Figure S5; see Materials and Methods for a description of chance level calculation). More precisely, more neurons exhibited the same preferred numerical value in the dot and shape protocols than expected assuming independence between the encoding of the two stimulus protocols.
Neurons that were ANOVA-selective in both protocols (especially those with the identical preferred numerical value in both protocols) constitute a potential neural substrate for long-term numerical associations. In addition to mere selectivity in the dot and shape protocols, however, neurons should have similar tuning functions for the (direct and associated) numerical values in both protocols. To test this hypothesis, and to investigate the time course of association, we performed a sliding cross-correlation analysis between each neuron's tuning functions in the shape and dot protocols for all 210 ANOVA-selective cells and derived the cross-correlation coefficients (CCs; see Figures S2–S4 for details). The significance of the CCs was evaluated by using a sliding receiver operating characteristic (ROC) analysis. For each neuron, we derived the ROC values of the difference between CCs and the shuffle predictors (SPs, which constitute chance CCs) in 25-ms time steps [28] (see Materials and Methods). Based on this analysis, 157 cells (157/692, or 23%) were significantly correlated and classified as “association neurons.” For instance, neuron 1 associated between visual shapes and numerical values during the sample onset phase, i.e., 175 ms after stimulus onset and, taking its response latency of 120 ms into account, 55 ms after its earliest visual response (Figure 3D and 3E). The associative neuronal responses of neuron 2 (Figure 3I and 3J) ranged from 250 ms (latency-corrected: 11 ms) after stimulus onset to 50 ms before the end of the delay period. As an example of a late-associating cell, neuron 3 associated throughout the entire memory phase (see Figure 3N and 3O). The time course of association shown in Figure 4A for the entire sample of association neurons revealed many neurons that associated the numerical values of shapes and dots early after sample onset. While individual cells coded the (direct and associated) numerical values during specific time phases in the trial (represented by the black bars in Figure 4A), the neuronal population represented the numerical association throughout the entire trial. When corrected for response latency, about half of the association neurons started to associate numerical values within the first 200 ms after neuronal response onset. One hundred and thirteen neurons began to associate during the sample phase, and 44 neurons during the delay phase (Figure 4B).
Interestingly, the tuning functions of association neurons showed a distance effect [22] for both protocols, i.e., a drop-off of activation with increasing numerical distance from the preferred numerical value (numerical distance 1 versus 3, dot protocol, p < 0.001, n = 104; shape protocol, p < 0.01, n = 91; Wilcoxon signed-rank test, two-tailed; see single-cell examples in Figure 3C, 3H, and 3M, and population analysis in Figures 4C and S6 ). The distance effect found in the shape protocol indicates that association neurons responded as a function of numerical value rather than visual shape per se. However, the neuronal response drop-off between the preferred and second-preferred numerical values was larger in the shape protocol (50%) than in the dot protocol (39.8%) (p = 0.016, n = 157, Wilcoxon signed-rank test, two-tailed). This might indicate a more precise encoding of numerical values represented by signs than by sets of dots.
Is the association of numerical values by single PFC neurons really relevant for the monkeys' behavior? If association neurons constitute a neuronal correlate for the monkeys' ability to link signs with numerosities, the tuning correlations for both protocols should be weakened whenever the monkeys failed to associate visual shapes with their corresponding numerosities in error trials. To address this issue, we calculated the CCs of association neurons between correct trials in the dot protocol and error trials in the shape protocol. Because of the monkeys' low overall error rates, error trials were only available for a subset of numerical values (e.g., 2, 3, and 4) for many neurons. Only neurons recorded during errors to two or more numerical values were included into the error trial analysis. This criterion was fulfilled by 153 out of the 157 association neurons. As shown in Figure 5A and 5B, the correlation patterns for individual neurons were disturbed in error trials, and the mean population CCs were significantly decreased in error trials during and after cue presentation (p < 0.001, n = 153, Wilcoxon signed-rank test, two-tailed). As expected, baseline correlation during the fixation period was unaffected (p = 0.44). These findings strongly argue for association neurons as a neuronal substrate of the semantic mapping processes between signs and categories.
During PFC recordings, we simultaneously recorded from 437 neurons in the fundus of the IPS (see Figure 6) and analyzed the neurons' responses in the same manner (i.e., two-factor ANOVA and cross-correlation analysis). In the IPS, we found many neurons encoding either the visual shapes or the numbers of dots separately (67/437, or 15%, and 62/437, or 14%, respectively, during the sample period and 58/437, or 13%, and 83/437, or 19%, respectively, during the delay period; see Figure 6B for a summary of sample and delay). The proportion of neurons showing stimulus condition and/or interaction effects in the dot and shape protocols was significantly higher in the IPS (118/437, or 27%, and 107/437, or 24%) than in the PFC (119/692, or 17%, and 133/692, or 19%) (p < 0.001 and p < 0.05, respectively; Chi-square test). This argues for a more abstract encoding of numerical values in the PFC and a more sensory-driven activity in the IPS.
In contrast to the abundance of significantly tuned IPS neurons for the shape and dot protocols, only very few IPS neurons were selectively tuned to both protocols (n = 19); even fewer turned out to have significant correlations (8/437; Figure 6B). Compared to the PFC, for the IPS, the proportion of association cells from the pool of all selective cells was significantly lower (p < 0.001, Chi-square test; Figure 6D). Nevertheless, the proportion of neurons with identical preferred numerical values in both protocols was slightly higher than expected by chance (p < 0.001, binomial test) (see Figure S5 and Materials and Methods for calculation of chance level and the distribution of preferred numerical values).
Correlation time course and correlation strength (as measured by the ROC values) were fundamentally different between PFC and IPS neurons (Figure 7). In the PFC, ROC values showed a sharp increase right after sample onset and remained elevated throughout the entire trial (Figure 7A). In the IPS, however, neuronal association was weak and occurred much later during the trial (Figure 7B); ROC values showed an increase around the end of the sample and delay period, but in contrast to values for the PFC, the IPS values were low during both periods. In summary, only PFC neurons seemed to be crucially involved in associating shapes with numerical magnitudes.
We trained monkeys to associate quantitative categories with inherent meaning (i.e., numerosities) with a priori meaningless visual shapes. After this long-term learning process was completed, a large proportion of PFC neurons (23%) encoded plain numerical values, irrespective of whether they had been presented as a specific number of dots or as a visual shape. The activity of association neurons predicted the monkeys' judgment performance; if the monkeys failed to match the correct number of dots to the learned shapes, discharge patterns were drastically de-correlated. The population of these PFC cells represented the numerical association throughout the entire trial, providing crucial information to bridge the association over time. In contrast, only 2% of all recorded IPS neurons associated signs with numerosities. These findings suggest the PFC as the prime source in the mapping process of visual shapes to cardinalities.
Previous studies showed that neurons in the PFC encode learned associations between two purely sensory stimuli without intrinsic meaning (e.g., the association of a certain color with a specific sound, or pairs of pictures) [29–31]. In the anterior inferotemporal cortex, Miyashita and co-workers found “pair-coding neurons” that responded to arbitrary pairs of images monkeys learned to match in a pair-association task [32], and evidence that the PFC is important for active retrieval of these associative representations [33]. Here we show, to our knowledge for the first time, that neurons in the PFC represent semantic long-term associations not only between pairs of pictures, but between arbitrary shapes and systematically arranged categories with inherent meaning (i.e., the ordered cardinalities of sets). Our results suggest that the PFC may not only control the retrieval of long-term associations, but may in fact constitute a cardinal processing stage for abstract semantic associations. The prefrontal region is strategically situated for such associations [34]; it receives input from both the anterior inferotemporal cortex, which encodes shape information [35], and the posterior parietal cortex, which contains numerosity-selective neurons [23,24].
The described association neurons and their response characteristics suggest such cells as neuronal correlates of semantic association. We observed that many neurons associated visual shapes with numerical values transiently, and not until the end of the delay period (Figure 4A), whereas prospective activity typically dominates near the end of the delay [29]. More importantly, a high proportion of neurons associated numerical values in the shape and dot displays right after sample onset (see Figure 4A and 4B). This argues for a direct involvement of these neurons in linking numerical values to shapes, rather than encoding upcoming match stimuli in a prospective manner. Finally, an analysis of error trials (see Figure 5) revealed that tuning correlation between both protocols was weakened whenever the monkeys failed to associate visual shapes with their corresponding numerosities. This again provides evidence that association neurons constitute a neuronal correlate for the monkeys' ability to link signs with numerosities.
While quantity representations are spontaneously developed [3,6], associations between visual shapes and numerical categories clearly have to be learned by mapping shape representations onto numerical categories. This neuronal learning could start with two classes of PFC cells: one class encoding visual characteristics of shapes (input possibly via inferotemporal cortex [35]), the other class representing numerical information most likely received from the IPS [23,24]. According to the Hebbian learning rule [36], the connections may be strengthened between these two classes of neurons so that cells encoding matching pairs (e.g., Arabic numeral 3 and three dots) are interconnected and become associative. This learning behavior could potentially be modeled via a recurrent neuronal network as has been done for pair-association encoding in inferotemporal neurons [37] or for somatosensory parametric working memory in PFC [38].
Even though numerosity-selective neurons in IPS are relatively abundant and encode numerical information earlier than PFC neurons [23], association neurons were surprisingly rare in the parietal lobe. Moreover, IPS neurons differentiated to a larger extent between the sensory features of the visual displays; they responded less abstractly than PFC neurons, which generalized across visual properties. At first glance, the sparseness of association IPS neurons in the nonhuman primate seems to be at odds with the well-known role of the posterior parietal cortex in adult humans for both non-symbolic [17,18] and symbolic numerical cognition [18–20]. Beyond possible species-specific differences between humans and monkeys, this difference might also be the consequence of training duration; our monkeys were trained for few months to match numerosities with visual shapes, whereas humans acquire symbols over years. Because of the monkeys' inferior proficiency, it is likely that the shape–numerosity association was not automatically executed in the monkey brain, but required a strong involvement of the PFC in order to manage the high cognitive demands [34].
Support for this assumption comes from recent functional magnetic resonance imaging studies with human children. In contrast to adults, preschoolers lacking proficiency with number symbols show elevated PFC activity when dealing with symbolic cardinalities [39–41]. Only with age and proficiency does the activation seem to shift to parietal areas. This frontal-to-parietal shift has been interpreted as being a result of increasing automaticity in number tasks. This shift of symbolic associations to the parietal lobe could release the limited cognitive resources of PFC for new demanding tasks [34]. The PFC could, thus, be ontogenetically and phylogenetically the first cortical area establishing semantic associations, which might be relocated to the parietal cortex in human adolescents [27,42] in parallel with the maturing language capabilities [43] that endow our species with a sophisticated symbolic system [42].
During cultural evolution, humans invented number symbols as mental tools. Number symbols endow our species with an exact understanding of cardinality and the ability to execute the most complicated calculations. Given that the first ancient number symbols have been dated back to only a couple of thousand years ago [44], it is impossible that the human brain has developed areas with distinct, culturally dependent number symbol functions [27]. It is more parsimonious to assume that existing brain structures, originally evolved for other purposes, are reused and built upon in the course of continuing evolutionary development (by a process called “exaptation” [45]), an idea captured by the “redeployment hypothesis” [46] (also termed “recycling hypothesis” [27]). According to this hypothesis, already existing simpler networks are largely preserved, extended, and combined as networks become more complex, instead of there being a de novo creation of intricate structures [47]. In the number domain, evidence suggests that existing neuronal components (located in PFC and IPS)—originally developed to serve nonverbal quantity representations—are used for the new purpose of number symbol encoding, without disrupting their participation in existing cognitive processes [18]. While monkeys use the PFC and IPS for non-symbolic quantity representations [23], only the prefrontal part of this network is engaged in semantic shape–number associations. Interestingly, this pattern of brain area use seems to be preserved in human children [39–41]. Moreover, we found that numerical values represented by signs were encoded more selectively as than analog set sizes. This sharpening of the tuning functions for signs was predicted by a recent network model [48] and might indicate the advent of a digital representation via symbol-like signs in the primate brain. We speculate that our data in the monkey provide a first glimpse of redeployment of the PFC for symbolic-like learning, thus paving the way for the neuronal quantity network to encode real number symbols in language-endowed humans.
We trained two monkeys to match either a set of dots with another set of dots (delayed match-to-sample task, or dot protocol; see Figure 1A) or a visual shape with a set of dots (delayed association task, or shape protocol; see Figure 1B). Stimuli were sets of black dots or black Arabic numerals pseudo-randomly varying in size and position and displayed on a gray background. A trial started when the monkey grasped a lever and fixated (± 1.75° of visual angle, monitored with an infrared eye tracking system) on a central target. After a monkey fixated for 500 ms, the sample appeared for 800 ms (multiple-dot display in the dot protocol; Arabic numeral in the shape protocol). The monkey then had to maintain fixation until the end of a 1,000-ms delay period, after which the test stimulus was presented (always a multiple-dot pattern). In 50% of cases the test stimulus was a match, i.e., it showed the same number of dots as cued during the sample period by a multiple-dot pattern or a shape. In the other 50% of cases the first test stimulus was a nonmatch, which showed—with equal probabilities—either a higher or lower numerosity than the sample display. After a nonmatch test stimulus, a second test stimulus appeared that was always a match. To receive a fluid reward, monkeys were required to release the lever as soon as a match appeared. Trials were pseudo-randomized and balanced across all relevant features (e.g., match versus nonmatch, dot versus shape protocol, standard versus control, etc.).
The stimuli for the dot protocol were randomly arranged black dots displayed on a gray background (diameter 6° of visual angle). For each session, 100 different images per numerosity were generated with pseudo-randomly varied visual features: the diameter of the dots ranged from 0.5 to 0.9° of visual angle, and their positions were restricted only by the border of the gray background circle and the fact that they were not allowed to overlap each other. Sample and test stimuli were never identical. All four quantities were presented in each session with one standard and one control condition. Controls in the dot protocol included dot displays with constant circumference (the summed circumference of the dots was constant, such that dot size decreased as dot number increased, as opposed to in the standard condition), linear configuration (i.e., all dots were linearly arranged), and constant density (i.e., constant mean distance between dots) across all presented quantities (see Figure 1C). These measures prevented the monkeys from memorizing visual patterns instead of using the numerical information to solve the task. For the shape protocol, a sample stimulus consisted of a black Arabic numeral on a gray background circle. Font size (range 26 to 42 points) and position of the shapes were varied pseudo-randomly from trial to trial. The font “Arial” was used for standard trials; “Times New Roman,” “Souvenir BT,” and “Lithograph Light” were control fonts (see Figure 1D). The test stimulus for the shape protocol consisted of sets of black dots in the style of the dot protocol. Standard and control trials as well as trials from the dot and shape protocols were pseudo-randomly intermingled and appeared with equal probabilities in each session. These measures ensured that the monkeys generalized to the overall shape characteristics instead of memorizing local features.
Recordings were made from one left and one right hemisphere of the ventral convexity of the lateral PFC and in the fundus of the IPS of two rhesus monkeys (Macaca mulatta) in accordance with the guidelines for animal experimentation approved by the Regierungspräsidium Tübingen, Germany. These areas were chosen because in preceding studies [21–26] they were shown to contain visual numerosity-selective cells and, from human studies, are known to be activated during numerosity-related tasks [17–20,39–41]. Single-cell recordings were made with arrays of tungsten electrodes (1–2 MOhm impedance). Recording sites were localized using stereotaxic reconstructions from magnetic resonance images. Recordings in the IPS were done exclusively at depths from 9 to 13 mm below the cortical surface (Horsley-Clark coordinates, anterior/posterior, −5 mm or 0 mm) [24]. No attempts were made to preselect neurons. Off-line sorting was routinely applied to separate single units. As of the publication of this article, both monkeys are still engaged in discrimination tasks.
To determine the neuronal response latencies, averaged spike density histograms were derived with a 1-ms resolution, smoothed by a sliding window with a kernel bin width of 10 ms for all sample stimuli. A 200-ms time window before stimulus onset was used as baseline. If five consecutive time bins after stimulus onset reached a value higher than the maximum of the baseline period, response latency was defined by the first of these time bins. A default latency of 100 ms was used if no value could be calculated.
Putative association neurons were preselected based on a two-factor ANOVA. To account for different temporal response phases, spike rates were tested in four adjacent, nonoverlapping time windows. The first window (400 ms) started at the beginning of the sample period and was shifted by the neurons' response latencies. The second window (400 ms) followed right after the first one and covered the rest of the sample period. The subsequent two windows (450 ms each) covered the first and second part of the delay period. Selectivity for numerical values was calculated based on these discharge rates separately for the dot and shape protocols using a two-way ANOVA with main factors “numerical value” (one to four) and “stimulus condition” (standard and control). Cells were considered to be numerosity-selective only if they showed a significant main effect to “numerosity” in one of the four analysis windows, but no significant “stimulus condition” or interaction effect.
To derive averaged numerosity-filter functions, the tuning functions of individual neurons were normalized by dividing all spike rates of the tuning functions by the maximum activity, thus setting the activity at the preferred numerical value to 100%. Pooling the resulting normalized tuning curves across the entire population of association cells resulted in averaged numerosity-filter functions (see Figures 4C, S6A, and S6B). The population tuning functions were calculated for the time windows during which association neurons were significantly tuned to numerosity as tested by the two-way ANOVA. If neurons were significantly tuned in more than one window the analysis was restricted to the window with the smallest p-value.
The correlation analysis aimed to extract tuning similarities of individual neurons to numerical values in the shape and dot protocols. Figure S2 describes the application flow of the analysis. For each protocol (Figure S2A and S2B), eight trials per numerical value were chosen in a random manner (Figure S2C). Tuning functions were built with the averaged spike rates of these trials (Figure S2D and S2E). Next, the CCs between these tuning functions were calculated. The same subset of trials was shuffled so that the relation between neural activity and numerical value was abolished (Figure S2F); with this shuffled dataset, we calculated dummy tuning curves (Figure S2G and S2H) and computed the CCs (termed SPs) between them. This procedure was repeated 1,000 times, always using a new random subset consisting of eight trials to create two distributions of CCs and SPs. We quantified the discriminability between these distributions by ROC analysis. This analysis was accomplished for each of the sliding windows separately (one exemplary window is shown by the shaded bars in Figure S2A and S2B). Each separate analysis step is described in more detail below.
Out of the set of all trials (Figure S2A and S2B), we randomly drew eight trials per numerosity and protocol (i.e., in total four numerosities × two protocols × eight trials = 64 trials per turn; Figure S2C). This was done 1,000 times with replacement. We took care that no trial combination occurred more than once. The CCs and the SPs were calculated for each turn of the bootstrapping algorithm. This method filters robust effects across trials and provides reliable distributions.
The tuning functions tshape and tdot were composed of the spike rates of a given neuron obtained in the shape and dot protocols, respectively. Spike rates were obtained by averaging across the raw spike trains for 100 ms (see shaded windows in Figure S2A and S2B). Each tuning function consisted of four spike rates (corresponding to the neuron's responses to numerical value n = 1, 2, 3, and 4 during the identical time window). The spike rates were combined into one tuning function by sorting them in ascending numerical order (Figure S2D and S2E).
The CCs provided a measure to quantify the similarity between tuning to the shape and dot protocols. The rationale behind this was the following. A neuron that was ANOVA-selective in both protocols constituted a potential neuronal association substrate between shapes and numerical values. In addition to the mere selectivity in the dot and shape protocols, however, neurons should have similar tuning functions for the (direct and associated) numerical values in both protocols. Neurons showing different tunings to the numerical values in the two protocols cannot be regarded as association neurons and should be excluded. The normalized cross-correlation is an appropriate method for filtering for these criteria. The cross-correlation takes a neuron's entire tuning functions tshape(n) and tdot(n) for the numerical values n ∈ [1, 2, 3, 4] for dot and shape protocols, respectively, into account, rather than just comparing the preferred numerosities. We calculated the cross-correlation between these tuning functions for the shape and dot protocols. It is scale-invariant, since the means t̄shape and t̄dot are subtracted from each spike rate, and has the advantage of normalization, which allows comparison across all cells. The normalized CC was calculated as follows:
The SP is supposed to represent the chance correlation level, irrespective of numerical values. For its calculation we abolished the relationship between neural activity and numerical value by randomly assigning each neural response a numerical value (Figure S2F). Based on the tuning functions of this shuffled dataset (Figure S2G and S2H), we calculated CCs. We termed the distribution of these CCs the SP. Since the SP was calculated within the bootstrapping algorithm (1,000 repetitions), it provides a robust estimate of non-numerical-related fluctuations. In other words, the SP takes accidental correlations into account (e.g., those occurring at phasic “on” responses) and can thus be regarded as baseline correlation irrespective of influences by the presented numerical values.
To determine whether a given cell in a given time bin responded more similarly to shape and dot stimuli than expected by chance, we performed a ROC analysis [28] that provided a measure of how well the distributions of CCs and SPs were separated. The SPs were taken as the reference distribution. ROC values greater than 0.5 indicated that the CCs of a given cell were higher for the original dataset, arguing for correlated responses in the two protocols. We determined a significance threshold based on the ROC values obtained during the fixation period, during which only random correlations might occur. A neuron was termed an “association neuron” if it reached an ROC value after stimulus onset that was higher than the mean ROC value during the fixation period plus three standard deviations [49].
It needs to be emphasized that significant correlations are not caused by similar overall response modulations in the dot and shape protocols without being related to numerical value. Figure S4A and S4B shows an example neuron that responded very similarly to both protocols. Nevertheless, the CCs were close to zero (see red line in Figure S4C), because this neuron did not show any tuning to numerical value. The SP was also characterized by values fluctuating around zero (see blue line in Figure S4C). Consequently, the ROC analysis did not reveal any significant deviations from chance level (Figure S4D). In contrast, the neurons in Figure 3 showed strong modulations of firing rates with numerical value. As a consequence, the CCs reached high values up to one (see red lines in Figure 3D, 3I, and 3N). At the same time, however, the SP hovered around zero (see blue line in Figure 3D, 3I, and 3N). Thus, the ROC analysis correctly detected the periods of meaningful correlations (see Figure 3E, 3J, and 3O).
We calculated the CCs, the SP, and the area under the ROC curve (AUROC) in sliding windows (100-ms duration, shifted by 25 ms; see shaded area in Figure S2A and S2B). This procedure allows a detailed analysis of correlation development over time (Figure 7A and 7B) and reveals the different temporal correlation patterns of individual neurons (Figure 4A). We obtained almost identical proportions of association neurons when the analysis was based on nonoverlapping windows of 100-ms duration (n = 167; values exceeding threshold in at least one window to reach significance).
We evaluated the link between neuronal responses and behavior by analyzing the influence of erroneous judgments on the neuronal association. To that aim, we calculated CCs between the neuronal tuning functions based on error trials in the shape protocol and neuronal tuning functions obtained from correct trials in the dot protocol. Since the monkeys made very few errors, we often did not collect error trials for all tested numerical values. In these cases we restricted the analysis to the numerical values for which we obtained neuronal data during error trials (at least two numerical values). We compared these error-related CCs with CCs based on correct trials (again restricted to the same numerical values).
Was the proportion of neurons tuned to the same numerical value in both the dot and shape protocols higher than expected by chance? Since some neurons were tuned to numerosity in the dot protocol while others were encoding numerical information in the shape protocol, neurons encoding both formats may simply emerge by chance. We therefore compared the actual frequency of neurons with identical preferred numerical values in both protocols to chance occurrence based on probability calculations. To that aim, we considered the following three events: a cell is shape-selective, a cell is dot-selective, and a cell is selective for shapes and dots, formally written as
Based on our dataset, we calculated the probabilities that a cell encodes a specific preferred numerical value n in one of the protocols alone, given that the cell was ANOVA-selective to any numerical value in both protocols (P(shape = n|sig in both) for the shape protocol and P(dot = n|sig in both) for the dot protocol). To obtain the probability that a cell is encoding the preferred numerical value n in both protocols, given that the cell is selective to any numerical value in both protocols (P((shape = n ∧ dot = n)|sig in both)), the two obtained probabilities were multiplied. This approach was legitimate, because the two probabilities P(shape = n|sig in both) and P(dot = n|sig in both) are independent because of the pseudo-randomized presentation protocol. Thus, we can phrase the probability that a cell by chance encodes a specific shape and a specific number of dots simultaneously given that the cell is significant in both formats as
In total, the overall probability that a cell encodes one of the n shapes and the respective associated number of dots by chance, given that the cell is significant in both protocols, is the sum of the probabilities for all n:
The predicted chance probability Ppred was compared to the observed probability calculated as the percentage of cells with the same preferred quantity in both protocols in the pool of cells that were ANOVA-selective in both the dot and shape protocols. We calculated binomial tests with Ppred as test proportion. The observed fractions in the PFC differed significantly from the test proportions during sample and delay period (p < 0.001, n = 93, Ppred = 0.30, and p < 0.001, n = 139, Ppred = 0.31, respectively). The fraction of neurons in the IPS with the same preferred numerical value in both protocols was very small but differed significantly from the predicted frequency during the sample and delay period (p < 0.001, n = 5, Ppred = 0.32, and p < 0.001, n = 16, Ppred = 0.25, respectively). The results are depicted as fractions of the entire sample of recorded neurons (both selective and unselective) in Figure S5E.
This analysis represents a parallel argumentation line to the cross-correlation analysis. It shows on a stochastic basis that associations of visual signs and numerical values is not a coincidence.
|
10.1371/journal.ppat.1004864 | Phospholipase D1 Couples CD4+ T Cell Activation to c-Myc-Dependent Deoxyribonucleotide Pool Expansion and HIV-1 Replication | Quiescent CD4+ T cells restrict human immunodeficiency virus type 1 (HIV-1) infection at early steps of virus replication. Low levels of both deoxyribonucleotide triphosphates (dNTPs) and the biosynthetic enzymes required for their de novo synthesis provide one barrier to infection. CD4+ T cell activation induces metabolic reprogramming that reverses this block and facilitates HIV-1 replication. Here, we show that phospholipase D1 (PLD1) links T cell activation signals to increased HIV-1 permissivity by triggering a c-Myc-dependent transcriptional program that coordinates glucose uptake and nucleotide biosynthesis. Decreasing PLD1 activity pharmacologically or by RNA interference diminished c-Myc-dependent expression during T cell activation at the RNA and protein levels. PLD1 inhibition of HIV-1 infection was partially rescued by adding exogenous deoxyribonucleosides that bypass the need for de novo dNTP synthesis. Moreover, the data indicate that low dNTP levels that impact HIV-1 restriction involve decreased synthesis, and not only increased catabolism of these nucleotides. These findings uncover a unique mechanism of action for PLD1 inhibitors and support their further development as part of a therapeutic combination for HIV-1 and other viral infections dependent on host nucleotide biosynthesis.
| Replication of all human viruses depends on building blocks derived from the metabolic pathways of the infected host cell. The production of progeny virions requires synthesis of viral nucleic acids from deoxyribonucleotide triphosphates (dNTPs). HIV-1 infection in resting T cells is limited, at least in part, because the levels of critical nucleotides are low. However, stimulation of T cells turns on their metabolic machinery to increase c-Myc expression and subsequent synthesis of these key components of RNA and DNA, which augments HIV-1 replication. We have identified PLD1 as a key molecular switch that couples stimulatory T cell signals to c-Myc-dependent nucleotide biosynthesis. We also found that a small molecule that inhibits PLD1 suppresses HIV-1 replication by limiting c-Myc-dependent effects of T cell activation that support efficient HIV reverse transcription. Our study provides insight into a novel way of targeting T cell activation-induced processes such as nucleotide biosynthesis that has potential to augment current therapeutics for HIV-1.
| HIV-1 replication in resting CD4+ T cells is restricted post-entry, but prior to integration [1]. Several groups have reported that suboptimal dNTP pools in these metabolically quiescent cells support only inefficient reverse transcription and subsequent integration [2,3]. Cellular activation, or addition of exogenous deoxyribonucleosides, relieves the post-entry block to HIV-1 infection in resting CD4+ T cells [2,3]. Decreasing dNTP pools in activated T cells with hydroxyurea (HU), a ribonucleotide reductase inhibitor, was also shown to suppress HIV-1 replication in vitro [4,5], although clinical trials were limited by serious toxicities [6]. More recently, glucose metabolism has been identified to play a fundamental role in providing a carbon source for both T cell function and HIV-1 replication [7]. Notably, glucose uptake and its metabolism via the pentose phosphate pathway produces ribose intermediates that are critical for the synthesis of all nucleotides [8]. Expression of Glut1, a glucose transporter, is also essential for HIV-1 infection of activated CD4+ T cells [9]. Finally, catabolism of dNTPs is one of the mechanisms implicated in the anti-HIV activity of sterile alpha motif—histidine-aspartic domain-containing protein 1 (SAMHD1) in resting, but not activated, CD4+ T cells [1].
Recent reports have supported a prominent role of the c-Myc oncogene as a “master regulator” of transcriptional regulation of genes needed for nucleotide biosynthesis and glucose metabolism essential for both cellular and viral processes [10,11]. In an elegant study utilizing acute conditional deletion of c-Myc in murine T cells, Wang and colleagues demonstrated that c-Myc is essential for metabolic reprogramming and nucleotide precursor accumulation in activated T cells [11]. Consistently, c-Myc was also found to be highly induced upon T cell activation and required for cell growth and proliferation [11]. Further, pharmacologic inhibition of the Ras/ERK pathway was found to abrogate expression of c-Myc after T cell activation [11]. Inhibition of either the Ras/ERK signaling module or c-Myc activity has been reported to suppress early steps of HIV-1 replication in activated T cells [12,13,14]. However, the mechanism by which T cell activation induces c-Myc expression to initiate this cascade remains undefined.
Interestingly, one pathway potentially involved in coupling T cell activation to c-Myc expression, phospholipase D (PLD)-mediated hydrolysis of phosphatidylcholine to choline and phosphatidic acid (PA) [15], is activated whether T cells are stimulated by the mitogenic lectin phytohaemagglutinin (PHA) or via antibody-mediated crosslinking of the T cell receptor (TCR). In humans, PLD exists as two isoforms derived from separate genes, PLD1 and PLD2 [16]. The two PLD isoforms have been implicated in a plethora of signaling pathways that influence numerous essential cellular functions, such as vesicular trafficking, exocytosis, autophagy, regulation of cellular metabolism, and tumorigenesis [16]. Furthermore, PA upregulates Ras/ERK [17], and increases expression of c-fos and c-Myc [18]. This occurs if PA is supplied either exogenously or endogenously through PLD1 or 2 activity [18]. Since PLD1 has been shown to mediate responses downstream of the T cell receptor [19], we tested the hypothesis that the PLD1 signaling pathway couples T cell activation to cellular processes essential for HIV-1 replication. Experiments undertaken here using pharmacologic and genetic inhibition of PLD1 provide evidence that PLD1 activity links T cell activation signals to the Ras/ERK/c-Myc signaling cascade required for metabolic reprogramming that expands dNTP pools. We also report that PLD1 inhibition blocks HIV-1 reverse transcription and replication.
Human resting CD4+ T cells were stimulated with PHA in the presence and absence of a PLD1-selective small molecule inhibitor, VU0359595 (PLD1i) [20]. Here, inhibition of PLD1 reduced ERK1/2 phosphorylation after T cell activation with PHA, in a similar manner as direct suppression of ERK activity with the selective MEK/ERK inhibitor U0126 (p-ERK in Fig 1A). Ras/ERK signaling also promotes site-specific phosphorylation of ribosomal protein S6 (S6) at Ser235/236 (p-S6, in Fig 1A) [21]. Notably, resting CD4+ T cells activated with PHA in the presence of either PLD1i or U0126 had a marked reduction in S6 phosphorylation (p-S6, Fig 1). Since PLD1 has been shown to activate the parallel pathway of mechanistic target of rapamycin (mTOR) / S6 kinase 1 (S6K1), perturbation of mTOR by PLD1i to decrease its activity could also contribute to a diminution in p-S6. To test this possibility, the levels of phosphorylation of specific targets of the mTOR pathway, carbamoyl-phosphate synthetase 2, aspartate transcarbamoylase, dihydroorotase (CAD) at Ser1859 (p-CAD) and translation repressor protein 4E-BP1at Ser65 (p-4E-BP1) were also determined. PHA stimulation increased the abundance of p-S6, p-CAD and p-4E-BP1, while rapamycin, an allosteric mTORC1 inhibitor known to suppress c-Myc expression, blocked these phosphorylation events (Fig 1A) [11]. Inhibition of either ERK or PLD1 also reduced the levels of p-S6, p-CAD, and p-4E-BP1, suggesting that PLD1i does indeed suppress mTOR activity in stimulated CD4+ T cells. PLD1i also abrogated induction of the total level of these proteins following T cell activation (S6, CAD, 4E-BP1; Fig 1A). Since depletion of c-Myc decreased levels of total S6, CAD, and 4E-BP1 [11,22], we hypothesized that PLD1 inhibition diminished the overall levels of these proteins by decreasing c-Myc expression. Therefore, we assessed expression levels of c-Myc in cells that were stimulated with PHA in the presence of PLD1 and ERK inhibitors. Levels of the c-Myc-dependent nucleotide biosynthetic enzymes thymidylate synthase (TS), large subunit of ribonucleotide reductase (RRM1), and small catalytic subunit of ribonucleotide reductase (RRM2) were also studied. Blocking PLD1, ERK, or mTOR impaired activation-dependent induction of c-Myc, TS and RRM2 (Fig 1A), as well as RRM1, proteins (S1 Fig). These results phenocopied the effects of a specific c-Myc inhibitor (10058-F4) (Myci in Fig 1A) [23]. PLD1i and Myci also inhibited activation-induced expression of RRM2 (Fig 1B) and RRM1 (S1 Fig) in primary CD4+ T cells. To confirm genetically that PLD1 is required for optimal c-Myc expression, ERK and mTOR activity in activated CD4+ T cells, we depleted PLD1 by siRNA-mediated silencing. We observed reduced expression of c-Myc, as well as reduced phosphorylation of ERK and mTOR target 4E-BP1 with two independent PLD1 siRNAs (Fig 1C). c-Myc depletion by siRNA also reduced expression of PLD1, as well as known c-Myc dependent targets (Fig 1C). These results suggest that PLD1 catalytic activity couples T cell activation signals to de novo nucleotide biosynthesis by augmenting ERK and mTOR-dependent c-Myc expression. These results are consistent with the effects of PLD1i here and those previously reported [24]. Taken together, these results suggest a positive correlation between the level of c-Myc and PLD1expression in activated CD4+ T cells consistent with a positive feedback loop (Fig 1D). We hypothesize that PLD1 activity increases expression of both c-Myc and previously described c-Myc-dependent genes, and that increases in PLD1 and c-Myc amplify each other’s expression (Fig 1D).
Stimulation of CD4+ T cells with PHA/IL2 for only 30 min was sufficient to induce a 2-fold increase in PLD activity, relative to the unstimulated control (2.07 ± 0.33 vs. 1 ± 0.49 relative fluorescence, P = 0.023). Importantly, this rapid burst of activity after T cell activation was reduced by PLD1i pretreatment, when compared to PHA/IL-2 stimulated cells (1.27 ± 0.35, P = 0.044), but was not significantly different when compared to unstimulated cells (P = 0.51). Others have documented that c-Myc expression is also rapidly induced, within 2 hours of T cell activation [11]. This may help explain why effects of PLD1i on c-Myc (Fig 1C) appeared more robust than those of siRNAs against PLD1 (Fig 1A). Maximal effects of siRNA on protein expression are not observed before 24 hours after transfection. Therefore, the greater decrease of c-Myc observed here with PLD1i than siRNA against PLD1 is consistent with more acute inhibition of PLD1 by the inhibitor than genetic silencing [11,25].
c-Myc also drives the expression of key nutrient transporters needed for cell growth and proliferation after T cell activation: Glut1 (for glucose); SNAT1 and SNAT2 (for glutamine); Slc7a5/LAT1 (for large neutral amino acids) [11]. Since CD28 stimulation is essential for optimal surface expression of Glut1 [26], resting CD4+ T cells were activated with anti-CD3/anti-CD28. Cells were then surface-stained for Glut1 and the activation marker CD25. Consistent with previous reports, inhibition of PLD1 suppressed expression of CD25 [27] (Fig 2A); however, PLD1i had little observable effect on expression of activation markers CD71 or CD98, suggesting that PLD1i does not lead to global inhibition of T cell activation (S2 Fig). PLD1i treatment prevented the upregulation of Glut1 surface expression on a sub-population of activated T cells, similarly to prior observations (Fig 2A) [28]. Additionally, we found that inhibition of PLD1 or c-Myc in activated CD4+ T cells reduced total cellular expression of nutrient transporters Glut1, SNAT1, and SNAT2. Importantly, inhibition of upstream mediators of c-Myc expression (ERK and mTORC1) also impaired expression of these nutrient transporters (Fig 2B). Quantitative PCR (qPCR) on RNA from CD4+ T cells transfected with siRNAs targeting c-Myc or PLD1 confirmed reduced mRNA expression of RRM2, SNAT1, SNAT2, and Slc7a5/LAT1 (Fig 2C). Slc7a5/LAT1 is required for both mTORC1 activity and c-Myc expression in activated T cells (Fig 2C) [29]. Knockdown of c-Myc and PLD1 again resulted in reduced expression of the alternate RNA (Fig 2C), consistent with data in Fig 1C that suggested a positive feedback loop as illustrated in Fig 1D. The observations of inhibition of PLD1 activity resulting in the impairment of coordinated expression of c-Myc, nucleotide biosynthetic genes, and nutrient transporters are consistent with PLD1 signaling being upstream of induction of c-Myc in activated T cells.
Activation of T cells leads to increased synthesis of biosynthetic precursors that enable cell proliferation [11]. To this end, c-Myc coordinates increased uptake of glucose and glutamine with nucleotide biosynthesis to facilitate metabolic reprogramming of activated CD4+ T cells. Furthermore, like genetic ablation of c-Myc activity, glucose or glutamine starvation severely compromises activation-induced proliferation of T cells [11]. Since inhibition of PLD1 activity also reduces both c-Myc (Fig 1) and c-Myc-dependent nutrient transporter expression (Fig 2), we investigated the effects of PLD1i on cell cycle distribution and proliferation of activated CD4+ T cells. First, CD4+ T cells were pretreated with indicated inhibitor and then stimulated for 72 h in the continued presence of inhibitor. Cell-cycle progression was determined by simultaneously staining for RNA (Pyronin Y) and DNA (7-AAD) followed by flow cytometry. Fig 3A shows the distribution of cell-cycle phases identified by this technique. Resting CD4+ T cells remain in G0, but increase their RNA content after stimulation and progress into G1a then G1b. Activated CD4+ T cells then initiate DNA synthesis and enter S phase followed by G2/M phase completion. We found that PLD1i-treated cells progressed to all stages of the cell-cycle; however, when compared to control cells, more PLD1i-treated cells were in G1b (24.5% versus 16.8%) (Fig 3B). This suggested that inhibition of PLD1 activity delayed the initiation of DNA synthesis at the G1b/S boundary. Consistent with this hypothesis, genetic ablation of RRM2, a c-Myc target gene suppressed by PLD1i (Fig 1A), was previously found to induce G1/S phase cell-cycle arrest [30]. We also directly assessed proliferation of PLD1i-treated CD4+ T cells by determining the dilution of CellTRACE Violet stain by flow cytometry 72 h after stimulation (Fig 3C). PLD1i suppressed activation-induced T cell proliferation in a concentration-dependent manner, albeit less so than did the c-Myc inhibitor (10058-F4). We also observed a delay of activation-induced proliferation by both U0126 and rapamycin (Fig 3C), as previously reported [31,32]. Cytotoxicity was not detected with PLD1i (S2 Fig, bottom panel).
To directly assess dNTP pool expansion, resting CD4+ T cells were stimulated with PHA/IL2 in the presence and absence of PLD1i, and dNTP levels were quantified by mass spectrometry. PHA stimulation of CD4+ T cells resulted in 3.66-, 1.6-, and 9-fold increase in dATP, dCTP, and dTTP, respectively. Inhibition of PLD1 activity potently restricted the expansion of the dNTP pools. Increases in both dATP and dCTP were nearly completely inhibited and dTTP levels only increased 2-fold in the presence of PLD1 inhibitor (Fig 4A–4C). Hydroxyurea (HU) treatment decreased only dATP levels (Fig 4A), as previously reported [33].
Since inhibition of PLD1 activity in activated CD4+ T cells limits dNTP pool expansion, we hypothesized that HIV-1 replication would be impaired. To test this hypothesis, resting primary CD4+ T cells were pretreated with PLD1i or vehicle and then stimulated with PHA/IL2. Cells were then infected with a single-round CXCR4-tropic envelope-pseudotyped GFP-expressing HIV-1. CXCR4-tropic virus, rather than CCR5-tropic virus, was used to limit assessment to effects of PLD1i on post-entry steps of HIV-1 replication. This is because PLD1i-mediated decreases in mTOR activity that diminish CCR5 surface expression and HIV-1 entry could confound analyses of CCR5-tropic virus [34]. PLD1i inhibited CXCR4-tropic HIV-1 infection by nearly 75% in CD4+ T cells from four independent donors (Fig 5A). The effects of PLD1i on HIV infection were rescued by adding exogenous deoxyribonucleosides (dN), which bypassed the need for ribonucleotide synthesis and reduction; degree of rescue varied in cells from different donors (Fig 5A). Exogenous dN had little effect on HIV-1 infection in control cells (Fig 5A). Quantitative PCR was used to measure viral early reverse transcripts (ERT), late reverse transcripts (LRT), and 2-LTR circles at 24 hours after infection (Fig 5B); the latter is an indicator of nuclear import of full-length viral cDNA. PLD1i had little effect on the level of ERT cDNA, consistent with normal levels of HIV-1 cell entry and initiation of reverse transcription (Fig 5B). Assessment of CD4 and CXCR4 surface expression on PLD1i-treated cells also confirmed lack of receptor or co-receptor down-regulation by PLD1i that could affect entry (S4 Fig). PLD1i suppressed the accumulation of LRT cDNA after HIV-1 infection (Fig 5B), consistent with a previous study of ERK inhibitors [13]. PLD1i also reduced the levels of 2-LTR circles more markedly than LRT cDNA. Treatment of cells with HU, known to limit HIV-1 reverse transcription and dNTP pools by inhibiting ribonucleotide reductase RRM2-dependent activity, also reduced the levels of LRT and 2-LTR circles in HIV-1 infected cells, with a similarly greater effect on 2-LTR circles (Fig 5B)[5].
To confirm and further define the requirement of PLD1-dependent processes for HIV-1 replication, in a separate experiment we determined the effects of PLD1i on the accumulation of viral cDNA products at 8, 16, and 24 h after infection (Fig 5C). Consistent with PLD1i-dependent effects on dNTP pools, the kinetics of reverse transcription was markedly delayed when compared to DMSO vehicle-treated cells. Reduced levels of LRT, and 2-LTR cDNA were again detected in PLD1i-treated cells at each time point. ERT cDNA levels were decreased at 8 and 16 hours, but only minimally decreased at 24 hours. Furthermore, inhibition of the PLD1 target c-Myc recapitulated these effects at the 24 h time point (other time points not studied with Myci) (Fig 5C). This effect on the kinetics of reverse transcription can cause a “bottle-neck” upstream of HIV-1 nuclear import and can explain, at least in part, the reduction in 2-LTR circle levels we observed in PLD1i-treated cells based on a delay in availability of completed reverse transcripts in the cytoplasm.
Since PLD1i-treated cells have reduced dNTP pools and exogenous dN have been shown to increase the kinetics of reverse transcription in resting CD4+ T cells [3], we determined the effects of dN addition on HIV-1 cDNA products in PLD1i-treated cells (Fig 5D). Exogenous dN increased the levels of LRT cDNA in PLD1i-treated cells, consistent with PLD1i’s mechanism of RT inhibition being due to its effects on dNTP pool expansion. Interestingly, dN addition did not reverse PLD1i-decreased 2-LTR formation (Fig 5D). This observation suggests that inhibition of PLD1 activity has an additional effect, not reversed by exogenous dN that diminishes HIV-1 cDNA nuclear import and/or 2-LTR formation in the nucleus.
This study demonstrates that PLD1 is required to couple activation of primary CD4+ T cells to the c-Myc-dependent coordinated upregulation of nutrient transporters, dNTP biosynthesis, and other biosynthetic pathways that we and others have previously reported to support HIV-1 replication [35]. Our results now also suggest a positive feedback loop between c-Myc and PLD1 not previously appreciated (Fig 1D). Loss of PLD1-mediated metabolic reprogramming caused dNTP-dependent delay in the accumulation of HIV-1 late reverse transcripts and other anti-HIV effects. PLD1-dependent downstream effects are also likely to be critical for replication of cells and other viruses, since c-Myc overexpression increases accumulation of nucleotides critical for DNA replication and cell division of cancer cells and adenovirus-infected cells [36,37].
Addition of exogenous dN rescues the PLD1i-mediated decreases in HIV-1 replication and accumulation of late reverse transcripts (Fig 5). This is evidence that PLD1i acts against HIV-1 through a specific effect that limits dNTP pool expansion following T cell activation, rather than via an off-target effect. It is of note that siRNA against PLD1 did not block HIV-1 reverse transcription in our hands. However, the demonstration here, and elsewhere, of rapid onset of PLD activity following T cell activation (in 30 minutes) indicates a technical limitation in using genetic silencing to confirm the specificity of the potent and very rapid effects of PLD1i, given that siRNAs do not decrease protein expression until 24 hours after T cell activation [25]. Results also show that limited dNTP pool expansion is not the only mechanism by which PLD1i decreases HIV-1 replication. The greater decrease in 2-LTR circles than LRT (Fig 5B–5D) and lack of reversal of the reduction in 2-LTR circles by added dN suggest that PLD1i has additional effects on nuclear import and/or 2-LTR circle formation that are independent of its effect on dNTP pools. In line with this postulate, it has been hypothesized that slowing reverse transcription may enhance the action of host cell restriction factors [38].
Toxicity profiles have been characterized and previously reported for the PLD2 preferring inhibitors [38], but to date detailed toxicological characterization has not been performed on the compound series showing preference for the PLD1 isoenzymes. Importantly though, compounds using the same chemical scaffold that were shown to be dual isoenzyme inhibitors, and similar to those used in this report, have been tested in human clinical trials and no overt toxicity was observed [39]. Given serious adverse events seen in clinical trials of HU-based regimens, it is important to further exclude potential toxicity of PLD1 inhibitors.
Glut1 expression in CD4+ cells is essential for HIV-1 replication in target CD4+ T cells, since knockdown of Glut1 inhibited early HIV-1 replication [9]. However, the mechanism through which Glut1 knockdown inhibited HIV-1 replication has not yet been delineated. The current results, and those previously reported, suggest the hypothesis that limiting both Glut1 and glutamine transporter expression may also indirectly decrease HIV-1 replication via host cell dNTP depletion. Moreover, increased Glut1 expression is observed in CD4+ T cells in HIV-infected patients and the magnitude of increase is directly associated with the pace of T cell depletion and disease progression, thus suggesting an additional rationale for targeting this pathway for therapeutic intervention [28].
The importance of expanded dNTP pools for HIV-1 replication is well established, and recent studies of SAMHD1 have added the suggestion that enhanced catabolism of dNTPs may also contribute to anti-HIV effects [39,40]. Earlier reports have clearly shown that inhibiting ribonucleotide reductase activity with HU following PHA activation suppresses early steps of HIV-1 replication, established limiting host CD4+ T cell dNTP synthesis as an antiretroviral strategy [4,5]. However, data shown in Fig 4 demonstrates that inhibition of PLD1-dependent biosynthetic pathways has a more robust effect on dNTP biosynthesis than RRM2 inhibition, the mechanistic target of HU. Taken together with results depicted in Fig 5, the data strongly support a mechanism where PLD-regulated nucleotide biosynthesis and other processes play a major role in supporting HIV-1 replication. We found that inhibition of PLD1 also limits CD4+ T cell activation-induced proliferation (Fig 3C). Limiting proliferation of T cells may also benefit anti-HIV strategies in ways that medications targeting viral processes cannot. Abnormal T cell activation/proliferation is hypothesized to contribute to “non-AIDS” adverse outcomes, as it persists even among patients with prolonged suppression of HIV replication by current medications. Indeed, this excessive activation/proliferation may not be ablated even when antiretrovirals are started in the earliest stages of acute infection (Utay NS, et al. Abstract 47, CROI 2015, presented February 24, 2015). In addition, recent reports indicate that latently infected resting memory CD4+ T cells may persist during antiretroviral therapy at least in part, because of HIV integrant-driven cellular proliferation [41,42]. If PLD1 inhibition is found to be safe in the future, it could be used to test if these pathogenic processes can be ameliorated. Importantly, certain PLD inhibitors also have demonstrated ability to traverse the blood brain barrier to target HIV-1 replication in myeloid-derived cells in the CNS, unlike some current anti-HIV drugs [43,44]. Furthermore, perturbation of nucleotide pools may be an additional factor beyond recently described effects on innate and adaptive immune responses contributing to PLD inhibitor-mediated blockade of influenza virus replication [45]. It is also provocative to speculate that short-term blockade of host cell synthesis of ribonucleotides and deoxyribonucleotides may provide a strategy for broad-spectrum activity against diverse RNA and DNA viruses. When compared to the anti-HIV activity of FDA-approved antiretrovirals, the effects of PLD1i are modest; however, this inhibitor constitutes only an early candidate in a search for more effective compounds for advancing to clinical development. Further development of PLD inhibitors holds promise as a potential therapeutic for viral infections that require host nucleotide pools for replication as well as cancers, although the roles of PA production, whether biophysical, transcriptional, or as a signaling molecule, in these therapeutic interventions have yet to be fully elucidated.
Peripheral blood mononuclear cells (PBMCs) were purified from healthy blood donor specimens obtained from Lifesource (Rosemont, IL) by Ficoll-Hypaque PREMIUM (GE Healthcare) gradient centrifugation. Resting CD4+ T cells were isolated from negatively-selected total CD4+ T cells (CD4+ T Cell Isolation Kit, Miltenyi Biotec) using CD25+ and HLA-DR+ microbeads (Miltenyi Biotec) and cultured in RPMI-1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (Hyclone), glutamine (2 mM) and antibiotics (100 U/ml penicillin, 100 mg/ml streptomycin). Cells were activated with PHA-L (5 μg/ml) (Roche) and IL-2 (20U/ml) (Roche) or anti-CD3/anti-CD28 beads (Invitrogen) (1 bead/5 cells).
For transfections, nontargeting or siRNAs targeting c-Myc (Santa Cruz Biotech) or PLD1 (Santa Cruz Biotech (PLD1 siRNA#1) and Dharmacon/ThermoFisher (PLD1 siRNA#2)) were nucleofected into resting CD4+ T cells using an AMAXA nucleofector apparatus. Transfection was performed with human T-cell Nucleofector kit (LONZA), following the manufacturer's instructions. Briefly, 240 pmol (~3μg) of siRNA was added to 1 x 107 cells resuspended in 100 μl of Nucleofector solution for each. Nucleofector program U-14 was used. Nucleofected cells were transferred into 2 ml of medium and incubated at 37°C for 24 h before medium was changed and cells resuspended in 1 ml of medium. Cells were then stimulated by adding anti-CD3/anti-CD28 beads (1:5; bead:cell ratio), cells and beads were pelleted for 5 minutes at 1200 x g in 96-U-well plates, and incubated at 37°C for 48 h before cells were harvested for analysis.
Cell cycle subcompartment determination by staining with 7-aminoactinomycin D (Invitrogen) and pyronin Y (Sigma) was performed as previously described [46]. For analysis of surface markers, cells were stained at 4°C for 30 min with antibodies against Glut1 (R&D Systems), CD25 (BD Bioscience), CD71 (BD Bioscience), and CD98 (BD Bioscience) in PBS containing 1% BSA. Flow cytometric data was obtained on a LSRFortessa (Becton Dickinson) and analyzed with FlowJo software (TreeStar).
To follow cell division, cells (107/ml) were pulsed with CellTRACE Violet (5μM) in PBS for 30 min at 37°C. Cells were then washed with PBS, resuspended in growth medium, and treated as indicated before stimulation by adding anti-CD3/anti-CD28 beads (1:5; bead:cell ratio). Cells and beads were pelleted for 5 minutes at 1200 x g in 96-U-well plates and incubated at 37°C for 72 h before cells were harvested for analysis by flow cytometry.
HIV-1 stocks were prepared by transfecting 293T cells as previously described [35]. All virus stocks were treated with TURBO DNase (Lifetechnologies) (100 U/ml) for 30 min at 37°C followed by 30 min at RT. CD4+ T cells were infected with virus (50 ng of p24 per 2 x 105 cells) by spinoculation (1,200 x g, 2 h), followed by incubation at 37°C. Where indicated, cells were pretreated with PLD1i (VU0359595), 1mM HU, or 50 μM deoxyribonucleosides before infection.
Cells were washed with ice-cold phosphate-buffered saline, harvested and whole lysates were prepared using RIPA buffer [50mM Tris-HCl pH 8.0, 150mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, and 1mM EDTA] with Protease Inhibitor Cocktail (Roche). Whole cell lysates were clarified (10,000 x g for 20 min at 4°C) and resolved by SDS-PAGE on 4–12% gradient Bis-Tris or 3–8% Tris-acetate polyacrylamide gel and transferred to a nitrocellulose membrane. The membrane was blocked with SuperBlock Blocking Buffer (Thermo Scientific) and incubated with indicated antibodies overnight at 4°C in SuperBlock. Blots were then incubated with anti-mouse or anti-rabbit antibody conjugated with horseradish peroxidase (Thermo Scientific) before detection (SuperSignal West Dura Chemiluminescent Substrate, Thermo Scientific).
Cells were treated as indicated and whole lysates were prepared using NP-40 lysis buffer [50mM Tris-HCl pH 7.5, 150mM NaCl, 1% Nonidet P-40, and 1mM EDTA] with Protease Inhibitor Cocktail (Roche). Whole cell lysates were clarified (10,000 x g for 20 min at 4°C) and equal cell equivalents of whole cell lysates were used to determine total PLD activity with the Amplex Red PLD Assay kit (Lifetechnologies), according to the manufacturer’s protocol.
Total RNA was isolated using RNeasy Plus Mini Kit (QIAGEN). Briefly, cDNA for qPCR was generated from total RNA using oligo dT primers (Promega) and M-MLV Reverse Transcriptase (Promega). Quantitative real-time PCR was performed on an iCycler (Bio-Rad) using iQSYBR Green (Bio-Rad) detection. Samples were analyzed in triplicate and normalized to actin RNA (ΔΔCt method). Primer pairs were:
Actin (GGACTTCGAGCAAGAGATGG, GGACTTCGAGCAAGAGATGG), RRM2 (CAAGCGATGGCATAGTAA, TGTAAGTGTCAATAAGAAGACT), SNAT2 (AAGACCGCAGCCGTAGAAG, CAGCCATTAACACAGCCAGAC), LAT1 (GTGCCGTCCCTCGTGTTC, GCAGAGCCAGTTGAAGAAGC), PLD1 (TGTCGTGATACCACTTCTGCCA, AGCATTTCGAGCTGCTGTTGAA), c-Myc (TCCAGCTTGTACCTGCAGGATCTGA, CCTCCAGCAGAAGGTGATCCAGACT), ASCT2 (ATCGTGGAGATGGAGGA, AAGAGGTCCCAAAGGCAG), SNAT1 (GGCAGTGGGATTTTGGGACT, TGACCAAGGAGAACAACACCC).
Total cellular DNA was isolated from HIV-1 infected cells (DNeasy DNA isolation kit, Qiagen). Real time PCR was performed using iQSYBR Green (Bio-Rad) detection (Bio Rad CFX96). Reaction mixtures contained 250 nM of each primer and 100 to 300 ng template DNA in a final volume of 25 μl. The sequence of the primers used for real time PCR for early reverse transcription (ERT), late reverse transcription (LRT), two LTR circle DNA (2LTR) and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) were: ERT (TTA GAC CAG ATC TGC GCC TGG GAG, GGG TCT GAG GGA TCT CTA GTT ACC), LRT (TGT GTG CCC GTC TGT TGT GTG A, GAG TCC TGC GTC GAG AGA TCT), 2LTR (AAC TAG GGA ACC CAC TGC TTA AG, TCC ACA GAT CAA GGA TCT CTT GTC), GAPDH (GAA GGT GAA GGT CGG AGT, GAA GAT GGT GAT GGG ATT TC).
Samples were analyzed in triplicate and normalized to GAPDH (ΔΔCt method).
Cellular analysis of dNTPs was performed as previously reported [47,48]. Briefly, after indicated treatments, cells were pelleted at 1000 × g, the supernatant aspirated and pellet resuspended in 500 μl of 70% methanol/water mixture at -20°C. Suspensions agitated for 4 minutes at 4°C, and then incubated at -20°C for one hour. Internal standards were then added; for dNTPs, 4 nmols of aminoallyl-UTP; for carbamoyl aspartate, 5 nmols of citrate-d4. Suspensions were agitated again at 4°C for one minute and centrifuged (18,000 × g, 10 minutes, 4°C). The supernatant was collected, transferred to a bullet tube and solvent evaporated under vacuum. Immediately prior to analysis, extracts were reconstituted in 100 μl of a 2 mM ammonium acetate, 3 mM hexylamine solution in water (pH 9.2).
dNTPs were quantified (adapted from [49]) by chromatography on Acquity I-class UPLC (Waters, Milford, MA) with detection by MDS SCIEX 4000QTRAP hybrid triple quadrupole/linear ion trap mass spectrometer (Applied Biosystems). Acquity BEH C18 column (2.1 x 50 mm, 1.7 μ) with a 10 μl sample injection was used for metabolite delivery and chromatographic resolution. Solvent A consisted of 2 mM ammonium acetate and 3 mM hexylamine in water (pH 9.2); solvent B was 100% acetonitrile. Flow rate of 0.6 ml/min was maintained with a linear gradient as follows: 0 minutes, 9% B; 2 minutes 16% B; 5 minutes, 16% B; 5.5 minutes, 100% B; 6.5 minutes, 100% B; 7 minutes, 9% B; 8 minutes, 9% B. For dNTP analysis, the mass spectrometer was operated in negative MRM mode; the following mass transitions were monitored: dATP, 490/159; dCTP, 466/159; TTP, 481/159. dGTP could not be reliably quantified with this method since its molecular fragmentation pattern and retention time were identical to ATP.
P values were calculated with Student’s t test. P values<0.05 were considered significant.
|
10.1371/journal.ppat.1006039 | The CXCL12/CXCR4 Signaling Pathway: A New Susceptibility Factor in Human Papillomavirus Pathogenesis | The productive human papillomavirus (HPV) life cycle is tightly linked to the differentiation and cycling of keratinocytes. Deregulation of these processes and stimulation of cell proliferation by the action of viral oncoproteins and host cell factors underlies HPV-mediated carcinogenesis. Severe HPV infections characterize the wart, hypogammaglobulinemia, infection, and myelokathexis (WHIM) immunodeficiency syndrome, which is caused by gain-of-function mutations in the CXCR4 receptor for the CXCL12 chemokine, one of which is CXCR41013. We investigated whether CXCR41013 interferes in the HPV18 life cycle in epithelial organotypic cultures. Expression of CXCR41013 promoted stabilization of HPV oncoproteins, thus disturbing cell cycle progression and proliferation at the expense of the ordered expression of the viral genes required for virus production. Conversely, blocking CXCR41013 function restored virus production and limited HPV-induced carcinogenesis. Thus, CXCR4 and its potential activation by genetic alterations in the course of the carcinogenic process can be considered as an important host factor for HPV carcinogenesis.
| Human papillomaviruses (HPV) are epitheliotropic tumor viruses causing mostly benign warts but that have developed strategies to establish persistent infections. Although host immune responses clear most infections, persistence of some HPV types causes ~5% of human cancers and severe pathogenesis in immunosuppressed individuals. How early events in HPV infection, determined by the interaction between viral and host proteins, might lead to viral persistence and pathogenesis is unknown. Here, we thought to investigate this issue by providing mechanistic insights into the selective susceptibility to HPV pathogenesis displayed by patients who are immunosuppressed as a consequence of mutations in the CXCR4 gene encoding for the receptor of the CXCL12 chemokine (WHIM syndrome). We previously unraveled the existence of a general interplay between the CXCL12/CXCR4 axis and HPV, which is hijacked toward cell transformation upon expression of the CXCR4 mutant. Here, using three dimensional epithelial cell cultures to analyze the HPV life cycle, we found that the CXCR4 mutant promotes cell hyperproliferation and stabilization of viral oncoprotein expression at the expense of virus production. Our results, which identify CXCR4 as an important gatekeeper of keratinocyte proliferation and as a new susceptibility factor in HPV pathogenesis, may be translated into anti-viral and anti-cancer strategies.
| Human papillomaviruses (HPVs) are a family of highly related non-enveloped epitheliotropic viruses that have co-evolved with their human host and developed powerful strategies to establish persistent infection [1]. Many reports have described HPVs as commensal viruses that can persist on healthy skin [2–4]. HPVs selectively infect basal keratinocytes of stratified epithelia and other discrete populations, including the cells located in the squamocolumnar junction of the cervix [5, 6]. These viruses undergo productive replication strictly in the terminally differentiated layers of the infected epithelium, and in most cases, cause no tissue damage or only benign warts [7]. Although host immune responses resolve most infections, instauration of persistent infections by the mucosal types of HPVs classified as high-risk, of which HPV16 and HPV18 are the most significant, is responsible for almost all cases of cervical carcinoma, a leading cause of cancer death in women. These viruses are also responsible for most anal cancers, as well as a fraction of vulval, vaginal, penile, and oropharyngeal cancers, causing nearly 5% of human cancers worldwide [8]. Cutaneous high-risk HPV types that normally persist without symptoms have also been associated with non-melanoma skin cancers in some rare genetic diseases or in immunosuppressed patients [9, 10]. The oncogenicity of high-risk HPVs appears in the context of asymptomatic persistent infections that can hinder host immune responses as a result of evasion and subversion strategies [11]. In support of this idea, clinical observations suggest that the frequency of HPV-associated cancers is increased in immunosuppressed patients [12, 13].
HPV-associated cancers are generally non-productive infections in which the viral E6 and E7 oncoproteins are abnormally expressed. In contrast, the timing and induction of E6 and E7 expression are tightly controlled during productive HPV infection, as the infected cells migrate towards the epithelial surface where the sequential appearance of the E4, L2, and L1 viral proteins allows virus release [14]. Although HPV oncoproteins are primarily responsible for the initiation and progression of cancer, only a small percentage of people infected with high-risk HPVs develop cancers. This indicates a contribution for host factors in HPV malignancy, although their mechanisms remain poorly understood [1].
One possible mechanism is that the changes in host cell signaling pathways that occur during disease progression arise from deregulated E6/E7 oncoprotein expression, which increases the risk of transformation [15]. HPV integration found in many HPV-positive carcinomas predominantly results in deregulated oncoprotein expression. Integration has also been associated with several recurrent genomic alterations and the elevated expression of host cell genes adjacent to the integration site [16]. Additionally, the oncoproteins of high-risk HPV types have been proposed to induce genomic instability through subversion of the cellular functions involved in DNA damage and repair responses [11, 17]. In addition, host factors might confer high susceptibility for the development of HPV-associated cancers, as cervical intraepithelial neoplasias and cancers related to carcinogenic HPV infection have been linked to infection of the cervical reserve cells [18] and/or the discrete population of cuboidal cells, which are located in the cervical squamocolumnar junction and express a unique panel of genes [5, 19]. Moreover, severe HPV-associated pathogenesis (e.g. persistent verrucosis, dysplasia, neoplasia, some cutaneous squamous cell cancers, and a high prevalence of genital cancer) is an underappreciated major manifestation of some primary immunodeficiencies and investigation of these clinical developments has provided clues to the host risk factors involved [20].
The wart, hypogammaglobulinemia, infection, and myelokathexis (WHIM) syndrome is associated with inherited gain-of-function mutations in the CXCR4 gene that encodes a receptor for the CXCL12 chemokine [21]. Binding of CXCL12 to CXCR4 triggers typical activation of Gαi protein-dependent pathways of a chemokine receptor that are regulated in a timely manner by β-arrestins, which preclude further G protein activation (i.e., desensitization) and also link CXCR4 to additional signaling pathways involved in cytoskeleton reorganization and anti-apoptotic signaling [22, 23]. In WHIM, the CXCL12/CXCR4 signaling pathways manifest by abnormally increased and prolonged G protein- and β-arrestin-dependent responses associated with an impaired desensitization of CXCR4. Such dysfunction are responsible for the characteristic panleukopenia [24] in WHIM patients and likely also account for the severity of HPV disease through mechanisms that involve target cells and systemic immunity. From an immunological perspective, the unprecedented remission of HPV-induced warts in a WHIM patient after spontaneous partial inactivation of CXCR4 and subsequent restoration of immune function [25] suggests a role for myeloid cells in HPV life-cycle in support of the anomalies observed in WHIM patients’ myeloid cells [26]. However, whether it indicates that myeloid cells participate to host defense against HPV or, that they contribute to HPV-induced disease, when altered in WHIM patients, as reported in instances of chronic inflammation [27], remains unknown. Evidence for the involvement of the CXCL12/CXCR4 pair in the HPV life cycle arose from the abnormal and specific expression of CXCL12 observed in keratinocytes of HPV-productive skin or mucosal lesions regardless of whether patients suffer from WHIM [28]. Expression levels of CXCL12 and its receptors, which increase in keratinocytes as a consequence of HPV genome expression, generate an autocrine signaling loop essential for keratinocyte proliferation and migration. This interplay is involved in the WHIM-associated gain-of-function CXCR4 mutant, which confers transforming capacity on HPV18-immortalized keratinocytes in mice [29]. This further supports the hypothesis that dysfunction of the CXCL12/CXCR4 signaling pathway contribute to the pathogenesis of HPV-associated cancer.
However, the mechanism accounting for this process and whether it affects viral replication is not known. Here, we investigated a possible role for the WHIM-associated gain-of-function mutant receptor CXCR41013 in the HPV life cycle using HPV18 in the context of three-dimensional organotypic raft cultures of keratinocytes, the sole replicative model for HPV [30]. Our data indicate that the CXCR41013 mutant receptor shifts the HPV18 life cycle toward carcinogenesis, which is reflected in a viral gene expression pattern that favors oncoprotein stabilization. Blockade of either the CXCL12/CXCR4 axis or downstream effectors restored the productive HPV life cycle. Thus a main finding is that the WHIM-associated mutant CXCR4 receptor has a keratinocyte-intrinsic effect on HPV life cycle, supporting the idea that not all the effects of this mutation are mediated through dysregulation of the immune system. Besides, our results demonstrate an important function for the CXCL12/CXCR4 axis in the control of keratinocyte proliferation and its role in triggering transformation by HPV upon signaling imbalances resulting from cell hyperproliferation and elevated levels of oncoprotein-induced signaling.
Expression of CXCR41013 but not the wild-type CXCR4 receptor (CXCR4wt) confers primary human keratinocytes immortalized by HPV18 with transforming capacity, such that they develop solid tumors in nude mice [29]. We therefore explored a role for CXCR41013 in the HPV vegetative cycle in experimental models. As HPV properly replicates only within stratified epithelium, we set up three-dimensional organotypic cultures (raft cultures) that form dermal and epidermal layers resembling those in skin. These cultures support the productive HPV life cycle because the infected epithelial cells differentiate during their migration towards the epithelial surface. We used spontaneously immortalized human keratinocytes (NIKS cells) that grow normally, and can undergo terminal differentiation when cultured in rafts [31]. In NIKS cells the expression levels of ectopically expressed wt and mutant CXCR4 receptors (CXCR4wt or CXCR41013) were in the same range at the RNA and protein levels (S1 Fig, panels A, B and C). Viral genome copy numbers were similar among the various NIKS cells in monolayer (i.e. expressing the endogenous CXCR4wt solely or with either CXCR4wt or CXCR41013 ectopically expressed receptors) and were increased in the same range upon differentiation of NIKS cells in organotypic cultures (S2 Fig, panel A). Viral transcript levels (E6E7 and E2) were also similar in NIKS cells expressing the exogenous wt or mutant forms of the receptor (S2 Fig, panel B). The global architectures and subcellular topologies of raft cultures expressing CXCR41013 or CXCR4wt (CXCR41013- or CXCR4wt-rafts) was apparently unchanged within CXCR41013-expressing raft cultures (Fig 1, panel A) as well as the expression pattern of CXCR4 receptors (endogenous CXCR4 and ectopically expressed wt and mutant forms) in basal and supra-basal layers (S1 Fig, panel D). CXCR41013- and CXCR4wt-rafts also exhibited normal expression patterns for keratin 10 in the spinous and granular layers, and the keratin filament-associated filaggrin protein in the granular layer, which are two prominent markers of epidermal differentiation (Fig 1, panels B-C). In contrast, replication of the viral genome was strongly diminished in CXCR41013-raft cultures compared to that in CXCR4wt-rafts (Fig 2, panel A). Consistent with its role in HPV replication [32], production of the E2 viral protein was also lower in CXCR41013-raft cultures (Fig 2, panel B). Production of viral proteins involved in the completion of the HPV life cycle in the upper epithelial layers, were also dramatically reduced (L1) or nearly undetectable (E4) in CXCR41013-rafts compared to their levels in CXCR4wt-rafts (Fig 2, panels C-D). The E4 expression pattern in native NIKS raft cultures expressing endogenous CXCR4 (S3 Fig) was comparable to that observed in CXCR4wt-raft cultures (Fig 2, panel C), further supporting the relevance of the CXCR4wt-raft model. We have then searched for the biosynthesis of infectious viral particles, which is the final step in the HPV life cycle. We found that CXCR4wt-rafts are producing virions, which were infectious in an HaCaT cell infection assay as detected by the presence of the HPV spliced E1E4 transcript [33], while the CXCR41013-rafts did not contain detectable infectious virions (S4 Fig). These data were correlated with the presence of koilocytic cells in the intermediate layers of the CXCR4wt-rafts (e.g. Fig 1, panels A-B). Collectively, the dramatic lowest production of E2, L1, and E4 proteins in CXCR41013-rafts together with the absence of any detectable infectious virions strongly suggest that this cell environment restricts replication and the productive HPV life cycle.
To gain further insight into the consequences of CXCR41013 expression on HPV life cycle, we analyzed expression of the E6 and E7 viral oncogenes and their surrogate markers. Western blot analyses of raft cultures showed that E6 and E7 protein production in CXCR41013-rafts was significantly higher than in CXCR4wt-rafts (Fig 3, panel A and control experiments for antibodies specificity in S5 Fig). Since we were unable to detect E7 and E6 proteins in raft cultures by immunohistochemistry because of high background, we used surrogate markers to investigate the expression of these proteins. E7 is known to disrupt proteins involved in cell cycle progression, resulting in substantial induction of a functionally inactive form of cyclin-dependent kinase inhibitor 2A (p16), which can be used as an indicator of deregulated E7-expression and HPV-associated dysplastic and neoplastic lesions [9]. The minichromosome maintenance (MCM) family of cell cycle proteins are essential for eukaryotic DNA replication, and MCM7 and MCM2 are widely used as molecular surrogates of E6/E7 expression and E6/E7-mediated cell cycle entry [34]. In agreement with this, MCM2 and p16 expression levels were higher in CXCR41013-rafts than in CXCR4wt-rafts. Compared to the limited distribution of p16 staining in basal layers, MCM2 staining was more widespread and intense, and extended into the upper epithelial layers (Fig 3, panels B-C-D). The significantly higher levels of oncogene expression observed in CXCR41013-rafts, were not paralleled by neither higher E6/E7 transcript levels nor changes in E2 transcripts levels (S6 Fig, panel A). This was consistent with our inability to detect any HPV genome integration (S6 Fig, panel B) or enhanced transcriptional activity of the HPV long control region (LCR) (S6 Fig, panel C) in CXCR41013-rafts.
To determine whether the E6 and E7 oncoproteins were stabilized in CXCR41013 cells, we quantified E6 and E7 protein levels over time in CXCR4wt- and CXCR41013-expressing keratinocytes induced to differentiate in medium containing a high concentration of calcium. This model permits the activation of late events and the productive phase of the HPV life cycle after 48−96 h of culture [35]. In NIKS cells differentiated for 96 h in high-calcium medium, E6 and E7 protein decay after 2 h of cycloheximide treatment was lower in CXCR41013 keratinocytes than in CXCR4wt keratinocytes, resulting in higher relative levels of E6 and E7 in CXCR41013 keratinocytes (S7 Fig, panel A). In contrast, in undifferentiated NIKS cells we observed no significant difference in the stability of oncogenes (S7 Fig, panel B). These results suggest that the viral oncoproteins were stabilized in the presence of CXCR41013 in differentiated cells.
An elevated proliferation rate is an early step in viral oncogenesis [8]. The Ki-67 antigen, which is expressed in all phases of the cell cycle except in G0, is widely used to assess proliferation and represents a biomarker for cervical cancer [34]. Immunohistochemical analyses of Ki-67 in raft cultures indicated higher levels of cellular proliferation in the basal and suprabasal compartments of CXCR41013 rafts than in those of CXCR4wt rafts, and a higher overall proportion of (Ki-67-positive) cells in cycle (Fig 4 panels A, C). Keratinocyte proliferation is normally restricted to the basal layers in CXCR41013-rafts in the absence of HPV infection (S8 Fig), indicating that CXCR41013 does not deregulate cell proliferation on its own but rather acts synergistically with HPV. Given the capacity of E6 to promote degradation of the p53 protein, a gatekeeper of aberrant cell cycle progression involved in cell cycle arrest and apoptosis [36], we quantified apoptotic cell numbers in raft sections by TUNEL assay and p53 protein levels by western blot (Fig 4). There were fewer TUNEL-positive apoptotic cells in sections of CXCR41013 rafts than in CXCR4wt rafts suggesting that the overall proportion of apoptotic cells can significantly lower in CXCR41013 rafts (Fig 4, panels B-C). Consistent with these results, p53 levels were significantly lower in CXCR41013-rafts than in CXCR4wt rafts (Fig 4, panel D). These results demonstrate that CXCR41013 plays an important role in driving the HPV18 viral life cycle toward carcinogenesis, notably by increasing the levels of the HPV oncoproteins.
HPV, like other viruses associated with human cancers, was recently shown to hijack DNA damage response pathways responsible for maintaining genomic integrity. Proteins involved in these pathways include the ataxia telangiectasia mutated (ATM) and Rad3-related protein kinases, as well as p53 and the CHK1 and CHK2 kinases, which are involved in downstream checkpoint pathways [17]. Activation of the ATM/CHK pathway in the course of HPV infection provides a suitable environment for viral replication in differentiated cells [37]. The mechanisms of this activation, which may be part of the HPV replication process itself, are not completely understood but may involve the E1, E2, and E7 proteins [38]. To investigate whether the ATM/CHK pathway can be differentially modified in a CXCR41013-expressing background as a result of altered viral replication or higher E7 expression levels, we analyzed the expression of the ATM and CHK proteins in keratinocytes differentiated in high-calcium medium. Progression of normal differentiation was indicated by increases in involucrin protein abundance over time in CXCR41013- and CXCR4wt-expressing NIKS cells cultured in high-calcium medium (Fig 5, panel A). In differentiated HPV18-positive NIKS cells, E6 protein levels were higher in cells expressing CXCR41013 than in those expressing CXCR4wt (Fig 5, panel B). These results confirmed and extended our findings in raft cultures (Fig 3). Confirming our results in raft cultures (Fig 4), p53 protein levels at 96 h were lower in CXCR41013-expressing cells than in CXCR4wt-expressing cells (Fig 5, panel A), which was likely related to the higher levels of E6 protein in CXCR41013-expressing cells. The expression patterns of ATM and its activated phosphorylated form (pATM) in control cultures (Fig 5, panel C, CXCR4wt-cells) were concordant with previous report [37]. After 96 h of culture, pATM levels were lower in CXCR41013-expressing cells than in CXCR4wt-expressing cells (Fig 5, panel C). Accordingly, the levels of the activated phosphorylated form of CHK2 (pCHK2) kinase were significantly lower in keratinocytes expressing CXCR41013 than in those expressing CXCR4wt at 96 h but also at 48 h (Fig 5, panel C). This correlates with the tendency of CXCR41013-expressing cells to have lower levels of activated ATM. Altogether these results suggest that the presence of CXCR41013 might lead to suboptimal activation of the ATM pathway and a reduced ability to support HPV replication.
The suboptimal HPV replication environment in keratinocytes expressing CXCR41013 prompted us to examine the consequences of normalizing the CXCR41013-enhanced signaling. One of the key pathways accounting for the CXCR41013 gain-of-function is the enhanced β-arrestin-mediated signaling, which is dependent upon the third intracellular loop (SHSK motif) of CXCR4 [39], which is also required for mobilization of intracellular calcium and G-protein-independent stimulation of JAK2/STAT3 in response to CXCL12 [40]. Therefore, we expressed CXCR41013 lacking the SHSK motif, CXCR41013&ΔSHSK in keratinocyte raft cultures (Fig 6). In leukocytes, CXCR41013&ΔSHSK exhibited normal β-arrestin–mediated signaling and CXCL12-induced chemotaxis [39]. In raft cultures expressing CXCR41013&ΔSHSK, E2 protein levels were significantly higher than in CXCR41013-rafts (Fig 6, panel A). E4 and L1 proteins were detected in the upper epithelial layers of the CXCR41013&ΔSHSK-rafts (Fig 6, panels B-C). Proliferating cells expressing the Ki-67 antigen remained confined to basal and suprabasal layers (Fig 6, panel D), as in CXCR4wt raft (Fig 4, panel A). Thus, CXCR41013&ΔSHSK, with normal β-arrestin-mediated signaling, lacks the ability to impair virus production, further supporting the capacity of CXCR4 to tune the viral life cycle through its downstream signaling pathways.
To further assess the role of CXCR4 activity in the HPV life cycle, CXCR4wt- and CXCR41013-rafts were treated with AMD3100, a selective and competitive antagonist of CXCR4 [41] that efficiently blocks CXCR41013 function [28]. The architecture and viral expression patterns of raft cultures treated with AMD3100 were compared to control ones (untreated rafts) by immunohistochemistry and western blot analyses (Fig 7). The stratified epithelium in CXCR4wt and CXCR41013-rafts treated with AMD3100 was thinner than in untreated rafts (Fig 7, panel A) but keratinocytes differentiation was apparently not affected given the normal expression pattern for keratin 10 (S9 Fig panel A). E4 and L1 proteins were readily detected in the upper layers of CXCR41013-rafts treated with AMD3100 (Fig 7, panel B) as in CXCR4wt-rafts treated with AMD3100 (S9 Fig panel B), while sparsely detected (L1) or undetectable (E4) in untreated CXCR41013-raft cultures (Fig 7, panel B). In contrast, AMD3100 treatment reduced expression of the E6 and E7 oncoproteins in CXCR41013- and CXCR4wt-rafts (Fig 7, panel C and D, respectively). Thus blocking CXCR41013-dependent signaling allows virus production, while reducing oncogene-expression and–driven neoplastic-like changes. Moreover, we found that tumors produced by injecting CXCR41013-expressing human keratinocytes immortalized by HPV18 [29] into nude mice were significantly smaller in AMD3100-treated mice than in untreated mice (Fig 8). Collectively, these results indicate that CXCL12/CXCR4 signaling impacts the balance between HPV replication and HPV-driven carcinogenesis.
Considering the role of the CXCL12/CXCR4-signaling in the migration and survival of HPV18-infected keratinocytes, we investigated the importance of this axis in the productive HPV life cycle in which the timely and coordinated expression of different viral genes occurs as infected cells move toward the epithelial surface where virions mature. In view of the consequences of axis deregulation in HPV-induced cell transformation via the CXCR41013 gain-of-function mutant, we investigated the impact of CXCR41013 expression on the productive HPV life cycle in an organotypic model of human epidermis.
Our results demonstrate that unregulated CXCR41013 function fosters a keratinocyte environment that restrains the productive HPV18 life cycle in contrast to CXCR4wt-expressing rafts that efficiently reproduce the complete HPV life cycle including the production of infectious virions. These findings in CXCR41013-rafts correlate with deregulated keratinocyte proliferation and the stabilization of the E6 and E7 oncoproteins together with dramatic decreases in production of the late viral proteins involved in HPV virion assembly. These results are especially relevant because CXCL12 is expressed in keratinocytes from HPV-infected raft cultures (S10 Fig). This extends previous studies detecting CXCL12 expression in epidermal keratinocytes from HPV-induced lesions but not from other skin pathologies and normal skin [28, 42]. Thus collectively, our results support the concept that the CXCL12/CXCR4 pathway controls the HPV productive life cycle, likely reflecting the normal function of this pathway as a regulator of keratinocyte proliferation and survival. When deregulated, as in the WHIM syndrome, the CXCL12/CXCR4 pathway can trigger HPV-induced transformation as a result of elevated levels of oncoprotein-induced signaling and down-regulation of DNA damage response pathways associated with cell hyperproliferation.
On the one hand, the severe pathogenesis in WHIM patients manifests as intractable genital warts that often develop into severe dysplasia and carcinoma [43, 44]. This pathogenesis is generally due to mucosal high-risk HPV types for which we can suggest that expression of CXCR41013 might enhance transforming capacity, partly through the elevation of E6/E7 proteins expression that drive proliferation of the infected cells, although this remains to be investigated directly in patients-derived cells. In some cases, patients’ dysplasia was found to be associated with low-risk HPV types, such as HPV6, which display potential oncogenicity [45]. On the other hand, the profusion and persistency of cutaneous warts is another major clinical manifestation of the WHIM-associated HPV pathogenesis. In this regard, it can be postulated a role for CXCR41013 in driving the initial proliferation of the infected cells, thus allowing expression and replication of cutaneous low-risk HPV types, which do not normally stimulate proliferation. Our study is providing the rational for future mechanistic investigations of the productive life cycle of cutaneous low-risk HPV, which remains underappreciated due to the lack of robust in vitro models.
We have found that CXCR41013 expression in keratinocyte raft cultures is associated with the stabilization of the E6 and E7 oncoproteins. This was not associated with an integration of the HPV genome further supporting that disruption of the E2 ORF is not the only mechanism of suppressing E2 and increasing E6 and E7 expression as previously reported in HPV16-induced carcinogenic progression [46]. Such modulations of oncoproteins expression might involve the ubiquitin-proteasome system, which is involved in the degradation of E6 produced by the high-risk types HPV18 and HPV16 [47]. Additionally, the capacity of E6 to interact with certain PDZ domain proteins and phosphoserine-binding proteins involved in cell signaling pathways is a mechanism for regulating E6 stability that is common to diverse high-risk HPV types [48–50]. Beside PDZ domain proteins, chemokine receptors including CXCR4 can interact with chaperone proteins, some being recently found to increase the steady-state levels and half-life of E6 and E7 oncogenes [51, 52]. As the interaction of E6 with either of these protein families depends on its phospho-regulation by various kinases (e.g. protein kinase A or B) and determines the fate of E6, different environmental conditions might have a significant impact on the likelihood of HPV infection progressing toward malignancy [53]. We propose that changes in cell signaling pathways in the context of CXCR41013 expression, and notably in the downstream kinases activated by CXCL12-CXCR4 signaling may differentially affect the stability of HPV18 E6. Some kinases, as well as the rate of ubiquitination, can also control the steady-state level of E7 by interfering with proteasome-dependent degradation, but these processes have been studied in the context of only a few HPV types (HPV16 and HPV6) [54, 55].
Increased production of E6 and E7 proteins in the context of CXCR41013 expression makes rafts prone to drive the viral lifecycle toward carcinogenesis as demonstrated by the altered levels of keratinocyte proliferation and apoptosis and by a disturbance in the ordered expression of viral gene products that normally leads to virus replication and production. Whether the fact that rafts derive from the spontaneously immortalized human NIKS keratinocytes might contribute to this process is not known and is awaiting the setting of rafts from primary human keratinocytes for HPV life cycle modeling.
Previous studies have clearly suggested that an increase in high-risk HPV protein levels can drive a more severe neoplastic phenotype [56]. Among these viral proteins, we observed a dramatic decrease in the level of E2 that might be related to enhanced degradation by the ubiquitin-proteasome system, which controls viral protein stability and was proposed to operate in cycling cells [57]. Aberrant cell cycle progression in CXCR41013-rafts might thus increase E2 degradation and diminish its activity in the late phases of the viral life cycle. Conversely, normalizing arrestin-dependent signaling downstream of CXCR41013 (CXCR41013&ΔSHSK) might stabilize E2 protein thus partially restoring the productive HPV life cycle. This shift toward viral production was revealed by the enhanced production of L1 and E4, which are primarily involved in genome packaging and virus release [1, 58, 59]. The stabilization of E2 in CXCR41013&ΔSHSK-rafts might also be accounted for the physical and functional interaction of E2 and E4 as the level of each protein is increased by the presence of the other [60].
Additionally, and consistent with the recently reported essential role of DNA damage responses in the viral replication [37], ATM/CHK pathway activation was found to be significantly lower in CXCR41013-rafts. The contribution of ATM/CHK pathway activation to viral replication has begun to be deciphered but its function in HPV-induced cancer development remains unclear, especially because E7 and E6 have important roles in promoting this activation [38, 61]. Decreased activation of the ATM/CHK proteins in CXCR41013-rafts, in spite of increased expression of E6 and E7, may appear paradoxical. However, such deregulation might include the JAK/STAT pathway, which is induced downstream of CXCR4 in a G protein-independent manner [62] and was shown to activate the ATM-dependent DNA damage responses [38]. Although the biological role of DNA damage responses in HPV-induced malignancy is still uncertain, decreasing the production or activation of ATM/CHK proteins would likely lead to the accumulation of DNA damage in CXCR41013-rafts.
In cervical disease, it is thought that the levels of E6 and E7 rise with cervical intraepithelial neoplasia (CIN) severity. Changes in gene expression underlie the neoplastic progression, with CIN1 lesions supporting the complete HPV life cycle in contrast to CIN3 lesions that are considered to be high-grade precancerous. We provide evidence that blocking CXCL12/CXCR4-dependent signaling in the course of raft culture differentiation allowed the productive HPV life cycle to proceed at the expense of the HPV-induced carcinogenesis in CXCR41013-rafts. These results provide molecular clues to the potential therapeutic effect of AMD3100 treatment for skin warts when combined with imiquimod [63] but also for HPV-induced carcinogenesis in a mouse preclinical model [64] and here in nude mice after injection of human keratinocytes immortalized by HPV18 and expressing CXCR41013. Whereas we think that the interplay between the HPV life cycle and CXCL12/CXCR4 in keratinocytes is the direct mediator element, the beneficial effect of AMD3100 might also be related to other components that are controlled by this signaling axis (e.g. resident and infiltrating immune cells in the skin, stem cell recruitment, or endothelial cell responses). The CXCL12/CXCR4 pair is indeed involved in the increased survival and/or proliferation of cancer cells from various types including virus-related cancers, as well as in the promotion of tumor metastasis and angiogenesis linked to tumor progression [65–69].
In light of the HPV-pathogenesis associated with the WHIM syndrome, it can be extrapolated that dysfunction of CXCR4 might be acquired from genetic errors accumulated during the multistep process of HPV-induced neoplasia, as demonstrated by the somatic WHIM-like CXCR4 mutations reported in Waldenström macroglobulinemia [70] or the anomalies in the effectors of the CXCL12/CXCR4-signaling pathway reported in patients with GATA2-deficiency [71–73]. Clues to the additional effectors of this potentially pathogenic process arise from the abnormal expression of CXCL12 in HPV-lesions in the general population of HPV infected individuals [28] and from the presence of CXCL12 (our data) among the panel of expressed genes unique to the discrete population of cuboidal cells located in the cervical squamocolumnar junction, which have been linked to HPV-related cervical intraepithelial neoplasia and cancers [5]. Consequently, the CXCL12/CXCR4 signaling pathway appears to be an important host factor in HPV-induced pathogenesis.
NIKS cells, near-diploid spontaneously immortalized human keratinocytes [31] (kindly provided by Dr Paul F. Lambert), were maintained at subconfluence on mitomycin C-treated 3T3-J2 feeder cells (kindly provided by Dr Paul F. Lambert) in F medium with all supplements as previously described [30]. Human foreskin fibroblasts (kindly provided by Dr Paul F. Lambert) were cultured in Ham’s F12 medium containing 10% fetal bovine serum and 1% penicillin-streptomycin, before use in raft cultures.
NIKS cells expressing CXCR4wt, CXCR41013 or CXCR41013&ΔSHSK were obtained with a lentivirus-mediated strategy as previously described [29, 39]. Expression of similar levels of each receptor in the different cell populations was checked by flow cytometry (S1 Fig, panel B). Recircularized HPV18 DNA was prepared as previously described [30]. The different NIKS cell populations (2.5 x 106 cells plated the day before the transfection) were cotransfected with 2 μg of recircularized HPV18 DNA and 0.5 μg of the blasticidin-resistance plasmid pcDNA6 using Effectene Transfection Reagent (QIAGEN). Blasticidin selection (7 μg/mL) was performed for 6 days.
HPV18-positive NIKS cells expressing CXCR4wt, CXCR41013, or CXCR41013&ΔSHSK were grown in raft cultures to induce the three-dimensional architecture of the stratified epithelium, as described previously [30]. Briefly, 1.5 x 106 NIKS cells were seeded onto a dermal equivalent composed of rat-tail collagen type 1 containing 1 x 106 human foreskin fibroblasts. Raft cultures were lifted onto transwell inserts submerged in deep well plates containing keratinocyte plating medium and cultured for 4 days. The transwell inserts were then raised by placing four cotton pads underneath them, thereby exposing the epithelial cells to the air-liquid interface. Raft cultures were fed by diffusion from below with cornification medium and were allowed to stratify for 14 days. When specified, AMD3100 (A5602, Sigma-Aldrich) was added to the cornification medium at a concentration of 20 μg/mL from day 4 to the end of the experiment. Rafts were then removed from transwell inserts and either fixed in formalin and embedded in paraffin for histological analyses, or frozen at −80°C for quantitative real-time PCR and western blot analyses.
To induce differentiation in medium containing 1.5 mM CaCl2 (high calcium), HPV18-positive NIKS cells expressing either CXCR4wt or CXCR41013 were cultured in absence of 3T3-J2 feeders in F medium for 24 h and then switched to F medium (without growth supplements) containing 1.5 mM CaCl2. Cells were then harvested at 0, 48, and 96 hours for protein extraction.
NIKS cells from monolayer cultures or rafts cultures were resuspended in protein lysis buffer (1% Triton X-100, 10 mM Tris-HCl pH 7.4, supplemented with Protease and Phosphatase Inhibitor Tablets (Pierce)). Protein concentration was measured with the BCA Protein Assay Kit (Pierce) according to the manufacturer’s protocol. Equivalent amounts of protein were separated on a SDS-polyacrylamide gel and transferred to a PVDF membrane. Primary antibodies were as follows: anti-HPV18-E6 (kindly provided by Arbor Vita Corporation), anti-HPV18-E7 (sc-1590, Santa Cruz), anti-HPV18-E2 (kindly provided by Dr. F. Thierry), anti-p53 (sc-126, Santa Cruz), anti-involucrin (I9018, Sigma-Aldrich), anti-GAPDH (14–9523, eBioscience); anti-pATM (Ser1981; #13050), anti-ATM (#2873), and anti-pCHK2 (Thr68) and anti-CHK2 (#2661 and #3440, respectively, Cell Signaling). Membranes were incubated with the appropriate secondary antibodies conjugated to HRP (GE Healthcare). Proteins were detected using the Immobilon Western Chemiluminescent HRP kit (Millipore).
Raft paraffin sections (5-μm thick) were stained with hematoxylin and eosin (HE), and with safran (HES) where indicated. Immunohistochemistry was performed on paraffin sections using primary antibodies for keratin 10 (MA1-35540, Thermo Scientific), filaggrin (VP-F706, Vector laboratories), HPV18-E2 (provided by Dr. F. Thierry), HPV18-E4 (provided by Dr. J. Doorbar), HPV18-L1, p16 (sc-56330, Santa Cruz), Ki-67 (18-0191Z, AbCys), and CXCL12 (K15C clone, MABC184, EMD Millipore). Bound antibodies were detected using the LSAB+/HRP kit (K0679, Dako) or the AEC+ High Sensitivity Substrate Chromogen Ready-to-Use System (K3461, Dako). For immunofluorescence, staining with the primary antibody for MCM2 (ab31159, Abcam) was followed by staining with a goat anti-rabbit Alexa Fluor 596 (Invitrogen). Tissues were counterstained with DAPI. For the TUNEL assay to detect apoptotic cells, we used the In Situ Cell Death Detection Kit, Fluorescein (11684795910, Roche Diagnostics) according to the manufacturer’s instructions. In situ hybridization was performed with the wide spectrum HPV biotinylated DNA probe sets able to detect 11 types of anogenital HPV (In Situ Hybridization Detection System, K0601, Dako).
Where indicated, slides were scanned by the digital slide scanner NanoZoomer 2.0-RS (Hamamatsu) allowing an overall view of the samples. Images were digitally captured from the scanned slides using the NDP.view2 software (Hamamatsu). All quantifications from the histological analyses were performed by counting 10 different fields on the scanned slides. Other slides were analyzed using a Leica DMLA microscope, in particular to visualize images at 100X magnification, and images were captured with a Leica DFC450 C digital microscope camera. Immunochemical stainings were interpreted simultaneously and independently by at least two investigators (FM, LC, or FG).
HK-HPV18-CXCR41013 tumors were established in nude mice as previously described [29]. Briefly, athymic female nude nu/nu 5-week-old mice (Harlan Laboratories) were injected subcutaneously with 2 x 107 HK-HPV18 cells expressing T7-GFP-tagged CXCR41013 in the right flank (six to seven mice per group). Mice were treated with 5 mg/kg of AMD3100 (intraperitoneal administration) on days 8, 13, 16, and 20 after tumor cell injection. Tumor volumes (V) were calculated as V = π/6 x (length x width2).
All experimental procedures were conducted in our animal facility (agreement n° B 92-023-01) in accordance with the European Union’s legislation and the relevant national legislation, namely the French “Décret no 2013–118, 1er février 2013, Ministère de l’Agriculture, de l’Agroalimentaire et de la Forêt” regarding the use of laboratory animals and were approved by the Committee on the Ethics of Animal Experiments (Comité d'Ethique en Expérimentation Animale Capsud or CEEA-26) under the authorization 2014_039 #2521.
Student’s t test was used to compare the significance between specified groups. All analyses were performed with GraphPad Prism software.
|
10.1371/journal.pgen.1002594 | PIF4–Mediated Activation of YUCCA8 Expression Integrates Temperature into the Auxin Pathway in Regulating Arabidopsis Hypocotyl Growth | Higher plants adapt their growth to high temperature by a dramatic change in plant architecture. It has been shown that the transcriptional regulator phytochrome-interacting factor 4 (PIF4) and the phytohormone auxin are involved in the regulation of high temperature–induced hypocotyl elongation in Arabidopsis. Here we report that PIF4 regulates high temperature–induced hypocotyl elongation through direct activation of the auxin biosynthetic gene YUCCA8 (YUC8). We show that high temperature co-upregulates the transcript abundance of PIF4 and YUC8. PIF4–dependency of high temperature–mediated induction of YUC8 expression as well as auxin biosynthesis, together with the finding that overexpression of PIF4 leads to increased expression of YUC8 and elevated free IAA levels in planta, suggests a possibility that PIF4 directly activates YUC8 expression. Indeed, gel shift and chromatin immunoprecipitation experiments demonstrate that PIF4 associates with the G-box–containing promoter region of YUC8. Transient expression assay in Nicotiana benthamiana leaves support that PIF4 directly activates YUC8 expression in vivo. Significantly, we show that the yuc8 mutation can largely suppress the long-hypocotyl phenotype of PIF4–overexpression plants and also can reduce high temperature–induced hypocotyl elongation. Genetic analyses reveal that the shy2-2 mutation, which harbors a stabilized mutant form of the IAA3 protein and therefore is defective in high temperature–induced hypocotyl elongation, largely suppresses the long-hypocotyl phenotype of PIF4–overexpression plants. Taken together, our results illuminate a molecular framework by which the PIF4 transcriptional regulator integrates its action into the auxin pathway through activating the expression of specific auxin biosynthetic gene. These studies advance our understanding on the molecular mechanism underlying high temperature–induced adaptation in plant architecture.
| Exposure of Arabidopsis to high temperature (29°C) results in a dramatic hypocotyl elongation. The basic helix-loop-helix transcription factor PIF4 and the phytohormone auxin play essential roles in high temperature–mediated induction of Arabidopsis hypocotyl elongation. However, the possible molecular linkage between PIF4 and the auxin pathway in regulating high temperature–induced adaptative growth remains unknown. Here, we report that high temperature–induced elevation of YUCCA8 (YUC8) transcripts and endogenous free IAA levels is dependent on the function of PIF4. In particular, we provide evidence that PIF4 directly activates the expression of YUC8 to upregulate auxin biosynthesis, as a consequence, achieves high temperature–induced hypocotyl elongation. In addition, we found that SHY2/IAA3 is an important component of the PIF4–auxin pathway in regulating high temperature–induced hypocotyl elongation. Overall, our results establish a direct connection between the PIF4 transcription factor and the auxin pathway in regulating high temperature–induced adaptation growth.
| Higher plants continually sense environmental conditions to adapt their growth and development. To a large extent, this is achieved through integrating environmental cues into the growth-regulating hormonal pathways. Exposure of Arabidopsis thaliana plants to high temperature (29°C) results in dramatic plant architecture changes including rapid hypocotyl elongation, leaf hyponasty, and early flowering [1]–[4]. High temperature-induced hypocotyl elongation of Arabidopsis plants provides an ideal model system to investigate the regulatory mechanisms underlying adaptive growth of plants to their ever-changing environments. Among the endogenous cues involved in the regulation of high temperature-induced hypocotyl elongation is the plant hormone auxin [3]. An early observation revealed a correlation between high temperature-induced hypocotyl elongation and high temperature-induced elevation of endogenous free indole-3-acetic acid (IAA) levels [3]. Genetic analyses found that high temperature-induced hypocotyl elongation is sharply reduced in Arabidopsis mutants defective in auxin biosynthesis, transport or signaling [3]. Together, these data attribute an essential role of the auxin pathway in mediating high temperature-induced hypocotyl elongation.
It is long-recognized that auxin has profound effects on plant growth and development. A combination of physiological, biochemical, pharmacological and molecular genetic studies provide an ever-growing body of insights on our understanding of the auxin biosynthesis pathway [5], [6]. It is generally believed that, IAA, the main auxin in higher plants, can be synthesized from tryptophan (Trp)-dependent and -independent pathways [5]. Among the best-characterized enzymes involved in the Trp-dependent auxin biosynthetic pathway are the YUCCA (YUC) family of flavin-containing monooxygenases [5], [7]–[9] and the TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS1/TRYPTOPHAN AMINOTRANSFERASE-RELATED (TAA1/TAR) family of aminotransferases [5], [10], [11]. A wealth of genetic evidence indicated that, while inactivating members of the YUC family genes causes dramatic developmental defects [8], [9], overexpression of the YUC family genes leads to auxin overproduction and long hypocotyl phenotype in Arabidopsis [7]. Although mutation of TAA1 or its close homologs (TAR genes) leads to developmental defects similar to those of the yuc mutants [10], [11], overexpression of TAA1/TAR does not cause obvious developmental phenotype, suggesting that TAA1/TAR probably do not catalyze a rate-limiting step in IAA biosynthesis [10], [11]. Interestingly, recent studies provide evidence that TAA1/TARs and YUCs may act in a common linear biosynthetic pathway for auxin production [6], [12], [13].
In addition to auxin, a family of phytochrome-interacting factors (PIFs), which encode basic helix-loop-helix (bHLH) transcription factors, have been shown to be central integrators of versatile environmental and hormonal signals during plant adaptive growth [14], [15]. Among the PIF family of transcriptional regulators, a selective function of PIF4 in high temperature-induced hypocotyl elongation has recently been reported [1], [16]. These studies revealed that high temperature induced a rapid elevation of PIF4 transcript levels and that the pif4 mutant largely lost the robust enhancement of hypocotyl elongation induced by high temperature [1].
In the context that both the transcription factor PIF4 and the phytohormone auxin are required for high temperature-induced hypocotyl elongation, a fascinating hypothesis is that PIF4 may directly link the auxin pathway in regulating plant adaptation growth to high temperature. We provide here evidence that, in response to high temperature, PIF4 directly activates YUC8 expression and thus elevates endogenous free IAA levels. We also show that the SHY2/IAA3 protein is a downstream component of the PIF4-auxin signaling pathway in regulating high temperature-induced hypocotyl elongation. Our results exemplify how a transcriptional regulator integrates environmental cues with endogenous hormonal signaling to mediate specialized developmental changes in regulating plant adaptive growth.
It has been shown that high temperature activates the expression of the transcription factor PIF4 [1], and elevates endogenous free IAA levels [3] in Arabidopsis. To explore the possible molecular linkage between PIF4 and the auxin pathway in regulating high temperature-mediated adaptation growth, we examined high temperature-induced expression of PIF4 and the YUCCA (YUC) family of auxin biosynthetic genes [5]. Consistent with previous reports [1], , when wild type (WT) seedlings grown at 22°C for 6 days were transferred to 29°C in continuous light over a 24 h time course, PIF4 transcript abundance was transiently elevated to a peak level at 3 h after transfer (Figure 1A). Correlating with an increased expression of PIF4, high temperature also markedly increased transcript abundance of YUC8 with a peak at 3 h in WT seedlings (Figure 1B). Closer observation with a narrower range of time points revealed that high temperature-mediated induction of YUC8 expression occurred generally later than that of PIF4 (Figure S1). Parallel experiments indicated that high temperature did not upregulate the expression of other YUCCA family genes tested (Figure S2). We then compared high temperature-induced YUC8 expression between WT and the pif4 mutant, which has been shown to be defective in high temperature-induced adaptations in plant architecture (Figure 1B). As shown in Figure 1B, the basal expression levels of YUC8 were already low in pif4 seedlings and, significantly, high temperature-induced upregulation of YUC8 expression was largely abolished in this mutant, indicating that the function of PIF4 is important for the basal- and high temperature-induced expression of YUC8.
The pif4 mutation impairs high temperature-induced upregulation of YUC8 expression suggests that this mutation may also affect high temperature-induced elevation of free IAA levels. To test this, we compared high temperature-induced elevation of free IAA levels in WT and pif4 seedlings. For these experiments, we grew seedlings at 22°C or 29°C in continuous light for 6 days and collected hypocotyls for IAA measurement. Consistent with a previous report [3], high temperature increased free IAA levels of WT seedlings by around 50% (Figure 1C). As expected, high temperature-induced elevation of free IAA levels was abolished in the pif4 mutant (Figure 1C), indicating that PIF4 is also required for high temperature-induced elevation of auxin biosynthesis. Together, these results suggest that PIF4 and YUC8 may function in linking temperature and auxin pathway in regulating hypocotyl elongation.
As a first step to test the possibility that PIF4 may directly regulate YUC8 expression during high temperature-induced adaptation growth, we examined YUC8 expression in transgenic plants overexpressing PIF4 (35S-PIF4). Like the reported yucca mutants which contain increased endogenous auxin levels [7], 35S-PIF4 plants show a long hypocotyl phenotype that resembles high temperature-grown WT seedlings (Figure S3). As revealed by quantitative reverse transcription-polymerase chain reaction (qRT-PCR) assays, the expression of YUC8 (Figure 2A), but not that of TAA1 (Figure S4), was substantially increased in 35S-PIF4 seedlings as compared to WT. We also generated PIF4-overexpression plants (pMDC7:PIF4) using the chemical inducible vector pMDC7 [17]. In the presence of the chemical inducer estradiol, pMDC7:PIF4 seedlings show increased expression of PIF4 (Figure 2B) and display a long hypocotyl phenotype like 35S-PIF4 seedlings (Figure S5). As expected, YUC8 expression was considerably elevated following estradiol induction (Figure 2C). Consistently, measurement of auxin revealed that the free IAA levels in 35S-PIF4 plants were increased by 50% as compared to those in WT plants (Figure 2D). In line with increased free IAA levels in 35S-PIF4 plants, the expression of the auxin responsive DR5:GUS, a widely used reporter of auxin response, was clearly enhanced in the basal region of 35S-PIF4 hypocotyls (Figure S6). These data together indicate that overexpression of PIF4 leads to increased expression of the auxin biosynthetic gene YUC8 and, as a result, elevated endogenous free IAA levels in planta.
Three lines of evidence support a scenario that the PIF4 transcription factor may directly regulate YUC8 expression during high temperature-induced adaptation growth. First, underlying high temperature-induced hypocotyl elongation, high temperature upregulates the expression of PIF4 in a similar fasion to that of YUC8. Second, high temperature-induced upregulation of YUC8 expression requires the function of PIF4. Third, overexpression of PIF4 leads to increased expression of YUC8 and elevated free IAA levels in planta. Given that PIF4 specifically binds to a core DNA G-box motif (CACGTG) of its target gene promoters [18], we searched for the presence of G-box motifs in the promoter regions of the 11 YUC family genes present in the Arabidopsis genome. As shown in Figure 3A, G-box motifs were found not only in the promoter of YUC8, whose expression was significantly induced by high temperature (Figure 1), but also in the promoters of YUC5, YUC9 and YUC10, whose expression was not or slightly induced by high temperature (Figure S2). To test the idea that PIF4 may actually bind to the G-box-containing regions of these YUC genes, we performed chromatin immuno-precipitaiton (ChIP) assays using a previously reported transgenic line expressing a fusion of PIF4 to the haemagglutinin (HA) antigen (PIF4-HA) [19] and anti-HA antibody (Abcam). PCR amplification of the promoter regions of the four YUC genes showed that PIF4-HA specifically bound to the G-box-containing promoter region of YUC8, but not to the G-box-containing promoter regions of YUC5, YUC9 and YUC10 (Figure 3B). These results suggest that PIF4 associates with the G-box DNA motifs in the promoter region of YUC8 in vivo. Further evidence supporting this conclusion came from electrophoretic mobility-shift assays (EMSA) using PIF4 protein expressed in vitro. As shown in Figure 3C, PIF4 bound to the G-box-containing DNA fragments present in the promoter region of YUC8 and, this binding could be effectively competed by the addition of excess amount of unlabeled G-box-containing DNA probes (Figure 3C). As a control, we showed that DNA probes containing a mutated G-box motif (CACGGG) failed to compete the binding of PIF4 to the G-box-containing DNA fragments (Figure 3C). Together, these results support that the PIF4 transcription factor regulates YUC8 expression by directly binding to its promoter region.
Next, using the well-established transient expression assay of Nicotiana benthamiana leaves, we verified the activation effect of PIF4 on the expression of a reporter containing the YUC8 promoter fused with the firefly luciferase (LUC) gene. When the pYUC8:LUC reporter was infiltrated into N. benthamiana, the LUC activity could be detected at lower level (Figure 4A, B). Coexpression of pYUC8:LUC with the 35S:PIF4 construct led to an obvious induction in luminescence intensity (Figure 4A, 4B), suggesting that ectopic expression of PIF4 can activate pYUC8:LUC expression in this transient expression assay. In a parallel experiment, pYUC8(mut):LUC, in which the two G-boxes of the YUC8 promoter were deleted and fused with LUC, togehter with 35S:PIF4 were co-infiltrated into N. benthamiana leaves. As shown in Figure 4, the activation effect of PIF4 on pYUC8(mut):LUC expression was abolished. Together, our transient expression assays in N. benthamiana leaves confirmed that PIF4 directly activates YUC8 expression in vivo.
To determine the genetic relationship between PIF4 and YUC8, we identified a yuc8 mutant (SALK_096110) which harbors a T-DNA insertion that markedly reduced the expression levels of the YUC8 gene (Figure S7). We show that the yuc8 mutant is defective in high temperature-induced hypocotyl growth (Figure 5A). We then introduced the above-described 35S-PIF4 construct into the genetic background of the yuc8 mutant through genetic crossing. As shown in Figure 5B, the yuc8 mutation substantially suppressed the long-hypocotyl phenotype of the 35S-PIF4 plants, supporting that YUC8 acts genetically downstream of PIF4 in regulating high temperature-induced hypocotyl elongation.
Several elegant observations have demonstrated the involvement of PIF4 and auxin in regulating adaptive growth of plants to high temperature [1], [16]. Our data presented here further revealed that, through directly activating of the YUC8 expression, PIF4 integrates its action into the auxin pathway in regulating high temperature-mediated hypocotyl elongation. To further identify auxin signaling components involved in this process, we employed a genetic approach to search for auxin-related mutations that can suppress the long-hypocotyl phenotype of the 35S-PIF4 plants. It has been shown that the shy2-2 mutant, which harbors a stabilized mutant form of the SHY2/IAA3 protein, displays a short hypocotyl phenotype [20], suggesting a role of SHY2/IAA3 in regulating auxin-mediated hypocotyl growth. We showed that shy2-2 seedlings are defective in high temperature-induced hypocotyl growth (Figure S3). Importantly, like the auxin signaling mutant axr1-12 (Figure S8), shy2-2 genetically suppressed the long-hypocotyl phenotype of 35S-PIF4 (Figure 6). In contrast, other gain-of-function mutations in different IAA proteins [21]–[24], including slr-1 (contains a gain-of-function mutation in IAA14), axr2-1 (contains a gain-of-function mutation in IAA7), axr5-1 (contains a gain-of-function mutation in IAA1) and iaa28-1, did not affect hypocotyl elongation in response to high temperature and failed to suppress the long-hypocotyl phenotype of 35S-PIF4 seedlings (Figure S9). These results demonstrate that the auxin signaling repressor SHY2/IAA3 is selectively involved in high temperature-induced hypocotyl growth.
As sessile organisms, plants have evolved remarkable ability to adapt their development to the ever-changing environmental conditions. Exposure of plants to high temperature results in dramatic changes in plant architecture, including elongation responses and leaf hyponasty. High temperatures can also considerably reduce plant biomass, raising concerns over future crop productivity and food security. Therefore, the modulation of plant architecture by high temperature is a subject of considerable agricultural significance, particularly with regard to global climate change. An ever-growing body of evidence in Arabidopsis has implicated that high temperature-induced plant architecture remodeling relies on the interplays between multiple external and internal cues including light, circadian clock, auxin, gibberellin and others [25], [26]. Particularly, recent studies reveal that a group of bHLH transcription factors play a central role in modulating developmental responses to both light and temperature [1], [14], [16], [27]–[29].
In this study, we discovered that, as a molecular integrator, the PIF4 transcription factor links high temperature to the auxin pathway in regulating high temperature-induced hypocotyl elongation. Several lines of evidence support this finding: First, underlying the long-standing observation that high temperature induces a dramatic elongation of the hypocotyl, we showed that high temperature triggers an elevation of the transcript abundance of both PIF4 and YUC8 (Figure 1). Second, high temperature-induced upregulation of YUC8 expression largely depends on the function of PIF4 (Figure 1). Third, overexpression of PIF4 leads to increased expression of YUC8 and elevated endogenous free IAA levels (Figure 2). Fourth, as revealed by ChIP and EMSA assays, PIF4 specifically binds to a core DNA G-box motif (CACGTG) present in the promoter of the YUC8 gene (Figure 3). Fifth, transactivation assays in N. benthamiana leaves support that PIF4 stimulates the activity of the YUC8 promoter fused with a reporter (Figure 4). Finally, the yuc8 mutation, which is defective in high temperature-induced hypocotyl elongation, is able to partially suppress the long-hypocotyl phenotype of the 35S-PIF4 plants (Figure 5). Together, these data support that, PIF4 selectively activates the expression of the auxin biosynthetic gene YUC8, thus integrates high temperature to the auxin pathway in regulating adaptive hypocotyl growth.
It is worthy of note that the yuc8 mutant still retains some response to high temperature in hypocotyl elongation and that this mutation fails to completely suppress the long-hypocotyl phenotype of 35S-PIF4 plants (Figure 5). A plausible explanation for this is that the yuc8 mutant used in this study shows reduced, but not loss of, YUC8 expression (Figure S7). Alternatively, we could not rule out the possibility that PIF4 may activate auxin biosynthetic genes other than YUC8, which act weakly in PIF4-mediated hypocotyl growth in response to high temperature. A very recent report hints that the PIF4 transcription factor could target TAA1 [29], which acts genetically upstream of the YUC family genes in IAA production [12], [13]. Considering that overexpression of TAA1 does not lead to any obvious developmental phenotype [11], [12] and that TAA1 and YUCs act in a common linear biosynthetic pathway for auxin production [6], [12], [13], it is reasonable to propose that TAA1 acts together with other auxin biosynthesis genes such as YUC8 to mediate high temperature-induced and PIF4-mediated hypocotyl elongation. However, our gene expression analyses reveal that overexpression of PIF4 alone fails to elevate TAA1 transcription (Figure S4).
PIF4-mediated activation of YUC8 expression in response to high temperature exemplifies a mechanism by which environmental cues manipulate auxin, the key endogenous modulator of plant architecture. Another known physiological process in which both PIF4 and auxin are involved is shade avoidance syndrome (SAS), plant adaptive growth responses to the light signal [14], [30], [31]. PIF4 is therefore emerging as a molecular “hub” to integrate both temperature and light signals to regulate plant architecture remodeling [14]. Accumulating evidence reveals that, unlike shade avoidance, where PIF4 acts redundantly with its homolog, PIF5, to regulate elongation growth, PIF4 appears to perform a dominant role in driving high temperature-induced adaptive growth [1], [14], [16], [32]–[34]. These studies suggest that, PIF4, and possibly other PIF family members, have specialized and overlapping functions in regulating plant adaptive growth to different environmental stimuli.
Our results support a scenario in which the auxin pathway acts downstream of the PIF4 transcriptional regulator in regulating high temperature-induced hypocotyl elongation. Supporting evidence for this hypothesis came from our genetic analysis showing that the axr1-12 mutation, which contains a mutation in a subunit of the heterodimeric RUB-E1 enzyme required for auxin signaling [35], completely suppressed the long-hypocotyl phenotype of 35S-PIF4 seedlings (Figure S8). Based on our current knowledge of the auxin signaling pathway, auxin mediates the expression of auxin responsive genes through the inactivation of AUX/IAA transcriptional repressors that negatively control the activity of AUXIN REPONSE FACTOR (ARF) transcription factors [36]. In the context that many gain-of-function aux/iaa mutations are associated with reduced response to exogenous auxin, but developmental defects among these mutants are frequently more specific [36], it is reasonable to speculate that specific Aux/IAA-ARF pair(s) may function in the PIF4-auxin pathway to mediate the specialized hypocotyl elongation process triggered by high temperature. In our genetic efforts to identify new components involved in the PIF4-auxin pathway in regulating high temperature-mediated hypocotyl elongation, we determined that SHY2/IAA3, but not other IAA proteins tested, has a specialized function in mediating high temperature-induced hypocotyl elongation. It is of interest in future studies to identify the ARF transcription factor(s) interacting with SHY2/IAA3 in regulating high temperature-induced hypocotyl elongation.
Arabidopsis thaliana ecotypes Columbia (Col-0), Ler and WS were used as wild types. The pif4 mutant used in this study was the reported null allele pif4-2 [1]. Other plant materials used in this study were previously described: DR5:GUS [37], 35S-PIF4 [19], 35S:PIF4-HA [19], yucca [7], axr1-12 [38], shy2-2 [2019], slr-1 [21], axr2-1 [22], axr5-1 [23] and iaa28-1 [24]. yuc8 (SALK_096110) was identified from the SIGnAL T-DNA collection [39].
All molecular manipulations were performed according to standard methods [40]. The PIF4 coding fragment was amplified by PCR and cloned into the AscI/PacI sites of the binary vector pMDC7 [17], resulting in a chemical-inducible PIF4 expression construct. The construct was then transformed into Agrobacterium tumefaciens strain GV3101 (pMP90), which was used for transformation of Arabidopsis plants by vacuum infiltration [41].
Seeds were surface-sterilized for 15 min in 10% bleach, washed four times with sterile water, and plated on half-strength Murashige and Skoog (MS) medium. Plants were stratified at 4°C for 2 d in darkness and then transferred to a phytotrone set at 22°C with a 16-h light/8-h dark photoperiod or in continuous light for specific experiments. For high temperature treatment, plants were directly grown at 29°C in continuous light or young seedlings were transferred to 29°C in continuous light for different times.
For qRT-PCR analysis, seedling were harvested and frozen in liquid nitrogen for RNA extraction. RNA extraction and qRT-PCR analysis were performed as previously described [37]. Primers used to quantify gene expression levels are listed in Table S1. The GUS activity assays were performed as previously described [37].
One gram of 6-d-old seedlings of 35S:PIF4-HA transgenic plants [19] and the anti-HA antibody (Abcam) were used in ChIP experiments. Chromatin immunoprecipitation (ChIP) assays were performed as previously described [42]. The enrichment of DNA fragments was determined by semi-quantitative PCR analysis. Three independent biological repeats were performed.
PIF4 and Luciferase (Luc) were synthesized by using the Rabbit Reticulocyte TNT system (Promega) [18], [43]. The 60-bp YUC8 promoter probes containing G-box motifs were synthesized and labeled with biotin at the 3′ end (Invitrogen). Cold competitor probes were generated from dimerized oligos of the YUC8 promoter region containing the wt-G-box (CACGTG) or mut-G-box (CACGGG) motifs, respectively. DNA gel-shift assays were performed as described [18], [43]. Probe sequences are shown in Table S1.
The transient expression assays were performed in N. benthamiana leaves as previously described [44]. The YUC8 promoter was amplified with the primer pairs 5-CACCATCCGATATGATAACGAT-3 and 5-TGGAAGTTGTATTGGAAA-3 and cloned into pENTR using the pENTR Directional TOPO cloning kit (Invitrogen). To generate YUC8 promoter with mutations, site-directed mutagenesis was used to delete the two G-boxes in the YUC8 promoter (Figure 3) using the TaKaRa MutanBEST kit. Then, the two YUC8 promoter versions were fused with the luciferase reporter gene LUC through the Gateway reactions into the plant binary vector pGWB35 [45] to generate the reporter constructs pYUC8:LUC and pYUC8(mut):LUC. The PIF4 effector construct was the 35S:PIF4. For this construct, the PIF4 coding fragment was amplified by PCR with the primer pairs 5-CACCATGGAACACCAAGGTTGGAG-3 and 5-GTGGTCCAAACGAGAACCGT-3. Five independent determinations were assessed. Error bars represent SD. The experiments were repeated at least five times with similar results.
For measurement of free IAA levels in wild-type and pif4 mutant hypocotyls in response to high temperature treatment, the hypocotyls of 6-d-old wild-type and pif4 mutant seedlings grown at 22°C and 29°C in continuous light, respectively, were harvested for free IAA measurement. For the wild-type seedlings grown at 29°C, the 2 mm length parts for each hypocotyl (above the junction between hypocotyl and root) were harvested for free IAA measurement. Eight-d-old seedlings of wild-type and 35S-PIF4 grown at 22°C in continuous light were harvested for free IAA measurement. Approximately 200 mg (fresh weight) of tissues were used for IAA extraction and measurement as previously described [46].
|
10.1371/journal.ppat.1005158 | Distinct Viral and Mutational Spectrum of Endemic Burkitt Lymphoma | Endemic Burkitt lymphoma (eBL) is primarily found in children in equatorial regions and represents the first historical example of a virus-associated human malignancy. Although Epstein-Barr virus (EBV) infection and MYC translocations are hallmarks of the disease, it is unclear whether other factors may contribute to its development. We performed RNA-Seq on 20 eBL cases from Uganda and showed that the mutational and viral landscape of eBL is more complex than previously reported. First, we found the presence of other herpesviridae family members in 8 cases (40%), in particular human herpesvirus 5 and human herpesvirus 8 and confirmed their presence by immunohistochemistry in the adjacent non-neoplastic tissue. Second, we identified a distinct latency program in EBV involving lytic genes in association with TCF3 activity. Third, by comparing the eBL mutational landscape with published data on sporadic Burkitt lymphoma (sBL), we detected lower frequencies of mutations in MYC, ID3, TCF3 and TP53, and a higher frequency of mutation in ARID1A in eBL samples. Recurrent mutations in two genes not previously associated with eBL were identified in 20% of tumors: RHOA and cyclin F (CCNF). We also observed that polyviral samples showed lower numbers of somatic mutations in common altered genes in comparison to sBL specimens, suggesting dual mechanisms of transformation, mutation versus virus driven in sBL and eBL respectively.
| Burkitt lymphoma is endemic in sub-Saharan Africa and affects primarily children of age 4–7 years. Historically, it was one of the first tumors associated with a virus (EBV) and bearing a translocation involving an oncogene, i.e. MYC. There are three distinct clinical variants of Burkitt lymphoma according to the World Health Organization: sporadic, endemic and immunodeficiency-related. Although there has been some recent work on the molecular characterization of sporadic Burkitt lymphomas, little is known about the pathogenesis of endemic cases. In this work, we analyzed 20 samples of RNASeq from Burkitt lymphoma collected in Lacor Hospital (Uganda, Africa) and validated in an extension panel of 73 samples from Uganda and Kenya. We identify the presence in the adjacent non-neoplastic tissue of other herpesviridae family members in 53% of the cases, namely cytomegalovirus (CMV) and Kaposi sarcoma herpesvirus (KSHV). We also demonstrate expression of EBV lytic genes in primary tumor samples and find an inverse association between EBV lytic expression and TCF3 activity. When studying the mutational profile of endemic Burkitt tumors, we find recurrent alterations in genes rarely mutated in sporadic Burkitt lymphomas, i.e. ARID1A, CCNF and RHOA, and lower numbers of mutations in genes previously reported to be commonly mutated in sporadic cases, i.e. MYC, ID3, TCF3, TP53. Together, these results illustrate a distinct genetic and viral profile of endemic Burkitt lymphoma, suggesting a dual mechanism of transformation (mutation versus virus driven in sBL and eBL respectively).
| Burkitt lymphoma (BL) is the first human cancer to be associated with the Epstein-Barr virus (EBV), the first tumor to exhibit a chromosomal translocation activating an oncogene (MYC), and the first lymphoma to be associated with human immunodeficiency virus (HIV) infection. The World Health Organization[1] classification describes three clinical variants of BL: endemic, sporadic, and immunodeficiency-related. These variants are similar in morphology, immunophenotype, and genetics. While the sporadic variant (sBL) occurs outside of Africa and is rarely associated with EBV infection, the endemic variant (eBL) arises mainly in Africa and is associated with malaria endemicity and EBV infection in almost all cases. Epidemiological studies have shown that malaria and EBV combined do not fully explain the distribution of eBL in high risk regions[2]. Malaria and EBV are in fact ubiquitous within the lymphoma belt of Africa, suggesting that other etiologic agents may be involved[3]. However, it is unclear what other epidemiological factors could play a role in the genesis of eBLs.
Three types of EBV latency have been described in EBV-related lymphomas according to the pattern of EBV nuclear antigen (EBNA) and the latent membrane protein (LMP) expression, namely latency I, II, and III[4]. Specifically, latency I is usually associated with eBL and it denotes a transcriptional program in which an EBV infection does not produce virions and expresses a single protein, EBNA-1. While the latency I program has been extensively characterized in vitro, a different form of latency has been recently reported in 15% of eBL that uses a different set of promoters. Termed Wp-restricted latency[5], this program shows a homogeneous host expression signature[6] characterized by down-regulation of BCL-6 and up-regulation of IRF-4 and BLIMP-1. Other reports have described latency program heterogeneity at single cell level[7] and low expression of LMP genes in a fraction of cases[8,9]. Heterogeneous EBV transcription profiles with LMP expression have been recently reported in some cases of AIDS-related and sporadic BL[10], but extensive data on endemic cases are not available yet. These studies indicate that the transcriptional EBV programs of primary eBL could be more complex than expected across cases and within individuals. Therefore, the exact role of EBV has remained elusive and further investigation is required.
The genetic hallmark of all three clinical variants of BL is the t(8;14) translocation involving the juxtaposition of the immunoglobulin heavy chain locus (IGH) with the MYC oncogene[11]. However, although transgenic mice expressing MYC under the control of the intronic IGH enhancer (Eμ) develop B cell lymphomas[12], successive molecular characterization demonstrated that this model does not fully recapitulate the human disease. The comparison between the gene expression profile (GEP) of BL and diffuse large B-cell lymphoma (DLBCL) highlighted a distinct signature of BL characterized by the expression of both MYC targets and germinal-center B-cell genes[13]. Furthermore, hypermutation and different breakpoint patterns of IGH/MYC translocation[14,15] suggests that the origin of human BL derives from aberrant class switching in the germinal center (GC), while transgenic IGH/MYC mice typically arise from precursor/naive B-cells. The more accurate PI3K/MYC transgenic mouse model by Sander et al[16] better recapitulates the human phenotype of BL and highlights the importance of the PI3K pathway in the disease. Moreover, GEP analysis has demonstrated that the transcriptional profile of eBL is different from that observed in sBL[17]. Recent studies have unveiled the genetic landscape of sBL characterized by mutations affecting the B-cell receptor (BCR) pathway and in particular the transcription factor TCF3, its negative regulator ID3, the cell-cycle G1/S regulator CCND3[18,19], and the chromatin-remodeling gene ARID1A[20]. On the contrary, very little is known about the spectrum of alterations in eBL, how it might differ from that of sBL, the correlations between host mutation and viral infection, and the specific viral/host transcriptional programs.
In this study, we aim to characterize the presence of other potential agents, to define the EBV transcriptional profile and to link these profiles to the mutational status of new and previously reported genes. We provide a characterization of the mutational and viral landscape of eBL using 20 cases from Uganda. RNA-Seq, in combination with targeted sequencing technology on a larger cohort of cases, allows the identification, validation and assessment of the recurrence of new somatic mutations. In addition, in contrast with earlier microarray-based expression studies, RNA-Seq provides the opportunity to identify and associate microbial and tumor mutational and expression profiles.
To identify new pathogens in eBL, we applied Pandora, a new pipeline for the characterization of tumor microbiomes, to a discovery cohort of 20 RNA-Seq samples. We established a read cutoff on the basis of those samples that tested positive for RNA in situ hybridization (ISH) of the EBER transcript. Since ISH validated all the RNA-Seq samples as positive, we established the threshold to call a virus present in a particular sample as the minimal number of reads detecting EBV (S1 Fig). Next, we established the EBV subtype by aligning RNA-Seq reads to the genomes of both EBV type I and type II and deduced type I as the closest genotype. In addition to EBV, RNA-Seq revealed the presence of other viruses. In particular, 5/20 cases contained human herpesvirus 5 (HHV5, cytomegalovirus, CMV), 4/20 human herpesvirus 8 (HHV8, Kaposi sarcoma herpes virus, KSHV), and 1/20 human T-lymphotropic virus 1 (HTLV-1) (Fig 1A, S2 Fig and S3 Fig). Human immunodeficiency virus (HIV) was not detected in any case, confirming that pediatric eBL is rarely associated with the immunodeficiency syndrome[21]. Nested PCR and immunohistochemical (IHC) analysis performed on all 20 original samples confirmed the presence of all the viruses in the discovery cohort (S1A Table). To assess whether RNA-Seq findings generalize for EBV, CMV, KSHV, and HTLV-1, we assayed for the presence of these four viruses in 20 additional cases from western Kenya by IHC (S1B Table). In this Kenyan cohort, EBV was detected in 20/20 samples, CMV in 8/20 samples (Fig 1B and S4 Fig), KSHV in 7/20 samples (Fig 1C and 1D, and S5 Fig), and HTLV-1 in 0/20 samples. Therefore, over the 40 cases, we report the overall viral infection frequencies of 40/40 (100%) for EBV, 13/40 (32.5%) for CMV, 11/40 (27.5%) for KSHV, and 1/40 (2.5%) for HTLV-1. IHC analysis demonstrated the presence of CMV in the stromal cells and macrophages localized within the tumors and in the adjacent reactive lymphoid tissue (Fig 1B, S2 Fig and S4 Fig). KSHV was identified not only in normal B-lymphocytes and endothelial cells from the adjacent reactive lymphoid tissue (S3 Fig and S5 Fig), but also in one case in about 5–10% of neoplastic cells (Fig 1C and 1D). HTLV-1 was detected in reactive T-lymphocytes in the only positive case of the discovery cohort. Sections of the samples incubated with the secondary antibody alone and sections of reactive lymphoid tissue were used as negative controls. Sections of lymph nodes with infectious mononucleosis were used as positive control for EBV. Next, we compared the viral landscape of endemic and sporadic cases by analyzing 27 RNA-Seq sBL samples from Schmitz et al.[19] with Pandora. The analysis showed the presence of EBV and HIV respectively in 4/27 (15%) and in 1/27 (4%) cases, consistent with several literature sources[22].
Beyond identification of EBV presence, RNA-Seq enabled us to quantitatively analyze the viral transcriptional program. In addition to EBER-1 and EBER-2 transcripts, expression analysis of the viral genes showed the expression of EBNA-1, a gene associated to latency I type, in 18/20 cases (Fig 2A). We also detected either LMP-1 or LMP-2A, characterizing the latency II type, in 13/20 samples (65%), and also EBNA-2 in 1/20 cases (5%). Interestingly, 2/20 cases (10%) were characterized by the expression of EBNA-3A/B/C/LP, together with the lytic gene BHRF-1, suggesting a Wp-restricted program[23]. However, the specific analysis of EBV isoforms showed the presence of H2-HF splicing event, which is hallmark of lytic BHRF-1 expression[24–26](S6 Fig). Unsupervised hierarchical clustering of expressed EBV genes demonstrated two main clusters distinguished largely by gene products involved in EBV replication (BALF-2, BCRF-1, BHRF-1, BILF-1, BMRF-1, BNLF-2a, BZLF-1). The expression of these genes suggests a non-canonical latency program of the virus with a subset of viral episomes initiating lytic reactivation[23].
Due to the heterogeneity of the viral transcriptional programs, we aimed to validate the latency type by performing RT-qPCR for the EBNA-1, LMP-1, LMP-2A, EBNA-2, EBNA-3C, and BHRF-1 transcripts across an additional series of 26 cases from an extended cohort of samples from Kenya. EBNA-1 was detected in 26/26 (100%), LMP-1 and LMP-2 in respectively 5/26 (20%) and 20/26 (75%) cases (S3 Table and S7A Fig), EBNA-2 in 0/26 (0%), and the combination of EBNA-3C and BHRF-1 in 4/26 (15%). These results are largely consistent with the RNA-Seq data with the exception of LMP-1 that has been detected at higher frequency in RNA-Seq (S2 Table). Next, we evaluated the lytic cycle activation and found BILF-1, BALF-4, and LF-2 in all 26 cases, whereas we observed the expression of BALF-2 in 23/26 (90%), BHRF-1 in 20/26 (80%), BZLF-1 and BMRF-1 in 15/26 (60%), BNLF-2a in 13/26 (50%), and BCRF-1 in 11/26 (45%) of the cases (S3 Table and S7B–S7D Fig).
We then validated the expression of all the available encoded-proteins by IHC using stringent positive and negative controls as reported in Materials and Methods. Overall, IHC evaluation confirmed a non-canonical latency associated program with the expression of some proteins characterizing latency II (i.e. LMP-1 in 2/26 and LMP-2A in 17/26 of the cases); however, there was heterogeneity in the intensity of protein staining and in the proportion of positive tumor cells. LMP-1 was detected in few cells, whereas LMP-2A was identified in a proportion of cells ranging from 25% to 50% (Fig 2B and 2C). EBV replication was assessed by nuclear expression of the immediate-early BZLF-1/ZEBRA and early BMRF-1/Ea-D, BHRF-1/Ea-R lytic proteins (Fig 2D–2I). There was positive staining in the neoplastic cells for BZLF1, BHRF-1/Ea-R and BMRF-1/Ea-D, respectively in 11/26 (40%), 16/26 (60%), and 13/26 (50%) of the cases (S3 Table).
Finally, we compared the patterns of latent and lytic gene expression between endemic and sporadic BLs using the 4 EBV-positive sBLs of the 27 RNA-Seq samples from Schmitz et al.16 We observed the expression of BHRF-1 and BMRF-1 in 1 case; BZLF-1 was present in 2 cases and LMP-2A in 4 cases.
To identify the genes that are somatically mutated in eBL, we applied the SAVI algorithm[27] to the cohort of 20 RNA-Seq samples (see Material and Methods for gene selection criteria). Our analysis identified 13 genes recurrently mutated in more than 4 samples. We confirmed the presence of mutations in genes previously reported in BL literature[18,19,28] (Fig 3A–3C and S4 Table), including MYC in 10/20 (50%), DDX3X in 7/20 (35%), ID3 in 6/20 (30%), ARID1A in 5/20 (25%), RHOA in 4/20 (20%), TCF3 and TP53 in 3/20 (15%), CCND3 in 1/20 (5%) of the cases. In addition, we found recurrent mutations in one gene not reported so far: CCNF, detected in 4 out of the 20 cases (20%). Since RHOA mutations have not been previously detected in eBL and CCNF mutation was a new discovery, their prevalence as specific mutations was further assessed using Sequenom technology on an extended panel of 66 neoplastic samples plus 7 cases with matched normal controls (S8A and S8B Fig). Recurrent mutations in RHOA were found in 6/73 eBL cases (8%), and in 0/7 normal samples. Two of the 6 RHOA mutations occurred in paired eBL/normal cases, confirming that the alterations are somatic (S5 Table). Recurrent mutations in codon 451 of CCNF were found in 14/73 eBL cases (19%), and in 0/7 normal samples. One of the 14 CCNF mutations occurred in a paired eBL/normal case, showing that also CCNF alteration is somatic. Direct sequencing of genomic DNA confirmed all the mutations identified by Sequenom tecnnology and RNAseq (S9 Fig and S10 Fig).
The distribution of somatic mutations and viral presence across both eBL and sBL samples exhibit two interesting features (Fig 3B). First, in eBL samples we observed lower mutational frequencies in the genes MYC, ID3, TCF3, DDX3X, CCND3 and TP53, as compared to their reported recurrence in sBL, and higher mutational frequencies in ARID1A, RHOA, and CCNF[18,19]. Second, in sBL cases an almost mutual exclusivity can be seen between EBV presence and mutations in TCF3/ID3 both known to be driver genes in sBL (p-value < 0.02, Fisher exact test). To explore this hypothesis, we performed a hierarchical clustering of both endemic and sporadic cases on TCF3 target genes (previously reported in Schmitz et al.[19]) and we demonstrated that the first bifurcation of the dendrogram classifies the samples into EBV-positive and EBV-negative BL independently on the specific subtype with an accuracy of 96% (45/47). (Fig 4A). The results show that the TCF3 pathway is more activated in EBV-negative cases, as indicated by the significant negative enrichment of TCF3 target genes in EBV-positive samples. Furthermore, we observe that when considering the overall panel of both endemic and sporadic BL samples, the mutually exclusivity between TCF3/ID3 mutations and EBV infection yields a more significant effect (p-value < 0.0008, Fisher exact test). To further investigate the host transcriptional programming related to EBV presence, we performed GSEA C2 analysis on genes differentially expressed between EBV-positive and EBV-negative cases. Interestingly, we detected a significant enrichment for the LMP-1 gene set signature, reported by Sengupta et al. [29] in nasopharyngeal carcinoma, which is consistent with the detected LMP-1 expression in RNA-Seq data (Fig 4B). Moreover, since 13/20 RNA-Seq cases were positive for LMP2A, we investigated the role of this viral gene in the context of eBL and GSEA C2 analysis has been performed on gene differentially expressed between LMP-2A positive and LMP-2A negative samples. Interestingly, the E2F, E2F3 and cell cycle G1/S gene sets presented the highest significant enrichment score (see S11 Fig), together with the down-regulation of retinoblastoma pathway. These results can be explained as an effect of the interaction between MYC and LMP2A. In fact, previous studies showed that LMP-2A promotes MYC-induced lymphomagenesis[30], and E2F is a know target of MYC during cell division and proliferation[31]. Moreover, several works associate LMP-2A expression to the PI3K/Akt pathway activation[32–35] and the study from Brennan et al. (Oncogene, 2002 [36]) shows that the activation of PI3K pathway in lymphoblastoid cell lines can promote E2F transcription activity to affect cell cycle and cellular proliferation.
Over the past few years, the concept that many diseases can be etiologically linked to infection by more than one pathogen has drawn increased attention[37–40]. Whether endemic Burkitt lymphoma should also be considered a polymicrobial disease and what role genetic alterations play in the tumor are still open questions. In this paper, we analyzed the presence of pathogens other than EBV in 40 eBL primary tumors by RNA sequencing, PCR, and immunohistochemistry, and found the presence of CMV and KSHV. We detected these viruses, which are frequently reported in the African population[41], primarily in the surrounding non-neoplastic tissue. Their prevalence in areas endemic for EBV, along with their absence in the sporadic cases, suggests that CMV or KSHV could contribute to the chronic antigenic stimulation in which eBL occurs.
The presence of these additional cofactors may also induce EBV lytic cycle through B-cell reactivation and spreading EBV infection out of its natural niche of memory B-cells, characterized by a latency 0/I program[42,43]. In fact, in our samples we showed a non-canonical latency program of the virus characterized by a large number of cases expressing LMP-1/-2A/-2B in a significant proportion of cells along with lytic reactivation. Our results are in agreement with recent studies showing more complex EBV protein expression in Akata and Mutu cell lines, commonly used to study the role of EBV in Burkitt lymphoma[44]. By using an alternative approach based on RT-QPCR array platform, Tierney et al. report a quantitative characterization of EBV transcripts in different experimental infection models that were validated in endemic Burkitt lymphoma samples[24]. Interestingly, in this study a significant expression of LMP-2 gene was revealed. Moreover, our results are in accordance with a previously published study in primary AIDS-related lymphomas (ARL) by Arvey and colleagues[10], although a rigorous comparison is limited by the small number of ARL BLs. All together, our findings confirm recent evidence that LMP-2A cooperates in reprogramming the function of normal B-lymphocytes and enhance MYC driven lymphomagenesis through the activation of PI3K-pathway[45,46]. This pathway is a crucial to MYC mediated transformation as shown by PI3K/MYC transgenic mouse that produces a model that represent a phenocopy of human tumors in terms of histology, gene and protein marker expression, and somatic hypermutations[47]. This scenario suggests that LMP-2A activation of PI3K is an alternative/convergent mechanism to the one driven by TCF3/ID3 mutations.
The expression of genes characterizing the lytic phase of EBV found by RNA-Seq was confirmed by IHC staining for the three main genes involved in the initiation of the lytic phase, BZLF-1/ZEBRA, BMRF-1/Ea-D and BHRF-1/Ea-R. In early latent infection, EBV can be induced to enter the lytic cycle by a variety of causes including B-cell receptor stimulation, Toll-like receptor-9 activation, hypoxia, and growth factors[48,49]. Although lytic infection kills the host cell, it also allows horizontal spread of EBV from cell to cell and may increase the pool of latently infected B-lymphocytes from which transformed cells arise. Additionally, lytically infected B-cells secrete factors that may promote tumorigenesis, including growth and angiogenesis factors and immunosuppressive cytokines. Recent evidence has challenged the view that only the latency phase of EBV infection is significant for the development of EBV-associated malignancies, proposing that lytic EBV replication may be of pathogenic relevance.[50] Humanized mice infected with lytic active viral strains develop more lymphomas than animals infected with replication-defective strains[39], suggesting that lytic EBV infection may be of importance also in the context of an active immune response. In the present study we gave evidence for the first time that this occurs in vivo in the neoplastic cells of the primary tumors. Physiologically, lytic gene products are expressed in three consecutive stages: immediate-early, early, and late. Immediate-early lytic gene products initiate the process by inducing the activation of transcription of the other genes. Early genes control replication and metabolism of neoplastic cells[51]. Fatty acid synthase expression is induced by the BRLF-1 immediate–early protein, and interestingly BL tumors are characterized by altered lipid metabolism[52]. Late gene products code for viral capsid antigens and proteins involved in immune evasion. BNLF-2A, detected in a significant number of our cases, may protect infected B-cells from immune recognition and elimination[53]. Finally, the EBV transcriptome during the reactivation may involve the contribution of a wide array of other virus-encoded RNAs, such as BARF-0, BARF-1, BcLF-1, and RPMSI-1[54], that are not translated and may function as non-coding RNA molecules which could participate in regulating gene expression[55]. Heterogeneity in lytic/latent expression programs can be observed not only between patients but also within individual tumors, on a cell-to-cell basis. Intra-patient heterogeneity might be related to the activation of the immune response following the expression of the viral genes. Therefore, the tumor is under selective pressure and needs alternative mechanisms to survive and proliferate[56].
Our data on the mutational landscape of eBL seems to support this hypothesis. In fact, eBL samples were characterized by a lower number of point mutations in genes previously found altered in sBL, including MYC, ID3, TCF3, DDX3X, CCND3, and TP53. These results are consistent with previous studies by Schmitz et al. [19]in which TCF3/ID3 mutations were more common in sBL (70% of the cases) than eBL (40%). In particular, we observed a near mutual exclusivity between TCF3/ID3 mutations and the presence of EBV, indicating that TCF3 pathway is more significantly activated in EBV-negative cases.
The inverse correlation we observed between the presence and expression of EBV and the number of cellular mutations in the different BL cases, may represent an in vivo picture of the dynamic process by which a neoplastic cell, initially dependent upon EBV, switches-off viral genes and switches-on cellular mutated genes to survive and proliferate. These results are consistent with previous analysis of pediatric BL[20]. Based on our findings, one should infer that eBL may arise from pathogenic pathways that are partially distinct from those driving sBL, suggesting dual mechanisms of transformation in BL, mutationally versus virally driven. On the other hand, ARID1A and RHOA were more often mutated in eBL than in sBL. ARID1A is one of the subunits of the Switch/Sucrose Non-Fermentable (SWI/SNF) chromatin remodeling complex and is currently thought to behave like a tumor suppressor gene. Consistently, ARID1A mutations frequently occur as insertion/deletion, and in most of our cases involved the amino acid G1630. This gene has been reported as frequently mutated in the context of pediatric BL, with a significant association to EBV negative cases[20], suggesting that the high prevalence in eBL compared to the sBL may be due to the pediatric nature of the endemic case. However, other EBV-associated cancer types show frequent deregulation of ARID1A[57–61]. In particular, in EBV-associated gastric cancer a strong correlation between ARID1A deactivation and EBV presence has been reported[62–64]. RHOA, which belongs to the Ras homolog family, is a small GTP-ase protein recently found to be mutated in three tumors associated with EBV infection, namely peripheral T-cell lymphoma (where it relates to follicular helper T-cells[65–67]), diffuse gastric carcinoma[68] and paediatric sBL[28]. The distribution of RHOA mutations in our cohort overlaps with the already reported mutations (codons 5, 17, 42 and 69) suggesting a similar functional role.
Finally we identified recurrent mutations involving the amino acid R451C in one gene not previously detected in endemic or sporadic BLs, CCNF, altered in 20% of our cases. CCNF encodes a member of the cyclin family belonging to the F-box protein family; it acts as an inhibitor of centrosome reduplication during G2 phase and protects the cell from genome instability[69]. Therefore, it is reasonable that CCNF mutations may cooperate in inducing lymphomagenesis by promoting chromosome instability and a hypermutator phenotype[70,71].
Understanding the mechanisms regulating EBV lymphomagenesis will hopefully lead to the development of highly specific therapies. To avoid the tumor evasion from the already available therapies, we need to identify and target the multiplicity of pathways that are deregulated in the neoplastic cells and decrease tumor survival and proliferation.
A total of 20 BL samples preserved in RNAlater (RNA stabilization Reagent-QIAGEN, Valencia, CA) were collected from the Department of Human Pathology of the Lacor Hospital (Uganda, Africa), in endemic areas. For all of them, formalin-fixed and paraffin-embedded (FFPE) samples have been available. All diagnoses were reviewed by 2 expert hematopathologists and were formulated according to the 2008 WHO classification. The clinical and histopathologic characteristics of the 20 BL cases are summarized in S6 Table. Briefly, all cases were t(8;14)-positive, and the immunophenotype was consistent with the diagnosis of BL (CD20 positive, CD10 positive, BCL-6 positive, Ki67> 98%, BCL-2 negative). Epstein-Barr virus was detected by using in situ hybridization with EBER probes (INFORM EBER, Roche Diagnostics, Basel, Switzerland). EBV infection in tumor cells was observed in 100% of the samples, assessed by strong nuclear expression of small EBV-encoded RNA genes, EBER-1 and -2. These cases have been previously studied for gene expression profile analysis and showed a molecular profile consistent with molecular BL[17].
We used two distinct series of cases for validation of RNA-Seq results. The first included 26 primary tumors collected at the Moi University, Eldoret (Western Kenya). Of these, 20 were used for virus data validation and 26 for EBV latency validation. The second was comprised of 66 neoplastic samples plus 7 cases for which matched normal controls were available (1 liver, 6 lymph nodes) collected from endemic area in Africa, and was used for Sequenom validation.
Total RNA extraction was perfomed by RNeasy Plus Mini Kit(QIAGEN, Valencia, CA) according to the manufacture instructions. The amount and quality of RNA were evaluated by measuring the optical density (OD) at 260 nm, the 260/230 and the 260/280 ratios using a Nanodrop spectrophotometer (ND-100, Nanodrop, Thermo Scientific, Celbio, Italy).
Paired-end libraries (2x75 base pair) were prepared according to the TruSeq RNA sample preparation v2 protocol (Illumina, San Diego, USA). Briefly, 2 μg of Poly(A)+ RNA was purified from total RNA using poly-T oligo attached magnetic beads and then used for fragmentation into 130–290 bp fragments. First, single stranded cDNA was synthesized using reverse transcriptase (SuperScript II, Invitrogen, Life Technologies,USA) and random hexamer priming, followed by generation of double-stranded cDNA. AmpureXP beads (Beckman Coulter, Brea CA) were used to purify the ds cDNA and end repair step was performed to convert the overhangs, resulting from fragmentation, into blunt ends by 3’ to 5’ exonuclease activity. A single “A” nucleotide was added to the 3’ ends of the blunt fragments to prevent them from ligating to one another during the adapter ligation reaction. This approach was adopted to ensure a low rate of chimera (concatenated template) formation. Subsequently, sequencing adapters were added to the ends of the ds cDNA fragment and a PCR reaction was used to selectively enrich those ds cDNA fragments that had adapter molecules on both ends, amplifying the amount of ds cDNA in the final libraries. Lastly, PCR library products were purified by AmpureXP beads and quality control analysis was assessed using a DNA-1000 (Agilent, USA). The quantification was performed by Quant-it PicoGreendsDNA Assay Kit according to manufacturer’s protocol (Invitrogen, Life Technologies,USA). The resulting libraries were sequenced on an Illumina HiScan SQ (Illumina, San Diego, USA) following the manufacturer's instructions.
Sequence variants were obtained using the SAVI (Statistical Algorithm for Variant Identification)[72,73] algorithm independently for each sample. Candidate somatic mutations were obtained by eliminating common germline variants (dbSNP 132 and variants from 10 reactive lymph nodes). Genes recurrently mutated in more than 4 samples and expressing the corresponding transcript with RPKM>3 were selected. Mutations occurring in the exact same position in more than 4 samples have been discarded. Conversely, genes previously reported in BL[19,74,75] were selected, even at low recurrence, to allow the comparison between endemic and sporadic subtypes (S3 Table). Sanger sequencing was used for technical validation.
Characterization of the tumor microbiome is accomplished with Pandora, a new RNA-Seq pipeline for pathogen identification and discovery (S12 Fig). The algorithm takes raw RNA-Seq data as input and outputs annotated microbial spectra present in the tumor sample. Pandora implements a subtractive algorithm consisting of discrete modules. First, the Host Removal phase sequentially aligns the input reads to the host reference using bowtie2[76], blastN[77] and Megablast,[78] and filters out the data originating from the host. Second, the unaligned (non-host) reads are passed as input to the Microbe Identification phase where the reads are aligned to curated sets of NCBI microbial sequences representing viruses/viroids, bacteria, fungi, and select taxa of eukaryotic parasites. Third, the NCBI records matching each non-host read are input to the Reporting phase where microbial load, gene expression, and relevant clinical parameters are computed as the final output. The microbial load is computed as the number of reads mapping to the organism or virus normalized by the genome length. Gene expression quantification is computed as transcript per million (TPM)[79], which provides a more accurate relative quantification of mRNA abundance compared to other normalization methods such as RPKM.
RNA-Seq reads were aligned to the GRCh37/hg19 reference genome using Bowtie2[76], Blastn[77] and Megablast[78]. Reads not aligning to homo sapiens (non-host reads) were mapped to human herpesvirus 4, type I (NCBI accession number NC_007605.1) using TopHat, a splicing aware alignment program [80] (S9 Table and S10 Table). EBV viral gene expression was normalized as transcripts per million (TPM)[79]. For each viral product the TPM expression was normalized by the expression of A73 genes, which is consistently expressed in all the HHV4 positive BL samples. Hierarchical clustering was computed with Pearson distance and Ward’s linkage method.
Gene expression analysis was performed on both endemic and sporadic[19] RNA-Seq samples of Burkitt Lymphoma. All the reads were aligned to human reference genome (GRCh37/hg19) by means of TopHat version 1.3.3. Transcript abundance quantification was computed as FPKM using Cufflinks, Cuffquant and Cuffnorm version 2.2.1[81]. Hierarchical clustering was performed with Pearson distance and Ward’s linkage method. Gene set enrichment analysis was obtained by running GSEA software on pre-ranked list of log2 ratio of the FPKM mean fold change between two conditions[82].
The DNA was extracted from formalin-fixed paraffin embedded (FFPE) of the original neoplastic samples using NucleoSpin Tissue (Machery-Nagel, Italy) following manufacture’s instructions. The amount and quality of DNA were evaluated by measuring the optic density (OD) at 260 nm, the 260/230 and the 260/280 ratios using a Nanodrop spectrophotometer (ND-100, Nanodrop, Thermo Scientific, Celbio, Italy).
To detect the presence of HTLV-1, CMV and KSHV, a nested PCR assay was performed on DNA of original tumor samples as previously reported[83],[84] (S7 Table and S8 Table). DNA from HTLV-1-positive cells, CMV-positive cells, and KSHV-positive cells were used as positive controls, whereas DNA from HeLa293 cells was used as negative control. Several precautions have been taken to prevent false-positive PCR results: (a) rooms for pre- and post-PCR procedures were physically separated; (b) reagents were prepared in large batches and stored in small aliquots; (c) equipment such as the microcentrifuge, water baths, pipettes, tube racks, and other small equipment was designated for PCR work only; (d) gloves were changed frequently; and (e) aerosol-barrier pipette tips, PCR tubes, and autoclaved, diethylpyrocarbonate-treated water were sterilized by UV irradiation prior to PCR. Finally, 15 μl aliquots of the PCR mixture were electrophoresed on a 2% agarose gel and directly visualized by ethidium bromide staining under ultraviolet light[85].
The MassARRAY Assay Design Suite software was used to design 8 different multiplex reactions for investigating 115 SNPs. Genotyping was performed using iPLEX Gold technology 57 MassARRAY high-throughput DNA analysis with matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (Sequenom), according to the manufacturer’s protocol. 66 neoplastic cases plus 7 samples with matched normal controls (1 liver, 6 lymph nodes) were analysed.
The expression of EBV-encoded genes (EBNA-1, EBNA-2, EBNA-3c, BALF-2, BALF-4, BCRF-1, BHRF-1, BILF-1, BNLF-2a, BMRF-1, BZLF-1, LMP-1, LMP-2A LF-2), which characterize the different latency programs, has been investigated on an additional series of 26 samples by RT-qPCR using the QuantiTect SYBR Green PCR Kit (Qiagen, CA) as previously reported (S7 Table). All samples were run in triplicate. The stably expressed housekeeping gene hypoxanthine-guanine phosphoribosyltransferase (HPRT) was used as an endogenous control and reference gene for relative quantification of each target gene. The relative expression is expressed as 2ΔCt, where ΔCt is defined as the difference in mean cycle thresholds of the gene of interest and HPRT[86]. The samples were defined as “not expressed” if the ΔCt value exceeded 50 cycles[87].
To further validate the presence of HTLV-1 and HHV-8, immunohistochemistry for viral products (HTLV-1-TAX 1: 70, Abcam, Cambridge, United Kingdom; HHV8-LANA 1: 50, Leica Biosystems, Newcastle Lid, United Kingdom; HHV8 clone AT4C11 1:50, Abnova, Taipey City, Taiwan) was performed on formalin-fixed paraffed-embedded (FFPE) sections of the original samples and in an additional series of 20 cases. CMV was detected using in situ hybridization (ISH) with Bond ISH Probe.
The protein expression of EBNA-1 (1:150, AbCam, Italy), EBNA-2 (1:100, AbCam, Italy), LMP-1 (1:100, Novus Biologicals, Italy), LMP-2A (1:100, AbCam, Italy), BZLF-1/ZEBRA (1:100, Novus Biologicals, Italy), BMRF-1/Ea-D (1:150, AbCam, Italy), BHRF-1/Ea-R (1:150, Novus Biologicals, Italy), was assessed by immunohistochemistry on FFPE sections of the original samples and on an additional series of 26 primary tumors. Sections of the samples incubated with the secondary antibody alone and sections of reactive lymphoid tissue were used as negative controls. Sections of lymph nodes with infectious mononucleosis were used as positive control. Immunoreactivity was performed on Bond Max automated immunostainer (Leica Microsystem, Bannockburn, IL, USA), with controls in parallel. No epitope retrieval was used. Ultravision Detection System using anti-Polyvalent HRP (LabVision, Fremont, CA, USA) and diaminobenzidine (DAB, Dako, Milan-Italy) as chromogen was used. Two independent investigators assessed immunoreactivity. Case were considered positive when more than 20% of the cells were stained for latent gene products and when more than 5% of the cells were stained for lytic gene products.
Ethics approval for this study was obtained from the Institutional Review Board at the University of Siena (Italy), from the Ethics and Research Committee at the Lacor Hospital (Uganda) and from the Ethics and Research Committee at Moi University, Eldoret (Kenya). Written permission and informed consent have been obtained before sample collection in accordance with the Declaration of Helsinki.
|
10.1371/journal.ppat.1003601 | Host Adaptation Is Contingent upon the Infection Route Taken by Pathogens | Evolution of pathogen virulence is affected by the route of infection. Also, alternate infection routes trigger different physiological responses on hosts, impinging on host adaptation and on its interaction with pathogens. Yet, how route of infection may shape adaptation to pathogens has not received much attention at the experimental level. We addressed this question through the experimental evolution of an outbred Drosophila melanogaster population infected by two different routes (oral and systemic) with Pseudomonas entomophila. The two selection regimes led to markedly different evolutionary trajectories. Adaptation to infection through one route did not protect from infection through the alternate route, indicating distinct genetic bases. Finally, relatively to the control population, evolved flies were not more resistant to bacteria other than Pseudomonas and showed higher susceptibility to viral infections. These specificities and trade-offs may contribute to the maintenance of genetic variation for resistance in natural populations. Our data shows that the infection route affects host adaptation and thus, must be considered in studies of host-pathogen interaction.
| Pathogens enter their hosts through several routes, the most common being ingestion (oral infection) and breaches in the cuticle (systemic infection). Several studies have shown that these infection routes strongly affect the evolution of pathogen virulence, though little attention has been given to the role of host evolution in this process. Here, we study the effect of infection route on the evolution of host defenses, using Drosophila melanogaster and its natural pathogen Pseudomonas entomophila. Profiting from the power of experimental evolution, in which the evolution of populations is followed in real time, we show that survival of D. melanogaster to an oral infection increases within the first 3 generations of selection, whereas the response to systemic infection is slower. Furthermore, we show that the evolved response is specific to the route of infection and to pathogen. Indeed, flies that resist bacteria through ingestion are not protected from systemic infection with the same bacteria species, and vice versa. Also, evolution of resistance to one pathogen does not extend to infections with bacteria of different genera via the same infection route. This degree of specificity calls for more attention onto pathogen infection routes in studies of host-parasite interactions.
| The transmission route taken by pathogens to infect their hosts has a profound impact on the evolution of host-pathogen interactions. A body of theory [1], [2], [3] and several experiments [4], [5], [6], [7] have addressed the effect of vertical or horizontal transmission on the evolution of pathogen virulence. Moreover, virulence in vector-borne or directly transmitted pathogens is expected to be differentially-affected by several factors, such as the timing of infection or inoculum size [8], [9], [10]. Recently, a meta-analysis has also shown that systemically-infecting pathogens are more virulent than those that infect via ingestion [11]. However rich this body of literature may be, it concerns the effect of transmission routes on the evolution of pathogens, not hosts (even though this implies measuring host traits, as pathogen virulence is defined as the harm imposed on hosts) [12], [13]. Pathogens that infect hosts via different routes (e.g., orally vs systemically) also trigger different physiological responses in hosts. This in turn may affect the evolution of host responses to pathogens, which will affect the outcome of the host-pathogen interaction. Therefore, addressing the evolutionary consequences of transmission route for host-parasite interactions calls for a characterization of its effects in the evolution of both pathogen and host.
It has been suggested that the immune response follows a hierarchical structure, starting with behavioural avoidance, through physical barriers and culminating in a humoral/cellular response [14], [15], [16]. Different infection routes will impact this cascade of events at different levels. Thus, the route taken by the pathogen will be crucial in defining the evolutionary consequences of infection to the individual and population. Yet, the distribution of variants across different levels in this cascade of events is unknown: which level is more likely to evolve in a population exposed to a particular immune challenge? If host adaptation occurs through changes in a shared downstream portion of the cascade such as the humoral effectors, then adapted populations are expected to show a positive correlated response to challenges acting on any part of the cascade. Conversely, if there is at least partial independence in the defence pathways activated by each infection route, then adaptation to pathogens infecting through different routes should be uncorrelated. Thus, testing host evolutionary responses to infection through different routes is crucial to ecological immunology as it will, (a) establish whether responses are general or specific for distinct routes of pathogen access and, (b) provide insight into which part of the defense cascade may be modified by evolution.
In recent years much attention has been given to the mechanistic distinction between resistance (capacity to limit pathogen loads) and tolerance (capacity to survive damage caused by a given pathogen load) [17], [18], [19]. Yet, although a few recent studies have determined if resistance or tolerance mechanisms are involved in insect host responses to pathogens [20], [21], [22], whether and how different transmission routes affect the evolution of these mechanisms is still unknown. Indeed, no study has yet addressed the consequences of different infection routes of horizontally-transmitted pathogens for the evolution of host responses.
Routes of infection observed in nature are paralleled by the infection protocols used in the Drosophila melanogaster laboratory model of insect immunity [23], [24], [25]. Traditionally, the study of Drosophila immunity is done with systemic infections [26], [27], [28], [29], but more recently, several studies have addressed the immune response to ingested bacteria [30], [31], [32], [33], [34], as the ecological relevance of this route of infection is most likely higher (for a review see [35]). These studies have shown that several responses are specific to the infection route, even if some overlap can be observed [30], [33], [36]. Indeed, to infect hosts, ingested pathogens need to avoid evacuation, resist oxidative burst and/or breach the epithelial gut barrier [32], [37], [38], [39]. For example, Kuraishi and co-workers [40] have found that loss of Drosocrystallin, a protein involved in the formation of the peritrophic matrix, leads to increased mortality after ingestion of P. entomophila and S. marcescens, but does not seem to play a role in systemic infections. Conversely, systemic infections bypass those defence levels [25] leading, in most cases, to virulence at much lower doses [31] and inducing melanisation responses that are not observed in oral infections [41]. However, besides the local specific response, oral infection may induce, a systemic response [31], [34], [38] although not always [30].
Because it is a model system for both invertebrate immunity [23], [42] and experimental evolution [43], Drosophila melanogaster stands out as the ideal organism to address the evolutionary consequences for hosts of different infection routes. In particular, recent years have witnessed the use of experimental evolution in Drosophila to unravel the evolution of host responses to pathogens [44], [45], [46], [47], [48]. However, all these studies concern host evolution to one specific immune challenge, and hence they do not address how different infection routes affect the host response. In the work here presented, we bridge this gap using experimental evolution on an outbred population of D. melanogaster responding to two routes of infection of the bacteria Pseudomonas entomophila. In brief, we will, (a) compare the rate of adaptation to each challenge, (b) test whether pathogen loads after infection changes with the evolutionary history of populations, (c) address whether adaptation is specific to each infection route and (d) test the generality of the response towards other pathogens.
In figure 1, we present the survival along of the selected and control populations across 24 and 34 generations of experimental evolution, upon exposure to the natural pathogen P. entomophila, by oral (Figure 1a) and systemic infection (Figure 1b).
Both the selection regime and selection regime by generation effects were significant (P<0.0001), either in the BactOral scenario (χ21 = 35.452 and χ217 = 60.522 for the selection regime and selection regime by generation effects, respectively) and the BactSys scenario (χ21 = 16.336 and χ225 = 265.756, respectively).
Upon oral infection, the mean number of live individuals at day 10 after infection rose from the control 33% to a stable 90% after approximately 5 generations (Figure 1a). This rise is quite spectacular in that in only 3 generations the number of alive orally-infected flies had doubled (Figure 1a). Concomitantly, pairwise comparisons at each generation reveal significant differences among selection regimes for this treatment starting at generation 3 (|z|>3.072; P<0.05 for all comparisons beyond that generation). In contrast, selection via systemic infection with the same bacterium, only led to significant differences at generation 13 (|z|>4.160; P<0.001). This difference was consistently significant after generation 16 (|z|>3.887; P<0.01), except for generation 20 (z = 3.065; P = 0.05), The lines selected in presence of the pathogen never exceeded 80% survival (Figure 1b).
Next, we asked whether the increased levels of survival observed after 24 generations of selection corresponded to differences in pathogen loads after infection. For both modes of infection and for the early time point corresponding to the onset of mortality (left bars on Figure 2a and 2b), the profile was the same, displaying a significantly higher number of bacteria in controls relatively to the evolved populations (|z| = 3.287 and 3.430, for oral and systemic infections, respectively, P<0.01 for both comparisons). At the later time point, after which no more death is observed between populations (right bars on Figure 2a and 2b), there were no statistical differences between bacteria titers in the two time points for each of the infection routes (|z|>0.175 for oral and systemic infections, respectively; P = 0.998 for both comparisons). The absolute number of bacteria was significantly reduced between the first and second time points in all treatments and selection regimes (|z|>4.883, P<0.001 for all pairwise comparisons) (Figure 2a and 2b). Under oral challenge, infection-free samples raised from 6/48 to 33/48 in control populations, and from 11/48 to 35/48 in selected populations. As for systemic infection, samples without bacterial counts increased from 0/48 to 11/22 in control populations, and 0/48 to 22/48 for selected populations.
We wondered how much of the adaptation to one route of infection would protect individuals infected through a different route. To address this, individuals of both sexes from control and selected populations were infected by pathogens via each of the two alternative routes of infection at two different time points (generations 14–15 and 24–25).
For both the oral and systemic infection treatments, there was a significant overall interaction effect between sex, selection regime and generation (χ26 = 67.795 and χ26 = 15.420, P<0.0001 and P<0.05 for oral and systemic infections, respectively). We therefore compared the hazard ratios between the selection regime and their respective controls, independently for the two time points and averaging the effect of sex.
Concurrently with the survival data obtained for generations 14–15 and 24–25 in Figures 1a and 1b, evolved populations tested in the conditions in which they evolved (hereafter homologous environment) had a significantly higher survival relative to their controls. This is shown by the significant departure from zero of their hazard ratios (Figure 3: oral infection: |z|>8.003, P<0.001 in both generations; systemic infection: |z|>6.229; P<0.0001 in both generations). In contrast, exposing the adapted populations to the challenge they have not evolved in (hereafter heterologous environment), revealed no difference between control and selected lines for the BactOral selection regime (|z|<1.292, P>0.784 in both generations). For the BactSys selection regime, a significant difference was found in generations 14–15 (in which Bactsys<control), but not in the later generations (|z| = 3.062, P<0.01, and |z| = 0.656, P = 0.939, respectively). Therefore, adaptation to P. entomophila through one infection route infection did not affect susceptibility to the same pathogen infecting from a different route.
Subsequently, we tested whether specificity of the evolved response could extend to other pathogens when infected via the same route (Figure 4).
Hazard ratios between the BactSys and ContSys populations after infection with the closely related species (same genus) P. putida were equivalent to those obtained with the original challenge, P. entomophila (|z| = 6.001 and 8.790, for P. entomophila and P. putida, respectively, P<0.001 in both comparisons). In contrast, challenges with other known Drosophila pathogens such as Serratia marcescens and Erwinia carotovora, also Gram-negative Gammaproteobacteria, or Enteroccocus faecalis, a Gram-positive bacterium, caused equal degrees of mortality between evolved populations and their controls (|z| = 0.670, P = 0.503; |z| = 0.031, P = 0.976 and |z| = 1.374, P = 0.170 for S. marcescens, E. carotovora and E. faecalis, respectively). We therefore conclude that the response obtained is specific, at least, to the Pseudomonas genus level but not for all Gammaproteobacteria. Finally, fly populations evolving with P. entomophila infection were more susceptible than control populations to infections with Drosophila C Virus (DCV) and Flock House Virus (FHV) (|z| = 4.043 and 2.855, P<0.001 and P<0.05 for DCV and FHV infections respectively).
Here, we report the first study addressing the impact of different infection routes taken by horizontally-transmitted pathogens on the evolutionary trajectories and outcomes of their hosts. Our main conclusions are:
Despite using the same pathogen in both infection protocols, we observed a lack of cross-resistance after a heterologous challenge with the same pathogen. Indeed, fly populations adapted to an oral infection by P. entomophila are equally susceptible to a systemic infection by the same bacterium species as populations evolved without the pathogen. The same holds true for populations evolved under a systemic infection challenged with an oral infection. This indicates that the response to each challenge has a different genetic basis.
Several genes and pathways are known to specifically participate in each infection route [23], [25], [33], [40] and our results are compatible with these findings. Yet, both humoral and epithelial responses may lead to the activation of anti-microbial peptides (AMPs) [25], [36], [49]. Moreover, the same pathways may be activated and required for survival in both infection routes. For instance, the Imd pathway has a role in protection against both orally and systemic infection with P. entomophila [38], [50]. Therefore, some of these effector elements could constitute a common target for selection and a general basis for adaptation to the pathogens, irrespective of infection route [51]. This is probably not the case, otherwise we would observe a positive correlation among responses.
A few studies have previously shown that evolution of the response to different pathogens in Drosophila occurs at a rapid pace [44], [46]. Our results confirm this rapid evolution but they also show that the rate of adaptation is contingent upon the infection route taken by this pathogen. Specifically, the increase in survival to oral infection in our fly population occurs within fewer generations than the response to systemic infection, and it reaches a higher plateau. Because this is the first study that compares adaptation to different infection routes, whether these differences in dynamics are a general feature remains to be established. It would be interesting in the future to compare other pathogens that can infect through these different routes.
The observed differences in the evolutionary dynamics of populations exposed to each challenge may be due to the different genetic bases underlying each adaptation process. However, other factors may account for different dynamics. For example, systemic infection may be associated with more environmental variance (Ve) than oral infection. These differences in Ve would lead to the observed differences in dynamics even in the absence of different genetic bases for the traits underlying adaptation to each challenge. Quantitative genetic designs allowing measures of environmental and additive genetic variance for these traits are needed to distinguish between such alternatives.
Interestingly, in our experiments the only aspect in which the adaptive responses to oral or systemic infections are parallel, regards the evolution of resistance (Figure 4a and 4b). Indeed, we find a significant difference between the bacterial counts of control and evolved lines at the onset of mortality for each selection regime. At a later time point (120 h), control and evolved flies have the same bacterial load. However, at this point, we are only measuring bacterial loads in flies that survive infection, hence this information is irrelevant to the clarification of the mechanism involved in the adaptation process. Our results thus reiterate the need to follow the infection dynamics to discriminate between resistance and tolerance. Yet, with our data, we cannot exclude a role for tolerance, as the infected flies from evolved and control populations that survive may have different abilities to cope with the infection (e.g., in terms of fecundity or subsequent mortality). Given that theory predicts different evolutionary outcomes depending on whether host responses involve tolerance or resistance [52], it is important to establish experimentally which of these mechanisms is acting in an evolving population.
The similarity observed among responses to each challenge does not imply an equivalence of mechanisms. The clearance of bacteria in fed versus pricked flies is likely bound to rely upon very different processes [33]. Bacterial loads are much lower in orally infected flies (two orders of magnitude) than in systemic infections (compare panels a and b of Figure 4), despite the fact that in the oral infection treatment the bacteria density administrated was four orders of magnitude higher than in systemic infections, indicating that elimination mechanisms are much more effective in this route of infection. This is consistent with published work showing that oral infection provokes strong epithelial responses namely by the modulation of physical barriers blocking pathogen access to the body cavity and of gut epithelium renewal, and there is limited crossing of the bacteria to the body cavity [33], [40], [41], [53]. In contrast, in a systemic infection the pathogen is inside the body cavity. Thus, any reduction in pathogen loads in the populations adapted to systemic infection must rely on active methods of identifying and eliminating bacterial invaders, namely through the canonical action of AMPs and plasmatocytes [23], [25], [42].
The evolved populations only respond to infections with the bacterium used for selection, P. entomophila, and to its close relative P. putida. Other bacteria cause the same levels of lethality as in controls. This genus-specific response is somewhat surprising in that systemic infection with different bacteria can induce a wide-spectrum of AMPs and other immune responsive genes with large overlaps, yet closely related pathogens induce considerably divergent responses [54], [55], [56]. Other studies using inbred lines have also established a lack of correlation between bacterial loads of different bacteria [57]. Finally, this specific adaptation to the Pseudomonas genus comes at a cost in survival to viral infections (Figure 3). Other studies provide contradictory evidence regarding the existence of trade-offs between susceptibility to different pathogens [54], [58], [59], [60]. This study, however, strongly points to the occurrence of a trade-off, where adapting to one pathogen increases susceptibility to others. This trade-off may underlie the maintenance of variation for resistance to Pseudomonas in the population.
Several studies have shown that infection routes affect the evolution of virulence in pathogens [4], [5], [6], [7], [11]. Here, we show that host adaptation to pathogens is also contingent upon those infection routes. Therefore, host responses may confound the conclusions drawn from studies on the evolution of virulence in pathogens in natural populations. For example, most pathogens that infect invertebrate hosts systemically are transmitted by vectors [14]. Several factors are expected to differentially affect virulence in vector-borne or directly-transmitted pathogens [8], [9], [10]. However, here we show that hosts adapt slower to a systemic than to an oral infection. This may confound the conclusions drawn from the observation of virulence patterns in natural populations. Hence, instead of merely observing patterns, studies on the effect of transmission modes in the evolution of host-pathogen interactions should follow the processes of adaptation in hosts and pathogens separately, to pinpoint the real cause underlying the observed patterns. In this sense, experimental evolution is a powerful, yet underexploited tool to unravel the selection pressures underlying host-pathogen interactions. Our findings reinforce the necessity of including the mechanism of pathogen access into the set of criteria used to categorize and study host-pathogen interactions in ecological immunity, physiology and evolution [14], [16].
An outbred population of Drosophila melanogaster was established in the laboratory in 2007, from 160 Wolbachia-infected fertilized females, collected in Azeitão, Portugal. Variability in this base population was assessed using multiple methods, based on 103 SNPs located in the left arm of the 3rd chromosome (supplementary methods). It contains high and relatively constant levels of polymorphism (SI, Figure S1). The population was kept in the laboratory cages for over 50 non-overlapping generations (generation time: three weeks) with high census (>1500 individuals). Flies were maintained under constant temperature (25°C), humidity (60–70%) and light-darkness cycle (12∶12), and fed with standard cornmeal-agar medium. Prior to the initiation of experimental evolution, the initial population was serially expanded for 2 generations to allow the establishment of 16 new populations used in this work (see below).
Experimental evolution of D. melanogaster populations was performed using Pseudomonas entomophila. In addition, we used other pathogens in some assays, namely, Pseudomonas putida, Serratia marcescens, Erwinia carotovora, Enterococcus faecalis, DCV (Drosophila C Virus) and FHV (Flock House Virus). For each round of infections, bacterial pathogens were grown in LB inoculated with a single bacterial colony, taken from solid medium cultures grown from glycerol stocks kept at −80°C and streaked in fresh (<1 week) Petri dishes. Excluding P. entomophila, grown at 30°C, all bacteria were prepared from an overnight culture grown exponentially at 37°C, centrifuged and adjusted to the desired OD (see below). The P. entomophila strain used for experimental evolution was a generous gift from Bruno Lemaitre. It is resistant to rifampicin, which was used as a marker trait. The remainder bacterial pathogens were generous gifts from Karina Xavier (P. putida), Dominique Ferrandon (S. marcescens) and Thomas Rival (E. carotovora and E. faecalis). Viruses were produced as described elsewhere [61] and aliquots were kept at −80°C and thawed prior to infection.
Lines of all treatments were derived from the same base population (four lines per treatment). Four selection regimes were created, to which the following treatments were applied: systemic infection, in which flies were pricked in the thoracic region [32] with P. entomophila (OD600 = 0.01) (BactSys regime); a control for injection, following the same procedure except that the needle was dipped in sterile LB as a control (ContSys regime); oral infection, in which the food plates were covered for 24 hours with filter papers soaked with a P. entomophila culture (OD600 = 100) diluted 1∶1 with sterile 5% sucrose solution (BactOral regime) (adapted from [41]); and control lines, where flies were kept in standard food (Control regime). The dose of P. entomophila for both bacterial treatments was determined at the start of the selection experiment to cause an average mortality of 66% in the base population, which corresponds to an OD of 0.01 for the systemic and of 50 for the oral infection treatments, respectively (SI, Figure S2).
These treatments were administrated at each generation to 310 males and 310 females (4–6 days old since eclosion). The subsequent generation was founded by all survivors at days 5 and 6 after treatment. The density of eggs was limited to 400 eggs in each cup, a density determined experimentally to enable optimal larval development. Each generation cycle lasted 3 weeks. Absence of transmission of the pathogen to the progeny was tested by plating whole pupae homogenates in LB agar plates supplemented with 100 µg/ml rifampicin. No evidence of transmission of the pathogen to the next generation was found for either infection route, as plating of the progeny of infected flies (pupae) resulted in no P. entomophila colony. Altogether, populations evolved in their specific treatments for 24 generations in the case of the BactOral regime and 34 generations in the case of the BactSys regime.
At each generation, a sample of individuals from each selection regime was used to monitor survival across generations. To this aim, individuals from each replicate population of the BactSys and the ContSys selection regimes were exposed to systemic infection with P. entomophila, whereas individuals from the BactOral and ContOral selection regime were exposed to oral infection with the same bacteria species, and their mortality was monitored in vials for at least 10 days. For systemic infections, 100 individuals were placed in vials of 10 individuals. For the oral infection treatments, 120 individuals were placed for 24 hours in groups of 20 in vials where the food was covered with a filter paper disk soaked in bacteria solution, and subsequently transferred to standard food vials. A mixed sample of 200 individuals of the four populations of the Control selection regimes (ContSys and ContOral) were used as controls in these experiments. To further confirm that persistent infection was not affecting the results, e.g., due to immune priming, at generation 20, these tests were also performed using individuals whose eggs were previously decontaminated in 50% bleach for 2 minutes. Evolved populations showed the same proportion of individuals surviving after infection with/without bleaching.
P. entomophila quantifications were performed in two assays at generations 23 to 25, as described in Nehme et al (2007) [30] with minor modifications. For these assays, 150 to 250 flies (males and females) from each control and selected population were infected as in the survival assays. Flies were collected at 14 and 120 hours after systemic infection for BactSys and ContSys regimes, and at 40 and 120 hours after oral infection, for the BactOral and Control regimes. These time points were selected as the ones that correspond to the point before the onset of mortality in both modes of infection, and to the first day of egg-laying, after which no further mortality occurs (Figure S2). Six replicates of pools of 3 infected flies were homogenized in 50 µL of sterile 1 mM MgCl medium and serially diluted. Homogenates (4 µl) were plated in triplicate on LB agar plates, supplemented with 100 µg/ml Rifampicin and incubated overnight. The next day, we counted the number of colony-forming units (CFUs) on those plates. To avoid possible artifacts due to different maternal effects, flies used in these tests were the progeny of unselected flies that spent one generation in a common environment.
To test how host adaptation to pathogens from one infection route affected the host response to pathogens from a different route, 100 individuals (males and females) from each of the replicate populations of the BactSys and BactOral selection regimes, and the matching controls were exposed to the environment they evolved in as well as to that of the heterologous selection regime (orthogonal assay), following the same protocol of the survival assays, at generations 15 and 25. To avoid possible artifacts due to different maternal effects, flies used in these tests were the progeny of flies that spent one generation without being exposed to pathogens, thus all in the standard environment of the base population.
To test how adaptation to a specific pathogen affected host responses to other pathogens, 100 individuals (males and females) from each replicate population of the BactSys and ContSys selection regimes were systemically infected with the following pathogens: Pseudomonas putida (OD600 = 10); Serratia marcescens (OD600 = 0.01); Erwinia carotovora (OD600 = 150); Enterococcus faecalis (OD600 = 3); DCV (TCID50 = 2×107); FHV (TCID50 = 5×106). These tests were performed between generations 27 and 30, and were repeated at least twice for each pathogen. The protocol followed was the same as that used for the cross-testing experiments. We could not perform this experiment with oral infections because we were unable to find another pathogen that caused mortality in our population via this infection route.
All statistical analyses were done using R (v 2.15).
To compare survival across generations in flies evolving with or without pathogens, the proportion of individuals surviving at day 10 after infection in each vial was first estimated using the Kaplan-Meier method. Individuals alive at the end of the experiment, stuck in the food or escaped from vials during the period of observation were counted as censored observations. Afterwards, the square root of the proportion of surviving individuals was arcsin transformed and analyzed using a general linear mixed model, with sex, generation and selection regime as fixed factors and replicate population as a random factor. To test for the effect of the selection regime, a model with sex and generation as fixed factors was compared with a model with sex, generation and selection line as fixed factors. To test the different effects of the selection line across generations a model with interaction between selection line and generation was compared with the model without this interaction. To compare the proportion of individuals surviving at each generation, each selection regime was contrasted with its control at a given generation and corrected for multiple comparisons using the Bonferroni correction.
To compare survival between the control and selected population in the homologous and in heterologous selection environment, and after infection with different pathogens, we used a Cox's proportional hazards mixed effect model. The model included sex, selection regime and generation as fixed factors and test vials nested into population as random factor, thus accounting for variation in survival rates between populations within each selection line and between vials [62].
To compare pathogen loads, a linear mixed model on the natural logarithm of bacterial counts was employed, with selection regime, time after infection and sex as fixed factors and population as random factor. Interactions among all fixed factors were included in the full model, and sequentially removed if non-significant (P>0.05). These tests were done using the R libraries lme4 (v0.999999, generalized and linear mixed models), coxme (v2.2, mixed effects Cox proportional hazards model) and glht (v1.2, multiple comparisons).
|
10.1371/journal.pgen.1003804 | Identification of 526 Conserved Metazoan Genetic Innovations Exposes a New Role for Cofactor E-like in Neuronal Microtubule Homeostasis | The evolution of metazoans from their choanoflagellate-like unicellular ancestor coincided with the acquisition of novel biological functions to support a multicellular lifestyle, and eventually, the unique cellular and physiological demands of differentiated cell types such as those forming the nervous, muscle and immune systems. In an effort to understand the molecular underpinnings of such metazoan innovations, we carried out a comparative genomics analysis for genes found exclusively in, and widely conserved across, metazoans. Using this approach, we identified a set of 526 core metazoan-specific genes (the ‘metazoanome’), approximately 10% of which are largely uncharacterized, 16% of which are associated with known human disease, and 66% of which are conserved in Trichoplax adhaerens, a basal metazoan lacking neurons and other specialized cell types. Global analyses of previously-characterized core metazoan genes suggest a prevalent property, namely that they act as partially redundant modifiers of ancient eukaryotic pathways. Our data also highlights the importance of exaptation of pre-existing genetic tools during metazoan evolution. Expression studies in C. elegans revealed that many metazoan-specific genes, including tubulin folding cofactor E-like (TBCEL/coel-1), are expressed in neurons. We used C. elegans COEL-1 as a representative to experimentally validate the metazoan-specific character of our dataset. We show that coel-1 disruption results in developmental hypersensitivity to the microtubule drug paclitaxel/taxol, and that overexpression of coel-1 has broad effects during embryonic development and perturbs specialized microtubules in the touch receptor neurons (TRNs). In addition, coel-1 influences the migration, neurite outgrowth and mechanosensory function of the TRNs, and functionally interacts with components of the tubulin acetylation/deacetylation pathway. Together, our findings unveil a conserved molecular toolbox fundamental to metazoan biology that contains a number of neuronally expressed and disease-related genes, and reveal a key role for TBCEL/coel-1 in regulating microtubule function during metazoan development and neuronal differentiation.
| The evolution of multicellular animals (metazoans) from their single-celled ancestor required new molecular tools to create and coordinate the various biological functions involved in a communal, or multicellular, lifestyle. This would eventually include the unique cellular and physiological demands of specialized tissues like the nervous system. To identify and understand the genetic bases of such unique metazoan traits, we used a comparative genomics approach to identify 526 metazoan-specific genes which have been evolutionarily conserved throughout the diversification of the animal kingdom. Interestingly, we found that some of those genes are still completely uncharacterized or poorly studied. We used the metazoan model organism C. elegans to examine the expression of some poorly characterized metazoan-specific genes and found that many, including one encoding tubulin folding cofactor E-like (TBCEL; C. elegans COEL-1), are expressed in cells of the nervous system. Using COEL-1 as an example to understand the metazoan-specific character of our dataset, our studies reveal a new role for this protein in regulating the stability of the microtubule cytoskeleton during development, and function of the touch receptor neurons. In summary, our findings help define a conserved molecular toolbox important for metazoan biology, and uncover an important role for COEL-1/TBCEL during development and in the nervous system of the metazoan C. elegans.
| Metazoans, or multicellular animals, represent the epitome of biological complexity. Prerequisite for generating this complexity was the development of a multicellular lifestyle, and the ability to coordinate cell division, migration and differentiation to optimize the overall fitness of the organism [1]. Multicellularity emerged several times during evolution (in algae, plants, fungi and metazoans); however, the one that emerged in metazoans is notable in terms of the extreme diversity of body plans and differentiated cell types that subsequently evolved [2].
Metazoans form a monophyletic group within the opisthokont lineage, a large taxonomic unit containing fungi and several groups of single-celled organisms, including choanoflagellates, ichthyosporea, filasterea and nucleariids [3] (Figure 1A). The properties of the last common ancestor of metazoans and identity of its closest relatives has been the subject of much debate, due to a lack of fossils and sequence data from a broad range of relevant species. This is currently being addressed by genome sequencing efforts such as the UNICORN project [4]. The release of the complete genome sequence of Monosiga brevicollis, now widely recognized as the closest known unicellular ancestor of metazoans, as well as the genomes of several early branching metazoan species—including Amphimedon queenslandica, Trichoplax adhaerens and Nematostella vectensis—have provided new insights into the genetic developments underlying metazoan evolution (Figure 1A) [5].
A comparison between the genomes of the sea sponge A. queenslandica and the choanoflagellate M. brevicollis has highlighted some of the most important genetic innovations that coincided with the metazoan multicellular transition—including those associated with cell growth, proliferation, adhesion, differentiation and immunity [6]. On the other hand, a consistent trend seems to be that exaptation of pre-existing genetic tools played an important role during metazoan evolution. For example, the genome of M. brevicollis contains genes associated with a multicellular lifestyle [7], while basal metazoans such as A. queenslandica or T. adhaerens, which lack differentiated neuronal cells, possess some proteins specifically required for nervous system function in modern eumetazoans (i.e. ‘true’ metazoans, excluding Porifera, Ctenophora and Placozoa) [6], [8].
Bearing in mind that limited phylogenetic sampling poses a challenge to classification of early-branching metazoan species, it appears as if basal metazoans lack some of the genetic innovations conserved throughout the eumetazoan lineage. For example, the body plan diversity and number of differentiated cell types within known poriferan species is limited, suggesting that the genetic toolkit possessed by sponges, such as A. queenslandica, does not code for the biological complexity displayed by their eumetazoan cousins [6]. Therefore, crucial genetic innovations likely occurred on the eumetazoan stem, after the divergence from a more basal metazoan ancestor, which were essential for the development of differentiated cell types such as neurons and muscle cells, and for overall body plan complexity.
In this study, we sought to uncover genes of central importance to multicellular metazoan biology, and to initiate the analysis of poorly studied or uncharacterized candidates using C. elegans as a model system. Employing a comparative genomics approach, we identified 526 metazoan-specific genetic innovations conserved across 24 metazoan species but absent from 112 non-metazoan, mostly single-celled eukaryotic and prokaryotic organisms. These 526 ortholog groups could be considered a set of core metazoan genes which define metazoan biology, since they are both specific to metazoans and highly conserved. As expected, many previously characterized metazoan-specific genes have functions associated with multicellularity. Numerous genes are also characterized by neuronal expression in C. elegans and a significant proportion of the human orthologs are linked to human diseases. We highlight 54 core metazoan genes whose biological functions are largely unknown and may represent high-priority targets for understanding fundamental animal biology, and for biomedical research. From our dataset, we chose the poorly studied tubulin folding cofactor E-like (encoded by TBCEL in H. sapiens, and coel-1 in C. elegans) as a case study for understanding its metazoan-specific character. Our findings reveal a novel role for C. elegans COEL-1 during development and in neuronal differentiation and maturation, functions that are unique to metazoans.
To uncover the conserved genetic innovations that specifically arose in the metazoan lineage, we identified genes that were highly conserved in metazoans but absent from non-metazoan genomes, including Monosiga brevicollis—the closest known unicellular outgroup to metazoans [7] (Figure 1A,B). Our approach is aimed at uncovering the most inclusive set of metazoan ortholog groups while retaining high stringency, and taking into account different levels of genome completeness [9] (Table S1; see also Materials and Methods). Briefly, ortholog predictions for 138 species, including 25 metazoans, were obtained from OrthoMCL-DB [10], [11]. We divided the metazoan species into their phylogenetic clades, and for an ortholog to be classified as metazoan-associated, we required it to be found in nearly all of the well-sequenced species in each metazoan clade, although potentially missing from a few species in separate clades. Also, we ensured that the genes were found exclusively in metazoans, while accommodating a limited number of falsely-predicted non-metazoan orthologs.
Our comparative genomic analysis identified 526 metazoan-specific ortholog groups (Table S2) conserved in a wide array of metazoan species, including Homo sapiens, Drosophila melanogaster and Caenorhabditis elegans. While we refer here to these genes as metazoan-specific, we acknowledge that some may turn out not to be unique to metazoans as more genomes are sequenced. Some species have undergone gene duplication, resulting in multiple proteins per ortholog group. In the 526 ortholog groups, there were 887 human proteins (1.6 proteins/group) compared to only 577 C. elegans proteins (1.1 proteins/group) (Table S2). It should be emphasized that while the genes in this dataset are highly conserved in eumetazoans, we did not require them to be conserved in the genomes of basal metazoans such as A. queenslandica. Our dataset may therefore contain genetic innovations that evolved in the last common ancestor of eumetazoans, after the divergence of basal metazoans. The recently sequenced placozoan T. adhaerens likely represents a phylogenetic intermediate between sponges and cnidaria, and as such, would be the closest known outgroup to the eumetazoans [12]. T. adhaerens is a morphologically simple organism with only four described cell types, lacking specialized neurosensory or muscle cells found in eumetazoans [8]. Given the simple morphology and putative position of T. adhaerens in the metazoan tree (Figure 1A), we identified ortholog groups that are either present (346) or absent (180) from this genome (Figure 1B, Table S2). The observation that most of our dataset (66%) was robustly conserved in T. adhaerens confirms that it is strongly enriched for core metazoan genetic innovations. We further reasoned that differences between these two groups of genes might provide insights into the evolution of biological processes that are unique to eumetazoans.
To provide global insights into the nature of the metazoan-specific genes, we performed several analyses. Namely, we (a) examined whether the genes are characterized to a significant degree or essentially unstudied; (b) positioned annotated human orthologs into functional categories and pathways; (c) examined RNAi phenotypes from published genome-wide C. elegans studies; (d) evaluated their involvement within interaction networks; and (e) assessed them for a causative role in human disease.
By searching for existing functional annotations of human genes in the UniprotKB database or C. elegans genes in the well-curated WormBase database, we found that 54 of the 526 ortholog groups (∼10%) appear to be completely or largely uncharacterized. This is a smaller proportion than is true for the whole human genome; however, it remains a significant number, considering the highly conserved nature and presumed fundamental biological roles of these genes. We provide this list in Table S3, which includes proteins with a wide range of predicted sequence motifs and domains, and an additional 13 groups that have only been very recently characterized.
As anticipated, human metazoan-specific orthologs with functional annotations were over-represented in functional categories (Figure 1C) and pathways (Table S4) deemed to be important for multicellular animals; these include development, cell-cell communication, and signal transduction. Several organ systems, including nervous, endocrine and circulatory, were also enriched for metazoan-specific genes. In contrast, processes such as DNA replication and repair, transcription, amino acid and carbohydrate metabolism were expectedly under-represented, as most genes implicated in such functions evolved prior to the emergence of the metazoan lineage. In other under-represented categories, such as sensory systems (corresponding in the KEGG database to olfactory, taste and phototransduction), most genes were excluded from our set of core metazoan-specific genes because they are only present in a subset of metazoan species or involve sensory-signaling pathways (e.g., cilium-based) widely conserved across unicellular and multicellular eukaryotes.
When we compared the representation in various categories of metazoan-specific genes with or without a T. adhaerens ortholog, glycan biosynthesis and metabolism was the only functional category that showed a significant difference, containing a greater proportion of genes that were absent in T. adhaerens (Table S4). This is consistent with respect to previous findings suggesting that proteins associated with the extracellular matrix are in some cases eumetazoan (i.e., Cnidaria and Bilatera) innovations [6]. All other categories, including cell-cell communication, signal transduction, development and nervous system, contained similar proportions of metazoan-specific genes with or without a T. adhaerens ortholog (Table S4).
Since T. adhaerens lacks a recognizable nervous system, we looked in more detail at the distribution of core metazoan-specific genes in functional pathways associated with the nervous system, namely neuroactive ligand-receptor interactions and axon guidance. A number of metazoan-specific G-protein coupled receptors (GPCRs) involved in human neuroendocrine pathways have orthologs in T. adhaerens (Figure S1). These include receptors for glycoproteins (FSHR, LHCGR, TSH), neurotransmitters (GRM1-7, GABBR2) and the neuropeptide galanin (GALR). Similarly, we found that many of the known axon guidance molecules are core metazoan-specific proteins, and several are also found in T. adhaerens, including slit, netrin, selected semaphorins, and the receptors robo and eph (Figure S2). However, the axonal guidance machinery is by no means complete in T. adhaerens, since several receptors are missing their canonical ligands and vice-versa. These results show that some genes that became associated with modern eumetazoan functions, such as the nervous system, already existed in the last common ancestor of T. adhaerens and eumetazoans (see also reference [8]).
We obtained additional insights into metazoan-specific gene function by querying genome-wide C. elegans RNAi data. RNAi phenotypes associated with metazoan-specific genes were compared to a control dataset, namely genes with widely conserved eukaryotic orthologs (Figure 1D) (see Materials and Methods for description of the control dataset). Core metazoan genes were collectively less likely to be associated with any particular RNAi phenotype compared with core eukaryotic genes (55% versus 74%). In addition, we classified RNAi phenotypes into three groups: (i) “essential” (embryonic/larval/adult lethality or sterility); (ii) “development” (growth and body shape defects); and (iii) “movement/behavior” (motility and egg-laying defects). Of the genes displaying an RNAi phenotype, the proportion of essential functions was significantly lower for metazoan than for widely conserved eukaryotic genes (56% versus 80%). In contrast, the proportion of genes causing developmental phenotypes and movement/behavior phenotypes was higher for the metazoan-specific group than the conserved-eukaryote group (17% versus 9% and 7% versus 3%, respectively). Therefore, when compared to core eukaryotic genes, core metazoan genes are more likely to cause post-embryonic defects in C. elegans. No differences were apparent when comparing metazoan genes with or without a T. adhaerens ortholog. Taken together, these results suggest that metazoan-specific genes might have emerged not to perform essential functions, but rather, were used to modify or enhance existing cellular pathways. We note that a potential caveat is that RNAi in C. elegans is less effective for genes expressed in neurons, potentially masking important functions of some genes important for this cell type.
To shed light on the functional relationships among core metazoan genes and more ancient eukaryotic genes (using the same control dataset described above), we performed interaction network analyses using the InnateDB database (See Materials and Methods). Our analysis indicates that although metazoan-specific genes are highly connected with each other (60% with one or more connections), they also make extensive connections with ancient eukaryotic genes (40%) (Figure 1E and Figure S3). This supports the idea that the metazoan biological innovations resulted in part from integrating novel components with existing, or evolutionarily more ancient, pathways.
The same evolutionary processes of gene duplication and mutation that drive functional innovation are also to blame for the accumulation of heritable diseases [13]. To estimate what proportion of the human orthologs of metazoan-specific genes are linked to human disease, we queried the curated OMIM (Online Mendelian Inheritance in Man) database. We found that ∼16% (142/887) have a clearly defined link to a documented pathology in humans (Table S5). Diseases linked to the human orthologs of metazoan-specific genes include nervous system disorders (e.g., Parkinson's disease (PARK2/pdr-1, PINK1/pink-1), Alzheimer's disease (APP/apl-1; TAU/ptl-1), and torsion dystonia (TOR1A/tor-1/tor-2/ooc-5)), as well as neoplasms, loss of sensory perception, and several other diseases (Table S5). On the whole, these diseases are associated with the proper regulation of cell proliferation and adhesion within certain tissues, and with the development and function of differentiated tissues such as the nervous system. In light of these findings, it is possible that as many as 8 (16% of 54) of the uncharacterized metazoan-specific genes (Table S3) may represent novel biomedical targets.
Reasoning that the expression patterns of metazoan-specific genes might reveal clues regarding their functions, in particular if restricted to particular tissues, we took advantage of the well-developed tools for examining transgenic expression in C. elegans. We generated promoter-GFP-bearing transgenic lines to analyze the expression patterns of 43 core metazoan genes lacking expression data (Table S6); these were prioritized by their relative lack of functional characterization and whether or not the human ortholog was implicated in disease. We also compared the previously determined expression patterns of core metazoan genes to those of core (widely conserved) eukaryotic genes (Figure 2A). For simplicity, expression patterns were categorized into neuronal, muscle, intestinal, secretory/excretory, hypodermal and reproductive tissues, and quantified using GExplore [14] (See Materials and Methods).
There was no significant difference in the overall proportion of genes with tissue-specific expression between core metazoan and core eukaryotic genes (Figure 2A, left panel). Thus, single tissue-specific expression per se does not appear to be a distinguishing feature of metazoan genetic innovations. However, when we compared the tissue types individually, the proportion of neuronal-specific expression was slightly higher for metazoan genes than eukaryotic genes (9% versus 6%). Fully 16% of our novel metazoan expression patterns were neuron-specific, suggesting a possible bias in our data set or a lack of detailed analysis in genome-wide studies. Examples of such metazoan-specific genes include D2092.5/macoilin, a transmembrane and coiled-coil domain-containing protein, which displays pan-neuronal expression (Figure 2B) and C15C8.4/LRPAP1, a low-density lipoprotein receptor-related protein that is expressed in a subset of specific neurons (Figure 2C).
When assessing expression by tissue type regardless of specificity, we observed that neurons expressed a greater proportion of core metazoan genes than core eukaryotic genes (72% versus 54%) (Figure 2A, right panel). Neurons also expressed 79% of the 43 additional metazoan genes examined in this study. For example, W09G3.7/WBSCR16, a predicted RCC1-like nucleotide exchange factor, is expressed in a pair of sensory neurons, and in the intestine and hypodermis (Figure 2D). C34C12.4/C4orf34, a completely uncharacterized gene with a predicted transmembrane domain, is nearly ubiquitously expressed (Figure 2E), as is F09G2.2/C2orf24, an uncharacterized gene with a cyclin domain (Figure 2F). In contrast, intestinal cells expressed a lower proportion of metazoan-specific genes than core eukaryotic genes, whereas secretory/excretory, hypodermal and reproductive tissues all had similar proportions of both types of genes expressed (Figure 2A). There was no significant difference when comparing genes present or absent from Trichoplax (data not shown). On the whole, our expression data are consistent with the idea that the unique demands of neuronal cell biology were an important raison-d'être for some metazoan-specific genes, and may help reveal other cell-specific functions.
Our expression analysis of metazoan-specific genes uncovered the tubulin folding cofactor E-like gene TBCEL as a potential case study for further analysis in C. elegans. TBCEL (COEL-1 in C. elegans) was first identified based on sequence similarity to tubulin folding cofactor E (TBCE) [15]. The two proteins share UBiquitin-Like (UBL) and Leucine-Rich Repeat (LRR) domains, but TBCEL lacks a cytoskeleton-associated protein-glycine-rich (CAP-Gly) domain present in its counterpart (Figure S4). TBCEL was shown to depolymerize microtubules when overexpressed in cultured cells by committing α-tubulin to proteasomal degradation, while suppression of its activity increased stable microtubule levels [15]. TBCEL is found in all metazoans, including N. vectensis, but interestingly, does not appear to be present in Trichoplax adhaerens; in contrast, its evolutionary precursor, TBCE, is conserved across all eukaryotes (Figure S4). Given its potential ‘housekeeping’ role in tubulin turnover, we expected the C. elegans coel-1 gene to be widely expressed across all cell types. Although expressed broadly during embryogenesis (Figure 3B), its expression became restricted to a subset of neurons during larval development (Figure 3C, D) and adulthood (Figure 3E). Co-expression of coel-1::GFP with odr-2::CFP [16] was observed in the AIZ interneuron (Figure 3D, white arrows), and based on the position of cell bodies relative to those expressing odr-2::CFP, other coel-1-expressing cells in the head of the animal are likely to be the AVK, AIY, AIM and RIB interneurons, the AWC amphid wing cells, SIBV neurons, OLL neurons and URB neurons. During later stages of larval development, transgenic coel-1 expression became even more restricted, so that by the adult stage expression was typically observed in ∼10 neurons in the head, the ALM touch receptor neurons along the body wall, and the PLM touch receptor neurons in the tail (Figure 3E). The neuronal-specific expression pattern of coel-1 was surprising given its presumed general role in microtubule regulation, yet in keeping with a metazoan-exclusive function in C. elegans.
To probe the function of C. elegans coel-1, we obtained several mutant strains predicted to interfere with coel-1 function, including coel-1(tm2136) and coel-1(gk1291), and we also generated an additional allele, coel-1(nx110), using Mos insertion mutagenesis (Figure 3A) (see Materials and Methods). The coel-1(tm2136) allele is predicted to encode a protein with a 77 amino acid deletion in the highly conserved LRR domain (Figure S5). coel-1(tm2136) animals do not show any obvious phenotypes; they are viable, fertile, move normally and have a normal life span (data not shown). Both coel-1(gk1291) and coel-1(nx110) alleles are predicted to remove the C-terminal UBL domain (Figure S5), and also exhibit superficially wild-type development and behavior. Since coel-1 is expressed in touch receptor neurons (TRNs), we carried out a gentle body touch assay. No significant difference was observed between wild-type and either coel-1(tm2136) or coel-1(nx110) animals (Figure 3F).
Given previous reports that cofactor E-like could regulate α-tubulin turnover in cultured mammalian cells [15], we examined α-tubulin levels throughout C. elegans development. Total, steady-state levels of α-tubulin do not appear significantly different in coel-1(tm2136) (Figure 3G) or coel-1(nx110) animals (data not shown) relative to wild-type. We then tested our available coel-1 mutant strains for changes in microtubule function using the microtubule-stabilizing drug paclitaxel (taxol). Eggs hatched on plates containing low doses of paclitaxel arrest their development at larval stages, and the number of animals that escape this arrest to reach adulthood decreases in a dose-dependent manner [17]. All three alleles of coel-1 (i.e. tm2136, gk1291 and nx110) cause a similar degree of hypersensitivity to paclitaxel compared to wild-type animals (Figure 3H). Together, these findings suggest that disruption of coel-1 function does not affect touch sensitivity or modulate global levels of α-tubulin, but that it has an impact on microtubule stability during development.
In an attempt to rescue the paclitaxel hypersensitivity of the coel-1 mutants, and assess the consequence of increased coel-1 levels, we created an integrated strain carrying additional copies of coel-1 (nxIs445), hereafter referred to as coel-1XS (coel-1 ‘excess’). We confirmed a significant overexpression of coel-1 from the coel-1XS allele at the RNA level by qPCR (Figure S6A). Interestingly, coel-1XS animals show a highly variable rate of late-stage embryonic lethality (among individual animals), despite extensive outcrossing. This phenotype has also been observed in animals carrying an extrachromosomal array (nxEx445) of the same coel-1 transgene (Figure S6B). As with coel-1 mutants, no significant change in the total α-tubulin level was observed in the coel-1XS progeny that escape lethality (Figure 3G). However, these animals do have a decreased egg-laying rate (Figure S6C), causing older adults to become full of eggs. We then attempted to test the effect of paclitaxel on coel-1XS animals; however, the variable embryonic lethality hindered our ability to carry out the assay. Contrary to coel-1 mutant animals, coel-1XS worms showed a significant reduction in their response to a mechanical stimulus (Figure 3F). Importantly, injection of coel-1XS animals with coel-1 dsRNA (to reduce coel-1 transcript levels by RNAi) rescued the egg-laying and touch sensitivity defects (Figure S6C, D). Together, these data suggest that the phenotypes observed are due to the overexpression of coel-1 rather than rearrangement of the transgene upon integration or gene interruption at the integration site. The phenotypic analyses show that overexpression of coel-1 causes defects in late embryonic development, egg-laying and touch sensation, and that these defects are more severe than coel-1 disruption. This could be due to the presence of partial redundancy which is able to compensate for lost function, but not able to compensate for vast overexpression. A similar pattern, whereby overexpression is less tolerated than disruption, can be seen with other microtubule regulatory proteins, such as stathmin [18]–[21]. Alternatively, or in addition, the relatively subtle defects caused by the coel-1 alleles studied here may be due to a partial loss of coel-1 function.
The response to gentle body touch in C. elegans is mediated by the touch receptor neurons (TRNs), and represents a well-studied model system for microtubule-dependent neuronal function [22]. The TRNs consist of two anterior lateral ALM neurons, an anterior ventral AVM neuron, two posterior lateral PLM neurons, and a posterior ventral PVM neuron; each has a single anteriorly-directed process extending from the cell body (Figure 4A). The ALMs are born posterior to the pharynx and migrate to the middle of the animal during embryogenesis, while the AVM and PVM are born post-embryonically and migrate to their final positions during the first larval stage. To examine the role of coel-1 in the TRNs, where the gene is expressed, we used a mec-4::GFP reporter which allowed us to visualize the morphology of these cells in living animals.
Measurements of the total length and cell body positioning of the TRNs revealed several subtle changes in coel-1 (tm2136) mutant animals. We observed a significant posterior misplacement of the ALM and the AVM cell bodies in coel-1 animals compared to wild- type (Figure 4B). We also observed an increased length of the PLM neurons in the coel-1 animals (Figure 4C), while cell body positioning was normal (data not shown), indicating an overgrowth of their neuronal processes. coel-1XS animals also exhibit a significant defect in AVM cell body positioning (Figure 4B,F,G). However, in contrast to the overgrowth of PLM processes in coel-1 mutant animals, coel-1XS animals show a reduced outgrowth of AVM processes, which terminate prematurely (Figure 4D,E). In addition, we observed missing or duplicated neurons for the ventral TRNs in coel-1XS animals (Figure 4H,I) and a variety of heterogeneous TRN morphology defects, including disorganized nerve ring branches (data not shown). Similar to other phenotypes detected with the integrated coel-1XS transgene, TRN development defects are also observed in worms carrying the extrachromosomal array (Figure S6C). These defects are probably severe enough to result in the partial touch insensitivity that we measured in animals overexpressing coel-1 (Figure 3F).
AVM and PVM are both descendants of a pair of bilateral Q neuroblasts that each gives rise through asymmetric cell divisions to three different neurons and two apoptotic cells (Figure S6D) [23]. To investigate the possibility that the abnormal ventral TRN cell number observed in coel-1 overexpressing animals is due to a lineage defect, we looked at AQR and PQR, another pair of neurons arising from the Q neuroblast lineage during post-embryonic development [23]. We found a similar AQR/PQR cell number defect as for AVM/PVM, albeit with an opposite frequency with respect to missing versus extra cells (Figure S6E). The inverse correlation between missing and extra AVM/PVM and AQR/PQR cells suggest a Q neuroblast lineage defect, such that when AQR/PQR are not generated, 2 AVM/PVM are made, and conversely.
Taken together, our results demonstrate that the alteration of the wild-type function of coel-1 interferes with normal neurodevelopmental processes that control cell fate, cell migration and neurite outgrowth of the TRNs. We did not attempt to address whether the TRN developmental effects of coel-1 overexpression or disruption were cell autonomous, and it therefore remains strictly possible that the TRN phenotypes we observed are due to coel-1 function in other cells or tissues. However, this possibility is less likely given the observed TRN expression of transgenic constructs under control of the coel-1 promoter and because of genetic interactions with genes that are also expressed in the TRNs (i.e., mec-17 and atat-2, see below).
To address the potential mechanisms behind the morphological and functional TRN defects we observed, we used serial-section TEM to ask whether coel-1 function might affect the structure and organization of the microtubule cytoskeleton in these cells. A unique morphological feature of the TRNs is that they contain 15-protofilament microtubules (MTs) arranged in closely packed bundles along their neurites, while most other MTs in C. elegans have 11 protofilaments and are not specifically organized [22].
To visualize individual protofilaments, we prepared wild-type, coel-1XS, and coel-1(tm2136) mutants using high-pressure freezing and a staining procedure previously developed for this purpose [24]. Using this approach, we found that on average, MTs had the same number of protofilaments (15) in coel-1, coel-1XS and wild-type PLM neurites (Figure 5A). However, serial reconstructions of neurite segments revealed that coel-1XS mutants had significantly fewer microtubules (26±3, mean±s.e.m., n = 4 reconstructions, total L = 8 µm) than wild-type animals (47±4, mean±s.e.m., n = 3, L = 7.3 µm) (Figure 5A,B). In contrast, the number of MTs per section in coel-1(tm2136) mutants was not statistically different from wild-type (35±6, n = 3, L = 7.85 µm). In wild-type animals, MTs were 11.4–20.0 µm in length, consistent with previous estimates [24], [25]. In both the coel-1XS background and coel-1 loss of function mutants MTs appeared to be shorter: 3–13.7 µm and 2.3–14.2 µm, respectively. Thus, coel-1 disruption or coel-1 overexpression appear to decrease MT length compared to wild-type, but the variance in these datasets was too high to infer a statistically significant effect of genotype on MT length (Figure 5C). These results indicate that coel-1 overexpression may reduce the MT content in the TRNs, whereas coel-1 disruption appears to have a more subtle effect on tubulin/MT function. Overall, our TEM data provides a link between cofactor E-like function and neuronal MT homeostasis.
Tubulins are subject to post-translational modifications that participate in fine-tuning the properties of MTs to suit their cellular functions [26]. α-tubulin acetylation at residue K40 is linked to MT stability and function [27]. In C. elegans, the only α-tubulin bearing K40 is MEC-12, and acetylated α-tubulin immunoreactivity is found in TRNs, the nerve ring, the VNC and in some ciliated neurons [28]. Given that the phenotypes associated with altered coel-1 activity are related to MT stability, mechanosensation, neuronal development and MT structure in TRNs, we sought a possible functional link between coel-1 and tubulin acetylation.
α-tubulin acetylation is regulated by the balance between acetyltransferases and deacetylases. HDAC6, which is well conserved in C. elegans, is a histone deacetylase that can deacetylate α-tubulin K40 [29]. We obtained the hdac-6(tm3436) strain, which carries a 476 base-pair deletion spanning exon 4 and intron 4 of hdac-6 and is superficially wild-type. We found that hdac-6 animals are very similar to coel-1 mutant animals. They display subtle TRN morphology defects (posteriorly displaced ALM cell body position and PLM termination sites) (Figure S7A,D) and respond normally to body touch (Figure 6F). The hdac-6 mutation did not alter the subtle TRN morphology defects associated with the coel-1(tm2136) allele (Figure S7A–D). In contrast, the hdac-6 allele partially rescued most of the phenotypes associated with the coel-1XS allele, including TRN morphology defects (Figures 6E,G) and touch insensitivity (Figure 6F). Similarly, the PLM defect associated with the hdac-6 mutation was reduced by the overexpression of coel-1 (Figure S7D).
Two C. elegans paralogs (mec-17 and atat-2) were recently shown to be redundantly responsible for acetylating the α-tubulin MEC-12 at K40 [30]–[32]. Both single and mec-17(ok2109);atat-2(ok2415) double mutant worms have body touch sensitivity defects [30]. As reported by Topalidou and colleagues [33], we found that older mec-17;atat-2 mutant animals display TRN morphology defects. These subtle AVM and PLM defects, caused by coel-1 deficiency, were reduced by the mec-17;atat-2 mutations (Figure 6A,D). In contrast, the acetyltransferase mutations enhanced most of the phenotypes associated with the coel-1XS allele, including TRN morphology defects (Figure 6A,B,E) and touch insensitivity (Figure 6F). Notably, we also observed the extension of the normally small or non-existent posterior ALM process that was suppressed by the coel-1XS allele, but not the coel-1(tm2136) allele (Figure 6C). Altogether, these genetic interactions imply that the activities of coel-1 and tubulin acetylation regulators overlap in the development and proper function of the touch receptor neurons.
Given the genetic interactions between coel-1 and regulators of tubulin acetylation, we assessed the relative amounts of acetylated tubulin in different strains by western blot analysis. As previously reported [30], [31], we found that K40-acetylated α-tubulin is detected in protein extracts from wild-type, but not mec-17;atat-2 mutant worms (Figure 6H). However, no significant change in the levels of acetylated α-tubulin was observed in lysates from hdac-6, coel-1 or coel-1XS animals (Figure 6H). Contrary to the complete loss of tubulin acetylation in the mec-17;atat-2 acetyltransferase-deficient animals, disrupting HDAC-6 appears to have little or no effect on the steady-state level of acetylated tubulin. This could be due, for example, to a partially redundant function with other tubulin deacetylases (e.g., SIRT2; [34]).
In summary, we conclude from our results that C. elegans COEL-1 influences microtubule homeostasis in the TRNs, and that its function in these cells relates to at least a subset of acetylated tubulin heterodimers. The requirements for tubulin acetylation/deacetylation and microtubule homeostasis likely vary during different stages of neuronal development and function, helping to account for the diverse effects on cell fate, cell migration, neurite outgrowth and mechanosensory behavior observed upon altering COEL-1 activity.
We have used comparative genomics to identify a core of 526 ortholog groups widely conserved among and unique to metazoans—the ‘metazoanome’. In contrast to previous studies, which have uncovered between 1147 to 1584 animal-specific gene families [5], [6], [35], we did not try to capture all the novelties in the gene repertoire of the metazoan ancestor. Our approach uncovered a much more restricted number of genes that are presumably critical for metazoan biology, by selecting for genes that were innovations in metazoans but also conserved in nearly all well-sequenced metazoan species within the clade. Since we did not require genes to be absolutely conserved in basal metazoans, our dataset includes some genes that may only have emerged in eumetazoans but were then maintained during the evolution of more complex species. As such, our analysis clearly excluded genes that are metazoan-specific but not highly conserved. This is the case, for example, with genes like mdm2, a negative regulator of p53. Mdm2 is conserved from T. adhaerens to human but is not found in C. elegans or D. melanogaster [36], suggesting that elements of the p53 pathway have been lost because they are dispensable in those organisms.
It should be emphasized that, as shown in our results and incorporated into our methodology, incomplete genomes represent a major challenge for the identification of conserved metazoan genes across diverse species. As a greater number of metazoan genome sequences and gene annotation are more robustly completed, additional genes may be identified as meeting the highly conserved metazoan-specific criteria. This is particularly true for the earlier-branching metazoan clades, where very few species have been sequenced. Therefore, our dataset undoubtedly omits some genes that could be of significant interest and significance to metazoan biology. Conversely, as additional genome sequences are obtained of closely related non-metazoans, some genes may need to be removed from this list. However, with the number and breadth of species that can now be investigated, we feel that we have identified a landmark set of genes that can be described as metazoan-specific or metazoan-associated, even though, for the reasons described above, the absolute specificity of the dataset cannot be conclusively determined.
Our global analysis revealed that the core metazoan genes are proportionately less essential than conserved eukaryotic genes (Figure 1D). We also found that, on the whole, core metazoan genes are deployed in multiple differentiated cell types. Our interaction network analysis suggests that metazoan-specific genes interact with each other and with ancient conserved eukaryotic genes (Figure 1E). Taken together, these results are consistent with the notion that a common property of emergent metazoan-specific proteins is that they evolved as partially redundant modifiers of existing cellular processes—in effect modifying these processes in specific cells to create novel functions in a multicellular context. Our work on C. elegans cofactor E-like, discussed below, illustrates a protein that emerged at the dawn of eumetazoans to influence pre-existing biological processes, including the microtubule cytoskeleton during embryogenesis, as well as differentiation, migration and neurite outgrowth of a subset of neurons required for behavior (mechanosensation).
Of the core metazoan-specific genes we identified, approximately one-third (34%) were absent from the genome of the basal metazoan T. adhaerens, and some of these may therefore have been eumetazoan innovations. However, when we compared those genes with the 346 genes conserved in T. adhaerens, we did not find a significant enrichment in any specific functional categories, with the exception of glycan biosynthesis and metabolism, which was underrepresented in T. adhaerens (Table S4). Glycans are a diverse group of molecules that are important components of the extra-cellular matrix (ECM), which is essential for many aspects of metazoan biology including cell adhesion, differentiation, morphogenesis and immunity [37]. Heparan sulfate proteoglycan (glypican) GPC6/gpn-1 is an example of a core metazoan-specific gene that is not found in T. adhaerens (Table S2). Although it is possible that the absence of GCP6/gpn-1 in T. adhaerens is due to genome incompleteness or a species-specific gene loss, it is equally plausible that its absence, along with other glycans (T. adhaerens does not produce a distinct ECM [38]), reflects an expansion and diversification of genes involved in production and maintenance of the ECM in eumetazoans. Interestingly, the precise roles of glypicans, including GPC6, remain poorly understood. GPC6 has a broad expression pattern which includes the developing brain, and defects in human GPC6 result in omodysplasia (severe limb shortening and facial dysmorphism) [39]. In C. elegans, glypican gpn-1 has been implicated in mediating the proper migration of neuronal precursors [40].
Even in the functional categories associated with higher eumetazoan-specific functions such as the nervous system, the representation of metazoan-specific genes with or without a T. adhaerens ortholog did not differ. For example, orthologs of some neuroendocrine G-protein coupled receptors and axon guidance molecules that have evolved specific neuronal functions in eumetazoans can be found in T. adhaerens, which otherwise lacks a nervous system (Figure S1, S2). This is consistent with an important contribution of exaptation to the evolution of new metazoan traits [6]. T. adhaerens has been shown to exhibit behavioral responses to stimuli [8], and the conserved neuroendocrine pathway components may be part of a primitive stimulus response signaling system that existed in the last common ancestor of T. adhaerens and eumetazoans. Furthermore, genetic analyses in C. elegans have shown that many of the axon guidance genes have other critical functions (e.g., ephrins and semaphorins in epithelial formation) [41]. This suggests that they could have evolved their axonal guidance activities from a more general cell-cell communication system present in basal metazoans prior to the emergence of the nervous system.
Of the 526 core metazoan ortholog groups we identified, approximately 10% are uncharacterized or very poorly characterized. For instance, C4orf34/C34C12.4 encodes a small transmembrane domain-containing protein of 99 residues with no functional annotations in any organism, including human, mouse, fly, fish or worm. Its predicted transmembrane domain architecture and our finding that the C. elegans ortholog has a broad expression pattern, suggest that it may be involved in a metazoan-specific signal transduction pathway or other conserved cellular process. Importantly, a number of core metazoan genes have been recently characterized (Table S3); these genes are associated with the nervous system, developmental function and human disease. For example, consistent with our expression analysis showing pan-neuronal distribution, macoilin/maco-1 has been shown to be involved in regulating neuronal functions in C. elegans [42], [43]. Another recently characterized metazoan-specific gene is TTC19/ddl-3. Its disruption causes mitochondrial complex III deficiency and neurological impairment in humans and flies [44]. These discoveries demonstrate the potential of our dataset as a source of candidate genes for novel functions that may be associated with the nervous system and/or human disease.
In addition, many metazoan-specific proteins, despite having some functional annotation, are poorly characterized and would benefit from analysis from a whole-organism perspective. For example, the protein attractin/mahogany is found in all sequenced metazoans, but its known interaction partner, MC4R, implicated in weight control, has a more restricted distribution [45]. This suggests that attractin/mahogany likely participates in other, potentially broader cellular roles that are yet-to-be-discovered; these could for example be explored in the genetically-amenable metazoan, C. elegans, where RNAi of the gene (F33C8.1) suggest roles in fertility and locomotion that may be relevant to synaptic transmission [46]. In this study, we used C. elegans to investigate the in vivo function of one such poorly characterized metazoan-specific protein, namely tubulin folding cofactor E-like (TBCEL/coel-1).
TBCEL/coel-1 is evolutionarily related to tubulin folding cofactor E (TBCE), which is conserved across all eukaryotes (Figure S4). In cell culture, TBCEL has been proposed to be a MT-destabilizing factor which disassembles the tubulin heterodimer and promotes the targeting of α-tubulin subunits to the proteasome for degradation [15]. As such, it appears to function together with tubulin folding cofactor E (TBCE) as part of a cellular tubulin quality control machinery that includes other tubulin-specific cofactors, and upstream molecular chaperones (CCT and prefoldin) required for protein folding/assembly [47]. Contrary to the previous cell culture studies on TBCEL, we did not observe a significant change in the overall, or global tubulin levels as an effect of cofactor E-like overexpression in C. elegans [15], [48]. This could be due to a more robust function of autoregulatory mechanisms controlling tubulin levels in vivo [49]–[51] or the possibility that COEL-1 acts only on a subset of tubulin isotypes. Regardless, we did find that, consistent with previous observations, C. elegans COEL-1 does regulate microtubule stability in vivo, as indicated by hypersensitivity of coel-1 mutant animals to the MT-stabilizing drug paclitaxel/taxol (Figure 3H) and the reduced MT number per section in PLM neurites overexpressing coel-1 (coel-1XS strain). These data underscore a net destabilizing role for COEL-1. The drug sensitivity phenotype could arise from a role for COEL-1 in embryonic or early larval development, when coel-1 is broadly expressed (Figure 3B). A function for COEL-1 in mitotic cells can be also inferred from aberrant AVM/PVM and AQR/PQR cell numbers observed in coel-1XS animals (Figure 4J–N), which could potentially be explained by a defect in cell polarity and/or asymmetric cell division [23], [52]. It is possible that additional cell types are missing or duplicated in coel-1XS animals. We also found that coel-1 deficiency altered the final position of ALM and AVM neurons, suggesting a defect in the migration of the ALM neuron during embryogenesis, and the AVM neuron during post-embryonic development. Cell polarity, asymmetric cell division and cell migration are crucial for the development of multicellular animals [53], and the function of coel-1 in embryos could impact these processes on a broader scale, which could explain the embryonic lethality observed when it is overexpressed (Figure S6B).
During larval and adult development, coel-1 becomes restricted to neuronal cells, including the TRNs for which we have shown that coel-1 activity influences cell migration and neurite outgrowth. Neurite outgrowth is the result of growth cone migration and guidance, in a process that is largely analogous to cell migration. These biological processes require the correct balance between microtubule-stabilizing and microtubule-destabilizing forces, as well as efficient microtubule-based transport [54], [55]. In fact, defects in neuronal migration and axon elongation have been associated with disruption of MAP1B and TAU, both microtubule stabilizing proteins [56]. Moreover, microtubule destabilizing factors like members of the stathmin family are also involved in neurite outgrowth and cell migration [20], [57]. Our results indicate that coel-1 deficiency and coel-1 overexpression have opposite effects on neurite outgrowth in the TRNs, whereby a subtle overgrowth of the PLM processes in coel-1(tm2136) animals was observed; conversely, animals carrying the coel-1XS allele displayed premature termination of AVM processes (Figure 4C,E). An opposite effect is also observed for cell migration, whereby the ALM cell body seems to migrate further than their wild-type position when coel-1 function is deficient, while the AVM cell body migrates less when coel-1 is overexpressed compared to wild-type. The results are consistent with a microtubule-destabilizing role for coel-1 necessary for proper cell migration and neurite outgrowth.
The overexpression of coel-1 decreases microtubule number, supporting a role of COEL-1 as a microtubule destabilizing factor. A plausible mechanism for this would be a tubulin heterodimer binding and disassembly activity for COEL-1, as demonstrated for the human ortholog [15], which, when overexpressed, would push tubulin partitioning from the polymerized microtubule to the free tubulin heterodimer. The functional connections between TRN morphology, their atypical 15-protofilament microtubules and mechanosensory behavior are not yet clear. However, these questions are being addressed [58]. Topalidou et al. [33] have shown for instance that touch response does not depend on the presence of intact 15-protofilament microtubules. In addition, Bounoutas et al. [59] revealed that the polymerization-state of microtubules can regulate protein expression in the TRNs via the p38 MAPK pathway. Based on the observation of reduced microtubule content in coel-1XS TRN neurites, it is possible that the TRN defects observed upon alteration of this gene are in part, or entirely, a consequence of misregulated expression of factors required for TRN development and function.
There is compelling evidence that the 15-protofilament microtubules of the TRNs are heavily modified by acetylation of α-tubulin at K40. Our data show that both the touch insensitivity and the TRN developmental phenotypes associated with the coel-1XS overexpression allele are partially suppressed by mutation of the tubulin deacetylase hdac-6, and enhanced by mutations in the acetyltransferases mec-17 and atat-2 (Figure 6). Conversely, hdac-6 has no effect on the TRN developmental phenotypes associated with coel-1, while the mec-17;atat-2 mutations suppress them. In addition, the coel-1XS allele, but not the coel-1(tm2136) allele, suppressed the hdac-6 deacetylase PLM phenotype. Collectively, these results suggest that, in the context of the TRN developmental phenotypes, COEL-1 function is antagonistic with respect to tubulin acetylation. However, in light of the recent data showing distinct enzymatic and structural functions for mec-17 [33] as well as the implications of HDAC-6 in multiple processes in the cell [60], not all the genetic interactions we identified may depend on tubulin acetylation. For example, we found that the coel-1XS allele suppressed the posterior ALM process phenotype of mec-17;atat-2 mutants, a phenotype shown to be independent of mec-17 enzymatic function.
We propose a general model for COEL-1 function whereby it binds and disassembles tubulin heterodimers, as demonstrated for the human ortholog [15], to recycle and/or degrade specific α-tubulin species (e.g., tubulin isotypes, modified tubulin, damaged tubulin) in cooperation with other factors. Our data suggests that the effects associated with COEL-1 deficiency are relatively minor compared to COEL-1 overabundance. In coel-1(tm2136) mutants, a reduction in the turnover or recycling rate of specific tubulin species, for example MEC-12 α-tubulin, could have subtle and diverse effects on microtubule function depending on the cellular and developmental context. In contrast, COEL-1 overabundance in coel-1XS animals may exert its more dramatic effect on microtubule stability/polymerization by reducing tubulin heterodimer availability below a critical threshold. The special relationship between COEL-1 function and MEC-12 in the TRNs is supported by our observations that eliminating MEC-12 acetylation (by mec-17;atat-2 mutations) suppressed effects of coel-1 disruption and enhanced effects of coel-1 overexpression in these cells. However, given the broad expression of coel-1 in developing embryos and in a variety of neurons in developing larvae, COEL-1 function may not be specific to MEC-12.
Tubulin, and the cellular machinery that regulates its acetylation, were ancient eukaryotic innovations [32], yet the evolution of microtubule regulatory mechanisms is an ongoing process. The microtubule-associated protein 6 (MAP6/STOP) family, for instance, has emerged relatively recently and is only found in vertebrates [61]. In this study, we identified several microtubule-related proteins as highly conserved metazoan innovations, including microtubule-associated protein TAU/ptl-1, the echinoderm microtubule-associated proteins (EML1-4/elp-1), MIT domain-containing protein 1 (MITD1/Y66D12A.10), microtubule-actin cross-linking factor 1 (MACF1/vab-10), and tubulin folding cofactor E-like (TBCEL/coel-1). We have shown that coel-1 acts broadly (i.e., in most cells of the animal) for a brief period during embryonic development and in a subset of differentiated neurons throughout the lifespan of the animal. Notably, both elp-1 and ptl-1 are expressed in TRNs of mature C. elegans and needed for full touch sensation [62], [63]. ptl-1 is required for both embryogenesis and touch sensation [62], [63]. The apparently dual roles of these microtubule regulatory factors may simply be indicative of an increased demand for remodeling of the microtubule cytoskeleton in these two situations. Furthermore, the elaboration of microtubule regulatory mechanisms may have been an important part of the evolution of metazoan embryonic development and the emergence of a differentiated neuronal cell type.
In conclusion, this study provides a current best estimate of a core metazoan-specific genetic toolkit (the ‘metazoanome’), as well as an overall assessment of some of its features. The detailed analysis of cofactor E-like in C. elegans, revealing a novel role in regulating microtubule homeostasis during development and neuronal differentiation and function, provides an experimentally-validated example of the metazoan-specific character of this dataset. Further study of the uncharacterized or poorly studied core metazoan-specific genes, as well as the interactions between them and with other evolutionarily conserved proteins, should provide important insights into the fundamental biology of multicellular animals and possible targets for neuropathies and other human disorders.
C. elegans strains were maintained and cultured at 20°C under standard condition [64]. Constructs were made using standard molecular biology techniques and fusion PCR as previously described [65]. Alleles used in the study include coel-1(tm2136), coel-1 (nx110), coel-1(gk1291), hdac-6(tm3436), mec-17(ok2109) and atat-2(ok2415).
RT-PCR analysis of the coel-1(tm2136) allele revealed that in contrast to wild-type animals, in which a major transcript of 1300 bp was detected, there were several transcripts of different sizes which were amplified in the coel-1(tm2136) mutant worms. Among those transcripts, only one could potentially encode a protein with a deletion of 77 amino acids which would remove the N-terminal half of the LRR domain.
We generated the nx110 allele of coel-1 using the Mos-mediated mutagenesis [66], we used PCR screen for imprecise excision events from ttTi16961 worms, which contains a Mos1 transposon inserted on chromosome X. This allele corresponds to a deletion of 1763 bp spanning exons 5 to 8. Before use, this strain was outcrossed at least 4 times. RT-PCR analysis of coel-1(nx110) showed that an mRNA is transcribed and is predicted to be translated in a protein lacking the UBL domain in the C-terminal region of COEL-1.
The allele gk1291 corresponds to an 873 bp deletion including exons 5 to 8 and it is predicted to delete the UBL domain in the C-terminal region of COEL-1.
The following integrated reporter transgenes were used: zdIs5[Pmec-4::GFP] expressing GFP in the touch receptor neurons, iaIs19[Pgcy-32::GFP] expressing in AQR, PQR and URX neurons and hdIs26[odr-2::CFP; sra-6::dsRED2] expressing fluorescent proteins in neurons (used for cell identification in the coel-1 expression analysis) [16].
The following transgenes were constructed by PCR and microinjected into worm gonads to generate transgenic lines: (1) nxEx401[Pcoel-1::GFP+dpy-5(+)], expressing an extrachromosomal transgenic array containing 1621 bp of 5′ sequence containing the putative coel-1 promoter fused to GFP; several lines were obtained from this injection and showed the same pattern of expression; (2) nxEx445[coel-1(+)+dpy-30::dsRED] expressing an extrachromosomal transgenic array containing the entire genomic sequence of coel-1, including 1621 bp of 5′ sequence and 778 bp of 3′ sequence; three lines were established and one was used for integration by X-ray integration; (3) nxIs445, the integrated version of [coel-1(+)+dpy-30::dsRED], is presumed to be on the X chromosome as nxIs445 males did not generate any nxIs445 male progeny during 6X backcrossing.
To confirm coel-1 overexpression and for molecular characterization of the coel-1 mutants, N2, coel-1XS, coel-1(tm2136) and coel-1(nx110) cDNAs were isolated by RT-PCR. Briefly, following suspension of mixed-staged worms in Trizol reagent (Invitrogen) and purification with RNeasy (Qiagen), first-strand cDNAs were generated with 1 µg of RNA using the Superscript First-Strand Synthesis System (Invitrogen) with a random hexamer.
For molecular characterization, PCR amplification was performed using primers annealing to the 5′ and 3′ ends of the coel-1 coding sequence. PCR fragments were incorporated into the PGEM-T Easy vector (Promega) and sequenced. To confirm the overexpression of coel-1, real-time qPCR reactions were set-up using the KAPA SYBR FAST master mix (KAPA Biosystem) following the manufacture's protocol. ΔΔCT values were calculated using cdc-42 and ama-1 as housekeeping genes [67].
For RNAi, single stranded RNA was synthesized from PCR product containing flanking T7 and SP6 sites using the RiboMax kit (Promega), annealed and injected at ∼1 mg/ml into both gonad arms of young adult animals.
Standard genetic crosses were used to introduce transgenes into different genetic backgrounds and to make double or triple mutant strains. Single-worm PCR reactions were used to genotype the different mutants. A list of all strains generated and used in this study is available in Table S7.
For generation of GFP-reporter transgenic strains, promoter-containing sequences were fused upstream of the GFP-coding region in the pPD95.67 GFP-coding cassette. The PCR constructs were injected into the syncytial region of the gonad. The final concentrations of the injection mix were 10 ng/µl of the target construct along with 100 ng/µl of the marker construct, pCeh361 (dpy-5(+)) [68], injected into the target strain dpy-5(e907) (CB907). Transgenic F1s (Dpy-5 rescued) were individually plated. Wild-type F2 lines were selected to establish the transgenic lines. When available, we analyzed a second, independent transgenic line.
To examine touch responsiveness, each worm was tested 10 times by alternately touching the anterior and posterior with an eyebrow hair [69]. Wild-type animals respond to anterior touch by moving backwards and to posterior touches by accelerating forward. Any worm failing to move a significant distance was counted as a non-response. 30–60 young adults were tested blind to genotype.
To measure the egg-laying rate, worms were reared at room temperature and staged by alkaline/hypochlorite treatment. Equal numbers of young adult worms of each genotype, including coel-1XS worms injected with coel-1 dsRNA, were placed on separate plates and after 36 hours, transferred to fresh plates and incubated at room temperature. Once the progeny on these plates grew to gravid adults, 60 young adult worms were picked onto 3 separate plates (i.e., 20 worms/plate) for each genotype and the number of eggs laid on each plate was counted after 2 hours at room temperature.
To measure embryonic lethality, one-day adult hermaphrodites were allowed to lay eggs at 20°C and removed from plates after 3 hours. The number of eggs laid were counted and monitored every day until hatching.
Worms were tested for their ability to develop into gravid adults in the presence of the microtubule drug paclitaxel (Sigma, T7191; 10 mM in methanol), essentially as previously described [17] except that worms were grown on solid medium instead of liquid.
Mixed-stage transgenic animals were examined for GFP expression using a Quorum WaveFX spinning disc system. Stacks of confocal images with 0.5–1 µm distance between focal planes were recorded, and image acquisition and analyses were done with the Volocity software package (Improvision). Cells were identified by location and morphology in comparison with reference images from Wormatlas (http://www.wormatlas.org/). Maximum intensity projections of all focal planes were used to generate images for the figures.
Touch receptor neurons of young adult worms were visualized by fluorescence microscopy using the zdIs5 transgene. Both fluorescence and DIC pictures were taken using a Nikon A1R laser scanning confocal system or Zeiss Axioskop 2 microscope. ALM, AVM and PLM total lengths were measured from the cell body to the end of the process. The distances from the vulva to the end of the PLM and from the back of the pharynx to the end of ALM and AVM were also evaluated to examine the site of termination of the processes. Cell body positioning was investigated by measuring the distance from ALM and AVM cell bodies to the back of the pharynx and from PLM and PVM cell bodies to the vulva. To control for individual variation in animal length, the different distances were expressed as ratios with respect to the length of either the anterior body part (tip of the nose to vulva) for ALM and AVM data or the posterior body part (vulva to tail) for PLM and PVM.
We also characterized AVM cell position defect scoring as defective any AVM found at the level of or further than the ALM cell bodies. A premature termination was counted when the anterior process terminates before or at the level of the anterior bulb of the pharynx.
Synchronized worms grown to the appropriate stage on plates at 20°C were collected, suspended in lysis buffer (100 mM Tris, pH 6.8, 4% SDS, 20% glycerol) and treated by 3 cycles of freezing (liquid nitrogen) and boiling. After centrifugation to pellet the insoluble debris, the protein concentration of the supernatants was determined using the BCA protein assay kit (Pierce). Alpha-tubulin and acetylated α-tubulin level were quantitated by fluorescent Western blot. 15 µg of total protein were suspended in Laemmli buffer, separated in 10% SDS-PAGE and electrotransferred to nitrocellulose membranes. Primary antibodies were used at 1∶1000 for the monoclonal mouse anti α-tubulin antibody (Sigma, #T6199) clone DM1A which recognizes amino acids 426–450 of chicken α-tubulins, residues which are highly conserved in C. elegans [70]; 1∶500 monoclonal mouse anti acetylated α-tubulin antibody (Sigma, #T7451); 1∶500 polyclonal rabbit anti actin antibody (Sigma, #A2066). Cy5-conjugated goat anti-mouse (GE Healthcare, #PA45009) and anti-rabbit (GE Healthcare, #PA45011) secondary antibodies were used in 1∶2500. Processed blots following manufacturer's protocols were scanned in a Typhoon Phosphorimager system (Molecular dynamics) and quantitated using ImageQuant software (Molecular Dynamics) or ImageJ software (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://rsb.info.nih.gov/ij/,1997-2009).
Adult nematodes were prepared for EM as previously described [71]. Briefly, animals were frozen using an EMPACT2 high-pressure freezer system (Leica, Vienna, Austria). A Leica AFS freeze substitution apparatus was used to preserve and embed nematodes in 2% glutaraldehyde plus 1% osmium tetroxide and in Eponate 12/Araldite 502. Serial, ultrathin (50 nm) sections were cut with a diamond knife on a Leica Ultracut S microtome and collected on Formvar-coated copper-slot grids. To enhance contrast, sections were post-stained in 3.5% uranyl acetate (30 sec) and Reynold's lead citrate preparation (3 min). The grids were imaged on a transmission electron microscope (JEOL TEM 1230, Tokyo, Japan) and images acquired with an 11 megapixel bottom-mounted cooled CCD camera (Orius SC1000, Gatan, Pleasonton, CA). Images of consecutive sections were aligned manually and analyzed with Reconstruct [72].
Our approach utilized the ortholog prediction tool OrthoMCL [10]. OrthoMCL initially predicts orthologous gene pairs by reciprocal best BLAST hit (RBBH) analysis and then clusters the RBBH pairs into highly connected multi-species ortholog groups. Ortholog predictions for 138 species, including 25 metazoan species were obtained from OrthoMCL-DB version 4 [11]. The metazoan species included 11 Craniate species, 1 Urochordate species, 8 Arthropod species, 3 Nematode species, 1 Cnidarian species and 1 Placozoan species (Figure 1, Table S1).
To obtain a list of metazoan-specific ortholog groups from the OrthoMCL groups, groups were selected using two main criteria. First, groups were required to have sufficient coverage across metazoan species. One of the challenges in identifying conserved orthologs across species is accommodating the varying degrees of genome completeness. To avoid missing valid metazoan-specific orthologs in cases where a potentially conserved gene was omitted due to incomplete genome sequence, we performed a cursory assessment of genome completeness in the metazoan species using a list of widely-conserved, low-copy number eukaryotic genes (Table S1). The presence or absence of these conserved eukaryotic genes provides an approximation of the gene coverage [9]. Coverage was on average 94% in the metazoan genomes used in this study, but was observed as low as 79% in the case of Ciona intestinalis. To accommodate this source of error, we adopted flexible criteria for selecting metazoan-specific orthologs. We divided the metazoan species into their phylogenetic clades and for an ortholog to be classified as metazoan-specific, we required for it to be found in the majority of species in every metazoan clade, but could be missing from a few species in separate clades. Groups had to have at minimum, orthologs in: 9 of 11 Craniate species, 6 of 8 Arthropod species, 2 of 3 Nematode species. In addition, since Nematostella vectensis and Ciona intestinalis were sole representatives of their particular phylogenetic clades, we required genes to be conserved in at least one of these genomes. Finally, ortholog groups had to be conserved in a combined total of at least 20 of 24 metazoan species. These criteria ensured a high degree of conservation throughout the metazoan lineage while permitting a limited number of false negatives (i.e. orthologs missing due to errors by OrthoMCL or incomplete genome information).
Because these modified criteria ensures that the gene is strongly represented in all metazoan clades, it is likely that we are more often selecting for groups where an ortholog is absent due to incomplete genome sequence rather than gene loss. The second criterion ensures that the group's orthologs are found exclusively in metazoan species, while accommodating a limited number of falsely predicted non-metazoan orthologs. OrthoMCL is an effective tool for identifying metazoan-specific orthologs. It does, however, generate small numbers of falsely predicted orthologs (false positives). In the OrthoMCL data, singular non-metazoan genes would occasionally be clustered with a group of metazoan-conserved orthologs. These genes would often only be predicted to be orthologs to one or two other metazoan species genes in the group based on existence of an RBBH relationship. From a phylogenetic perspective, it is more likely that these weakly-related, singleton genes are false positives rather than true orthologs. To prevent these groups from being excluded in our list of metazoan-specific orthologs, a second criterion was added: an ortholog group could have at maximum, orthologs in 2 of 112 non-metazoan species, provided that those orthologs had 3 or less RBBH connections to the metazoan orthologs in the group. These weakly connected, singular non-metazoan orthologs likely represent false predictions by OrthoMCL.
Pathways were obtained from InnateDB [73] and functional categories from KEGG Brite database [74]. To identify pathways and functional categories that were over- or under-represented with metazoan-specific human genes, a hypergeometric test was used. Multiple hypothesis correction was performed using a Benjamini-Hochburg procedure and results were considered significant if the corrected p-value was less than 0.05. The set of genes used for comparison in this analysis comprised all human genes with pathway or category annotations in these databases. Differences in the proportions of human metazoan-specific genes with or without a Trichoplax ortholog in each category were tested using the Fisher's exact test.
To determine the genes that were completely uncharacterized we combined several approaches. First, using Genealacart, the batch querying application based on Genecards database (www.genecards.org) [75], we searched for human genes that had no functional annotation from the UniprotKB database. Second, using wormart, the wormbase implementation of Biomart (http://caprica.caltech.edu:9002/biomart/martview/), we identified the C. elegans genes that had no description. Next, we compared the results between human and C. elegans to obtain a list of uncharacterized genes in those two species. Finally, we searched manually for genes that had no associated papers in Pubmed.
In order to assess metazoan-specific genes characteristics, we compared them with a set of highly conserved eukaryotic genes that were identified on OrthoMCL-DB by selecting genes found in H. sapiens, C. elegans, D. melanogaster, S. cerevisiae, A. thaliana and absent in prokaryotes. The total number of eukaryote-specific ortholog groups is 1004, consisting of 1237 C. elegans and 1630 human genes.
To analyze and quantify the functional interactions of metazoan-specific genes and compare them to conserved eukaryotic genes, we used the InnateDB database (http://www.innatedb.com) [73]. Specifically we used the “data analysis” page to retrieve experimentally-verified molecular interactions for each gene. This database allowed us to obtain interactions only between genes of the same set. We identified interactions of metazoan-specific genes with each other, interactions of conserved eukaryotic genes with each other and compared them with the list of interactions between metazoan-specific and conserved eukaryotic genes. Differences in the proportion of interactions were tested using the Fisher's exact test. Cytoscape [76] was used for the visualization of the interaction network.
Wormart (http://caprica.caltech.edu:9002/biomart/martview/), the wormbase implementation of Biomart, was used to retrieve the RNAi phenotypes associated with the metazoan-specific and conserved eukaryotic genes. Specifically, the WS220 gene dataset was used and the genes filtered by their “gene ID.” For the second dataset, we used the RNAi dataset and the “phenotype” filter.
To identify genes that have RNAi data we then selected the filter “limit to RNAi that have one or more scored phenotype”. To identify genes that are associated with any RNAi phenotype we chose the filter “limit to RNAi that have one or more observed phenotype”. To identify the genes associated to particular phenotypes, the filters “phenotype annotation includes observed phenotype” have been selected and the wanted phenotypes have been entered in the “limit to phenotype ID” filter case. The phenotypes analyzed were Emb = embryonic lethal, Ste = sterile, Stp = sterile progeny, Lvl = larval lethal and Adl = adult lethal corresponding to the “Essential” category; Gro = slow growth, Lva = larval arrest, Dpy = dumpy, Bmd = body morphology defect, Bli = blistered, Slm = slim, Lon = long, Sma = small, Pvl = protruding vulva, Muv = multivulval for the “Development” category and Unc = uncoordinated, Prl = paralysed, Rol = roller and Egl = egg-laying defective associated with the “Movement/behaviour” category.
To identify the numbers of disease-associated genes among the metazoan-specific ones, we queried the OMIM database through BioMart and filtered the output to include only diseases whose molecular basis was known (i.e. containing #3 in the Phenotype map key column of the MIM Gene Map output), as described [13].
The GExplore tool (http://genome.sfu.ca/gexplore/) [14] was used to compare the annotated expression patterns of metazoan-specific and conserved eukaryotic genes in Wormbase. The total number of genes with annotated expression patterns were identified by an ‘expr’ full-text search. For simplicity, anatomical expression patterns were classified as neuronal (nerv OR neuron), muscle (muscle OR myo), intestinal (gut OR intestin), secretory/excretory (gland OR secretory OR excretory), hypodermal (hypoderm OR epiderm) and reproductive (uter OR gonad OR germ OR sperm OR oocyt OR reproduct). This classification scheme necessarily omits certain cell types (e.g., coelemocytes, male-specific sexual organs, etc.) and some overlap exists between categories (e.g., uterine muscle). Nevertheless, due to the limitations of a text search and the variable annotation of expression data, this scheme was deemed sufficient for a crude, global quantitative comparison between sets of genes. Specific patterns of expression (e.g., tissue-specific, combinations of tissues) was determined by simple Boolean searches. Results were plotted as proportions of the number of genes with annotated expression data for each group (i.e., 376 (333 in Wormbase, 43 novel expression patterns in this study) for metazoan-specific genes and 479 for conserved eukaryotic genes). Differences between metazoan-specific and conserved eukaryotic gene sets were tested using Fisher's exact test.
|
10.1371/journal.pntd.0002537 | Isolation of Saint Louis Encephalitis Virus from a Horse with Neurological Disease in Brazil | St. Louis encephalitis virus (SLEV) is a causative agent of encephalitis in humans in the Western hemisphere. SLEV is a positive-sense RNA virus that belongs to the Flavivirus genus, which includes West Nile encephalitis virus, Japanese encephalitis virus, Dengue virus and other medically important viruses. Recently, we isolated a SLEV strain from the brain of a horse with neurological signs in the countryside of Minas Gerais, Brazil. The SLEV isolation was confirmed by reverse-transcription RT-PCR and sequencing of the E protein gene. Virus identity was also confirmed by indirect immunofluorescence using commercial antibodies against SLEV. To characterize this newly isolated strain in vivo, serial passages in newborn mice were performed and led to hemorrhagic manifestations associated with recruitment of inflammatory cells into the central nervous system of newborns. In summary this is the first isolation of SLEV from a horse with neurological signs in Brazil.
| St. Louis encephalitis virus (SLEV), a member of the Flavivirus genus, which includes West Nile encephalitis virus, Japanese encephalitis virus, Dengue virus, and other medically important viruses, is a cause of encephalitis in humans and animals. SLEV is considered endemic in the Americas, and currently there is no vaccine or specific treatment available for controlling of preventing SLEV-induced encephalitis. In this study we describe the first isolation of SLEV from an adult male horse with neurologic disease, which was further characterized by molecular and serological methods. Phylogenetic analysis of a 903 base pairs amplified sequence from partial Envelope (E) gene region indicated that the isolate from the horse was within the cluster of the VB genotype. In addition, inoculation of the SLEV isolate intracranially in newborn mice resulted in circulatory and neurological changes. This is the first report of isolation of SLEV from a horse with neurological disease in Brazil.
| St. Louis encephalitis virus (SLEV) is a mosquito-borne virus that causes human and animal encephalitis in the Western hemisphere. SLEV is considered endemic in the Americas, with encephalitis cases being diagnosed from Canada to Argentina [1]–[3]. There is no vaccine or treatment available for St. Louis encephalitis.
SLEV is a single-stranded positive sense RNA virus, with approximately 50 nm in diameter and a genome of 11 kb. SLEV is a member of the Flavivirus genus in the Flaviviridae family, together with several important pathogens such as West Nile virus (WNV), Japanese encephalitis virus (JEV), Dengue virus (DENV), Yellow fever virus (YFV) and others [4], [5]. Viral life cycle is enzootic and birds are the natural amplifying host [6]. Other vertebrates (e.g. wild animals, horses, and humans) are considered accidental/final hosts [7]–[9].
Human infections with SLEV are mostly asymptomatic. Infected individuals can present mild malaise or flu-like symptoms, especially young or middle-aged patients [6], [10]. Severe cases are clinically characterized by high fever, neurological dysfunction, altered consciousness, and headache; which are accompanied by encephalitis or meningoencephalitis that affects more often the elderly [11]–[13]. Lethality rates in severe cases can reach 30%, and are associated to direct damage to the central nervous system (CNS) [3]. Acute illness can be followed by prolonged convalescence with cognitive and psychosocial deficits for over a year [6], [14]. Disease in wild or domestic animals has not been described, although many species are infected or are serologically positive for SLEV in endemic areas [6], [15]–[19].
SLEV has been detected in Brazil for over 40 years, isolated from arthropods [19] or by serological surveys in birds [20] and mammals [18], [21]. SLEV was isolated from two patients in the Amazon region in 1970's [22], [23] and isolated again from a dengue-suspected patient in Southeastern Brazil, in the early 2000's [2]. Interestingly, SLEV infections in humans were identified in southeast Brazil in the following years, under an outbreak of DENV-3, together with the first a human case of DENV-3 and SLEV co-infection [24], [25].
Here we describe the first isolation of SLEV from a horse with neurological signs in Brazil. SLEV identity was confirmed by molecular and serological techniques, and by inoculation of newborn mice. Our findings highlight the importance of effective arboviral surveillance.
Our animal study followed national guidelines (Law number 11.794, 8/10/2008), which governs the use of animals for experimental procedures. All experimental procedures were approved and complied with the University of Minas Gerais (UFMG) Committee for Ethics in Animal Experimentation (CETEA) regulations, under protocol number 163/2011.
Pregnant female mice were acquired from Centro de Bioterismo (CEBIO) of UFMG (Belo Horizonte, Brazil). Newborn Swiss mice (24 hours old) were used in animal model development experiments. All mice were kept under controlled temperature (23°C) with a strict 12 h light/dark cycle, food and water available ad libitum, under specific pathogen-free conditions, in the animal warren at the Departamento de Clínica e Cirurgia Veterinárias of UFMG.
Brain tissue from horses that presented neurological symptoms before death were sent to Laboratório de Saúde Animal at Instituto Mineiro de Agropecuária (LSA/IMA) in Belo Horizonte, Brazil. Samples that were PCR negative for Rabies virus were sent to Laboratório de Patologia Molecular at UFMG and stored at −80°C. Tissue samples were processed for RNA extraction and screened by nested RT-PCR. There were no tissue samples available for histopathology. Reaction parameters and primers were used as described by Ré and colleagues [26]. Primers used for the initial amplification were SLE 1497 (+) RRYATGGGYGAGTATGGRACAG, SLE 2517 (−) CTCCTCCACAYTTYARTTCACG, and primers for the final amplification were SLE (+) 2002 TGGAYTGGACRCCGGTTGGAAG and SLE (−) 2257 CCAATRGATCCRAARTCCCACG.
SLEV RT-PCR amplicon band was purified from an agarose gel and sequenced in Megabace 1000 sequencer. Edited sequences were aligned by CLUSTAL/W, using the BioEdit program, version 5.09. The resulting sequence was deposited in GenBank (accession number KF718857).
A phylogenetic tree was generated using the Molecular Evolutionary Genetics Analysis software (MEGA - www.megasoftware.net), version 4 (MEGA 4) [27]. The neighbor-joining method was used to generate bootstrap of 1,000 replications using p-distance. Nucleotide sequences were used to perform a similarity search in sequence databases, using BLAST algorithm (http://blast.ncbi.nlm.nih.gov/). Construction of the phylogenetic was based on 39 other SLEV sequences available at GenBank, which were distributed within the following genotypes: IA, IB, IIA, IIB, IIC, IID, IIE, IIF, IIG, III, IV, VA, VB, VI, VII, VIIIa; VIIIB and 3 out-groups: WNV, JEV, and DENV, as detailed in Table 1. The isolated viral strain was categorized within genotypes as previously described [19], [28]–[31].
A tissue fragment of the SLEV-positive brain was homogenized and clarified by centrifugation. The brain homogenate was inoculated on monolayers of the mosquito cell lineage C6/36 and cultures were monitored for cytopathic effect (CPE) daily. Viral stocks were collected for up to five days post-infection and re-inoculated (500 µL of supernatant on a fresh culture) two times, for a total of three passages. The viral titer from supernatant of the third passage was determined by focus immunodetection assay (FIA) as previously described [32]. We obtained SLEV stocks at 1×103 focus forming units, or FFU, per mL of culture supernatant.
To characterize the isolated SLEV strain, 40 µL of SLEV stocks were inoculated in newborn mice by intracranial route (frontal left region of the brain). Homogenates of a pool of brain samples (10% in PBS) from the littermates were used for preparing the inoculum for the next passage. Infected newborn brains were collected at the onset of neurological disease or at day 7 post-infection, to produce new virus stocks or to process for histological analysis. Brain suspensions were passed seven times in newborn mouse brains and relevant experimental controls were maintained. Clinical alterations were assessed daily in newborns after each SLEV passage and representative pictures were taken.
Tissue samples, including brain, kidney, liver, lung, heart, and fragments of the thoracic and pelvic limbs, were obtained from two newborn mice at each SLEV passage, immediately fixed in 4% buffered formaldehyde, processed and embedded in paraffin. Tissue sections (4 µm thick) were stained with hematoxylin and eosin (HE), and examined under light microscopy. Micrographs were taken using a Spot Insight color camera coupled to Olympus BX41 microscope.
C6/36 cells were harvested and seeded in 24-well plates with gelatin-coated coverslips, and incubated at 28°C for at least 3 hours. Cells were infected with a virus stock derived from the 7th SLEV passage in newborn mice, at a multiplicity of infection (MOI) of 1, for 1 h. At 24 hours post-infection, cells were fixed, permeabilized and stained with a mouse anti-SLEV monoclonal antibody (MSI-7, clone 6B6C-1, MAB8744; Merck Millipore, USA) followed by Alexafluor 488-labelled secondary anti-mouse (Molecular Probes, Invitrogen, USA). Experiments were performed with control group with or without primary antibodies. Stained coverslips were mounted in Mowiol 4–88 (Polysciences, Inc., USA) and analyzed using an Olympus BX61WI microscope equipped with a FV300 confocal scanning unit. Images were analyzed with imageJ software.
From a total of 170 brain samples from horses with neurological disease received and analyzed by PCR at the Laboratório de Patologia Molecular at UFMG, one sample was identified as positive for SLEV. This brain sample was obtained in March 2009 (late summer) from a 12 years-old male horse of undefined breed, which died 72 hours after presenting neurological signs. Those neurological signs were described as incoordination, depression, and flaccid paralysis of the hind limbs. The horse came from a farm in Abaeté, countryside of Minas Gerais State, 207 kilometers from the capital, Belo Horizonte. In the same farm there were two additional horses that remained clinically healthy. Importantly, this SLEV-positive brain sample was negative for other pathogens commonly associated with encephalitis in horses, including Rabies virus, Equine Herpesvirus-1, Equine Herpesvirus-4, West Nile virus, Eastern equine encephalitis virus, Western equine encephalitis virus, Venezuelan encephalitis virus, and Sarcocystis neurona.
The SLEV amplicon originated by the RT-PCR reaction was sequenced and deposited in GenBank (Submission ID #1663732), referring to SLEV strain MG150.
Phylogenetic analysis of a 903 bp amplified sequence from partial Envelope (E) gene region [26] indicated that the isolate from the horse was within the cluster of the VB genotype (Figure 1). A higher degree of nucleotide identity (97–98%) was observed among ten SLEV strains from the Brazilian Amazon region, of the VB genotype (BRA-71, BRA-78, BRA-72, BRA-60, BRA-84B, BRA-74B, BRA68B, BRA-84A, BRA-74D, BRA-73E) by comparison with nucleotide sequences previously deposited in GenBank using BLAST algorithm (http://blast.ncbi.nlm.nih.gov/). Among the foreign isolates with a similar level of nucleotide identity (i.e. 98%), one (F72M022), also genotype VB originated in Florida from opossum in 2006 [33].
After confirmation of the virus identity by sequencing, we focused our efforts on isolating SLEV from the horse tissue. A brain fragment from the SLEV-positive horse was homogenized and inoculated in C6/36 mosquito cells, which is a cell lineage suitable for arbovirus propagation. All passages were tested for SLEV by RT-PCR, and were all positive beginning at the second passage. The virus was isolated after three passages, obtaining a SLEV stock of 1×103 FFU/mL of supernatant.
To characterize this newly isolated strain in vivo, we performed serial passages of SLEV by intracranial inoculation of newborn mice, which allowed us to gather some data on MG150 strain pathogenicity. Increase of clinical signs and circulatory changes were associated with increased mortality that reached 100% at the 7th passage (Figure 2). At third passage, edema and necrosis at the distal extremity of the hind limbs and at the tip of the tail were observed in SLEV infected newborn mice (Figure 3). After the fourth passage SLEV infection was associated with behavioral changes ranging from excitability to apathy and neurological changes including tremors, loss of proprioception, and walking in circles, which were accompanied by the same circulatory changes as observed at the third passage (Figure 3). Importantly, neurological changes in the final SLEV passages were accompanied by hemorrhage in the CNS and peritoneum (Table 2). Neither mortality nor clinical signs were observed in uninfected control mice.
Histological analysis indicated hyperemia and discrete multifocal hemorrhage in CNS from all infected newborn mice at the 4th passage. Multifocal mild lympho-hystiocytic inflammatory infiltrate in the leptomeninges and around blood vessels in the brain was observed in one newborn mouse at the 4th passage (Figure 4A). Focal mild neuronal degeneration, and multifocal hemorrhage were also observed (Figure 4B). Circulatory changes such as hyperemia and multifocal hemorrhage were noticed in several organs, including brain, kidney, liver, lung, and heart at the 5th passage (Figure 4C). Newborn mice at the 6th passage also developed hyperemia in the kidney and lung, and multifocal hemorrhage in liver and brain (Figure 4D). Newborn mice at the 7th passage had hyperemia in the kidney, liver, brain, and lung (Figure 4E). No histological changes were detected in uninfected control mice.
Together, these data indicates that SLEV can cause disease in newborn mice. Virus adaptation to the murine host increased after each passage, resulting in neurological and circulatory changes consistent with Flavivirus infection, providing further evidence that our isolated virus strain was indeed SLEV.
Inocula (i.e. CNS tissue homogenate pools) resulting from each passage in mice were submitted for RNA extraction for confirmation of viral detection by RT-PCR (data not shown). The same procedure was performed with organs that had gross changes. Pool of organs, including the liver, heart, kidney, and lung from the 4th to the 7th passage were analyzed. CNS and other organs were positive in all RT-PCR assays, confirming the presence of viral RNA in tissues.
To further confirm SLEV identity, we performed an immunofluorescence assay, using a commercial monoclonal antibody to detect SLEV proteins in cell culture. Monolayers of C6/36 mosquito cells were infected with the 7th SLEV passage, fixed and stained 24 hours post-infection. Infected cells were positively stained with the anti-SLEV antibody, which was not observed in mock-infected cells, confirming the identity of this SLEV strain in an antibody-based test (data not shown).
The first isolation of SLEV from a horse that died due to a neurological disease in Brazil is a significant event. Virus isolation from horse brain tissue, together with molecular and immunofluorescence data, confirms that SLEV was the agent that caused disease and, ultimately, the horse death in this case. To our knowledge, this is the first observation that SLEV can cause disease in wild or domestic animals, which indicate that some aspects of SLEV viral cycle and its ability to cause disease need further studies. Furthermore, a model of newborn mice infection for characterization of SLEV was thoroughly described.
In terms of public health and epidemiology, the first identification and isolation of SLEV in the State of Minas Gerais adds to previous reports regarding SLEV detection in Brazil [2], [19], [21], [23], [24], and neighboring South American countries [34], [35], which strongly indicates that SLEV circulates in Brazil. Importantly, for SLEV epidemiological surveillance purposes, dengue is endemic and is an important health problem in Brazil. Antigenic similarity between SLEV, DENV and other flaviviruses, especially in terms of their envelope protein, generates cross-reactive antibodies that make serological detection of SLEV infections problematic, especially during the frequent dengue outbreaks [36]. Furthermore, SLEV infection can cause febrile illness and even hemorrhagic manifestations that are indistinguishable from mild and severe dengue fever cases, respectively [37]. In spite of these difficulties, molecular screening methods are available and could be employed to monitor SLEV circulation, allowing for preparedness in case of virus re-emergence or SLEV encephalitis outbreak, which has taken place in Argentina [38] and several times in the United States [39]. Considering that the SLEV strain isolated in this study came from a horse with neurological signs, and that the virus was able to induce systemic and neurological sings in mice, the virulence of the circulating strains should be evaluated. A serological survey involving five Brazilian states, including Minas Gerais, resulted in a prevalence of 36% with a total of 753 horses sampled [40].
Despite the existence of eight lineages and fifteen subtypes of SLEV, namely IA, IB, IIA, IIB, IIC, IID, IIG, III, IV, VA, VB, VI, VII, VIIIA, and VIIIB, phylogenetic studies based on the E gene indicate that genotypes I and II are found predominantly in North America, whereas genotypes III to VIII have been isolated in South and Central Americas [19]. Brazilian SLEV isolates have been classified within the genotypes II, III, V (A and B), and VIII (A and B). Genotypes V and VIII are predominately Brazilian Amazon region, whereas genotypes II and III have been isolated in the State of São Paulo (Southeast Region). In this study, the isolate from a horse had a higher degree of identity (97–98%) with the VB genotype, suggesting that this sample was likely originated from the Brazilian Amazon Region. The circulation of SLEV from the Amazon Region in the Southeast Region of Brazil suggests a possible involvement of migratory birds in disseminating the virus, since SLEV has been detected in 49 species of wild birds in Brazil, many of which are migratory [19], [21]. SLEV strains genotype VB were isolated since 1960s from wild birds and mosquitoes or sentinel animals at a surveillance site for arboviroses in a forested area of Pará state [19]. Migratory birds may have also been related to the periodic introduction of South American SLEV genotype V in Florida (USA) in 2006, originated possibly from Brazil, Mexico, or Panama [33].
In our efforts to characterize SLEV strain MG150 in vivo, to study its pathogenicity, we noticed the virus progressively adapted to serial passagens in newborn mice. The mouse has been previously used as a model for assessing SLEV virulence [41]. Disease presented by newborn mice infected with the last SLEV passages had some similarities to human disease, such as the development of hemorrhagic manifestations [19], mortality, and neurological changes [12], [14]. Importantly, lesions in organs such as the liver and heart are likely to reflect a systemic circulatory change rather than any specific viral tropism for these organs.
|
10.1371/journal.ppat.1004883 | The Proteome of the Isolated Chlamydia trachomatis Containing Vacuole Reveals a Complex Trafficking Platform Enriched for Retromer Components | Chlamydia trachomatis is an important human pathogen that replicates inside the infected host cell in a unique vacuole, the inclusion. The formation of this intracellular bacterial niche is essential for productive Chlamydia infections. Despite its importance for Chlamydia biology, a holistic view on the protein composition of the inclusion, including its membrane, is currently missing. Here we describe the host cell-derived proteome of isolated C. trachomatis inclusions by quantitative proteomics. Computational analysis indicated that the inclusion is a complex intracellular trafficking platform that interacts with host cells’ antero- and retrograde trafficking pathways. Furthermore, the inclusion is highly enriched for sorting nexins of the SNX-BAR retromer, a complex essential for retrograde trafficking. Functional studies showed that in particular, SNX5 controls the C. trachomatis infection and that retrograde trafficking is essential for infectious progeny formation. In summary, these findings suggest that C. trachomatis hijacks retrograde pathways for effective infection.
| The important human pathogen Chlamydia trachomatis causes 100 million new infections each year world-wide. It replicates inside the infected host cell in a unique vacuole, the inclusion. Currently, the nature, and specifically the protein composition of the inclusion, is poorly defined. Here, we described the host cell-derived inclusion proteome by quantitative proteomics using a newly established method to purify inclusions from infected epithelial cells. We showed that the inclusion is a complex intracellular trafficking platform that is well embedded into the organellar network and interacts with host cells’ antero- and retrograde trafficking pathways. Particularly, SNX1, 2, 5 and 6, components of the retromer, are recruited to the inclusion and seem to control the infection. We found also that retrograde trafficking is essential for Chlamydia progeny formation. Our study provides new insights into how the obligate intracellular bacterium C. trachomatis interacts with the eukaryotic host cell and subverts host cell functions for productive infection.
| With 100 million new infections per year, Chlamydia trachomatis is the most frequently sexually transmitted bacterial pathogen world-wide [1]. C. trachomatis replicates inside a membrane-bound vacuole, the inclusion, and has a unique cycle of development, alternating between two distinct bacterial forms. The elementary body (EB) is spore-like, infectious but non-dividing. In contrast, the reticulate body (RB) is non-infectious but replicative. After internalization of the EB, the bacteria are found inside the inclusion, which is segregated from the lysosomal degradation pathway. EBs then differentiate into RBs, which replicate inside the growing inclusion. At mid-infection time points the inclusion is packed with replicating RBs that start to re-differentiate into EBs [2]. The surrounding inclusion membrane is the interface between the bacteria and the host cell. This membrane is actively modified by insertion of bacterial proteins and is not permissive for diffusion of molecules of 520 Da and larger [3]. It contains classical bacterial inclusion proteins of the Inc-protein family as well as non-classical Inc proteins [4]. Furthermore, a growing number of cellular proteins have been described to associate with the Chlamydia inclusion, but a global picture of proteins contributing to the inclusion is currently missing.
Membranes compartmentalize the eukaryotic cell into different organelles, including those of the secretory pathway and the endo-lysosomal system. In the secretory pathway, cargo is modified to address it to and then to transport it to its designated destination. The endo-lysosomal system functions in internalization of molecules from the plasma membrane (PM) or the extracellular space, followed by sorting of these molecules either for degradation in the lysosomes or for retrograde transport to different organelles, including the Golgi apparatus (GA). The human retromer is a multi-protein complex essential for recycling of cargo receptors into the tubular endosomal network and transports them to the trans-Golgi network (TGN) [5]. In human cells, the retromer consists of a membrane-deforming and a cargo recognition subcomplex, which are composed of the sorting nexins (SNX) 1, 2, 5, 6 and the vacuolar protein sorting-associated proteins (VPS) 26, 29, 35, respectively [6]. On endosomes, SNX dimers bind to phosphatidylinositol phosphates (PIPs) via their phox homology (PX)-domains. Additionally, these SNXs contain a Bin-Amphiphysin-Rvs (BAR) domain that recognizes membranes with high curvature and induces membrane tubulation, which is thought to support sorting of retrograde receptors out of the endo-lysosomal pathway [7]. Interaction with the cargo recognition subcomplex eventually leads to vesicle formation and the enclosed cargo is transported along microtubules to the TGN [8,9].
Proteomic studies of phagosomes isolated using latex-beads have greatly increased our knowledge about the biogenesis and function of these organelles [10–12]. Furthermore, the biochemical purification of vacuoles containing Salmonella enterica, Mycobacterium avium, Rhodococcus equi and Legionella pneumophila also fostered our understanding of the host cell protein composition of these unique intracellular compartments [13–17].
Here, we describe a two-step protocol for the isolation of high purity C. trachomatis serovar L2 inclusions at mid-cycle. Using LC-MS/MS based proteomics combined with ss isotope labeling by amino acids in cell culture (SILAC), we identified 351 host cell proteins that are significantly enriched in the proteome of isolated inclusions, representing the host cell-derived Chlamydia inclusion proteome. Enrichment analysis of this data showed that the C. trachomatis inclusion is a complex intracellular compartment that interacts with components of the retromer. Confocal studies confirmed the recruitment of SNX1, 2, 5 and 6 to the inclusion and further suggested that the retromer subcomplexes are at least partially separated at the inclusion membrane. Functional analyses of the retromer by RNA interference and by treatment with Retro-2, an inhibitor of retrograde transport of toxins and viruses, revealed that knockdown of SNX5 resulted in an increase in infectious progeny whereas Retro-2 treatment inhibited the formation of infectious bacteria. Taken together, these results show a previously unknown association of SNXs with C. trachomatis inclusions and provide evidence for a new role of SNXs during bacterial infections that appears to be independent of the classical SNX-BAR retromer complex.
We established an isolation method for C. trachomatis inclusions at mid-infection time points, based on a two-step protocol originally described for the isolation of Legionella-containing vacuoles from amoebae (Fig 1A) [16]. Infected HeLa cells were lysed and the obtained cell lysate containing inclusions was separated on a self-forming Percoll gradient. Gradient fractions were taken and analyzed for presence of bacterial and cellular proteins by immunoblotting and for presence of intact inclusions by phase contrast microscopy (S1A and S1B Fig). The high density fractions harboring intact inclusions (S1A and S1B Fig) were collected, pooled and further purified by magnet assisted cell sorting (MACS) using an antibody specific for IncA, a bacterial transmembrane protein located in the inclusion membrane [18]. Presence and numbers of inclusions were monitored by phase contrast microscopy (Fig 1A and 1B). Counting of visually intact inclusions at each purification step showed that ~50% of C. trachomatis inclusions present in the cell lysate could be isolated (Fig 1B). The purity of the different fractions was assessed by immunoblotting, using antibodies specific for marker proteins of different cellular compartments and for chlamydial proteins (Fig 1C). Lysate of infected and uninfected HeLa cells showed presence of organelles such as the nucleus, endoplasmic reticulum (ER), lysosomes, mitochondria, cytosol and the PM (Fig 1C). After separation by Percoll gradient, inclusions were enriched as indicated by an increase in IncA and Hsp60 signals, accompanied by a decrease in signals for cellular compartments. MACS purification resulted in a fraction that contained chlamydial inclusions that were nearly completely devoid of cellular contaminants as monitored by immunoblotting (Fig 1C). Obtained inclusion fractions were then analyzed by electron and fluorescence microcopy (Fig 1D and 1E). Transmission electron microscopy (TEM) demonstrated the presence of inclusions that contained both bacterial forms surrounded by the inclusion membrane (Fig 1D). To validate the presence of cellular proteins in the isolated inclusion fraction, inclusions were purified from cells expressing a Rab11A-eGFP fusion protein that is known to be associated with C. trachomatis inclusions [19]. Immunofluorescence (IF) staining and confocal microscopy of isolated inclusions revealed that Rab11A-eGFP signal co-localized with IncA in a rim-like pattern (Fig 1E). In summary, these data show that we are able to isolate C. trachomatis inclusions at mid-infection time points.
To identify host cell proteins specifically associated with isolated C. trachomatis inclusions, SILAC was applied [20]. Using this method, we were able to control for non-specific, co-purifying proteins during the isolation procedure (Fig 2A). The proteins that are bona fide constituents of the inclusion were expected to have a high ratio of L label vs. H label (SILAC ratio) of one peptide species, whereas contaminants were expected to have SILAC ratios close to 1 in the inclusion fraction (Fig 2A). The abundance of inclusion-associated proteins in enriched fractions and proteins in total cell lysates was calculated using iBAQ (intensity based absolute quantification) which estimates the abundance of proteins based on the sum of peak intensities of all peptides matching to a specific protein, divided by the number of theoretically observable peptides [21]. Despite limited accuracy, this method provides additional information especially for highly abundant proteins in addition to the SILAC based exclusion approach. Based on this method, we quantified the relative contribution of each protein to the total proteome of the lysate and the inclusion using sum total normalization for the proteins in each fraction. Only proteins that passed the SILAC exclusion approach were considered for the inclusion proteome. The quotient of the values for the inclusion and the lysate resulted in the enrichment score for proteins which were overlapping in the two datasets (iBAQ enrichment score) (Fig 2B and S1 Text). For proteins that were not found in our lysate proteome, we used a recently published very high coverage dataset of the HeLa proteome [22] for approximation of the protein abundance in the cell lysate. We performed experiments in three biological replicates. Analysis of the raw data by MaxQuant resulted in the identification of 1400 host cell proteins in the inclusion fraction (Fig 2C) and 2002 host cell proteins in the cell lysate. To characterize potential organellar contaminants, subcellular localization data of all proteins in the inclusion fraction was retrieved from UniprotKB [23] and annotations were plotted according to their SILAC ratios (Fig 2D). This data clearly showed that proteins from mitochondria, the nucleus and the PM appeared at SILAC ratios of 1 and lower, and therefore are most likely contaminants of the inclusion fraction. The majority of proteins annotated with the terms cytoplasmic vesicle, ER, ER-Golgi intermediate compartment (ERGIC), GA and lysosome were separated from the contaminants with a SILAC ratio above 1.5, demonstrating an enrichment of these proteins in the inclusion isolation procedure of infected cells vs. uninfected cells (Fig 2D). Statistical testing based on the SILAC ratio distribution in the lysate and in the inclusion fractions revealed 351 host proteins that were significantly enriched in the inclusion fraction, of which 253 were highly reliable due to the presence of high ratios in all three replicates, resulting in small multiplicity adjusted p values of below 0.01 (S2A Fig). An additional 98 proteins were qualified as enriched with reduced statistical confidence (multiplicity adjusted p value < 0.05, S2B Fig). These 351 host proteins are thus considered to be inclusion associated (S1 Table). Of the approximately 50 host proteins known to be recruited to Chlamydia inclusions, 23 were identified in our analysis (S2 Table). These proteins included 14-3-3 ß, CERT, VAP-A, VAP-B, Rab1, Rab6A, Rab11A and Rab14 [19,24–27]. These known inclusion-associated proteins were distributed across the SILAC ratios, further increasing our confidence in the generated inclusion proteome data set (Fig 2C).
We next validated the obtained data by confocal microscopy. To this end, 26 newly found inclusion-associated proteins with different SILAC ratios were chosen. Proteins of interest were either detected after ectopic expression of tagged fusion proteins or by visualizing endogenous proteins using specific antibodies (S3 and S4 Figs). Non-fused eGFP was used as control. Localization of these proteins in infected cells was assessed after IF staining counterstained with an IncA-specific antibody to visualize the inclusion membrane and were then analyzed by laser scanning confocal microscopy (LSCM) (Figs 2E and S3 and S4). To confirm the presence of the fluorescently tagged proteins in the inclusion fraction, inclusions were also isolated from cells transiently expressing the respective fusion proteins (Figs 2E and S3). In total, 26 proteins were included in the validation process. From these 26 proteins, 19 proteins were validated positively, either by inclusion isolation or by immunofluorescence microscopy. Among these positive hits were YFP-RAB3D wild-type, VCP-eGFP, eGFP-SYNGR2, eGFP-Rab8A, GFP-Syntaxin 7, STIM1 and Sec22b. As expected, no co-localization of eGFP was observed (S3 Fig). Five proteins were evaluated as false-positive including eGFP-Cofilin-1, Sequestosome-1 and Arginase-1 (S3 and S4 Figs). For two proteins the localization to the inclusion as monitored by fluorescence microscopy was ambiguous (S3 and S4 Figs). Furthermore, recruitment of Rab3D appears to be an active process, as the dominant negative form of Rab3D (YFP-RAB3D T36N) was not found at the inclusion (Fig 2E). Taken together, we have identified 351 host cell proteins that are significantly enriched in the isolated inclusion fraction and thus contribute to the host cell-derived inclusion proteome.
Based on this core host cell-derived inclusion proteome, we analyzed the contribution of cellular organelles to the proteome of isolated inclusions. Subcellular localization data of the identified proteins was retrieved from UniprotKB to calculate the relative contribution of different organelle types to the obtained proteomes. We observed a clear enrichment of proteins annotated as components of the ER, the PM, the ERGIC, the GA, endosomes and cytoplasmic vesicles (Fig 3A). As expected, relative depletion was seen for proteins annotated as nuclear and mitochondrial (Fig 3A).
Next, we performed a gene ontology (GO) enrichment analysis based on GO of biological processes (GOBP) (S3 Table). The most highly enriched single term apart from ER-specific processes was `establishment of protein localization´ (GO:0045184) with a p value of 3.94 x 10–13 and a total of 86 proteins contributing to this category. Proteins from this term were analyzed for specific complexes of interacting proteins using STRING 9.1 [28]. This interaction map revealed four clusters of highly interacting proteins including a cluster composed of the SNX-BAR retromer, a complex involved in retrograde trafficking from endosomes to the TGN (Fig 3B). The most granular (i.e. highly resolved) GO term apart from ER-related processes was `vesicle-mediated transport´ (p = 1.66 x 10–10, GO:0016192, n = 58; n = 72 including child terms). To further characterize these trafficking pathways that are putatively involved in the function of the inclusion, we analyzed the contribution of proteins involved in anterograde and retrograde transport to the proteome (Fig 3C). Proteins involved in retrograde trafficking constitute 39% of these proteins, with retrograde transport from endosomes to the GA being the largest group within the retrograde trafficking group (17% of total). Strikingly, components of the human retromer were highly enriched in the host cell-derived inclusion proteome compared to total cell lysates, including proteins of the SNX family and the retrograde-transport cargo protein Ci-M6PR, which are among the 25% most highly enriched proteins (Fig 3D). In summary, the host cell-derived proteome of C. trachomatis inclusions reveals a complex intracellular compartment enriched for SNX-BAR retromer and suggests that the inclusion interacts with multiple cellular trafficking pathways, including this retrograde transport pathway.
Based on the high enrichment of retromer components on C. trachomatis inclusions, we performed IF studies using antibodies specific for SNX1, SNX2, VPS35 and Ci-M6PR to confirm localization of these proteins to the inclusion using LSCM (Figs 4A and S5A). SNX5 and SNX6 localizations were analyzed after ectopic expression of eGFP-SNX fusion proteins (Figs 4B and S5B). In uninfected HeLa cells, signals for SNX1 and SNX2, were found in punctuated structures in the cytosol consistent with the reported endosomal localization of these SNXs (S5A Fig). In contrast, in C. trachomatis-infected HeLa cells, SNX1, SNX2, eGFP-SNX5 and eGFP-SNX6 were detected as a rim-like staining pattern that partially co-localized with the bacterial inclusion marker, IncA (Fig 4A and 4B). Recruitment of these SNXs was specific, as other members of the SNX family (SNX3 and SNX12) did not co-localize with the inclusion membrane (S6 Fig). Furthermore, these SNXs were also found in IncA-positive fibers emanating from the inclusion body (Fig 4C). Interestingly, VPS35 and Ci-M6PR did not show a rim-like inclusion-staining pattern, but rather were depicted as small punctuated structures adjacent to the inclusion, suggesting that the membrane-deforming and receptor-recognition subcomplex of the human retromer are at least partially disconnected at the inclusion (Fig 4A). To confirm the separation of these two subcomplexes, SNX2 and VPS35 were simultaneously localized in infected and uninfected cells (Figs 4D and S7). Interestingly, at the inclusion, a separation of the two signals was observed. Co-localization of the two signals in defined punctuated structures at the inclusion was rarely seen (Fig 4D). In contrast, in uninfected cells, signals for both subcomplexes were clearly co-localized (S7 Fig). Pearson's correlation coefficient also suggested only a moderate co-localization of the two signals at the inclusion, whereas a strong correlation was detected in punctuate-structures in the cytoplasm of either infected or uninfected cells (S7 Fig). To avoid artifacts due to overexpression of eGFP-SNX2, we also performed experiments in cells expressing eGFP-VPS35 and stained for endogenous SNX2 (S7B Fig), confirming that the retromer subcomplexes do not co-localize at the inclusion, indicating separation or dissociation of the retromer complex. No difference in protein abundance for all tested retromer components was detected in C. trachomatis-infected cells compared to control cells (Fig 4E). These observations demonstrate that during C. trachomatis infection SNX-BAR proteins become recruited to the inclusion and the localization of the two retromer subcomplexes is dramatically changed.
Given that SNX-BAR proteins of the retromer are recruited to the C. trachomatis inclusion at 24 h p.i., we tested whether knockdown of retromer components by RNA interference (RNAi) affects C. trachomatis infection including inclusion formation and development of infectious EBs. We used pools of small-interfering RNAs (siRNAs) to target SNX1, 2, 5 and 6. Silencing of these proteins did not affect the formation of inclusions as analyzed by inclusion size and numbers (Fig 5A and 5B). Interestingly, silencing of SNX5 resulted in a clear increase in infectious EBs compared to control transfections (Fig 5C). SNX1, 2 and 6 knockdown also increased infectious progeny, albeit only marginally (Fig 5C). Genome copy numbers upon silencing of the different SNX proteins were slightly affected, showing the strongest increase in genome copy numbers in SNX5 knockdown cells (Fig 5D). Immunoblotting confirmed that upon knockdown, the targeted SNX-BAR proteins were drastically reduced compared to control treated cells (S8A Fig). We confirmed published data that silencing of SNX5 also resulted in a decrease in protein level of SNX1 (S8A Fig). To elucidate if the observed increase in infectious progeny in SNX5 knockdown cells is dependent on co-regulating the abundance of the other SNX proteins, we silenced SNX5 in combination with SNX1, 2 or 6 and measured infectious progeny formation (S8B Fig). Number of infectious bacteria was increased under all combinational knockdown conditions compared to control, suggesting that other SNX proteins do not contribute to the observed increase in infectious progeny formation in SNX5 knockdown cells. Knockdown efficiencies in these double knockdown cells were confirmed by immunoblotting (S8A and S8C Fig). Taken together, these results suggest that individual SNX-BAR proteins might have distinct functions in addition to controlling the retrograde transport of specific receptors. SNX5 in particular might be a rate-limiting factor and involved in intracellular replication of C. trachomatis, most likely independently of the other SNX-BAR retromer components.
Retro-2 was identified in a high-throughput screen for small molecules that inhibit the toxicity of the plant toxin ricin in cell culture and was additionally found to efficiently protect cells from secreted bacterial toxins, including Shiga-like toxin and cholera toxin by inhibiting retrograde trafficking of these toxic agents from the endosomes to the GA or the ER without affecting trafficking of endogenous cellular retrograde-transport cargo proteins including Ci-M6PR [29].
SNX1, SNX2 and eGFP-SNX5 recruitment to the inclusion was detected starting from 12 h p.i. Interestingly, association of eGFP-SNX6 with the inclusion was detected slightly later (S9 Fig). At 16 h p.i. all inclusions were positive for the four different SNX proteins, coinciding with the expansion of the inclusion (S9 Fig). Taking this into account, we treated cells prior to SNX recruitment (8 h p.i.) with different concentrations of Retro-2 and assessed the formation of infectious EBs by re-titration at 48 h p.i. Treatment of C. trachomatis-infected cells with Retro-2 resulted in a dose-dependent decrease by more than one order of magnitude in EB numbers compared to the vehicle control (Fig 6A). Reducing the treatment duration from 40 h to 28 h by shifting the time point of Retro-2 addition to 20 h p.i. still showed a decrease in infectious progeny formation albeit to a much lesser extent (S10 Fig). The progression of the chlamydial developmental cycle was not affected as EB formation peaked at 48 h p.i. under both conditions, even though fewer EBs were recovered from the Retro-2 treated sample (Fig 6B). Retro-2 treatment reduced the size of C. trachomatis inclusions at 24 h and 48 h p.i. by about 40% without changing the shape of the inclusions (S11 Fig). Pretreatment of EBs with high Retro-2 concentrations (200 μM) before infection did not reduce infectious progeny compared to vehicle control (Fig 6C) and numbers of bacterial genomes were only slightly affected by the inhibitor (Fig 6D). To elucidate the effect of Retro-2 treatment on induction of chlamydial persistence, the ultrastructure of Retro-2 treated and control infected cells were determined by electron microscopy (Fig 6E). No signs of persistence in Retro-2-treated infections, as characterized by the appearance of larger aberrant Chlamydia forms were observed. Quantification of bacterial numbers confirmed that Retro-2 treatment affects replication of the bacteria which is in agreement with Retro-2 effects on genome copy numbers (S12 Fig and Fig 6D). Interestingly, we also detected a slight increase in numbers of intermediate bodies and ghosts in C. trachomatis inclusion grown in Retro-2 treated cell cultures compared to solvent control (S12 Fig).
A recovery assay in which infected cells were treated with Retro-2 from 8–48 h p.i., followed by removal of the inhibitor and additional incubation for 48 h in the absence of the inhibitor, confirmed that Retro-2 does not induce chlamydial persistence (Fig 6F). These experiments demonstrated that treatment of C. trachomatis infected cultures with Retro-2 strongly reduced the number of infectious bacteria at 48 h p.i. and upon removal the number of infectious bacteria remained on a low level. In contrast, the bacteria nearly completely recovered after removal of the well-known persistence inducer, penicillin G (Fig 6F).
In summary, our data show that C. trachomatis infections are Retro-2 sensitive resulting in smaller inclusions with slightly less bacteria inside, but with a strong defect in the generation of infectious EBs without induction of persistence.
We have shown that SNX5 and Retro-2 act on C. trachomatis infections, albeit with opposite effects on the bacteria. To further determine which effect is dominant, cells were treated with siRNA pools specific for SNX5, SNX1 and luciferase. Luciferase was used as non-targeting control while SNX1 knockdown served as additional control, as it did not significantly increase the EB numbers (Fig 5C). Infected knockdown cells were either treated with a single dose of Retro-2 at 8 h p.i. or mock-treated. Infectious progeny number was determined 48 h p.i. (Fig 6G). As expected, in vehicle-treated SNX5 knockdown cells, the characteristic increase in EB numbers upon knockdown of SNX5 was observed (Fig 6G). Interestingly, this increase in EB numbers in comparison to SNX1 knockdown and non-targeting control was lost upon Retro-2 treatment (Fig 6G). To assess whether Retro-2-sensitive retrograde transport is involved in recruiting SNX proteins to the inclusions, the localization of SNX proteins after Retro-2 treatment was analyzed at 12 h, 16 h and 24 h p.i. by confocal microscopy. In these imaging studies, no change in SNX localization was observed (S13 Fig). These data show that the increase in numbers of infectious EB after the silencing of SNX5 is Retro-2 sensitive whereas recruitment of SNX proteins to the inclusion appears to be Retro-2 insensitive.
The previous inability to isolate Chlamydia inclusions enforced severe experimental constraints and impeded progression in our comprehension of virulence mechanisms and the development of novel anti-chlamydial therapies. For example, recruitment of cellular proteins to the inclusion could only be addressed by microscopy. Direct biochemical evidence for the association of these factors with the inclusion membrane was therefore missing. To overcome this limitation, we established a method to isolate C. trachomatis inclusions at 24 h p.i. and analyzed isolated inclusions using a quantitative proteomics approach to decipher the host-derived C. trachomatis inclusion proteome.
We used the recently described protocol for the isolation of LCV from D. discoideum [16] as a starting point, but due to the fragile nature of the C. trachomatis inclusion, this protocol was heavily modified. As a result, we retained a two-step protocol but started with a Percoll-based gradient followed by immuno-magnetic separation using an IncA-specific antibody. One of the critical steps in the isolation protocol was the lysis of the infected host cells. We carefully tested different buffer and infection conditions, but the majority of inclusions were ruptured at this step resulting in a maximum recovery of 15% of the calculated initial numbers of inclusions. The yield in the following steps (gradient and MACS) was about ~50% amounting to a total recovery rate of about 8%. This recovery rate is in the range or even slightly higher than the yields obtained for Legionella containing vacuole isolations [16,30]. The second challenge was to find an optimal strategy for initial purification of the visually intact inclusions from cellular debris. We used isopycnic density gradient centrifugation to separate inclusions from host cell debris. We recovered the majority of inclusions in solution by fractionation of the gradient, but apparently the buoyant density of inclusions is very diverse, distributed across the range of densities of intracellular organelles, thus a subpopulation escaped our analysis which was distributed over the whole gradient without apparent peaks. It seems likely that these are inclusions that either contained large amounts of glycogen [31] or lipid droplets which are known to be translocated into the lumen of inclusions [32]. This translocation could have a considerable effect on their overall density. This speculation is supported by the absence of markers for lipid droplets in our proteome analysis. Moreover, we detected inclusions ranging in size from 3 μm up to 10 μm, representing the majority of expected inclusion sizes, possibly with a slight bias towards smaller inclusions, which could result from an increased fragility of larger inclusions.
The high sensitivity of modern LC-MS/MS-based proteomics demands an experimental design which includes a strategy to distinguish between bona fide components of the isolated compartment as well as co-purified contaminations. To this end, we used a SILAC-based exclusion approach in combination with label-free absolute quantification. A similar method was successfully used in a recent study to identify contaminants in purified latex bead-containing phagosome preparations [33].
Underlining the success of the purification and SILAC exclusion approach, we found a significant proportion of previously reported inclusion-associated proteins in our dataset. To further investigate the sensitivity of our assay, we ranked the proteins detected in a deep proteome of HeLa cells [22] by the iBAQ value of tryptic peptides, to see if highly abundant proteins are over-represented in the overlap with previously known inclusion-associated proteins (S14 Fig). Our limit for reliable detection of proteins with more than one peptide is slightly above the median iBAQ intensity in the HeLa cell lysate (S14 Fig). This is satisfying, considering the technical difficulties due to massive amounts of bacterial peptides present in our samples. However, based on these data, the true number of inclusion-associated proteins might be significantly higher than what we report here, probably around two times greater than the reported number based on known host proteins associated with inclusions. Furthermore, the SILAC exclusion approach has also some limitations, for example with proteins that have a high dissociation constant, which reduces the SILAC ratio due to exchange of L- for H-labeled proteins during the extended incubation time in cell lysate before MACS separation, thereby increasing the number of false negative classifications. These factors influence the number of reported proteins, but are all likely to reduce the reported number rather than to lead to false positives.
Whereas originally the inclusion was thought to be a separated compartment that acts as a niche devoid of host proteins [34], this picture has changed dramatically in recent years as indicated by the extensive interaction with cellular organelles and recruitment of specific proteins, often mediated by bacterial effectors, which was first described for 14-3-3 ß [27]. Interestingly, proteins annotated as nuclear, mitochondrial and lysosomal were significantly depleted in the Chlamydia inclusion proteome. Proteins assigned to other cellular organelles contributed significantly to the inclusion proteome, suggesting the inclusion is embedded in the intracellular trafficking network of the host cell. This conclusion supports the view that the C. trachomatis inclusion is a complex intracellular trafficking platform that exploits different pathways to foster optimal intracellular growth, rather than that of an isolated niche. For an obligate intracellular pathogen that lacks a number of genes for the biosynthesis of essential nutrients, this integration into the host cell organellar network seems reasonable to secure intracellular survival [35]. We noted redundancy in interactions which could reflect robustness of the intracellular lifestyle, which is further supported by the fact that C. trachomatis can infect and grow in an array of different cell types.
Detailed analysis of the host cell-derived inclusion proteome showed that C. trachomatis inclusions interact with the retromer, an important complex regulating retrograde transport of different cellular receptors and a pathway also hijacked by bacterial and plant toxins and distinct viruses to intoxicate and infect cells [6,36–38]. In Chlamydia-infected cells, the SNX-BAR proteins SNX1, 2, 5 and 6, are recruited to the inclusions decorating the inclusion in a rim-like staining pattern and are additionally found on IncA-laden fibers emanating from the inclusion body. In this context, it is interesting to note that Salmonella enterica serovar Typhimurium acquire SNX1 and SNX3, and SNX1 is found on spacious vacuole-associated tubules early in the infection process [39,40]. In uninfected cells, the PX and BAR domains of SNX-BAR proteins target these proteins to phosphoinositide-enriched, high-curvature membranes [41,42]. Phosphatidylinositol-4-phosphate (PI4P) has also been detected in the inclusion membrane by expression PIP-sensitive reporter proteins [43]. Whether the detected PI4P or additional bacterial proteins such as Inc proteins that are present in the inclusion membrane are involved in recruiting the SNX-BAR proteins to the inclusions is currently not known. Interestingly, the cargo recognition subcomplex of retromer showed only a punctual localization at the inclusion membrane. Consequently, there is partial separation of the two retromer subcomplexes at the inclusion membrane but not in other locations of infected cells. These observations support recent findings on the structure and function of the cellular retromer. Firstly, whereas the retromer complex is a stable hetero-pentamer in yeast cells, this association is much more transient in mammalian cells [44] and secondly, the two subcomplexes and the individual SNX-BAR proteins are involved independently of each other in trafficking of distinct cargo [45–47]. Functional analysis of SNX-BAR proteins using RNAi showed that in particular SNX5 knockdown resulted in an increase in infectious progeny. This may indicate that SNXs, and in particular SNX5, become segregated by recruitment to the C. trachomatis inclusion, thereby affecting the cellular retrograde trafficking pathways. The activity of the retromer complex has often been linked to processes controlling the sorting of cellular receptors including the epidermal growth factor receptor (EGFR) and M6PR [48,49]. SNX5 in particular has been implicated in EGFR trafficking and signaling in uninfected cells [48]. For C. trachomatis infections it has recently been demonstrated that EGFR activity is important for maturation of the inclusion by controlling calcium signaling and actin remodeling [50]. In light of these and our findings it is tempting to speculate that SNX5 recruitment to the inclusion alters e.g. EGFR transport and signaling inside the cells which in turn triggers calcium release and F-actin rearrangements. These changes then support the development of a proper C. trachomatis inclusion and are thus important for a successful infection. Alternatively, distinct SNX-BAR proteins control a currently not well-defined Retro-2-sensitive retrograde trafficking pathway that delivers distinct nutrients to the bacteria or alternatively could be related to factors controlling innate immunity. The idea of an innate immunity-related function of the retromer is further supported by the recently published observation in Drosophila that retromer can also control the Toll pathway [51].
The observed sensitivity towards the retrograde inhibitor Retro-2 also supports the view that retrograde transport is important for C. trachomatis progeny formation. The molecular target of Retro-2 is currently unknown but treatment results in displacement of the three t-SNAREs syntaxin (Stx) 5, 6 and 16 from membranes of the Golgi apparatus. These t-SNAREs are essential for retrograde transport of different cargo molecules to the TGN [52]. Interestingly, the localization of Stx6 to the inclusion has also been documented using microscopy and lack of Stx6 slightly but significantly reduced C. trachomatis infectious progeny [53,54]. Whether the strong inhibitory effect of Retro-2 treatment on C. trachomatis growth and infectious progeny formation is a result of mislocalization of different t-SNAREs from the inclusion or if additional proteins are also targeted by the treatment remains to be determined. Experiments are in progress to address Retro-2 dependent changes on a global level to determine these factors, which will potentially identify the molecular target of Retro-2 and might also uncover novel functions of the evolutionarily highly conserved retromer complex.
In summary, we have deciphered the core host cell-derived proteome of the C. trachomatis inclusion 24 h p.i. by quantitative proteomics of isolated inclusions. This data set describes the inclusion as a highly complex and interactive compartment that amongst others recruits proteins normally forming the membrane-binding subcomplex of the cellular SNX-BAR retromer. Of the subset of SNX-BAR proteins, SNX5 controlled the formation of infectious Chlamydia progeny in a Retro-2 sensitive pathway highlighting the importance of distinct SNX-BAR proteins and the retrograde transport for C. trachomatis infections. Thus, the development of a technique to isolate Chlamydia inclusions fosters our understanding of the inclusion composition, the contribution of cellular factors to inclusion formation and maintenance. This may pave the way for the development of axenic culture conditions and novel anti-chlamydial strategies.
HeLa cells were grown in Roswell Park Memorial Institute medium (RPMI, Gibco) 1640 supplemented with 10% fetal calf serum (FCS, Biochrom) at 37°C and 5% CO2 in a humidified incubator. The cells were routinely tested for Mycoplasma contamination via polymerase chain reaction (PCR) using the VenorGeM kit (Biochrom) according to manufacturer’s instructions. C. trachomatis L2 lymphatic isolate 434 Bu (ATCC: VR-902B) was propagated in HeLa cells. For more details on infections, determination of infectious progeny formation, the quantification of relative bacterial genome copy number, infection recovery assay, bacterial morphology assay and measurement of inclusion size, see S1 Text.
For plasmid transfections, HeLa cells were grown to 80% confluency and transfected with Lipofectamine 2000 reagent (Invitrogen) according to manufacturer’s instructions. For knockdown of target host cell proteins, HeLa cells were transfected with pools of target specific siRNAs as described in S1 Text.
For the standard procedures TEM, IF, SDS-PAGE, immunoblotting, molecular cloning as well as used reagents, plasmids and oligonucleotides, see S1 Text.
HeLa cells were infected with C. trachomatis (MOI 4) at 70–90% confluence. For standard isolations, 6 x 107 cells were used. All steps were done on ice or in a cold room at 4°C. Cells were washed once with PBS and subsequently with ice cold HSMG buffer (20 mM HEPES, 250 mM sucrose, 1.5 mM MgCl2, 0.5 mM EGTA, pH 7.4). Cells were scraped into 6 ml lysis buffer (33% Percoll solution (Sigma), HSMG) supplemented with cOmplete EDTA free protease inhibitors (Roche). Lysis was performed by repeated passage through a ball homogenizer (Isobiotech) using 16 μm clearance and 11–13 passages.
The lysate was then separated on a self-forming Percoll gradient in a total volume of 16 ml by centrifugation at 35’000 x g for 30 minutes at 4°C (Beckmann RC-6 with Thermo Scientific F21-8x50y rotor). The lower 6 ml of the gradient were either used for MACS purification or crude inclusions were diluted six fold in HSMG and pelleted at 1500 x g for 10 minutes, followed by another wash and centrifugation at 1200 x g for 10 minutes.
For MACS separation, crude inclusions were incubated with rabbit αIncA (1:1000) antibody [55] for 1.5 h at 4°C, followed by incubation with MACS secondary goat anti-rabbit antibody (1:100, Miltenyi) for another 1.5 hours. Inclusions were mixed gently by inversion every 30 minutes. The crude inclusions were loaded on a MACS LS separation column (Miltenyi) column in steps of 2 ml and washed with three times the input volume of HSMG buffer. Inclusions were then eluted with 3 ml HSMG buffer after removal of the magnet, aided by gentle pushing using the supplied plunger. Counting of inclusions, the small scale isolation procedure for validation and processing of inclusions for IF and TEM are described in supporting information (S1 Text).
For SILAC experiments, cells were grown in SILAC DMEM (PAA) containing dialyzed FCS (Biochrom), supplemented with H labeled L-arginine(13C615N4) and L-lysine (13C615N2) (Silantes) or non-labeled amino acids (L). Inclusions were isolated as described above but H labeled mock infected cells were mixed with equal amounts with L labeled infected cells prior to cell lysis. Inclusion samples were prepared for LC-MS/MS. 10% of the sample was used for direct injection after desalting. The remaining peptides were separated by strong anion exchange chromatography into 6 fractions before desalting, followed by LC-MS/MS. Lysate samples were prepared for LC-MS/MS without pre-fractionation. For more details, see S1 Text.
Tryptic peptides were analyzed using a data dependent method on a Q Exactive mass spectrometer (Thermo) coupled to a Ultimate 3000 nHPLC (Dionex) for separation by reverse phase chromatography. The resulting. raw files were analyzed in MaxQuant 1.3.0.5 [56]. Protein groups that had less than two unique + razor peptides in at least one experiment were filtered. See S1 Text for more details on SILAC enrichment analyses, abundance analyses using iBAQ and further bioinformatics analyses.
|
10.1371/journal.pgen.1001302 | Pervasive Adaptive Protein Evolution Apparent in Diversity Patterns around Amino Acid Substitutions in Drosophila simulans | In Drosophila, multiple lines of evidence converge in suggesting that beneficial substitutions to the genome may be common. All suffer from confounding factors, however, such that the interpretation of the evidence—in particular, conclusions about the rate and strength of beneficial substitutions—remains tentative. Here, we use genome-wide polymorphism data in D. simulans and sequenced genomes of its close relatives to construct a readily interpretable characterization of the effects of positive selection: the shape of average neutral diversity around amino acid substitutions. As expected under recurrent selective sweeps, we find a trough in diversity levels around amino acid but not around synonymous substitutions, a distinctive pattern that is not expected under alternative models. This characterization is richer than previous approaches, which relied on limited summaries of the data (e.g., the slope of a scatter plot), and relates to underlying selection parameters in a straightforward way, allowing us to make more reliable inferences about the prevalence and strength of adaptation. Specifically, we develop a coalescent-based model for the shape of the entire curve and use it to infer adaptive parameters by maximum likelihood. Our inference suggests that ∼13% of amino acid substitutions cause selective sweeps. Interestingly, it reveals two classes of beneficial fixations: a minority (approximately 3%) that appears to have had large selective effects and accounts for most of the reduction in diversity, and the remaining 10%, which seem to have had very weak selective effects. These estimates therefore help to reconcile the apparent conflict among previously published estimates of the strength of selection. More generally, our findings provide unequivocal evidence for strongly beneficial substitutions in Drosophila and illustrate how the rapidly accumulating genome-wide data can be leveraged to address enduring questions about the genetic basis of adaptation.
| Characterizing the nature of beneficial changes to the genome is essential to our understanding of adaptation. To do so, researchers identify and analyze footprints that beneficial changes leave in patterns of genetic variation within and between species. In order to teach us about adaptive evolution, these footprints need to be specific to positive selection as well as rich enough to allow for reliable inferences. Here, we identify such a footprint: a pronounced trough in the average levels of genetic diversity surrounding amino acid substitutions throughout the D. simulans genome. Based on this pattern, we infer that approximately 13% of amino acid substitutions were beneficial, a minority of which (3%) conferred a large selective advantage of nearly 0.5% and the majority of which (10%) conferred a much smaller advantage of about 0.01%. These findings offer insights into the distribution of selection effects driving beneficial changes to the D. simulans genome and suggest how the widely varying estimates obtained in previous studies of Drosophila may be reconciled. Moreover, the approach that we introduce is readily applicable to other taxa and thus should help to gain important insights into how the rate and strength of adaptive evolution vary depending on life-history, population size, and ecology.
| A central challenge of evolutionary biology is to elucidate the nature of adaptive changes to the genome: do they comprise a negligible or substantial fraction of differences among species? When they occur, are they driven by strong positive selection or are they fine-tunings of minor consequence to fitness? In Drosophila, perhaps the most studied taxon in these respects, there are conflicting accounts regarding the intensity of selection driving adaptations [1]–[4] but accumulating lines of evidence suggest that adaptation may be prevalent [5]–[7].
The evidence is based primarily on two kinds of signatures that beneficial substitutions leave in their wake. The first is an excess of divergence at functional sites compared to that expected under neutrality, detected using the approach introduced by McDonald and Kreitman [8]–[11]. Numerous studies based on extensions of this approach indicate that approximately one in two amino acid and one in five non-coding differences between Drosophila species may be adaptive [7], [11]–[14]. These findings remain tentative, however, because other factors, and notably plausible demographic scenarios, could cause a substantial overestimation of the fraction of beneficial substitutions [7], [8], [15]–[17]. Moreover, McDonald-Kreitman based approaches can provide only very limited information about the strength of positive selection.
The second footprint of adaptation is in diversity patterns. When a rare or new allele is favored and fixes in the population, it drags closely linked neutral alleles to loss or fixation. This “selective sweep” leads to a transient reduction in levels of neutral diversity around a beneficial substitution, where the size of the affected region decreases with the recombination rate and increases with the intensity of positive selection [18]–[20]. In accordance with a model of recurrent selective sweeps, levels of synonymous diversity across the genomes of a number of Drosophila species increase with rates of crossing over [21]–[23] and decrease with increasing numbers of amino acid substitutions [2], [3].
Making reliable inferences about adaptation based on these relationships has been challenging, with two decades of effort focused on distinguishing the effects of positive selection from those of background (i.e., purifying) selection and from possible mutagenic effects of recombination [5], [24]–[29]. By necessity, previous studies relied on limited summaries of the data, thereby losing much of the information carried by the spatial signature of beneficial fixations. In particular, measurements of diversity, recombination, and functional divergence were taken in arbitrarily chosen window sizes, making it harder to distinguish the effects of adaptation from other evolutionary forces [29], [30], and likely biasing estimates of adaptive parameters of interest (e.g., the rate and intensity of selection) [7]. As an illustration, based on the relationship between diversity levels and amino acid divergence seen in 100 kb windows, Macpherson et al. [3] inferred few beneficial amino acid substitutions with a large selective coefficient of ∼1%; in contrast, focusing on the same relationship in individual genes, Andolfatto [2] inferred many beneficial amino substitutions with a selective coefficient of ∼10−3%; the two studies differed in other regards, but the disparate conclusions may reflect in part the choice of window size [7]. In summary, despite accumulating evidence that adaptation may be widespread in Drosophila, we still lack characterizations that capture genome-wide signatures that are specific to adaptive evolution and do not rely on an a priori choice of scale.
Here, we take advantage of genome-wide variation data from Drosophila in order to produce a readily interpretable characterization of the effects of positive selection that overcomes a number of limitations. To do so, we consider the average level of neutral diversity as a function of distance from amino acid substitutions. Our reasoning is as follows: Beneficial amino acids that fixed in the recent evolutionary past (∼Ne generations [20]) should create a trough in diversity levels around them, whereas amino acid substitutions that were selectively neutral or occurred farther in the past should have little effect on diversity patterns. If we consider the effects of all amino acid substitutions in the genome jointly, and a non-negligible fraction of amino acid fixations were favored – as McDonald-Kreitman based estimates suggest – then we should expect a trough in the average level of neutral diversity around amino acid substitutions. The depth of this trough is expected to increase with the fraction of beneficial amino acid substitutions, and its width will reflect the intensity of selection driving these substitutions. In contrast to previous approaches, this characterization does not depend on an a priori choice of window size, and captures much more of the footprint of adaptive substitutions.
To generate this plot, we use autosomal amino acid substitutions on the lineage leading from the common ancestor of Drosophila simulans and D. melanogaster to D. simulans, relying on the genomes of D. erecta and D. yakuba as outgroups [31]. As a measure of neutral diversity, we consider the number of synonymous polymorphisms divided by the overall number of codons at a given distance from an amino acid substitution. The polymorphism levels in D. simulans are measured using a recent dataset of six inbred lines [5], down-sampled to have a uniform sample size of 4 lines at ∼50% of the codons in the genome. Ideally, we would like to plot diversity levels as a function of genetic distance from amino acid substitutions, since the expected reduction in diversity depends on genetic rather than physical distance from the selected loci. Since there are no high-resolution estimates of recombination rates in D. simulans, we use physical distance instead, but consider only regions for which the homologous regions in D. melanogaster have an estimated recombination rate above 0.75cM/Mb. The collated plot in Figure 1A (red) thus obtained is averaged over n = 26,834 amino acid substitutions.
Because the plot is constructed by conditioning on a substitution at the center, diversity patterns could be distorted even in the absence of adaptive evolution. Namely, if mutation rates vary across the genome then they might, on average, be elevated near substitutions. Considering the average synonymous divergence between D. melanogaster and D. yakuba as a proxy for the mutation rate confirms this expectation, as it reveals a small increase near substitutions (Figure 1B). To correct for this elevation in rates, we divide the average level of diversity around amino acid substitutions at a given distance by the average divergence (Figure 1C). Moreover, as a control, we compare the patterns around amino acid substitutions with plots that were constructed analogously but around synonymous substitutions instead (Figure 1A–1C: black) [28].
As predicted by a model of recurrent selective sweeps, we find a clear reduction in diversity levels around amino acid substitutions relative to the synonymous control. This reduction is statistically significant within a window of ∼15kb around amino acid substitutions (at the 1% level, as assessed by bootstrapping; see Text S1). Farther from substitutions, where sweeps are unlikely to have an effect on diversity, the curves for synonymous and amino acid substitutions are indistinguishable. This pattern is robust to the effects of synonymous codon usage bias (Figure 4 in Text S1), as well as to changes in the recombination rate threshold (Figure 5 in Text S1), and to the choice of outgroup used to correct for the mutation rate (not shown). In addition, we see similar patterns when we examine the substitutions that occur on any one of the autosomal chromosome arms (Figure 6 in Text S1).
This pattern is a distinctive signature of adaptive evolution. Demographic processes would not lead to systematically decreased diversity around amino acid substitutions. In turn, for background selection to generate the observed trough centered on amino acid substitutions, its effects in regions of the genome with moderate to high recombination rates would have to be strong enough to lead to both a substantial reduction in diversity and to the fixation of many weakly deleterious amino acid mutations. Modeling indicates that, given plausible parameters for Drosophila, this is highly unlikely [32].
Our analyses also reveal that amino acid substitutions are clustered near one another (Figure 2A: red). This clustering is greater and more localized than the clustering of synonymous substitutions around amino acid substitutions (Figure 2A: black), implying that it is caused by more than the spatial distribution of exons in the genome and an elevated mutation rate near amino acid substitutions. The difference between the clustering of amino acid and synonymous substitutions further suggests that variation in constraint and possibly in adaptability among and within genes contribute to the pattern for amino acid substitutions ([33]; also see Text S1).
Aside from being an interesting finding in itself, this clustering could influence the observed reduction in diversity. If two amino acid substitutions occur in close proximity and one led to a recent selective sweep, the reduction in diversity that it caused will also be observed around the other substitution. This effect will reduce diversity around both non-synonymous and synonymous substitutions, but it will have a larger effect around amino acid substitutions because the density of amino acid substitutions nearby is on average greater (Figure 2A). Indeed, the level of synonymous diversity decreases strongly with the density of amino acid substitutions surrounding a substitution (Figure 2B; Figure 8 in Text S1; Spearman's ρ = −0.93 for amino acid substitutions and ρ = −0.88 for synonymous substitutions; p<10−15 for both), consistent with previous studies [2], [3]. We also find, however, that the average level of synonymous diversity around amino acid substitutions is consistently lower than that around synonymous substitutions when the two are matched for the density of amino acid substitutions in their vicinity (Figure 2B; Figure 8 in Text S1; signs test p<10−4). In other words, there is a substantial relative reduction in diversity around amino acid substitutions that is not explained by the amplifying effects of clustering.
In addition to providing compelling evidence for the prevalence of beneficial amino acid substitutions, the collated plot carries information about selection parameters, as the shape of the trough in diversity is indicative of the rate of adaptive protein evolution and of the distribution of selective effects of fixations. To learn about these parameters, we develop a coalescent-based model for average diversity levels as a function of distance from an amino acid substitution, accounting for their clustering (see Text S1). Using this model, we infer adaptive parameters by jointly maximizing the composite-likelihood of diversity patterns as a function of different distances from the focal substitution (i.e., the likelihood of points along the entire curve), thus mining a richer summary of the data than previous approaches. When we assume that a fraction α of beneficial substitutions were driven by a selection coefficient s and the rest were neutral, we estimate that ∼5% of the substitutions were beneficial with a relatively strong selection coefficient of ∼0.4% (Table 5 in Text S1). Using a Gamma distribution for the selection coefficients, α increases to ∼6.5% and the average selection coefficient remains similarly high; despite the additional parameter, the likelihood is barely higher (Table 5 in Text S1). These estimates are relatively insensitive to assumptions about other parameters (with the exception of the assumptions about recombination rates, as discussed below); in particular, simulations suggest that the estimated strength of selection is robust to demographic assumptions (see Text S1 for details).
A visual comparison suggests a reasonable fit of these models to the data (Figure 3A). However, the inference based on models with one selection coefficient, or even a Gamma distribution of coefficients, might be dominated by the broad features of the plot, such that any narrower trough caused by beneficial substitutions with weaker selection coefficients could be overlooked. A closer look around the focal substitutions supports this notion, revealing a small trough inside the main trough, on the scale of several hundred bps, which is not captured by either of the two models (Figure 3B). We therefore consider another model, with two beneficial selection coefficients. Using it, we estimate that ∼13% of the substitutions were beneficial, ∼3% with a large selective advantage of ∼0.5% and the rest with a much weaker effect, of approximately one hundredth of a percent (Table 5 in Text S1). A mixture model with two exponentials reveals a similar picture: ∼4% of substitutions are estimated to come from a distribution with a mean selective coefficient of ∼0.5% and 11% from a distribution with a mean of ∼4·10−5 (Table 5 in Text S1). Importantly, both models provide a substantially better fit to the data (Table 5 in Text S1) and they capture the smaller as well as the larger troughs in diversity (Figure 3A and 3B). In turn, estimates under a model with three beneficial selective coefficients are similar to those obtained in model with only two and offer no improvement to the fit (Table 5 in Text S1). Taken together, these findings indicate that selective sweeps are driven by two classes of beneficial fixations: a minority with large beneficial effects that account for most of the reduction in diversity and a majority with much weaker effects. Moreover, they help explain why previous inferences based on the signatures of sweeps in Drosophila yielded markedly different estimates (ranging over three orders of magnitudes) [1]–[4].
Our estimates of the fraction of beneficial amino acid substitutions (∼13%) are on the same order of magnitude but lower than previous McDonald-Kreitman based estimates (∼50%; cf. [7]). Some of this difference might arise from violations of the assumptions on which the inferences rely; in particular, in our approach, that adaptive parameters have remained constant in the D. simulans lineage, or in McDonald-Krietman based inferences, that the efficacy of purifying selection has not changed markedly [8], [16], [34].
An intriguing alternative is that the two approaches are actually estimating parameters of somewhat different modes of adaptation. Our inference is based on the effects of beneficial substitutions that arise from new mutations and likely misses some contribution of adaptation from standing variation. Specifically, a subset of beneficial substitutions could stem from previously neutral or deleterious alleles that were segregating in the population before a change in the environment rendered them beneficial. If these alleles were young when the environment changed, they would still generate the signature of a selective sweep and contribute, at least partially, to our estimated fraction of beneficial substitutions. This is likely for alleles that were previously deleterious and at mutation-selection balance, but also possible for neutral alleles [35]–[37]. If, however, the segregating alleles were older when they became beneficial and at higher frequency in the population, they would lead to a negligible effect on diversity and would therefore not contribute to the signature on which our inference relies. These beneficial substitutions would nonetheless contribute to an excess of non-synonymous divergence compared to the neutral expectation, and should therefore be picked by the McDonald-Kreitman based inferences, leading to higher estimates of adaptive substitutions than obtained by our approach. Other modes of adaptation, such as polygenic selection, may also contribute differentially to the two inference methodologies [38].
We note that a current limitation of our inference is its reliance on rough estimates of the recombination rate, and its assumption of a constant rate per base. In the logistic approximation to the trajectory of a beneficial allele, the expected reduction in diversity as a function of distance from the beneficial substitution depends on s/r, where s is the selection coefficient and r is the genetic distance to the substitution (Equation 2 in Text S1). This implies, for example, that if our inference relies on a recombination rate consistently two-fold greater than the real rate, our estimated selection coefficient will be two-fold overestimated (see Table 3 in Text S1). We therefore consider our estimates of selection coefficients to be rough approximations. In addition, heterogeneity in the recombination rate, such as is known to exist in other taxa (e.g., [39], [40]), could also affect our inferences. The heterogeneity would have to be of a highly specific nature in order to account for our finding of two markedly different scales of selection coefficients, but at the moment, we cannot rule out the possibility. For these reasons, it would be important to revisit the inference once we possess high-resolution genetic maps in D. simulans.
In summary, our findings establish a distinctive, genome-wide signature of adaptation in D. simulans, suggesting that many amino acid substitutions are beneficial and are driven by two classes of selective effects. Enabled by a richer summary of diversity patterns that avoids an a priori choice of scale, these conclusions offer a coherent interpretation of the results of previous inferences. It will now be interesting to see whether similar findings emerge in other Drosophila species, which vary in their recombination rates, effective population sizes, and ecology.
We reconstructed the sequence of the ancestor of D. melanogaster and D. simulans in order to identify substitutions along the D. simulans lineage. For that purpose, we use a four species alignement from the 12 Drosophila genomes project [31] consisting of D. simulans, D. melanogaster, D. yakuba and D. erecta, and removed codons containing gaps in either of them. We then inferred the ancestral sequences using PAML, with the CODEML model and the ((D. mel, D. sim), (D. yak, D. ere)) tree [41]. To measure polymorphism levels at coding regions of the D. simulans genome, we used resequencing data from six inbred lines of D. simulans and their alignment with D. melanogaster [5]. We applied quality control filters and randomly down-sampled the remaining codons to four, in order to maintain a uniform sample size in measuring polymorphism. In the end, we retained ∼50% of all protein-coding DNA. Unless otherwise noted, our analysis was performed on data from autosomal regions, for which the sex-averaged recombination rate in the homologous region of D. melanogaster was greater than 0.75cM/Mb (using the genetic map as in [3]). See Section 1 in Text S1 for more details.
We used synonymous polymorphisms to measure the average levels of diversity as a function of distance from amino acid and synonymous substitutions along the D. simulans lineage. To measure the average level of diversity at distance x, we divided the number of codons segregating for a synonymous polymorphism by the overall number of codons observed in the D. simulans polymorphism dataset at distance x from one of the amino acid (or synonymous) substitution. In order to control for variation in the neutral mutation rate around substitutions, we calculated the average synonymous divergence around both amino acid and synonymous substitutions. For that purpose, we identified synonymous substitutions between D. melanogaster and D. yakuba and measured the average level of divergence at distance x by dividing the number of codons exhibiting a synonymous substitution between D. melanogaster and D. yakuba by the overall number of codons observed in the alignment of these species at distance x from one of the amino acid (or synonymous) substitutions. For further details and the robustness analysis, see Sections 2–4 in Text S1.
The shape of the collated plot around amino acid substitutions carries information about the rate of adaptive protein evolution and the intensity of selection driving it, two parameters of long-standing interest. To learn about these parameters, we developed a model describing the expected neutral diversity levels around substitutions, which relies on Gillespie's pseudohitchhiking coalescent model [42]. We then used a composite likelihood approach [43] to estimate the parameters. For a description of the approach and assessments of its reliability, see Section 6 in Text S1.
|
10.1371/journal.pgen.1007071 | Profiling RNA-Seq at multiple resolutions markedly increases the number of causal eQTLs in autoimmune disease | Genome-wide association studies have identified hundreds of risk loci for autoimmune disease, yet only a minority (~25%) share genetic effects with changes to gene expression (eQTLs) in immune cells. RNA-Seq based quantification at whole-gene resolution, where abundance is estimated by culminating expression of all transcripts or exons of the same gene, is likely to account for this observed lack of colocalisation as subtle isoform switches and expression variation in independent exons can be concealed. We performed integrative cis-eQTL analysis using association statistics from twenty autoimmune diseases (560 independent loci) and RNA-Seq data from 373 individuals of the Geuvadis cohort profiled at gene-, isoform-, exon-, junction-, and intron-level resolution in lymphoblastoid cell lines. After stringently testing for a shared causal variant using both the Joint Likelihood Mapping and Regulatory Trait Concordance frameworks, we found that gene-level quantification significantly underestimated the number of causal cis-eQTLs. Only 5.0–5.3% of loci were found to share a causal cis-eQTL at gene-level compared to 12.9–18.4% at exon-level and 9.6–10.5% at junction-level. More than a fifth of autoimmune loci shared an underlying causal variant in a single cell type by combining all five quantification types; a marked increase over current estimates of steady-state causal cis-eQTLs. Causal cis-eQTLs detected at different quantification types localised to discrete epigenetic annotations. We applied a linear mixed-effects model to distinguish cis-eQTLs modulating all expression elements of a gene from those where the signal is only evident in a subset of elements. Exon-level analysis detected disease-associated cis-eQTLs that subtly altered transcription globally across the target gene. We dissected in detail the genetic associations of systemic lupus erythematosus and functionally annotated the candidate genes. Many of the known and novel genes were concealed at gene-level (e.g. IKZF2, TYK2, LYST). Our findings are provided as a web resource.
| It is well acknowledged that non-coding genetic variants contribute to disease susceptibility through alteration of gene expression levels (known as eQTLs). Identifying the variants that are causal to both disease risk and changes to expression levels has not been easy and we believe this is in part due to how expression is quantified using RNA-Sequencing (RNA-Seq). Whole-gene expression, where abundance is estimated by culminating expression of all transcripts or exons of the same gene, is conventionally used in eQTL analysis. This low resolution may conceal subtle isoform switches and expression variation in independent exons. Using isoform-, exon-, and junction-level quantification can not only point to the candidate genes involved, but also the specific transcripts implicated. We make use of existing RNA-Seq expression data profiled at gene-, isoform-, exon-, junction-, and intron-level, and perform eQTL analysis using association data from twenty autoimmune diseases. We find exon-, and junction-level thoroughly outperform gene-level analysis, and by leveraging all five quantification types, we find >20% of autoimmune loci share a single genetic effect with gene expression. We highlight that existing and new eQTL cohorts using RNA-Seq should profile expression at multiple resolutions to maximise the ability to detect causal eQTLs and candidate genes.
| The autoimmune diseases are a family of heritable, often debilitating, complex disorders in which immune dysfunction leads to loss of tolerance to self-antigens and chronic inflammation [1]. Genome-wide association studies (GWAS) have now detected hundreds of susceptibility loci contributing to risk of autoimmunity [2] yet their biological interpretation still remains challenging [3]. Mapping single nucleotide polymorphisms (SNPs) that influence gene expression (eQTLs) can provide meaningful insight into the potential candidate genes and etiological pathways connected to discrete disease phenotypes [4]. For example, such analyses have implicated dysregulation of autophagy in Crohn’s disease [5], the pathogenic role of CD4+ effector memory T-cells in rheumatoid arthritis [6], and an overrepresentation of transcription factors in systemic lupus erythematosus [7].
Expression profiling in appropriate cell types and physiological conditions is necessary to capture the pathologically relevant regulatory changes driving disease risk [8]. Lack of such expression data is thought to explain the observed disparity of shared genetic architecture between disease association and gene expression at certain autoimmune loci [9]. A much overlooked cause of this disconnect however, is not only the use of microarrays to profile gene expression, but also the resolution to which expression is quantified using RNA-Sequencing (RNA-Seq) [10]. Expression estimates of whole-genes, individual isoforms and exons, splice-junctions, and introns are obtainable with RNA-Seq [11–18]. The SNPs that affect these discrete units of expression vary strikingly in their proximity to the target gene, localisation to specific epigenetic marks, and effect on translated isoforms [18]. For example, in over 57% of genes with both an eQTL influencing overall gene expression and a transcript ratio QTL (trQTL) affecting the ratio of each transcript to the gene total, the causal variants for each effect are independent and reside in distinct regulatory elements of the genome [18].
RNA-Seq based eQTL investigations that solely rely on whole-gene expression estimates are likely to mask the allelic effects on independent exons and alternatively-spliced isoforms [16–19]. This is in part due to subtle isoform switches and expression variation in exons that cannot be captured at gene-level [20]. A large proportion of trait associated variants are thought to act via direct effects on pre-mRNA splicing that do not change total mRNA levels [21]. Recent evidence also suggests that exon-level based strategies are more sensitive than conventional gene-level approaches, and allow for detection of moderate but systematic changes in gene expression that are not necessarily derived from alternative-splicing events [15,22]. Furthermore, gene-level summary counts can be biased in the direction of extreme exon outliers [22]. Use of isoform-, exon-, and junction-level quantification in eQTL analysis also support the potential to not only point to the candidate genes involved, but also the specific transcripts or functional domains affected [10,18]. This of course facilitates the design of targeted functional studies and better illuminates the causative relationship between regulatory genetic variation and disease. Lastly, though intron-level quantification is not often used in conventional eQTL analysis, it can still provide valuable insight into the role of unannotated exons in reference gene annotations, retained introns, and even intronic enhancers [23,24].
Low-resolution expression profiling with RNA-Seq will impede the subsequent identification of causal eQTLs when applying genetic and epigenetic fine-mapping approaches [25]. In this investigation, we aim to increase our knowledge of the regulatory mechanisms and candidate genes of human autoimmune disease through integration of GWAS and RNA-Seq expression data profiled at gene-, isoform-, exon-, junction-, and intron-level in lymphoblastoid cell lines (LCLs). This is firstly performed in detail using association data from a GWAS in systemic lupus erythematosus, and is then scaled up to a total of twenty autoimmune diseases. Our findings are provided as a web resource to interrogate the functional effects of autoimmune associated SNPs (www.insidegen.com), and will serve as the basis for targeted follow-up investigations.
Using densely imputed genetic association data from a large European GWAS in systemic lupus erythematosus (SLE) [7], we performed integrative cis-eQTL analysis with RNA-Seq expression data profiled at five resolutions: gene-, transcript-, exon-, junction-, and intron-level. Expression data were derived from 373 healthy European donors of the Geuvadis project profiled in lymphoblastoid cell lines (LCLs) [18]. See S1 Fig for a summary of how expression at the five resolutions was quantified. A total of 38 genome-wide significant SLE loci (S1 Table) were put forward for analysis. To test for evidence of a single shared causal variant between disease and gene expression at each locus, we employed the Joint Likelihood Mapping (JLIM) framework [9] using summary-level statistics for SLE association and full genotype-level data for gene expression. Using JLIM, cis-eQTLs were defined if a nominal association (P<0.01) with at least one SNP existed within 100kb of the SNP most associated with disease and the transcription start site of the gene was located within +/-500kb of that SNP (as defined by authors of JLIM). JLIM P-values were corrected for multiple testing by a false discovery rate (FDR) of 5% per RNA-Seq quantification type (i.e. at exon-level, JLIM P-values were adjusted for total number of exons tested in cis to the 38 SNPs). Causal associations of the integrative cis-eQTL SLE GWAS analysis across the five RNA-Seq quantification types are available in S2 Table and the full output (including non-causal associations) are available in S3 Table. The distribution of JLIM P-values across the five RNA-Seq quantification types are depicted in S2 Fig.
We found the number of causal cis-eQTLs was markedly underrepresented when considering conventional gene-level quantification (Table 1). Only two of the 38 SLE susceptibility loci (5.3%) were deemed to be causal cis-eQTLs at gene-level for three candidate genes. This is a similar proportion observed by Chun et al [9] who found that 16 of the 272 (5.9%) autoimmune susceptibility loci tested were cis-eQTLs driven by a shared causal variant in the Geuvadis RNA-Seq dataset using gene-level quantification (based upon the seven autoimmune diseases interrogated—not including SLE).
Of note, transcript-level quantification did not increase the number of causal cis-eQTLs (Table 1). Transcript-level analysis did, however, yield a greater number of candidate genes (seven individual transcripts derived from a total of four genes). Both junction- and intron-level quantification increased the number of causal cis-eQTLs to four (10.5% of the 38 total SLE loci). Using exon-level quantification, we were able to detect seven significant cis-eQTLs driven by a single shared causal variant (18.4%). Exon-level analysis also produced the greatest number of candidate gene targets: nine unique genes derived from 24 individual SNP-exon pairs (Table 1). Therefore, even with the severe multiple testing burden, we firstly conclude that exon-, junction-, and intron-level analysis detects more causal cis-eQTLs than gene-level.
By combining all five types of RNA-Seq quantification (gene, transcript, exon, junction, and intron) we classified nine of the 38 SLE susceptibility loci (24%) as being driven by the same causal variant as the cis-eQTL in LCLs (Table 1). This value, derived from interrogating only a single cell type, is almost equal to the total number of causal autoimmune cis-eQTLs detected by Chun et al [9] (~25%) across three different cell types (CD4+ T-cells–measured by microarray, CD14+ monocytes–microarray, and LCLs–RNA-Seq gene-level).
We found that when considering the specificity of cis-eQTLs and target genes across the five RNA-Seq quantification types, both gene- and transcript-level quantification were redundant with respect to exon-level data; i.e. there were no causal cis-eQTLs or target genes detected at gene- or transcript-level that were not captured by exon-level analysis (S3 Fig). Both junction- and intron-level quantification captured a single causal cis-eQTL each that was not captured by exon-level. We conclude that profiling at all resolutions of RNA-Seq is required to capture the full set of potentially causal cis-eQTLs.
We compared the detection of cis-eQTLs using a pairwise comparison between the five RNA-Seq quantification types for matched SNP-gene cis-eQTL pairs (Fig 1). We only considered matched SNP-gene cis-eQTL association pairs that had a nominal cis-eQTL association P-value < 0.01 in both quantification types, and to be conservative, when multiple transcripts, exons, junctions, and introns were annotated with the same gene symbol, we selected the associations that minimized the difference in JLIM P-value between matched SNP-gene cis-eQTLs across RNA-Seq quantification types. There were over 250 matched SNP-gene cis-eQTL pairs per comparison. We firstly observed that the correlation of both cis-eQTL association P-values from regression and JLIM P-values across RNA-Seq quantification types reflected the methods in which expression quantification was obtained (Fig 1A). Both cis-eQTL and JLIM P-values between matched SNP-gene pairs at gene- and transcript-level were highly correlated as gene-level estimates are obtained from the sum of all transcript-level estimates for the same gene. Exon-level and junction-level associations were also highly correlated due to split-reads being incorporated into the exon-level estimate. As expected, intron-level cis-eQTL and JLIM P-values for matched SNP-gene pairs were only weakly correlated against other quantification types—as reads mapping to introns are not included in the other quantification models. Interestingly, although cis-eQTL association P-values for matched SNP-gene pairs between transcript-level and junction-level were found to be relatively high (r2 = 0.70), we found the JLIM P-values for the matched pairs to be comparatively low (r2 = 0.29); suggesting that whilst the statistical significance of matched cis-eQTLs maybe similar between these quantification types, the underlying causal variants driving the disease and cis-eQTL association are likely to be independent.
By plotting the JLIM P-values for matched SNP-gene pairs between different quantification types, we found many instances of P-values distributed along the axes rather than on the diagonal (Fig 1B). Our findings therefore suggest that often, one quantification type is more likely to explain the observed disease association than the other. When we compared conventional gene-level cis-eQTL analysis against exon-level results (Fig 1C), we found that of the 296 matched SNP-gene cis-eQTL associations (P<0.01), eleven (4%) shared the same causal variant at both gene- and exon-level using a nominal JLIM P-value threshold <0.01. Only three of the 296 matched SNP-gene cis-eQTL associations (1%) were captured by gene-level only—in contrast to the 26 (9% of total associations) captured uniquely at exon-level. As expected, the overwhelming majority of cis-eQTL associations (86%) did not possess a single shared causal variant at either gene- or exon-level. We performed this analysis for all possible combinations of quantification types (Table 2). In each instance, gene-level analysis detected only the minority of nominally causal associations for matched SNP-gene association pairs (JLIM P<0.01). Exon-level and junction-level analysis consistently detected more causal cis-eQTL associations than gene-, transcript-, and intron-level. In fact, when combined, exon- and junction-level analysis explained the most nominally causal associations for all significant SNP-gene cis-eQTL association pairs (24%).
We functionally dissected the 12 candidate genes taken from the nine SLE associated loci that showed strong evidence of a shared causal variant with a cis-eQTL in LCLs (Table 3). We systematically annotated these genes using a combination of cell/tissue expression patterns, mouse models, known molecular phenotypes, molecular interactions, and associations with other autoimmune diseases (S4 Table). We found the majority of novel SLE candidate genes detected by RNA-Seq were predominately expressed in immune-related tissues such as whole blood as well as the spleen and thymus. Based on our annotation and what is already documented at certain loci, we were sceptical on the pathogenic involvement of three candidate genes (PHTF1, ARHGAP30, and RABEP1). Although the cis-eQTL effect for these genes is evidently driven by the shared causal variant as the disease association, it is possible that these effects of expression modulation are merely passengers that are carried on the same functional haplotype as the true causal gene(s) and do not contribute themselves to the breakdown of self-tolerance (detailed in S4 Table). We show the regional association plots and the candidate genes detected from cis-eQTL analysis in S4 Fig.
The causal cis-eQTL rs2736340 for genes BLK and FAM167A was detected at all RNA-Seq profiling types. It is well established that the risk allele of this SNP reduces proximal promoter activity of BLK; a member of the Src family kinases that functions in intracellular signalling and the regulation of B-cell proliferation, differentiation, and tolerance [26]. The allelic consequence of FAM167A expression modulation is unknown. We found multiple instances of known SLE susceptibility genes that were concealed when using gene-level quantification. For example, we defined rs7444 as a causal cis-eQTL for UBE2L3 at transcript- and exon-level—but not at gene-level (Table 3). The risk allele of rs7444 has been associated with increased expression of UBE3L3 (Ubiquitin conjugating enzyme E2 L3) in ex vivo B-cells and monocytes and correlates with NF-κB activation along with increased circulating plasmablast and plasma cell numbers [27]. Similarly, the rs10028805 SNP is a known splicing cis-eQTL for BANK1 (B-cell scaffold protein with ankyrin repeats 1). We replicated at exon-, and junction-level this splicing effect which has been proposed to alter the B-cell activation threshold [28]. Again, this mechanism was not detected using gene-level quantification.
IKZF2 (detected at the exon-level only) is a transcription factor thought to play a key role in T-reg stabilisation in the presence of inflammatory responses [29]. IKZF2 deficient mice acquire an auto-inflammatory phenotype in later life similar to rheumatoid arthritis, with increased numbers of activated CD4+ and CD8+ T-cells, T-follicular helper cells, and germinal centre B-cells, which culminates in autoantibody production [30]. Of note, other members of this gene family, IKZF1 and IKZF3, are also associated with SLE and can hetero-dimerize (S4 Table) [7]. We also believe LYST, ATG4D, and TYK2 to also be intriguing candidate genes. LYST encodes a lysosomal trafficking regulator [31] whilst ATG4D is a cysteine peptidase involved in autophagy and this locus is associated with multiple sclerosis, psoriasis, and rheumatoid arthritis [32]. TYK2 is discussed in greater detail in the following section.
Interestingly, for the three causal SNP-gene pairs detected at gene-level (rs2736340 –BLK, rs2736340 –FAM167A, and rs7444 –CCDC116), we found that at exon-level, all expressed exons possessed causal cis-eQTLs. For example, rs2736340 is a causal cis-eQTL for all thirteen exons of BLK and for all three exons of FAM167A (S5 Table). These data suggest that gene-level analysis is capturing associations where all—or the majority of exons—are modulated by the cis-eQTL.
We found that within the SLE associated loci that showed evidence of a shared causal variant with a cis-eQTL (Table 3), there were many instances in which the proposed causal cis-eQTL modulated expression of only a single expression element. This enabled us to resolve the potential regulatory effect of the causal cis-eQTL to a particular transcript, exon, junction, or intron (S5 Table). We were able to resolve to a single expression element in nine of the twelve candidate SNP-gene pairs. For example, rs9782955 is a causal cis-eQTL for LYST at junction-level for only a single junction (chr1:235915471–235916344; cis-eQTL P = 1.3x10-03; JLIM P = 2.0x10-04). We provide depicted examples of this isolation analysis for candidate genes IKZF2 (S5 Fig), UBE2L3 (S6 Fig), and LYST (S7 Fig).
We provide a worked example of resolving the causal mechanism(s) using RNA-Seq for the novel association rs2304256 with TYK2 (Fig 2). The top panel of Fig 2A shows the genetic association to SLE at the 19p13.2 susceptibility locus tagged by lead SNP rs2304256 (P = 1.54x10-12). Multiple tightly correlated SNPs span the gene body and the 3′ region of TYK2 –which encodes Tyrosine Kinase 2—thought to be involved in the initiation of type I IFN signalling [33]. In the panel below, we plot the gene-level association of all SNPs in cis to TYK2 and show no significant association of rs3204256 with TYK2 expression (P = 0.18). At exon-, and intron-level, we were able to classify rs2304256 as a causal cis-eQTL for a single exon (chr19: 10475527–10475724; cis-eQTL P = 2.58x10-09; JLIM P<10−04) and a single intron (chr19: 10473333–10475290; P = 2.20x10-08; JLIM P = 2x10-04) of TYK2 respectively as shown in the bottom two panels of Fig 2A. We show the exon and intron labelling of TYK2 in further detail in S8 Fig. We found strong correlation of association P-values of the SLE GWAS and the P-values of TYK2 cis-eQTLs against at exon-level and intron-level, but not at gene-level (Fig 2B). The risk allele rs2304256 [C] was found to be associated with decreased expression of the TYK2 exon and increased expression of the TYK2 intron (Fig 2C). By plotting the cis-eQTL P-values alongside the JLIM P-values for all exons and introns of TYK2 against rs2304256 (Fig 2D), we clearly show that only a single exon and a single intron of TYK2 colocalize with the SLE association signal–marked by an asterisk (note that rs2304256 is a strong cis-eQTL for many introns of TYK2 but only shares a causal variant with one intron). We show the genomic location of the affected exon and intron of TYK2 in Fig 2E (exon 8 and the intron between exons 9 and 10). Intron 9–10 of TYK2 is clearly expressed in LCLs according to transcription levels assayed by RNA-Seq on LCLs (GM12878) from ENCODE (Fig 2E).
Interestingly, rs2304256 (marked by an asterisk in Fig 2E) is a missense variant (V362F) within exon 8 of TYK2. The PolyPhen prediction of this substitution is predicted to be benign and, to the best of our knowledge, no investigation has isolated the functional effect of this particular amino acid change. We do not believe the cis-eQTL at exon 8 to be a result of variation at rs3204256 and mapping biases, as the alignability of 75mers by GEM from ENCODE is predicted to be robust around exon 8 (Fig 2E). In fact, rs3204256 [C] is the reference allele yet is associated with decreased expression of exon 8.
In conclusion, we have found an interesting and novel mechanism that would have been concealed by gene-level analysis that involves the risk allele of a missense SNP associated with decreased expression of a single exon of TYK2 but increased expression of the neighbouring intron. Whether the cis-eQTL effect and missense variation act in a combinatorial manner and whether the intron is truly retained or if it is derived from an unannotated transcript of TYK2 is an interesting line of investigation.
We re-performed our integrative cis-eQTL analysis with the Geuvadis RNA-Seq dataset in LCLs using association data from twenty autoimmune diseases. This was to firstly reiterate the importance of leveraging RNA-Seq in GWAS interpretation and to secondly demonstrate that our findings in SLE persisted across other immunological traits. As the raw genetic association data were not available for all twenty diseases, we were unable to implement the JLIM pipeline which requires densely typed or imputed GWAS summary-level statistics. We therefore opted to use the Regulatory Trait Concordance (RTC) method, which requires full genotype-level data for the expression trait, but only the marker identifier for the lead SNP of the disease association trait (see methods for a description of the RTC method). We stringently controlled our integrative cis-eQTL analysis for multiple testing to limit potential false positive findings of overlapping association signals. To do this, we applied a Bonferroni correction to nominal cis-eQTL P-values separately per disease and per RNA-Seq quantification type. We rigorously defined causal cis-eQTLs, as associations with PBF < 0.05 and RTC ≥ 0.95. An overview of the analysis pipeline is depicted in S9 Fig and S10 Fig. Using an r2 cut-off of 0.8 and a 100kb limit, we pruned the 752 associated SNPs from the twenty human autoimmune diseases from the Immunobase resource (S6 Table) to obtain 560 independent susceptibility loci.
Our findings confirmed our previous results from the SLE investigation, and again support the gene-level study using the JLIM package. As before, we found that only 5% (28 of the 560 loci) of autoimmune susceptibility loci were deemed to share causal variants with cis-eQTLs using either gene- or transcript-level analysis (Fig 3A). Exon-level analysis more than doubled the yield to 13% (72 of the 560 loci) with junction-, and intron-level analysis also outperforming gene-level (10% and 8% respectively). When combining all RNA-Seq quantification types, we could define 20% of autoimmune associated loci (110 of the 560 loci) as being candidate causal cis-eQTLs—which corroborates our previous estimate in SLE using JLIM (24%).
By separating causal cis-eQTL associations out by quantification type, we found over half (65%) were detected at exon-level, and considerable overlap of cis-eQTL associations existed between both types (Fig 3B). Unlike in our SLE analysis, gene- and isoform-level analysis did capture a small fraction of causal cis-eQTLs that were not captured at exon-level. Our data therefore suggest that although exon- and junction-level, and to a lesser extent intron-level analysis, capture most candidate-causal cis-eQTLs. It is necessary to prolife gene-expression at all quantification types to avoid misinterpretation of the functional impact of disease associated SNPs.
We mapped the causal cis-eQTLs detected by all RNA-Seq quantification types back to the diseases to which they are associated (Fig 3C). Interestingly, we observed the diseases that fell below the 20% average comprised autoimmune disorders related to the gut: celiac disease (7%), inflammatory bowel disease (14%), Crohn’s disease (16%), and ulcerative colitis (18%). We attribute this observation as a result of the cellular expression specificity of associated genes in colonic tissue and in T-cells [34]. Correspondingly, we observed an above-average frequency of causal cis-eQTLs detected in SLE (22%) and primary biliary cirrhosis (37%); diseases in which the pathogenic role of B-lymphocytes and autoantibody production is well documented [34]. Note that there are 60 SLE GWAS associations in this analysis as these originate from three independent GWA studies (S6 Table). We further broke down our results per disease by RNA-Seq quantification type (Fig 3D) and in all cases, the greatest frequency of causal cis-eQTLs and candidate genes were captured by exon- and junction-level analyses.
We provide the results from our analysis as a web resource (found at www.insidegen.com) for researchers to lookup causal cis-eQTLs and candidate genes from the twenty autoimmune diseases detected across the five RNA-Seq quantification types. The data are sub-settable and exportable by SNP ID, gene, RNA-Seq resolution, genomic position, and association to specific autoimmune diseases. See methods for a walkthrough of how to access results.
By implementing a mixed model test of heterogeneity that accounts for the dependency structure arising from within-individual and within-gene expression correlations, we attempted to distinguish causal cis-eQTLs at transcript-, exon-, junction-, and intron-level that fitted either a systematic gene-model (characterized by a similar effect on expression across all elements within a gene) or a heterogeneous gene-model (where the cis-eQTL signal is only evident in a subset of expression elements). The full results of this analysis are found in S7 Table.
We found that across each RNA-Seq profiling type, the majority of causal cis-eQTLs exhibited heterogeneous effects on gene expression; indicative of alternative isoform usage (Fig 4A). Junction-level causal cis-eQTLs had the greatest proportion of heterogeneous associations (49 of 65 causal cis-eQTLs were heterogeneous—75%). Both systematic and heterogeneous causal cis-eQTLs were then stratified by whether or not they were also causal at gene-level. As expected, we observed that causal cis-eQTLs that were also detected at gene-level (Fig 4B) showed a greater proportion of systematic effects on gene expression than associations not detected at gene-level (Fig 4C). In both cases however, the heterogeneous model was more apposite. Interestingly, we found that the greatest frequency of systematic associations, which were not captured at gene-level, were observed at exon-level (42 of 76: 55%). This implies that exon-level analysis captures a near equal proportion of both systematic and heterogeneous effects that are not detected by gene-level analysis. We show four examples of systematic and heterogeneous causal cis-eQTLs stratified by their detection at gene-level quantification in Fig 5.
A previous investigation has suggested that causal variants of gene-level and transcript-level cis-eQTLs reside in discrete functional elements of the genome [18]. We therefore investigated whether this notion held true across the five RNA-Seq quantification types tested in this study. To accomplish this, we selected the causal cis-eQTLs from the twenty autoimmune diseases interrogated, and per quantification type, tested for enrichment of these SNPs across various chromatin regulatory elements taken from the Roadmap Epigenomics Project in LCLs (using both the Roadmap chromatin state model and the positions of histone modifications). We implemented the permutation-based GoShifter algorithm to test for enrichment of causal cis-eQTLs and tightly correlated variants (r2>0.8) in genomic functional annotations in LCLs (see methods) [25]. Results of this analysis are depicted in Fig 6. We found the 28 gene-level cis-eQTLs were enriched in two chromatin marks: strong enhancers (P = 0.036) and H3K27ac occupancy sites–a marker of active enhancers (P = 0.002). Transcript-level cis-eQTLs were also enriched in H3K27ac occupancy sites (P = 0.039) but were not enriched in any other marks. The 72 exon-level cis-eQTLs were additionally enriched in active promoters (P = 0.017). Interestingly, the 54 causal cis-eQTLs detected at junction-level were found to be enriched in weak enhancers only (P = 0.002); whilst the 43 intron-level cis-eQTLs were enriched in chromatin states predicted to be involved in transcriptional elongation (P = 0.001; 83% of intron-level cis-eQTLs). Disease relevant cis-eQTLs detected at different expression phenotypes using RNA-Seq clearly localise to largely discrete functional elements of the genome.
We quantified the number of causal cis-eQTLs and tightly correlated variants (r2>0.8) per quantification type that were predicted to be alter splice site consensus sequences of the target genes (assessed by Sequence Ontology for the hg19 GENCODE v12 reference annotation). We found only two of the 28 (7%) gene-level cis-eQTLs disrupted consensus splice-sites for their target genes compared to the 14% and 13% detected at exon- and junction-level respectively (Fig 6C). Our data suggest that although exon- and junction- level analysis leads to the greatest frequency of causal cis-eQTLs, the majority at this resolution cannot be explained directly by variation in annotated splice site consensus sequences (splice region/donor/acceptor/ variants).
We extended our investigation and performed genome-wide cis-eQTL analysis for all SNPs against gene-, transcript-, exon-, junction-, and intron-level quantifications. As with our integrative analysis of autoimmune risk loci, we found the greatest number of genome-wide significant cis-eQTLs and target genes (at a genome-wide FDR threshold of 5%) were detected at exon-level, followed by junction- and intron-level; with gene- and transcript-level being thoroughly outperformed (S8 Table and S11 Fig). We confirmed that all of the causal cis-eQTL associations detected in our integrative analysis with autoimmune risk loci reached genome-wide significance—owing to the stringent Bonferroni multiple testing correction adopted (S9 Table).
Elucidation of the functional consequences of non-coding genetic variation in human disease is a major objective of medical genomics [35]. Integrative studies that map disease-associated eQTLs in relevant cell types and physiological conditions are proving essential in progression towards this goal through identification of causal SNPs, candidate-genes, and illumination of molecular mechanisms [36]. In autoimmune disease, where there is considerable overlap of immunopathology, integrative eQTL investigations have been able to connect discrete aetiological pathways, cell types, and epigenetic modifications, to particular clinical manifestations [2,34,36,37]. Emerging evidence however has suggested that only a minority (~25%) of autoimmune associated SNPs share casual variants with basal-level cis-eQTLs in primary immune cell-types [9].
Genetic variation can influence expression at every stage of the gene regulatory cascade—from chromatin dynamics, to RNA folding, stability, and splicing, and protein translation [21]. It is now well documented that SNPs affecting these units of expression vary strikingly in their genomic positions and localisation to specific epigenetic marks [18]. The eQTLs that affect pre-transcriptional regulation—affecting all isoforms of a gene—differ in the proximity to the target gene and effect on translated isoforms than their co-transcriptional trQTL (transcript ratio QTL) counterparts. Where the effect size of eQTLs generally increases in relation to transcription start site proximity, trQTLs are distributed across the transcript body and generally localise to intronic binding sites of splicing factors [18,21]. In over 57% of genes with both an eQTL influencing overall gene expression and an trQTL affecting the ratio of each transcript to the gene total, the causal variants for each effect are independent and reside in distinct regulatory elements of the genome [18]. In fact, three primary molecular mechanisms are thought to link common genetic variants to complex traits. A large proportion of trait associated SNPs act via direct effects on pre-mRNA splicing that do not change total mRNA levels [21]. Common variants also act via alteration of pre-mRNA splicing indirectly through effects on chromatin dynamics and accessibility. Such chromatin accessibility QTLs are however more likely to alter total mRNA levels than splicing ratios. Lastly, it is thought that only a minority of trait associated variants have direct effects on total gene expression that cannot be explained by changes in chromatin. As RNA-Seq becomes the convention for genome-wide transcriptomics, it is essential to maximise its ability to resolve and quantify discrete transcriptomic features so to expose the genetic variants that contribute to changes in expression and isoform usage. The reasoning for our investigation therefore was to delineate the limits of microarray and RNA-Seq based eQTL cohorts in the functional annotation of autoimmune disease association signals.
To map autoimmune disease associated cis-eQTLs, we interrogated RNA-Seq expression data profiled at gene-, isoform, exon-, junction-, and intron-level, and tested for a shared genetic effect at each significant association. As we had densely imputed summary statistics from our SLE GWAS, we opted to use the Joint Likelihood Mapping (JLIM) framework [9] to test for a shared causal variant between the disease and cis-eQTL signals. This framework has been rigorously benchmarked against other colocalisation procedures. Summary statistics were not available for the remaining autoimmune diseases and therefore we implemented the Regulatory Trait Concordance (RTC) method for these diseases and set a stringent multiple testing threshold to define causal cis-eQTLs. We found the estimates of causal cis-eQTLs were near identical between the two methods used (Table 1 and Fig 3A). Exon- and junction-level quantification led to the greatest frequency of causal cis-eQTLs and candidate genes (exon-level: 13–18%, junction-level: 10–11%). We conclusively found that associated variants were in fact more likely to colocalize with exon- and junction-level cis-eQTLs when applying a nominal JLIM P-value threshold of <0.01 (Fig 1B and Table 2). Gene-level analysis was thoroughly outperformed in all cases (5%). Our findings that gene-level analysis explain only 5% of causal cis-eQTLs corroborate the findings from Chun et al [9] who composed and used the JLIM framework to annotate variants associated with seven autoimmune diseases (multiple sclerosis, IBD, Crohn’s disease, ulcerative colitis, T1D, rheumatoid arthritis, and celiac disease). They found that only 16 of the 272 autoimmune associated loci (6%) shared causal variants with cis-eQTLs using gene-level RNA-Seq (with the same Geuvadis European cohort in LCLs as used herein). In our investigation, we argue that it is necessary to profile expression at all possible resolutions to diminish the likelihood of overlooking potentially causal cis-eQTLs. In fact, by combining our results across all resolutions, we found that 20–24% of autoimmune loci were candidate-causal cis-eQTLs for at least one target gene. Our study therefore increases the number of autoimmune loci with shared genetic effects with cis-eQTLs in a single cell type by over four-fold. Interestingly, using microarray data from CD4+ T-cells Chun et al classified 37 of the 272 autoimmune loci (14%) as causal cis-eQTLs [9]—strengthening the hypothesis that autoimmune loci (especially those associated with inflammatory diseases of the gut) are enriched in CD4+ T-cell subsets and the cells themselves are likely to be pathogenic [25,34]. Microarray data are known to underestimate the number of true causal cis-eQTLs [10]. If we assume that by leveraging RNA-Seq we can increase the number of steady-state causal cis-eQTLs four-fold, we hypothesise that as many as ~54% of autoimmune loci may share causal cis-eQTLs with gene expression at multiple resolutions in CD4+ T-cell populations. A large RNA-Seq based eQTL cohort profiled across multiple CD4+ T-cell subsets will therefore be of great use when annotating autoimmune-related traits. Immune activation conditions further increase the number of causal cis-eQTLs detected in autoimmune disease [38]. We reason that although using relevant cell types and context-specific conditions will undoubtedly increase our understanding of how associated variants alter cell physiology and ultimately contribute to disease risk; it is clearly shown herein that we are only picking the low hanging fruit in current eQTL analyses. We argue it necessary to reanalyse existing RNA-Seq based eQTL cohorts at multiple resolutions and ensure new datasets are similarly dissected. Despite the severe multiple testing burden, we also argue that expression profiling at multiple resolutions using RNA-Seq may be advantageous even when looking for trans-eQTL effects. As trans-eQTLs are generally more cell-type specific and have a weaker effect size, we decided not to perform such analyses using the Geuvadis LCL data. Large RNA-Seq based eQTL cohorts in whole-blood will be more suitable for such analysis [19].
As well as biological reasons for using multiple expression phenotypes for integrative eQTL analysis, there are also technical factors to consider. Gene-level expression estimates can generally be obtained in two ways–union-exon based approaches [14,17] and transcript-based approaches [11,12]. In the former, all overlapping exons of the same gene are merged into union exons, and intersecting exon and junction reads (including split-reads) are counted to these pseudo-gene boundaries. Using this counting-based approach, it is also possible to quantify meta-exons and junctions easily and with high confidence by preparing the reference annotation appropriately [13,15,39]. Introns can be quantified in a similar manner by inverting the reference annotation between exons and introns [18]. Of note, we found intron-level quantification generated more candidate-causal cis-eQTLs than gene-level (Fig 3A). As the library was synthesised from poly-A selection, these associations are unlikely due to differences in pre-mRNA abundance. Rather, they are likely derived from either true retained introns in the mature RNA or from coding exons that are not documented in the reference annotation used. Transcript-based approaches make use of statistical models and expectation maximization algorithms to distribute reads among gene isoforms—resulting in isoform expression estimates [11,12]. These estimates can then be summed to obtain the entire expression estimate of the gene. Greater biological insight is gained from isoform-level analysis; however, disambiguation of specific transcripts is not trivial due to substantial sequence commonality of exons and junctions. In fact, we found only 5% of autoimmune loci shared a causal variant at transcript-level.
The different approaches used to estimate expression can also lead to significant differences in the reported counts. Union-based approaches, whilst computationally less expensive, can underestimate expression levels relative to transcript-based, and this difference becomes more pronounced when the number of isoforms of a gene increases, and when expression is primarily derived from shorter isoforms [20]. The Geuvadis study implemented a transcript-based approach to obtain whole-gene expression estimates. Clearly therefore, a gold standard of reference annotation and eQTL mapping using RNA-Seq is essential for comparative analysis across datasets. Our findings support recent evidence that suggests exon-level based strategies are more sensitive and specific than conventional gene-level approaches [22]. Subtle isoform variation and expression of less abundant isoforms are likely to be masked by gene-level analysis. Exon-level allows for detection of moderate but systematic changes in gene expression that are not captured at gene-level, and also, gene-level summary counts can be shifted in the direction of extreme exon outliers [22]. It is therefore important to note that a positive exon-level eQTL association does not necessarily mean a differential exon-usage or splicing mechanism is involved; rather a systematic expression effect across the whole gene may exist that is only captured by the increased sensitivity. By implementing a mixed model test of heterogeneity that accounts for the dependency structure arising from within-individual and within-gene expression correlations we found that causal cis-eQTLs captured by exon-level analysis that are not detected at gene-level, are derived from both systematic and heterogeneous effects on gene expression in almost equal proportions (Fig 4). Additionally, by combining exon-level with other RNA-Seq quantification types, inferences can be made on the particular isoforms and functional domains affected by the eQTL which can later aid biological interpretation and targeted follow-up investigations [10]. We clearly show this from our analysis of SLE candidate genes IKZF2 (S5 Fig), UBE2L3 (S6 Fig), LYST (S7 Fig) and TYK2 (Fig 2). For TYK2 we reveal a novel mechanism whereby the associated variant rs2304256 [C] leads to decreased expression of a single exon and increased expression of a neighbouring intron (Fig 2). By isolating particular exons, junctions, and introns, one can design more refined follow-up investigations to study the functional impact of non-coding disease associated variants. We show how our findings can be leveraged to comprehensively examine GWAS results of autoimmune diseases. We found nine of the 38 SLE susceptibility loci were causal cis-eQTLs (Table 3) for 12 candidate genes which we later functionally annotated in detail (S4 Table).
Taken together, we have provided a deeper mechanistic understanding of the genetic regulation of gene expression in autoimmune disease by profiling the transcriptome at multiple resolutions using RNA-Seq. Similar analyses leveraging RNA-Seq in new and existing datasets using relevant cell types and context-specific conditions (such as response eQTLs as shown in [38]) will undoubtedly increase our understanding of how associated variants alter cell physiology and ultimately contribute to disease risk.
RNA-Sequencing (RNA-Seq) expression data from 373 lymphoblastoid cell lines (LCLs) derived from four European sub-populations (Utah Residents with Northern and Western European Ancestry, British in England and Scotland, Finnish in Finland, and Toscani in Italia) of the Geuvadis project [18] were obtained from the EBI ArrayExpress website under accession: E-GEUV-1. The 89 individuals of the Geuvadis project from the Yoruba in Ibadan, Nigeria were excluded from this analysis. All individuals were included as part of the 1000Genomes Project. Expression was profiled using RNA-Seq at five quantification types: gene-, transcript-, exon-, junction-, and intron-level (the files downloaded and used in this analysis have the suffix: ‘QuantCount.45N.50FN.samplename.resk10.txt.gz’). Full methods of expression quantification can be found in the original publication and on the Geuvadis wiki page: http://geuvadiswiki.crg.es/)). We have also provided a breakdown of the quantification methods in S1 Fig. Expression data downloaded represent quantifications that are corrected for sequencing depth and gene/exon etc length (RPKM). Only expression elements quantified in >50% of individuals were kept and Probabilistic Estimation of Expression Residuals (PEER) had been used to remove technical variation [40]. We transformed all expression data to a standard normal distribution.
In summary, transcripts, splice-junctions, and introns were quantified using Flux Capacitor against the GENCODE v12 basic reference annotation [16]. Reads belonging to single transcripts were predicted by deconvolution per observations of paired-reads mapping across all exonic segments of a locus. Gene-level expression was calculated as the sum of all transcripts per gene. Annotated splice junctions were quantified using split read information, counting the number of reads supporting a given junction. Intronic regions that are not retained in any mature annotated transcript, and reported mapped reads in different bins across the intron to distinguish reads stemming from retained introns from those produced by not yet annotated exons. Meta-exons were quantified by merging all overlapping exonic portions of a gene into non-redundant units and counting reads within these bins. Reads were excluded when the read pairs map to two different genes.
SNPs genetically associated to systemic lupus erythematosus (SLE) were taken from the Bentham and Morris et al 2015 GWAS in persons of European descent [7]. The study comprised a primary GWAS, with validation through meta-analysis and replication study in an external cohort (7,219 cases, 15,991 controls in total). Independently associated susceptibility loci taken forward for this investigation were those that passed either genome-wide significance (P<5x10-08) in the primary GWAS or meta-analysis and/or those that reached significance in the replication study (q<0.01). We defined the lead SNP at each locus as either being the SNP with the lowest P-value post meta-analysis or the SNP with the greatest evidence of a missense effect as defined by a Bayes Factor (see original publication). We omitted non-autosomal associations and those within the Major Histocompatibility Complex (MHC), and SNPs with a minor allele frequency (MAF) < 0.05. In total, 38 independently associated SLE associated GWAS SNPs were taken forward for investigation (S1 Table). Each susceptibility locus had previously been imputed to the level of 1000 Genomes Phase3 using a combination of pre-phasing by the SHAPEIT algorithm and imputation by IMPUTE (see original publication for full details) [7].
Primary trait summary statistics file. A JLIM index file for each of the 38 SLE associated SNPs was firstly generated by taking the position of each SNP (hg19) and a creating a 100kb interval in both directions. Summary-level association statistics were obtained form the Bentham and Morris et al 2015 European SLE GWAS (imputed to 1000Genomes Phase 3). We downloaded summary-level association data (chromosome, position, SNP, P-value) for all directly typed or imputed SNPs with an IMPUTE info score ≥0.7 within each of the 38 intervals. The two-sided P-value was transformed into a Z-statistic as described by JLIM.
Reference LD file. Genotype files in VCF format for all 373 European individuals of the Geuvadis RNA-Seq project were obtained from the EBI ArrayExpress under accession: E-GEUV-1. The 41 individuals genotyped on the Omni 2.5M SNP array had been previously imputed to the Phase 1 v3 release as described [18]; the remaining had been sequenced as part of the 1000 Genomes Phase1 v3 release (low-coverage whole genome and high-coverage exome sequencing data). Using VCFtools, we created PLINK binary ped/map files for each of the 38 intervals and kept only biallelic SNPs with a MAF >0.05, imputation call-rates ≥ 0.7, Hardy–Weinberg equilibrium P-value >1x10−04 and SNPs with no missing genotypes, we also only included SNPs that we had primary trait association summary statistics for. These are referred to as the secondary trait genotype files. We then used the JLIM Perl script fetch.refld0.EUR.pl to generate the 38 reference LD files from the 373 individuals (the script had been edited to include the extra 95 Finnish individuals).
Cis-eQTL analysis. We created a separate PLINK phenotype file (sample ID, normalized expression residual) for each individual gene, transcript, exon, junction, and intron in cis (within +/-500kb) to the 38 lead SLE GWAS SNPs. We only included protein-coding, lincRNA, and antisense genes in our analysis as classified by Ensembl BioMart. Using the chromosome 20 genotype VCF file of the 373 European individuals (E-GEUV-1), we conducted principle component analysis (PCA) and generated an identity-by-state matrix using the Bioconductor package SNPRelate (S9 Fig) [41]. Based on these results, we decided to include the first three principle components and the binary imputation status (as 41 individuals had been genotyped on the Omni 2.5M SNP array were imputed to the Phase 1 v3 release) of the European individuals (derived from Phase1 and Phase2 1000Genomes releases) in the cis-eQTL analysis so to minimize biases derived from population structure and imputation status.
We used PLINK to perform cis-eQTL analysis using the ‘—linear’ function, including the above covariates, for each expression unit (phenotype file) in cis to the 38 loci (secondary trait genotype files). We performed 10,000 permutations per regression and saved the output of each permutation procedure. In cis to the 38 SLE SNPs were: 439 genes, 1,448 transcripts (originating from 456 genes), 3,045 exons (400 genes), 2,886 junctions (332 genes), and 1,855 introns (443 genes).
Joint likelihood mapping (JLIM) and multiple testing correction. Per RNA-Seq quantification type, a JLIM configuration file was created using the jlim_gencfg.sh script and JLIM then run using run_jlim.sh–setting the r2 resolution limit to 0.8. We merged the configuration files and output files to create the final results table which included the primary and secondary trait association P-value, the JLIM statistic, and the JLIM P-value by permutation. Multiple testing was corrected for on the JLIM P-values per RNA-Seq quantification type using a false discovery rate (FDR) as applied by the authors of JLIM. A JLIM P-value <10−04 means that the JLIM statistic is more extreme than the permutation (10,000). We classified causal cis-eQTLs as SLE associated variants that share a single causal variant with a cis-eQTL based on the following: if there existed a nominal cis-eQTL (P<0.01) with at least one SNP within 100kb of the SNP most associated with disease, the transcription start site of the expression target was located within +/-500kb of that SNP, and the FDR adjusted JLIM P-value of the association passed the 5% threshold. Candidate genes modulated by the causal cis-eQTL.
Using publically available resources, we systematically annotated the twelve SLE associated genes that were classified as being modulated by causal cis-eQTLs. The expression profiles at RNA-level across multiple cell and tissue types were interrogated in GTEx [42] and the Human Protein Atlas [43]—with the top three cell/tissue types documented per gene. We noted using Online Mendelian Inheritance in Man [44] any gene-phenotype relationships by caused by allelic variants and any immune-related phenotypes of animal models. Protein-protein interactions of candidate genes were taken from the BioPlex v2.0 interaction network (conducted in HEK293T cells) [45]. Using the ImmunoBase resource (https://www.immunobase.org/), we looked up each gene and noted if the gene had been prioritized as the ‘candidate gene’ within the susceptibility locus per publication. Finally, we counted the number publications from PubMed found using the keywords ‘gene name AND SLE’.
Autoimmune associated SNPs were taken from the ImmunoBase resource (www.immunobase.org). This resource comprises summary case-control association statistics from twenty diseases: twelve originally targeted by the ImmunoChip consortium (ankylosing spondylitis, autoimmune thyroid disease, celiac disease, Crohn's disease, juvenile idiopathic arthritis, multiple sclerosis, primary biliary cirrhosis, psoriasis, rheumatoid arthritis, systemic lupus erythematosus, type 1 diabetes, ulcerative colitis), and eight others (alopecia areata, inflammatory bowel disease, IgE and allergic sensitization, narcolepsy, primary sclerosing cholangitis, Sjogren syndrome, systemic scleroderma, vitiligo).
The curated studies and their corresponding references used in this analysis are presented in S6 Table. For each disease, we took the lead SNPs which were defined as a genome-wide significant SNP with the lowest reported P-value in a locus. Associations on the X-chromosome and within the MHC and SNPs with minor allele frequency < 5% were omitted from analysis, leaving 752 associated SNPs. We pruned these loci using the ‘—indep-pairwise’ function of PLINK 1.9 with a window size of 100kb and an r2 threshold of 0.8, to create an independent subset of 560 loci.
An overview of the integration pipeline using the twenty autoimmune diseases against the Geuvadis RNA-Seq cohort in 373 European LCLs is depicted in S10 Fig. Genotype data of the 373 individuals were transformed and quality controlled as previously described in the above methods sections (biallelic SNPs kept with a MAF >0.05, imputation call-rates ≥ 0.7, Hardy–Weinberg equilibrium P-value >1x10−04).
We opted to use the Regulatory Trait Concordance (RTC) method to assess the likelihood of a shared causal variant between the disease association and the cis-eQTL signal [46]. This method requires full genotype-level data for the expression trait but only the marker identifier for the lead SNP of the disease association trait. SNPs within the 560 associated loci for the expression trait were firstly classified according to their position in relation to recombination hotspots (based on genome-wide estimates of hotspot intervals) [47]. Normalized gene expression residuals (PEER factor normalized RPKM) for each quantification type were transformed to standard normal and the first three principle components used as covariates in the cis-eQTL model as well as the binary imputation status (as previously described above). All cis-eQTL association testing was performed using a liner regression model in R. Cis-eQTL mapping was performed for the lead SNP and all SNPs within the hotspot recombination interval against protein-coding, lincRNA, and antisense expression elements (genes, transcripts, exons etc.) within +/-500kb of the lead SNP. In cis to the 560 loci were: 7,633 genes, 27,257 transcripts (originating from 7,310 genes), 52,651 exons (5,435 genes), 48,627 junctions (4,237 genes), 34,946 introns (6,233 genes).
For each cis-eQTL association, the residuals from the linear-regression of the best cis-asQTL (lowest association P-value within the hotspot interval) were extracted. Linear regression was then performed using all SNPs within the defined hotspot interval against these residuals. The RTC score was then calculated as (NSNPs—RankGWAS SNP / NSNPs). Where NSNPs is the total number of SNPs in the recombination hotspot interval, and RankGWAS SNP is the rank of the GWAS SNP association P-value against all other SNPs in the interval from the liner association against the residuals of the best cis-eQTL.
We rigorously adjusted for multiple testing of cis-eQTL P-values using a Bonferroni correction per quantification type (corrected for number of genes, isoforms, exons, junctions, and introns tested) and per disease–as we wanted to keep our analysis as close to the authors of JLIM who themselves also adjusted per cell type and per disease. We stringently defined causal cis-eQTLs as associations with expression PBF < 0.05 and an RTC score ≥ 0.95. Candidate genes are modulated by the cis-eQTL.
Expression of gene elements (for example exons) within a gene are naturally correlated, as are expression data from the same individual. We therefore applied a linear mixed-effects model approach within each RNA-Seq quantification type to test for heterogeneity in cis-eQTL signal strength of causal associations. We firstly fitted a systematic gene-model containing a SNP allele dosage main effect (encoded 0, 1, 2) and two random effects terms indexing each individual (1|Sample) and each expression element found within the same gene (1|Target). We then fitted a heterogeneous gene-model containing the same terms plus a set of fixed-effect SNP dosage * expression element interaction terms. Both models were fitted via restricted maximum likelihood (REML = FALSE) using the lmer() function of the lme4 R package. A likelihood ratio test was used to determine significance (anova). P-values were corrected for multiple testing using a Bonferroni correction, correcting for all tests (n = 230) across all quantification types. PBF < 0.05 was deemed significant for the heterogeneous model.
To test for enrichment of causal cis-eQTL associations in chromatin regulatory elements we implemented the Genomic Annotation Shifter (GoShifter) package [25]. Chromatin regulatory elements were divided into two categories: chromatin state segmentation and histone marks. The genomic coordinates of the fifteen predicted chromatin state segmentations (active promoter, strong enhancer, insulator etc.) for LCLs (in the GM12878 cell-line) were downloaded from the UCSC Table browser (track name: wgEncodeBroadHmmGm12878HMM). Histone marks and DNase hypersensitivity sites were obtained from the NIH Roadmap Epigenomics Project for LCLs (GM12878) in NarrowPeak format. Sites were filtered for genome-wide significance using an FDR threshold of 0.01 and peak widths harmonised to 200bp in length centred on the peak summit (as used in the GoShifter publication).
We obtained all SNPs in strong LD (r2 > 0.8) with the causal cis-eQTLs by using the getLD.sh script from GoShifter (interrogating the 1000Genomes Project for Phase3 Europeans). Per quantification type, we then calculated the proportion of loci in which at least one SNP in LD overlapped a chromatin regulatory element (conducted one at a time per chromatin mark). The coordinates of the chromatin marks were then randomly shifted, whilst retaining the positions of the SNPs, and frequency of overlap re-calculated. This was carried out over 1,000 permutations to draw the null distribution. The P-value was calculated as the proportion of iterations for which the number of overlapping loci was equal to or greater than that for the tested SNPs (P < 0.05 used as significance threshold).
Genome-wide cis-eQTL analysis was performed using the normalized expression residuals for each quantification type, four population principle components, and quality controlled SNP genotype data of the 373 European individuals as already described. Cis-eQTL association analysis was performed using the MatrixeQTL R package fitting the linear-model function for all SNPs within +/-500kb of protein-coding expression targets [48]. The total number of SNPs, genes, targets, and SNP-gene targets tested are documented in S8 Table and S11 Fig. The issue of multiple testing was addressed by calculating a False Discovery Rate for each SNP-target pair per quantification type and thresholding associations below 5%.
R version 3.3.1 and ggplot2 was used to create heatmaps, box-plots, and correlation plots. Genes were plotted in UCSC Genome Browser [49] and regional association plots in LocusZoom [50]. To access the online results table, visit www.insidegen.com and follow the link ‘Lupus’ then ‘data for scientists’. The table is found under title ‘Expression data associated with different autoimmune diseases’.
|
10.1371/journal.pcbi.1005381 | Unifying view of mechanical and functional hotspots across class A GPCRs | G protein-coupled receptors (GPCRs) are the largest superfamily of signaling proteins. Their activation process is accompanied by conformational changes that have not yet been fully uncovered. Here, we carry out a novel comparative analysis of internal structural fluctuations across a variety of receptors from class A GPCRs, which currently has the richest structural coverage. We infer the local mechanical couplings underpinning the receptors’ functional dynamics and finally identify those amino acids whose virtual deletion causes a significant softening of the mechanical network. The relevance of these amino acids is demonstrated by their overlap with those known to be crucial for GPCR function, based on static structural criteria. The differences with the latter set allow us to identify those sites whose functional role is more clearly detected by considering dynamical and mechanical properties. Of these sites with a genuine mechanical/dynamical character, the top ranking is amino acid 7x52, a previously unexplored, and experimentally verifiable key site for GPCR conformational response to ligand binding.
| The biological functionality of several receptors and enzymes depends on their capability to sustain large-scale structural fluctuations and adopt different conformational states in response to ligand binding. This is the case for G protein-coupled receptors (GPCRs), the largest superfamily of signaling proteins in mammals and a primary pharmaceutical target. To better understand the functional dynamics of GPCRs, we have analysed the inter-residue distance variations across the available structures for several receptors of the rhodopsin-like family (class A). We first reconstructed the network of mechanical, rigid-like couplings between nearby amino acids and then identified those acting as dynamical/mechanical hubs. These were the sites whose virtual removal led to a significant softening of the overall mechanical network. After validating the biological relevance of these sites by comparison against known key functional sites, we singled out those regions which emerge as prominent mechanical hubs and yet have an otherwise still unknown functional role. The most relevant of such novel putative functional sites, which could be probed by mutagenesis experiments, is at interface of two transmembrane helices and we expect it to be crucial for assisting GPCRs conformational response to agonist binding.
| Mammalian G protein-coupled receptors (GPCRs) are the largest family of signaling proteins, with approximately ∼850 unique members up to now identified in the human genome [1, 2]. Given the size of this family, their ubiquitous expression, and their involvement in virtually every (patho)physiological process in mammals, it is not surprising that human GPCRs are targeted by more than half of current drugs [3].
GPCRs share a distinctive structural signature, namely seven α-helical transmembrane (TM) domains [4]. Such common structural organization strongly contrasts with the structural diversity of the agonists: these range from subatomic particles (a photon), to ions (H+ and Ca++), to small organic molecules, to peptides and proteins [4]. The presence of an agonist (or a photon in the case of rhodopsin) triggers specific downstream G protein-dependent signaling pathways.
The mechanisms that precisely control GPCR agonist binding and the following receptor activation have until very recently been hindered by a lack of crystallized active receptor states and receptor-ligand complexes. However, significant advances in crystallization has recently permitted the structural determination of several class A receptors in active state. Moreover, several mutagenesis and assay procedures were performed in an attempt to identify functionally important residues [5], along with specific micro-switches, i.e. small groups of residues that undergo conformational change during receptor activation [6, 7].
Despite a consolidated list of residues important for binding and/or function emerged, the findings are limited by their individualized nature [8].
Indeed, GPCRs are not rigidly switching between the alternative agonist-bound and inactive forms. They rather adopt a series of intermediate conformations influenced not only by association with ligands, but also by other receptors, signaling and regulatory proteins, by post-translational modifications, and by environmental cues [2]. The capability of GPCRs to engage with such diverse signaling machinery strongly depends on their conformational flexibility. All these diverse signaling events are indeed accompanied by dynamic conformational changes. Each state is likely represented by an ensemble of conformations [9].
A characterization of the conformational and structural dynamics of these proteins is therefore critical for understanding the molecular mechanisms underlying their function. A suitable comparative analysis of the available structures for these receptors ought to give insight into their structure–function relationship by clarifying the functional-oriented character of their internal dynamics [10].
While the inspection of GPCRs’ and G proteins’ structures has been essential to map out the accessible distinct signaling states, our knowledge is still limited regarding the internal dynamics of such states and the pathways that link them [11].
To our knowledge this problem has not yet been addressed systematically. The reason for its challenging character lies, at least in part, in the high structural heterogeneity of the conformers that bridge between the active and inactive forms. Such structural diversity, for instance, limits a priori the scope of general methods, such as elastic networks and normal mode analysis, which can otherwise be profitably used to identify low-energy collective modes from near-native fluctuations [12, 13].
Here, we introduce and apply a novel comparative tool that can single out those sites that act as hubs in the network of mechanical connections between the receptor residues, i.e. that are crucial for maintaining the integrity of the protein’s large-scale dynamics and mechanics.
We present and discuss this strategy, which is otherwise general and transferable, for the members of a specific GPCR class, namely the class A. This functional group was chosen precisely because of its well-populated and structurally diverse repertoire of conformers.
We analyzed the structural fluctuations across representative conformers to identify those residues that are central for the network of mechanical couplings, and hence the functional dynamics, of the receptors. Such sites have good overlap with known key residues, including those established by purely static structural considerations, but involve additional sites whose functional relevance, that is experimentally verifiable, emerges more clearly from a dynamical perspective.
We focus on GPCRs belonging to the rhodopsin-like class A. This class has currently the broadest structural coverage spanning between active, or partially active, and inactive forms. The set includes six different types of receptors, namely: A2A adenosine, β2 adrenergic, M2 muscarinic, μ-opioid, neurotensin NTS1 and rhodopsin (see Table 1).
The mechanical hubs of these receptors were identified with a three-step strategy described below and sketched in Fig 1, see Methods for further details.
As a first step, for each receptor we first retrieved all available PDB structures covering its conformational repertoire (Fig 1a). Next, for each pair of residues in spatial proximity (within 12Å on average), we computed their distance variations over the structural set. The variation amplitude is a measure of rigidity, and the residues’ pairwise distance variance can be used as an inverse measure of residues mechanical couplings [14–19]. Hence, this step allows to define the local mechanical network that underpins the receptors functional dynamics (Fig 1b and 1c).
In the final step, each amino acid is profiled based on how much its virtual “mutation”, performed by deleting from the network its local mechanical interactions, changes the network’s connectivity, an approach similar and alternative to measuring the centrality of a particular node in a network (Fig 1d). The higher is the perturbation induced on the network, the higher is the dynamical impact of the considered amino acid. The returned quantity is a measure of the relevance of each residue in establishing indirect couplings between structural fluctuations across distant parts of the receptors. For this reason we shall refer to it as the “mechanical bridging score”.
As we shall discuss later, amino acids with high mechanical bridging score are typically located at the hinge or interface regions between quasi-rigid protein domains and are accordingly well-suited to affect the long-range propagation of structural fluctuations, including functionally-oriented ones. Note that, because we consider intrinsically dynamical properties (structural fluctuations), our notion of bridging score can aptly complement previous GPCRs’ profiling based on networking properties defined from single, static, structures [20, 21].
For a robust identification of the aforementioned mechanical hubs, we combined the six mechanical bridging profiles of the different receptors (Fig B and C in S1 Supporting Information) into a single, average one. The average was taken over the set of corresponding residues (with same GPCRdb numbers [22]) that are shared by all considered receptors. The resulting profile is shown in Fig D in S1 Supporting Information along with its estimated error, which is significantly smaller than the profile variations.
The structure of rhodopsin, color-coded according to the average profile, is shown in Fig 2. One can see that the highest average bridging scores are found at the interface between transmembrane helices that are known to be relevant for the receptor activation, namely: TM3, TM6 and TM7 [7, 23]. Note that, compared to these helices, TM4 appears to be much less involved in the large-scale conformational variations of the receptors (see also Fig D in S1 Supporting Information).
The functional relevance of sites with high average bridging score can be shown more stringently by cross-referencing them with the list of currently known key residues for class A receptors based on the survey of Tehan et al. [23]. This list of residues was recently compiled by combining sequence- and structure-based selection criteria, that is by singling out residues that are both highly conserved as well as located along the pathway that structurally connects the orthosteric site and the G protein docking site. This connecting region coincides with a hydrophobic core that is central to the helix bundle. The top ranking sites for the average bridging score and those reported in ref. [23] are given in Table 2.
The overlap between our top ranking sites and the known key functional residues reported by Tehan et al. [23] was assessed by using the receiver operating characteristic (ROC) curve in Fig 3a. The curve shows that by running through our ranked list of residues, the “discovery” of the known functional sites occurs at a significantly higher rate than expected for a random reference case (the plot diagonal).
This is an indication that the average bridging score is able to capture with a significant degree of sensitivity those residues likely to be involved in the functionality of class A GPCRs.
This conclusion is further supported by comparing the ranking based on the average bridging score with one based on a purely static structural criterion. To this end, we ranked the amino acids based on their number of contacts. This allows for a transparent and equal-footing comparison, since the criterion exclusively considers the average amino acid connectedness, regardless of whether a contact is associated to a strong (i.e. rigid-like) mechanical coupling or not. This structure-based ranking criterion is inspired by previous works on GPCRs [20, 21] that demonstrated a correlation between sites with functional relevance and graph properties of the static contact map build on single receptor structures. This is confirmed by the marked deviation of the corresponding ROC curve from the diagonal in the plot of Fig 3a. The key observation that is relevant here is that the average bridging score ROC curve is well in line with the structure-based one, thus underscoring the functional significance of the mechanics-based ranking criterion. In addition, it prompts to understand the different insight that it can offer over pure structural approaches.
To clarify the latter point, we show in Fig 3b and 3c and Fig D in S1 Supporting Information the profiling of residues according to the dynamical or structural criteria. The comparative inspection indicates that the differences are mostly localized at specific portions of TM6 and TM7, which are high ranking for the mechanical bridging score, but not for the structural one. These regions, therefore, appear to have a key role across class A members that is genuinely tied to the receptors’ functional mechanics and hence cannot be detected from static structural observables.
The 10 sites with the highest average bridging score (Table 2) include residues forming the so-called hydrophobic hindering mechanism (HHM: 6x44, 3x43 and 6x40). Mutagenesis experiments have shown that this conserved hydrophobic triplet, that is contacted by other listed residues, namely 3x40, 6x43 and 3x44, is essential for the activation process of class A GPCRs [23]. The HHM triplet plus the proximal site 3x40, which has the second highest score, all take part in the structural rearrangements bridging the inactive and active state. The latter, in fact, depends on the HHM opening for establishing the water channel in the active conformation [23]. Residue 3x40 additionally participates in the transmission switch [7] and is highly conserved as a branched hydrophobic residue as well, see Table A in S2 Supporting Information [23].
Residue 7x42 is, instead, involved in a different molecular switch, i.e. the TM3-TM7 lock [7]. This is the main mechanism responsible for activation in rhodopsin and possibly one of the first switches triggered by ligand binding in other GPCRs. Position 7x45 is one of the most conserved residue in TM7 (Table A in S2 Supporting Information) [7]. Finally, the 3x36 position, though not conserved, was shown by site-directed mutagenesis experiments to have a stabilizing role for the inactive state [7].
Most of the top scoring residues listed in Table 2 are therefore sites with a demonstrated involvement in class A GPCRs activity. This validates the viability of dynamical profiling approaches in general, and the mechanical bridging score in particular, for singling out functionally important residues and providing a rationale for their relevance. Given these premises, of particular interest are those sites that have a high bridging score, but are not yet known as functionally relevant.
This is the case for site 7x52, that has the highest score in our analysis. This amino acid is part of the well-conserved motif NPxxY(x)5,6F, but is otherwise not particularly central in the static network of contacts, see Fig 3c and Fig D in S1 Supporting Information. Its functional relevance therefore has not been fully investigated before, though its possible participation in stabilising the TM6–TM7 interhelical interaction has been suggested by [24]. Mutations at position 7x52 were shown to constitutively activate the TSH (thyroid stimulating hormone) receptor [5, 25] by possibly disrupting the packing between TM6 and TM7. We therefore suggest site 7x52 as a putative novel site crucial for functionality. Again, the fact that its relevance does not emerge from structural considerations indicates that its role is likely to be a genuinely dynamical, or mechanical one.
We finally note that the highest scoring sites in Fig 2 are immediately adjacent to the region that the latest studies of refs. [26, 27] have identified as the most structurally affected by the activation/inactivation transitions. In particular, by comparing class A GPCRs with different activation states, Venkatakrishnan et al. [27] identified three G protein-coupling residues, 3x46, 6x37 and 7x53, whose contacts are disrupted during activation, and that are exposed to the G protein-binding pocket by the dislocation of the cytoplasmic side of TM6 away from the helix bundle.
A comprehensive and annotated list of sites so far addressed in mutagenesis experiments of class A GPCRs is provided in Table B in S2 Supporting Information. Further mutagenesis probings of residue 7x52, though for non-class A GPCRs, are given in Table C in S2 Supporting Information. The data in Table B and C, while not necessarily transferrable to a different class, are still fully consistent with our conclusion that site 7x52 has a key functional role and ought to be a good candidate for future mutagenesis experiments.
This conclusion is further supported from the bioinformatics analysis of the degree of evolutionary conservation of the key residues identified in this study (Table A in S2 Supporting Information). In particular, the physico-chemical characteristics of the residue in position 7x52 are highly conserved in all class A GPCRs from eukariotes. Specifically, in more than 80% of the sequences, the corresponding amino acid is branched and hydrophobic. This underscores the functional relevance of this position from an evolutionary point of view. Similar conservation trends are found for other residues of Table A in S2 Supporting Information, that are key for the functional mechanics, particularly the activation, of the receptors.
The conclusions of the previous section are supported by two complementary extensions of the analysis above. Specifically, we first repeated the bottom-up mechanical profiling of residues for a single receptor using an ensemble of structures obtained from a molecular dynamics simulation. Finally, we examined the mechanical role of residue 7x52 by using a top-down approach based on the quasi-rigid domain decomposition of all receptors.
For the first extension, we applied our protocol to conformers sampled by extensive atomistic molecular dynamics (MD) simulations of the μ-opioid receptor [28] started from both the inactive state and the ligand-bound active one. The MD ensemble provides a richer sampling of the active and inactive conformers and hence allows to capture the internal dynamics and mechanics with greater fidelity than from the sole pair of available crystal structures.
The results of the single-residue analysis for the μ-opioid receptor (Fig 4a) are well consistent with those of Fig 2, based on the cumulated profiles of all six receptors. Specifically, the highest scoring residues, highlighted in Fig 4a and listed in the caption, include conserved residues of helices TM3, TM6 and TM7, two residues of the HHM (6x40 and 6x44) and, again, site 7x52.
The analysis of the μ-opioid receptor MD simulation helps clarify a further important question, that is how sensitive is the bridging score profile to the size of the conformational ensemble. To this end, we measured the Pearson correlation coefficients between the profile computed from the combined active and inactive MD trajectories and the profile obtained from the two available experimental structures for μ-opioid receptor, corresponding to its active and inactive forms. In spite of the very different size of the two datasets, the profiles, shown in Fig E in S1 Supporting Information, are remarkably similar and their Pearson correlation coefficient is as high as 0.80.
A similar analysis has been performed on additional MD simulations run for M2 muscarinic receptor, including a 190ns-long simulation of the inactive state (PDB ID: 3UON) and a 200ns-long one for the active state (PDB ID: 4MQS) (for more details about the MD simulations setup, see the relative section in S2 Supporting Information). The resulting comparison is reported in Fig E in S1 Supporting Information as well, and again a very high correlation (0.87) with the original score based on crystal structures has been observed.
More insight into this result is provided by the additional analysis reported in Fig F in S1 Supporting Information which conveys, in the form of a color-coded matrix, the Pearson correlation coefficients between the profile computed from the combined trajectories and the profile computed from various pairs of snapshots picked at various points of either or both simulations. The matrix vividly shows that, despite the dataset size differences, the consistency of the profiles can be very high as long as the two snapshots are diverse enough to represent both the active and inactive forms.
Analogous conclusions hold for the other five receptors, see Fig G in S1 Supporting Information, for which an equally meaningful bridging score could be derived based solely on a single pair of active-inactive conformations. As an immediate consequence of this fact, the comparison between the score profiles from all the possible pairs of active-inactive structures of a receptor allowed us to assign error-bars to each data point, which are consistently smaller than the local profile variations we are interested in measuring (see Fig H in S1 Supporting Information).
We finally turn to the top-down analysis based on the quasi-rigid domain decomposition of the six class A receptors. To this purpose we used the SPECTRUS webserver [19]. This performs an optimal domain decomposition based on the internal distance variations across a set of representative structures. The analysis, an example of which is illustrated in Fig 4b for rhodopsin, presented two salient features that recurred across the different receptors.
First, the intracellular half of TM helix 6 was systematically identified as a quasi-rigid domain, consistent with its role in the internal rearrangements accompanying the receptors’ activation [23].
The second feature is that residue 7x52 is often assigned to the same rigid domain as TM6. Such domain association is interesting because intuitively one would otherwise always assign 7x52 to the TM7-based domain, to which it structurally belongs, see Fig 4b. As a matter of fact, site 7x52 is recognised part of the TM6 dynamical domain in a sizeable fraction (∼25%) of the subdivisions from 2 to 10 domains of the receptors, including the μ-opioid receptor MD simulations, see Fig I in S1 Supporting Information. This means that the displacements of 7x52, unlike other sites in TM7, are appreciably coupled with those of the cognate helix, TM6. Accordingly, 7x52 appears to act as an interface, bridging site between the two distinct mobile TM6- and TM7-based domains, as it is illustrated in Fig 4b for rhodopsin.
The recurrent difference of the dynamics- and structure-based assignment is consistent with the other evidence presented above that residue 7x52, whose functional role is still largely unexplored, is likely relevant for the mechanical response of class A GPCRs.
The current understanding of GPCRs functionality, and primarily the response to ligand binding, has been significantly shaped by the analysis of the growing number of their structures solved with X-ray or NMR [29]. Though such structures give valuable clues for the active states of GPCRs, they still include a limited set of snapshots of the likely conformational states induced by agonist and G protein binding. In addition, both experiments and atomistic MD simulations indicate that the receptors are capable of adopting multiple conformations, depending on the nature of the bound ligand. Our insight into the agonist- and G protein-initiated conformational changes is therefore still limited.
As a step towards clarifying this open problem, we devised and applied a strategy for identifying key sites presiding the functional dynamics and mechanics of class A GPCRs. This is the largest subclass and it has arguably the widest structural coverage, with conformers from 6 different receptor types (including rhodopsin) in different activation states. We analysed the internal structural fluctuations across the dataset. In particular, we focussed on the pairwise distance variations of corresponding amino acids which were used to infer the network of local mechanical couplings that underpin the large scale, and arguably functionally-oriented conformational changes. The mechanical network was finally analyzed to locate the few sites that most contribute to GPCR’s collective mechanics. To do so we identified the residues whose virtual deletion leads to the strongest softening of the overall mechanical response.
The viability of the approach to single out the most relevant functional sites was validated by the significant overlap between key sites for mechanical response and those known to be crucial for function based on independent and different criteria.
On the one hand, this result provides a concrete and vivid illustration of the relevance of dynamics- and mechanics-based criteria for locating key sites for enzyme functionality and hence prompts their use in combination with other more established structure-based static criteria.
On the other hand, the validation revealed that mechanically-relevant sites at interface between transmembrane helices 6 and 7 were not included in the list of previously known functionally-relevant positions. This was particularly the case for site 7x52, which is among the highest ranking ones for the mechanical response, and whose relevance is supported by the analysis of both atomistic MD simulations of the μ-opioid receptor as well as the analysis of GPCR’s rigid-domain decompositions.
Based on these convergent indications, we conclude that site 7x52 likely plays a key role in the conformational dynamics of class A GPCRs. Its functional relevance, as well as that of other sites in the central region of the transmembrane helical bundle, ought to be experimentally verifiable, e.g. with site-directed mutagenesis experiments.
The receptors’ mechanical network was inferred from the analysis of distance variations between pairs of amino acids. These, in fact, are key elements to define the subparts of the proteins that interact in such a concerted manner that they behave as quasi-rigid domains [19]. The distance variation fa,b between two residues a and b is computed as the standard deviation of the distances da,b between their Cα atoms over two or more structures (PDB entries or snapshots from MD simulations):
f a , b = ⟨ d a , b 2 ⟩ - ⟨ d a , b ⟩ 2 . (1)
The strength (rigidity) of the pairwise mechanical couplings is then quantified with a Gaussian weighting of the corresponding distance variations
σ a , b = exp ( - f a , b 2 / 2 f ¯ 2 ) , (2)
Because we are interested to define the receptors’ mechanical network in terms of physical, local coupling between amino acids, we set σa,b = 0 for amino acids whose Cα’s are at an average distance larger than 12Å, see Fig J in S1 Supporting Information. The value of the sensitivity parameter, f ¯, in Eq 2 is then set as the average of fa,b over the amino acids pairs closer than 12Å.
To define the key mechanical bridging sites, or hubs, of the receptors, we resort to the spectral clustering analysis of the mechanical network [30, 31].
Specifically, given the matrix, σ, of couplings between N amino acids, we characterize the spectrum of the symmetric Laplacian matrix,
L = I - D - 1 / 2 σ D - 1 / 2 , (3)
where I is the identity matrix and D is the degree matrix Da,b = δa,b ∑c σa,c. Its non-negative eigenvalues 0 = λ0 ≤ … ≤ λi ≤ … ≤ λN−1 provide information about how well the network is neatly partitioned in distinct clusters (mechanical domains) and, accordingly, are typically used to define optimal subdivisions of the network.
Here, the eigenvalues will be used for a different goal, namely to ascertain how important is each node to maintain the overall mechanical connectedness of the network. This amounts to measuring how much the network Laplacian spectrum changes when the connections, or couplings, of a node with its neighbors (excluding the connections corresponding to bonded interactions) are deleted. This response for residue k is given by the mechanical bridging score:
Δ k = Ω k - Ω 0 . (4)
where Ω 0 = ∑ ˜ i = 1 N - 1 1 λ i is the sum of the inverse eigenvalues (the tilde superscript denotes the omission of zero eigenvalues) for the full network, and Ωk is the same quantity but calculated for the network where the couplings relative to the kth node have been deleted.
The bridging score profile is computed separately for each receptor using its available structural representatives. The average bridging score is then obtained by averaging the bridging score over all equivalent positions of the various receptors.
The structures used for the analysis are listed in Table 1. Among the receptors whose structure is reported in the Protein Data Bank, we selected those for which both active and inactive conformations were known. These include the following receptors: A2A adenosine, β2 adrenergic, M2 muscarinic, μ-opioid, neurotensin NTS1, rhodopsin. Moreover, we applied the same analysis on an MD trajectory as well, obtained by merging two simulations of the μ-opioid receptor [28], starting from the inactive state (PDB ID: 4DKL [32]) and the active state bound to the agonist BU72 (PDB ID: 5C1M [33]).
Each of the six receptors included in our dataset had a minimum of two crystal structures (μ-opioid receptor) and a maximum of 21 (rhodopsin), including both active and inactive conformations.
The GPCRdb numbering scheme [22] has been used to match the residue positions common to all receptors. This scheme consists of the combination of two numbers in the form AxBB, where the first one is the helix number, while the second one is a progressive number chosen so that the most conserved residue in each helix has the value of 50.
Note that, because our main goal is to identify the key residues that are common across the various GPCR types, the analysis must necessarily focus on those amino acids that are in one-to-one correspondence across the heterogeneous GPCR set. This requirement lead, de facto, to exclude the residues involved in EL/IL loops, though one should be aware that their role in receptors’ activation is increasingly acknowledged [34]. Likewise, when defining the set of common positions, those residues, close to the intra- and inter-cellular regions, for which the process of cutting the surrounding connections could lead to unwanted disconnections of the network, were not included. Consequently, the remaining set of positions correspond to the transmembrane region of the receptors, with numbering: 1x36–1x56, 2x40–2x63, 3x24–3x54, 4x42–4x61, 5x38–5x60, 6x34–6x57, 7x36–7x43, 7x45–7x55.
|
10.1371/journal.pcbi.1000098 | How To Record a Million Synaptic Weights in a Hippocampal Slice | A key step toward understanding the function of a brain circuit is to find its wiring diagram. New methods for optical stimulation and optical recording of neurons make it possible to map circuit connectivity on a very large scale. However, single synapses produce small responses that are difficult to measure on a large scale. Here I analyze how single synaptic responses may be detectable using relatively coarse readouts such as optical recording of somatic calcium. I model a network consisting of 10,000 input axons and 100 CA1 pyramidal neurons, each represented using 19 compartments with voltage-gated channels and calcium dynamics. As single synaptic inputs cannot produce a measurable somatic calcium response, I stimulate many inputs as a baseline to elicit somatic action potentials leading to a strong calcium signal. I compare statistics of responses with or without a single axonal input riding on this baseline. Through simulations I show that a single additional input shifts the distribution of the number of output action potentials. Stochastic resonance due to probabilistic synaptic release makes this shift easier to detect. With ∼80 stimulus repetitions this approach can resolve up to 35% of individual activated synapses even in the presence of 20% recording noise. While the technique is applicable using conventional electrical stimulation and extracellular recording, optical methods promise much greater scaling, since the number of synapses scales as the product of the number of inputs and outputs. I extrapolate from current high-speed optical stimulation and recording methods, and show that this approach may scale up to the order of a million synapses in a single two-hour slice-recording experiment.
| The circuitry of the brain is defined by the connections (synapses) between its cells. Synapses are very small, so it is difficult to identify more than a few at a time using standard methods like electron microscopy or high-precision electrical recordings from cells. This study shows that it is possible to measure single synapses using low-precision methods such as optical recordings from neuronal cell bodies. I model optical or electrical stimulation of many inputs to trigger a visible response from neurons, and find single synapses by testing how this response is modulated when a single additional input synapse is triggered as well. I predict that it should be possible to record from as many as a million synapses using new optical recording and stimulation methods. It is believed that memories are encoded in synaptic connection patterns, so such connectivity data may give us a picture of how memories are encoded. We now know a great deal about how individual neurons behave, so a synapse-level wiring diagram would go a long way to fill out the picture of how neurons work together in the brain to interpret sensory information and plan actions.
| The neuronal wiring diagram of many mammalian brain regions is known in a statistical sense, but not at the level of individual neurons [1]. The hippocampal CA3 to CA1 region is a particularly simple circuit with considerable functional relevance in memory, and is therefore an interesting test case for working out detailed connectivity. An idealized way to work out the neuronal connection matrix (Figure 1F) is to stimulate one input neuron at a time, record the outputs of each CA1 neuron, and enter these values as the weights of that row of the connection matrix. The major exercise of this paper is to analyze how to detect individual synapses, despite experimental limitations that complicate this idealized approach.
The first limitation is stimulus specificity. How can we stimulate exactly one input neuron at a time? Recent optical stimulation experiments using localized glutamate uncaging [2]–[4] have been used to estimate spatial connectivity profiles in the hippocampus and cortex. These studies provide high spatial resolution, which approaches single neuron resolution. Optogenetics provides another approach [5],[6]. By inserting the channelrhodopsin-2 (ChR2) gene into hippocampal neurons, it is possible to stimulate cells with <5 ms precision [7]. Current ChR2 constructs have not been reported to be used with 2-photon excitation to obtain single-neuron specificity in the slice, but the method does provide for genetic targeting to specific neuronal populations (reviewed in [8]). High resolution is also possible using minimal stimulation on arrays of electrodes [9], though it is difficult to scale this to more than a few hundred inputs.
The second limitation is output sensitivity. Whole-cell patch recordings have long been used as sensitive measures of synaptic responses. Modeling and experimental studies have used patch-clamp data in the presence of spontaneous activity to obtain distributions of synaptic conductances [10],[11]. When coupled to electrical [12],[13] or optical [14] stimulation, patch recordings have been used to obtain spatial distributions of synaptic connectivity in networks. Individual synaptic data are more difficult to obtain in neural circuits. By performing large numbers of pair wise patch-clamp recordings it is possible to measure single synaptic connections in circuits (e.g., [15]). As many as seven cells have been reported to be patched simultaneously (e.g., [16]), and the current technical limit is ∼12. However, patch recordings do not scale well. Demanding experiments such as multi-patch recording can give connectivity information for tens of synapses, but one would like to work out circuitry for many times this number, and to do so in an individual slice. The alternative is to turn to more scalable but less sensitive methods. These include Ca2+ recordings and single-unit extracellular electrode recordings.
Ca2+ dye recordings have been used to simultaneously monitor hundreds of individual neurons [17]. Implanted extracellular electrodes have also been used to record from hundreds of isolated single units [18],[19]. Such recordings require the neuron to fire and generate a Ca2+ transient or an extracellular spike. This is a highly nonlinear process. Even with intracellular recordings, spiking nonlinearities greatly complicate the estimation of synaptic conductances [20]. It typically takes 50 or more simultaneous synaptic inputs to elicit an action potential in CA1 pyramidal neurons [21],[22]. In other words, we would not be able to see a response to a single input fiber stimulus. In this study we overcome this by adding a baseline stimulus to bring the output neuron near threshold, and then monitor how the addition of the single fiber stimulus changes the response.
The third limitation is stochasticity in synaptic release [23],[24]. Even with perfect optics and recordings, there is a modest (50% or smaller) probability that any given action potential will elicit a postsynaptic response at a given synapse. Furthermore, this probability is strongly history dependent. This introduces variability in neuronal responses to identical stimuli. While this variability complicates estimates of synaptic connection strength, it is also an essential requirement for the proposed method. Stochasticity in synaptic release helps the measurement by providing fluctuations near the action potential threshold, so that repeated samples reveal differences in distribution due to the addition of the single synaptic stimulus. A similar process has been proposed for physiological neuronal responses in the context of subthreshold background input, and has been shown to be effective in improving detection of single synaptic inputs [25].
In this study I perform a series of in silico experiments on the hippocampal CA1 network to design and analyze a synaptic estimation method that only needs extracellular or low-resolution optical recordings. These computational ‘experiments’ have the advantage that the correct synaptic weight matrix can be directly read out from the model definition, as well as from more physiologically practical readouts such as Ca2+ responses. This gives an unambiguous assay of the accuracy of synapse prediction. The method scales as the product of number of stimulus points and readout neurons. With current techniques this method should be able to resolve thousands of synapses, and it has the potential to scale to around a million.
I simulated hippocampal slice optical recording experiments designed to obtain synaptic weight matrices. The basic design of these experiments was to deliver a background stimulus to a block of Schaffer collaterals or CA3 neurons so as to bring postsynaptic CA1 cells above firing threshold. A probe stimulus was delivered to a single input neuron, over this background stimulus. By comparing responses to background and background+probe stimuli, the presence and potentially the strength of synaptic connections could be determined. In principle the background stimulus could be delivered directly to the output neurons using ChR2 or glutamate uncaging, but this seemed unnecessarily complex because the input axons/CA3 neurons would already be set up for the stimulation procedure.
I first calibrated the basic properties of the models. Then I explored different contributions to noise in the network and readouts. Finally I simulated a set of complete experiments including multiple sources of noise and several variants on the full network, to obtain an experimental and analysis design capable of reading the wiring of a large network.
I simulated 10000 input fibers and 100 CA1 output neurons with 19 compartments, multiple channel types, and Ca2+ dynamics (See Methods and Figure 1). The model size was determined by two considerations. First, many axons must be stimulated to elicit detectable Ca2+ responses. Assuming 5% connectivity, and a requirement for ∼50 simultaneous inputs, we would need 50/0.05 = 1000 axons to trigger an output action potential (Figure 1G). In order to have a reasonable number of such sets, the simulation used 10,000 input fibers. Second, the calculations needed a representative sample of output neuron properties, and enough output neurons that the random synaptic connectivity would form a representative distribution of inputs.
As a first pass approach to testing the feasibility of the approach, I modeled the 10000-axon, 100 neuron network without noise and with identical neurons (Figure 1F). I delivered inputs to a group of 1000 consecutive input fibers, numbered 1 to 1000. I selected one more fiber on which to deliver the probe stimulus in addition to the background 1000 inputs. I then advanced the entire stimulus set by 1 axon, so that the input block was from 2 to 1001. The probe stimulus was also advanced by 1 axon. I repeated this process for the entire set of 10,000 axons, so that all axons had been probed individually, and had contributed to the background 1,000 times.
To find the synapses I compared the response to background (RB) to the response to the background+probe (RP). Whenever RP>RB, it was inferred that a synapse was present between the probe axon and neuron. This was used to build up a synaptic connection matrix (Figure S1). This matrix was accurate but missed a few synapses (170 out of ∼50,000) because of premature truncation of the Ca2+ signal in the simulations. However, in the presence of as little as 0.5% Gaussian readout noise this method was completely inaccurate (Figure S1).
I considered four sources of noise in the experimental system:
Notably, the first three are inherent properties of the biological system and were incorporated into the model (Methods). Only the readout noise is under the technical control of the experimenter. It was added at analysis time.
In initial Monte Carlo calculations I computed the noise arising from probabilistic synaptic release for the baseline stimulus (Dataset S1, Figure 2A). I used estimates for single and double action potential releases [23], with the following parameters: probability of release on first pulse = 0.4, on subsequent pulses if no earlier release = 0.9, and on subsequent pulses after an earlier release = 0.55. While this was a simplistic model and did not account for all forms of synaptic and release variability (e.g.,[24],[26]), it did result in considerable stochasticity. The calculations showed that the signal-to-noise ratio was substantially better for paired action potentials (Figure 2A). Therefore the baseline stimulus protocol was designed to use paired pulses.
In a similar manner, I computed the distributions of actual number of inputs as a function of number of pulses in the probe stimulus (Figure 2B). There was a large standard deviation of ∼35% for 6 pulses.
Instead of using baseline stimuli, an alternative approach could be to use KCl to raise the cellular resting potential near threshold [13]. In principle, probe stimuli riding on a near-threshold depolarization should be able to elicit action potentials and Ca2+ transients. I modeled this experiment by altering the reversal potential of all modeled K+ channels, and depolarizing the membrane potential Em in all compartments in the neuronal models. The modeled network incorporated cellular variability and probe synaptic release variability. These simulations showed that probe stimuli elicited responses only in a very narrow window of ∼1 mV resting potential (Figure 2C, D, E). Above this window cells tended to go into bursting activity. Because of variability between cells, this window differed between cells. Thus, at least for CA1 pyramidal neuron physiology, KCl depolarization did not appear to be a viable approach.
To design effective stimulus patterns, I performed a series of simulations to characterize the distributions of responses without and with probe, RB and RP. I used only a single probe position, but carried out the RB and RP simulations 1000 times each. These runs excluded instrumentation noise but included probabilistic synaptic release and cell-to-cell variability. The simulations used 1300 baseline axons, and 6 pulses for the probe stimulus (Figure 3A).
I first assessed which of two readouts of neuronal responses (Ca2+ amplitude and timing) were most informative. I examined the raw distributions of RP and RB for sample neurons that were known from the model definition to be connected to the probe axon. As expected from spike-triggered Ca2+ influx, the amplitude responses were clustered into a small number of bins (Figure 3B, C). Similar multi-peak distributions have also been seen experimentally (Parameshwaran and Bhalla, unpublished data). RP distributions had more samples with larger responses, indicating that the probe stimuli occasionally elicited additional action potentials. If baseline synaptic noise was eliminated from the simulations, the difference in distributions was much smaller (Figure 3C–F). This suggested that the difference in distributions was amplified by stochastic resonance.
In the case of the timing responses, there was a small shift of 5–10 ms in the position of the RP vs. the RB distribution (Figure 3G, H). 10 ms is at or below the resolution of most Ca2+ recording methods, due to slow kinetics of most dyes. However, electrical recordings such as extracellular recordings, have <1 ms time resolution and would be well suited to using timing data. The current analysis is restricted to optical readouts and therefore does not use timing data.
I performed an initial survey of the amplitude data using a separation measure S based on means and standard deviations:(1)Here NR is the number of responding neurons, Ntot the total, 〈RP〉 and 〈RB〉 are the means of RP and RB, respectively, and σP and σB are the standard deviations of their distributions. This measure was not optimal given the strongly non-Gaussian nature of the response, but was useful for an initial characterization of the data. I considered four parameters that could affect the separation between responses to baseline and probe stimuli:
Overall the best value of separation S was ∼0.25. As a rough estimate, this should improve as √N, where N is the number of repetitions. The target accuracy is 0.05% errors, to achieve an error of less than 1 in 100 of connected synapses, which in turn are 5% of total possible connections. This requires around 4 standard deviations. If the baseline variability σB can be eliminated, the separability requirement is halved, so we would need a total of ∼64 repetitions.
Based on these data, I designed a stimulus procedure to resolve synapses. This stimulus design is shown schematically in Figure 5A, B. A movie of the stimulus applied to a reduced version of the network is shown in Video S1. The key features of this stimulus were as follows:
This total stimulus set was very large, requiring 800,000 stimuli for 80 trials per probe position. If we were to deliver 3 stimuli per second this would take ∼67 hours. Later I discuss how to reduce this to experimentally feasible durations.
I looked for differences between baseline and baseline+probe responses using two methods: standard errors (mean/SEM test), and a variant on the Kolmogorov-Smirnov (KS) test (see Methods). Each of these measures was able to resolve ∼10,000 synapses out of ∼50,000 in the network, with <100 false positives (Figure 5C, D). These estimates were for a Sorra-Harris distribution of weights [27], with 20% neuronal variability and 20% instrumentation noise. Although the mean/SEM test could resolve slightly more synapses with optimal threshold settings, these settings were not as consistent across noise levels and repeats as the p-value of the modified KS test. Importantly, this p-value enabled an estimate of the number of false positives (Methods).
I tested how the number of identified synapses scaled with the number of trials, as this was a key consideration in the experimental design (Figure 5E). As expected, there was a steady improvement in numbers. If the neuronal population had 50% cell-to-cell variability, this caused only a small reduction in the ability of the KS test to classify synapses. I then considered how experimental readout noise affected the number of classified synapses (Figure 5F). Increased noise degraded classification, but the falloff was graceful. Classification was ∼20% of synapses for 80 repetitions, for experimentally achievable noise levels of 10 % to 20%. For extremely low noise levels the original KS test was able to classify over 50% of synapses, but it failed with even modest noise levels (Figure S2).
I now had a Boolean synaptic weight matrix, with 0 or 1 entries to indicate absence or presence of synapses. This led to two questions: First, was I picking up only the stronger synapses? Second, could I estimate synaptic weights?
I first examined the distribution of weights of identified synapses, and compared this with the distribution for all synapses (Figure 6A, B). I found that most of the reported synapses were strong, but that there were a substantial number of strong synapses that were false negatives. The classification success was strongly dependent on position of baseline stimulus and probe (Figure 6C), consistent with the single-neuron analysis of threshold responses (Figure 1G). It reached a peak of nearly 40% in the middle of the dendrite. This suggests that the stimulus design could be further refined to give uniformly high classification. While it was possible to fine-tune the current stimulus design, the simplicity of the CA1 pyramidal neuronal models in the current study limits the utility of such fine-tuning. A more careful analysis would require more detailed neuronal models and experimental input. There was a correlation between synaptic weight and the P-value for significance of the KS test, suggesting that synaptic weights could be estimated by this approach (Figure 6D).
The current method was designed for the hippocampal slice-preparation. This preparation has very little recurrence in the CA1 and very low basal activity in typical low-potassium media, and these are reflected in the simplified feed-forward design of the model. To test the applicability of the method to a broader range of neuronal circuits and experimental contexts, I considered background activity, recurrence, feed-forward inhibition, and plasticity.
I first introduced random background synaptic input to represent the use of the method in an active network context. Random synaptic activity was added at 10 Hz and 70 Hz per apical compartment in separate simulations, without modifying the existing synaptic weights. As there were 12 apical compartments, this came to 120 and 840 inputs per cell per second, respectively. The 10 Hz input did not elicit action potentials, and it actually improved synaptic resolution to nearly 36% in the presence of 20% instrumentation noise. The 70 Hz input resulted in 1–3 spikes/second in the CA1 neurons, and completely abolished the ability of the method to resolve synapses.
I next considered circuit elaborations including recurrence and inhibitory interneurons (Figure 7A) along with the 10 Hz background activity. These circuit elaborations were purely a way to introduce complications into the simple feedforward network and were not meant to be accurate models of specific biological circuits. As expected, inhibitory interneurons fired very reliably following the input volley (Figure 7B), and reduced the number of spikes that the volley elicited in CA1 neurons (Figure 7C, D). Likewise, ∼60% recurrence alone had the expected effect of eliciting a burst of action potentials, which was truncated if inhibition was also present (Figure 7C, D). Surprisingly, the synapse detection was fairly insensitive to each of these circuit elaborations (Figure 7E). There was about a 40% reduction in ability of the method to resolve synapses when recurrence was present alone and 20% for inhibition alone. These drops in synapse detection were not due to circuit complexity, but to overall excitability. When both recurrence and inhibition were present (R+I) the excitability was similar to the original model (Figure 7C, D) and the drop was only 12%. Thus the method was good at identifying first-order synapses and ignoring polysynaptic input, at least in the circuit configurations tested.
A further generalization was to consider the effects of synaptic plasticity. This is a major concern of this approach, since the method relies on large numbers of volley stimuli that trigger a postsynaptic action potential. I analyzed spike timings of the postsynaptic cells following volley input, and found that most spikes occurred after 30 ms (Figure 7F). There was little overlap with standard STDP curves [28]–[30]. Based on the shape of experimental curves, I ignored the tail of the STDP curve beyond 30 ms. Taking the product of STDP with the number of spikes for the region before 30 ms, I found that the cumulative amount of potentiation was ∼12% for the case without background activity (0 Hz). Furthermore, these paired spikes could be spread out over a 2-hour period if the block stimuli were interleaved. Hence there would be considerable intervening uncorrelated activity which may act to restore the synaptic weight toward its set point [31]. I also analyzed spike timings with the 10 Hz background, and in this case the overlap was about twice as large and potentially more likely to introduce plasticity. However, the high background activity may again serve to balance out the plasticity over long periods.
The final step in the study was to analyze the scalability of the approach, assuming idealized optical recording capabilities. A specific target was to design the most informative 2-hour slice recording experiment. The design for this experiment was constrained by the characteristics of optical stimulation and recording, by neuronal projection patterns, by plasticity, and by the number of trials needed to build up statistical confidence. Plasticity effects are likely to be relatively small, as calculated above. I consider the number of trials here, and the remaining points in the discussion section.
As a baseline for this analysis, I considered the synapse selectivity achieved so far. Out of a possible 1 million synaptic contacts (10,000 inputs and 100 output neurons) the actual simulated circuit had ∼50,000 synapses, of which ∼11,000 (about 22%) were resolved using the modified KS method. Most of these were the strong synapses (Figure 8A, B). In separate simulations using 10Hz background activity the method gave ∼35% coverage of synapses (Figure 7E). However, both these simulated experiments required too many stimulus trials to be practical.
I analyzed the tradeoffs between number of trials, statistical confidence, and number of stimulated axons. To improve the detection of synapses, it was necessary to maximize the number of trials, by minimizing the duration of each stimulus cycle. The 10 ms interval between stimulus pulses was close to the minimum set by ChR2-stimulated firing rates [6], and also by the dynamics of synaptic priming and release [23]. This set 60 ms as the shortest time for the 7 probe pulses. Following these pulses, the Ca2+ response itself took ∼150 ms to complete (Figure 1). To allow for some settling, I considered total trial durations of 300 ms.
I analyzed a tradeoff that could increase the number of synaptic measurements by an order of magnitude. I considered supra-minimal electrode stimulation of X probe neurons, or equivalently, optical stimulation of groups of X neurons expressing ChR2. A stimulus would be unambiguous if there were either zero or one synaptic contacts per CA1 neuron out of this set of X axons (Figure 8C). For 5% synaptic connectivity, and 10% ambiguous synapses, X could be as large as ten axons.
I analyzed an optimally designed experiment of 2 hours, grouping probe axons into sets of ten as described above. I scaled the number of stimulus repetitions inversely with the number of axonal probes, so as to retain the same total experiment time. To do so I performed additional simulations with up to 240 repetitions, on a reduced network with the same 10,000 inputs but only 12 CA1 neurons because of computational limitations. I used the appropriate number of trials taking samples from the 240 and 80 repetition cases for subsequent calculations. In each case I stipulated that the number of false positives was less than 1% of the number of reported synapses. I first analyzed how the fraction of reported synapses per axon scaled with noise and number of axons (Figure 8D). Nearly 50% of synapses were identified for 10% noise and 1000 axons. I then considered how many synapses were reported per target neuron (Figure 8E). Here, the coverage of potential synapses rises with number of axons, but because of time limitations the number of repeats falls with greater numbers of axons. Overall, the number of synapses per neuron peaked at about 50 synapses for 6000 stimulating axons. Thus a 2-hour experiment, recording from 10,000 neurons with 10% instrumentation noise should resolve approximately 500,000 synapses.
This study shows that single synaptic inputs can modulate a suprathreshold background input to produce a measurable shift in the distribution of action potential firings, and consequent calcium transients. The method relies on stochastic resonance between the noisy baseline synaptic input and sub-threshold synaptic events, and generates a readout of action potentials which can be monitored using extracellular electrodes or calcium recordings. Current electrical and optical methods should already be technically capable of using such shifts to record hundreds of single synaptic weights. This study further predicts that new optical stimulation and optical recording methods may be deployed to obtain very large connectivity matrices with single-synapse resolution.
In order to validate the proposed approach, a conclusive experimental method for identifying synapses must be combined with this high-throughput experimental analysis. One possible experimental design would be to perform patch recordings in conjunction with bipolar electrode stimulation and dye recording from the target patched neuron. The patch recordings would detect putative single synaptic inputs to compare with the statistical analysis from the optical recording method. Paired patch recordings may be necessary to show that the input is from precisely one neuron. This experiment would allow us to test if the predicted true and false positives are as accurate as these simulations suggest.
Using this approach we can at best sample from about 50 synapses (<1%) per neuron, from perhaps 10,000 neurons in a slice (∼2% of hippocampal CA1 neurons) [32],[33]. How useful is such a sparse sampling of synaptic connectivity? While it is difficult to anticipate outcomes of these proposed experiments, there are grounds to expect that even a sparse functional wiring diagram would be very informative. First, known hippocampal representations of space are distributed and broad [34]. Recent direct experiments on hippocampal memory indicate that some aspects of memory traces may be observed even from a small number of recording electrodes [35]. Thus a sparse sample may cover a substantial number of synapses involved in ‘memory engrams’. Second, even a sparse circuit measurement may reveal signatures of repeated neuronal microcircuits (e.g.,[3],[12],[36]). Indeed, almost all current knowledge of vertebrate circuitry has been obtained from sparse sampling methods combined with neuroanatomy [1],[32]. Third, the coupling of precise but sparse functional data with new anatomical methods such as block-face sectioning [37] and multicolor genetic labeling [38] may build a more complete picture of neural circuitry than either approach on its own. Such a combination is especially important because geometrical connectivity does not always translate to functional connectivity [39].
This analysis was done on the relatively simple neuronal circuit in the CA1, and ignores interneurons. Other brain regions with more complex circuits will require their own stimulus designs and the deployment of multiple kinds of optogenetic or electrical stimuli. The KS analysis should be effective for inhibitory as well as excitatory inputs, but would not work well for weak synapses (Figure 6, 8).
In many neuronal circuits (e.g., cortex) there are many local circuits in addition to long-range fiber tracts. In such cases, interneurons and recurrence complicate the analysis, which is too slow to resolve polysynaptic effects. Our preliminary calculations (Fig 7) suggest that the method may be able to resolve monosynaptic input in the presence of as much as 60% recurrence. Nevertheless, it will require a cortex-specific study to better understand the capabilities of this approach in the far more complicated cortical circuit. A possible experimental approach to reversibly ‘simplify’ such networks is to transiently silence interneurons using pharmacological blockers or halorhodopsin [40].
While the current analysis assumes the use of brain slices, the general multi-input/multi-output approach is readily carried over to in-vivo recordings. Our data suggest that modest levels of background activity would be tolerated by the method. Optical methods have already been employed in vivo [5], and electrode recordings routinely monitor hundreds of neurons [18]. Electrode recordings have the additional advantage of fine time resolution, which allows the use of spike-timing data that was discarded in this study.
The fundamental benefit as well as difficulty of this approach is its scalability. The benefit is that the number of monitored synapses scales as the product of recorded and stimulated neurons. The difficulty is due to the increasingly stringent timing and accuracy requirements at larger scales. Current array electrodes have ∼60 contact points [9], each of which could be used with near-minimal stimulation to address a set of around 10 axons. Current optical methods can readily record from 100 individual cells in the slice [17]. With the assumptions of 5% connectivity and a 50% synapse detection rate, it should be possible to record from ∼1500 synapses in this configuration. This contrasts favorably with the current maximum of ∼12 patch electrodes, which should yield about 6 synapses assuming 5% connectivity. Beyond these current capabilities, a major goal of this study was to extrapolate from existing methods and set technical targets that would enable high-throughput recording.
The slice configuration itself would require some optimization. Neuronal projection patterns in the hippocampal slice are well known. With careful selection of the plane of slicing, it is possible to establish unbroken connections between CA3 and CA1 neurons. Nevertheless, it is challenging to retain enough connections to achieve several thousand intact axonal projections.
Both stimulation and recording may require optical techniques to scale up to very large network reconstructions. Methods already exist to do so for up to 1000 neurons in cortex [41]. In the hippocampus one may have to use separate CA3 stimulation and CA1 recording scanning optics. Two-photon methods are likely to be required for sufficient resolution in each case. This is currently feasible for the recordings, and for glutamate uncaging, but to my knowledge two-photon stimulation of optogenetic constructs is yet to be demonstrated. One possible configuration may use paired inverted and upright optical assemblies. Another possible configuration could utilize light guides [42] to provide the stimulus.
These simulations suggest a target of ∼80 repeats per input to achieve around 20% accuracy in synaptic identification. To deliver the required stimulus, the stimulating apparatus must generate reliable action-potential trains with ∼10 ms resolution, applied to ∼10,000 CA3 neurons. While this level of accuracy has been achieved with illumination of single neurons [40], it will require fast and precise scanning methods [43] to illuminate many neurons within this time window. The block design of the method relaxes these constraints significantly, so that only a handful (one to 10) axons/neurons must be stimulated precisely (within a ∼10 ms window) for any given trial, and the rest can be activated together using broad illumination or supraminimal electrical stimulation. In sum, the stimulus scaling targets appear achievable.
The technical issues with scaling up the number of recorded CA1 neurons are familiar ones of scanning speed versus signal-to-noise versus photobleaching. The suggested 2-hour experiments are feasible for a small number of neurons without much photobleaching using enhanced CCD cameras (Parameshwaran and Bhalla, unpublished data). It is more challenging to perform long recordings using 2-photon imaging, but improved calcium reporters may extend the duration of such recordings as well. There are already Ca2+ recording methods which can monitor ∼1000 individual neurons [17]. This suggests that it should be possible to scale up recordings to several thousands of neurons. Accurate algorithms for estimating spike counts from experimental Ca2+ waveforms have been developed [44],[45], and these may give higher classification accuracy with better noise immunity that the methods in this study. Combining ∼50-synapses per neuron (Figure 8E), and ∼10,000 recorded neurons, the target of almost a million synapses in a 2-hour experiment should be an ambitious but achievable technical goal. Such data would be a significant step toward reconstructing the functional wiring diagram of large neuronal circuits.
Input (CA3) neurons were modeled as single compartment passive cells with a spiking threshold and a 2 ms refractory period. Inputs were provided as a brief (60 microsecond) current pulse to represent electrical stimulation, but were also tested to give equivalent spiking output with smaller but longer current pulses representing light input to ChR2. Output (CA1) neurons were modeled as 19-compartment neurons slightly modified from Traub et al. [46] with the inclusion of NMDA and AMPA receptors. These neurons included Na, K, K_Ca and L-type Ca2+ channels and incorporated simple pump-based Ca2+ dynamics (Figure 1, Dataset S2). The time-courses of calcium in these cell models are too fast, for two reasons. First, the modeled neurons are based closely on models by Traub et al. [46], which use relatively rapid Ca2+ kinetics. Second, experimental recordings use Ca2+ indicators, which act as chelators and are therefore relatively slow. Optical recordings from CA1 somas have time-courses of the order of 150 msec (Parameshwaran, Madhavan, and Bhalla, unpublished data). However, the calcium time-course should not affect these calculations, because the analysis is based on the total area of the calcium transient.
Input axons were connected onto the NMDA and AMPA receptors of the CA1 neurons using a 5% connection probability [22]. Synaptic weights were set up using one of two Gaussian-based distributions: 1. A flat distribution with an upper cutoff (standard deviation = 1.0, upper cutoff = 1.0 standard deviations). 2. A narrower distribution with standard deviation = 0.5 and upper cutoff = 2.0 standard deviations, based on synaptic area estimates of Sorra and Harris [27]. The mean weight was set to 0.006 (arbitrary units) so as to give the response profile in Figure 1, where approx. 50 inputs were required to elicit an action potential. The requirement to keep this number of inputs around 50 meant that the synaptic conductances were somewhat smaller than estimated for CA1 synapses, because of the short length constant of the single simulated apical dendrite. The peak synaptic conductance reached following input on a single synapse was:(See Dataset S2.) I used a known random number seed for the network setup, so as to generate the same weight matrix to compare across many simulations. I used the same weights for NMDA and AMPAR conductances, but if the AMPA conductance was less than half of the mean it was set to zero to represent silent synapses.
I used two readouts for the Ca2+ response: (1) the area under the curve of the Ca2+ signal from 10 to 300 ms; (2) the time of the first Ca2+ transient, measured as time when the Ca2+ signal crossed a preset threshold.
Most runs used 10,000 CA3 neurons as inputs, and 100 CA1 output neurons, but for >80 repeats I reduced the model to 12 CA1 neurons because of computational limitations. Action potential propagation velocity was set to 1.0 m/s. The 10,000 Schaffer collaterals were distributed in the proximal 240–740 microns of the dendrite and ran in parallel.
I modeled variability between cells by scaling key passive and active properties of all neurons in the network using the equation(2)
I did not alter the channel kinetics or reversal potentials.
I modeled stochastic synaptic transmission using a simple Monte Carlo method based on measured release probabilities and facilitation [23]. I used the presence of inputs to the model CA3 neuron as a surrogate for probabilistic synaptic release on all synapses on the axon of that neuron. Different CA3 neurons used independent release calculations. I triggered stimuli with a 40% probability on the first pulse. After the first pulse, synaptic release probability was 90% if no release had yet occurred. If one or more releases had occurred, synaptic release probability was 55%. Note that the entire simulated axon was triggered with this probability, though in reality the individual synapses should function independently. This simplification should not affect the primary results as the CA1 neurons are independent in most of the calculations. In the recurrent circuits there may be some effects of this correlation across inputs but it is unlikely to affect synaptic detection.
This was added to all Ca2+ responses as a Gaussian distribution with a mean of zero, and a standard deviation set to the desired scaling factor. The random number generator was the Mersenne Twister [47].
All simulations were run using the GENESIS simulator [48] using a 50 microsecond timestep. Large calculations were run on a 260-CPU cluster of Opteron processors (Sun microsystems/Locus computing) running the Linux operating system. The simulation source files are provided as Dataset S3.
To analyze the responses I looked for differences between responses for each individual block+probe response vs. the combined responses for the entire block, as a reference. As an initial analysis I used means and standard errors of each of these distributions. Given the strongly non-normal distribution of responses, I then used the Kolmogorov-Smirnov (KS) test.
For the mean/SEM analysis, I used two parameters to tune the sensitivity:(3)I categorized a response as due to a synapse if S was greater than a threshold.
Here baseSEM was the first parameter, and the threshold the second.
For the KS test, I used the standard incomplete gamma function estimator (Q) for probability of obtaining the observed difference between baseline and probe distributions. I categorized a response as due to a synapse if the probability P was less than the threshold.
In both cases I set the threshold according to the criterion that less than 1% of the identified synapses should be false positives. The 1% false positive rate was picked as a conservative cutoff, because in circuit reconstruction false positives would be more problematic than false negatives.
This meant that the threshold had to be adjusted depending on the number of reported synapses.
The KS test provided a P value which mapped to the number of false positives more consistently than the 2-parameter mean/SEM test. The modified KS test also worked consistently with the inclusion of a scale factor:(4)This equation made it possible to obtain a good estimate of false positives, and hence to maintain accurate synapse selectivity from the data. The only additional datum required was an estimate of the number of potential synapses, which is the product of the synaptic connectivity and the number of stimulated axons. The synaptic connectivity value has been estimated for many systems and is around 5% for CA3 to CA1 projections [22].
I found that a scale factor of 10 was quite conservative. So, for ∼50,000 synapses in the simulations, there should be <80 false positives for a P-threshold of 0.00016. The actual value of false positives for P = 0.00016 was in the range of 30 to 50 for several variants of the model and at several values of instrumentation noise. Based on these estimates, the criterion of under 1% false positives would be met if there were over 8000 reported synapses for a P-threshold of 0.00016.
All statistical tests were custom coded in C++. The implementation of the KS test was based on Press et al. [49].
The original KS test was too sensitive to instrumentation noise. For extremely low noise the KS test gave very good results, but for even moderate levels of noise the test failed. This was because the algorithm was classifying responses based on subtle differences in peak amplitudes rather than on the number of action potentials. I therefore implemented a variant on the KS test that selected cases where the difference between the distributions spanned a wide response amplitude range (Figure S2). The specific modification to the KS test was that the maximum vertical difference used for the test should only be considered if the difference between the distributions had the same sign over a certain minimum amplitude (x-axis) range. This x-axis range had a value of 1.0+10% of the maximum amplitude in the distribution. For comparison, typical single calcium spikes had an amplitude of ∼10 units. Overall, this modification biased the KS statistic toward robust and large shifts in Ca2+ signal, such as might be expected for different numbers of action potentials.
I also tested how to combine responses for the same probe when it was stimulated along with different background blocks. I tried several ways of combining such responses, including taking logical combinations (AND and OR) of individual probe classifications, and summing the P or S values from the individual probes. Although combining probe information usually did improve synaptic classification, the improvement was less than simply running twice as many repeats on the same probe (data not shown). So the most economical way of obtaining good classifications seemed to be to simply use a single probe position.
|
10.1371/journal.ppat.1000939 | Requirement for Ergosterol in V-ATPase Function Underlies Antifungal Activity of Azole Drugs | Ergosterol is an important constituent of fungal membranes. Azoles inhibit ergosterol biosynthesis, although the cellular basis for their antifungal activity is not understood. We used multiple approaches to demonstrate a critical requirement for ergosterol in vacuolar H+-ATPase function, which is known to be essential for fungal virulence. Ergosterol biosynthesis mutants of S. cerevisiae failed to acidify the vacuole and exhibited multiple vma− phenotypes. Extraction of ergosterol from vacuolar membranes also inactivated V-ATPase without disrupting membrane association of its subdomains. In both S. cerevisiae and the fungal pathogen C. albicans, fluconazole impaired vacuolar acidification, whereas concomitant ergosterol feeding restored V-ATPase function and cell growth. Furthermore, fluconazole exacerbated cytosolic Ca2+ and H+ surges triggered by the antimicrobial agent amiodarone, and impaired Ca2+ sequestration in purified vacuolar vesicles. These findings provide a mechanistic basis for the synergy between azoles and amiodarone observed in vitro. Moreover, we show the clinical potential of this synergy in treatment of systemic fungal infections using a murine model of Candidiasis. In summary, we demonstrate a new regulatory component in fungal V-ATPase function, a novel role for ergosterol in vacuolar ion homeostasis, a plausible cellular mechanism for azole toxicity in fungi, and preliminary in vivo evidence for synergism between two antifungal agents. New insights into the cellular basis of azole toxicity in fungi may broaden therapeutic regimens for patient populations afflicted with systemic fungal infections.
| Systemic fungal infections impose a significant threat to public health and therapeutic options to treat these diseases remain limited. Azoles represent the largest category of anti-fungal drugs and repress fungal growth by inhibiting biosynthesis of ergosterol, an important constituent of fungal membranes. Despite the wide use of azoles in the clinic for decades, the cellular basis for their antifungal mechanism remains elusive. In this study, we use a range of genetic, cellular and biochemical approaches to reveal a requirement for ergosterol in vacuolar H+-ATPase function. V-ATPase plays essential roles in diverse cellular processes, and is required for fungal virulence. Concomitant ergosterol feeding restores vacuolar acidification and growth in cells treated with fluconazole. These results suggest that the critical requirement for ergosterol in V-ATPase function may underlie the antifungal activity of azoles. Moreover, we show in a mouse Candidiasis model that combining an ion homeostasis-disruptive drug with azole is an effective approach to treat fungal infections.
| Pathogenic fungal species, including Aspergillus, Candida, Histoplasma and Cryptococcus among others, cause infections ranging from mucocutaneous disorders to life-threatening invasive diseases that can involve any organ. In the past two decades, expanding populations of immunocompromised patients and increased use of invasive devices and implants have led to an increase in the incidence of fungal infections [1], [2]. Currently, four major categories of antifungal therapeutics are available to treat invasive fungal infections: polyenes, azoles, echinocandins and flucytosine [3]. Azole drugs are the most widely deployed in clinics, and inhibit the biosynthesis of ergosterol, the fungal-specific sterol. The primary molecular target of azole drugs is Erg11p (Entrez GeneID: 856398), a P450 cytochrome that catalyzes 14α-demethylation of lanosterol in the ergosterol biosynthesis pathway [4]. Besides azoles, a number of other drugs such as allylamines and morpholines used in medicine and agriculture also inhibit ergosterol biosynthesis [5], [6].
Ergosterol is an important constituent of membrane lipids, similar to vertebrate cholesterol, and modulates the fluidity, permeability and thickness of the membrane. These sterols preferentially associate with sphingolipids in microdomains that have been postulated to have important roles in membrane organization and function [7], [8]. Ergosterol is most abundant in the plasma membrane and has been implicated in several cellular processes including sporulation, pheromone signaling and plasma membrane fusion during mating and endocytosis [9], [10]. Discernable amounts of ergosterol have also been found in membranes of intracellular organelles including peroxisomes, mitochondria, vacuoles and ER [11]. Some studies have ascribed a regulatory role at these intracellular compartments, including homotypic vacuole fusion [12], mitochondrial biogenesis and inheritance, and protein sorting along exocytosis and endocytosis pathways [13], [14]. The absence of ergosterol in mammals and suppression of fungal proliferation by a battery of ergosterol biosynthesis inhibitors emphasize the importance and utility of ergosterol as an effective target in antifungal chemotherapy. Yet, despite nearly two decades of use and the general recognition of the importance of ergosterol to fungal cells our understanding of the specific cellular processes disrupted by ergosterol deprivation following azole therapy remains minimal.
The limited categories of antifungal agents and emergence of resistance to existing antimycotics have prompted a search for compounds with alternative modes of action. The anti-arrhythmia drug, amiodarone, was recently documented to exhibit fungicidal activity [15], [16]. This cationic amphipathic compound inserts into the lipid bilayer where it elicits membrane hyperpolarization, and influx of H+ and Ca2+ into the cytoplasm [15], [17]. Within minutes, amiodarone also elicits a transcriptional response to starvation and blocks cell cycle progression [18]. A screen of the yeast haploid deletion library for amiodarone hypersensitivity revealed multiple vma genes encoding subunits of the vacuolar membrane H+-ATPase [15]. The V-ATPase is critical for generation of a pH gradient that drives secondary transporters to maintain cellular ion homeostasis. Since the fungicidal activity of amiodarone appears to be tightly coupled to ion stress [19], hypersensitivity of vma mutants was ascribed to defects in ion homeostasis. Notably, deletion mutants of several erg genes in the ergosterol biosynthesis pathway were also identified in the screen [15], although the underlying mechanism for their amiodarone hypersensitive phenotype was unclear. Meanwhile, a screen of the yeast haploid deletion mutant library for strains with alterations in vacuolar pH revealed that erg mutants, like vma mutants, had severely alkaline vacuoles (Brett, C.L., Rao. R. et al., unpublished data). A separate study showed that sphingolipid, the other major membrane lipid component found associated with ergosterol in detergent-resistant microdomains, was required for the structural integrity of V-ATPase domains [20]. In light of these observations, we investigated a potential link between ergosterol and V-ATPase function, which led to a mechanistic basis for the antifungal activity of azole drugs. As a first step in exploiting our observations for improved management of invasive fungal diseases, we assessed the efficacy of combining fluconazole with ion homeostasis-disruptive agent amiodarone in a murine Candidiasis model.
A genome-wide screen of the S. cerevisiae haploid deletion collection for hypersensitivity to the antifungal agent amiodarone revealed multiple vma and erg mutants [15]. Given our previous observation that amiodarone triggered Ca2+ and H+ influx leading to fungal death from ion stress [15], [19] and the importance of V-ATPase in ion homeostasis, we considered the possibility that ergosterol may be important for V-ATPase function. A systematic examination of viable erg null mutants (erg2Δ [Entrez GeneID: 855242], erg3Δ [Entrez GeneID: 850745], erg6Δ [Entrez GeneID: 855003] and erg24Δ [Entrez GeneID: 855441]) revealed multiple vma− phenotypes, with erg24Δ displaying the most severe defects. In addition to hypersensitivity to amiodarone (Fig. 1A), erg24Δ was unable to grow at alkaline pH (Fig. 1B), a defining phenotype of vma mutants indicative of the inability to acidify vacuoles. Furthermore, erg24Δ exhibited hypersensitivity to Zn2+ toxicity and to the calcineurin inhibitor FK506, consistent with broad ion homeostasis defects characteristic of vma mutants (Fig. 1C–D). Yeast strains defective in trafficking of chitin synthase, including vma mutants, are more sensitive to toxicity from calcofluor white, an antimicrobial agent that binds to cell wall chitin. We showed that erg24Δ shared calcofluor white hypersensitivity with vma2Δ (Fig. 1E). Poor growth of vma mutants on high concentrations of non-fermentable carbon sources has been ascribed to oxidative stress from respiration. Although erg24Δ was able to grow on non-fermentable carbon sources, growth was significantly impaired (Fig. 1F). Overall, the novel observation that ergosterol biogenesis mutants largely phenocopy vma mutants suggests that cellular ergosterol content may be important for the function of V-ATPase.
V-ATPase hydrolyzes ATP and acidifies vacuolar compartments. To assess a possible requirement for ergosterol in V-ATPase function, we first measured the vacuolar pH in erg mutants using the pH-sensitive fluorescent dye BCECF. The acetoxy methyl ester of BCECF is taken up by cells and de-esterified in the vacuole where it accumulates [21]. While the vacuolar pH of wild-type cells was 6.0, vacuoles of vma2Δ (Entrez GeneID: 852424) cells were significantly more alkaline, around pH 7, as would be expected for loss of proton pump capacity (Fig. 2A). Vacuolar pH of all viable erg mutants closely resembled that of the vma mutant, as shown for erg24Δ (Fig. 2A). Next, we purified intact vacuolar vesicles from wild type, erg24Δ and vma2Δ strains, and compared V-ATPase function, including rates of proton pumping and ATP hydrolysis. There was no V-ATPase activity detectable in vma2Δ vacuoles as expected, whereas in vitro ATPase and H+ pumping activity were both diminished to about 40% of wild type levels in erg24Δ (Fig. 2B). Examination of sterol profiles in purified vacuoles confirmed the presence of ergosterol in wild type vacuoles and its absence in erg24Δ (not shown). Taken together, these results provide evidence for a role for ergosterol in V-ATPase activity.
The P-type H+-ATPase Pma1 (Entrez GeneID: 852876) has been documented to associate with ergosterol enriched domains [22], [23]. Upon glucose activation, it pumps protons out of cells to acidify the extracellular medium. To assess the effect of ergosterol depletion on Pma1 function, we examined extracellular acidification upon glucose activation in erg24Δ cells. As shown in Fig. 2C, the kinetics of medium acidification was substantially similar between the wild type and erg24Δ, while a previously characterized PMA1 mutant, pma1-105, reported to have a 65% reduction in activity, exhibited slower acidification rate as expected [24]. These data suggest that ergosterol is not required for Pma1 function and support the specificity of the ergosterol effect on V-ATPase.
To investigate the mechanism underlying the requirement of ergosterol for optimal V-ATPase function, we first asked if V-ATPase localization was altered in the erg24Δ mutant. The V-ATPase is made up of 14 subunits organized into the Vo sector, integral to the membrane, and the cytoplasmic V1 sector that reversibly dissociates from the membrane [25]. Figure 3A & 3B show that representative subunits from the Vo sector (Vph1-GFP) and the V1 sector (Vma5-GFP) colocalized with the vacuolar membrane stain (FM4-64) in erg24Δ cells, similar to the isogenic wild type. It was possible that the ergosterol biogenesis defect significantly decreased V-ATPase expression or caused the dissociation of V1 from Vo domain. However, analyses of representative V-ATPase subunits in vacuolar vesicles purified from the wild type and erg24Δ showed similar expression levels and V1/Vo ratios that were identical between the two strains (Fig. 3C & 3D).
The oligosaccharide Methyl-β-Cyclodextrin (MβCD) extracts sterols from cellular membranes. We observed a dose dependent loss of V-ATPase activity following treatment of purified vacuolar vesicles with MβCD, which could be blocked by preloading cholesterol into MβCD (Fig. 4A). Both ATP hydrolysis rates and H+ pumping declined at similar rates, suggesting an inhibition of the intact V1Vo complex. This was verified by immunoblot analysis of vacuolar membranes collected by centrifugation after MβCD treatment (Fig. 4B): we did not observe a loss of either V1 (Vma2p) or Vo (Vph1p) subunits, suggesting that the two sectors remain associated after ergosterol extraction. In contrast, a previous study pointed to a role for sphingolipids in maintaining structural integrity of the V-ATPase enzyme complex [20]. We conclude that ergosterol constitutes a critical component in the lipid membrane environment for V-ATPase function.
Azole drugs exert their fungistatic effect by inhibiting ergosterol biosynthesis, specifically targeting lanosterol demethylase (Erg11p), which is the enzymatic step immediately upstream from Erg24p. Despite the wide spread use of azole antifungals, our understanding of the specific cellular pathways disrupted by azoles is limited. In addition to ergosterol depletion, fluconazole treatment results in accumulation of lanosterol (substrate of Erg11p) and its derivative 14-methyl-3,6 diol [26]. Based on analysis of sterol profiles in fluconazole susceptible and resistant strains, it was concluded that 14-methyl-3,6 diol toxicity [26], [27] was responsible for azole-mediated growth arrest. To specifically assess the functional effect of ergosterol depletion following azole treatment, we used the S. cerevisiae strain WPY361 with a gain-of-function mutation in UPC2 (Entrez GeneID: 851799), upc2-1, that allows overexpression of ATP-binding cassette transporters required for uptake of exogenously added sterol under aerobic condition [28], [29]. Table 1 shows that growth inhibition caused by fluconazole treatment in WPY361 can be reversed by exogenous supply of ergosterol. To examine whether ergosterol feeding represses endogenous sterol metabolism and reduces accumulation of intermediates and derivatives, we analyzed sterol profiles in upc2-1 cells after six hours of exposure to fluconazole and exogenous ergosterol. As expected, fluconazole alone caused reduction of ergosterol content (nine-fold) and accumulation of lanosterol and a major derivative, likely to be 14-methyl 3,6-diol (Fig. S1). Compared with cells treated with fluconazole alone, cells treated with fluconazole and ergosterol had three-fold higher ergosterol content, yet the level of lanosterol and its derivatives remained the same (Fig. S1). These data indicate that ergosterol feeding did not reduce accumulation of intermediates and derivatives. Thus, depletion of ergosterol, rather than the toxicity of intermediates and derivatives, is a plausible mechanism for the antifungal activity of fluconazole.
Based on our evidence that ergosterol was required for optimal V-ATPase function, we predicted that azole treatment would similarly impair activity of the V-ATPase. Indeed, we show that fluconazole treatment resulted in a dose-dependent alkalinization of vacuolar pH, consistent with depletion of ergosterol from the vacuolar membrane (Fig. 5A). Furthermore, vacuolar membrane vesicles purified from fluconazole treated cells showed significant reductions in both H+ pumping rates and ATPase activity (Fig. 5B). Next, we evaluated the effect of ergosterol feeding on vacuolar pH in the upc2-1 mutant. Not only did exogenous ergosterol restore growth in fluconazole treated cells, vacuolar acidification closely resembled that of untreated cells (Fig. 5C & 5D). This correlation strengthens the hypothesis that V-ATPase inhibition contributes to the cellular mechanism of azole activity.
To extend these observations in the human pathogen Candida albicans, we monitored vacuolar uptake of the fluorescent weak base quinacrine. Previous studies have demonstrated pH-dependent vacuolar accumulation of quinacrine, which was abolished in the homozygous vma7−/− mutant [30]. We observed robust quinacrine fluorescence in C. albicans vacuoles, colocalizing with FM4-64 staining of vacuolar membranes. Fluconazole treatment drastically reduced vacuolar accumulation of quinacrine in most cells, indicative of impaired vacuolar acidification (Fig. 5E). Additionally, trafficking of FM4-64 to the vacuolar membrane was impaired, consistent with endocytosis defects seen in vma mutants [30]. We note that following 6 h of fluconazole treatment, cells failed to divide but continue to increase in size, as previously reported [9]. These data suggest that the requirement of ergosterol for V-ATPase function is conserved in fungi. Given the importance of V-ATPase function and vacuolar acidification in diverse cellular processes, we conclude that disruption of V-ATPase function plays a critical role in antifungal activity of azole drugs. Consistent with this conclusion, both vma7−/− and erg24−/−mutants of C. albicans exhibit defective virulence in murine models of Candidiasis [30], [31].
We showed previously that amiodarone triggered a cytosolic H+ and Ca2+ surge in the baker's yeast and that mutants defective in ion homeostasis were hypersensitive to amiodarone toxicity [15]. Given the pivotal role of the V-ATPase in maintaining intracellular cation homeostasis, we predicted that ergosterol depletion by azole treatment would impair V-ATPase function and exacerbate disruption of cation homeostasis by amiodarone.
We first tested this hypothesis in the S. cerevisiae model. We used pH sensitive GFP (pHluorin) to monitor changes in cytosolic pH in wild type, vma2Δ and erg24Δ strains upon exposure to amiodarone [32]. Cytosolic acidification was most pronounced in the vma2Δ mutant consistent with a loss in the ability to transport H+ from the cytosol to the vacuole. Depletion of ergosterol in erg24Δ mutant or by fluconazole treatment also exacerbated cytosolic acidification, relative to wild type, upon amiodarone addition (Fig. 6A). We have previously demonstrated defective clearance of cytosolic Ca2+ in vma mutants following exposure to amiodarone [15]. We now show that pretreatment with fluconazole exacerbates the cytosolic Ca2+ surge elicited by amiodarone, consistent with defective V-ATPase function (Fig. 6B).
The proton gradient established by the V-ATPase drives vacuolar sequestration of excessive cytosolic Ca2+ by H+/Ca2+ exchange mechanisms. The importance of the V-ATPase is demonstrated by the inability of purified vacuolar vesicles to sequester 45Ca2+ following treatment with the V-ATPase inhibitor concanamycin A (Figure 6C). As predicted by our hypothesis, vacuolar vesicles purified from erg24Δ or fluconazole-treated wild-type S. cerevisiae both showed similar impairment in their ability to sequester Ca2+ (Fig. 6C). Likewise, vacuolar vesicles purified from C. albicans cells treated with fluconazole were also impaired in sequestering Ca2+ (Fig. 6D). Thus, our observations provide a mechanistic basis for previous reports of synergism between azoles and amiodarone against pathogenic fungi in vitro [15].
To investigate the potential clinical application of this synergism, we studied the effect of combining fluconazole and amiodarone in a murine candidiasis model. The microbial burden of Candida albicans in kidneys was assessed 3 days following intravenous infection of Balb/C mice. AMD treatment doses were presented at 5.0 and 25 mg/kg, while FLC doses were 0.5 and 1 mg/kg with treatments given once daily for three days after the first dose. In the absence of amiodarone, there was a significant dose-dependent effect of FLC on C. albicans (ratio of geometric means of the cfu count per 1 mg/kg of FLC was 0.5914, p = 0.01). In the absence of FLC, amiodarone did not confer a significant antifungal activity at the concentrations tested (ratio of geometric means of the cfu count per 1 mg/kg of amiodarone was 0.9996, p = 0.95). Yet in combination with fluconazole, the two doses of amiodarone significantly reduced C. albicans infection above and beyond the dose-dependent effect of fluconazole (ratio of geometric means of CFU with amiodarone versus without, for the same dose FLC = 0.4969, p<0.001) (Fig. 6E). These data provide proof of principle that combining azole drugs with other antifungal compounds that disrupt intracellular cation homeostasis could be a promising therapeutic strategy to treat systemic fungal infections.
In fungal cells, V-ATPase acidifies intracellular compartments including the vacuole, endosomes, and late-Golgi. Mutants lacking V-ATPase exhibit characteristic phenotypes of growth sensitivity to alkaline pH, calcofluor white, Ca2+ and metal ion stress, and are unable to grow on high concentrations of non-fermentable carbon sources [25]. We show that mutants defective in ergosterol biosynthesis exhibit most of these characteristic vma− phenotypes. Furthermore, we showed a reduction of vacuolar acidification by ergosterol depletion, restoration of vacuolar acidity by ergosterol feeding, and used biochemical assays of H+ pumping with purified vacuolar vesicles to collectively demonstrate the requirement of ergosterol for optimal V-ATPase function. This functional link explains simultaneous identification of multiple ergosterol biosynthesis genes (erg) and V-ATPase subunit genes (vma) in a number of genome-wide screens, including sensitivity to low Ca2+, alkaline stress, and acid stress [33]–[35]. V-ATPase mutants and ergosterol biosynthesis mutants also share defects in endocytosis [9], [36]. Additionally, in pathogenic fungal species, both vma and erg mutants are avirulent [30], [31], [37]. Disruption of V-ATPase function by ergosterol deprivation provides a mechanistic basis for these similarities.
To investigate the underlying basis for the requirement of ergosterol in V-ATPase function, we first checked the localization and abundance of V-ATPase in erg24Δ. Fluorescent microscopy and immunoblot analysis ruled out possible mislocalization and reduced abundance of V-ATPase in erg24Δ vacuolar membrane. In yeast, the V-ATPase complex undergoes rapid reversible dissociation into non-functional V1 and Vo sectors in response to glucose withdrawal [38]. A study with Baby Hamster Kidney cells suggested that increasing ratio of Intact V1/Vo along the endocytic pathway effectively increased acidity along the compartments in the pathway [39]. Analyses of immunoblot results in this study show that V1 and Vo domains are still associated, and V1/Vo ratio remains unchanged upon ergosterol deprivation by treating cells with fluconazole and treating vacuolar vesicles with MβCD. These data rule out dissociation of V1 from Vo domain as the cause of reduced V-ATPase function, and indicate that ergosterol directly modulates the activity of V-ATPase. Sphingolipid, another major component of lipid raft, was thought to affect V-ATPase function by maintaining the structural integrity of V-ATPase because V1 subunits (Vma1p, Vma2p and Vma5p) dissociate from Vo domain during Ficoll gradient procedure in the sphingolipid mutant sur4Δ. In contrast, Vma2p remained associated with Vph1p after Ficoll gradient procedure in erg24Δ mutant and after ergosterol extraction by MβCD in the wild type. Thus, these two key membrane lipid components play distinct roles in maintaining V-ATPase function.
The precise molecular basis of V-ATPase regulation by ergosterol remains to be determined. In mammalian cells, V-ATPase has been shown to associate with cholesterol-rich microdomains, with loss of vesicular acidification reported upon treatment with β-methylcyclodextrin [40]. A number of studies showed that inhibitors could interact with lipid bilayer and affect V-ATPase function by restricting its structural flexibility [41], [42]. Mechanisms proposed to explain the regulation of Ca2+-ATPase and Na+,K+-ATPase by membrane lipids include lowering of free energy of activation and proper packing at protein-protein interfaces [43], [44]. Altered sterol compositions are known to affect membrane packing and rigidity: fluorescence anisotropy probes have revealed increased membrane fluidity and permeability upon fluconazole treatment [45], consistent with alterations in activity of membrane-localized pumps and transporters, and a critical role for ion homeostasis mechanisms in drug treated cells. It is also possible that ergosterol may affect V-ATPase function indirectly by modulating regulatory interactions with other proteins. The complex multi-subunit structure of V-ATPase and its intimate association with sterols and spingolipids indicate that sophisticated regulatory mechanisms must be in place to ensure proper assembly, configuration, and communication among these components. Ergosterol depletion may affect other membrane functions besides vacuolar acidification. The plasma membrane H+-ATPase, Pma1, is a major efflux mechanism for protons, and is also found associated with ergosterol-rich microdomains [22], [23]. However, we found that Pma1-mediated proton pumping function was not altered in erg24Δ, in contrast to the pronounced effect seen on the V-ATPase. This suggests that membrane proteins have specific lipid requirement for their functions.
While the molecular target of azole drugs, Erg11p, has been extensively characterized, not much is known about the cellular basis of fungal growth inhibition. Inhibition of Erg11p by fluconazole results in accumulation of lanosterol and its derivative 14-methyl-3,6 diol. In azole resistant erg3 mutants, 14-methyl fecosterol accumulates upon treatment with fluconazole [26], [27]. This has led to the notion that toxicity of sterol derivatives such as 14-methyl-3,6 diol mediates the action of fluconazole [45], [46]. We exploited the ability of a recently described upc2-1 mutation that allows uptake of ergosterol under aerobic conditions to distinguish between the effects of byproduct accumulation and ergosterol depletion on cell growth. The ability of exogenous ergosterol to reverse growth inhibition by fluconazole supports a plausible alternative hypothesis that antifungal activity of azoles is due to ergosterol depletion. Although we ruled out a corresponding decrease in lanosterol and other derivatives in the ergosterol fed cells, we cannot exclude the possibility that a specific ratio of ergosterol to other sterols may counter potential toxic effects. This possibility could be tested by varying ratios of sterols in upc2-1 feeding experiments; however, 14-methyl-3,6 diol is not commercially available and its potential toxicity cannot be directly assessed at this time. Furthermore, our results warrant more careful examination of the role of 14-methyl fecosterol in the potential compensation of ergosterol function in membranes.
Some reports suggest a role for ROS in the toxicity of azoles to fungal cells [47]. It is worth noting that cellular ROS level in these studies was measured with the fluorescent dye DCFH-DA (2′,7′-dichlorofluorescin-diacetate) which has also been documented to respond to pH alterations [48]. Given the effect of azole drugs on pH homeostasis, the role of ROS in azole-induced growth inhibition may need to be re-evaluated.
The far-reaching effect of V-ATPase is illustrated by diverse phenotypes exhibited by V-ATPase mutants. By disrupting the function of V-ATPase through ergosterol deprivation, azole drugs can affect a wide range of cellular processes, including cation homeostasis, protein sorting, processing and degradation. Importantly, V-ATPase function and vacuolar processing of virulence factors are required for pathogenesis [30], [37]. Although we cannot preclude additional cellular targets of azole toxicity, disruption of V-ATPase function is sufficient to repress growth and attenuate fungal virulence and is likely to be a critical mechanism underlying antifungal activity of azole drugs.
Amiodarone exhibits antimicrobial activity against a wide range of fungi and protozoa through disruption of H+ and Ca2+ homeostasis [15], [16], [49]. Additionally, in vitro studies showed azoles were synergistic with amiodarone against fungal pathogens, e.g. C. albicans and C. neoformans [15]. Interestingly, amiodarone interacts with fluconazole synergistically against fluconazole-resistant clinical isolates of A. fumigatus and C. albicans [50], [51]. Moreover, the synergy of amiodarone and posaconazole has been shown on the protozoan T. cruzi both in vitro and in vivo [49]. Uncovering the mechanism underlying this synergism may provide insight guiding the design of more potent antifungal therapy.
Data in this study show that ergosterol is required for the optimal function of V-ATPase, a central player in maintaining H+ and Ca2+ homeostasis. Therefore, depletion of ergosterol would be expected to exacerbate the disruption of H+ and Ca2+ homeostasis upon amiodarone treatment. Indeed, upon ergosterol depletion either by erg mutations or by azole treatment, 45Ca uptake by purified vacuolar vesicles was reduced while cytosolic H+ and Ca2+ surges increased upon exposure to amiodarone. Thus, we conclude that disruption of V-ATPase function in maintaining cation homeostasis by azoles contributes to the synergy between azoles and amiodarone.
Recently, we demonstrated that a combination of amiodarone and fluconazole in C. albicans resulted in dampening of the transcriptional response to either drug alone [52]. This effect could potentially stunt cellular stress responses occurring downstream of drug toxicity and thereby contribute to synergistic effects of the drugs. These data reveal an additional mechanism contributing to the synergism that is distinct from the ion homeostasis defects presented here. We and others have argued that non-overlapping but complementary insights can be obtained from phenotype versus transcriptional profiling [18]. Thus, genes involved in membrane integrity and ion homeostasis such as VMA and ERG, determine phenotype of growth sensitivity to amiodarone. These genes tend to be constitutively expressed, have a non redundant function and represent pathways upstream of the transcriptional response. On the other hand, genes that are differentially regulated in response to drug appear to play a collective, rather than individual, response to adaptation to stress. Taken together, these mechanisms contribute to a more complete picture of the complex cellular response to drug toxicity. Finally, in this study, we evaluated the clinical potential of combining fluconazole and amiodarone in treating fungal infections in a murine Candidiasis model. The synergy demonstrated in this experiment is proof-of-principle that combining azoles with agents disruptive to cation homeostasis is a promising approach to better manage fungal infections.
S. cerevisiae erg− and vma− mutant strains are from MATα deletion library (Invitrogen, Carlsbad, CA). WPY361 (MATa upc2-1 ura3-1 his3-11,-15 leu2-3,-112 trp1-1) was kindly provided by Dr. Will Prinz (NIDDK). Yeast cells were grown in standard SC (synthetic complete) medium or YPD (yeast extract, peptone and dextrose) medium at 30°C with shaking at 250 rpm unless specified otherwise. Media with non-fermentable carbon source contains 1% Bacto-yeast extract, 2% Bacto-peptone, 3% glycerol (v/v) plus 2% ethanol (v/v), or 2% sodium lactate. Antibodies against Vph1p and Vma2p were purchased from Invitrogen (Carlsbad, CA) or provided by Dr. Patricia Kane (Upstate Medical University, New York).
Plasmid pZR4.1 with pHluorin gene under TEF1 promoter was constructed to measure yeast cytosolic pH. Briefly, TEF1 promoter sequence was amplified from BY4742 genomic DNA with primers XbaITEF1 and BamHITEF1, which incorporate restriction sites of XbaI and BamHI. CYC1 terminator sequence was amplified with primers BamHICYC1 and EcoRICYC1, which incorporate restriction sites of EcoRI and BamHI. pHluorin gene sequence was amplified from pCB190YpHc plasmid with primers BamHIPhluo and BamHIPhluoR, which incorporate BamHI restriction site. The amplicons were digested with corresponding restriction enzymes and ligated sequentially with the backbone of pYEplac181, resulting in the gene for pHluorin being flanked by TEF1 promoter and CYC1 terminator. pZR4.1 was transformed to yeast strains to monitor cytosolic pH change upon exposure to amiodarone.
Wild type (mating type a) Vph1-GFP and Vma5-GFP strains in which GFP was fused to the C-terminus of the two proteins were purchased from Invitrogen. The Vph1-GFP and Vma5-GFP fragments were amplified from these strains with primers Vph1L2 + Vph1R1, and Vma5L2 + vma5R1, respectively. The amplicons were transformed to erg24Δ to replace the endogenous VPH1 and VMA5 genes by homologous recombination. Primer sequences are available upon request.
Ergosterol (Sigma) was dissolved in Tween 80∶ethanol (1∶1) as 5 mM stock. Stocks of ergosterol, fluconazole or their combination were added at the same time to WPY361 cells in YPD. For growth assay, stationary cultures were grown in 96-well plates for 30 hours at 30°C. For vacuolar pH measurement and sterol analysis, log phase cells were treated with ergosterol and fluconazole for 6 hours. Total sterols were extracted from ∼5×107 log-phase cells after washing twice with water and analyzed with an Agilent 6850 gas chromatograph with an HP-1 column and FID as described previously [53]. Retention times for cholesterol, ergosterol and lanosterol were determined using standards. Cholesterol was added to each sample to normalize extraction efficiency.
Vacuolar vesicles were prepared as described previously except that 10% Ficoll, instead of 8% Ficoll, was used in the second ultracentrifugation step to facilitate purification of vesicles from erg24Δ and fluconazole treated cells [54]. For 45Ca uptake assay, vacuolar vesicles were incubated in reaction buffer containing 5 µM CaCl2 with tracer quantities of 45CaCl2. After 5 min of incubation, vacuolar vesicles were filtered onto nitrocellulose membranes. The filters were washed and processed for liquid scintillation counting. Concanamycin A was added to 0.5 µM to assess the dependence of 45Ca uptake by vacuolar vesicles. For MβCD treatment, vacuolar vesicles were incubated with MβCD or MβCD preloaded with cholesterol (cholesterol to MβCD ratio 1∶20 by weight, Sigma, C4951) for 30 min at 4°C with gentle shaking. Vesicles were spun down and suspended in buffer C with proteinase inhibitors (aprotinin 2 µg/ml, leupeptin 1 µg/ml, pepstatin 1 µg/ml, chymostatin 2 µg/ml) for V-ATPase function assays and SDS-PAGE [20], [54].
ATP-dependent proton pumping was assayed by monitoring change of ΔA490–540 [36]. The initial H+ pumping rates were calculated from the absorbance change in the first 60 seconds. ATPase activity was assessed by an enzyme-coupled assay monitoring depletion of NADH through oxidation at 340 nm. ATPase activity was calculated based on the absorbance decrease between three and six minutes after initiating the reaction. Concanamycin A was added at a final concentration of 0.1 µM to assess the specificity of V-ATPase.
Cytosolic pH was measured using pHluorin, a pH-sensitive GFP [32], under TEF1 promoter in plasmid pZR4.1. Early log phase cells harboring plasmid pZR4.1 were grown for 6 hours to mid log phase (OD ∼1) in SC minus leucine medium with or without fluconazole. Cells were collected by centrifugation and suspended in SC to OD 3 in a 96-well plate, with 200 µl culture in each well. Amiodarone (stock 300 µM in H2O) was injected to give the specified final concentration. Fluorescence emission at 520 nm was measured in triplicate with excitation at 485 nm and 410 nm in a Fluostar Optima plate reader. Cytosolic pH was calculated based on the ratio of emission at 520 nm excited at 485 nm and 410 nm against a calibration curve covering pH from 4 to 8 established as described previously [32].
Vacuolar pH was measured with BCECF AM, a pH-sensitive fluorophore that accumulates in the yeast vacuole [32]. Strains were grown to mid log phase (OD ∼1) in SC medium. Cells were collected by centrifugation and incubated in SC containing 50 mM BCECF AM for 25 min. The cells were washed twice, suspended in SC to OD 2 and transferred to a 96-well plate. Fluorescence emission at 520 nm was measured in triplicate with excitation at 485 nm and 450 nm in a Fluostar Optima plate reader. Vacuolar pH was calculated based on the ratio of emission at 520 nm at dual excitations of 485 nm and 410 nm against a calibration curve covering pH from 4 to 8.5.
Proton efflux was assayed in wild-type and erg24Δ cells as described previously [55]. Briefly, cells were grown to ∼OD 1 in YPD. 1.5×108 cells were pelleted by centrifugation and washed twice with distilled water. The washed cell pellet was resuspended in distilled water and placed on ice for 2–3 h. Just prior to use, the cells were centrifuged and resuspended in 3 ml of water. Once a stable pH base line was established, glucose was added to 2% to initiate proton efflux.
Log phase cells of C. albicans (SC5314) were diluted to OD 0.025 and grown in YPD with or without fluconazole (1 µg/ml) for 5 hours. FM4-64 was added to 5 µM and the cultures were grown for another 30 min. Cells were washed twice in YPD, suspended in YPD with or without fluconazole, and shaken for 20 min. Quinacrine was then added to 100 µg/ml and the cultures were shaken for another 5 min. The cells were collected and washed twice in YPD and examined by fluorescence microscopy immediately.
All animal experimentation was conducted following the United States Public Health Service guidelines for housing and care of laboratory animals and performed in accordance with Institutional regulations after pertinent review and approval by the Institutional Animal Care and Use Committee of the University of Medicine and Dentistry of New Jersey.
Female BALB/c mice (age, 6 to 8 weeks; weight, 17 to 20 g) (Charles River Laboratories, Wilmington, MA) were used throughout the experiments. The mice were housed in micro-isolator cages with five animals per group and had access to food and water ad libitum. All animal experiments were conducted in the PHRI-UMDNJ Research Animal Facility. Disseminated infection with C. albicans (ATCC 36082) was induced by injection of 5.0×105 blastoconidia (∼5 times the median lethal dose) in 0.1 ml of sterile saline via the lateral tail vein of female BALB/c mice. Therapy was initiated 3 hours after challenge. Mice were treated with vehicle, Fluconazole (LKT Laboratories Inc., St. Paul, MN) (0.1 to 60 mg/kg of body weight/dose), Amiodarone (Henry Schein Animal Health, Melville, NY) (5.0 or 25 mg/kg of body weight/dose) or Fluconazole (0.1 to 60 mg/kg of body weight/dose) + Amiodarone (5.0 or 25 mg/kg of body weight/dose) administered intraperitoneally once a day for a total of 4 days. On day 4 post-infection, kidneys from euthanized mice (5 per group, unless specified) were aseptically removed, weighed, and homogenized in sterile saline using an IKA Works ULTRA TURRAX Tube Disperser Workstation (IKA Works Inc., Wilmington, NC). Serial dilutions of homogenized kidneys were plated onto YPD agar plates containing chloramphenicol (70 µg/ml) and ampicillin (50µg/ml) to eliminate bacterial cross-contamination. Culture plates were incubated at 30°C for 48 h, after which the CFU were counted and the number of CFU per gram of tissue was calculated. The method was sensitive for detection of ≥10 CFU/g. The culture-negative plates were counted as having 0 CFU/g.
The association of C. albicans proliferation with treatment group was assessed using multiple linear regression of log-transformed CFU counts. Preliminary model fit analysis was conducted to confirm that a parsimonious model which included dose-response slopes for AMD and FLZ, and a term for synergistic effect of dual drug treatment fit the data as well as the saturated model, where each of the nine treatment and dosage combinations were independent predictors (likelihood ratio test p = 0.47). The regression coefficients for drug dose are interpreted as slopes of the dose response curves. For the synergistic term, the exponentiated coefficient is the ratio of geometric means of the CFU for any given dose of each drug when the other drug is present versus when the other drug is absent. The significance of the association of CFU with predictors was assessed using Wald t-statistics for the regression coefficients.
|
10.1371/journal.pcbi.1004914 | Detecting Mild Traumatic Brain Injury Using Resting State Magnetoencephalographic Connectivity | Accurate means to detect mild traumatic brain injury (mTBI) using objective and quantitative measures remain elusive. Conventional imaging typically detects no abnormalities despite post-concussive symptoms. In the present study, we recorded resting state magnetoencephalograms (MEG) from adults with mTBI and controls. Atlas-guided reconstruction of resting state activity was performed for 90 cortical and subcortical regions, and calculation of inter-regional oscillatory phase synchrony at various frequencies was performed. We demonstrate that mTBI is associated with reduced network connectivity in the delta and gamma frequency range (>30 Hz), together with increased connectivity in the slower alpha band (8–12 Hz). A similar temporal pattern was associated with correlations between network connectivity and the length of time between the injury and the MEG scan. Using such resting state MEG network synchrony we were able to detect mTBI with 88% accuracy. Classification confidence was also correlated with clinical symptom severity scores. These results provide the first evidence that imaging of MEG network connectivity, in combination with machine learning, has the potential to accurately detect and determine the severity of mTBI.
| Detecting concussion is typically not possible using currently clinically used brain imaging, such as MRI and CT scans. Magnetoencephalographic (MEG) imaging is able to directly measure brain activity at fast time scales, and this can be used to map how various areas of the brain interact. We recorded MEG from individuals who had suffered a concussion, as well as control subjects who had not. We found characteristic alterations of inter-regional interactions associated with concussion. Moreover, using a machine learning approach, we were able to detect concussion with 88% accuracy from MEG connectivity, and confidence of classification correlated with symptom severity. This potentially provides new quantitative and objective methods for detecting and assessing the severity of concussion using neuroimaging.
| Detection of mild traumatic brain injury (mTBI) using neuroimaging remains a challenge, as no abnormalities are typically apparent using routine MRI [1, 2]. Accordingly, diagnosis is usually a clinical judgement based on self-report measures and behavioural assessments. Despite the lack of apparent injury on conventional clinical scans, many patients with mTBI suffer post-concussive symptoms (PCS). Although such symptoms typically resolve within a few months, a subset of individuals continue to experience long-term cognitive and behavioural impairments [3–5], underscoring the need for quantitative and objective methods for detecting and determining the severity of mTBI. The presence of lingering PCS indicates the presence of subtle brain injuries, with significant functional consequences that cannot be detected using current clinical techniques; there is a need to develop new imaging approaches for the detection of mTBI using quantitative and objective evidence.
Recent advances in magnetoencephalographic (MEG) imaging indicate that identification of mTBI is possible through detection of excessive slow-wave activity [6] and that this approach can localize the foci of the damage [7]. MTBI is associated with altered white matter microstructure as indicated by diffusion tensor imaging (DTI), in agreement with the view that mTBI results in axonal injury [8]. The focal excessive MEG slow wave activity has been shown to be related to the location of white matter injury, consistent with the supposition that oscillatory slowing can occur from deafferentation [9].
Disruption of inter-regional oscillatory synchrony in mTBI has been reported using EEG [10]. Oscillatory synchrony among brain areas is understood to play a vital role in network connectivity supporting cognition and behaviour [11, 12], and the expression of such neurophysiological network connectivity at rest relates to the intrinsic organization of brain activity pertinent for brain function and its dysfunction in clinical populations [13]. Converging evidence now indicates that traumatic brain injury is associated with diffuse axonal injury, which disrupts intrinsic functional network connectivity, thereby contributing to associated cognitive sequelae [14].
Machine learning approaches have been successfully combined with imaging of intrinsic functional brain connectivity during resting state to accurately classify single individuals [15, 16]. Moreover, electrophysiological recordings from subjects have also been shown to be effective for accurately determining group membership of individuals [17, 18]. We used MEG to investigate alterations in resting state oscillatory network synchrony in adults with mTBI, and investigated the hypothesis that machine learning algorithms could accurately detect mTBI in individual subjects.
Two groups participated in this study: adults with mTBI and adult healthy controls. Resting state MEG data were recorded, and neuromagnetic activity was reconstructed at anatomically-guided, a priori defined locations representing cortical and subcortical brain sources from the Automated Anatomical Labeling (AAL) atlas. The time-frequency representation of the source dynamics was derived using wavelet decomposition. Functional connectivity between the neuromagnetic sources was estimated in terms of frequency-specific phase-locking values. A linear support vector machine (SVM) algorithm in combination with cross-validation was applied to classify the subjects into two groups: mTBI and controls. Classification analyses were performed on different sub-sets of features, associated with different oscillatory frequencies. Classification accuracies were estimated. Associations between confidence of classification and self-reported clinical scores were investigated in the mTBI group. In addition, Partial Least Squares analysis was applied to evaluate the statistical reliability of group differences in functional connectivity.
Frequency and source resolved imaging of resting MEG network synchrony was able to accurately detect whether individual participants had been diagnosed with mTBI or not. Fig 1 shows the predictive power of phase synchrony measured at 30 specific frequencies points covering the range between 1Hz and 75Hz. Specificity remains stable around 80% and fluctuates slightly (76–83%) within a relatively wide range of frequencies: from 3Hz to 50Hz. Conversely, sensitivity and hence total accuracy have a local maximum around 8-13Hz (α rhythms), reaching 80%.
Fig 2A provides a more aggregated view of the results shown in Fig 1. Specifically, prediction accuracy is given as functions of frequency bands, each including the features from several frequency points (wavelets). Fig 2B illustrates the same prediction accuracies with respect to the random chance prediction, wherein the distributions of accuracies were generated by shuffling the labels (mTBI or not) across subjects, and repeating the same procedure 500 times. With an accuracy of 80% (p < 0.01), inter-regional resting state phase synchrony in the α band carries the most discriminative information for inferring the presence or absence of mTBI within a single individual. Accordingly, the remainder of the results presented in this section pertain to phase synchrony in the α band, using features that individually provided the highest separability between mTBI and controls under the ROC criterion.
The list of ranked features reflects an estimate of how valuable a given feature was found to be for classification. We can choose the best number of features, i.e. the number that maximizes prediction accuracy. In this case, the dimensionality of the feature space will correspond to the number of source-pairings within the alpha range. Note that in this study the features were ranked using training data at each round of leave-one-out cross-validation. While the number of features to keep was set a priori, the best features themselves were determined within each round of cross-validation. Fig 3 shows accuracy values as functions of the number of best features selected for classification analysis with cross-validation. As can be seen from Fig 3C, classification accuracy can be improved with a proper threshold on the number of variables k with two peaks around k = 8−15 and k = 30−35. For example, for k = 33 accuracy is 88%, with 90% specificity and 85% sensitivity (all p < 0.01).
To investigate potential relations between SVM classification and symptom severity obtained from the concussion assessment tool (SCAT2), we quantified the distance to the decision boundary for each subject, and correlated these values with clinical scores for participants within the mTBI group. Note that the larger the distance that an individual is from the decision boundary, the higher our confidence that a subject with mTBI is classified correctly as mTBI. Similar to the procedure shown in Fig 4, for each subset of features k = 1, …, 100, we computed the distances to the decision boundary for mTBI patients, and correlated these distances with the severity and symptom scores, shown in Fig 4A and 4B, respectively. Two scatter plots with superimposed least-squares regression lines illustrate relations between these variables at two peaks, k = 11 for severity (Fig 4C), and k = 33 for symptoms (Fig 4D). Note that negative distances at the scatter plots reflect cases of misclassification, when the learning function F(x) projects the feature vectors x of mTBI subjects to other side of the optimal hyperplane, corresponding to controls. Moreover, the confidence of classifying a subject as mTBI positively correlated with the self-reported severity scores (Fig 4A), reaching a local maximum (r = 0.54, p < 0.05) at k = 11. It also correlated positively with the symptoms scores with a peak of r = 0.34 (p-value< 0.10) around k = 33.
Finally, Fig 5 illustrates a distribution of the connections extracted from the pool of the best k = 33 features in the α band. It plots connections within a transparent template of the brain in the MNI space, using the BrainNet Viewer [19]. The width of the connections represents the weights are between −1 and 0, where being close to −1 implies a robust contribution of a specific connection to classification, and zero means no contribution. Specifically, for each wavelet frequency within the α range and each round of cross-validation (m = 1, …, 41), we assigned −1 to a connection if this feature survived the threshold and participated in classification, otherwise it was 0, subsequently averaging across subjects and wavelets frequencies. The ability to predict evidence of injury of a subject is largely based on synchrony between frontal and parietal/temporal sites, located mainly in the left hemisphere.
We also employed PLS to characterize and test the statistical reliability of differences in resting state network synchrony between adults with and without mTBI. This analysis revealed the existence of one significant latent variable (p = 0.002) which indicated alterations of resting MEG network synchrony in mTBI (Fig 6A). The overall distribution of all the bootstrap ratio values, each associated with a pair-wise connection between the sources and frequencies, is shown in Fig 6B. As can be seen, there are relatively large positive and negative bootstrap ratio values, which reflect phase-locking and phase scattering effects, respectively, in controls with respect to mTBI.
The difference between increased and decreased phase locking is broken down further in Fig 6C and 6D, which shows how the strength of these effects varies across frequencies. Specifically, we identified the 1% tails, cut off by the 0.01− and 0.99-quantiles of the overall distribution of the bootstrap ratio values in Fig 6B. At each frequency, the number of connections with the bootstrap ratio values larger than the 0.99-quantile (right tail) was computed and plotted in Fig 6C. The strongest effects are robustly expressed at δ and lower γ frequencies, directly supporting higher phase locking in controls compared to mTBI at these frequencies. Similarly, the number of connections in the left tail defined by the 0.01-quantile is plotted in Fig 6D, as a function of frequency. These connections also support the contrast in Fig 6A, but in a reverse way, representing hyper-connectivity in the mTBI which were strongest at α frequencies.
Pair-wise connections that show decreased phase synchrony in the δ and lower γ bands in mTBI are depicted in Fig 7. The bootstrap ratio values were averaged across wavelet frequencies within corresponding frequency bands. A threshold of >1 was used for the figures to emphasize the spatial distribution. Reduced δ and γ resting phase synchrony in mTBI was most pronounced between occipital areas and other brain regions, and also preferentially involved temporal lobe connections. Similar to Figs 7 and 8A was created with a threshold of <−1, and shows the distribution of pair-wise connections associated with increased phase synchrony in mTBI at α frequencies. It is interesting to note that the distribution of connections which carry discriminative information between mTBI and controls, as illustrated on the transparent brain in the MNI space (Fig 5) and its matrix version (Fig 8B), is part of the spatial pattern representing hyper-connectivity of α rhythms in mTBI (Fig 8A), which involved numerous temporal and parietal connections.
To quantitatively compare the contribution of individual frequency bands to the contrast depicted in Fig 6A, we performed a series of steps testing difference in proportions. First, we identified the wavelets closest to the central frequencies of five canonical frequency bands: 2 Hz (delta), 6 Hz (theta), 11 Hz (alpha), 23 Hz (beta), 48 Hz (lower gamma). Then, for a given z-sore threshold (1% tails), at each central wavelet frequency, we counted connections (out of the total 90*89/2 = 4005) within the positive and negative tails of the overall distribution of z-scores (see Fig 6C and 6D).
For the effects defined by the tail with negative z-scores, where we observed a peak around 8 Hz (Fig 6D), we ran two-sample proportion z-tests between the alpha and other frequencies. Specifically, we tested if the numbers of connections within the negative tail at two frequency points were statistically different. We found that the number of connections was significantly higher at alpha relative to delta (p = 0.0013), beta (p = 0.0238), and lower gamma (p<0.0001), but not theta (p = 0.757).
For the positive tail of z-scores (Fig 6D), where we identified two peaks around 2 Hz and 75 Hz, we performed a series of similar two-sample proportion z-tests. We found that the number of connections from the positive tail was statistically higher at delta relative to theta (p = 0.0035) and alpha (0.0013), whereas the number of connections at gamma was higher than theta (p<0.001), alpha (p<0.001), and beta (0.0015), but not delta (p = 0.1211).
In addition, Fig 9 provides an example of the effects shown in Fig 6C and 6D, indicating the range of absolute values of PLV for specific connections at the characteristic frequencies. Specifically, Fig 6 depicts a spatiotemporal interplay between synchronizations and de-synchronizations in the delta, gamma, and alpha frequency bands, and we chose three connections with the largest z-scores to illustrate the effects: i) between the left middle occipital gyrus (Occipital Mid L) and the left median cingulate and paracingulate gyri (Cingulum Mid L) at 2 Hz; ii) between the temporal pole of the left middle temporal gyrus (Temporal Pole Mid L) and the left gyrus rectus (Rectus L) at 8 Hz; and iii) between the left inferior temporal gyrus (Temporal Inf R) and the right calcarine fissure and surrounding cortex (Calcarine R) at 75 Hz.
Finally, we explored the effect of the length of time between injury and scan acquisition on resting MEG connectivity. We applied the behavioural PLS analysis to correlate the phase locking value with the time between brain injury and scanning. PLS analysis revealed a significant latent variable (LV) with p = 0.016, which is plotted in Fig 10A as an overall correlation (first component of LV) and a distribution of all the bootstrap ratio values (second component of LV), each associated with a unique combination of frequency and source pairing. The right (red) and left (blue) tails of the histogram in Fig 10B represent robust positive and negative correlations, respectively, between the length of time between injury and scan and phase synchronization. Frequency-specific number of connectins in these tails are shown in Fig 10C and 10D, respectively. As can be seen from Fig 10D, the effect for negative correlations between the connectivity at alpha frequencies and time of scanning is strongest at alpha frequencies. In other words, the more time that has passed since injury, the less connectivity we observed in the alpha frequency band. It is worth noting that mTBI patients, when compared to controls, were characterized by increased connectivity at alpha frequencies (Fig 6D).
The present study provides the first evidence for altered resting state neuromagnetic phase synchrony in a group of patients with mTBI, and showed that these alterations were associated with the amount of time elapsed between injury and scan acquisition. More importantly, we demonstrate that atypical MEG network connectivity, in combination with SVM learning, can accurately detect mTBI. This is an important step forward as mTBI is typically not detectible using conventional imaging. Our findings indicate that neurophysiological network imaging using MEG may provide an objective method for detection of mTBI. Moreover, we show that the distance of individual participants from the classification decision boundary was correlated with clinical symptom severity. These results demonstrate that MEG imaging of resting state functional connectivity may offer new approaches for assessing and tracking injury severity in mTBI.
Using a data-driven approach, we showed that group differences can be characterized in terms of interplay between synchronizations and desynchronizations at different frequencies. Specifically, we observed more increases in connectivity around theta/alpha frequencies in mTBI, whereas more decreases in connectivity in mTBI were detected for delta rhythms. This fits the hypothesis that processing of information in the brain requires both phase synchrony and phase scattering. Speculatively, phase synchronization can be viewed as a mechanism for long-range integration, whereas phase scattering can be a strategy to allow different local neural ensembles to share the same frequency channel by assigning specific neural signals to their own timeslots. Furthermore, we also found that the length of time elapsed between injury and scan tended to be negatively correlated with alpha synchronization and positively correlated with delta connectivity. These results may indicate that brain plasticity, a fundamental property for functional recovery from brain injury [20], may potentially be described in terms of redistribution of phase synchronyzation and phase scattering at different rhythms.
A similar pattern of the interplay between increases and decreases in functional connectivity was reported in an MEG study of TBI patients in two conditions: following an injury and after a rehabilitation treatment [21]. Noticeably, the study reported an opposite pattern, as increases in connectivity at higher frequencies such as alpha and beta, and conversely decreases in connectivity for delta and theta rhythms were associated with recovery from TBI. One of the key differences between the two studies was the time since injury. In our study, MEG data were recorded from mTBI patients, who were all within 3 months of injury (on average, one month). In [21], the mean time since injury was almost 4 months, and the rehabilitation program lasted for about 9 months.
Prior studies have indicated that resting state MEG can be used to detect mild and moderate TBI at the level of single individuals, but rather than focusing on inter-regional oscillatory synchrony, such research focused on the regional expression of excessive slow-wave activity [6, 7]. It has been proposed that axonal sheering caused by rapid deceleration and rotational forces plays a critical role in the pathology of TBI as well as its impact on functional networks and cognition [14]. Interestingly, regional expression of increased slow-wave activity has been shown to be either proximal to white matter abnormalities revealed by DTI, or in some cases, remote if micro-structural abnormalities occur in a major tract innervating that region [8]. Furthermore, this implies that excessive slow-wave activity reported in prior studies may be related to alterations in functional connectivity reported in the present investigation. Recent evidence indicates that regional concentrations of oscillatory slowing also correspond to particular symptoms expressed [7], raising the question of whether region-specific differences in functional connectivity may relate to specific patterns in post-concussive symptoms.
Research using EEG has also reported that electrophysiological interactions among brain regions are atypical in mTBI. Reduced inter-hemispheric phase synchrony among EEG scalp electrodes has been reported, and it was shown that such connectivity reductions in the beta and gamma frequency ranges were associated with alterations in white matter microstructure [10]. The network organization of resting state EEG connectivity has also been shown to be altered in mTBI [22]. An MEG investigation of patients with mild, moderate and severe TBI reported functional network disconnection in this group [23]. Using the data set employed in the present study, we previously showed that resting state correlations in the amplitude envelope of MEG activity is elevated in the delta, theta and alpha bands in mTBI, and that these alterations are associated with cognitive and affective sequelae in this group [24]. Interestingly, this pattern of alteration is different from MEG network alterations associated with PTSD (which is often a co-morbidity of mTBI) which was associated with high-frequency increases in resting phase synchrony [25]. Neural oscillations and their synchronization among brain areas are thought to play a critical role in cognition [11, 21], and resting neuromagnetic synchrony and amplitude correlations are presently thought to reflect intrinsic functional networks underpinning cognition, perception and their disturbance in clinical populations [13]. EEG research has also indicated that reduced electrophysiological interactions among brain areas may contribute to cognitive and behavioural problems associated with PCS. Reduced EEG coherence, for example, has been observed during visuospatial working memory in mTBI [26] and disrupted organization of network synchronization during episodic memory processing has also been reported [27]. Such reports of altered task dependent connectivity are congruent with reports of atypical electrophysiological and hemodynamic responses during cognitive processing following mTBI [28].
MRI studies have indicated altered functional network connectivity in mTBI [29–31], in the very low hemodynamic frequency oscillations measured by fMRI, which have been related to cognitive problems and recovery in this group [32]. During resting state, fMRI abnormalities have been reported which encompass visual, limbic motor and cognitive networks [29]. Altered default mode network connectivity [32] and regulation have been reported in mTBI. Spontaneous BOLD correlations have also been shown to be atypical in thalamocortical networks in mTBI patients, and these alterations are correlated with both clinical symptomatology and cognitive performance [30]. That altered connectivity is prominent in both neurophysiological and hemodynamic imaging studies is not surprising, as damage to white matter tracts in the form of diffuse axonal injury is common in severe brain injury [32–35]. Investigations of brain microstructure in such populations indicate altered axonal structure in both gray and white matter [36–38].
The present study capitalizes on rapidly emerging methods combining analysis of brain network connectivity with machine learning approaches supporting classification at the level of individual participants. This provides new insights into complex spatiotemporal shifts in intrinsic coupling in neurophysiological brain networks following mTBI. More importantly, the present work provides potentially clinically translatable methods that will permit the detection of mTBI in single individuals where conventional radiological imaging approaches are inconclusive. The finding that classification confidence is associated with self-reported symptom severity indicates that these methods may provide quantitative and objective measurements of brain changes underlying PCS. This could have significant impact on current clinical practice. An objective, quantitative method for diagnosing brain dysfunction after mTBI would allow identification of patients at risk for a subsequent injury, be invaluable for developing parameters around return to play / work / duty, and assist in developing guidelines for providing care, monitoring treatment efficacy and tracking recovery.
MEG data were recorded from 20 men with mTBI (21–44 years of age, mean = 31±7 years, 2 left-handed), all within three months of injury (days since injury = 32 ± 18 days). Participants with mTBI were recruited through the Emergency Department of Sunnybrook Health Science Centre in Toronto. The inclusion criteria were: concussion symptoms while in emergency; Glasgow Coma Scale ≥13 (within 24 hours of injury); if loss of consciousness occurred, then less than 30min; if post-traumatic amnesia occurred, then less than 24 hours; causes of head injury were clear (e.g., sustaining a force to the head); no skull fracture; no abnormalities on Computer Tomography (CT) scan and no previous incidences of concussion. Participants in the mTBI group completed the Sports Concussion Assessment Tool 2 (SCAT2) Symptom Checklist and Symptom Severity Score; were able to tolerate the enclosed space of the MRI; were English speaking and able to complete tasks during MEG and MR scans and able to give informed consent. The mean Severity score of mTBI patients was 20 ± 19, whereas the Symptom score was 9 ± 6. The MEG and MRI scans were obtained, on average, on 32nd day since injury: 32 ± 18 days. Potential participants were screened prior to recruitment and none of the mTBI participants reported any post-traumatic stress disorder, neurological or psychiatric symptoms, and psychoactive medication use. All of the MRI scans were read by a neuroradiologist, and there were no abnormalities noted.
An age- and sex-matched control group without any history of TBI included 21 participants (20–39 years of age, mean = 27±5 years, 1 left-handed). The control group had no history of TBI (mild, moderate or severe), no neurological or psychiatric disorders, and were not on psychoactive medications. None of the participants had MRI contraindications such as metallic implants or metal dental work. Data acquisition was performed with the informed consent of each individual and with the approval of the Research Ethics Board at the Hospital for Sick Children (SickKids).
MEG data were acquired in a magnetically shielded room at SickKids using a whole-head CTF system (MISL Ltd., Coquitlam, BC, Canada) with 151 axial gradiometers as well as reference sensors for gradient correction. For each subject, 5 minutes of MEG data were continuously recorded at 600Hz using third-order spatial gradient noise cancellation. 60Hz and 120Hz notch filters were applied to MEG recordings. Data were also band-pass filtered between 1Hz and 150 Hz with a fourth-order Butterworth digital filter applied first in a forward, and then in a reverse direction so as to produce zero phase distortion. Head position during testing was monitored via three localization coils, positioned at the nasion, and the left and right pre-auricular points.
Anatomical MRI was performed on the same day at SickKids on a 3T MR scanner (MAGNETOM Tim Trio, Siemens AG, Erlangen, Germany) with a 12-channel head coil. The three fiducial coils used in the MEG were replaced with radio-opaque markers for all participants. These markers can be seen on their T1-weighted images for co-registration of the MEG source locations to the MRI images. Anatomical images were collected by whole-brain T1-weighted MRI scans (3D SAG MPRAGE: GRAPPA = 2, TR/TE/TI/FA = 2300/2.96/900/9, FOV/Res = 192x240x256, 1mm isotropic voxels).
Individual MRI scans were normalized into Montreal Neurological Institute (MNI) space based on the ICBM 2009c Nonlinear Symmetric 1 × 1 × 1mm template [39]. We applied a nonlinear diffeomorphic registration, as implemented in the ANTS toolbox [40,41]. This transformation to MNI space was additionally used to warp a manually segmented inner skull surface from the MNI ICBM template to subject space. Using this inner skull surface, a multi-sphere head model was fit for each subject [42].
MEG data were co-registered to each participant’s individual anatomical MRI to constrain neuromagnetic sources to subject-specific head shape and structural anatomy. To reconstruct neuromagnetic source activity, we first selected 90 seed locations in MNI space, which represented all cortical and subcortical brain regions in the Automated Anatomical Labeling (AAL) atlas [43]. Regions specified by the AAL atlas and located in the cerebellum were excluded from the further analysis. For visualization purposes, the regions were re-ordered according to which lobe each region belongs to. The new order of the regions is given in Table 1 (the left region goes first, followed by the right one). Specifically, for each region from the AAL parcellation, the seed location was defined as a voxel within the region, which was closest, in the mean-square sense, to the means of x-, y-, and z-coordinates, averaged across all the voxels in this brain region [44]. Source estimation was performed at these 90 locations, using an adaptive spatial filter (vector beamformer) [45]. For each subject, 27 non-overlapping epochs of 10 seconds duration were extracted such that head motion within each epoch did not exceed 3mm in any direction for any of three head location coils.
The time-frequency representation of the original time series for each reconstructed source was derived from the wavelet decomposition, using a time-frequency toolbox [46]. Thirty frequency points equally spaced on a logarithmic scale were selected to cover the range between 1Hz and 75Hz. The analysis of phase synchronization between the neuromagnetic sources was performed on spectrally decomposed data. We computed phase-locking values [47], which are known in the literature under different names such as mean phase coherence [48] or phase synchronization index [49]. Phase synchronization emerged from studying coupled nonlinear systems [50], and is based on an idea that the existence of correlations between the phases of coupled systems does not imply correlation between their amplitudes.
A common method for obtaining phase dynamics for analyzing phase synchronization between brain signals is based on wavelet transformation [51]. A signal can be decomposed into a set of brief oscillatory patterns called wavelets. Specifically, wavelet coefficients Wx(τ, f) at time τ and frequency f are obtained by convolving a given signal x(t) with a zero-mean special function or wavelet ψτ, f(t):
Wx(τ,f)=∫−∞+∞x(t)ψτ,f(t)dt
(1)
where ψτ, f(t) is a short segment of a oscillatory signal (wavelet) obtained from an elementary function called the mother wavelet by dilutions and translations. Often, a specific form of the mother wavelet is used, known as the Richer wavelet or Mexican hat function, which is defined as the negative normalized second derivative of a Gaussian function. To decompose a signal at a specific frequency f and time τ, the mother wavelet is compressed or dilated, and then translated such that ψτ, f(t) is centered at time τ. To maintain a consistent frequency resolution, the bandwidth of the envelope is set to be inversely proportional to f, such that each wavelet contains the same number of cycles.
In general, the coefficients Wx(τ, f) are complex numbers. The transformation Eq (1) thus defines both the amplitude of signal x(t) and the phase over a range of times τ and frequencies f. The instantaneous phase ϕx(τ, f) is the angular component (phase angle) of Wx(τ, f).
The relative phase Δϕx(τ, f) of two signals, x(t) and y(t), is defined as a time series of the difference between the instantaneous phase of each signal, namely
Δϕx,y(τ,f)=ϕx(τ,f)−ϕy(τ,f)
(2)
which can be computed from the wavelet coefficients at time τ and frequency f from
eiΔϕx,y(τ,f)=Wx(τ,f)Wy*(τ,f)|Wx(τ,f)||Wy(τ,f)|
(3)
where Wy*(τ,f) is the complex conjugate of Wy(τ, f).
The phase differences can be projected as a series of two-dimensional vectors onto the unit circle, one per time point τ = τ1, …, τN. The phase-locking value PLVx, y(f), which reflects the amount of phase-synchrony between two signals across time, is computed as the length of the resultant (mean) vector:
PLVx,y(f)=〈eiΔϕx,y(τ,f)〉τ=|1N∑k=1NeiΔϕx,y(τk,f)|
(4)
By construction, PLVx, y(f) is limited between 0 and 1. When the relative phase distribution is concentrated around the mean, the PLV is close to one, whereas phase scattering will result in a random distribution of phases and PLV close to zero.
For each epoch, for all pairs of 90 regions of interest (ROIs), frequency-specific phase differences were computed as functions of time. The phase-locking value, PLVx, y(f), was calculated as relative stability of the phase differences between two signals at a given frequency, subsequently averaging across epochs. Thus, 30 90-by-90 matrices were produced for each subject, representing functional connectivity in terms of phase-locking between 90 neuromagnetic sources at 30 frequency points.
In the present study, Support Vector Machine (SVM) learning was used to predict the clinical status Y of a subject (mTBI or control) from a set of features X obtained from the subject’s MEG data [52]. These features are represented by frequency-specific phase-locking values (PLV) between the neuromagnetic activity reconstructed for 90 regions of interest (ROIs). Each of the samples (subjects) i = 1, …, m, where m = 41, can be treated as a point xi in a n-dimensional feature space, where n is the total number of features—unique combinations of all the connections and frequencies of interest.
A learning machine can be seen as a function F, which determines a learning model:
F:X→Y
(5)
The function F transforms vectors xi from the feature domain X to the set Y of possible outcome values. When Y is a set of only two symbols (mTBI and control), the learning problem Eq (5) is called a binary classification, and Y is called the set of class labels.
Learning machines encompass many computational approaches. For classification problems, they can produce models with various types of decision borders. In this study, we applied a linear version of a SVM to determine a linear border between the classes [52]. Depending on which side of the border the sample xi is located, it can be assigned to one of two classes:Y = {1,−1} coding mTBI and control groups, respectively. Samples used to define the border are called training data. The clinical status of new cases (test data) can be predicted based on their locations with respect to the decision border. If we know the true status of the test data, we can estimate the accuracy of that prediction. In practise, the entire data with known labels are typically divided into two sets: training data to learn the function (5) and test data to validate it.
Mathematically, learning the model (5) with a linear SVM is equivalent to finding the optimal hyperplane ωTx + b = 0 in the feature space, where ω is an n-dimensional weight vector, and–b defines the threshold. Optimal here means separating the two classes Y = {1,−1} with maximal margin. Mathematically, training the model (5) is reduced to an optimization problem, maximizing the minimum distance between vectors xi and the hyperplane:
maxw,b min{ ‖x−xi‖ such that wTx+b=0, i=1,…,m }
(6)
If a vector x satisfies F(x) = ωTx + b > 0, then the model (5) will assign the label 1 (class mTBI) to it, otherwise the label −1 (class Controls) is assigned. The distance from the decision boundary F(x) = ωTx + b can serve as a measure of confidence in the classification.
Leave-one-out cross-validation was used in this study to estimate prediction accuracies of the classification. During this procedure, all samples xi, i = 1, …, j − 1j + 1, …, m except one xj were designated as the training data to determine the optimal model (5) for separating the classes, whereas ability of this model (5) to correctly predict the outcome was tested with the remaining sample xi. This procedure was repeated m times such that each subject served as the test sample only once. The prediction accuracies such as sensitivity and specificity were then computed by comparing the predicted and true statuses of m subjects.
Until now, we assume that all features, i.e. all frequencies and connections, were used for classification. However, predictive accuracy could be improved by selecting the most relevant and informative features. In general, feature construction and selection is a critical step in classification. In practise, it is essentially heuristic. Fig 11 schematically illustrates one round of cross-validation used to learn a model from the training data, and then predict the group status of the test data. Feature selection was based on supervised learning, wherein the features were the phase synchrony estimates with three feature selection schemes: i) individual wavelet frequencies, ii) canonical frequency bands, and iii) best representative features within the α band. As a first-pass analysis, contribution of individual wavelets to the classification was estimated. Then, five frequency bands, namely δ (1-4Hz), θ (4-8Hz), α (8-14Hz), β (14-28Hz), and lower γ (28-75Hz) were a priori selected, and all the wavelets representing more fine grained frequency bins were assigned to one of these canonical bandwidths. Further, for the frequency band that carried the most discriminative information (namely, α), the features were ranked in a univariate manner. Specifically, for each feature (PLV for a given frequency and connection), overlapping probability distribution functions for two classes were compared, and the area under the resulting receiver operating characteristic (ROC) was computed [53]. The area under the ROC (AUR) provides an estimate of how valuable a feature can be for separating the two classes.
Accordingly, the feature selection can be summarized as follows. First, the features that were computed to quantify the brain state were separated into 30 sub-sets, each associated with a wavelet frequency. Classification analysis with leave-one-out cross-validation was applied separately for each subset, using linear SVM [54] as implemented in a Matlab statistics toolbox (MATLAB and Statistics Toolbox Release 2012a, The MathWorks, Inc., Natick, Massachusetts, USA). Next, total accuracy as well as specificity and sensitivity were computed. The features were then regrouped in an alternate way, and the classification process was repeated. Specifically, the features were separated into 5 subsets, each associated with a frequency band (δ, θ, α, β, lower γ), containing phase-locking values calculated for a set of wavelets within a specific frequency range. Further, the features representing phase synchrony in the α band were ranked, computing the area under the ROC curve (AUR). In the next step, leave-one-out cross-validation was applied for every number k = 1, …, 100 of features with the highest AUR, and classification accuracy was estimated as a function of the number of selected features. Significance of accuracy values was tested with respect to the distribution created by shuffling 500 times the labels (mTBI and Control) among the subjects.
Partial Least Squares (PLS) analysis was used to further explore possible group differences in connectivity (PLV) between the neuromagnetic sources across groups, as well as how these differences are expressed across frequencies and specific connections [55]. In contrast to the prediction analysis with linear SVM, wherein the learning model F was estimated for a subset of subjects (training data), PLS analysis was performed on the entire data in a single analysis.
PLS is a multivariate technique, which decomposes the covariance between the neurophysiological data and a discrete variable coding a contrast (between groups, for example) or a continuous variable (such as the time since injury) into mutually orthogonal factors (latent variables), similar to the principal component analysis [55]. In practice, PLS analysis can identify data-driven contrasts between groups or test specific a priori contrasts, and finds optimal relations among these contrasts and features (combinations of individual connections and frequencies in our case). Significance of the contrast can be tested with permutation tests, whereas the robustness of the contribution of specific connections and frequencies to the identified contrast can be tested with bootstrap procedures. Here we give a brief description of the technique [55–58], which was previously applied in a number of EEG and MEG studies to characterize changes in the brain signals [59–61].
PLS operates on the whole data matrix at once. Typically the rows of the data matrix correspond to participants within groups, whereas the columns correspond to voxels in functional MRI, electrodes in EEG, sensors or sources in MEG. These features (voxels, electrodes, sensors) can be called the elements. In our case, the elements were represented by all the possible combinations of a pair of neuromagnetic sources and a frequency point. Specifically, to prepare for the PLS analysis, the data matrices were organized in the form of subjects within groups by elements, each associated with a connection and a frequency point (30 × 90 × 89/2 = 120, 150 elements in total). Thus, the neuroimaging data were organized as a matrix: subjects within groups by all the possible combinations of connections and frequencies. Then, the covariances were computed between the data matrix and the vectors representing either the contrast between groups or the length of time elapsed between injury and scan.
Next, singular value decomposition (SVD) was used to project the covariances to a set of orthogonal latent variables (LVs), mathematically described as a products of three vectors: the left-singular vectors, the non-zero singular values, and the right-singular vectors. Each latent variable (LV) thus had three components: (a) a singular value, representing how much variance can be explained by this LV, similar to principal component analysis; (b) weights within the left singular vector, representing an underlying contrast between groups or an overall correlation between imaging and clinical data; (c) weights within the right singular vector (element loadings), representing the robustness of contribution of all the elements to the group contrast or overall correlations.
The overall significance of each LV and the importance of the individual elements within a specific LV was assessed using resampling procedures. First, we randomly reassigned subjects between groups, performing a permutation test. This global permutation test assessed the overall significance of a given LV, measuring how it is different from random noise. Specifically, we computed a measure of significance as the number of times the singular values from permuted data were higher than the observed singular value (500 permutations). In the second step, we tested the element loadings for stability across subjects by bootstrap resampling of subjects within groups (500 bootstrap samples). A measure of stability (bootstrap ratio value) was calculated as the ratio of the original element loading to the standard error of the distribution of the element loadings generated from bootstrapping. This is approximately equivalent to a z-score: a bootstrap ratio value of 3 or -3 corresponds to 95%-confidence under the assumption of a Gaussian distribution. Elements (all the combinations of connections and frequencies) with positive bootstrap ratio values directly support the contrast or overall correlation associated with the left-singular vector of a given LV. Negative bootstrap ratio values also indicate the robustness of the effects, but in the reverse direction. In other words, to correctly interpret the output, the bootstrap ratio values (or z-scores) need to be reported with respect to the contrast or overall correlation in order to correctly understand the direction of the loadings.
We distinguish two types of PLS analysis: so called “contrast” and “behavioural” PLS [55,56]. In the contrast PLS, there are groups of subjects (in our case, mTBI and healthy controls), and the PLV data are projected to an a priori defined contrast. In this case, the weights within the left singular vector are equivalent to the group contrast. The “behavioural” PLS, which is typically based only on one group of subjects, explores the covariance between the brain data and some continuous, subject-specific variables, such as time of scanning since injury. In this case, the weights within the left singular vector represent the overall correlations (one correlation per variable) between the PLV and time-of-scanning matrices. Both in the contrast and behavioural PLS, the right singular vector reflects the contribution of the individual elements to the tested effects.
One assumption of our study is that the volume conduction effects do not represent a significant confounding factor. It is not entirely true that MEG is not sensitive to effects of volume conduction. It has been shown, however, that secondary currents resulting from volume conduction do not contribute to the radial component of the magnetic field under the assumption of a dipolar source in a spherical homogeneous conductor [62]. For our study, we used a first order axial gradiometer system, which is mainly sensitive to the radial component of the magnetic field (that is, the field of a source dipole with tangential orientation). In this setting, estimating PLV, which may capture the couplings with a phase shift close to zero, seems reasonable. Using a more conservative measure such as weighted phase lag index (PLI) would further minimize the volume conduction effects, but it would also remove some physiologically meaningful couplings, which may reduce both the sensitivity and specificity.
Another point is related to the segmentation of MEG recording. The original data were epoched into 10s segments. We choose 10s as a compromise between our intent to estimate phase synchrony at the lowest frequencies and to increase the robustness of the estimation by averaging across different epochs. Specifically, we believe that 10s is, on the one hand, long enough to robustly estimate the phase locking effects at the frequencies close to 1Hz, and on the other hand, is short enough to allow us to extract relatively large number of segments. The latter helps to increase the robustness of the results by averaging the phase synchrony across segments. Furthermore, the segments should be relatively short to not introduce large movement artefacts. Please note that the segments were extracted from 5 minutes of recordings using a rather conservative threshold of less than 3mm of movement.
|
10.1371/journal.pntd.0005717 | The clinical burden of human cystic echinococcosis in Palestine, 2010-2015 | Cystic echinococcosis (CE) is classified by the WHO as a neglected disease inflicting economic losses on the health systems of many countries worldwide. The aim of this case-series study was to investigate the burden of human CE in Palestine during the period between 2010 and 2015.
Records of surgically confirmed CE patients from 13 public and private hospitals in the West Bank and Gaza Strip were reviewed. Patients’ cysts were collected from surgical wards and formalin-fixed paraffin-embedded (FFPE) blocks were collected from histopathology departments. Molecular identification of CE species /genotypes was conducted by targeting a repeat DNA sequence (EgG1 Hae III) within Echinococcus nuclear genome and a fragment within the mitochondrial cytochrome c oxidase subunit 1, (CO1). Confirmation of CE species/genotypes was carried out using sequencing followed by BLAST analysis and the construction of maximum likelihood consensus dendrogram. CE cases were map-spotted and statistically significant foci identified by spatial analysis. A total of 353 CE patients were identified in 108 localities from the West Bank and Gaza Strip. The average surgical incidence in the West Bank was 2.1 per 100,000. Spot-mapping and purely spatial analysis showed 13 out of 16 Palestinian districts had cases of CE, of which 9 were in the West Bank and 4 in Gaza Strip. Al-Khalil and Bethlehem were statistically significant foci of CE in Palestine with a six-year average incidence of 4.2 and 3.7 per 100,000, respectively.
To the best of our knowledge, this is the first confirmation of human CE causative agent in Palestine. This study revealed that E. granulosus sensu stricto (s.s.) was the predominating species responsible for CE in humans with 11 samples identified as G1 genotype and 2 as G3 genotype. This study emphasizes the need for a stringent surveillance system and risk assessment studies in the rural areas of high incidence as a prerequisite for control measures.
| Cystic echinococcosis (CE) is a neglected disease caused by a parasite called Echinococcus granulosus. The dog as a definitive host plays a major role in transmitting the infection to human. The study aimed to investigate the clinical burden of human CE in Palestine during the period between 2010 and 2015. Thirteen hospitals in the West Bank and Gaza Strip were targeted. CE patients’ records were reviewed. Patients’ samples were collected including cysts following surgery and formalin-fixed paraffin-embedded (FFPE) blocks from histopathology departments. Molecular identification of CE species /genotypes was conducted by targeting nuclear and mitochondrial DNA and confirmed by identifying the DNA sequence and comparing it with those in the Genebank. CE cases were spot-mapped and statistically significant foci identified. A total of 353 CE patients were identified in 108 localities in Palestine from the West Bank and Gaza Strip. The average surgical incidence in the West Bank was 2.1 per 100,000 with Al-Khalil district reporting the highest incidence in Palestine. This study revealed that E. granulosus sensu stricto (s.s.), G1 genotype was the main species responsible for CE in humans. There is need for a surveillance system and risk assessment studies in the rural areas as a prerequisite for control measures.
| Cystic Echinococcosis (CE) is a zoonotic parasitic infection caused by the metacestode larval stage of species within the genus Echinococcus. CE as the most frequently encountered disease is caused by Echinococcus granulosus sensu lato (s.l.) with the dog as the main definitive host and ungulates, mainly sheep as the intermediate host. Humans are accidental hosts infected following the ingestion of eggs shed in dog faeces. Although the classification and taxonomy of the genus Echinococcus is still controversial, however recent classification recognizes nine species, five of which belong to E. granulosus sensu lato (s.l.) namely E. granulosus sensu stricto (s.s.) (genotypes G1-G3), E. felidis (lion strain), E. equinus (genotype G4), E. ortleppi (genotype G5), and E. canadensis (genotypes G6/7-G8 and G10) [1, 2]. CE is considered by WHO as an important food-borne parasitic disease with estimated 1–3 million Disability adjusted life years (DALYs) for cystic echinococcosis (accounting for underreporting), [3]. CE is common worldwide with hyperendemic areas exceeding 50 cases per 100,000 in some countries like Argentina, Peru, East Africa, Central Asia, and China and causing annual loss of approximately $3 billion, yet still listed as one of the 18 neglected diseases in the world [4–6]. In the Mediterranean region where CE is endemic, Tunisia and Morocco reported the highest incidence of 12.5 and 5.1 surgical cases per 100,000, respectively [7–12]. In Turkey, an ultrasonography-based survey among children revealed a prevalence of 0.2% with E. granulsus sensu stricto (s.s.) (G1/G3) as the main cause [13, 14]. Historical records in Palestine put the disease incidence at 1 and 5/100,000 during 1922–1935 and 1959, respectively [15, 16]. More recently, the incidence of human CE in 2015 was officially reported to be 1.6 per 100,000 [17]. This rate depends exclusively on surgical incidence reported by government hospitals following surgery. Pilot studies showed that the incidence rate in dogs as definitive hosts using copro-PCR was 18% [18]. The most common genotype in the definitive (dog) and intermediate hosts (sheep) from Palestine is E. granulosus G1 genotype [18, 19]. In adjacent areas like the city of Rahat and Bir-Al Saba’, CE was predominant among Bedouin community compared to Jewish residents with an incidence of 2.7 and 0.4 per 100,000, respectively [20].
In this study we aimed to investigate the clinical burden of human cystic echinococcosis in Palestine (West Bank and Gaza Strip) using retrospective hospital records during the period between 2010 and 2015 supported by molecular methods through amplification of DNA from human cysts and formalin-fixed paraffin embedded (FFPE) blocks. In this study surgical incidence was defined as the frequency of operated CE cases per 100,000 inhabitants per year.
This study was ethically approved by the Ministry of Health (MoH) in Palestine (162/2044/2015). All patients’ data were securely archived and anonymized.
A case series observational study on surgically-confirmed human CE cases in Palestine (The West Bank and Gaza Strip) was carried out during a six year period between 2010 and 2015.
CE patients’ records were retrieved from public (government) and private hospitals in the West Bank and Gaza Strip. All reviewed CE cases had been diagnosed using ultrasonography or computed tomography (CT) and histopathologically confirmed following surgery. Thirteen hospitals were included in the study namely Jenin Government Hospital, Al-Amal Hospital (Jenin), Zakat Hospital (Jenin), Rafidia Government Hospital in (Nablus), Al-Khalil Government Hospital, Al-Ahli Hospital in Al-Khalil, Al-Mezan Hospital in Al-Khalil, Beit-Jala Government Hospital in Bethlehem, Jericho (Ariha) Government Hospital, Ramallah Government Hospital, Al-Makassed Hospital in East Jerusalem (Al-Quds), Al-Shifa Government hospital in Gaza, and Gaza European Hospital in Rafah. This included all government hospitals and major private hospitals that serve a population of ca. 4.6 million Palestinians in the West Bank and Gaza Strip. Retrieved data included patients’ demography such as names, age, address, date of birth, sex, site of infection, and diagnosis based on histopathology reports. Missing demographic and clinical data were obtained by conducting phone interviews with patients facilitated by the Ministry of Interior in Palestine. Human CE material removed from surgically-confirmed patients was collected from the histopathology departments in the form of FFPE pathology blocks. In addition, CE cysts surgically-removed during the duration of this study were provided by the relevant wards and stored at -20°C.
Kulldorff’s SaTScan programme v9.4.3 was used to assign spatial and space-time distribution of cases in Palestine based on number of cases per locality, year of diagnosis, population size of locality at time of diagnosis, and the exact latitude-longitude coordinates of each location. Data were analyzed based on a discrete Poisson model with the level of statistical significance considered at P-value ≤ 0.05 [24]. Spot mapping of cases and frequencies were analyzed using Epi Info statistical package (CDC free-software). The level of statistical significance was considered at P-value ≤ 0.05. Evolutionary analysis, genetic relationship, and multiple alignments were conducted in MEGA 7 [25].
During 2010–2015 a total of 353 CE patients were identified in 13 hospitals in the West Bank and Gaza Strip, the vast majority of whom (n = 282) were diagnosed during this study period (Fig 1). Cystic echinococcosis was reported in 108 localities in the West Bank (94%, 319/338) and Gaza Strip (6%, 19/338) (Fig 2a). The average surgical incidence for the disease in the West Bank, Gaza Strip, and Palestine as a whole was approximately 2.1, 0.13, and 1.1 per 100,000, respectively. Demographic habitats of CE cases were shown to include villagers (78%), city-dwellers (20%), refugee camp residents (2%) and Bedouins living in encampments (0.3%). The female-to-male ratio was 1.13 (187:166) which was not statistically significant (Chi square = 1.2, P = 0.26). The age of CE patients ranged from 2 to 86 years old, with the two age groups 10–19 and 20–29 having significantly high number of reported hydatid cyst cases than expected (Chi square = 145.5, P = 0.0001). After this, CE cases decreased gradually until reaching 4 in the age group 80–89 years (Table 1).
Human cases of CE started to appear systematically on the Palestinian Ministry of Health (MoH) annual report from 2006 with a peak in 2012. The MoH annual report also reported Gaza Strip as CE-free area except for 1 case in 2011, while this study revealed 19 cases in the same period. Of the 353 CE cases, the residential premises of 338 were known and were therefore used for spot-mapping of CE cases. CE was found to be present in thirteen out of 16 Palestinian districts, 9 regions of which were in the West Bank and 4 in the Gaza Strip (Fig 2a). District-wise, Al-Khalil and Bethlehem were the main foci of CE in Palestine with a six-year average incidence of 4.2 and 3.7 per 100,000, respectively. Purely spatial analysis identified villages of Yatta in Al-Khalil District and Z’atara, and Ash-shawawra in Bethlehem District to be statistically significant foci for the disease and the most prevalent urban areas in Palestine with an annual CE incidence of 9.6, 13.9, and 23.3 per 100,000, respectively (Fig 2b). On the other hand, in the space-time analysis two foci were spotted in certain years over the study period, one in Al-Khalil and another in Bethlehem (Fig 2c).
Of the 261 CE patients, information regarding localization of CE infection was known for 271 cysts. Liver cysts were exclusively found in 158 (58%) of CE cases whereas pulmonary infection was recorded in 27% (74/271) of cases. Approximately 3.4% (9/261) of CE cases had multiple site infection with two or more organs being involved (Table 2). Of the 299 cases for which symptoms were known, the most frequent symptoms were abdominal pain (42%) and dyspnea (13%) (Table 3).
In this study, 14 cysts were collected from patients immediately following surgery and 68 additional FFPE CE samples were collected from histopathology departments. Of these, 86% (12/14) and 82% (56/68) respectively were positive for Echinococcus species as identified through the amplification of the diagnostic tandem repeat product (269bp). In addition, a partial fragment of cox 1 gene was successfully sequenced for 11 of the 14 cyst samples, which were identified using BLAST as E. granulosus s.s. (9 samples as G1 and 2 as G3 genotype). Nucleotide sequences generated here were deposited in the GenBank under the accession numbers depicted in Fig 3 showing the genetic clustering. The dendrogram showed that all 11 isolates from Palestine clustered in one group. Seven of the 353 cases were classified at hospital level as originated by Echinococcus multilocularis based solely on clinical picture without molecular confirmation. These 7 samples were not available in this study, thus was not possible to confirm or rule out the presence of this parasite.
In Palestine, despite the fact that CE is a notifiable disease, information on the magnitude of infection is usually generated by surgical wards from public government hospitals. Using this approach, an average surgical incidence of CE in the West Bank during 2010–2015 was reported to be 1.6 per 100,000 compared to the 2.1 per 100,000 generated in this study. An earlier study in which researchers scanned hospitals in the West Bank between 1990 and 1997 reported a relatively high average incidence of 3.1 per 100,000 [26]. In addition, the number of CE cases with known year of surgery included in this study was greater than that reported by the MOH in five of the six-year study period (Fig 1). A potential explanation for this discrepancy may be related to the inaccuracy in the surveillance system since the reporting of CE in official records only began in 2006, as CE was not reportable disease by law before then (Fig 1) [17]. At the same time, the Israeli health authorities reported 38 cases between 1991 and 1995 [27]. In Jordan, the surgical incidence rate between 1985 and 1993 was 2.9 per 100,000. [26, 28, 29]. Surgical incidence studies appear to underestimate the actual burden of CE. CE data collated from surgical procedures reflects the tip of the iceberg as the majority of CE cases are asymptomatic, and thus do not come to the attention of clinicians. Surgery is normally performed on symptomatic patients with complicated CE or on individuals who are inadvertently diagnosed as having the disease [30]. Furthermore, it is normal practice for patients to be referred for CE surgery to hospitals outside Palestine, such as Jordan. Those cases were not included in Palestinian annual health report. Other explanations might be due to incorrect diagnosis or underreporting by physicians or hospitals.
The distribution of CE cases by sex with slight predominance in females, but not statistically significant, is in agreement with other studies [26, 28]. Cystic echinococcosis surgical incidence was higher in the younger age groups (10–29 years). This is in congruence with other studies using surgical incidence that reported the highest number of CE cases in young patients (10–13 years) [20, 26, 28]. In rural and Bedouin areas, herds of sheep are commonly accompanied by several dogs which are in strict contact with children, increasing their exposure to this parasite. Furthermore, it’s a habit by villagers, young and adults, to eat leafy plants such as mallow (Malva parviflora) or those growing in the yard such as lattice, spinach and onions, thus increasing the possibility of contracting this infection.
CE has a long latency period and subsequently can be detected years after infection, which is often at an older age [31]. However, the appearance of disease among adolescents (10–20 years) Palestinian patients would suggest an infection sustained at a very early stage in life. Most of CE cases were reported in rural areas, which is in agreement with studies worldwide [29, 32, 33] and is one of the potential risk factors for acquiring CE identified in a recent systematic review for acquiring CE [34]. In contrast, it should be stressed that surgical incidence may introduce bias in the engagement of patients in this study since young age groups, especially children, could be more likely to seek medical attention compared to older age groups.
In Palestine, a study that identified high incidence of E. granulosus infection among dogs was recently published [18]. The proximity of humans to free-roaming dogs which have access to infected offal is a known potential risk factor for CE infection [34].
This study identified seventeen sites of CE infection; however 85% had a predilection for the liver and the lungs (Table 2). The liver is widely known to be the most infected organ for cystic echinococcosis in Palestine and elsewhere as a result of the portal blood flow [26, 28, 35–37]. Double-site infection was rare with mostly the liver involved, and a multiple-site infection was reported only in one case. Abdominal pain as a result of bile duct compression and dyspnea resulting from irritated lung membranes were the main symptoms reported by patients, which in turn reflects the predominance of liver and lung infections [37].
Of the 14 human CE cysts, 86% were positive for the Hae III E. granulosus repetitive gene sequence and two were negative with one having been preserved for over a month in 10% formalin, a potent DNA degrader. Similarly, the infection rate of FFPE samples was 85%. Nine (81.8%) and 2 (18.2%) of the Palestinian patients’ included in this study were molecularly confirmed as having been infected with E. granulosus G1 and G3 genotype respectively. To the best of our knowledge, this is the first molecular identification of the human CE causative agent in Palestine. E. granulosus sensu stricto (s.s.) had been previously confirmed from sheep [14] and dogs [13] and the findings of the current study point to the free circulation of this species within Palestine and demonstrate the active involvement of these hosts in the perpetuation and transmission of this parasite [18, 26]. Results obtained through the construction of the maximum likelihood tree showed E. granulosus sensu stricto (s.s.) (G1/G3) nucleotide sequences generated in this study to group within a single cluster along with E. granulosus s.s. (G1/G3) from the Genbank demonstrating genetic uniformity. This is consistent with other studies from Italy, Jordan, Iran, India, China and Peru that investigated nucleotide sequence variation of DNA extracted from CE material derived from humans, livestock and dogs confirming the low nucleotide-diverse nature of E. granulosus sensu stricto (s.s.) worldwide [18, 19, 38–41].
CE is widely spread in Palestine with the majority (94%) of cases in the West Bank and only 6% in Gaza. The low incidence in Gaza Strip may be a reflection of the total share of livestock in Palestine which is 20.2% for Gaza Strip and 79.8% for the West Bank [42]. Spatial and space-time distribution showed Al-Khalil district to be the main focus of the disease in Palestine. Al-Khalil district is the most populous district and holds the largest share of livestock animals in Palestine (21.1%) including sheep (25.2%) and goats (21%) [42]. Rural areas within Al-Khalil such as Yatta, Idhna and Dura villages appear to be the hot spots for CE; as this has been the case for the last 3 decades with an incidence of 16.8 per 100,000 in Yatta village between 1990 and 1997 (Fig 2) [26]. The seven cases of E. multilocularis identified at hospital level based only on clinical picture are the first reports in Palestine. However, in the absence of molecular confirmation, multiple cysts of E. granulosus s.s., such as CE2 according to WHO-IWGE (Informal Working Group on Echinococcosis) classification, may be erroneously identified as those of E. multilocularis [43].
In conclusion, this study has shown that E. granulosus sensu stricto (s.s.), is the most prevalent species causing human CE in the West Bank and Gaza Strip and identified Al-Khalil district as the main focus for CE infection. Risk assessment studies in the rural areas such as Yatta are a prerequisite for control measures. For this reason, we would encourage the Ministry of Health, Ministry of Agriculture, and local health authorities to implement control measures aiming at decreasing the burden of CE in humans in Palestine and interpolating a stringent surveillance system. In addition, sensitizing the Palestinian citizens by community health awareness campaigns and upgrading the level of health service by training the medical team on CE epidemiology and detection are prerequisites for effective surveillance and control of this neglected disease.
|
10.1371/journal.ppat.1003495 | Mutated and Bacteriophage T4 Nanoparticle Arrayed F1-V Immunogens from Yersinia pestis as Next Generation Plague Vaccines | Pneumonic plague is a highly virulent infectious disease with 100% mortality rate, and its causative organism Yersinia pestis poses a serious threat for deliberate use as a bioterror agent. Currently, there is no FDA approved vaccine against plague. The polymeric bacterial capsular protein F1, a key component of the currently tested bivalent subunit vaccine consisting, in addition, of low calcium response V antigen, has high propensity to aggregate, thus affecting its purification and vaccine efficacy. We used two basic approaches, structure-based immunogen design and phage T4 nanoparticle delivery, to construct new plague vaccines that provided complete protection against pneumonic plague. The NH2-terminal β-strand of F1 was transplanted to the COOH-terminus and the sequence flanking the β-strand was duplicated to eliminate polymerization but to retain the T cell epitopes. The mutated F1 was fused to the V antigen, a key virulence factor that forms the tip of the type three secretion system (T3SS). The F1mut-V protein showed a dramatic switch in solubility, producing a completely soluble monomer. The F1mut-V was then arrayed on phage T4 nanoparticle via the small outer capsid protein, Soc. The F1mut-V monomer was robustly immunogenic and the T4-decorated F1mut-V without any adjuvant induced balanced TH1 and TH2 responses in mice. Inclusion of an oligomerization-deficient YscF, another component of the T3SS, showed a slight enhancement in the potency of F1-V vaccine, while deletion of the putative immunomodulatory sequence of the V antigen did not improve the vaccine efficacy. Both the soluble (purified F1mut-V mixed with alhydrogel) and T4 decorated F1mut-V (no adjuvant) provided 100% protection to mice and rats against pneumonic plague evoked by high doses of Y. pestis CO92. These novel platforms might lead to efficacious and easily manufacturable next generation plague vaccines.
| Plague caused by Yersinia pestis is a deadly disease that wiped out one-third of Europe's population in the 14th century. The organism is listed by the CDC as Tier-1 biothreat agent, and currently, there is no FDA-approved vaccine against this pathogen. Stockpiling of an efficacious plague vaccine that could protect people against a potential bioterror attack has been a national priority. The current vaccines based on the capsular antigen (F1) and the low calcium response V antigen, are promising against both bubonic and pneumonic plague. However, the polymeric nature of F1 with its propensity to aggregate affects vaccine efficacy and generates varied immune responses in humans. We have addressed a series of concerns and generated mutants of F1 and V, which are completely soluble and produced in high yields. We then engineered the vaccine into a novel delivery platform using the bacteriophage T4 nanoparticle. The nanoparticle vaccines induced robust immunogenicity and provided 100% protection to mice and rats against pneumonic plague. These highly efficacious new generation plague vaccines are easily manufactured, and the potent T4 platform which can simultaneously incorporate antigens from other biothreat or emerging infectious agents provides a convenient way for mass vaccination of humans against multiple pathogens.
| Plague, also known as Black Death, is one of the deadliest infectious diseases known to mankind. Yersinia pestis, the etiologic agent of plague, is a Gram-negative bacterium transmitted from rodents to humans via fleas [1]. The bite of an infected flea results in bubonic plague which can then develop into secondary pneumonic plague, resulting in person-to-person transmission of the pathogen through infectious respiratory droplets [2]. Pneumonic plague can also be caused by direct inhalation of the aerosolized Y. pestis, leading to near 100% death of infected individuals within 3–6 days [2], [3]. Due to its exceptional virulence and relative ease of cultivation, aerosolized Y. pestis poses one of the greatest threats for deliberate use as a biological weapon [4]. Since the disease spreads rapidly, the window of time available for post-exposure therapeutics is very limited, usually 20–24 h after the appearance of symptoms [3]. Although levofloxacin has recently been approved by the Food and Drug Administration (FDA) for all forms of plague (http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm302220.htm), prophylactic vaccination is one of the most effective means to reduce the risk of plague.
Stockpiling of an efficacious plague vaccine has been a national priority since the 2001 anthrax attacks but no vaccine has yet been licensed. Previously, a killed whole cell (KWC) vaccine was in use in the United States, and a live attenuated plague vaccine (EV76) is still in use in the states of former Soviet Union [5]. However, the need for multiple immunizations, high reactogenicity, and insufficient protection made the KWC vaccine undesirable for mass vaccination, and, consequently, it was discontinued in the United States [6]. In fact, the live-attenuated vaccine may not meet FDA approval because of the highly infectious nature of the plague bacterium and the virulence mechanisms of vaccine strains have not been fully understood [6], [7]. A cautionary tale related to this is the recent fatality of a researcher as a result of exposure to the attenuated pigmentation-minus Y. pestis strain, KIM/D27 (http://en.wikipedia.org/wiki/Malcolm_Casadaban].
The focus in the past two decades, thus, has shifted to the development of recombinant subunit vaccines [3], [6], [8], [9] containing two Y. pestis virulence factors, the capsular protein (Caf1 or F1; 15.6 kDa, Figure 1A and B) and the low calcium response V antigen (LcrV or V; 37.2 kDa, Figure 1A and C), which is a component of the type 3 secretion system (T3SS). F1 assembles into flexible linear fibers via a chaperone/usher mechanism [10], forming a capsular layer that allows Y. pestis to adhere to the host cell and escape phagocytosis [11] (Figure 1A and 1B). The V antigen forms a “pore” at the tip of the “injectisome” structure of the T3SS needle, creating a channel that delivers a range of virulence factors, also known as the Yersinia outer membrane proteins (Yops), into the host cytosol (Figure 1A) [12]. The V antigen is also critical for impairment of host's phagocytic responses [13]. Abrogation of these functions by F1 and V antibodies appears to be one of the mechanisms leading to protection of the host against lethal Y. pestis infection.
Two types of F1/V recombinant vaccines have been under investigation, one containing a mixture of F1 and V antigens [14], and another, a single F1-V fusion protein [15], [16]. Although both induce protective immunity against Y. pestis challenge in rodent and cynomolgus macaque models, protection of African Green monkeys was insufficient and highly variable [6], [17]. A phase I clinical trial in humans showed that a vaccine consisting of a mixture of F1 and V proteins was immunogenic, however, the antibody titers varied over a wide range leading to concerns about the consistency of vaccine efficacy [18].
One of the problems associated with the current plague vaccines is that the naturally fibrous F1 polymerizes into heterodisperse aggregates, compromising the quality and overall efficacy of the vaccines [15], [19], [20], [21], [22]. Second, the subunit vaccines do not induce adequate cell-mediated immune responses that also appear to be essential for optimal protection against plague [23]. Third, it is unclear if inclusion of other Y. pestis antigens such as the YscF, the structural unit of the injectisome needle (Figure 1A and 1D), can boost the potency of the F1/V vaccines. This is particularly important as F1-minus strains of Y. pestis exist in nature which are as virulent as the wild-type strains [24], [25] and significant diversity in the LcrV sequence of these strains might render the current F1/V vaccines ineffective [26], [27]. Finally, the reported immunosuppressive property of V antigen [13], [28] and whether it could compromise the innate immunity of humans, is a significant concern. These questions must be addressed to generate a next generation plague vaccine that could pass licensing requirements, as well as be manufactured relatively easily for stockpiling.
Recently, we have developed a novel vaccine delivery system using the bacteriophage T4 nanoparticle [29], [30], [31], [32]. The T4 capsid (head) is an elongated icosahedron, 120 nm long and 86 nm wide, composed of three essential capsid proteins: a major capsid protein, gp23*; vertex protein, gp24*; and a portal protein, gp20 (Figure 1E). It is decorated with two non-essential proteins, Soc, the small outer capsid protein, and Hoc, the highly antigenic outer capsid protein. Binding sites for these proteins appear following head “expansion,” a major conformational change that increases the outer dimensions of the capsid by ∼15% and inner volume by ∼50% [33].
Approximately 870 molecules of the tadpole- shaped Soc protein (9 kDa) assemble into trimers at the quasi three-fold axes, clamping to adjacent capsomers and forming a reinforced cage around the shell (Figure 1E) [34]. This stabilizes an already stable head that can withstand harsh extracellular environment (e.g., pH 11) [34]. Hoc, on the other hand, is a linear “fiber” containing a string of four domains, three of which are immunoglobulin (Ig)-like [35]. One hundred and fifty five copies of Hoc fibers, with their NH2-termini projected at ∼160 Å distance from the capsid assemble at the center of each capsomer (Figure 1E). Hoc binds to bacterial surfaces, apparently enriching the phage near its host for infection [36]. Although Soc and Hoc provide survival advantages, they are completely dispensable under laboratory conditions showing no significant effect on phage productivity or infectivity [37]. Purified Soc (or Hoc) protein binds to Hoc− Soc− capsid with high specificity and nanomolar affinity, properties that are not compromised by attachment of a pathogen antigen at the NH2- and COOH-termini [29], [30], [31], [32]. Individual domains, or full-length proteins as large as 90 kDa, or multilayered oligomeric complexes of >500 kDa fused to Soc can be arrayed on T4 capsid, making it a robust antigen delivery platform [29], [30].
Here, we describe two basic approaches to generate next generation plague vaccines, structure-based immunogen design and T4 nanoparticle delivery (Figure 1). We designed an F1 mutant that retained the T cell epitopes but folded into a soluble monomer rather than into an insoluble fiber (Figure 1B). The mutated F1 was fused to V antigen to produce a bivalent F1mut-V immunogen that was also expressed as a soluble monomer. We then constructed an oligomerization deficient YscF mutant (Figure 1D) as well as a V mutant without the putative immunomodulatory sequence (Figure 1C). The mutated antigens were fused to Soc and arrayed on phage T4 nanoparticle (Figure 1F). The F1mut-V monomer induced robust immunogenicity, and the T4-decorated F1mut-V without any adjuvant, in addition, induced balanced TH1 and TH2 responses. Both the soluble and T4 decorated F1mut-V provided 100% protection to mice and rats against intranasal challenge with high doses of Y. pestis CO92. Inclusion of YscF showed a slight enhancement in the potency of F1-V plague vaccine, whereas replacement of V with V10 mutant, which lacks the putative immunosuppressive sequence, did not significantly alter vaccine efficacy. These results provided new insights into plague vaccine design and produced next generation plague vaccine candidates by overcoming some of the concerns associated with the current subunit vaccines.
The X-ray structure and biochemical studies established that F1 polymerizes into a linear fiber by head to tail interlocking of F1 subunits through a donor strand complementation mechanism [10] (Figure 1B). Each subunit has an Ig-like domain consisting of a four-stranded anti-parallel β-sheet. Of the four β-strands, three belong to one subunit forming a cleft into which the NH2-terminal β-strand of the “n+1” subunit locks in, resulting in a bridge connecting adjacent subunits (inter-molecular complementation) (Figure 1B). Stringing of subunits in this fashion leads to assembly of linear F1 fibers of varying lengths. Caf1M chaperone is required for this process because prior to filling the cleft, a “spare” β-strand of Caf1M temporarily occupies the cleft until it is replaced by the β-strand of the incoming subunit with the assistance of an outer membrane usher protein, Caf1A. Over-expression of the F1 gene in a heterologous system such as E. coli (Figure 2) exposes the unfilled hydrophobic cleft, resulting in uncontrolled aggregation of F1 subunits into insoluble inclusion bodies. This is demonstrated in Figure 2B in which all of the over-produced F1 protein partitioned into the pellet (lane 8) and none was detected in the supernatant (lane 7). Denaturation of the insoluble protein recovered some of the F1 protein into the soluble fraction but it still aggregated rapidly leading to precipitation (in the Histrap column) upon removal of the denaturant. Similar aggregation behavior of F1 was observed in previously published studies [38], [39].
We hypothesized that shifting of the NH2-terminal β-strand of F1 to the COOH-terminus should reorient the β-strand such that it fills its own cleft (intra-molecular complementation (Figure 1B), and furthermore, it should no longer require the assistance of chaperone or usher proteins. To test this hypothesis, we constructed an F1 mutant (F1mut1) by deleting the NH2-terminal donor strand [amino acid (aa) residues 1–14] and fusing it to the COOH-terminus with a short (Serine-Alanine) linker in between (Figure 2A). The recombinant F1mut1, as predicted, folded into a soluble protein in the absence of Caf1M or Caf1A, and approximately 70% of the protein partitioned into the cell-free lysate (Figure 2B, lanes 9 and 10). In addition, for reasons unknown, the mutated F1 protein was expressed at significantly higher levels than that of the native F1 protein after IPTG induction (Figure 2B, compare lane 5 with lane 3). The gel filtration profile showed that the F1mut1 eluted as a symmetrical peak corresponding to a molecular mass of ∼19 kDa (Figure 2C), a monomer, suggesting that the interlocking mechanism had shifted from inter- to intra-molecular interactions.
A bioinformatics approach was used to determine if the strand shifting might have disrupted the NH2-terminal epitopes of F1. The aa residues 7 to 20 are reported to contain a mouse H-2-IAd restricted CD4+ T cell epitope [40]. Of the fifty-three predicted 9-mer CD8+ T cell epitopes that encompassed 46 human MHC-I alleles (Table S1 in Text S1), four peptides (aa residues: 9–17, 10–18, 11–19 and 13–21) fell in this region, and of the 9 peptides predicted to contain CD4+ T cell epitopes (Table S2 in Text S1), only one (aa residues 1–18) belonged to this region. We determined that the integrity of these potential linear epitopes could be restored by extending the sequence of the switched strand by up to the aa residue 21, which would duplicate the residues 15 to 21 at the COOH-terminus. Thus, the F1mut2 was constructed (Figure 2A) and tested. The F1mut2 behaved in a similar manner as the F1mut1 with respect to over-production and solubility (Figure 2B, lanes 12–15) and was also purified as a monomer (data not shown).
Fusion of F1mut2 to V would generate a bivalent plague vaccine. Consequently, a mutated F1-V fusion protein (F1mut-V) was produced by fusing F1mut2 to the NH2-terminus of V with a two aa linker in between (Figure 3A), and its solubility was compared to that of the native polymeric F1-V. The native F1-V protein, as reported previously [20], [22], was insoluble and partitioned into inclusion bodies (Figure 3B; lanes 5 and 6). Denaturation and refolding solubilized some of the protein but it also eluted, as was reported previously [20], over a wide range of high molecular weight sizes in a gel filtration column (Figure 3C, red profile). F1mut-V protein, on the other hand, was nearly 100% soluble (Figure 3B; lanes 7 and 8) and eluted as a symmetrical peak corresponding to a molecular weight of ∼64 kDa, equivalent to the mass of monomeric F1mut-V fusion protein (Figure 3C, blue profile). The yield of F1mut-V was quite high, ∼20 mg pure protein per liter of the E. coli culture. Furthermore, its stability to trypsin digestion was similar to that of the native F1-V (Figure 3D).
The Y. pestis V antigen has been reported to induce interleukin (IL)-10 and suppress the production of pro-inflammatory cytokines such as interferon (IFN)-γ and tumor necrosis factor (TNF)-α, which could lead to lowering of innate immunity in vaccinated individuals [41]. A truncated V in which the COOH-terminal 30 aa residues (271–300) were deleted (referred to as “V10” mutation) was reported to lack this immunomodulatory function [41]. A mutated F1mut-V10 recombinant was therefore constructed by deleting these residues (Figure 3A). This mutant protein was also over-produced in E. coli, which was also highly soluble and could be purified as a monomer (Figure 3C, green profile).
Inclusion of YscF might expand the breadth of efficacy of F1-V plague vaccine formulation to Y. pestis strains containing variant V antigens [26], or of those strains devoid of capsule but highly virulent in nature [24], [25]. Since YscF is a structural component of the injectisome, over-production of this protein caused aggregation [42]. A mutant YscF was constructed by mutating the aa residues Asn35 and Ile67, that are involved in oligomerization (Asn35 changed to Ser, and Ile67 changed to Thr) (Figure 4A) [43]. The resultant YscF35/67 mutant protein was soluble and the gel filtration profile showed two peaks, a high molecular weight aggregate near the void volume, and a second peak corresponding to a molecular mass of ∼22 kDa, which is equivalent to a dimer (Figure 4B, blue profile; C). The native YscF, on the other hand, eluted over a wide range of high molecular weight sizes consistent with the formation of heterodisperse aggregates (Figure 4B, red profile). The mutant dimer did, however, show slow aggregation during concentration and storage, as evident by the appearance of small amounts of precipitates.
A large number of F1, V, F1-V, and YscF recombinant proteins, both in native and mutated forms, were fused to the NH2- and/or the COOH-termini of either phage T4 Soc or the T4-related phage RB69 Soc and screened for their solubility as well as ability to bind to T4 phage (Figure 5A, and data not shown). Our previous studies showed that the RB69 Soc binds to T4 capsid at nearly the same affinity as T4 Soc [34]. The RB69 Soc-fused plague antigens, with the exception of the native F1-Soc, produced soluble proteins whereas the T4 Soc-fused antigens were insoluble. Several of the RB69 immunogens were purified (Figure 5B) and tested for binding to T4 using our previously established in vitro assembly system. A typical result is shown in Figure 5C and D, which also exemplifies the versatility of the T4 nanoparticle display. Consistent with the crystal structure of Soc, which showed that both the NH2- and COOH-termini are exposed on the capsid surface, the plague immunogens F1mut and V could be efficiently displayed as an F1mut-V fusion protein that in turn was fused to the NH2-terminus of Soc (Figure 5C). At the same time, its COOH-terminus could be fused to YscF35/67, and the resultant F1mut-V-Soc-YscF35/67 fusion protein containing all three plague immunogens could be displayed on T4 capsid (Figure 5G, lane 4).
The 66 kDa F1mut-V-Soc bound to T4 even at a relatively low 1∶1 ratio of F1mut-V-Soc molecules to Soc binding sites (Figure 5C, red arrow). Binding increased with increasing ratio and reached saturation at 20–30∶1. The copy number of bound F1mut-V-Soc per capsid (Bmax) was 663, which meant that ∼76% of the Soc binding sites were occupied, and its apparent binding affinity (Kd) was 292 nM, which was ∼4-fold lower than that of Soc binding (Kd = 75 nM) [34] (Figure 5D). This is consistent with the expectation that the 66 kDa F1mut-V-Soc, unlike the 10 kDa Soc, would encounter steric constraints to occupy all the binding sites on the capsid exterior. Given this limitation, the observed copy number was remarkably high, with the capsid surface presumably tightly packed with the F1mut-V molecules (model shown in Figure 1F) and exposing, consequently, the plague antigen epitopes for presentation to the immune system. Indeed, cryo-electron microscopy showed that these T4 capsids, unlike the wild-type capsids (Figure 5E), were decorated with a layer of F1mut-V molecules, seen as fuzzy protrusions around the perimeter of the capsid wall (Figure 5F). A series of nanoparticle decorated plague immunogens were prepared, including all three plague immunogens displayed on the same capsid using the F1mut-V-Soc-YscF35/67 fusion protein (Figure 5G, lane 4).
The immunogenicity of mutated F1 immunogen was tested in a mouse model. The animals (Balb/c) were immunized according to the scheme shown in Figure 6 A and B and antibody titers in the sera were determined by ELISA. The data showed that all the three plague antigens adjuvanted with alhydrogel induced antigen-specific antibodies (Figure 6C). The V antigen induced the highest titers with the end point titer reaching as high as 7×106. The YscF antigen was the least immunogenic (Figure 6C, panel III), with the endpoint titers about 1–2 orders of magnitude lower than that of F1 and V antigens (Figure 6C, panels I and II). No significant differences in F1-specific antibody titers were observed among the various groups (i.e., F1-V versus F1mut-V versus F1-V+YscF; panel II). Importantly, the monomeric F1mut-V induced comparable antibody titers as the native polymeric F1-V, suggesting that the capsular structure of F1 per se did not afford a significant advantage to induction of antibodies. However, unexpectedly, the V-specific IgG titers were at least an order of magnitude higher when YscF was also included in the vaccine (p<0.001) (Figure 6C, panel I; compare F1-V to F1-V+YscF).
Intranasal challenge of animals with 90 LD50 of Y. pestis CO92 [1 LD50 = 100 colony forming units (CFU) in Balb/c mice], one of the most lethal strains, showed that all the control mice died by day 3. However, the mice immunized with native V immunogen showed 83% survival (two of twelve mice died), whereas the mice immunized with F1-V, F1mut-V, or F1-V plus YscF were 100% protected (Figure 6D). The surviving mice were then re-challenged with a much higher dose, 9,800 LD50, of Y. pestis CO92 on day-48 post-first challenge. The purpose of re-challenge was to determine if a strong adaptive immunity was generated after first infection with Y. pestis, which should in turn confer a much higher level of protection against subsequent challenges. Indeed, our data showed that all of the mice survived the re-challenge except two mice in the native F1-V group that succumbed to infection (83% protection) (Figure 6D). All of the naïve animals of same age which were used as a re-challenge control died as expected. These efficacy results showed that the monomeric F1mut-V was as efficacious as or even slightly better than the native F1-V polymer.
The immunogenicity of nanoparticle decorated plague antigens was tested by vaccinating mice with phage T4 particles (Figure 7A). The amount of the antigen was kept the same as that of the soluble preparations (Figure 6); however, the T4 formulations contained no adjuvant. The data showed that the T4 displayed plague antigens induced comparable antibody titers as the adjuvanted soluble antigens (Figure 7B). The challenge data showed that all the T4 decorated plague antigens, including the V alone group, provided 100% protection to mice against intranasal challenge with 90 LD50 of Y. pestis CO92; all the control animals died by day 4. Upon re-challenge on day 48 post-first challenge with 9,800 LD50 (Figure 7C), all of the mice were completely protected. As expected, the control re-challenge group of mice succumbed to infection. Overall, these data suggested that the T4 nanoparticle arrayed plague antigens might be more potent than the soluble antigens, as two deaths in each of the V and F1-V groups of mice occurred with the soluble vaccines (Figure 6D) but not with the T4 vaccines.
Stimulation of both arms of the immune system, humoral (TH2) and cellular (TH1), is probably essential for protection against Y. pestis infection [6], [23], [44], [45]. In mice, the TH1 profile involves induction of antibodies belonging to IgG2a subclass whereas the TH2 profile primarily involves the induction of IgG1 subclass. To determine the specificity of antibodies induced by soluble vs T4 displayed antigens, the subclass IgG titers were determined by ELISA (Figure 8). These data showed that the soluble antigens and the T4 displayed antigens induced comparable IgG1 titers (TH2 response) (Figure 8A) whereas the T4 antigens evoked 1–2 orders of magnitude higher IgG2a titers than the soluble antigens (TH1 response) (Figure 8B). These results suggested that the T4 decorated plague immunogens stimulated stronger cellular responses as well as humoral responses, whereas the soluble antigens showed a bias towards the humoral responses as was observed in the previous studies [46].
The immunogenicity and protective efficacy of F1mut-V vs F1mut-V10 was evaluated by three criteria: F1- and V-specific antibody titers, cytokine responses, and protection against Y. pestis CO92 challenge. Both the F1- and V-specific IgG antibodies (Figure 9B) and subclass IgG titers (Figure 8A and B) were not significantly different between the F1mut-V and F1mut-V10 immunized groups of mice when the immunogen used was soluble and alhydrogel-adjuvanted. However, when decorated on phage T4 nanoparticle with no adjuvant, F1mutV elicited higher total IgG (Figure 9B) and IgG1 titers (Figure 8A) than F1mutV10 (p<0.05). These trends were also reflected in the production of the TH2 cytokines, IL-4 and IL-5, by splenocytes of immunized mice stimulated ex vivo with F1-V. Similar levels of IL-4 and IL-5 were produced by the soluble F1mut-V and F1mut-V10 antigens or the T4-displayed F1mut-V, whereas the T4 displayed F1mut-V10 showed slightly reduced levels (Figure 10). The induction of proinflammatory cytokines, such as IL-1α and IL-1β was also similar, irrespective of whether the antigens were soluble or T4 displayed (Figure 10). However the levels of TNF-α, an inflammatory mediator that synergistically acts with IFN-γ to help bridge the gap between innate and cell-mediated immune responses, were significantly higher in mice immunized with soluble F1mut-V10 than those immunized with F1mut-V (Figure 10). However, the trend was opposite when F1mut-V and F1mut-V10 immunogens were T4 displayed, although the data did not reach statistical significance. In fact, the T4 displayed F1mut-V10 induced overall weaker IFN-γ and cytokine responses when compared to its F1mut-V counterpart.
With respect to animal survival, both the F1mut-V and F1mut-V10 immunogens, either soluble or T4 displayed, provided 100% protection to mice upon intranasal challenge with 5,350 LD50 of Y. pestis CO92 (Figure 9C), with the control animals dying by day 3. When the mice were re-challenged with an extremely high LD50 (20,000) on day 88 post-first challenge, all the groups showed 100% protection except the T4-displayed F1mut-V10 group in which one mouse died (92% protection) (Figure 9C). All of the naïve re-challenge control animals died by day 4.
To further test the efficacy of the mutated inmunogens, a rat study was conducted. Rats [47], the natural host of Y. pestis, were vaccinated with alhydrogel adjuvanted F1mut-V, and F1mut-V10 as well as the T4 nanoparticle displayed F1mut-V and F1mut-V10 (Figure 11A). The same immunization scheme as shown in Figure 6B was used and the animals were challenged with a 5,000 LD50 of Y. pestis CO92. The data showed that all the control animals died by day 4 whereas all the F1mut-V and F1mut-V10 immunized animals were 100% protected (Figure 11B).
Since the deadly anthrax attacks in 2001, stockpiling of recombinant anthrax and plague vaccines to protect masses against a potential bioterror attack became a national priority. However, no plague vaccine has yet been licensed. The reasons include poor stability, insufficient immunogenicity, and/or manufacturing difficulties associated with the current formulations. New immunogen designs and vaccine platforms that could overcome some of these problems would be of great interest not only to stockpile efficacious biodefense vaccines but also to develop vaccines against a series of infectious diseases of public health importance. Here, by using structure-based immunogen design and T4 nanoparticle delivery approaches, we have engineered new and efficacious plague vaccines that could be manufactured relatively easily and provide complete protection against pneumonic plague in two rodent models.
The surface-exposed Y. pestis antigens F1 and V have been the leading candidates for formulating a subunit plague vaccine for nearly two decades [14], [15], [16], [17]. Although poorly immunogenic by themselves, their immunogenicity could be enhanced by adjuvantation with Alum [15] or by fusion with a molecular adjuvant such as flagellin [19]. While complete protection was observed in rodent models [17], these vaccines impart partial and varied protection in African Green monkeys [6], [17]. Another concern has been that the naturally polymeric F1 has high propensity to aggregate (Figure 2). When produced in a heterologous system such as E. coli, the recombinant F1-V protein partitions into insoluble inclusion bodies [15], [19], [20], [21], [22] (Figure 3). Although it can be partially recovered in soluble form by denaturation and re-folding, the preparation still consists of a mixture of heterogenous aggregates and varying amounts of the misfolded protein [20]. These might also trap contaminants, compromising the overall purity, stability, and efficacy of the vaccine. Attempts to produce a monomeric vaccine by mutating the lone cysteine residue in V have not been successful [20].
We proposed three hypotheses to design a soluble monomeric plague vaccine, yet retaining its structural and epitope integrity. First, we hypothesized that the β-strand that connects the adjacent F1 subunits requires repositioning. This was achieved by transplanting the NH2-terminal β-strand to the COOH-terminus in such a way that the reoriented β-strand fitted into its own β-sheet cleft rather than that of the adjacent F1 subunit. It also eliminated the need for chaperone and usher mediated oligomerization as there would no longer be an unfilled β-sheet pocket exposed in the F1 subunit. Second, by using epitope predictions, the NH2-terminal aa residues 15–21 of F1 flanking the β-strand were duplicated at the COOH-terminal end to restore any potential T-cell epitopes that might have been lost during the switch. This is important because in a previous study, a simple β-strand switch produced a less stable monomer with diminished immunogenicity [48]. Third, the mutated F1 was fused to the NH2-terminus of V with a flexible linker in between to minimize interference between the F1 and V domains. The bivalent F1mut-V immunogen thus produced showed a remarkable shift in solubility, from an insoluble F1-V polymer to a completely soluble monomer (Figure 3). The monomer could be purified from cell-free lysates at high yields, ∼20 mg of pure protein from a liter of E. coli culture, which we believe could be substantially increased under optimized conditions in a fermentor.
Several lines of evidence demonstrated that the F1mut-V monomer was as efficacious as, if not better than, the native F1-V polymer. In four separate immunization studies and two animal models (Figures 6, 7, 9, and 11), F1mut-V induced robust immunogenicity and protective efficacy. It showed similar levels of F1- and V-specific antibody titers as the native F1-V, and no significant differences were observed in TH1 vs TH2 specific IgG subclass titers. Furthermore, F1mut-V overall showed stronger cytokine responses and conferred 100% protection in vaccinated mice and rats, including when very high doses of Y. pestis CO92, ∼5,350 LD50 for first challenge and ∼20,000 LD50 for re-challenge, were administered by the intranasal route (Figure 9). The native F1-V, on the other hand, showed slightly lower protection (∼83%) upon re-challenge (Figure 6).
The possibility of increasing the breadth and potency of F1-V vaccine by inclusion of YscF was tested by constructing an oligomerization deficient YscF35/67 mutant [43]. Such a vaccine might be effective even against those Y. pestis strains that contain variant V antigens or lack the capsule, but are highly virulent [26]. The mutated protein, purified as a soluble dimer, elicited YscF-specific antibodies on its own, and, when it was mixed with F1-V, it enhanced the induction of V-specific antibody titers as well as survival rate in mice (Figure 6). While these results indicated enhanced potency of F1-V vaccine in the presence of YscF, more studies are needed to determine if the cost of an additional protein can be justified for vaccine manufacture. On the other hand, the T4 displayed trivalent vaccine, F1mut-V-Soc-YscF (Figures 5 and 7), might offer an alternative to incorporate YscF into the plague vaccine formulation.
Y. pestis infection stimulates IL-10 production which in turn suppresses the production of proinflammatory cytokines IFN-γ and TNF-α. Both IFN-γ and TNF-α are important for innate immunity, as well as to elicit TH1 immune responses that might be essential for protection against pneumonic plague [49], [50], [51]. These immunomodulatory functions, in part, were attributed to the V antigen, specifically to the NH2-terminal aa residues 31–49 [49]. Deletion of these residues, or of the COOH-terminal aa residues 271–300 (V10 mutation), have been reported to abrogate the suppression of IFN-γ and TNF-α [41], presumably by preventing the interaction of V with toll like receptor 2 (TLR2) and CD14, the receptors of the innate immune system [49], [52]. Our studies showed that both the F1mut-V and F1mut-V10 immunogens produced similar levels of IFN-γ and other proinflammatory cytokines, such as IL-1α and IL-1β, upon stimulation ex-vivo of splenocytes from immunized mice with F1mut-V. However, TNF-α was induced to significantly higher levels in the F1mut-V10 group (Figure 10), consistent with the published report [41]. However, the T4 nanoparticle decorated F1mut-V10 showed opposite trend, producing much reduced levels of TNF-α as well as IFN-γ and other cytokine responses than its F1mut-V counterpart, a result also correlated with lower protection against re-challenge [92% protection with T4 displayed F1mut-V10 vs 100% protection with T4 displayed F1mut-V upon re-challenge with 20,000 LD50 (Figure 9C)]. Thus, our data did not show consistent enhancement of proinflammatory cytokines by the V10 mutation, hence it is questionable that replacing native V with V10 mutant would lead to a significant beneficial effect in a new plague vaccine design. On the other hand, from a structural standpoint, deletion of the aa residues 271–300 disrupts the coiled coil bridge between the NH2- and COOH-domains of V [12], which would likely make V10 mutant a conformationally more flexible molecule and could adversely affect vaccine stability and efficacy.
Although humoral immune responses are critical for protection against plague, several studies have shown that cell-mediated immunity also plays important roles [23], [53], [54]. Wang et al. [53] established the role of CD8+ T cells in protection of mice against pneumonic plague evoked by Y. pestis KIM 1001 strain. This study corroborated the earlier report of Parent et al. [23], which concluded that plague vaccines that generate both humoral- and cell-mediated immune responses will be most effective. Likewise, Philipovskiy and Smiley (3) reported that mice vaccinated with a live Y. pestis vaccine primed both CD4+ and CD8+ T cells, which when passively transferred to naïve mice, provided protection against pulmonary Y. pestis infection [54]. The adjuvant-free T4 nanoparticle decorated F1mut-V induced robust F1- and V-specific antibody responses, as well as provided 100% protection to mice and rats against very high doses of Y. pestis challenge (Figures 7, 9 and 11). In addition, T4 delivery induced balanced TH1 and TH2 responses with a potent TH1 response, as evident from the induction of subclass IgG2a specific antibodies. Similar patterns were observed in our previous studies with the T4 displayed HIV-1 p24 immunogen [32]. Presumably, the large size of the T4 phage particle (capsid, 120 nm×86 nm; tail, 100 nm) allows for its efficient uptake by the antigen presenting cells and cross-presentation to both MHC-I and MHC-II molecules, stimulating both the humoral and cellular arms of the immune system. It is also possible that the T4 phage DNA containing CpG might potentially serve as a TH1-type of adjuvant. Indeed, studies have shown that F1-V vaccine adjuvanted with CpG or poly IC (also a TH1 type adjuvant), given by the intranasal route, induced both TH1 and TH2 responses, providing better protection to mice against bubonic and pneumonic plague [55], [56]. Thus, T4 might be a particularly useful platform for plague vaccine design since clearance of the Y. pestis bacterium may require a balanced response that is generally not seen with the current F1-V vaccines [46]. We also note that, although the mechanistic basis for T4 responses is currently unknown, no adverse effects to T4 vaccination have been observed in many preclinical studies performed in mouse, rat, rabbit, and rhesus macaque models [31], [57], [58], or in a human trial where T4 phage was given orally [59].
There has been a considerable urgency to develop a recombinant plague vaccine, but several concerns precluded licensing of current formulations. Our studies have established that the F1mut-V recombinant vaccine is efficacious and easily manufacturable and should be seriously considered as a next generation plague vaccine. Future studies would include preclinical evaluation of protection against Y. pestis infection in cynomolgus macaques as well as African Green monkeys, potentially leading to human clinical trials. Although the soluble F1mut-V vaccine adjuvanted with alum would be relatively easy to manufacture, the phage T4 nanoparticle-decorated F1mut-V vaccine offers certain advantages. First, the T4 formulation provided enhanced vaccine potency in small animal models. Second, the T4 vaccine would not require an extraneous adjuvant, and third, additional antigens from other biodefense pathogens, such as the protective antigen (PA) from Bacillus anthracis could be incorporated into the same formulation generating a dual vaccine against both inhalation anthrax and pneumonic plague. Our recent study demonstrated that the T4 displayed PA provided complete protection to Rhesus macaques against aerosol challenge with Ames spores of B. anthracis [51]. Fourth, the large interior of T4 head which has the capacity to package ∼171 kb DNA can also be used to deliver DNA vaccines [60]. By combining protein display outside and DNA packaging inside the T4 nanoparticles can simultaneously deliver vaccine antigen(s) as well as vaccine DNAs, similar to that of the prime-boost strategy, potentially inducing robust and long-lasting immune responses. Finally, such prime-boost vaccines could be targeted to antigen-presenting dendritic cells (DCs) by displaying a DC-specific ligand on the capsid using Hoc, further stimulating the cell-mediated immunity. One or two doses of such potent nanoparticle vaccines might be sufficient to afford protection against multiple biothreat agents. With the recent data demonstrating the proof of concept [60], we are currently developing these novel vaccine platforms, not only to defend against biowarfare pathogens but also to generate efficacious vaccines against complex infectious agents such as HIV-1 and malaria.
This study was conducted in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were reviewed and approved by the Institutional Animal Care and Use Committees of the University of Texas Medical Branch, Galveston, TX, (Office of Laboratory Animal Welfare assurance number: A3314-01) and The Catholic University of America (Office of Laboratory Animal Welfare assurance number: A4431-01).
The T7 promoter containing E. coli expression vector pET28b (Novagen, MA) was used for recombinant plasmid construction. The template DNAs containing Y. pestis F1, V, or YscF were kindly provided by Dr. Richard Borschel from the Walter Reed Army Institute of Research (Silver Spring, MD). E. coli XL-10 Gold cells (Stratagene, CA) were used for the initial transformation of clones. The plasmid DNAs were then re-transformed into E. coli BL21 (DE3) RIPL (Novagen, MA) for expression of recombinant proteins. The Hoc− Soc− phage T4 was propagated on E. coli P301 and purified by CsCl gradient centrifugation.
The DNA encoding F1, V, or YscF were amplified by PCR using primers containing appropriate restriction site(s) (NheI/XhoI for F1 and YscF, and NheI/HindIII for V). The PCR products were purified, digested with appropriate restriction enzymes, and ligated with pET-28b vector DNA digested with the same restriction enzymes. The resulting plasmids had F1, V, or YscF coding sequences fused in-frame with the 23 aa vector sequence containing a hexa-histidine tag at the NH2-terminus. The YscF mutant, YscF35/67, which contained point mutations at aa 35 (Asn to Ser) and 67 (Ile to Thr) was amplified by overlap PCR [61] followed by digestion with NheI and XhoI enzymes. YscF35/67 DNA was then ligated into the linearized pET28b vector. The F1mut1, in which the first 14 aa residues were deleted and fused to the COOH-terminus with a two aa (SA) linker, was constructed by two rounds of PCR. The first round of PCR was performed to amplify F1 fragment in which the NH2-terminal 14 aa residues were deleted. This PCR product was used as a template for the second round of PCR using a forward primer containing NheI restriction site and a reverse primer containing the NH2-terminal 14 aa residues and XhoI restriction site. The PCR fragment was then inserted into NheI and XhoI linearized pET28b vector.
To construct F1mut2 in which aa residues 15 to 21 were duplicated at the COOH-terminus, a reverse primer with a 5′-tag corresponding to the 15 to 21 aa sequence and XhoI restriction site was used for PCR amplification. The F1mut2 fragment was then inserted into NheI and XhoI linearized pET28b vector. To construct F1-V recombinants, V was first amplified and inserted into BamHI and HindIII linearized pET28b vector to generate the pET-V clone. F1 and F1mut2 were amplified with primers containing NheI and BamHI restriction sites, digested with NheI and BamHI, and ligated with the pET-V vector DNA digested with the same restriction enzymes. The resulting F1-V and F1mut-V plasmids contained F1 or F1mut in-frame fusion with V and a 23-aa vector sequence containing the hexa-histidine sequence at the NH2-terminus of F1. The F1mut-V10 was amplified by overlap PCR using F1mut-V as the template and the mutated DNA was inserted into the NheI and HindIII linearized pET28b vector.
T4 Soc gene or RB69 Soc gene was fused with V, F1, or YscF with two aa (GS) linker by overlap PCR and the amplified DNA was inserted into the pET28b vector. The fused products V-T4 Soc, F1-T4 Soc, V-RB69Soc, and F1-RB69 Soc were further fused to YscF by overlap PCR to generate V-Soc (T4 or RB69)-YscF and F1-Soc (T4 or RB69)-YscF. Two aa residues, GS, were used as a linker between Soc and YscF. To construct F1-V-Soc clones, RB69 Soc gene was first amplified with end primers containing HindIII and XhoI restriction sites and inserted into the HindIII and XhoI linearized pET28b vector. This clone was then linearized by digestion with NheI and HindIII restriction enzymes. F1mut-V and F1mut-V10 DNAs were amplified by using the end primers containing NheI and HindIII restriction sites and inserted into the above plasmid. The resulting clones contained F1mut-V or F1mut-V10 fused in-frame to the NH2-terminus of RB69 Soc and also the flanking vector sequences containing two hexa-histidine tags at both NH2- and COOH-termini. The F1mut-V-Soc was then fused with YscF by overlap PCR with a two aa linker, GS, between Soc and YscF. All of the clones were sequenced (Retrogen, CA) and only the clones containing 100% sequence accuracy were used for protein purification. The primer sequences used and clones generated in these studies will be available upon request.
The structural models of F1, V, YscF, and T4 phage nanoparticle (Figure 1) were constructed using Chimera version1.4.1 [62]. The T cell epitopes were predicted using MetaMHC, a new web server which integrates the outputs of leading predictors by several popular ensemble strategies [63]. This was shown to generate statistically significant results that were more reliable than the individual predictors [63]. For the CD4+ T cell epitope prediction, F1 protein sequence was screened against 14 human MHC-II alleles. Peptides identified as positive ones by at least one predictor method were considered as potential CD4+ T cell epitopes. For the CD8+ T cell epitope prediction, F1 was screened against 57 human MHC-I alleles. Peptides identified as positive by at least one ensemble predictor approaches were considered to be potential CD8+ T cell epitopes. Default values were used for both the T cell epitope predictions.
The E. coli BL21 (DE3) RIPL cells harboring various plague recombinant plasmids constructed as above were induced with 1 mM IPTG for 1 to 2 h at 30°C. The cells were harvested by centrifugation at 4,000 g for 15 min at 4°C and the pellets were resuspended in 50 mM Tris-HCl (pH 8.0). Solubility analysis was carried out using bacterial protein extraction reagent (B-PER) (Thermo Fisher Scientific Inc., Rockford, IL). The cells were lysed with B-PER and centrifuged at 12,000 g for 10 min. The soluble supernatant and insoluble pellet fractions were analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) as follows. The samples were boiled in a buffer containing SDS and β-mercaptoethanol, and were electrophoresed on a 12% or 15% (w/v) polyacrylamide gel. Since the protein aggregates will be dissociated into monomers under these conditions. The molecular weight differences observed in Figure 2B reflect sizes of the polypeptide chains of F1, F1mut1, and F1mut2. For example, F1mut1 and Fmut2 are approximately 1.6 kDa and 2.2 kDa larger than F1 because F1mut1 has a two amino acid linker (SA) and an eight amino acid His-tag (LEHHHHHH) (orange) at the C-terminus. F1mut2, in addition, has the duplicated T cell epitope (EPARITL) (blue) (Figure 2A).
For protein purification, the cells were resuspended in binding buffer (50 mM Tris-HCl pH 8.0, 300 mM NaCl and 20 mM imidazole) containing proteinase inhibitor cocktail (Roche, USA). The cells were lysed by French press (Aminco, IL, USA) at 12,000 psi and the soluble fractions containing the His-tagged fusion proteins were isolated by centrifugation at 34,000 g for 20 min. The supernatants were filtered through 0.22 µm filters (Sartorius Stedim Biotech, Germany) and loaded onto 1 ml HisTrap column (AKTA-prime, GE Healthcare) pre-equilibrated with 20 ml of binding buffer. After washing with the binding buffer containing 50 mM imidazole, the proteins were eluted with 20–500 mM linear imidazole gradient. The peak fractions containing the desired protein were concentrated by Amicon Ultra-4 centrifugal filtration (10 kDa cut-off; Millipore). The proteins were further purified by gel filtration on Hi-load 16/60 Superdex 200 column (AKTA-FPLC, GE Healthcare) in a buffer containing 20 mM Tris-HCl, pH 8.0 and 100 mM NaCl. The peak fractions containing the purified proteins were concentrated and stored at −80°C. The native F1 recombinant proteins were purified from the pellet containing the insoluble inclusion bodies. The pellet was dissolved in the binding buffer containing 8 M urea and loaded onto 1 ml HisTrap column (AKTA-prime, GE Healthcare) pre-equilibrated with the same buffer. The proteins were renatured by washing the column with a decreasing urea gradient (8 to 0 M) in the binding buffer. The bound proteins were then eluted with 20–500 mM linear imidazole gradient. If necessary, the peak fractions from the HisTrap column were concentrated by Amicon Ultra-4 centrifugal filtration (10 kDa cut-off). The proteins were further purified by gel filtration on Hi-load 16/60 Superdex 200 column as described above.
The levels of lipopolysaccharide (LPS) contamination in the purified recombinant Y. pestis antigens from E. coli; F1, LcrV, YscF, and F1mut-V, were determined using Endosafe PTS system (Charles River Laboratories International, Inc.,Wilmington, MA). This system consists of a handheld spectrophotometer and utilizes FDA approved disposable cartridges. At least three batches of each antigen were tested. The endotoxin levels ranged from 0.05 to 0.8 EU/ml, substantially lower than the maximum recommended in gene vectors and subunit vaccines, 10 and 20 EU/ml respectively, for preclinical research [64].
In vitro binding of plague-Soc fusion protein on Hoc− Soc− T4 phage was carried out as previously described [29], [30], [32]. About 3×1010 phage particles were sedimented for 45 min at 34,000 g in LoBind Eppendorf tubes and resuspended in phosphate-buffered saline (PBS) buffer (pH 7.4). Various Soc fusion proteins were incubated with the resuspended Hoc− Soc− phage at 4°C for 45 min. The phage particles were sedimented at 34,000 g for 45 min and the supernatant containing the unbound protein was discarded. The phage pellet containing the bound plague antigen(s) was washed twice with excess buffer containing 20 mM Tris-HCl pH 8 and 100 mM NaCl. The final pellets were resuspended in PBS buffer (pH 7.4) and analyzed by SDS-PAGE. The gels were stained with Coomassie Blue R250 (Bio-Rad, USA) and the protein bands were quantified by laser densitometry (PDSI, GE Healthcare). The density of Soc fusion protein, gp23*, and gp18 (major tail sheath protein; 70 kDa) bands were determined for each lane separately and the copy number of bound plague antigen molecules per capsid was calculated using the known copy numbers of gp23* (930 molecules per capsid) or gp18 (138 molecules per capsid). A saturation binding curve relating the number of bound plague protein-Soc molecules per capsid (Y) and the concentration of unbound protein in the binding reaction (X) was generated by SigmaPlot software. The apparent Kd (association constant) and Bmax (maximum copies of Soc fusion protein bound per capsid) were determined using the equation Y = BmaxX/(Kd+X) as programmed in the SigmaPlot software.
Six to eight weeks female Balb/c mice (17–20 g) were purchased from Jackson Laboratories (Bar Harbor, Maine) and randomly grouped and acclimated for 7 days. Equivalent amounts of plague immunogen molecules, either soluble or phage-bound, were used for each immunization. For immunization of soluble antigens, the purified proteins (10 µg/mouse/immunization) were adsorbed on alhydrogel (Brenntag Biosector, Denmark) containing 0.19 mg of aluminum per dose. For the T4 displayed antigens, the phage particles were directly used without any adjuvant (10 µg of plague antigen/mouse/immunization). On days 0, 21 and 42, mice were vaccinated via the intramuscular route. Alternate legs were used for each immunization. Blood was drawn from each animal on days 0 (pre-bleeds), 35 and 49 and the sera obtained were stored frozen at −70°C. On day 56, mice were intranasally challenged with Y. pestis CO92 BEI strain [65] using the indicated LD50. Animals were monitored and recorded twice daily for mortality or other symptoms for 48 to 88 days. The animals that survived were re-challenged intranasally at 48 or 88 days post-first challenge with the indicated LD50 and monitored twice daily for a further 48 days.
Female Brown Norway rats (50–75 g) were purchased from Charles River (Houston, TX). Upon arrival, animals were weighed and randomized into the treatment groups and were acclimated for several days before manipulation. The plague immunogens were prepared as described above. On days 0, 21 and 42, rats were vaccinated via the intramuscular route with 15 µg antigen in 50 µl PBS buffer. Alternate legs were used for each immunization. On day 56, animals were intranasally challenged with 5,000 LD50 of Y. pestis CO92 BEI strain and were monitored twice daily for 30 days and clinical symptoms of disease and survival recorded.
The IgG titers were determined by ELISA. Briefly, 96-well microtiter plates (Evergreen Scientific, Los Angeles, CA) were coated with 10 ng/well of purified F1, V, YscF, F1-V or F1mutV antigen at 4°C overnight. Following blocking and washing, sera from naïve and immunized mice were serially diluted and incubated with the affixed antigens for 1 h at room temperature. Following several washes, horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG secondary antibody was added to the wells at a dilution of 1∶10,000. After incubation for 1 h at room temperature, the unbound antibody was removed and the wells were washed several times and the TMB (3,3′,5,5′-tetramethylbenzidine) substrate was added. Following a 20 min incubation to develop the color, the reaction was quenched by the addition of 2 N H2SO4 and the absorbance was read at 450 nm using an ELISA reader. For IgG subtypes, horseradish peroxidase-conjugated goat anti-mouse IgG1 or IgG2a secondary antibodies were used.
Seven days after the second boost (day 49), mice were sacrificed and spleens were harvested to prepare splenocytes using the lymphocyte separation medium. The isolated lymphocytes were adjusted to ∼5×106 cells/ml and 1 ml of lymphocytes seeded into each well. Triplicate cultures from each group were stimulated with purified F1-V (10 µg/ml). Additional control stimulators included medium only and concanavalin A (5 µg/ml). After approximately 48 h incubation at 37°C in a humidified (5% CO2 in air) incubator, culture supernatants were collected. Cytokines were measured using a multiplex assay (Millipore, Billerica, MA). The results were analyzed in Prism and the statistical significance was determined by one way ANOVA with Bonferroni correction.
F1 capsule antigen (caf1) [GeneID: 1172839, Sequence: NC_003134.1 (85950..86462)], lcrV [GeneID: 1172676; Sequence: NC_003131.1 (21935..22915, complement)], yscF [GeneID: 1172700, Sequence: NC_003131.1 (41026..41289)], and soc (RB69Soc) [GeneID: 1494143, Sequence: NC_004928.1 (14980..15216, complement)].
|
10.1371/journal.ppat.1004711 | Suppression of RNAi by dsRNA-Degrading RNaseIII Enzymes of Viruses in Animals and Plants | Certain RNA and DNA viruses that infect plants, insects, fish or poikilothermic animals encode Class 1 RNaseIII endoribonuclease-like proteins. dsRNA-specific endoribonuclease activity of the RNaseIII of rock bream iridovirus infecting fish and Sweet potato chlorotic stunt crinivirus (SPCSV) infecting plants has been shown. Suppression of the host antiviral RNA interference (RNAi) pathway has been documented with the RNaseIII of SPCSV and Heliothis virescens ascovirus infecting insects. Suppression of RNAi by the viral RNaseIIIs in non-host organisms of different kingdoms is not known. Here we expressed PPR3, the RNaseIII of Pike-perch iridovirus, in the non-hosts Nicotiana benthamiana (plant) and Caenorhabditis elegans (nematode) and found that it cleaves double-stranded small interfering RNA (ds-siRNA) molecules that are pivotal in the host RNA interference (RNAi) pathway and thereby suppresses RNAi in non-host tissues. In N. benthamiana, PPR3 enhanced accumulation of Tobacco rattle tobravirus RNA1 replicon lacking the 16K RNAi suppressor. Furthermore, PPR3 suppressed single-stranded RNA (ssRNA)—mediated RNAi and rescued replication of Flock House virus RNA1 replicon lacking the B2 RNAi suppressor in C. elegans. Suppression of RNAi was debilitated with the catalytically compromised mutant PPR3-Ala. However, the RNaseIII (CSR3) produced by SPCSV, which cleaves ds-siRNA and counteracts antiviral RNAi in plants, failed to suppress ssRNA-mediated RNAi in C. elegans. In leaves of N. benthamiana, PPR3 suppressed RNAi induced by ssRNA and dsRNA and reversed silencing; CSR3, however, suppressed only RNAi induced by ssRNA and was unable to reverse silencing. Neither PPR3 nor CSR3 suppressed antisense-mediated RNAi in Drosophila melanogaster. These results show that the RNaseIII enzymes of RNA and DNA viruses suppress RNAi, which requires catalytic activities of RNaseIII. In contrast to other viral silencing suppression proteins, the RNaseIII enzymes are homologous in unrelated RNA and DNA viruses and can be detected in viral genomes using gene modeling and protein structure prediction programs.
| RNA interference (RNAi) is a cellular mechanism activated by double-stranded RNA (dsRNA). Cellular dsRNA-specific RNaseIII enzymes (Dicer) recognize dsRNA and process it into double-stranded small interfering RNAs (ds-siRNAs) of 21–25 nucleotides (nt). siRNAs guide RNAi to degrade also single-stranded RNA homologous to the trigger. RNAi regulates gene expression, controls transposons, and represents an important antiviral defense mechanism. Therefore, viruses encode proteins dedicated to countering RNAi. In this study, the RNaseIII enzymes of a fish DNA virus (PPIV) and a plant RNA virus (SPCSV) were compared for suppression of RNAi in non-host organisms. The fish iridovirus RNaseIII suppressed RNAi in a plant and a nematode. It also enhanced accumulation of an RNAi suppressor deficient virus in plants, and suppressed antiviral RNAi and could rescue multiplication of an unrelated, RNAi suppressor-defective virus in nematodes. In contrast, the plant virus RNaseIII could suppress RNAi only in plants. Our results underscore that the active viral RNaseIII enzymes suppress RNAi. Their activity in suppression of RNAi seems to differ for the spectrum of unrelated organisms. Understanding of this novel mechanism of RNAi suppression may inform means of controlling the diseases and economic losses which the RNaseIII-containing viruses cause in animal and plant production.
| Eukaryotic RNA interference (RNAi) pathways are activated by double-stranded RNA (dsRNA) and function in gene regulation and antiviral defense [1–5]. In invertebrates, genes can be silenced via dsRNA as demonstrated in the nematode Caenorhabditis elegans [6] and the fruit fly Drosophila melanogaster [7], whereas in plants post-transcriptional gene silencing also can be induced by homologous antisense or positive-sense single-stranded RNA (ssRNA) [8]. Induction of sense-mediated RNAi typically requires the activity of a cellular RNA-dependent RNA polymerase (RdRp) for synthesis of dsRNA on the sense RNA transcript [9]. Class 3 RNaseIII endoribonucleases known as Dicers contain a dsRNA-binding domain, two catalytic domains (RNaseIII signature motifs), an N-terminal helicase, and a PAZ domain [10, 11]. Dicers recognize dsRNA and process it into double-stranded small interfering RNAs (ds-siRNAs) that are 21–25 nucleotides (nt) long [1, 12]. siRNAs bind to and guide the cellular RNase AGO to cleave complementary ssRNA molecules [13, 14]. RdRp helps to amplify RNAi via production of secondary triggers of RNAi derived from cleaved RNA in plants and nematodes (C. elegans) [12, 15] and hence also contributes to the generation of secondary siRNAs acting as mobile signals for systemic RNAi in plants, nematodes, and possibly insects (D. melanogaster) [5, 16].
Replicating viruses are vulnerable to RNAi, because the double-stranded replicative intermediates of viral RNA genomes, the secondary structures of RNA transcripts, and the sense-antisense transcript pairs resulting from bidirectional transcription of DNA viruses can be recognized by Dicers. Therefore, viruses encode proteins dedicated to countering RNAi and protecting viral RNA from degradation. Viral RNAi suppressor proteins interfere with the host antiviral RNAi pathway, for example by binding to the dicing complex, dsRNA or siRNA, by preventing assembly of the AGO-containing silencing complex, or inhibiting production of secondary ds-siRNA [16, 17].
The viral RNAi suppressor proteins identified to date do not contain conserved amino acid motifs or other structural features, but a variety of different types of viral proteins can suppress RNAi and target the same molecular components and steps in the RNAi pathway [17, 18]. It is therefore not possible to recognize an RNAi suppressor without carrying out pertinent experiments. However, certain plant and animal viruses encode homologous dsRNA-specific Class 1 RNaseIII enzymes [19–22], of which the dsRNA-specific Class 1 RNaseIII endoribonuclease termed RNase3 (designated as CSR3 in this paper) of Sweet potato chlorotic stunt virus (SPCSV) suppresses RNAi [21]. SPCSV contains a positive-sense ssRNA [(+)ssRNA] genome, but also iridoviruses (family Iridoviridae) which have large dsDNA genomes and infect invertebrate or poikilothermic vertebrate animals [19], encode putative Class 1 RNaseIII enzymes. Indeed, the RNaseIII of rock bream iridovirus is a dsRNA-specific endoribonuclease [20], but little is known about its role in infection and suppression of RNAi. Heliothis virescens ascovirus 3e (HvAV-3e, family Ascoviridae) contains a DNA genome and infects insects. It encodes an RNaseIII that was shown to suppress gene silencing following expression from a recombinant baculovirus in infected insect cells [22].
Compared with Dicers, Class 1 RNaseIIIs have a simple structure. Similar to the Class 1 RNaseIII of Escherichia coli [23], CSR3 contains a single catalytic domain and a dsRNA-binding domain and cleaves long dsRNA molecules in an Mg2+-dependent manner [21]. CSR3 cleaves ds-siRNA, suppresses sense-mediated RNAi, and counteracts antiviral RNAi in plants [24]. The RNAseIII of HvAV-3e also cleaves ds-siRNA [22]. However, it is not known whether the iridovirus RNaseIII can suppress RNAi, and therefore we compared RNAi suppression potential between the Pike-perch iridovirus (PPIV) Class 1 RNaseIII (PPR3) and CSR3 in plant and animal tissues (Fig. 1A). We were also interested to find out whether these proteins have broad spectrum of activity allowing suppression of RNAi in both animal and plant kingdoms. Our results reveal that the viral Class 1 RNaseIII enzymes have conserved functions in RNAi suppression, making it possible to identify this class of RNA suppressors using bioinformatics approaches, but the spectrum of unrelated organisms in which they are active differs.
The ability of PPR3 to suppress sense-mediated RNAi in Nicotiana benthamiana was tested using an agroinfiltration assay of leaves of transgenic N. benthamiana (line 16c) that constitutively expressed the jellyfish green fluorescent protein (GFP) under the Cauliflower mosaic virus 35S promoter [24–26]. Leaves were co-infiltrated with a liquid culture of Agrobacterium tumefaciens engineered with a 35S-GFP transgene and A. tumefaciens expressing 35S promoter—driven PPR3, CSR3, or GUS (β-glucuronidase; negative control). Consequently, GFP fluorescence and gfp mRNA level were initially enhanced but decreased substantially to the level of the constitutive expression of the gfp transgene in the leaves co-infiltrated to express GFP and GUS, as expected and consistent with sense-mediated silencing of gfp (Fig. 1B). In contrast, gfp mRNA level and GFP fluorescence increased and remained high by 3 days post-infiltration (d.p.i.) in leaf tissues co-infiltrated to overexpress GFP and PPR3 or CSR3 (Fig. 1B), consistent with suppression of gfp silencing. The accumulation of gfp mRNA-derived siRNA correlated inversely with gfp mRNA accumulation, as expected (Fig. 1B).
The endoribonuclease signature motif of Class I RNaseIII enzymes is conserved, and the structure of the catalytic domain of E. coli Class 1 RNaseIII and the amino acid residues critical for catalytic activity have been elucidated [22]. We have shown that when the corresponding critical residues are replaced with alanine in CSR3 (E37A and D44A; mutant CSR3-Ala), the RNaseIII and RNAi suppression activities of CSR3 are abolished [24]. In the current study, the corresponding mutations (E44A and D51A) were introduced to the endoribonuclease signature motif of PPR3 to yield the mutant PPR3-Ala (Fig. 1A). PPR3, PPR3-Ala, CSR3, and CSR3-Ala were expressed in E. coli, and the His-tagged recombinant proteins were purified (Fig. 2A). CSR3 and PPR3 processed long dsRNA (Fig. 2B, lanes 3 and 5, respectively) and synthetic ds-siRNA in vitro (Fig. 2C). While PPR3-Ala retained endoribonuclease activity for long dsRNA despite of the two mutations (Fig. 2B, lane 6), it failed to cleave ds-siRNA in vitro (Fig. 2C). In contrast, CSR3-Ala could not process either long dsRNA (Fig. 2B, lane 4) or ds-siRNA (Fig. 2C) as previously [24]. When GFP was co-expressed with CSR3-Ala or PPR3-Ala in leaves of N. benthamiana 16c as above, only constitutive or slightly higher expression levels, respectively, were observed at 3 d.p.i. (Fig. 1B), indicating that the endoribonuclease activities of CSR3 and PPR3 were needed to protect gfp mRNA from degradation. Results with PPR3 suggested that cleavage of ds-siRNA was particularly important for suppression of RNAi.
PPR3 and CSR3 were tested for suppression of dsRNA (hairpin RNA)-induced gene silencing. The agroinfiltration assay was carried out as above, except that an A. tumefaciens strain expressing hairpin-gfp RNA was used instead of GFP. PPR3 was able to suppress dsRNA-induced silencing of gfp, similar to the p22 RNAi suppressor protein of SPCSV [21] (Fig. 1D). In contrast, PPR3-Ala, CSR3, and CSR3-Ala failed to interfere with dsRNA-mediated silencing (Fig. 1D).
Certain viral RNAi suppressors cannot reverse RNAi after RNAi is initiated [18]. To test this aspect with PPR3 and CSR3, sense-mediated silencing of gfp was induced in leaves of N. benthamiana line 16c by agroinfiltration as above. After 24 h, the same leaves were agroinfiltrated for expression of PPR3, PPR3-Ala, CSR3, or CSR3-Ala. Expression of PPR3 enhanced GFP fluorescence in the leaves at 3 d.p.i., similar to the helper component proteinase (HCpro) RNAi suppressor of plant potyviruses (family Potyviridae) (Fig. 1E) known to reverse silencing [25]. In contrast, PPR3-Ala, CSR3, and CSR3-Ala failed to reverse gfp silencing (Fig. 1E).
Owing to systemic spread of silencing in C. elegans, large numbers of the nematodes actively performing RNAi can be obtained by feeding them bacteria engineered to express high levels of a specific dsRNA [6]. Because CSR3 suppressed only sense-mediated RNAi in N. benthamiana, however, an E. coli strain engineered to express high levels of gfp mRNA was used as an RNAi inducer in C. elegans. These bacteria were fed to four transgenic strains of C. elegans expressing gfp under different tissue-specific promoters (Fig. 3A, S1 Table). gfp silencing was observed in all four strains, whereas no detectable reduction of GFP fluorescence was observed in nematodes fed bacteria transformed with an empty (no insert) plasmid or with a promoterless plasmid as tested with strain RT476 (S1A,B Fig.). The greatest (ca. 5-fold) reduction of GFP fluorescence was observed in strain RT476 (Fig. 3A, S1B Fig.) expressing gfp under the intestine-specific promoter vha-6 [27]. Strand-specific reverse transcription—PCR (RT-PCR) detected exclusively gfp sense transcripts (mRNA) in the gfp-transformed bacteria, and no antisense gfp transcripts were detected (S1C Fig.), suggesting that efficient sense-mediated silencing of gfp could be achieved in intestine tissue of C. elegans. Hence, strain RT476 was chosen for use in the RNAi suppression experiments.
Strain RT476 was stably transformed with a gene encoding PPR3, PPR3-Ala, CSR3, or CSR3-Ala placed under the heat shock—inducible promoter mtl-2 [28]. Two independent transgenic lines of the same strain expressing each protein were used for the experiments. Silencing of gfp was induced by feeding the nematodes with bacteria expressing gfp mRNA, and 72 h later expression of the viral protein was induced by heat shock. GFP fluorescence was observed 24 h after inducing viral protein expression. In four independent experiments, PPR3 restored GFP expression, as indicated by GFP fluorescence intensity that was similar to the gfp-transgenic strain RT476 fed bacteria containing a plasmid lacking an insert (control; Fig. 4). PPR3-Ala partially restored GFP expression, with GFP fluorescence intensity attaining ~75% of the level measured in the control. In contrast, no enhancement of GFP fluorescence was observed in the gfp-silenced nematodes that were transformed with PPR3 or PPR3-Ala but not induced to express these proteins (Fig. 4).
Expression of CSR3 or CSR3-Ala did not result in recovery of GFP fluorescence in gfp-silenced C. elegans, as observed 24 h post-induction of the viral protein expression (Fig. 5A). When CSR3 and CSR3-Ala were expressed as a fusion product with the red fluorescent protein dTomato [29] in C. elegans strain RT476, the readily detectable red fluorescence occurring 24 h after induction indicated high levels of protein expression. As observed before, GFP fluorescence did not recover in the four independent experiments (Fig. 5B). Immunoblotting revealed high amounts of dTomato-CSR3 in the nematodes by 3 h post-induction followed by a gradual decline (Fig. 5C), which is consistent with previous studies carried out using the same promoter [30, 31]. Taken together, these results suggested that CSR3 was unable to reverse sense-mediated silencing in C. elegans.
CSR3 was tested for interference with induction of sense-mediated RNAi by expressing CSR3 simultaneously with induction of gfp silencing in C. elegans. In two independent experiments, GFP fluorescence declined similarly with time (0 to 72 h) irrespective of whether the nematodes were induced to express CSR3 (Fig. 5D), providing no evidence of obstruction with RNAi.
We tested CSR3 and PPR3 also for their ability to interfere with antisense-mediated RNAi of LacZ in the S2 cells of D. melanogaster. The LacZ gene was co-expressed in sense and antisense orientations from two plasmids, which induced antisense-mediated silencing against the gene. Analysis of proteins extracted from the cell cultures 72 h post-induction showed that antisense-mediated silencing significantly reduced LacZ expression, as compared with controls including co-expression of LacZ and the luciferase gene (luc), or co-expression of sense LacZ transcripts from two plasmids (Fig. 6). However, transfection of S2 cells with an additional plasmid expressing CSR3, PPR3, or the mutant CSR3-Ala or PPR3-Ala did not interfere with LacZ silencing and enhance LacZ expression in D. melanogaster (Fig. 6). These results indicated that CSR3 and PPR3 were unable to suppress antisense-mediated RNAi in D. melanogaster.
The transgenic C. elegans strain 123 contains a heat shock-inducible Flock House virus (FHV, family Nodaviridae) RNA1 replicon termed FR1gfp [32, 33]. In FR1gfp, the coding sequence of FHV B2, an RNAi suppressor, is replaced with GFP coding sequence, making the virus unprotected against RNAi [32]. However, in the presence of an RNAi suppressor, the replication of FR1gfp is restored to produce green fluorescence, making 123 an ideal strain to identify viral RNAi suppressors in C. elegans [32, 33]. The C. elegans strain 123/FHVB2 contains both the FR1gfp replicon transgene and a heat inducible FHV B2 transgene which is able to rescue FR1gfp replication and thus GFP expression in pharynx and muscle cells 24 h post heat shock (Fig. 7A, positive control) [33]. The C. elegans strain 123/rde-4 contains a null allele of rde-4. Because rde-4 encodes a dsRNA-binding protein that plays an essential role in RNAi, FR1gfp replication and GFP production in 123/rde-4 are restored (Fig. 7A, positive control) [32]. We transformed worms of strain 123 with the genes for CSR3, CSR3-Ala, PPR3 and PPR3-Ala placed under the heat-shock—inducible promoter mtl-2 and included two independent progeny lines of each in further experiments (lines 123/CSR3, 123/CSR3Ala, 123/PPR3 and 123/PPR3Ala, respectively). Strong GFP fluorescence was observed in the intestine of the progeny of line 123/PPR3 at 24 h post-heat shock, indicating suppression of RNAi-based antiviral defense (Fig. 7A). The heat shock-treated worms of line 123/PPR3 had stunted growth and fewer progeny were obtained than for other transgenic lines, suggesting harmful physiological effects of PPR3. GFP expression also was detected in the progeny of lines 123/FHVB2 and 123/red-4, as expected (Fig. 7A). The remaining lines displayed no GFP fluorescence, but faint signals of the pharyngeal cyan fluorescent protein used as a co-injection marker (e.g., line 123/CSR3Ala in Fig. 7A). The heat-induced expression of transgenes was verified by RT-PCR in all transgenic lines (Fig. 7B).
CSR3 suppresses antiviral RNAi efficiently in plants [24], but whether PPR3 is able to do the same was tested in this study. Tobacco rattle tobravirus (TRV) has a bipartite (+)ssRNA genome and infects a wide range of plant species. TRV RNA1 contains coding sequences for a replicase and two RNAi suppressors, namely 16K, the main suppressor of RNAi [34–36], and 29K that also acts as a viral cell-to-cell and long distance movement protein [37]. TRV RNA1 can infect plants systemically without RNA2, but RNA2 enhances accumulation of RNA1 in plant tissues by an unknown mechanism [37]. TRV-M1 is a TRV RNA1 replicon lacking the coding sequence of 16K and is only weakly protected against RNAi [37]. One half in the full-grown leaves of N. benthamiana was agroinfiltrated for co-expression of TRV-M1 and PPR3, or TRV-M1, TRV RNA2 and PPR3. The other half of the leaf was agroinfiltrated with controls, namely TRV-M1 only, or with TRV-M1 and TRV RNA2, respectively. Samples of infiltrated leaf tissue had to be collected at 3 d.p.i., because longer co-expression of TRV-M1 and PPR3 induced extensive necrosis of the leaf tissue. The concentrations of TRV RNA1 were estimated by quantitative reverse transcription PCR, which showed that TRV-M1 amounts were enhanced in the presence of PPR3, as compared with the controls (S2 Fig.). Although the necrotic symptoms enforced an early termination of the trials and consequently the differences were not statistically significant, there was a clear and consistent tendency in all three experiments of enhanced TRV RNA1 accumulation in the presence of PPR3.
The cellular RNaseIII—like endoribonucleases belonging to Class 3 (called Dicers) play a key role in RNAi, which functions as a non-virus-specific, basal antiviral defense mechanism in plants [1] and invertebrates such as C. elegans [2, 3, 38] or insects including D. melanogaster [39, 40]. Recent studies suggest also an antiviral role of RNAi in vertebrates [41, 42]. Dicers recognize long dsRNA molecules, including those of viral origin, and cleave them to yield 21- to 25-nt siRNAs pivotal in targeting RNAi to silence the homologous viral genomes. PPIV and SPCSV are unrelated viruses differing in genome structure and gene content. PPIV belongs to iridoviruses (family Iridoviridae) containing a dsDNA genome that typically codes for ca. 100 proteins [19]. SPCSV, in turn, has a bipartite (+)ssRNA genome encoding up to 12 proteins [43]. However, common to PPIV and SPCSV are the Class 1 RNaseIII homologs (PPR3 and CSR3, respectively) produced by both viruses. dsRNA-degrading activity develops in the host cells at an early stage of infection with Frog virus 3, the type species of genus Ranavirus (Iridoviridae) encoding a homolog of RNaseIII [19], and the RNaseIII activity of a homologous protein of rock bream iridovirus was recently reported [20]. In SPCSV, the subgenomic RNA coding for CSR3 is expressed early at infection of plants [43]. The early activation of RNaseIII expression is consistent with its role in interferences with antiviral RNAi, which is supported by our results. They show that PPR3 suppresses antiviral RNAi in C. elegans and N. benthamiana and hence enhances accumulation of the RNAi suppression deficient FHV and TRV replicons, respectively. Furthermore, our previous studies have shown that RNAi is suppressed and antiviral defense against many unrelated viruses is eliminated in virus-resistant sweet potato plants transformed to express CSR3 [24]. Finally, it is remarkable that PPR3 (this study) and CSR3 [24] suppress antiviral RNAi autonomously, in the absence of PPIV or SPCSV infection, respectively, which excludes a role for other viral proteins, or perturbation of cellular homeostasis caused by virus infection, in the observed interference with RNAi.
PPIV and SPCSV have unrelated host ranges including poikilothermic animals and plants, respectively. Class 1 RNaseIII—like proteins exhibiting endoribonuclease activity on dsRNA have been reported in Paramecium bursaria chlorella virus that infects algae [44], Diadromus pulchellus ascovirus [45] and Heliothis virescens ascovirus (HvAV-3e) [22] that are DNA viruses infecting insects, and rock bream iridovirus that infects fish [20]. Involvement of the HvAV-3e RNaseIII in suppressing RNAi has been suggested based on results showing that the dsRNA-induced silencing of gfp was suppressed in an insect cell line (Spodoptera frugiperda Sf9) infected with a baculovirus engineered to express HvAV-3e RNaseIII [22]. Also, many other viruses in the family Iridoviridae that infect poikilothermic animals, insects, or other invertebrates [19] and the viruses of family Ascoviridae that infect larvae of lepidopteran insects [46] have genes predicted to encode Class 1 RNaseIII—like proteins containing the conserved motifs of Class 1 RNaseIIIs, but the overall amino acid sequence similarity is low [22]. Taken together, interference with RNAi has now been demonstrated with RNaseIIIs of SPCSV, PPIV and HvAV-3e representing unrelated taxa of RNA and DNA viruses. These results suggest that, in general, viral RNaseIII endoribonucleases function as suppressors of RNAi. They represent the first group of RNAi suppressors that are homologous in unrelated viruses.
Mutations in the highly conserved RNase signature motif [22] in CSR3-Ala and PPR3-Ala prevented cleavage of ds-siRNA by both enzymes, but PPR3-Ala could still cleave long dsRNA substrates and partially suppress sense-mediated silencing in N. benthamiana and reverse silencing in C. elegans. PPR3 and CSR3 differ greatly for the size (371 and 229 residues, respectively) and amino acid sequence, but share conserved domains characteristic of Class 1 RNaseIII enzymes. Whereas CSR3—with a single dsRNA-binding domain—resembles the Class 1 RNaseIII of E. coli [21, 22], PPR3 contains two putative dsRNA-binding domains and resembles AtRTL2 of Arabidopsis thaliana [47]. On the other hand, little sequence variability is observed among the RNaseIII proteins of the different iridoviruses or between the isolates of SPCSV. The RNaseIII proteins of PPIV (NCBI sequence database accession KC191670), Frog virus 3 (AY548484), Rana grylio iridovirus (JQ654586), Soft-shelled turtle iridovirus (EU627010.1), Andrias davidianus ranavirus isolate 2010SX (KF033124) and Chinese giant salamander iridovirus (KF512820) vary only at three amino acid positions that are not situated in the RNA-binding and catalytic domains. The deduced CSR3 amino acid sequences characterized in 69 isolates of SPCSV were 98.7–100% identical and no sequence variability was observed in the RNA-binding and catalytic domains [48]. Therefore, PPR3 and CSR3 used in this study stood for the typical sequences of the respective types of RNaseIII enzymes. The difference in RNAi suppression, however, may be explained by the two dsRNA-binding domains of PPR3 as compared with a single dsRNA-binding domain in CSR3. Two dsRNA-binding domains may allow more efficient sequestering of ds-siRNA, and/or competition with Dicer/AGO for binding long dsRNA, which could be significant because siRNA binding is a common mechanism by which viral proteins suppress RNAi [17, 49]. Despite these differences, results with PPR3 and CSR3 were consistent in that the similar mutations introduced to the RNase signature motif abolished cleavage of ds-siRNA and debilitated RNAi suppression, suggesting that cleavage of ds-siRNA is required for efficient suppression of RNAi. Cleavage products of Class 1 RNaseIII are shorter (~15 bp) than those generated by Dicers and are not functional in RNAi [50, 51]. The suggested RNAi suppression activity of the HvAV-3e RNaseIII may function similarly because it cleaves 21-nt ds-siRNA [22].
Our results suggest that PPR3 is more versatile than CSR3 for suppressing RNAi. Whereas PPR3 and CSR3 both suppressed sense-mediated RNAi in plant tissue, PPR3 also suppressed dsRNA-mediated silencing. Furthermore, in contrast to CSR3, PPR3 was able to reverse silencing in plant tissue and C. elegans. These results suggest a different level of host specificity in the action of CSR3 and PPR3. However, both PPR3 and CSR3 failed to suppress RNAi in D. melanogaster. Previous studies have shown that the non-homologous proteins P15, P19, and P21 encoded by unrelated plant RNA viruses, the B2 protein of FHV (insect virus), and the 1A protein of Drosophila C virus suppress dsRNA-mediated RNAi in D. melanogaster [52]. All these plant viral proteins may suppress RNAi by binding and sequestering virus-derived siRNAs, which has been particularly well characterized for P19 [53]. Additionally, B2 binds dsRNA, which 1A may bind preferably [52]. These results are consistent with the conclusion that RNAi suppression exhibited by the RNaseIII enzymes was connected to their catalytic activity on dsRNA and ds-siRNA rather than RNA binding, which is not efficient enough to suppress RNAi. Furthermore, while the RNAi pathways in plants, C. elegans, and D. melanogaster conferring resistance to viruses [1–3, 38, 40, 54] include conserved components, there are also differences. For example, the RdRp mediated amplification of RNAi involves distinct mechanisms in plants and animals [12, 15, 55]. Non-optimal pH [56] and possible incompatible interactions between the viral silencing suppressors and cellular proteins [49] may affect catalytic activity of CSR3 and PPR3 and explain the differences in suppression of RNAi in a non-host cellular environment. Incompatible host interactions may also cause the necrotic response observed in the leaves co-expressing PPR3 and the TRV replicon.
Taken together, our results suggest a hypothesis in which the viral Class 1 RNaseIII may compete with Dicer for processing the long virus-derived dsRNA and AGO for binding ds-siRNAs, but especially destroys the ds-siRNA to prevent loading of RNA-induced silencing complexes (RISC), production of secondary siRNAs and amplification of RNAi [4, 12, 16, 32, 57]. The versatile functions of PPR3 in plants and animals suggest that it is less dependent on host factors than CSR3. Furthermore, the results of our study show that the viral Class 1 RNaseIIIs are unique among viral RNAi suppressors because they are homologous in unrelated RNA and DNA viruses and can be detected in viral genomes using gene modeling and protein structure prediction programs. Our results underscore the importance of their catalytic activities in suppressing RNAi. Understanding of this novel mechanism of RNAi suppression may inform means of controlling the diseases and economic losses which the RNaseIII-containing viruses cause in animal and plant production [58–60].
The binary plasmids containing the RNaseIII gene (CSR3) of SPCSV and the corresponding mutated gene (CSR3-Ala) have been described [21, 24]. In the binary vector (pKOH200), the coding region is flanked by the Cauliflower mosaic virus 35S promoter and the 3′ terminator region (3’ nos) of the nopaline synthase gene.
PPIV-infected tissue was obtained from newly hatched pike-perch (Stizostedion lucioperca) provided by H. Tapiovaara, Finnish Food Safety Institute Evira, Finland. Multiplication and isolation of the virus were carried out in a blue fry gill (BF-2) cell line [61]. Cell culture supernatants were collected and the DNA purified using the QIAamp DNA Mini kit (Qiagen). The RNaseIII ORF (PPR3) was PCR-amplified from PPIV using primers PPR3 NotI fwd and PPR3 FseI +StrepII rev (S1 Table) designed based on the flanking conserved genome regions identified by comparison with the genome sequences of Frog virus 3 (FV-3; GenBank AY548484) [62], Tiger frog virus (AF389451), and Ambystoma tigrinum virus (NC_005832) [63]. The amplified PCR product was treated with NotI and FseI, which allowed cloning of PPR3 in an intermediate E. coli plasmid (pKOH122) containing the 35S promoter and 3′nos sequences [64]. The gene was sequenced (GenBank as accession no. KC191670). PPR3 was mutated using the QuickChange Site-Directed Mutagenesis System (Stratagene) with primers PPR3-Ala fwd and rev (S1 Table) to introduce two amino acid substitutions (E44A and D51A) into the conserved catalytic site and express the mutant PPR3-Ala analogous to CSR3-Ala [21].
The cassettes ‘35S-PPR3-NosT’ and ‘35S-PPR3-Ala-NosT’ were excised from pKOH122 and subcloned into the binary vector pKOH200 for plant transient expression [64], resulting in plasmids pKOH200 PPR3 and pKOH200 PPR3-Ala. For immunodetection, a StrepII-tag (amino acid sequence WSHPQFEK) was fused with the C-terminal end of both PPR3 and PPR3-Ala.
For RNaseIII expression in E. coli, the full-length ORFs were amplified from the respective pKOH200 vectors (see above) and cloned fused with a C-terminal His-tag in pET11d+ vector as described [21, 24]. Amplification and cloning of the genes PPR3 and PPR3-Ala followed a similar procedure except an N-terminal His-tag fusion was used. The vectors (pET11d+ His-PPR3, pET11d+ His-PPR3-Ala, pET11d+ CSR3-His, pET11d+ CSR3-Ala-His) were transformed into E. coli strain BL21 (DE3 PLUS RIL; Qiagen) and expressed (for details, see below).
The coding regions for PPR3 and PPR3-Ala were amplified from pET11d+ His-PPR3 and pET11d+ His-PPR3-Ala using primers 54_01 PPR3 XbaI fwd and 54_01 PPR3 NheI rev (S1 Table). The PCR products were digested with XbaI and NheI and introduced in the corresponding sites of vector pPD54_01 to generate pPD54_01 PPR3 and pPD54_01 PPR3-Ala, respectively.
The full-length CSR3 and the CSR3-Ala coding regions were amplified from pET11d+ CSR3-His and pET11d+ CSR3-Ala-His, respectively, using primers 54_01 CSR3 XbaI fwd and 54_01 CSR3 XmaI rev (S1 Table). The constructs were cloned downstream of the mtl-2 promoter of the pPD54_01 expression vector (Plasmid 1507; Addgene, Cambridge, MA) using the XbaI and XmaI sites, resulting in plasmids pPD54_01 CSR3 and pPD54_01 CSR3-Ala. In addition, CSR3 and CSR3-Ala fusion proteins were generated using the dTomato tag [29]. Briefly, amplification of the dTomato coding sequence from plasmid pRSET dTomato was carried out with the primer pair 54_01 dTomato-CSR3 XbaI fwd and 54_01 dTomato-CSR3 XbaI rev (S1 Table). The PCR products were digested with XbaI and ligated in the corresponding sites of pPD54_01 CSR3 and pPD54_01 CSR3-Ala. All constructs were verified by sequencing.
The promoterless construct pET24b+ GFPopt—ΔT7 was obtained by digesting pET24b+ GFPopt with EcoRI and SgrAI. The insert was omitted, and the plasmid was blunt-ended with Klenow polymerase (NEB) and religated using T4 ligase (NEB). This promoterless plasmid was used as a negative control for gene expression in the transformation of C. elegans. All constructs and mutations were verified by sequencing (Haartman Institute, DNA sequencing unit, University of Helsinki).
Strains of C. elegans were maintained as described [65]. For analyzing the silencing effects in different tissues, four different strains were employed (S1 Table). For silencing suppression experiments, the C. elegans strain RT476 Is[Pvha-6:gfp::RAB-7] was used for stable transformation. In this strain, gfp was constitutively expressed under the intestine-specific promoter vha-6 and marked the RAB-7-positive endosomes in C. elegans intestinal cells. Vector pPD54_01 (Addgene) was employed for the stable germline transformation allowing heat shock—inducible expression of the protein in intestine. The constructs pPD54_01 CSR3, pPD54_01 CSR3-Ala, pPD54_01 PPR3, and pPD54_01 PPR3-Ala were stably introduced to the nematodes through germline transformation using the standard microinjection method [66] delivering 50 ng/μl of the construct of interest. In all experiments, 50 ng/μl pRF4 plasmid was co-injected. pRF4 harbors the dominant rol-6 (su1006) allele that causes a readily distinguishable roller phenotype in transgenic animals and serves as a co-transformation marker [67]. The transgenic strains FR1gfp (denoted as 123/N2), 123/FHVB2 and 123/rde-4 of C. elegans have been described [32, 33]. 123/N2 animals were injected with pPD54_01 CSR3, pPD54_01 CSR3-Ala, pPD54_01 PPR3 or pPD54_01 PPR3-Ala (100 ng/μl). The lines are denoted as 123/CSR3, 123/CSR3Ala, 123/PPR3 and 123/PPR3Ala, respectively. The plasmid encoding Pmyo-2::cfp pharyngeal CFP [68] was co-injected as an injection marker (50 ng/μl). Transgenes were maintained as extra-chromosomal arrays, and two independent lines carrying each transgene were isolated and analyzed.
E. coli strain HT115 was transformed with pET24b+GFPopt or with (i) the pET24b+ empty vector (Novagen) or (ii) pET24b+ GFPopt—ΔT7 as controls. In pET24b+ GFPopt, gfp was inserted between the T7 promoter and terminator sequences. Transcription was induced in HT115 by growing the bacteria on nematode growth medium plates containing 1 mM IPTG and appropriate antibiotics (as detailed below) leading to the production of sense-gfp transcripts.
RNAi was initiated by feeding nematodes with sense-gfp-expressing bacteria. For RNAi suppression experiments, Is[Pvha-6:gfp::RAB-7] (S1 Table) and RNaseIII transgenic animals were used. To synchronize the stage of the examined individuals, 4–6 adult nematodes were transferred to each feeding plate for 6 h. The hatched eggs grew to adults within 3 days, and heat shock was carried out by incubating the nematodes 2–3 h at 33–34°C (BioRAD Mini Incubator). GFP fluorescence was recorded at different times after heat shock induction, depending on the experiment. All lines of the transgenic nematode strains were treated and tested in duplicate in each assay in at least three independent experiments.
To test suppression of RNAi, expression of PPR3, PPR3-Ala, CSR3, CSR3-Ala and FHV B2 was induced by heat shock of worms at the temperature mentioned above, as described [32]. The experiment was carried out twice.
Static microscopic images were acquired using an OLYMPUS AX70 microscope with a 20× objective. The fluorescence signal was quantified and analyzed using ImageJ software (NIH). The total pixel intensity in wild-type worms was set to an arbitrary fluorescence unit of 1.0 to enable comparison with other strains. Representative images of the C. elegans intestine were taken with 488 nm excitation (emission 520 nm) for quantification of GFP fluorescence at different time points and with 568 nm excitation (emission 595 nm) for visualization of dTomato fluorescence. Imaging data analysis was conducted using IGOR Pro (Wavemetrics) or EXCEL (Microsoft) software. The mean value ± S.E.M. of the indicated experiments was calculated. Statistical significance was evaluated using the Student’s t-test.
In the study of suppression of RNAi, worm progeny were imaged 24 h post-heat shock with a Zeiss Axioplan 2 microscope using a 10× 0.3 NA objective.
E. coli strain HT115 harboring empty pET24b+ or pET24b+ GFPopt was grown in liquid Luria Bertani medium containing 50 mg/l tetracycline and 50 mg/l kanamycin overnight at 37°C. Thereafter, 200 μl of the culture was used to inoculate 20 ml of fresh medium and grown until OD600 = 0.6. Transcription under the T7 promoter was then induced by addition of 0.1 mM IPTG for 5 h at 37°C. Bacteria were harvested and RNA extracted using the RNeasy Midi kit (Sigma-Aldrich) following the manufacturer’s instructions for total RNA isolation from Gram-negative bacteria. Initial lysis of cells was achieved in 0.25 M Tris-HCl containing 5 mM EDTA (pH 8) including 1 mg/ml lysozyme (Sigma-Aldrich). After elution from the RNA-binding column, DNA was removed by treatment with RNase-free DNase (Promega). After heat-deactivation of DNase, strand-specific cDNA was produced by employing specific primers able to anneal to gfp mRNA either for synthesis of sense (pET24b+ EcoRI rev primer, S1 Table) or antisense (pET24b+ XhoI fwd primer, S1 Table) transcripts. RT-PCR was done using M-MLV-H-mutant reverse transcriptase (Promega) at 55°C. PCR (38 cycles) using Dynazyme II DNA polymerase (Thermo Fisher) was carried out at 57°C with primers pET24b+ XhoI fwd and pET24b+ EcoRI rev to detect sense and antisense strands.
Expression of transgenes was tested in C. elegans by RT-PCR. A total of 15–30 progeny of the transgenic line were collected following heat shock and stored at -80°C. Total RNA was isolated using a Trizol-based extraction method [69]. cDNA was synthesized using random oligo (dT)18 primers (Maxima First Strand cDNA Synthesis kit, Thermo Scientific), and 100 ng was used as template in PCR. Primers rtPPR3fwd (5′-TTGGTTGGGAAACTTGCTCG-3′) and rtPPR3rev (5′-CACTCTTGGGCGTAAACACC-3′) were used to detect the transcripts of PPR3 and PPR3-Ala (amplicon size 101 bp), whereas primers rtCSR3fwd (5′-GAGAATCGTTGGTTGGTTGG-3′) and rtCSR3rev (5′-GGGCAGGTTTCTTAATGTGG-3′) were used to detect the transcripts of CSR3 and CSR3-Ala (amplicon size 107 bp). Expression of the transgene FHVB2 was tested with primers rtB2fwd (5′-ACAACCACGCCACATAACAC-3′) and rtB2rev (5′-GACCATCATCACCGCACTTC-3′) (amplicon size 127 bp). Actin mRNA was detected as an internal control using primers targeting exon 1 (act-1 fwd. 5′-TCGGTATGGGACAGAAGGAC-3′ and act-1 rev. 5′-CATCCCAGTTGGTGACGATA-3′) (amplicon size 108 bp). PCR products were verified by sequencing.
Bacteria carrying the expression plasmid were grown and recombinant proteins expressed as described in The QiaExpressionist (Qiagen). Briefly, protein expression was induced by adding 0.1 mM IPTG in the culture medium growing bacteria overnight at 16°C. Bacterial cells were lysed using lysis buffer (50 mM Na2H2PO4, pH 8.0, 300 mM NaCl, 10 mM imidazole) supplemented with a protease inhibitor cocktail [Roche; 2.5 ml of stock solution (1 tablet/10 ml lysis buffer) in 10 ml extract] and lysozyme (Sigma-Aldrich; 1 ml of 10 mg/ml stock solution in 10 ml buffer) and incubated for 2 h on ice. Cells were additionally disrupted and nucleic acids degraded using sonication (50% duty cycle, 5 × 15 sec; SONIFIER, Branson, Cell Disruptor B15, Hielscher). Recombinant proteins were purified by affinity chromatography with Ni-nitrilotriacetic acid agarose (Ni2+-NTA) beads (Qiagen) according to the manufacturer’s protocol and finally loaded onto polypropylene columns (Qiagen). Wash buffer (50 mM Na2H2PO4, pH 8.0, 300 mM NaCl) containing increasing concentrations of imidazole (20–50 mM) was used to obtain pure protein. Bound proteins were eluted with strong elution buffer (50 mM Na2H2PO4, pH 8.0, 300 mM NaCl, 500 mM imidazole). Fractions containing high concentrations of pure protein were collected, 25% (v/v) glycerol was added, and the purified proteins were stored at -20°C for later use. The Bradford colorimetric method (Protein Assay, Dye Reagent concentrate, Bio-Rad) was used for protein quantification. Coomassie Brilliant Blue reagent [10% (v/v) glacial acetic acid, 40% (v/v) methanol, 1% (w/v) Coomassie Brilliant Blue G] was used to visualize proteins in SDS-PAGE gels.
The protein extraction method to isolate recombinant proteins from transgenic C. elegans for detection by western blot analysis was optimized to prevent protein degradation. CSR3 was prone to degradation at low pH for protein extraction from crude extracts of C. elegans. Protease inhibitors have proved ineffective in preventing degradation, for example with aspartic (acidic) proteases exhibiting pronounced activity at low pH in C. elegans extracts [70]. However, stable buffering of the extraction environment at pH 11 hindered protein degradation during extraction. The extraction buffer contained 2% SDS and 100 mM N-cyclohexyl-3-aminopropanesulfonic acid (Sigma) adjusted to pH 11 with 2 M NaOH.
After heat shock induction of protein expression, samples of ~1000 nematodes were collected at different time points (0 to 24 h) and proteins extracted using 25 μl of the extraction buffer and incubation at 100°C with occasional vortexing. After 5 min of centrifugation at13,000× g at room temperature, the supernatant was subjected to analysis with SDS-PAGE (12% acrylamide; running buffer: 0.025 M Tris-HCl, 0.192 M glycine, 0.1% SDS). Separated proteins were electrotransferred to a PVDF membrane (Amersham Hybond-P, GE healthcare) at 70 V (300 mA) in buffer containing 0.0275 M Tris-HCl, 2.4 M glycine, and 20% (v/v) methanol. Immunodetection was conducted with primary polyclonal anti-CSR3 as described [1]. To eliminate nonspecific cross-reactions of the polyclonal antibodies, 10 ml of 2.5% (w/v) BSA in Tris-buffered saline containing 0.1% (v/v) Tween-20 (which included the primary antibody diluted 1:500) was incubated with 1% (w/v) wild-type nematode extract at 4°C for 1 h with shaking. A horseradish peroxidase—conjugated donkey anti-rabbit IgG (GE Healthcare) secondary antibody was used.
Ectopic expression of proteins in plants was confirmed by crushing leaf material (200 mg) in liquid nitrogen and boiling in the presence of 200 μl 2× protein sample buffer for SDS-PAGE [50 mM Tris-HCl (pH 6.8), 100 mM DTT, 2% SDS, 0.005% bromophenol blue, and 10% (v/v)] glycerol]. Protein extract (20 μl) was subjected to SDS-PAGE (12% acrylamide) and transferred to a Hybond-P membrane by electroblotting. Anti-CSR3 and anti-StrepTagII (Stratagene) were used as primary antibodies, respectively, together with a 1% (w/v) extract of healthy N. benthamiana leaves to lower nonspecific binding. Secondary ECL anti-rabbit IgG conjugated with horseradish peroxidase was used (Amersham).
Signals were detected via ECL chemiluminescence using the SuperSignal West Pico Chemiluminescent Substrate (Pierce/Thermo Fisher Scientific), and membranes were exposed to film (Kodak).
Long dsRNA was produced according to the manual of Replicator RNAi kit (Finnzymes) using T7 GFP fwd and Phi6 GFP rev primers (S2 Table) based on the gfp sequence using pKOH GFP as a template for PCR, which produced a dsDNA intermediate. The dsDNA was subsequently transcribed and the complementary strand added by using T7 and Phi-6 polymerases, respectively, in a single reaction.
Short dsRNA processing was screened with the synthetic 24-nt gfp ds-siRNA: sense, GAGAGGGUGAAGGUGAUGCUACAC; antisense, GUAGCAUCACCUUCACCCUCUCAG. Equal volumes of sense and antisense RNA strands (Sigma; stock 0.1 μg/ml) were incubated at 85°C for 15 min and subsequently cooled to room temperature. The resultant double-stranded oligos contained typical 3′overhangs, 5′OH, and no further modifications.
The cleavage assays were done at least three times with PPR3, PPR3-Ala, CSR3, and CSR3-Ala using optimized endoribonucleolytic cleavage reactions at 37°C, as described [56]. Samples were analyzed by agarose gel electrophoresis (1% for long dsRNA; 4% for short dsRNA), stained with ethidium bromide and visualized by a UV photoimager (Molecular Imager, Gel Doc XR+, Bio-Rad).
The transgenic N. benthamiana (line 16c) plants expressing GFP [54] were grown in growth chambers (temperature 18–22°C, 70% relative humidity) with a 16-h photoperiod using sodium high-pressure lamp illumination. The binary vectors were transformed into competent A. tumefaciens C58C1 (pGV3850) cells by the freeze and thaw method [71] and agroinfiltration was carried out as previously described [24]. Co-infiltrations [72] were done with a mixture of A. tumefaciens overnight cultures (1:1 ratio) carrying constructs encoding pBIN35S GFP [73] or pKOH200 hpGFP [21] (OD600 = 0.5) and one of the test constructs expressing the wild-type and mutant CSR3 proteins (OD600 = 0.5) or PPR3 proteins (OD600 = 0.1). Agrobacterium cultures for expression of PPR3 had to be diluted more than for expression of other proteins, because PPR3 expression caused necrosis, which was not observed with the PPR3-Ala. Dilution of the Agrobacterium culture to OD600 = 0.1 (rather than only OD600 = 0.5) delayed development of necrosis until 4 d.p.i. A. tumefaciens expressing the β-glucuronidase gene (GUS) was included as a negative control, and the viral silencing suppressors HCpro and p22 [21] were included as positive controls. GFP expression was monitored by epi-illumination using a hand-held UV-lamp (B-100 AP; UVP, Upland, CA) up to 8 d.p.i.
Reversion of silencing was tested by inducing gfp silencing with pBIN35S GFP infiltration in 16c transgenic plants, and expressing viral proteins subsequently (after 24 h) by agroinfiltration in the same leaf spots. Reversion of silencing was monitored up to 6 d.p.i. Images were acquired with a CANON EOS 40D digital camera with an EFS 17–85 mm objective and processed using CorelDRAW Graphics Suite X3 and Adobe Photoshop software.
Total RNA was extracted from leaf tissue using TRIzol LS Reagent (Invitrogen), also allowing the recovery of the siRNA fraction. High- and low-molecular-weight RNA fractions were separated and the quality was determined by electrophoresis. The RNA samples were subjected to northern blot analysis using gfp—specific probes labeled with [32P]UTP as previously described [24]. Radioactive signals on membranes were detected by exposure to X-ray film.
The genes encoding CSR3, CSR3-Ala, PPR3, and PPR3-Ala were amplified from the corresponding binary vectors using Phusion high-fidelity DNA polymerase (Finnzymes) and primers containing appropriate restriction sites for cloning (CSR3 EcoRI fwd, CSR3 XhoI rev, PPR3 EcoRI fwd, and PPR3 XhoI rev; S1 Table). The purified PCR products were cloned into the Drosophila expression vectors pAc5.1 and pMT (Invitrogen) in-frame with a C-terminal V5 (GKPIPNPLLGLDST) and hexahistidine tags. Similarly, LacZ in the antisense or sense orientation and luc (encoding luciferase) in the antisense orientation were cloned into the vectors. All clones were verified by sequencing.
Each RNaseIII plasmid was co-transfected with antisense or sense LacZ or with antisense luc plasmid into Drosophila S2 cells at 2 × 106 cells/ml in 12-well plates with a total of 1 μg of the indicated plasmids using FugeneHD transfection reagent (Roche). Additionally, in some experiments, co-transfection of pMT-EGFP-V5—6xHis plasmid (constructed as above) was used to control transfection efficiency. Expression of the metallothionin promoter of the constructs was induced with 0.6 mM CuSO4 for 3 days, after which the cells were collected by centrifugation and lysed in lysis buffer [Tris-buffered saline pH 7.5, 1% (w/v) Triton X-100, 20 mM NaF, 1 mM EDTA] supplemented with Complete Proteinase Inhibitor Coctail (Roche). Protein concentration of the lysate was measured with Bradford Protein Assay (Bio-Rad). Approximately 10 μg of total protein per sample was subjected to 10% SDS-PAGE followed by immunoblotting and detection with anti-V5 (Invitrogen).
Sequence data from this article can be found in the GenBank database under the following accession numbers: PPR3 (KC191670), SPCSV (AJ428554).
|
10.1371/journal.pgen.1002238 | Regulation of Caenorhabditis elegans p53/CEP-1–Dependent Germ Cell Apoptosis by Ras/MAPK Signaling | Maintaining genome stability in the germline is thought to be an evolutionarily ancient role of the p53 family. The sole Caenorhabditis elegans p53 family member CEP-1 is required for apoptosis induction in meiotic, late-stage pachytene germ cells in response to DNA damage and meiotic recombination failure. In an unbiased genetic screen for negative regulators of CEP-1, we found that increased activation of the C. elegans ERK orthologue MPK-1, resulting from either loss of the lip-1 phosphatase or activation of let-60 Ras, results in enhanced cep-1–dependent DNA damage induced apoptosis. We further show that MPK-1 is required for DNA damage–induced germ cell apoptosis. We provide evidence that MPK-1 signaling regulates the apoptotic competency of germ cells by restricting CEP-1 protein expression to cells in late pachytene. Restricting CEP-1 expression to cells in late pachytene is thought to ensure that apoptosis doesn't occur in earlier-stage cells where meiotic recombination occurs. MPK-1 signaling regulates CEP-1 expression in part by regulating the levels of GLD-1, a translational repressor of CEP-1, but also via a GLD-1–independent mechanism. In addition, we show that MPK-1 is phosphorylated and activated upon ionising radiation (IR) in late pachytene germ cells and that MPK-1–dependent CEP-1 activation may be in part direct, as these two proteins interact in a yeast two-hybrid assay. In summary, we report our novel finding that MAP kinase signaling controls CEP-1–dependent apoptosis by several different pathways that converge on CEP-1. Since apoptosis is also restricted to pachytene stage cells in mammalian germlines, analogous mechanisms regulating p53 family members are likely to be conserved throughout evolution.
| Germ cell apoptosis helps to ensure that only healthy germ cells contribute to the next generation. The C. elegans p53 family member CEP-1 plays an important role in inducing apoptosis in damaged germ cells. CEP-1 protein is maximally expressed in late-stage pachytene cells, which are the only cells of the germline that undergo apoptosis. Restricting CEP-1 to late pachytene cells is thought to ensure that apoptosis does not occur in cells at earlier stages of meiosis where meiotic recombination occurs. Through an unbiased genetic screen, we uncovered a role for the Ras/MAP kinase signaling pathway as a novel regulator of DNA damage–induced, CEP-1–dependent apoptosis. We show that the Ras/MAP kinase pathway is required for DNA damage–induced apoptosis by regulating the expression of CEP-1 in late pachytene cells. In addition, MAP kinase signaling might be directly involved in apoptosis induction, as the pathway that is activated in response to IR and MAP kinase directly interacts with CEP-1. We postulate that p53 family members might be regulated by analogous mechanisms in mammals.
| The p53 family of transcription factors is conserved throughout animal evolution [1], [2]. In vertebrates the founding member, p53, is a key tumour suppressor and is the most commonly mutated gene in human tumours. Two paralogues, p63 and p73, have diverse roles in development and in responding to cellular stress [3]. Based on sequence similarity it appears that the majority of invertebrate p53 family members are most closely related to mammalian p63 and it has been postulated that an ancient function of the p53 family might be the regulation of germ cell apoptosis [4]. The sole C. elegans p53 homologue CEP-1 was implicated in regulating germ cell apoptosis in response to DNA damage and meiotic recombination failure [5], [6]. Interestingly, more recent reports indicate that the TAp63 specific isoform is required to eliminate damaged meiotic germ cells in the mammalian female germline [4].
The C. elegans hermaphrodite germline consists of two U-shaped gonads, in which the germ cells are organised in a gradient of maturation. In the distal part of the germline cells proliferate mitotically before entering meiosis in the transition zone. Cells go through the various stages of meiosis as they progress through the germline. Once they have progressed into diplotene and diakenesis they begin oocyte differentiation. Apoptosis is only observed in cells in the late pachytene stage where homologous chromosomes are synapsed and meiotic recombination has been largely completed. A number of different stimuli can induce apoptosis in the germline and all require the same core apoptotic machinery used during C. elegans somatic development, including the Bcl-2 family member CED-9, which acts to inhibit the Apaf-1 homologue CED-4, that in turn activates the caspase CED-3 [7]. A low background level of CEP-1 independent death, termed physiological apoptosis, is thought to maintain tissue homeostasis in the germline. In contrast, DNA damage induced apoptosis specifically involves CEP-1 activation by the DNA damage response pathway and the subsequent CEP-1 dependent transcriptional induction of the BH3 only (Bcl-2 homology domain 3) gene egl-1. This mechanism is analogous to IR-induced p53 dependent transcriptional induction of NOXA and PUMA in mammals [8], .
The extracellular signal-related kinase ERK is downstream of the MAP kinase signaling pathway that includes the Ras GTPase, and is involved in many aspects of animal development and homeostasis. In C. elegans, LET-60 (the Ras homologue), MPK-1 (the ERK homologue), and several other members of the pathway are conserved and are important for many aspects of somatic and germline development and function. During somatic development this pathway is part of an inductive signal required to specify the fate of the vulva [10]. Within the germline Ras/ERK signaling is involved in germline proliferation, meiotic progression, and oocyte maturation and growth [11], and is also required for physiological apoptosis [12], [13]. MAPK phosphatases (MKPs) are important regulators of this signaling pathway, and function by dephosphorylating and deactivating MAPKs. In C. elegans, genetic studies have implicated the MKP LIP-1 as an inhibitor of MPK-1 signaling in both the vulva and the germline [14]–[16].
The observation that apoptosis only occurs in cells in the late pachytene stage of meiosis indicates that there must be particular signals or regulatory mechanisms that make only these particular germ cells competent for apoptosis and that prevent apoptosis in all other germ cells. Restricting apoptosis to late pachytene stage cells could prevent the inappropriate loss of cells in both the transition zone and the early pachytene stage where SPO-11 dependent double strand breaks are formed [17]–[19] and meiotic recombination occurs [20], respectively. One way to restrict apoptosis to late pachytene cells is via control of CEP-1 expression in the germline. We previously reported that GLD-1 represses the translation of CEP-1 in early stage meiotic cells and that CEP-1 expression gradually increases as GLD-1 levels decrease in late pachytene [21]. It is likely that further developmental signals are also involved in establishing apoptotic competency, possibly by regulating entry into late pachytene, or regulating the expression of CEP-1 or other apoptotic factors. One such developmental signal is likely to be mediated by MPK-1 activation, which is required for entry into late pachytene [11].
Here we report that MPK-1 signaling regulates CEP-1 dependent, DNA damage induced apoptosis. Using an unbiased genetic screen we found that excessive MAP kinase signaling, conferred by mutations of the MAP kinase phosphatase LIP-1 and by an activating allele of Ras, leads to excessive DNA damage dependent germ cell apoptosis. Conversely, the absence of MPK-1 inhibits DNA damage induced apoptosis. We provide evidence that MPK-1 signaling acts developmentally to regulate apoptosis competency by controlling CEP-1 expression levels in late pachytene cells. Furthermore, we show that MPK-1 signaling is triggered by IR, and that this might directly activate CEP-1.
We previously implicated the translational repressor GLD-1 as a negative regulator of CEP-1 via a genetic screen for mutants showing an enhanced IR induced apoptosis phenotype [21]. To find further negative regulators of cep-1 we continued this genetic screen and isolated the gt448 mutant that contains significantly more apoptotic corpses than wild type (N2) worms following low dose IR treatment (30 Gy) (Figure 1A, 1D and 1E). Genetic analyses showed that the increased apoptosis is cep-1 dependent and is not caused by a DNA repair defect (see below). Mapping with a polymorphic strain, CB4856, positioned gt448 on linkage group IV, and three-factor mapping located gt448 between dpy-13 and unc-31. Fine mapping using a dpy-13 gt448 unc-31 triple mutant strain and CB4856 placed gt448 between the single nucleotide polymorphisms CE4-139 and CE4-140 (Figure 1B). Six cosmids map to this region, one of which contains the lip-1 (C05B10.1) locus. Sequencing of the coding region of the lip-1 phosphatase identified a C>T change leading to the conversion of Arg 170 to a stop codon, resulting in a truncated protein lacking the phosphatase catalytic domain (Figure 1C). Non-complementation between gt448 and the lip-1 (zh15) deletion allele [16] confirmed that increased IR induced apoptosis in the gt448 mutant is due to loss of lip-1 function (data not shown).
Both lip-1(gt448) and lip-1(zh15) mutant worms show slightly enhanced levels of apoptosis without irradiation at 20°C (Figure 1D and 1E). However, following low dose IR treatment (15 or 30 Gy) very high levels of CEP-1 dependent apoptosis are observed (Figure 1D and 1E). Previous reports indicate that lip-1 mutants show enhanced apoptosis when shifted to 25°C (without DNA damage) but no data were shown for growth at 20°C [22]. We also observed increased apoptosis when lip-1(zh15) and lip-1(gt448) mutants were shifted to 25°C, but this was cep-1 independent (data not shown).
The LIP-1 protein has been reported to be a MPK-1 phosphatase, based on its sequence homology with mammalian MAPK phosphatases and genetic analyses that implicated it as an inhibitor of mpk-1 [14]–[16]. To ascertain that the excessive apoptosis phenotype of lip-1 mutants is indeed linked to MPK-1 activation we first wished to confirm that LIP-1 acts as an MPK-1 phosphatase. We thus carefully assessed the phylogenetic relationship between LIP-1 and other known dual specificity protein phosphatases, including MAPK phosphatases, and tested directly which MAPK family members are inactivated by LIP-1. LIP-1 clusters with the mammalian ERK specific phosphatases DUSP6, 7, and 9 and Drosophila Mkp3 (Figure 2A, reviewed in [23]), whereas the other C. elegans MAPK phosphatase orthologue VHP-1 (F08B1.1), clusters with DUSP 16 and 8, both of which show substrate specificity for the JNK and p38 MAPKs (Figure 2A, reviewed in [23]). In agreement with our phylogenetic analysis, and extending the in vitro study performed by Mizuno et al., which indicated that LIP-1 shows specificity towards human ERK in vitro [24], our in vivo analysis in Cos-1 cells established that the expression of epitope-tagged LIP-1 leads to the inactivation of endogenous ERK1 and ERK2 but not of either the p38 or JNK MAPKs (Figure 2B). Furthermore, LIP-1 activity towards ERK is absolutely dependent on the integrity of a conserved Kinase Interaction Motif (KIM) located within the non-catalytic amino-terminal domain of LIP-1 (Figure 2C). LIP-1 thus shares a common mechanism of substrate recognition and catalysis with the mammalian ERK-specific phosphatase DUSP6/MKP-3, and likely acts to specifically inhibit the C. elegans ERK MPK-1 [25].
Having demonstrated that LIP-1 directly antagonizes ERK, we next tested whether activation of MPK-1 by a gain of function let-60/Ras allele results in increased apoptosis. At 20°C (the temperature used in these experiments) let-60(ga89) acts as a weak gain of function allele, while at 25°C it acts as a strong gain of function allele showing both somatic and germline phenotypes [26]. Similar to loss of lip-1, let-60(ga89) worms raised at 20°C show greatly elevated levels of IR induced apoptosis (Figure 3A and 3B). Interestingly, worms mutant for let-60(n1046), another gain of function allele, do not show enhanced apoptosis following low doses of IR but do show elevated apoptosis after higher levels (120 Gy) (Figure 3C). let-60(n1046) is a constitutive mutant allele reported to lead to excessive vulva formation but which has no effect on germline development [11].
Since we observed a difference in apoptosis induction in the two let-60 gain of function alleles we examined MPK-1 protein and phosphorylation levels in these mutants by immunoblotting with antibodies recognising mammalian ERK and phosphorylated ERK that cross react with MPK-1 [11], [27], [28]. MPK-1 is expressed as two isoforms that result from alternative splicing [29], [30]: MPK-1A is the smaller isoform that appears to be predominantly somatic, whereas MPK-1B is larger and is expressed only in the germline [31]. let-60(ga89) mutants show reduced levels of total MPK-1A and B (Figure 3D). Despite the reduction in total protein levels, both MPK-1 isoforms are hyperphosphorylated in this mutant (Figure 3D, the ratio of phosphorylated to total protein is ∼2.4 times greater than wild type for MPK-1A and ∼2.2 times for MPK-1B). On the other hand, let-60(n1046) shows reduced phosphorylation of MPK-1B (only 0.5 times that of wild type) but hyperphosphorylation of MPK-1A (∼1.6 times) and no change in total protein levels (Figure 3D). Thus, hyperphosphorylation of the MPK-1B germline isoform correlates with the hyperinduction of apoptosis following low dose irradiation. Our findings are consistent with a previous report indicating that the pattern of germline MPK-1 phosphorylation varies in let-60(ga89) and let-60(n1046) mutants [11].
In the C. elegans germline IR induced apoptosis is mediated by cep-1 (p53 homologue) dependent transcription of the BH3 only protein egl-1 (Figure 4A) [21]. In contrast, physiological apoptosis does not require either cep-1 or egl-1 [5], [12]. We were therefore interested in determining whether the increased IR dependent apoptosis observed in lip-1(lf) and let-60(ga89) mutants was mediated by the cep-1 pathway. For this, we generated double mutant strains containing combinations of either the lip-1(lf) or let-60(ga89) alleles with mutant alleles of apoptotic pathway components. We found that the enhanced apoptosis following irradiation observed in the lip-1(lf) and let-60(ga89) mutants is suppressed by the absence of cep-1 (Figure 4B and 4C) and egl-1 (Figure 4D) function. To test whether cep-1 dependent egl-1 transcription is enhanced in lip-1(lf) and let-60(ga89) mutants, we measured egl-1 RNA levels by quantitative PCR. In both lip-1(lf) and let-60(ga89) mutants there is increased IR-induced egl-1 transcription (Figure 4E and 4F, [21]). The gld-1(op236) mutant, which we have previously shown to cause excessive egl-1 transcription, was used as a positive control [21]. All germline apoptosis requires the Apaf1 homologue, ced-4, and the caspase ced-3 (Figure 4A). Therefore, as expected, in the absence of either ced-4 or ced-3 function no apoptosis is observed in lip-1(gt448) or let-60(ga89) mutants (Figure 4G). Interestingly, however, loss of either ced-4 or ced-3 enhances the small oocyte phenotype of lip-1(gt448) and let-60(ga89) mutants (Figure 4H). Old ced-3 and ced-4 worms have been reported to lay small oocytes of poor quality with the quality decreasing as the worms age, indicating that germ cell apoptosis is necessary to contribute to oocyte growth and viability by allocating scarce resources to the developing oocyte [12], [32]. Our finding that loss of both lip-1 and either ced-3 or ced-4 results in a much larger number of small oocytes indicates that both proper levels of apoptosis and MPK-1 activation independently regulate oocyte growth.
Defective DNA repair results in an enhanced apoptotic phenotype following IR due to the persistence of DNA double strand breaks that continually activate damage response pathways. To ensure that the enhanced apoptosis in lip-1(lf) and let-60(ga89) mutants is not due to a defect in DNA repair following IR, we examined the survival rate of progeny laid by irradiated mothers. Mutants that are defective in DNA repair (e.g. mrt-2(e2663) [33]) show a marked reduction in progeny survival rate following IR (see Table 1) due to the inheritance of broken chromosomes from their mothers. Unlike mrt-2(e2663) mutant worms, the survival rate of progeny arising from normal (i.e. not small) eggs from lip-1 and let-60 mutant mothers is not significantly different from that of wild type worms (Table 1). As reported previously [15] and confirmed above (Figure 4H), lip-1 mutant worms also lay small eggs and unfertilised oocytes (that can be identified by their flattened and brown appearance due to a lack of an eggshell). We also observed this phenotype in let-60(ga89) but not let-60(n1046) worms (Table 1). The rate at which these abnormal eggs/oocytes were laid was not changed by irradiation. However, the survival rates of progeny from small eggs did decrease in lip-1(lf) and let-60 (ga89) mutants following irradiation. Nevertheless, the extent of survival reduction was less than observed for mrt-2(e2663). Since these eggs are already abnormal the cause of the change in their survival rate is unclear, but is unlikely to be related to a reduced DNA repair capacity. In summary, our data show that the enhanced apoptosis observed in lip-1(lf) and let-60(ga89) worms is not due to a DNA repair defect as the survival rate of progeny derived from normal sized eggs is not affected by irradiation.
As both loss of lip-1 and gain of let-60 activity results in increased MPK-1 signaling, we tested whether the enhanced apoptosis observed in these mutants was dependent on enhanced MPK-1 activity. To do this, we generated double mutants of either the lip-1(lf) or let-60(ga89) alleles with the mpk-1(ga111ts) allele. ga111ts is a weak loss of function allele containing a mutation in the MEK binding site, which likely reduces the rate at which MPK-1 is phosphorylated and activated, and which at the restrictive temperature of 25°C results in an incomplete pachytene arrest phenotype [30]. In contrast, at the permissive temperature of 20°C mpk-1(ga111ts) worms appear wild type [30] and have normal levels of IR induced apoptosis (Figure 5A and 5B). Interestingly, at 20°C the mpk-1(ga111ts) allele could fully suppress the enhanced IR induced apoptosis observed in lip-1(gt448) and let-60(ga89) worms (Figure 5A and 5B), indicating that partially functional MPK-1 is sufficient to suppress the enhanced apoptosis phenotype. This finding demonstrates that elevated MPK-1 activity is required for the enhanced apoptosis induction observed following irradiation in the lip-1(lf) and let-60(ga89) mutants.
We had previously shown that CEP-1 protein expression occurs in a distinct pattern within the germline [21]. CEP-1 is expressed distally in the mitotic zone and proximally in late pachytene, diplotene, and diakinesis stage meiotic germ cells (Figure 6A). CEP-1 expression in the proximal region of the germline is regulated by the translational repressor GLD-1 [21]. Since MPK-1 signaling is required for progression into late pachytene [11], we wondered if the enhanced CEP-1 dependent apoptosis observed in the lip-1(lf) mutants could involve increased expression of CEP-1 in the proximal germline. We therefore examined the expression pattern of CEP-1 in dissected germlines by immunofluorescence using an anti-CEP-1 antibody [21] that shows specificity for CEP-1 (Figure S1). Both lip-1 loss of function mutants show increased overall CEP-1 expression, with CEP-1 being detected at earlier stages of pachytene compared to wild type (Figure 6A). Quantification of the range of CEP-1 expression, done by measuring the number of rows of nuclei from the beginning of a discernable CEP-1 fluorescent signal in pachytene to the first diplotene nuclei, confirms this finding (Figure 6A, right panel). The pattern and extent of CEP-1 expression is not affected by irradiation in any of the three genotypes examined (data not shown).
To further explore the relationship between MPK-1 signaling and CEP-1 germline protein levels we examined CEP-1 expression in mpk-1(ga111ts) mutants. As expected, based on the apoptotic phenotype of these mutants (Figure 5), the pattern of CEP-1 expression was indistinguishable from wild type worms raised at 20°C (data not shown, but for statistical analysis see Figure 6D). However, when raised at the restrictive temperature of 25°C the mpk-1(ga111ts) mutant germlines clearly have less CEP-1 in the pachytene region (Figure 6B, 6C and 6D). Representative images are shown to illustrate the patterns of CEP-1 expression observed in this mutant. Occasionally a normal looking germline with normal CEP-1 expression was observed. However, most germlines had either no pachytene but some diplotene expression or a patchy/small amount of pachytene expression. In summary, loss of lip-1 and loss of mpk-1 activities have opposing effects on CEP-1 germline expression. We noted that while IR has no effect on CEP-1 expression in wild type germlines (Figure 6A, 6B and 6D and [21]), mpk-1(ga111ts) germlines show rescue of CEP-1 expression: more germlines show a wild type or overexpression pattern (>12 nuclei rows) or partial rescue (0–12 nuclei rows) (Figure 6C) and the average extent of CEP-1 expression (as measured by nuclei rows) approached wild type levels (Figure 6D) (see below).
Our data clearly show that MPK-1 signaling influences CEP-1 expression in the pachytene region of the germline. We previously reported that the translational repressor GLD-1 regulates CEP-1 expression in late pachytene [21], and it has been reported that GLD-1 protein does not disappear in the proximal region of the germline in mpk-1 mutants [11], raising the possibility that control of CEP-1 expression by MPK-1 signaling is mediated by GLD-1. To test whether GLD-1 expression may be regulated by MPK-1 signaling we generated GLD-1 specific antibodies (Figure S2) and examined GLD-1 protein levels by immunoblotting. In accordance with a previous published report [11], mpk-1(ga111ts) mutants raised at 25°C show increased levels of GLD-1, whereas at the permissive temperature of 20°C GLD-1 levels are the same as wild type (Figure 7A). These findings indicate that GLD-1 levels are influenced by MPK-1 signaling and that this may form part of the mechanism controlling CEP-1 levels.
Since GLD-1 levels and CEP-1 germline expression are both affected by MPK-1 signaling we asked whether MPK-1 regulation of CEP-1 protein levels is mediated by GLD-1. As described above, we have observed that CEP-1 levels are low in mpk-1(ga111ts) mutants raised at 25°C and that IR treatment can restore CEP-1 levels to those of wild type (Figure 6). If CEP-1 expression is solely mediated by GLD-1 then we would expect that irradiation should reduce the heightened GLD-1 levels observed in the mpk-1(ga111ts) mutant. Against expectation we observe that GLD-1 levels remain unchanged in the mpk-1(ga111ts) mutants upon IR, even 24 hours post treatment (Figure 7A). These findings indicate that even though the reduced CEP-1 expression of mpk-1(ga111ts) mutants raised at 25°C can be rescued by IR, GLD-1 levels are not altered, and suggest that MPK-1 regulation of CEP-1 expression is not solely mediated by GLD-1.
Our finding that radiation rescues CEP-1 expression levels in mpk-1(ga111ts) mutants independent of changes in GLD-1 protein levels led us to examine the effect of IR on MPK-1 activity. Since mpk-1(ga111ts) is a partial loss of function allele, even at 25°C, it appeared possible that IR activates MAPK signaling, leading to more CEP-1 expression. If this hypothesis is correct IR might be able to restore mpk-1(ga111ts) activity, potentially leading to a rescue of the developmental germline defects associated with mpk-1(ga111ts) mutants.
To examine the effects that IR has on MPK-1 signaling we analysed mpk-1(ga111ts) mutant worms that had been raised at 25°C. Unirradiated worms show an incomplete pachytene arrest phenotype with approximately 70% of mpk-1 (ga111ts) mutant worms containing germlines arrested at the pachytene stage with no oocytes or embryos [11], [30] (Figure 8A). However, if mpk-1 (ga111ts) worms are irradiated and allowed to recover at 25°C, a dose dependent rescue of the pachytene arrest is observed as the proportion of intact, fully developed germlines increases (Figure 8A). These data indicate that IR activates MPK-1 signaling. Another allele of mpk-1, oz140, is functionally null and shows a fully penetrant pachytene arrest that is not temperature sensitive [30]. We did not observe a rescue of the pachytene arrest in irradiated mpk-1(oz140) mutant worms (data not shown), indicating that the rescue observed in the ga111ts worms is due to increased MPK-1 activity rather than through bypassing the requirement for MPK-1 in pachytene progression.
Since IR can rescue both the pachytene arrest phenotype and CEP-1 expression levels of mpk-1(ga111ts) mutants we were interested in measuring the apoptotic response of these worms to assess whether pachytene progression and CEP-1 expression was sufficient for a normal apoptotic response. When the irradiated pachytene-rescued worms were examined for apoptosis induction they were found to contain fewer corpses than wild type worms (Figure 8B). Even at the high dose of 120 Gy, where almost 100% of the mutant worms exhibit normal germlines, the level of apoptosis was greatly reduced compared to wild type, indicating that full MPK-1 activation is needed for apoptosis induction. The inability to induce wild type levels of apoptosis in the mpk-1(ga111ts) worms was not due to hypoproliferation of the germline as dissected germlines from mpk-1(ga111ts) mutant worms were of the same size as those of wild type, both with and without irradiation (Figure S3A, the smaller germline observed in the wild type after 120 Gy of irradiation is likely due to the high levels of apoptosis induced under these conditions) and there is no difference in the number of phospho-histone H3 positive M phase cells [34], [35] in the mitotic zone of mpk-1(ga111) germlines compared to wild type (Figure S3B). Since we observed reduced apoptosis induction in irradiated mpk-1(ga111ts) worms despite almost normal levels of CEP-1 expression, we tested whether MPK-1 signaling plays a direct role in apoptosis induction following IR. To do this, we examined egl-1 transcriptional induction in wild type and mpk-1(ga111ts) worms raised at both 20°C and 25°C treated with either 0 Gy or 60 Gy. At 20°C mpk-1(ga111ts) worms show normal apoptosis induction following IR treatment (see Figure 5) correlating with wild type levels of egl-1 induction with and without irradiation treatment (Figure 8C). When raised at 25°C unirradiated ga111ts mutant worms have equivalent levels of egl-1 mRNA to wild type worms, but greatly reduced levels of egl-1 transcriptional induction following irradiation (Figure 8C), indicating that high MPK-1 activity is required for egl-1 transcriptional induction by CEP-1 following irradiation. Thus, (I) high levels of MPK-1 signaling are required to trigger CEP-1 dependent egl-1 transcription upon IR, and the restoration of wild type levels of CEP-1 in mpk-1(ga111ts) worms rescued by ionising irradiation is not sufficient to trigger apoptosis, and (II) MPK-1 signaling plays an additional role in activating CEP-1 dependent apoptosis. In summary, our findings imply that the reduced apoptosis observed is due neither solely to an inability to enter into late pachytene where apoptosis occurs nor to defects in germline proliferation. Rather MPK-1 plays two roles: one in pachytene progression (and CEP-1 expression) and another in DNA damage dependent apoptosis induction.
The finding that irradiation rescues CEP-1 expression and pachytene progression in mpk-1(ga111ts) mutants indicates that irradiation may activate MPK-1 signaling. To directly test whether this is the case, we took advantage of an antibody that specifically recognises phosphorylated MPK-1 (P-MPK-1) in dissected germlines, and which can be used as a read-out for activated MPK-1 [11], [27], [28]. In wild type worms MPK-1 shows a distinctive phosphorylation pattern: phosphorylation occurs in early to mid pachytene, is absent in late pachytene and early diplotene, and resumes in oocytes, with highest phosphorylation levels observed in the oocyte closest to the spermatheca [11], [15], [28]. We first confirmed this phosphorylation pattern in unirradiated wild type worms (Figure 9A: the bend region is shown by the arc, mid pachytene by *, and late pachytene by **). We next demonstrated in lip-1 mutants that P-MPK-1 occurs in late pachytene cells residing in the germline bend as previously reported, indicating that LIP-1-mediated dephosphorylation is responsible for the absence of P-MPK-1 in this region of the germline (for representative images see, Figure 9B and 9C) [15]. We note that lip-1(lf) mutants have a reduced level of the MPK-1B germline isoform, which correlates with a lower level of total MPK-1B phosphorylation (Figure S4). Nevertheless, we consistently detected P-MPK-1 in the bend region of lip-1(gt448) and lip-1(zh15) mutant germlines (Figure 9B and 9C). Given that the bend region only comprises a small part of the germline our cytological data does not contradict our observations of total MPK-1B phosphorylation. To see whether IR induces MPK-1 activation in the germline, we dissected wild type and lip-1 mutant germlines 2–3 hours following irradiation. Interestingly, unlike unirradiated wild type germlines, irradiated wild type germlines show P-MPK-1 throughout the bend region of the germline (Figure 9D) indicating that MPK-1 is activated in the late pachytene/early diplotene region in wild type germlines following IR. Irradiation of the lip-1 mutant germlines resulted in no obvious change in P-MPK-1 fluorescence compared to unirradiated lip-1 mutant germlines (Figure 9E and 9F), indicating that lip-1 mutation likely results in a high level of MPK-1 phosphorylation which cannot be further enhanced by IR. Taken together, our data indicate that the presence of active MPK-1 in late pachytene germ cells correlates with apoptosis induction and that IR activates MPK-1 signaling in the germline.
Since we observe activation of MPK-1 in wild type germlines following irradiation we were interested in examining whether we could also detect increased phosphorylation of MPK-1 in the mpk-1(ga111ts) mutant, which would support our conclusions that MPK-1 is also activated in this mutant. As previously mentioned, the ga111ts mutation affects the MEK binding site and is predicted to reduce the rate of MPK-1 activation by MEK [30]. At 20°C this mutant shows no obvious phenotypic defects and this correlates with the almost wild type levels of P-MPK-1 staining observed in germlines from animals raised at 20°C (Figure 10C). The intensity of staining consistently appears to be slightly reduced in the ga111ts mutants and a low level persists in the bend region, indicating that MPK-1 may be involved in a negative feedback loop to control its own downregulation. Despite these differences, MPK-1 is clearly activated in ga111ts mutants following irradiation as the intensity of staining consistently increases in the bend region and in the developing oocytes (Figure 10D). These findings correlate with the observations that mpk-1(ga111ts) mutants raised at 20°C show no phenotypic differences from wild type, including a normal apoptotic response.
We then next examined germlines from animals raised at 25°C. Interestingly, while some wild type germlines showed the characteristic wild type staining pattern without irradiation (Figure 10E), some germlines showed activation of P-MPK-1 in the bend region even without irradiation treatment (Figure 10F, quantified in 10L: ‘normal’ describes the wild type pattern without IR, ‘activated’ describes the wild type pattern following IR, and ‘background’ means no discernable staining). This finding implies that P-MPK-1 may be activated due to stress caused by the elevated temperature. However, upon irradiation more germlines showed phosphorylation in the bend region (Figure 10G, quantified in 10L) indicating that at 25°C MPK-1 still becomes active following irradiation in wild type germlines. Germlines from mpk-1(ga111ts) animals raised at 25°C showed two patterns, they either had faint staining throughout the proximal part of the germline (Figure 10H) or they showed a background level (Figure 10I, quantified in 10L). Upon irradiation, more germlines showed some faint staining (Figure 10J) but some still showed background levels (Figure 10K, quantified in 10L), indicating that MPK-1 is activated in these mutants upon IR. However the level of activation in the bend region never approaches wild type levels. These findings correlate with the conclusions we have drawn from our genetic experiments. In the mpk-1(ga111ts) mutant worms raised at 25°C MPK-1 activity is greatly reduced resulting in an incomplete pachytene arrest phenotype and very low levels of CEP-1 expression. Upon IR MPK-1 is activated (but not to wild type levels) and this is sufficient to induce pachytene progression and CEP-1 expression but insufficient to induce proper egl-1 transcription and apoptosis.
IR induced cell cycle arrest and apoptosis is dependent on signaling by the DNA damage signaling pathway [36]. To test whether MPK-1 activation by irradiation is also dependent on the DNA damage signaling pathway, we assessed P-MPK-1 levels by immunofluorescence in germlines from atm-1(gk186) [37], atl-1(tm853) [38] and mrt-2(e2663) [33] mutants. atm-1 and atl-1 encode the homologues of the mammalian phosphatidylinositol 3-kinase proteins ATM and ATR, respectively. These proteins act to sense and signal DNA damage, with ATM responding primarily to double strand breaks and ATR to replication stress. However, there is increasing evidence for cross talk between the two signaling pathways (for recent reviews see [39], [40]). In C. elegans atm-1 and atl-1 are required for cell cycle arrest and apoptosis induction in the germline following IR [37], [38]. In addition, atl-1 is essential for embryogenesis and mutants exhibit mitotic catastrophe and defects in the S-phase checkpoint in mitotic germ cells [38]. mrt-2 encodes a component of the 9-1-1 complex, which is recruited to sites of DNA damage and is required for full ATR activation [41], [42]. In C. elegans mrt-2 is required to sense and signal DNA damage in the germline resulting in cell cycle arrest and apoptosis [33], [36].
P-MPK-1 activation in atm-1 mutants appeared wild type, with no P-MPK-1 detected in the bend region without IR (Figure 11C) but significant levels following IR (Figure 11D). In contrast, atl-1 mutants showed clear MPK-1 phosphorylation in the bend region with and without IR treatment (Figure 11E and 11F). P-MPK-1 is also detected in the bend region of mrt-2 mutants with and without IR treatment. However, the degree of activation is lower than in the middle pachytene region for both treatments (Figure 11G and 11H, compare the ** region with the * region in the images). These findings indicate that atm-1 and atl-1 are dispensable for MPK-1 activation by IR. However, the loss of atl-1 (but not atm-1) function in the absence of IR results in the activation of MPK-1 in late pachytene/early diplotene. The activation of MPK-1 could be a result of the high levels of chromosomal instability exhibited by atl-1 mutants [38] or other defects. Like atl-1, mrt-2 mutants also exhibit chromosomal instability [33] and also have activated levels of MPK-1 in the bend region of the germline. However, the levels of P-MPK-1 are lower than that observed in the atl-1 mutants and do not significantly increase upon IR, indicating that mrt-2 may play a role in the activation of P-MPK-1 in response to DNA damage but is not absolutely required.
Given our finding that MPK-1 is active in the late pachytene region of the wild type germline and that a high level of MPK-1 signaling is required for efficient CEP-1 dependent apoptosis induction, we next asked whether MPK-1 could directly activate CEP-1. To test whether CEP-1 could directly interact with MPK-1 we performed a yeast two-hybrid assay. For this, we generated a plasmid containing a fusion between the Gal-4 activation domain (GAD) and cep-1 cDNA and another set of plasmids with the Gal-4 binding domain (GBK) fused to either cep-1, lip-1, mpk-1a, or mpk-1b cDNAs, and generated yeast strains by pairwise matings between GAD and GBK strains. We tested for an interaction by (I) growth on selective media (-His -Ade) and (II) increased beta-galactosidase activity (Figure 12A and 12B). As positive controls, we examined known interactions between LIP-1 and the two MPK-1 isoforms, the mammalian ERK, the sevenmaker version of ERK (which binds phosphatases less efficiently), JNK, and p38, as well as between DUSP6 and the MPK-1 isoforms, ERK, sevenmaker, JNK, and p38. In accordance with our in vivo dephosphorylation data, both LIP-1 and DUSP6 interact strongly with the MPK-1 isoforms and with ERK, less so with sevenmaker, and not at all with JNK or p38 (Figure S5). As the controls showed the expected interaction we next tested for an interaction between CEP-1 and MPK-1. We observed an interaction between CEP-1 and each of the MPK-1 isoforms, with MPK-1B showing a stronger interaction in both assays (Figure 12A), indicating that CEP-1 and MPK-1A/B do directly interact in yeast cells. While we could not independently confirm this result via co-immunoprecipitation of exogenously expressed CEP-1 and MPK-1 due to an inability to express CEP-1 in mammalian cells, our combined genetic and biochemical evidence suggests that MPK-1 dependent phosphorylation might directly regulate CEP-1 activity.
Using an unbiased genetic screen we found that MAP kinase signaling affects CEP-1 dependent DNA damage induced apoptosis. We provide clear evidence that CEP-1 dependent germ cell apoptosis is increased in mutants with increased MPK-1 activity. Conversely, reduction of MPK-1 activity in mpk-1(ga111ts) mutants leads to reduced DNA damage dependent apoptosis. We show that MPK-1 signaling plays important developmental roles in pachytene progression and in regulating CEP-1 expression in pachytene, and a possible direct role in DNA damage induced apoptosis. We postulate that MPK-1 signaling controls DNA damage induced apoptosis through several genetic pathways that all appear to converge on CEP-1. Firstly, MPK-1 signals that germ cells are in late pachytene and that CEP-1 expression can occur. Secondly, MPK-1 signaling regulates GLD-1, which in part could account for the upregulation of CEP-1 expression. Thirdly, MPK-1 is activated in response to IR and this appears to contribute to CEP-1 dependent apoptosis, possibly by direct activation of CEP-1 by MPK-1.
Only cells that are in late pachytene are competent for apoptosis in the C. elegans germline. Our results clearly demonstrate that MPK-1 signaling plays a developmental role in establishing apoptotic competency by regulating CEP-1 levels in late pachytene. It does this by regulating the levels of GLD-1, a known translational inhibitor of CEP-1 [21] and by other unknown mechanism(s) independent of GLD-1 levels. Diagrams depicting GLD-1, CEP-1, and P-MPK-1 expression patterns in wild type, lip-1, and mpk-1 mutant germlines are shown in Figure 13A–13E. Our finding that IR can rescue the pachytene arrest phenotype of mpk-1 worms raised at 25°C has allowed us to examine the role that pachytene progression plays in CEP-1 expression, apoptosis induction, and GLD-1 regulation. Since CEP-1 expression is also rescued in the IR treated mpk-1 worms it appears that pachytene progression and low MPK-1 activity is sufficient for CEP-1 expression to occur in late pachytene (and to overcome or bypass GLD-1 mediated translational repression). Conversely, enhanced MPK-1 signaling leads to increased CEP-1 expression. Increased CEP-1 expression alone is not sufficient to induce an apoptotic response, as lip-1(lf) and let-60(ga89) mutants don't show high levels of CEP-1 dependent apoptosis without irradiation (at 20°C). Rather MAP kinase mediated CEP-1 expression primes the cells to respond to a DNA damage signal, and the more cells expressing CEP-1, the greater the apoptotic response. In addition, the rescue of CEP-1 expression in mpk-1(ga111ts) mutants raised at 25°C by irradiation does not lead to a wild type apoptotic response or egl-1 transcriptional induction, indicating that low MPK-1 activity (as shown by the low levels of P-MPK-1 in these germlines (Figure 10)) or restored CEP-1 expression alone are not sufficient to trigger a full apoptotic response. Rather, a normal apoptotic response requires a higher level of MPK-1 activity (see below).
Findings presented in this study indicate that CEP-1 expression in late pachytene, associated with the establishment of apoptotic competency, is under developmental control to ensure that germ cells with damaged DNA or defects in meiotic recombination are culled prior to oogenesis. Interestingly, apoptotic competency in late meiotic prophase seems to be evolutionary conserved: rat and mouse diplotene/diakinetic staged oocytes are more sensitive to IR than oocytes at earlier meiotic stages [43], [44], and p63 expression is also restricted to late pachytene and diplotene staged mouse and human oocytes [45] and pachytene staged mouse spermatocytes [46]. It is thus likely that in mammals p63 expression is subject to analogous developmental control mechanisms to those we have observed for C. elegans. While there is no reported evidence that ERK signaling impacts on p63 expression in the mammalian germline, the finding that ERK expression is observed in meiotic prophase in mouse spermatocytes [47], [48] lends weight to the idea that ERK signaling may play a role in regulating apoptosis in the mammalian germline.
Two questions arise from our finding that MPK-1 is phosphorylated in the late pachytene region in response to irradiation: how is MPK-1 phosphorylated in late pachytene, and what role does active MPK-1 play in this region? There are two possible explanations for the first question: either phosphorylated MPK-1 persists in cells progressing from earlier in pachytene (indicating that LIP-1 dependent dephosphorylation is inhibited by IR), or MPK-1 is activated anew by upstream signaling pathway responding to IR. In mammals ERK is activated in response to IR either through the EGF receptor [49]–[51], or by the inhibition of MAPK phosphatases by elevated levels of free radicals (reviewed in [52]–[54]). The finding that P-MPK-1 does not increase beyond a high basal level in lip-1(lf) mutants upon IR suggests that the mechanism of IR induced MPK-1 phosphorylation may occur via inhibition of LIP-1. However, it is possible that in lip-1 mutants maximal MPK-1 activation may already be reached and any enhancement due to upregulation of the signaling pathway is not detectable by immunofluorescence. At this stage our data do not allow us to differentiate between these possibilities. The DNA damage response pathway plays an important role in the cellular response to DNA damaging agents such as IR and it is possible that this pathway is responsible for MPK-1 activation in late pachytene. Our data clearly show that this is not the case as MPK-1 is still phosphorylated in the absence of sensing (mrt-2) and signaling (atm-1 and atl-1) gene products (Figure 11). However, the observation that in the absence of mrt-2 MPK-1 is not strongly phosphorylated in late pachytene upon IR indicates that MRT-2 may be required for full MPK-1 activation.
The second question arising from our observations that MPK-1 is phosphorylated and activated by IR regards its possible role in the damage response pathway. We show that in the absence of strong MPK-1 signaling (in the mpk-1(ga111) worms raised at 25°C) apoptosis and egl-1 transcriptional induction are reduced following IR (Figure 8) despite almost wild type pachytene progression and CEP-1 expression (Figure 6). It is possible that even though the rescue in pachytene progression and CEP-1 expression is almost wild type, they are still not sufficient to induce a wild type apoptotic response in these worms. However, another interpretation of the findings is that the activation of MPK-1 following IR is required for an IR induced cellular response and that it is possible that IR activated MPK-1 facilitates or directly regulates CEP-1 dependent apoptosis. MPK-1 activation occurs within two hours of IR treatment in the late pachytene, where apoptosis occurs [12], [36] and CEP-1 is expressed [21], and this timing correlates with egl-1 induction (first detected one to two hours post IR [55]). Our finding that MPK-1 and CEP-1 physically interact in a yeast two-hybrid assay lends support to the idea that MPK-1 may directly regulate CEP-1. It will be important to test the possibility that direct MPK-1 dependent phosphorylation is required for CEP-1 activation in future studies. In this study our inability to express CEP-1 in mammalian systems prevented us from using co-immunoprecipitation of heterologous proteins to independently confirm the two-hybrid interactions. Also, our attempts to immunoprecipitate either endogenous CEP-1 or MPK-1 from worm extracts using currently available antibodies failed. Nevertheless, CEP-1 contains a number of putative MAPK phosphorylation and also potential docking sites (data not shown), required to provide high-affinity binding sites between MAPKs and their substrates [56]. The discovery of such consensus sites suggests that CEP-1 could be a possible MPK-1 substrate and future experiments could address this question. Mammalian ERK can phosphorylate and activate p53, leading to cell cycle arrest or senescence [54]. In addition, there is a growing body of evidence indicating that ERK-mediated p53 serine 15 phosphorylation and activation can mediate apoptosis induction following treatment with DNA damaging agents such as doxorubicin [57], [58], cisplatin [59], and UV [60]. We therefore speculate that a conserved mechanism for MPK-1 in mediating damage induced apoptosis through direct CEP-1 phosphorylation may exist in C. elegans. If this mechanism does exist it functions in addition or parallel to the activation of CEP-1 by the DNA damage response pathway as induction of the DNA damage response pathway is still required for apoptosis induction in lip-1 mutants.
Understanding how p53 and p63 are regulated is of vital importance for understanding tumour progression and germline development, respectively. In this work we have begun to dissect the complex relationship between MAPK signaling and p53 dependent apoptosis in the germline. We show that C. elegans germline CEP-1 dependent apoptosis is regulated both developmentally and more directly by MAPK signaling in C. elegans, and we expect that these mechanisms of regulation could be conserved throughout evolution.
Worms were maintained at 20°C on NGM plates unless otherwise stated. The strains used were LG I cep-1(lg12501) [21], gld-1(op236) [21], atm-1(gk186) [37], LG III mpk-1(ga111ts) [30], mpk-1(oz140) [30], ced-4(n1162) [61], mrt-2(e2663) [33], LG IV lip-1(zh15) [16], lip-1(gt448) (this study), let-60(ga89) [26], let-60(n1046) [62], ced-3(n717) [63], LG V egl-1(n1084n3082) [64], atl-1(tm853) [38], CB4856 [65].
Mutants were generated using standard mutagenesis protocols and F2 progeny were screened for enhanced apoptosis 28–30 hr following irradiation using acridine orange staining [21]. lip-1(gt448) was backcrossed five times and mapped using standard genetic methods.
For the temperature sensitivity assays, mpk-1(ga111ts) and wild type worms were shifted to 25°C as L1 larvae, allowed to develop to the L4 larval or young adult stage, then treated and allowed to recover at 25°C for the time indicated.
Cos-1 cells were transiently transfected with either C. elegans pSG5-LIP-1-Myc or pSG5-LIP-1-KIM-Myc as previously described [66]. The LIP-1 KIM mutant in which both Arg59 and Arg60 were mutated to Ala was generated by overlap extension PCR [67]. Briefly, two independent PCR reactions were performed using pSG5-LIP-1-Myc as template and either primer pair 1 5′-GGCGAATTCTATTTTCAGATGAC-3′ and CCGCCCATTAAAGCGGCTTGAAG-3′, or primer pair 2 5′-CTCTCCTTCAAGCCGCTTTAATG-3′ and 5′-TTCCTCGAGAACTGCAGTTTCG-3′ (nucleotide substitutions are underlined). The PCR products were then mixed and used as template for a third PCR reaction using primers 5′-GGCGAATTCTATTTTCAGATGAC-3′ and 5′-TTCCTCGAGAACTGCAGTTTCG-3′ to generate the mutant reading frame. This amplicon was then subcloned into pSG5-myc as before and verified by DNA sequencing before transfection. As a positive control either pSG5-DUSP1-Myc or pSG5-DUSP5-Myc was used to inactivate ERK, or pSG5-DUSP1-Myc to inactivate p38 and JNK. To activate endogenous ERK cells were serum starved for 16 hrs and then stimulated by addition of 15% FBS. To activate endogenous p38 and JNK cells were exposed to anisomycin (5 µg/ml for 30 min). Following treatment, cells were lysed and proteins analysed by SDS-PAGE and Western blotting using antibodies that detect either the phosphorylated or total amount of the relevant MAPK. Tubulin levels were analysed as a loading control.
DNA damage induced apoptosis, radiation (rad) sensitivity, and egl-1 transcription assays have been previously described [68], [69]. A caesium-137 source (IBL437C, CIS Bio International) was used for the irradiation.
Rabbit anti-GLD-1 antiserum was raised against recombinant MBP-His-tagged GLD-1 amino acids 155–463 purified using TALON resin (Clontech). For antibody affinity-purification, GST-GLD-1 STAR domain (135–336) fusion protein was coupled to Affi-gel 15 resin (Bio-Rad) according to the manufacturer's guidelines. Rabbit anti-GLD-1 antiserum was incubated with the resin overnight at 4°C, and purified antibody was eluted rapidly using 100 mM glycine pH 2.5 and the pH was neutralised with 1 M Tris, pH 8.8. Once all antibody had apparently been eluted, the resin was then incubated with a further volume of glycine pH 2.5 for 1 hr at 4°C before elution and neutralisation to obtain higher affinity antibodies. Purified antibody was stored in 1% BSA, 10% glycerol and 0.02% thimerosal at −80°C.
Worms were grown until young adults (24 hrs post L4 larval stage), irradiated, and protein was harvested at the indicated times by adding an equal volume of lysis buffer (20 mM Tris HCl pH 8.0, 40 mM Na pyrophosphate, 50 mM NaF, 5 mM MgCl2, 100 µM Na vanadate, 10 mM EDTA, 1% Triton X-100, 0.5% deoxycholate). Zirconia/silica beads were added (0.7 mm, BioSpec Products) and the worms were homogenised by beating (3×30 sec, with 30 sec in between) in a Mini-Beadbeater-8 (BioSpec) at 4°C. The homogenate was incubated on ice for 30 min and then centrifuged to remove debris and resulting supernatant was stored at −80°C. An equal amount of protein extract (1 µg for tubulin, 20 µg for total MPK-1, and 40 µg for phosphorylated MPK-1, 10 µg for GLD-1) was boiled in 1× SDS loading buffer and separated on 10% for MPK-1 or 4–12% for GLD-1 Bis-Tris SDS-PAGE gels (Invitrogen). Western blot analysis was performed using ERK (K-23, Santa Cruz 1∶2000, rabbit) and phosphorylated ERK (clone MAPK-YT, Sigma, 1∶2000, mouse) specific antibodies that cross react with MPK-1 and phosphorylated MPK-1 [11], [27], [28] or anti-GLD-1 (1∶500, rabbit, this study) and HRP conjugated secondary antibodies (anti-rabbit-HRP and anti-mouse-HRP, DakoCytomation 1∶2000). Antibody to α-tubulin (DM1A, Sigma, 1∶2000, mouse) was used to control for loading. Band intensity quantification was performed using Image J software.
Germlines were extruded into dissection buffer (27.5 mM HEPES pH 7.4, 130 mM NaCl, 53 mM KCl, 2.2 mM CaCl2, 2.2 mM Mg Cl2, 0.01% Tween20, 0.2 mM levamisole) and fixed with 1.8% (for P-MPK-1 and CEP-1) or 0.5% (for P-H3) formaldehyde (27.5 mM HEPES pH 7.4, 130 mM NaCl, 53 mM KCl, 2.2 mM CaCl2, 2.2 mM MgCl2) for 5 min (P-MPK-1 and CEP-1) or 4 min (for P-H3). Following freeze cracking, they were post-fixed in 100% methanol (for P-MPK-1 and P-H3) or 50∶50 methanol∶acetone (for CEP-1) for 10 min at −20°C and permeabilised in 0.1% Triton X-100 (for P-MPK-1 and P-H3) or 1% Triton X-100 (for CEP-1) in PBS (4×10 min). Immunofluorescence was performed using antibody to phosphorylated ERK (clone MAPK-YT, Sigma, 1∶100) and anti-mouse AlexFluor-568 (Invitrogen, 1∶500), antibody to CEP-1 ([21], 1∶200) and anti-goat AlexFluor-488 (Invitrogen, 1∶200), or antibody to phosphorylated histone H3 (Ser 10, Millipore, 1∶500) and anti-rabbit AlexFluor-568 (Invitrogen 1∶200). DAPI (1 µg/µl) was used to stain chromatin. Images of P-MPK-1 and CEP-1 stained germlines were taken using an Axioskop 2 (Zeiss) microscope fitted with a RTke camera and accompanying SPOT analysis software (Diagnostic Instruments) using the same exposure settings for each channel. Brightness and contrast of the resulting images were modified to more clearly see the staining patterns but no other changes were made. Images of P-H3 stained germlines were taken using a Leica LMF Spectris microscope and deconvolved using SoftWorx (Applied Precision).
Yeast two-hybrid assays were performed as described previously [70]. Briefly, open reading frames encoding cep-1, lip-1, and human DUSP6 were subcloned into the Gal4 DNA binding domain-fusion (bait) vector pGBK-T7 (Clontech), while the C. elegans MAP kinases mpk-1a/1b, mammalian ERK2, and the ERK2 sevenmaker mutant were subcloned into the Gal4-activation domain-fusion (prey) vector pGAD-T7 (Clontech). pGBK and pGAD fusion constructs were then transformed into yeast strains pJ69-4A and pJ69-4alpha [71] respectively, using the rapid method of Gietz and Woods (2002) [72]. Transformed yeast were selected on auxotrophic media lacking tryptophan (pGBK-fusions) or leucine (pGAD-fusions) respectively. Transformants were mated overnight in 200 µl non-selective YPDA rich medium, of which 50–100 µl of suspended yeast were plated onto dual-selective media lacking leucine and tryptophan. Interactions were probed by growth on media lacking leucine/tryptophan (LT) or leucine/tryptophan/histidine/adenine (LTHA) respectively. Growth on LTHA medium was assessed after 72 hrs of culture at 30°C and considered indicative of an interaction. Semiquantitative analysis of two-hybrid interactions was performed by beta-galactosidase assay as described previously [70].
|
10.1371/journal.pgen.1000299 | Novel Low Abundance and Transient RNAs in Yeast Revealed by Tiling Microarrays and Ultra High–Throughput Sequencing Are Not Conserved Across Closely Related Yeast Species | A complete description of the transcriptome of an organism is crucial for a comprehensive understanding of how it functions and how its transcriptional networks are controlled, and may provide insights into the organism's evolution. Despite the status of Saccharomyces cerevisiae as arguably the most well-studied model eukaryote, we still do not have a full catalog or understanding of all its genes. In order to interrogate the transcriptome of S. cerevisiae for low abundance or rapidly turned over transcripts, we deleted elements of the RNA degradation machinery with the goal of preferentially increasing the relative abundance of such transcripts. We then used high-resolution tiling microarrays and ultra high–throughput sequencing (UHTS) to identify, map, and validate unannotated transcripts that are more abundant in the RNA degradation mutants relative to wild-type cells. We identified 365 currently unannotated transcripts, the majority presumably representing low abundance or short-lived RNAs, of which 185 are previously unknown and unique to this study. It is likely that many of these are cryptic unstable transcripts (CUTs), which are rapidly degraded and whose function(s) within the cell are still unclear, while others may be novel functional transcripts. Of the 185 transcripts we identified as novel to our study, greater than 80 percent come from regions of the genome that have lower conservation scores amongst closely related yeast species than 85 percent of the verified ORFs in S. cerevisiae. Such regions of the genome have typically been less well-studied, and by definition transcripts from these regions will distinguish S. cerevisiae from these closely related species.
| The budding yeast Saccharomyces cerevisiae, because of the relative ease of its genetic manipulation and its ease of handling in the laboratory, has long served as a model on which studies in higher organisms have been based. To more fully understand how eukaryotic cells express their genomes, we sought to identify RNA species that are transcribed at very low levels or that are rapidly degraded. We created mutants deficient in the ability to degrade RNA, with the expectation that this would increase the relative abundance of such RNAs, and then used high-resolution microarrays and sequencing technologies to locate and identify from where these RNAs are transcribed. Using this approach, we have identified 365 transcripts that do not appear in the most current list of annotated S. cerevisiae RNA transcripts; of these, 185 are unique to our study. Many of these novel transcripts derive from regions of the genome that are poorly conserved between S. cerevisiae and other closely related yeast species, suggesting that these RNAs may play an important role in the divergent microevolution of S. cerevisiae.
| Twelve years ago, in a landmark study resulting from the collaborative work of hundreds of scientists around the world, the budding yeast Saccharomyces cerevisiae became the first eukaryote to have its genome fully sequenced [1]. The initial analysis of the genome utilized the following (necessarily) arbitrary rules for defining whether an Open Reading Frame (ORF) was a protein-coding gene (a “genic ORF”) or not: 1) a genic ORF had to start with ATG and have at least 100 sense codons, and 2) if two ORFs of more than 100 sense codons overlapped one another by more than 50% of their lengths, then the longer was picked as being a genic ORF, while the shorter was discarded. In this way, it was determined that the sequence of 12,068 kilobases contained 5,885 potential protein-coding genes. In addition, non-protein-coding genes consisting of approximately 140 ribosomal RNA genes, 40 small nuclear RNA genes, and 275 transfer RNA genes were identified using various criteria, resulting in a total of approximately 6,340 genes.
Early analyses of the predicted protein-coding genes showed that about 35% had no known function or homolog [2], leading to questions about the validity of the rules used to identify genic ORFs. Various algorithmic methods have predicted fewer genes in the yeast genome than the originally predicted number of 6,340, based on a variety of criteria [3]–[7], while other methods have found and verified new ones, especially non-coding genes [8],[9]. Comparative genomics [10]–[12], and various experimental methods [13]–[17] have also resulted in significant changes to the primary annotation of the yeast genome, introducing hundreds of newly predicted genic ORFs, while marking many others as ‘dubious’. However, new genes added by one study are frequently marked as ‘dubious’ by another, as recorded within the Saccharomyces Genome Database (SGD) [18], indicating the speculative nature of many of these annotations. Additionally, a recent study [19] has shown that the use of comparative genomics alone to determine whether or not a genomic region is likely to harbor a genic ORF can result in false negatives, since many transcribed elements may not be conserved across even closely related species. It has been suggested that such ORFs may be important for the micro-evolutionary divergence between species. Clearly, even in a genome as simple as, and containing as few introns as that of S. cerevisiae, it is still not straightforward to identify all of the genes simply based on the DNA sequence.
Hybridization of RNA to tiling microarrays (microarrays containing overlapping, offset probes that tile across the entire genome) has been used to generate genome-wide transcript profiles and to detect previously unannotated transcripts. While this technique has its own caveats, it overcomes the limitations of many previous attempts to find undiscovered transcripts, by providing direct experimental support with high-resolution data. Tiling array studies have revealed more than 5,000 novel transcripts in Arabidopsis [20] and rice [21], and more than 10,000 previously unknown transcripts in human cells [22]–[24]. In yeast, tiling array experiments performed by David et al. [25], using RNA isolated from a single experimental condition, identified almost 800 novel (i.e., not annotated in SGD [18]) transcripts.
Recently, Miura et al. [26], also working with S. cerevisiae, performed large-scale sequencing of vector-capped cDNA clones [27],[28] from two cDNA libraries to accurately map over 11,000 transcriptional start sites (TSSs). Of these predicted transcripts, 667 were novel (many of which were also identified by David et al.), and contained ORFs corresponding to 100 amino acids or less and thus would have been missed in the original annotation. Furthermore, they discovered 45 new introns, 367 novel antisense transcripts, and showed that most yeast genes have two or more TSSs, demonstrating that the transcriptional potential of the yeast genome is more complex than previously thought. In total, their analysis detected only 3,599 of the more than 6,000 currently annotated genic ORFs, suggesting either that many genes were missing from their cDNA library, or that many of the annotated genic ORFs are not correct.
Recent advances in sequencing technology [29]–[32] have allowed an unprecedented look at the transcriptome, using a method known as RNA-Seq [33]. This method can yield millions of sequence reads from cDNA libraries, and has been used to discover and validate transcribed regions of the genome in various organisms [34]–[36]. Most recently, RNA-Seq has been used to identify additional transcripts expressed in S. cerevisiae growing in rich medium [37], and transcripts expressed in S. pombe growing under several different conditions, including a meiotic time course [38]. From tens of millions of sequence reads, 204 novel transcripts were identified in S. cerevisiae, and 453 novel transcripts in S. pombe; additionally, many transcript boundaries were refined, and novel introns identified. The functions of these novel transcripts remain unknown, with few expected to be protein-coding [38].
There exist various mechanisms by which RNA is processed, surveyed, and turned over. In S. cerevisiae, there are two major pathways that play a role in the decay of mRNAs in the cytoplasm, both of which involve deadenylation (Figure 1). In the first pathway, deadenylation is followed by the removal of the 5′ m7G cap by Dcp1p and Dcp2p, which is then followed by degradation in the 5′ to 3′ direction by Xrn1p [39]–[44]. In addition to Dcp1p and Dcp2p, there exists a group of proteins that function as activators for decapping, including Pat1p, the Lsm1-7p complex, and Dhh1p [45]–[49]. In the second pathway, deadenylated mRNAs are degraded in the 3′ to 5′ direction by the exosome and the Ski complex (consisting of Ski2p, Ski3p, and Ski8p) [50],[51]. In the nucleus, mRNAs that are unspliced, improperly processed, and/or otherwise unable to leave the nucleus are degraded in pathways using the same machinery [52]–[55]. Rrp6p, a nuclear-only component of the exosome which has 3′ to 5′ exonuclease activity [56],[57], plays a major role in the nuclear degradation of mRNAs as well as CUTs ([58] and reviewed in [59],[60]).
As described above, genome-wide screens for novel transcripts have revealed the existence of many non-coding, intergenic, and/or antisense RNAs. Such RNAs are poorly understood, sometimes being referred to as ‘transcriptional noise’, whose expression may be initiated from inadvertent binding of RNA polymerase complexes to DNA sequences that bear resemblance to ‘real’ transcriptional promoters. In S. cerevisiae, some of these transcripts are rapidly degraded and have been labeled as cryptic unstable transcripts or CUTs (Figure 1; [58] and reviewed in [59],[60]). While the roles of these CUTs are unclear, the mechanism by which these RNAs are degraded has been elucidated and it has been shown that they are specifically targeted for degradation via polyadenylation by the non-canonical polyadenylation protein Trf4p, a component of the TRAMP complex [58],[61],[62]. Why these RNAs are transcribed at all, and why a specific degradation pathway exists for them in the budding yeast remains speculative.
To identify additional novel transcripts in the yeast S. cerevisiae, we have employed both tiling microarrays and RNA-Seq, with the explicit goal of identifying those transcripts that are either short-lived and/or occur in low abundance. Such transcripts may include previously unrecognized protein-coding transcripts and non-coding transcripts, as well as cryptic unstable transcripts and ‘transcriptional noise’. To allow better detection of these types of transcripts, we have analyzed RNA isolated from three strains containing various combinations of deletions of six genes that play a role in RNA processing (RRP6, XRN1, PAT1, LSM1, SKI2 and SKI8), with the hypothesis that the most unstable and/or least abundant transcripts would show the greatest relative change in abundance in such mutants. The mutant-derived RNA was compared to RNA from wild-type cells, using Affymetrix strand-specific tiling microarrays. Novel strand-specific transcripts were identified by segmentation of the relative expression measures from the tiling arrays and subsequently validated using Illumina's Solexa sequencing platform. Using a combined tiling array and RNA-Seq approach, we have identified a total of 365 transcripts that are currently unannotated in SGD. Comparison of our data to various recently published transcriptome studies [25],[26],[37],[63] reveals that of these unannotated transcripts, 185 are novel and unique to our study.
Our primary goal was the discovery of novel transcripts based on comparing RNA from mutants deficient in RNA degradation pathways to RNA from a wild-type strain. We wanted to provide, in a high-throughput fashion, distinct and complementary lines of evidence for the existence of each putative transcript. We thus selected two technologies as being appropriate for this aim: tiling arrays and high-throughput sequencing. We used the tiling arrays to discover novel transcribed segments, with their strand of origin information. This approach has been used successfully in previous studies [25] and there are well-established computational and statistical methods for analyzing tiling array data. Tiling arrays, as opposed to high-throughput sequencing, provide an even spacing of measurements across the entire genome, making them more amenable to off-the-shelf segmentation algorithms. In addition, an entire population of molecules is hybridized to the microarray, whereas a sequencing based approach is inherently a sampling strategy, limited by the depth to which one can afford to sequence, and by the complexity of the sample being sequenced. However, high-throughput sequencing provides an independent experimental platform well-suited for transcript validation as each read provides distinct evidence for the presence of a transcribed segment.
Tiling microarray analysis of mRNA from yeast grown under a diverse set of several different conditions suggested that the greatest fraction of known transcripts are detectable in the presence of high salt (0.8 M NaCl) (our unpublished results); we thus chose high salt as the growth condition used in the experiments described herein. All our deletion strains (the ‘mutant’ strains) and the wild-type strain (see Table 1 for strain details) were shocked with high salt for 30 minutes; total RNA was isolated from each strain, from which a poly A+ RNA sample was also purified, resulting in two different RNA preparations per strain. These RNAs were then labeled and hybridized to both forward and reverse strand Affymetrix yeast genome tiling microarrays (see Materials and Methods).
Only perfect match (PM) probes mapping uniquely to the genome were used in the analysis; mismatch probes were discarded. In order to correct for probe-specific effects and to detect only those transcripts that were differentially expressed between a mutant and the wild-type, we used as expression measures the log ratio of mutant PM intensities to wild-type PM intensities. We segmented the log ratios using a piecewise constant change point model as implemented in the ‘segment’ function in the R package ‘tilingArray’ [64] from Bioconductor [65]. Following Huber et al., we utilized the Bayesian information criterion (BIC) penalized likelihood to select the number of transcribed segments. Poly A+ RNA and total RNA microarray data were segmented separately. Based on a visual assessment of the resulting segmentation it appeared that BIC overestimated the number of segments (also noted by Huber et al.).
Oversegmentation makes downstream validation of the segments more challenging, as putative segments are judged in pieces as opposed to their entirety. Thus, we post-processed the segmented data to: (1) join adjacent segments with similar expression measures, (2) drop segments that are not differentially expressed, using a threshold of <0.5 on the log2 scale, (3) remove segments overlapping known annotation on the same strand, (4) remove segments containing fewer than 5 probes, and (5) remove segments opposite known annotation if they had a log2 fold change less than 2, or there was detectable transcription on the opposite strand (see Materials and Methods for a detailed discussion). For the sake of consistency, we will now refer to our post-processed segments as clusters, as they may refer to one or more original segments. After segmentation and post-processing of the tiling microarray data, we identified 892 candidate clusters in the poly A+ RNA data (826 of which were intergenic) and 338 from the total RNA data (324 of which were intergenic). Our criteria in analyzing the microarray data were somewhat liberal, with the aim of being as inclusive as possible; however, we coupled this with more stringent criteria for subsequent validation by sequencing, with the expectation that many of these clusters identified from the tiling microarrays would not be subsequently validated. All subsequent analyses were done at the cluster level.
Following identification of these clusters from the tiling array data, we sought to validate them using sequencing. The same RNAs harvested for the tiling microarray experiments were used to generate cDNA libraries for ultra high-throughput sequencing on a Solexa 1G Genome Analyzer. Libraries were generated from double polyA purified RNAs (see Materials and Methods) from both the wild-type and the mutant strains, and were run on four lanes each of a Solexa flow cell. Reads that passed Solexa's software filters were aligned to the genome using ELAND, allowing up to two mismatches per read. For subsequent analyses, we retained only reads mapping to a unique location, and in total, we generated more than 50 million uniquely mappable reads across all four strains. The wild-type library generated a total of 14,103,067 uniquely mapped reads from four lanes, the Δrrp6Δlsm1Δpat1 mutant library generated 14,745,813 reads, the Δski2Δski8Δrrp6 mutant library 14,973,577 reads, and the Δxrn1Δrrp6Δlsm1Δpat1 mutant library 10,714,094 reads. Following an assessment of the inter-lane variation we combined data across lanes for each strain (see Materials and Methods and Figure S2).
In order to determine whether the sequence reads generated from the cDNA libraries contained sufficient coverage and depth of the transcriptome, we determined the coverage at each base within the following classes: Verified ORFs, Uncharacterized ORFs, Dubious ORFs, Introns, and Background regions. Background regions were defined as regions that were intergenic on both strands, with the following additional regions removed: novel regions identified in David et al., Davis and Ares, Miura et al., and Nagalakshmi et al. [25],[26],[37],[63], as well as putative novel regions identified in this study using the tiling array. For each of these categories we determined the percentage of total bases sequenced to a depth of 3 or greater (see Figure 2 and Figures S3 and S4). For comparison, we have included the publicly available data from Nagalakshmi et al [37].
Figure 2 demonstrates that with an increase in sequencing effort there would be a diminishing return in terms of percentage of bases sequenced to a certain depth. Figure 2 also illustrates that an increase in sequencing effort results in an increase in the percentage of bases sequenced from both background and intronic regions (see discussion). This is the case in our data as well as those of Nagalakshmi et al. This implies that any method for declaring a gene as “detected” must evaluate the data in the context of the reads observed in these regions.
Figure 3 shows ROC-like curves depicting the tradeoff between detecting ORFs and detecting background regions, as we vary the detection cutoff. These plots demonstrate that the choice of a detection cutoff imposes a sample specific tradeoff between detecting annotated ORFs and background regions. For subsequent analyses, we chose a cutoff corresponding to calling 20% of background regions detected. Using this cutoff, we detected on average 75% of the Verified ORFs across all four experiments.
A GO analysis [66] of the Verified ORFs that were not detected above background indicated a significant enrichment for ORFs whose gene products are involved in the cell cycle and sporulation. The lack of sporulation gene expression is not surprising, as the cells would not be expected to be undergoing sporulation under these conditions; as for cell cycle gene expression, presumably the salt shock shuts off the cell cycle, and those transcripts are no longer detected at these thresholds by the time we collected the cells (30 minutes after exposure to salt).
In addition, we also analyzed our sequence reads to look at the dynamic range of detected transcripts. By considering Verified ORFs (>50 unique bp) that were detectable above background in the sequence data, the most abundantly expressed transcript in every mutant, and the wild-type, in terms of number of mapped reads per unique base was that of HSP12 (YFL014W), which is known to be induced under conditions of osmotic stress. Its average number of reads per unique nucleotide was ∼400 in every case. The least abundant transcript was different in each mutant, but with an average number of reads per base of less than 1. Thus, transcript abundances of the Verified ORFs (as measured by sequencing) span at least 3 orders of magnitude (see Table S3 for read counts and RPKMs [33] for all annotated ORFs).
As another measurement of the validity of the sequenced libraries, we determined how many known introns we were able to detect by looking for reads that spanned exon-exon junctions. To detect these intron spanning reads, we identified those reads that mapped to the set of spliced genic ORFs but did not map to the unspliced genome. The wild-type and mutant libraries each generated sequence reads that map to exon-exon junctions, which, when combined, confirm splice junctions in 244 (86%) of the 284 known spliced ORFs reported in the current SGD annotation. In the most extreme case (RPL28) we saw 1399 reads that mapped to the exon-exon junction in the data from the Δrrp6Δlsm1Δpat1 mutant. Of those forty genes whose exon-exon junctions we failed to detect, two were in mitochondrial genes, and 16 were in Dubious or Uncharacterized ORFs. Of the remaining 22, six of the genes are expressed in meiosis, and fourteen have an initial exon of only a few residues. These were less likely to have been detected by our strategy, as we looked for reads that matched the ORF sequence and not the genome, which would have had to start at a few specific residues to be detected. Subsequent analysis, by inclusion of 5′ UTR sequence to capture such exon boundary spanning reads, was able to identify these remaining introns. Thus, only two Verified ORFs, YER014C-A/BUD25 and YPL075W/GCR1, which were not meiosis specific, failed to have reads detected that spanned their exon junctions. BUD25 is opposite two other Verified ORFs in the genome, while Nagalakshmi et al [37] also noted that they were unable to identify exon-exon boundary spanning reads for GCR1. Indeed, we were able to identify reads that spanned the 5′ exon-intron junction, and the 3′ intron-exon junction, suggesting that the intron is misannotated.
We then examined an integrated dataset consisting of our tiling array and sequencing data as well as data from other published high-resolution studies. Various statistics of the potentially novel transcripts were computed to determine our proposed changes to the set of transcripts produced from the yeast genome. Firstly, we required that a cluster had to contain at least 50% uniquely mappable bases. For every potential novel transcript identified by our microarray data in a particular mutant, we employed the following criteria to Solexa data originating from the same mutant to validate the transcript: (A) the transcript detectable above background level, (B) the transcript differentially expressed between the mutant and the wild-type, and (C) the transcript differentially expressed when compared to its surrounding regions (see Materials and Methods for detailed explanation of precise criteria and cutoffs used for determination of validity).
In addition, we analyzed our data for the presence of reads containing a putative poly A+ tail, which would allow us to infer both the strand of origin as well as a precise 3′ boundary, however, very few such reads were present in our dataset most likely due to our use of random priming as opposed to oligo dT priming.
Following validation of individual clusters, we determined which clusters were common across the different mutants and as well as our poly A+ and total RNA hybridizations. 240 of our validated clusters were found in data from only one microarray, 79 were found in 2, 26 in 3, and 20 were found in 4 or more of the six microarrays, resulting in 365 validated transcripts (see Table S1), identified by virtue of differential transcript abundance between one or more mutants and the wild-type strain. Of these, 204 were found exclusively in the poly A+ RNA, 86 were found exclusively in the total RNA fraction, and 75 were detected in both. Several of these overlap with novel transcripts identified in recent studies: 67 with David et al. [25], 116 with Miura et al. [26], 46 with Nagalakshmi et al. [37], and 43 with Davis et al [63]. Beyond these, our 365 validated transcripts includes 185 additional previously undescribed transcripts, which we were able to discover by down-regulating RNA degradation. The majority of these novel transcripts (140 of 185) were found and validated in a single mutant only, with only 45 of them being identified and validated on two or more mutants (Figure 4).
For each of the potential novel transcripts, their immediate surrounding regions were plotted (e.g. see Figures 5 through 10 and Figures S5 and S6) along with a track of the current annotation from SGD [18], and data from David et al. [25], Miura et al. [26], and Nagalakshmi et al. [37]. Additional tracks representing nucleosome positioning [67] and the degree of conservation between Saccharomyces cerevisiae and other closely related yeast species [68] were also plotted. In addition, the transcript's chromosome and its strand of origin are shown at the bottom of each plot. Six examples of transcripts unannotated in SGD and identified in this study can be seen in Figures 5 through 10, all of which are located in regions currently described as intergenic. Plots for all 365 currently unannotated transcripts identified in this study can be found in Figures S5 and S6.
Of the 185 transcripts novel to this study, more than 80% have an average conservation score lower than 85% of the Verified ORFs (see Figure 11, as well as Figures 5 through 9 for five such examples; see also Figure S7). This implies that the vast majority of these transcripts could not have been found using comparative genomics.
Figures 5 through 8 show four novel transcripts unique to this study that are all located in regions of the genome that show poor conservation across different Saccharomyces species, as indicated by the conservation track at the bottom of each plot. Both our tiling microarray data and our UHTS data clearly show that the transcripts in Figures 5 through 8 are only seen in the one or more of the mutant strains and not in the wild-type, which was the criterion that enabled us to identify them. Prior transcript discovery studies, however, were only able to identify transcripts that are present in the wild-type, and in Figures 5 through 8, there are no data from David et al., Miura et al., or Nagalakshmi et al. to suggest that they could detect these novel transcripts. In some cases the nucleosome track is suggestive of transcriptional potential, due to there being low occupancy immediately upstream of the potential transcript. In Figures 5 through 8 there is a nucleosome dip immediately upstream of the identified segment, which is frequently observed in connection with transcribed regions [67].
Figures 9 and 10 illustrate two examples of intergenic transcripts found in this study that have been found in at least one other study (we considered a transcript to be one found by another study if there was a 25% overlap between the transcripts on the same strand); one of these falls in a conserved region (Figure 10), while the other does not (Figure 9). Additionally, in both examples, it is clear in our UHTS data that these transcripts were present in the wild-type strain, though at lower levels than within our mutants, indicating that they could readily be detected in the other studies, as they indeed have been. Figure 9 shows a transcript on the Crick strand that is upstream of a verified ORF and is seen in all three of the other studies (though Nagalakshmi et al. do not call it). There is a large region of low nucleosome occupancy just upstream of it, suggesting that the region is indeed transcribed, and the transcript itself overlaps with the nucleosome dip of the downstream ORF, suggesting that this new transcript may play a role in the transcriptional regulation of the ORF downstream of it. Figure 10 shows a relatively long transcript (1,721 bp) on the Watson strand that is also seen in David et al. and Nagalakshmi et al. It is highly conserved and the presence of a nucleosome dip upstream suggests that this region is transcribed.
We analyzed all of our novel transcripts for potential open reading frames, to determine if any were likely to be protein-coding. In each case, the longest open reading frame was translated and blasted against the non-redundant protein dataset (nr) from GenBank. The shortest novel transcript identified was 47 nucleotides long (intergenic), while the longest was 1,869 nucleotides in length (also intergenic), though the longest ORF that it contains only has the potential to encode a peptide 80 amino acids in length. The longest ORF that we discovered within all of our novel transcripts was within an ∼438 bp transcript on the Watson strand of chromosome 7 (coordinates 23,339–23,777), with the potential to encode an 87 amino acid polypeptide. However, this potential peptide showed no significant similarity when BLASTed against the GenBank non-redundant protein dataset. The remaining longest ORFs within each novel transcript were all shorter, with no significant similarities to any known proteins. It is not clear whether this means they do not encode proteins, or whether they encode novel, short proteins, which are currently uncharacterized due to their low conservation. We also analyzed each of our novel transcripts for any matches to known RNA structures present in the RFAM database [69],[70], but none of the sequences showed matches to any RFAM entries.
Using our detected above background statistic, we sought to determine the percentage of recently published novel transcripts present in our sequencing data. It should be noted that non-detection based on our data does not imply non-existence of these transcripts due to the differing experimental conditions as well as the distinct assays. Using our wild-type data, we detected 18.1% of the 487 Nagalakshmi et al. transcripts, 43.7% of the 784 David et al. transcripts, and 16.3% of the 667 Miura et al. transcripts. Using our Δrrp6Δlsm1Δpat1 data, we detected 65.3% of the 176 Davis and Ares transcripts (see Table S2 and Table S2 for a discussion of which transcripts were used from each study).
In this study, we have clearly demonstrated that there is still much we do not know about the transcriptome of S. cerevisiae, despite its deserved reputation as the most well-characterized eukaryote. Unbiased genome-wide studies of the budding yeast transcriptome [25],[26],[37] have yielded a remarkable amount of information, regarding new transcripts, new introns, the presence and location of antisense transcripts, and corrections to the current annotation. As described here, we have utilized tiling microarrays in conjunction with “next-generation” technologies to sequence cDNA libraries, with which we generated more than 50 million uniquely mappable reads from a wild-type and four mutant strains. Using these data, we have identified and validated 365 transcripts, the majority of which are more abundant in one or more of the RNA turnover mutants than in the wild-type strain (with a minority being less abundant), all of which are currently unannotated in SGD. The functions of these new RNAs remain unknown, though it is possible that many of the newly discovered transcripts correspond to CUTs, which normally would have been targeted for degradation by the TRAMP complex, but have been stabilized in the mutant background. Others may correspond to novel functional transcripts. These novel transcripts do not contain long ORFs capable of encoding proteins with recognizable similarity to known proteins; it is not clear whether this means they do not encode proteins or whether they code for hitherto unknown proteins with no known homologs. They also do not contain any recognizable RNA structures found in the RFAM database.
While our work described here has much in common with the work described in David et al. and Nagalakshmi et al., our use of RNA turnover mutants resulted in the finding of an additional 185 novel transcripts that may have otherwise remained undiscovered. Miura et al.'s use of vector-capped cDNA clone libraries is powerful in that it has a single nucleotide resolution, as opposed to our tiling microarray resolution of 4 nucleotides, allowing these authors to map transcriptional start sites to the exact nucleotide, in a high throughput manner. The use of overlapping, but non-identical, techniques among all these studies (including this one) has resulted in an ever more detailed knowledge of the yeast transcriptome.
In our approach, we utilized a high-throughput discovery and validation pipeline. Clearly, much work needs to be done to characterize and understand the transcripts discovered here as well as those discovered in previous studies, however a first step in characterizing the transcripts is localization and then validation. In our computational analysis we employed a strategy of being lenient in identification of putative novel transcripts (differentially expressed at 0.5 on the log2 scale). This was followed by a strict validation step (at our thresholds, on average 75% of annotated Verified ORFs were detected in our 3 mutant experiments as described by the ROC-like curves in Figure 3). Many (∼55%) of the clusters found in the microarray analysis were not validated by these stringent thresholds. These tended to be shorter, be less differentially expressed and included many clusters that were less abundant in the mutants as compared to wild-type. By using distinct assays with rigorous criteria for transcript validation, we have elucidated more of the regions of the yeast genome that are transcribed.
In our attempt to find low abundance and transient transcripts by restricting our search to transcripts that were present in differential relative abundance in our RNA processing mutants, we may have missed transcripts that are present in the mutant and the wild-type at the same abundance. This was a caveat we had to consider in the pursuit of transcripts that we believed would otherwise be difficult to detect, and the discovery of 185 novel transcripts despite the work of other comprehensive genome-wide transcriptome studies shows that our strategy was a fruitful one. By utilizing the strand-specific tiling array were able to localize transcripts to their strand of origin, something that was not possible (without introducing a 3′ bias to the data by priming the labeling reaction with oligo-dT) with the current protocols for RNA-Seq using the Solexa 1G Genome Analyzer. It is likely that modified protocols will soon address this shortcoming, and indeed such protocols for the ABI SOLiD sequencing system have been recently published [71].
We can now ask the important and obvious question: has the yeast transcriptome been completely described, and what does completion mean? It is possible that if we sequence deeply enough, we may observe that every nucleotide within the genome is transcribed at some level (see Figure 2), though clearly this is not a strict enough criterion to allow us to identify a transcribed segment. The genome-wide studies that have set out to discover new transcripts in yeast in an unbiased fashion have so far used a limited set of experimental conditions. Thus, it seems likely that deep sequencing of RNA from dozens of possible conditions (which must be carefully chosen to span as much of the “expression space” as possible) will yield yet more new transcripts, or show new variations in existing ones. It will be of particular interest to profile all of these novel transcripts under a variety of conditions to see how they are regulated and co-regulated, as well as to determine whether they encode proteins or functional RNAs, and whether their absence results in a detectable phenotype.
Since many of the recently discovered transcripts (including those in this study) have been found in regions of the genome where there is little or no sequence conservation (though the conservation scores from Siepel et al. [68] do not indicate whether these regions are evolving neutrally, or under positive selection), it will be informative to profile different and diverse strains of S. cerevisiae to determine if these transcripts are ubiquitous within the species, and to determine whether the syntenic (but non-conserved) regions within closely related species within the Saccharomyces sensu stricto are also transcribed. With such data, we can hope to discover and hopefully appreciate not only how each of these species are related to one another, but also how their transcriptional potential and networks have diverged.
Since the landmark publication of the S. cerevisiae genome sequence 12 years ago, more than 25,000 research publications on yeast have appeared, yet we are still adding to our knowledge of the transcriptome of S. cerevisiae. While arguably the most well-understood eukaryote, we still do not have a complete understanding of such a fundamental concept as “what and where are all of its genes.” New technologies such as high resolution tiling microarrays and ultra high-throughput sequencing are opening up new avenues of research, and it is clear that the quantity of data that these technologies allow us to generate will only increase. This study (and others like it) underscores how much work remains to be done in understanding and cataloging the transcriptomes of even the most well-studied model organisms.
All deletions were created in a diploid Saccharomyces cerevisiae strain which was created by crossing strains FY23 and FY86 [72], which are isogenic to the sequenced strain S288C and carry the auxotrophic markers: his3-Δ200, leu2-Δ1, trp1-Δ63, and ura3-52. All deletions were created using the Geneticin antibiotic resistance marker, utilizing the system described in [73]. Specifically, primers specific to regions to be deleted by homologous recombination were designed to utilize the plasmid pFA6-kanMX6 as a PCR template in order to replace the regions of interest with the gene encoding for resistance against the antibiotic Geneticin.
PCR was performed (see Table 2 for primers), generating approximately 1.5 kb DNA fragments in agreement with the size of the Geneticin resistance gene, which were then transformed using standard lithium acetate transformation techniques into the diploid cells grown in YPD at 30°C at mid-log phase. Cells were selected on YPD agar plates with 300 µg/ml working concentration of Geneticin. Deletions were confirmed by PCR (see Table 2 for primers) and the diploids were sporulated and their tetrads dissected to generate haploid segregants carrying the deletions of interest.
Different deletions strains were mated to generate diploids, which were then sporulated and tetrads were dissected. Because only the Geneticin marker was used to generate these deletions, PCR analysis was used to confirm all newly generated double mutant strains. The process was repeated to generate the triple and quadruple mutants (see Table 1 for resulting strains used in this study). Some deletion combinations could not be generated, suggesting they are synthetically lethal, and thus were not used in this study. For instance, Δxrn1 and Δski8 are synthetically lethal, as any attempt to combine strains with these deletions was unsuccessful. Haploid strains exhibiting phenotypes suggesting the accumulation of suppressor mutations were not used for further study. Originally the decapping factor DHH1 and the Ski complex component SKI3 were selected to be included, but strains carrying either Δdhh1 or Δski3 showed a propensity to accumulate suppressor mutations when combined with other deletions from this study and thus were dropped from the analysis. The Affymetrix tiling array data as well as the sequencing data confirmed that there was no expression signal corresponding to the genetic loci of the deleted genes.
Our unpublished studies suggested that among two dozen or so different conditions that we have assayed, exposure to high salt (0.8 M NaCl) results in the expression of the greatest fraction of known and novel transcripts, and thus was chosen as the experimental condition to use to find previously unannotated and low abundance transcripts. Cells were grown at 30°C in YPD to approximately 1×107 cells/ml as determined by a Beckman Coulter Z2 Particle Count and Size Analyzer. 1.6 M NaCl (in YPD) was added in an equal volume of YPD prewarmed to 30°C (final concentration 0.8 M). Cells were harvested after 30 minutes by filtration, frozen in liquid nitrogen, and kept at −80°C until RNA extraction and purification.
RNA was extracted from the cells using a slightly modified version of the traditional hot phenol protocol [74] followed by ethanol precipitation and washing. Briefly, 5 ml of lysis buffer (10 mM EDTA pH 8.0, 0.5% SDS, 10 mM Tris-HCl pH 7.5) and 5 ml of acid phenol were added to frozen cells and incubated at 60°C for 1 hour with occasional vortexing, then placed on ice. The aqueous phase was extracted after centrifuging and additional phenol extraction steps were performed as needed, followed by a chloroform extraction. Total RNA was precipitated from the final aqueous solution with 10% volume 3 M sodium acetate pH 5.2, and ethanol, and resuspended in nuclease-free water.
All microarray analyses were carried out using Affymetrix GeneChip S. cerevisiae Tiling 1.0R Array (Reverse) (part number: 900645) for Watson strand expression or GeneChip S. cerevisiae Tiling 1.0F Array (Forward) (part number: 520286) for Crick strand expression.
The arrays each contain more than 2.5 million perfect match probes, which are offset from one another by 4 bases across the genome (21 bp overlap). Thus, each residue in the genome is interrogated on average by 6 oligonucleotide probes.
Total RNA samples were prepared following the protocol exactly as described in David et al. [25].
PolyA RNA samples were prepared as follows. 500 µg of total RNA were PolyA purified using Qiagen Oligotex suspension to produce approximately 10 µg of PolyA RNA as determined by OD260/280. 2 µg of the PolyA purified RNA were then used in the generation of cDNA as per Affymetrix First Strand and Second Strand Synthesis protocols utilizing a T7-Oligo(dT) as the primer for the First Strand, followed by in vitro transcription to generate biotin labeled cRNA, as outlined by Affymetrix protocols. The cRNA was fragmented as described by Affymetrix, and then sent for hybridization and scanning by the PAN facility at Stanford (http://cmgm.stanford.edu/pan/) according to standard Affymetrix protocols.
Our goal was to identify short-lived transcripts based on measured intensities of probes tiling the genome. It is well known that probe affinities significantly bias the relationship between measured intensity and actual transcript abundance. In David et al. this was addressed by effectively forming log ratios between wild-type and genomic DNA hybridization. In order to highlight the changes between mutants and wild-type transcription and to reduce the effect of probe affinities we formed log ratios between mutant and wild-type intensities. This approach has the same effect on probe affinities as the approach used by David et al., see Figures S1 and S2.
The probes on the tiling array were mapped to the yeast genome, as downloaded from the Saccharomyces Genome Database on May 19th 2008, using MUMmer [75]. Only perfect match (PM) probes mapping to a unique region were retained for further analysis. For each mutant RNA hybridization, log ratios of mutant PM intensities to wild-type PM intensities were calculated.
The resulting data were segmented using the ‘segment’ function in the R package ‘tilingArray’ [64] from Bioconductor release 2.1, which performs a simple change-point analysis. The log ratios of mutant compared to wild-type for total RNA and poly A+ purified mRNA extractions for each mutant and chromosome strand were segmented separately. An open question in any segmentation analysis is the selection of the number of segments. We followed Huber et al. (2006) in using the Bayesian information criterion (BIC) penalized log-likelihood, noting that this tends to overestimate the number of segments (see below).
Following the segmentation we were left with a set of segments for each of the three mutants and two RNA sample types (total RNA or poly A purified RNA). Our analyses indicated that transcripts are often split into a number of segments due to various artifacts of the array data (outliers, incomplete probe-affinity correction, cross-hybridization). At this stage, we wished to both join appropriate segments into adjacent co-expressed segments (clusters) as well as filter out a priori uninteresting clusters. The pipeline for constructing clusters from segments and producing a set of putative clusters to be validated using the sequencing data worked as follows:
This process resulted in a set of putative clusters that were subsequently considered for validation by Solexa sequencing.
In order to generate libraries for the Solexa platform, various reagents and kits were required. At the time that these experiments were performed, Illumina did not have an RNA-Seq specific kit, and thus parts of various kits were utilized. Note that not all of the reagents from the kits provided by Illumina were used, as these kits were adapted for use in the protocol below and not necessarily used as described in the instructions that came with the kit. They are as follows:
For protocols desiring PolyA purified RNAs:
Also required was a magnetic stand that can accommodate 1.5 ml microcentrifuge tubes. The protocol as described below was done using DNase/RNase certified free siliconized 1.5 ml microcentrifuge tubes.
Strains used for our UHTS experiments are GSY147 and GSY1289 (see Table 1). GSY147 was derived from DBY10146 (a gift from David Botstein) (which itself was derived from an FY background [72]) which was backcrossed by Katja Schwartz to FY2 and FY3 [72] to generate a wild-type S288C strain that had no auxotrophies or mutations.
Two consecutive purifications using oligo dT conjugated magnetic beads were performed as follows. 100 µg of Total RNA were diluted in a final volume of 100 µl water and heated at 65°C for two minutes and then placed on ice. 200 µl of beads were equilibrated by two consecutive 100 µl washes in binding buffer (mixed gently by hand), using a magnetic stand to separate the beads from the buffer. The beads were then resuspended in 100 µl of binding buffer. The RNA was added to the beads, and the tube was mixed gently by hand for 5 minutes at room temperature and then placed on the magnetic stand to separate the beads from the supernatant. The supernatant was discarded, and the beads underwent two consecutive washes with 200 µl washing buffer. The beads were resuspended in 10 mM Tris-HCl pH 7.5, and the tube was heated at 80°C for two minutes and then immediately placed on the magnetic stand where the supernatant was transferred to a new tube. The beads were saved and prepared for the second round of PolyA purification by washing them once with 200 µl washing buffer. The entire process was then repeated once for a second round of purification, beginning with the dilution of the RNA and the denaturing of the RNA secondary structure.
PolyA purified treated RNA samples were then fragmented to ensure an unbiased binding of the random hexamers during cDNA synthesis. 5× Fragmentation Buffer (200 mM Tris Acetate pH 8.2, 500 mM Potassium Acetate, 150 mM Magnesium Acetate) was made, of which 5 µl was added to the RNA sample, and the total reaction was brought up to 25 µl. The sample was heated at 94°C for 2.5 minutes and immediately placed on ice. The sample was then run through a G-50 spin column that has been equilibrated with 3×400 µl of nuclease free water to remove ions from the fragmentation. The sample was concentrated to 10.5 µl with a Micron filter.
First Strand Synthesis:
10.5 µl of fragmented RNAs were transferred to a PCR tube and 1 µl of random hexamer (3 µg/µl) was added. The tube was heated to 65°C for 5 minutes and then placed on ice. The following reagents from the Illumina kit were then added: 4 µl 5×1st strand buffer, 2 µl 100 mM DTT, 1 µl 10 mM dNTP, and 0.5 µl RNaseOUT (40 U/µl). The tube was mixed and left at room temperature for 2 minutes. 1 µl SuperScript III (200 U/µl) was added, and the sample was placed in a thermocycler with the following program: 25°C for 10 minutes, 42°C for 50 minutes, 70°C for 15 minutes, 4°C hold.
Second Strand Synthesis:
The first strand synthesis reaction was transferred to a 1.5 ml siliconized microcentrifuge tube and placed on ice. 61 µl nuclease free water was added to the sample, along with the following reagents from the Illumina kit: 10 µl 2nd strand buffer, 3 µl 10 mM dNTPs, 1 µl RNase H (2 U/µl), and 5 µl DNA Pol I (10 U/µl). The sample was vortexed and placed in an Eppendorf Thermomixer R (set at 16°C and programmed to spin at 1400 rpm for 15 seconds and stand for 2 minutes) overnight (minimum 2.5 hours).
The newly synthesized cDNA was purified with a QIAquick PCR spin column as per Qiagen protocols and eluted in 30 µl EB solution.
The following reagents from the Illumina kit were added to the 30 µl sample as follows: 45 µl nuclease free water, 10 µl T4 DNA ligase buffer with 10 mM ATP, 4 µl 10 mM dNTPs, 5 µl T4 DNA polymerase (3 U/µl), 1 µl Klenow DNA polymerase (5 U/µl), 5 µl T4 PNK (10 U/µl). The sample was vortexed and incubated at 20°C for 30 minutes. Afterwards, the sample was purified with a QIAquick PCR spin column as per Qiagen protocols and eluted in 32 µl EB solution.
The following reagents from the Illumina kit were added to the 32 µl sample as follows: 5 µl Klenow buffer, 10 µl 1 mM dATP, and 1 µl Klenow 3′ to 5′ exonuclease (5 U/µl). The sample was vortexed and incubated at 37°C for 30 minutes. Afterwards, the sample was purified with a MinElute spin column as per Qiagen protocols and eluted in 10 µl EB solution.
The following reagents from the Illumina kit were added to the 10 µl sample as follows: 25 µl DNA ligase buffer, 2 µl adaptor oligo mix, and 5 µl DNA ligase (1 U/µl). The sample was vortexed and incubated at 25°C for 15 minutes. Afterwards, the sample was purified with a MinElute spin column as per Qiagen protocols and eluted in 10 µl EB solution.
The 10 µl sample was loaded onto a 1% TAE agarose gel at least one lane away from a 100 bp ladder. The sample was run sufficiently far enough and a gel slice corresponding to approximately 200 bp+/−50 bp was excised out of the gel with a scalpel (note that no cDNA may be visible on the gel). The cDNA was purified using a Zymo Research Zymoclean Gel DNA Recovery Kit and eluted in 10 µl nuclease free water.
The 10 µl sample was transferred to a PCR tube. The following reagents from the Illumina kit were added to the 10 µl sample as follows: 27 µl nuclease free water, 10 µl 5× cloned Phu buffer, 1 µl oligo 1.1, 1 µl oligo 2.1, 0.5 µl 25 mM dNTPs, 0.5 µl Phu polymerase. The sample was then run on a thermocycler using the following program: 98°C hold for 30 seconds, 98°C for 10 seconds, 65°C for 30 seconds, 72°C for 30 seconds, 72°C hold for 5 minutes, 4°C hold, for 50 cycles. The sample was purified with a QIAquick PCR spin column as per Qiagen protocols and eluted in 30 µl EB solution. The sample was then run through a G-50 spin column that had been equilibrated with 3×400 µl of nuclease free water to remove any remaining unincorporated nucleotides that would interfere with the concentration determination of the library. The DNA was concentrated through the use of a Speed Vac until the final volume of the library was 10 µl. The cDNA was quantified using a Nanodrop. A concentration range between 10–100 ng/ml final concentration of an RNAseq library is required for good quality sequencing. The sample was then sent for sequencing in the Genetics Department Solexa machine at Stanford.
Sequence reads that passed Solexa's quality filters were aligned to both the yeast genome and the spliced yeast ORF set (allowing up to 2 mismatches), downloaded from the Saccharomyces Genome Database (SGD) [76] on May 19th, 2008, using ELAND, which is part of the Solexa analysis pipeline [77] (we used version 0.3.0). Only reads mapping uniquely to the genome were retained.
We examined the goodness of fit for a simple Poisson model described below, using the chi-square goodness of fit statistic (see [78]). QQ-plots of the observed statistic for each known gene against the theoretical distribution are shown in Figure S2 and show a remarkably good fit. Based on this model, we aggregated data for each strain across the multiple lanes on the Solexa flow cell.
In order to validate each putative transcript identified by tiling array data analysis, we investigated the following three criteria:
An important consideration in all subsequent analyses was that certain areas of the genome are unmappable due to repeated sequences. We defined a base as non-unique if the 25mer starting at that position occurs elsewhere in the genome. We excluded all such bases from consideration in subsequent analyses.
We applied the detected above background statistic described above with a cutoff of .8. Results are available in Table S2.
All raw data have been deposited in the GEO database with accession number GSE11802.
|
10.1371/journal.pbio.1000489 | Broca's Region: Novel Organizational Principles and Multiple Receptor Mapping | There is a considerable contrast between the various functions assigned to Broca's region and its relatively simple subdivision into two cytoarchitectonic areas (44 and 45). Since the regional distribution of transmitter receptors in the cerebral cortex has been proven a powerful indicator of functional diversity, the subdivision of Broca's region was analyzed here using a multireceptor approach. The distribution patterns of six receptor types using in vitro receptor autoradiography revealed previously unknown areas: a ventral precentral transitional cortex 6r1, dorsal and ventral areas 44d and 44v, anterior and posterior areas 45a and 45p, and areas op8 and op9 in the frontal operculum. A significant lateralization of receptors was demonstrated with respect to the cholinergic M2 receptor, particularly in area 44v+d. We propose a new concept of the anterior language region, which elucidates the relation between premotor cortex, prefrontal cortex, and Broca's region. It offers human brain homologues to the recently described subdivision of area 45, and the segregation of the ventral premotor cortex in macaque brains. The results provide a novel structural basis of the organization of language regions in the brain.
| Broca's region is involved in many aspects of language processing in the brain. Such detailed functional diversity, however, is in contrast to its classical anatomical subdivision into only two cortical areas. Since the regional distribution of neurotransmitter receptors has been proven to be a powerful indicator of functional segregation, we revised the subdivision of Broca's region by analyzing the distribution of six different receptor types in the human brain. On the basis of these results, we propose a novel map of Broca's and neighboring regions with several, previously unknown areas. Moreover, a significant left-sided interhemispheric asymmetry of receptors was found, mainly for the cholinergic muscarinic M2 type. This asymmetry correlates with the well-known left-sided dominance for language. Finally, we present a model of the molecular organization of the anterior human language region and neighboring prefrontal and motor areas on the basis of similarities in their receptor patterns. This model contributes to our understanding of the relation between motor areas and classical Broca region. Our results are important for future studies of the functional segregation and the role of mirror neurons in the human brain, and are relevant for revealing homologies between human and macaque brains.
| For more than a century, Broca's region in the posterior part of the inferior frontal gyrus has been considered essential for speech production [1]. Effortful, telegraphic speech, impairment in articulation and melodic line, semantic and phonemic paraphasias are some of the symptoms associated with lesions of this region and subsequent Broca's aphasia [2],[3]. Mohr et al. [4], however, showed that an infarction limited to Broca's region does not cause chronic speech production deficits, and thus, differs from the clinical characteristics in Broca aphasia. They concluded that Broca's aphasia is observed after damage that extends beyond Broca's region. Broca's pioneering study illustrates on the one hand the power of the clinico-anatomical approach, i.e., relating language functions to a brain region, but also demonstrates its limitations. Consequently, the anatomical correlates of Broca's region cannot be identified by lesion studies alone.
According to Brodmann's map [5], the posterior part of the inferior frontal gyrus represents Broca's speech region. Brodmann's areas 44 and 45 at the opercular and triangular parts of the inferior frontal gyrus are its putative cytoarchitectonic correlates [6],[7]. Neighboring areas include premotor area 6 at the ventral precentral gyrus, dorso-lateral prefrontal areas 9 and 46, area 47 at the orbital part of the inferior frontal gyrus, and the anterior insula (Figure 1). Brodmann's map became a widely distributed anatomical reference for the interpretation of functional imaging studies although it represents only a schematic 2-D sketch of a putative “typical” human brain; i.e., it considers neither intersubject variability in brain anatomy nor interhemispheric asymmetries.
In contrast to the rather simple parcellation of the inferior frontal lobe shown in Brodmann's map, recent functional imaging studies suggest a complex segregation of Broca's region and neighboring areas of the inferior frontal cortex [8]–[16]. The whole region is involved in various aspects of language including phonological and semantic processing, action execution and observation, as well as music execution and listening (for an overview see e.g., [17]–[20]). A meta-analysis suggested that the opercular part (area 44) is particularly involved in syntactic processing [21]. However, activation during processing of syntactically complex sentences was also assigned to area 45 (triangular part) in studies using semantic plausibility judgment tasks or sentence picture-matching tasks [22],[23]. Other studies showed activation in area 44 in production [10] and comprehension [11],[12]. A recent study crossing the factors of semantics and syntax demonstrated that area 44 and more anterior areas (45/47) were active during sentence comprehension; area 44 carried the main effect of syntactic complexity independent of semantic aspects, whereas semantic relatedness, as well as its interaction with syntax, was located more anteriorly [24]. In addition, the deep frontal operculum was shown to be segregated from the inferior frontal gyrus during processing of syntactic sequences [25]. Finally, activations during motor tasks were also observed near Broca's region, e.g., during imagery of a motion task [14].
In many cases, the Brodmann map does not enable a localization of functional clusters of activations, in particular when they are found buried in the sulci, where architectonic borders have not been mapped. The localization of activation clusters using 3-D probabilistic cytoarchitectonic maps of areas 44 and 45 [26], and the adjoining motor areas [27], demonstrated that some of the clusters did not only overlap with area 44, but with the neighboring Brodmann area 6 [14].
A frequent finding in neuroimaging is a functional activation spot covering the adjoining border regions of two or more Brodmann areas, which cannot be assigned unequivocally to a cytoarchitectonic area. This situation may be caused by methodical problems of generating functional activation maps (e.g., spatial normalization to a template, smoothing, mislocalization of the BOLD signal due to venous flow) or by biological reasons (e.g., intersubject variability). Beside these arguments, it must also be asked whether Brodmann's map adequately represents the cytoarchitectonic segregation of this region, or whether uncharted cortical areas lead to the observed mismatch between functional data and cytoarchitecture as provided by Brodmann's map.
This line of argument is further supported by architectonic studies in the macaque brain. Recently, a new map of the ventral motor-prefrontal transitional region of the macaque cortex has been proposed; it showed that area F5 consists of three subareas: F5c, F5p, and F5a [28],[29]. Area F5 plays a major role in the mirror neuron system and has been interpreted as a putative correlate of human area 44 [30], whereas other authors disagreed [31],[32]. The complex segregation of the macaque ventral frontal cortex (and area F5 in particular) as compared to the rather simple subdivision of the human cortex provides further arguments to question Brodmann's parcellation.
Quantitative receptor autoradiography, a method that demonstrates the inhomogeneous regional and laminar distribution patterns of neurotransmitter receptor binding sites in the brain [33]–[35] has been proven to be a powerful mapping tool [34],[36]–[38]. The quantitative analysis of the density of multiple receptors in each cortical area highlights the regionally specific balance between different receptor types, and the differences between cortical areas. It reveals a functionally relevant parcellation, since receptors play a crucial role in neurotransmission [34].
Our aim was, therefore, to establish a receptor-based architectonic parcellation of the posterior inferior frontal cortex with focus on Broca's region, its right hemispheric homologue, and the adjoining areas on the frontal operculum, as well as the ventral premotor cortex. We studied the distribution patterns of six different receptor binding sites of four neurotransmitter systems: glutamatergic AMPA and kainate receptors, GABAergic GABAA receptors, cholinergic muscarinic M1 and M2 receptors, and noradrenergic α1 receptors in autoradiographs of eight human brains (Table 1). Neighboring sections were stained for cell bodies in order to identify the cytoarchitecture in this region. Observer-independent receptor and cytoarchitectonic mapping methods [34] combined with multivariate statistics were applied to analyze the similarity and dissimilarity of receptor patterns between the cortical areas. As a result, three previously unknown areas and a further segregation of the classical Broca areas 44 and 45 were found. The study leads to a new organizational concept of the cortical areas in Broca's region. It demonstrates that motor cortex, Broca's region, and prefrontal areas differ in their regionally specific receptor expression patterns, and thus in their signal processing properties.
Eight architectonically defined cortical areas were identified in the posterior inferior-frontal and precentral cortex. In addition to the Brodmann areas 44, 45, 4, 6, and 47, three new areas, areas op8 and op9 in the frontal operculum and area 6r1 in the ventral part of the precentral sulcus (Figure 2), were found and delineated by quantitative cytoarchitectonic and receptor architectonic mapping (Figure 3). The Brodmann areas 44 and 45 could be subdivided into 44d and 44v, as well as 45a and 45p. Furthermore, five new areas adjoining our region of interest were identified, but not completely delineated in the present study: areas 6v1 and 6v2 as parts of premotor area 6, area ifs1 located in the inferior frontal sulcus, and areas ifj1 and ifj2 located at the junction of the inferior frontal and the precentral sulcus (Figures 4–7).
Both opercular areas op8 and op9 were dysgranular, i.e., they showed a faint but recognizable layer IV (Figure 2). In this respect, they were similar to area 44 [26], but different from area 45, which has a well-developed inner granular layer (i.e., granular type of a cortical area). Sporadically, large pyramidal cells were found in layer III of area op8; they were smaller, however, than those of area 44. The columnar and laminar arrangement was less regular in area op8 than in area 44. Compared to the dorsally adjoining area 45, op9 showed a higher cell density, and a less regular cellular distribution. Layer III of area op9 contained pyramidal cells that were smaller and less frequent than those in area 45.
In contrast to the neighboring, purely agranular ventral area 6, area 6r1 was almost dysgranular; it displayed a subtle layer IV (Figure 2). In comparison with rostrally adjacent, typical dysgranular area 44, layer IV of area 6r1 was even thinner and not continuous. Large pyramidal cells in deep layer III were found that were similar to those of areas 6 and 44. Similar to area 6, the laminar differentiation of area 6r1 was weak, i.e., all cortical layers from layer II to VI showed an approximately similar cell packing density.
Receptor architectonic borders were identified by differences in density and lamination patterns of the receptor binding sites using an observer-independent method (Figure 3; Table 2) [39].
Area 44 was divided by receptorarchitectonic differences into two areas—a more dorsal 44d and a ventral 44v. Additionally, area 44v appears more posterior than 44d, and 44d reached out to more anterior levels than 44v. Most pronounced differences between both areas were found in muscarinic M2, AMPA, and α1 (Figures 4b, 4c, 7 and 8) receptors; the remaining receptors and the cytoarchitecture did not clearly separate these two areas (Figures 7 and 8).
The posterior border of area 44v with caudally adjacent area 6r1 was particularly well delineated by kainate (Figure 3b), GABAA, and α1 receptors. The supragranular layers of area 44v have considerably higher densities of glutamatergic AMPA and GABAergic GABAA receptors compared to those of the adjacent area op8 (Figure 4c and 4d). The borders between areas 44v and op8 were found at precisely the same localization in all receptor types indicative of this border (Figure 4b–4d).
The dorsally adjacent area of the inferior frontal junction region (ifj1) had lower receptor densities of M2, AMPA, and GABAA receptors than area 44d (Figure 4b–4d).
The border between area 45 and area 44 was detected by all receptors. The receptor densities revealed a subdivision of area 45 into an anterior (area 45a) and a posterior (area 45p) part, indicated by a lower density of M1 and AMPA receptors (Figure 5c and 5e) in the supragranular layers of 45p compared to 45a. Furthermore, the receptor density of the noradrenergic α1 receptor in 45p was lower than in 45a (Figure 8). The laminar distribution pattern in area 45 was similar to that of area 44 (Figure 5), but lower mean (averaged over all cortical layers) densities of the α1 (Figure 8), AMPA, and M1 (Figure 5) receptors clearly separated 45p from 44d. 45p had a higher concentration of M2, kainate, and α1 receptors than the dorso-rostral neighboring area ifs1 (Figure 4f–4h). The ventral border of area 45p with area op9 was indicated by higher M2 and kainate receptor densities (Figure 4f and 4g). 45a had higher M1, kainate, AMPA (Figure 5c–5e), and α1 receptor densities in the supragranular layers than the rostrally adjacent prefrontal cortex.
The border between area 6r1 and area 6v1 was revealed by higher M2 (Figure 6) and lower α1 (Figure 8) receptor densities in 6r1, whereas the border of area 6r1 with 44v was indicated by changes in kainate (Figure 3) and α1 receptors.
Cytoarchitectonic borders coincided with changes in the laminar distribution patterns of several or all receptor binding sites. For example, the border between areas 44 and 45 was identified in all six receptor types (AMPA, kainate, GABAA, M1, M2, and α1). However, not all borders could be demonstrated by changes in the laminar distribution patterns of all receptors; e.g., the border between areas 6r1 and 6v1 was reflected by changes in M2, α1, and kainate receptors, but less well by GABAA and M1 receptors. The border between areas 44d and 44v was labeled by α1 (Figure 8) and muscarinic M2 (Figure 7) receptors, but less visible in the autoradiographs of kainate and GABAA receptors (Figure 7).
The topography of areas and their spatial relationship is illustrated in a series of four coronal sections of a complete hemisphere (Figure 8). The border between ventral area 6 and caudally adjoining area 4 was located in the anterior wall of the central sulcus or the posterior portion of the precentral gyrus. Area 6 always occupied the free surface of the precentral gyrus. In some cases, it reached the lateral fissure. The receptor distribution showed a subdivision of the ventral part of area 6: two new areas, 6v1 and 6v2, were defined in addition to area 6r1 (Figure 8). Both areas were agranular, and showed the typical cytoarchitectonic laminar pattern of area 6 as described by Brodmann [5]. They differed, however, in their receptorarchitecture, e.g., by the noradrenergic receptor (Figure 8). Dorsally to 6r1, area 6v1 was found, which differed itself from the more dorsally adjoining premotor area 6v2 by a lower α1 receptor density (Figure 8, level 40). Rostrally of areas 6v1 and 6v2, area 6r1 was located within the precentral sulcus (Figure 6). Area 44 had common caudal borders with 6r1, 6v1, and 6v2. Medio-ventrally, 6r1 was adjacent to the new opercular area op6 (Figure 8, level 40). Area 6r1 separated area 44v from the ventral and more posterior parts of area 6 on the free surface of the brain (Figure 9).
44v adjoined 6r1 rostrally and covered the free surface of the opercular part of the inferior frontal gyrus. The position of the border between both areas varied in the anterior wall of the inferior precentral sulcus. The dorsal border of 44d was found in the ventral wall of the inferior frontal sulcus; the dorsal neighbors were ifj2 (at more caudal levels) and ifj1 (at more rostral levels). The ventral border of 44v with the opercular area op8 was located at varying positions deep in the frontal operculum (levels 32 and 26 of Figure 8).
Area 45 occupied the triangular part of the inferior frontal gyrus anterior to area 44. The border between areas 44 and 45 (Figure 5) was found either within the ascending branch of the lateral fissure or on the free cortical surface of the inferior frontal gyrus, e.g., between the diagonal sulcus and the ascending branch as illustrated in Figures 8 and 9. The ventral border of 45a with the opercular area op9 (Figures 4f–4h and 8 at level 19) was located at varying positions at the entrance to the Sylvian fissure. Areas op8 and op9 were regularly found ventral to 44v and 45a, respectively. Area 47 occupies the orbital part of the inferior frontal gyrus; it reached only the most rostral part of area 45, and was located rostral to area op9. Thus, the part of area 47 at the border to area 45 was most likely area 47/12l as described by Öngür et al. [40]. Area 45 bordered dorsally to areas within the inferior frontal sulcus; an example (area ifs1) is shown at level 19 of Figure 8. Area ifs1 differed in its receptor pattern from dorsally adjacent lateral prefrontal areas, and was restricted to the depths of the inferior frontal sulcus.
Each of the areas in the inferior frontal and precentral gyri showed a distinct receptor pattern as defined by six receptor types. A canonical analysis of receptor densities in all brains and hemispheres demonstrated differences and similarities in the receptor distribution pattern, and quantified receptor architectonic differences by multivariate distances (Figure 10). The hierarchical cluster analysis showed that the prefrontal area 47 was most different from all the other areas, i.e., areas 4, 6, 44, 45, 6r1, op8, and op9 (Figure 10). On subsequent levels of the cluster tree, area 4 differed from the remaining areas. In a next step, areas 6 and 6r1 appeared in one cluster, separated from areas op8, op9, 44, and 45. On the lowest level, a distinction into two subclusters was found: one comprising areas op8 and op9, and the other areas 44 and 45.
Interhemispheric differences in receptor densities were tested in three steps. First, we tested the left–right difference of all areas and receptors together using a discriminance analysis (Wilks Lambda). The densities differed significantly between the left and the right hemispheres: the overall p-value indicated a significant effect of hemisphere on the receptor density (p = 0.0091). Second, this overall interhemispheric difference (left over right) was mainly caused by the cholinergic muscarinic M2 receptors. It showed a left-larger-than-right asymmetry, as demonstrated by a subsequent univariate F-test (p = 0.003; Table 3). Left–right differences of each of the remaining receptors did not reach significance (p>0.05) if tested for each receptor type separately (Table 3). Third, if the areas were studied separately, M2 receptor densities of areas 44, 45, 6v1, and 6r1 were left > right, whereas area 4 showed an inverse pattern (Figure 11). Among these areas, the left–right difference for area 44 was most pronounced (p<0.05).
The cerebral cortex is subdivided into structurally and functionally distinct cortical areas. Areas 44 and 45 of the anterior speech zone, Broca's region, are supposed to represent the cytoarchitectonic correlates. Homologues of these two areas have been described in nonhuman primates. Comparative studies in macaque brains provided evidence, however, that a simple subdivision of this region into two areas is not sufficient and obscures the highly differentiated organization: (i) area 45 is parcellated into an anterior and a posterior part, which differ in their connectivity [41],[42]; (ii) the transitional zone from motor cortex to Broca's region contains areas within F5, possibly involved in different aspects of motor control and cognitive functions [28],[29]. Thus, we hypothesized that Broca's region of the human brain shows a more complex segregation than assumed until now.
The present study provided a combined analysis of six transmitter receptors and cytoarchitecture in Broca's region and the frontal operculum in order to test this hypothesis. The ventral premotor cortex and neighboring prefrontal areas have also been included in order to achieve a more comprehensive view of the inferior frontal cortex and its segregation from the neighboring motor and prefrontal cortex.
The selection of the areas of the present study aimed to consider the relevant regions, and to provide an anatomical correlate of different concepts regarding the functional segregation of the inferior premotor and neighboring Broca region. Activations in the vicinity of areas 44 and 45 have been reported not only in language, but also in motor tasks [13],[19], in experiments focusing on the integration of semantic information from speech and gestures [43], and other tasks requiring cognitive control [44],[45]. For an overview about the role of motor and premotor cortices in language processing see [46]. A recent study argued that the human action observation—action execution mirror circuit—is formed by the inferior section of the precentral gyrus plus the posterior part of the inferior frontal gyrus (plus the inferior parietal lobule) [47]. As a consequence, parts of the ventral area 6 and area 44 would belong to the mirror system. The inferior frontal cortex, including Broca's region and the ventral premotor cortex, has been conceptualized as a region representing complex, systemic dependencies, regardless of modality and use: Fadiga and coauthors have speculated that this capacity evolved from motor and premotor functions associated with action execution and understanding, such as those characterizing the mirror neuron system [20]. Others proposed that the role of this region is associated with complex, hierarchical or hypersequential processing [48]. Morin and Grèzes provided arguments, on the basis of a review of 24 fMRI studies examining activations in areas 4 and 6, that the ventral precentral gyrus with area 6, and not area 44, shares the visual properties of mirror neurons found in area F5 of the macaque brain [32].
The present receptorarchitectonic study resulted in a novel parcellation of the inferior frontal cortex. Three new areas, op8, op9, and the ventral precentral transitional area 6r1, were identified. Their borders were proven by significant changes in the laminar patterns of cyto- and receptorarchitecture using an algorithm-based method for the detection of borders [39]. Both opercular areas, op8 and op9, were separated from the dorsally adjoining areas 44 and 45 by their receptor distribution pattern. Previous studies have shown that areas of similar functions show similar receptor patterns and differ from those with other properties [34]. The higher the functional similarity between two cortical areas, the more similar are their receptor distribution patterns [35]; similarities in receptor architecture between areas 44 & 45 on the one hand, and areas op8 & op9 on the other, suggest a corresponding functional segregation. Indeed, functional representations of hierarchically and nonhierarchically structured sentences [25] correlate with the clustering based on receptor architecture: Whereas the deep frontal operculum (where op8 and op9 are located) was activated during the processing of nonhierarchically and hierarchically structured sequences, areas 44 and 45 were only activated during the processing of hierarchically structured sequences that mimicked the structure of syntactically complex sentences in natural languages [25]. A diffusion-weighted magnetic resonance imaging study revealed a separation of Brodmann area 44, 45, and the deep frontal operculum on the basis of differences in their connectivity [49].
The analysis of the receptor distribution patterns using hierarchical clustering supports the notion that areas 44 and 45 are closely related. It disagrees with those concepts, which attributed Broca's region solely to either area 44 [50] or area 45 [51], or to a cortical assembly combining areas 44 and 45 with area 47 [52].
Area 47 was most distinct from any of the analyzed areas as shown in the cluster analysis, thus suggesting a different functional involvement. The present data, therefore, imply that it is not meaningful to attribute activation clusters obtained in functional imaging studies to a region labeled as “45/47,” since these are two independent, structurally and functionally, completely different cortical areas.
The newly described area 6r1 showed cyto- and receptorarchitectonic features that places it in between area 44 and area 6. The area was called 6r1 in order to underline that it is located rostrally from premotor area 6; “1” indicates that this is the first area of a group of areas that we expect to be located rostrally to the precentral area 6; this belt of areas is located at the transition of the motor domain to the prefrontal cortex. Because of the higher microstructural similarity of area 6r1 with the classically described Brodmann area 6 than to 44, it was labeled as “6r1.” When analyzing the neighborhood of area 6r1 it became obvious, that the ventral part of area 6 consists of several areas, not yet described in the human brain. At least two more areas, 6v1 and 6v2, have been identified in the present study on the basis of receptor and cytoarchitectonic criteria. This finding supports data of a recent study analyzing the connectivity of the premotor cortex in the human brain [53]. Studies of the macaque brain already resulted in detailed parcellation schemes (for an overview of parcellation schemes see figure 1 in Belmalih et al. [28]). However, the topography and the sulcal pattern of the ventral frontal cortex differ considerably between macaque and human brains.
There are, on the other hand, also similarities of the present parcellation of the inferior frontal cortex with a parcellation found in a recent study in macaque monkeys [28]. The authors described an area F5a in the inferior arcuate sulcus bordering area 44. F5a may correspond to area 6r1 not only by its location but also by its cytoarchitectonic features. Even though area F5a is part of the agranular frontal cortex, it shows transitional features displaying granular cells as well as a relatively prominent layer V [28]. Further cytoarchitectonic studies will be necessary to compare the subdivisions of macaque F5 with human 6r1 in detail. If the abilities associated with Broca's region have evolved from premotor functions [54], area 6r1 may be interpreted as some kind of “transitional” area between the motor cortex and Broca's region. The identification of area 6r1 implies that area 44 does not border the ventral premotor area 6 over its full extent as supposed by other maps [5],[41]. Future cytoarchitectonic mapping studies would help to understand the extent of the inferior frontal lobe areas and its intersubject variability.
New areas were also found in dorsa-caudally adjacent areas of area 44. Two areas, ifj1 and ifj2, were distinguished (Figure 7), which are located immediately rostrally to premotor area 6. Both were found at the junction of the inferior frontal and the precentral sulcus, and, therefore correspond to the previously described inferior frontal junction region [55]–[57]. In contrast to earlier observations, however, here we identified two new areas instead of one, which had been hypothesized on the basis of functional imaging experiments, for example during task switching [56],[58]. The functional difference between ifj1 and ifj2 remains to be further elucidated.
Additional new neighboring areas (e.g., ifs1) were located in the depths of the inferior frontal sulcus where, according to Brodmann's map, areas 46 or 9 would be expected (Figures 4 and 7b at level 19). The present analysis of the complete coronal sections demonstrates that a series of small areas occupies the sulcus. These areas in the inferior frontal sulcus are different by their receptorarchitecture from the dorsally adjacent areas of the dorso-lateral prefrontal cortex, and, therefore, have not been labeled as areas 46 and 9, but ifs1, etc. The analysis and mapping of these new areas, again, represents an independent research project, which would exceed the present study.
We provided evidence for a further parcellation within area 44 and area 45. Differences in the laminar receptor distribution patterns of AMPA and M1 receptors argue for a subdivision of area 44 into a ventral and dorsal part extending earlier cytoarchitectonic findings [26]. A dorso-ventral subdivision of area 44 is a putative correlate of functional differentiation within this area as indicated by recent imaging studies: Molnar-Szakacs et al. [59] reported activations in the dorsal part of area 44 during observation and imitation of actions, whereas the ventral part was activated during imitation, but not during observation of actions. The ventral, but not the dorsal part, was activated during the imagery of movement [14]. Finally, an activation in the ventral part of area 44 was found for syntactic processing during language production [10] and comprehension [25], whereas the dorsal opercular part (where 44 is found) was involved in phonological processing [9].
The laminar receptor distribution patterns subdivided area 45 into an anterior and a posterior part on the basis of differences in the density of noradrenergic α1 M1, AMPA GABAA receptors. The subdivision of area 45 agrees with a recent study comparing the cytoarchitectonic organization in the human and macaque cortex [60]: Petrides and Pandya divided area 45 into a more anterior part (area 45 A) and a more posterior part (area 45 B, located anterior to area 44) using the width of layer II as the distinguishing feature (being narrower in area 45 A than in 45 B). This finding was further supported by demonstrating differences in connectivity [41].
The outcome of the present study is a considerably detailed parcellation of Broca's region and the immediately surrounding cortex. Some of the new units described here can be assigned to regions covered by Brodmann areas and defined by his nomenclatural system [5]. In such cases, we keep Brodmann's numbering system and define the new units by Brodmann's number and an additional letter and/or number (e.g., 6r1, 44a, 44p). In other cases, new cortical units could not be reliably assigned to a Brodmann area, e.g., op8 and op9. Since our new parcellation is based on an observer-independent approach and statistical tests of the significance of regional differences, we will call all cortical units “areas.”
The question, however, of how a cortical unit is defined as “area,” and what makes it special as compared to a unit called “subarea,” or an intra-areal specialization, remains. Examples of intra-areal specializations would be somatotopies in sensory and motor areas and ocular dominance columns, i.e., structures that are regionally specific to a certain degree, but subserve a common function. Currently, the concept of a “subarea” is vaguely defined, and is used inconsistently in the literature. Therefore, we adopt the term “area” throughout the article.
A central question to any study devoted to Broca's region is that of lateralization. Several studies have provided evidence that cytoarchitecture [26],[50],[61]–[64], fiber tracts [65], and macroscopical anatomy of this region are asymmetric [66]–[68]. For overviews see [69] and [70]. These structural asymmetries were interpreted as putative correlates of functional lateralization. The present study revealed significant interhemispheric differences in the receptor concentrations when all six receptor types were taken together. A subsequent analysis was performed in order to identify the receptor type that contributed most to this finding. The cholinergic M2-receptor showed the only significant left–right difference. Interhemispheric differences of receptors in Broca's region have not been reported up to now.
In conclusion, the novel parcellation of the ventro-lateral frontal cortex and Broca's region provides a new anatomical basis both for the interpretation of functional imaging studies of language and motor tasks as well as for homologies between human and macaque brains. It will, therefore, contribute to the understanding of the evolution of language. The analysis of the receptor distribution sheds new light on the organizational principles of this region. This direction is a further step from a rigid and exclusively cytoarchitectonic parcellation scheme as introduced by Brodmann 100 years ago [71] towards a multimodal and functionally relevant model of Broca's region and surrounding cortex.
Adult post mortem brains of body donors were removed from the skull within less than 24 h post mortem in accordance with legal requirements (Table 1). None of the subjects had clinical records of neurological or psychiatric disorders. Six hemispheres were dissected into coronal slabs of approximately 30 mm thickness (Figure S1). Tissue blocks containing the posterior part of the inferior-frontal cortex were dissected from six hemispheres of three brains and sectioned horizontally. The tissue was frozen and stored at −70°C. Serial sections (thickness 20 µm) were prepared at −20°C using a large-scale cryostat microtome. The sections were thaw mounted onto glass slides (Figure S1).
The following receptor binding sites were studied: glutamatergic AMPA and kainate receptors, GABAergic GABAA receptors, cholinergic muscarinic M1 and M2 receptors, and noradrenergic α1 receptors (Table S1). Alternating brain sections were incubated with the receptor-specific tritiated ligands only, the tritiated ligands, and respective nonradioactive compounds (for measurement of nonspecific binding), or were stained for the visualization of cell bodies [72]. Thus, a group of serial sections at the same sectioning level demonstrates the different receptor types, and the regional cytoarchitecture (Table S1; for details see Zilles et al. [35]). Since nonspecific binding was less than 10% of the total binding in all cases and receptor types, the total binding was accepted as an estimate of the specific binding. The labeled sections were coexposed with plastic standards of known concentrations of radioactivity (Amersham) to β-sensitive films. The films were developed after 10–12 wk of exposure depending on the receptor type, and digitized using the KS400 image analyzing system (Zeiss). The grey value distribution in the autoradiographs is nonlinearly correlated [35] with the local concentrations of radioactivity (Figure S1), which represent the regional and laminar distribution of receptor binding sites. Therefore, the known concentration of radioactivity of the coexposed standards (Figure S1d, bottom right) enables the nonlinear transformation of grey values into receptor binding site concentrations in fmol/mg protein (linearized images). For improved visualization of the regionally different receptor concentrations, the linearized images were contrast enhanced, smoothed, and pseudo-color coded in a spectral sequence (Figure S1f).
Neighboring sections were stained for cell bodies to demonstrate the cytoarchitecture. Rectangular regions of interest (ROIs) containing area 44 and 45 of Broca's region and neighboring areas were defined. Images (1,376×1,036 pixels; spatial resolution 1.02 µm per pixel) of the ROIs were acquired using a microscope equipped with a digital camera (Axiocam MRm, Zeiss) and a scanning stage. A high-resolution image of the total ROI was then assembled from the individual tiles employing the KS 400 system (Zeiss; Figure 3a I). Grey level index (GLI) images of the ROIs were calculated by adaptive thresholding with a spatial resolution of 16×16 µm. The resulting GLI image (Figure 3a II) represents in each pixel the local volume fraction of cell bodies [73].
Borders between cortical areas were identified in the receptor autoradiographs as well as in the cell body–stained sections using an algorithm-based approach and multivariate statistical analysis [39]. Therefore, laminar profiles of the GLI distribution were extracted in the cell body–stained sections using MATLAB-based software (MATLAB 7.2) (Figure 3a III). Laminar profiles were also obtained for the binding site densities in the autoradiographs (Figure 3b III). A feature vector was calculated for each profile, which described the shape of each profile, i.e., the cyto- or receptorarchitecture [39]. Differences in the shape of the profiles were quantified by a multivariate distance measure, the Mahalanobis distance. A subsequent Hotelling's T2 test with Bonferroni correction for multiple comparisons was applied for testing the significance of the distance. Profiles sampled from one and the same cortical area were similar in shape, resulting in small Mahalanobis distances. Profiles sampled from different sides of a cortical border differed in shape and resulted in large distances.
To improve the signal-to-noise ratio, distances were calculated not between single profiles, but blocks of ten to 20 adjacent profiles. The position of a significant maximum in the Mahalanobis function was interpreted as a cortical border, if it was found for different block sizes (Figure 3b VI), and if it was reproduced in a similar position in adjacent sections. These criteria allowed the rejection of borders caused by artifacts due to tissue processing, or blood vessels.
For each receptor, the density averaged over all layers of a cortical area was calculated in a set of sections/autoradiographs of each hemisphere separately. These mean receptor densities were averaged over all hemispheres resulting in a mean areal density value for each area and receptor type.
The density values of all six receptors studied were combined into a receptor feature vector for each area. A hierarchical cluster analysis (MATLAB 7.2) was performed in order to analyze receptor architectonic similarities and dissimilarities between the different areas (Euclidean distance, Ward linking). The higher the similarity between two cortical areas, the smaller was the Euclidean distance between their feature vectors.
A one-way ANOVA analysis (Systat 12) was performed to test for interhemispheric differences in receptor densities of all areas and receptors together. The factor “hemisphere” had two levels: left and right. Cases with missing values were excluded from the analysis. A post hoc univariate F test was performed in order to identify receptor types that contributed mostly to overall interhemispheric differences. Finally, we tested interhemispheric differences for each cortical area and receptor. The p-level was set to 0.05.
|
10.1371/journal.ppat.1003132 | Atomic Model of Rabbit Hemorrhagic Disease Virus by Cryo-Electron Microscopy and Crystallography | Rabbit hemorrhagic disease, first described in China in 1984, causes hemorrhagic necrosis of the liver. Its etiological agent, rabbit hemorrhagic disease virus (RHDV), belongs to the Lagovirus genus in the family Caliciviridae. The detailed molecular structure of any lagovirus capsid has yet to be determined. Here, we report a cryo-electron microscopic (cryoEM) reconstruction of wild-type RHDV at 6.5 Å resolution and the crystal structures of the shell (S) and protruding (P) domains of its major capsid protein, VP60, each at 2.0 Å resolution. From these data we built a complete atomic model of the RHDV capsid. VP60 has a conserved S domain and a specific P2 sub-domain that differs from those found in other caliciviruses. As seen in the shell portion of the RHDV cryoEM map, which was resolved to ∼5.5 Å, the N-terminal arm domain of VP60 folds back onto its cognate S domain. Sequence alignments of VP60 from six groups of RHDV isolates revealed seven regions of high variation that could be mapped onto the surface of the P2 sub-domain and suggested three putative pockets might be responsible for binding to histo-blood group antigens. A flexible loop in one of these regions was shown to interact with rabbit tissue cells and contains an important epitope for anti-RHDV antibody production. Our study provides a reliable, pseudo-atomic model of a Lagovirus and suggests a new candidate for an efficient vaccine that can be used to protect rabbits from RHDV infection.
| Rabbit hemorrhagic disease (RHD), first described in China in 1984, causes hemorrhagic necrosis of the liver within three days after infection and with a mortality rate that exceeds 90%. RHD has spread to large parts of the world and threatens the rabbit industry and related ecology. Its etiological agent, rabbit hemorrhagic disease virus (RHDV), belongs to the Lagovirus genus in the family Caliciviridae. Currently, the absence of a high-resolution model of any lagovirus impedes our understanding of its molecular interactions with hosts and successful design of an efficient anti-RHDV vaccine. Here, we use hybrid structural approaches to construct a pseudo-atomic model of RHDV that reveals significant differences in the P2 sub-domain of the major capsid protein compared to that seen in other caliciviruses. We identified seven regions of high sequence variation in this sub-domain that dictate the binding specificities of histo-blood group antigens. In one of these regions, we identified an antigenic peptide that interacts with rabbit tissue cells and elicits a significant immune response in rabbits and, hence, protects them from RHDV infection. Our pseudo-atomic model provides a structural framework for developing new anti-RHDV vaccines and will also help guide use of the RHDV capsid as a vehicle to display human tumor antigens as part of anti-tumor therapy.
| Rabbit hemorrhagic disease (RHD) is extremely contagious in adult rabbits and is often associated with liver necrosis, hemorrhaging, and high mortality [1]. It was first described in China in 1984 [2], and within a few years had spread worldwide [3]. RHD outbreaks still occur on almost every continent and cause significant mortality rates, being endemic in Europe, Asia, Africa, and Australia [4]. This disease has a significant impact on the rabbit industry and ecology [4].
The etiological agent of RHD is rabbit hemorrhagic disease virus (RHDV), which has a single-stranded, positive-sense, polyadenylated RNA genome of ∼7.5 kb [5]. Mature RHDV virions are spherical, non-enveloped particles with a T = 3, icosahedral capsid whose outer diameter varies between 32 and 44 nm and whose structure is defined by characteristic, cup-shaped depressions [6]. The only capsid protein present in RHDV, VP60, is composed of three domains, which include the N-terminal arm (NTA), the shell (S), and the protrusion (P), the latter of which is further divided into P1 and P2 sub-domains [7].
RHDV belongs to the genus Lagovirus of the family Caliciviridae, which also includes the genera Norovirus, Nebovirus, Sapovirus and Vesivirus [8], [9]. Previous structural studies of caliciviruses include three-dimensional (3D) cryo-electron microscopic (cryoEM) reconstructions of virus-like particles (VLPs) of Murine Norovirus (MNV, Norovirus) and Feline calicivirus (FCV, Vesivirus) at 8- and 16-Å resolution, respectively [10], [11], and determination of the crystal structures of the Norwalk virus (NV, Norovirus) capsid at 3.4 Å [12], native FCV virions at 3.6 Å [13], and native virions of San Miguel sea lion virus (SMSV, Vesivirus) at 3.2 Å [14]. CryoEM reconstructions of the RHDV VLP at 8 Å [15] and the native RHDV virion at 11 Å [7] have been computed and a Cα homology model of RHDV was built based on the VLP cryo-reconstruction by using the crystal structure models of SMSV and FCV [16]. However, a more complete atomic model of RHDV is still lacking. Furthermore, the P domain of VP60, which is responsible for antigenicity and binding to host tissue [17], varies considerably across different Caliciviridae species, and hence this stimulated us to crystallize and obtain a high resolution crystal structure of this domain to provide a model that is more reliable than could be gleaned from any homology modeling approach.
It is worth noting that noroviruses infect hosts by recognizing histo-blood group antigens (HBGAs) that are important host susceptibility factors [18], and RHDV also agglutinates human erythrocytes and attaches to epithelial cells in the upper respiratory and digestive tracts of rabbits by binding to HBGAs [19]. HBGAs have recently been shown to act as attachment factors that facilitate infection and RHDV isolates from six different genetic groups bind specifically to different HBGAs [20].
Here, we report a pseudo-atomic model of the RHDV capsid derived through a combination of X-ray crystallography, cryoEM reconstruction, and molecular dynamics flexible-fitting (MDFF) [21]. We find that RHDV VP60 has a P2 sub-domain that differs from other caliciviruses. Furthermore, our new model reveals that certain aspects of the P2 and NTA domain structures that were previously reported [16] need reinterpretation. We also examined the putative HBGA binding sites in RHDV by mapping isolate–related sequence variations onto the P domain structure. Finally, we show that a peptide derived from a putative HBGA binding site can interact with hosts and stimulate the production of virus antibody. The new, high-resolution model of a Lagovirus presented here provides a solid framework for developing an efficacious antigen presenting system. The model yields also new insights regarding the molecular mechanisms of RHDV-host interactions.
Highly purified RHDV virions (Figure 1A) obtained from the livers of infected domestic rabbits were used for crystallization trials and cryoEM studies (Figure 1B). Unfortunately, we were unable to obtain any crystals of RHDV suitable for X-ray diffraction owing to its propensity to degrade with time. From cryoEM micrographs (Figure 1B), consistent with previous observations [7], [22], two distinct classes of particles were observed: intact virions containing whole genomic RNA (high density inside) and “empty” virions containing sub-genomic RNA (low density inside). The presence of these two types of particles was confirmed by image classification (Figure S1A). The cryoEM structure of RHDV that we computed from ∼36,000 images of individual particles (Figure 1C and S1B) was estimated to reach a resolution limit of 6.5/4.8 Å (Figure S1C) based on Fourier shell correlation (FSC) cutoff thresholds of 0.5 and 0.143, respectively [23], [24]. Considerably more detail was resolved in this RHDV cryo-reconstruction compared to that in our previous one at 11 Å [7]. In addition, the resolution achieved in the RHDV inner shell (radii between ∼130 and 150 Å) reached 5.5 Å (FSC0.5; Figure S1C) compared to 7.0 Å (FSC0.5) for structural features at larger radii (between ∼150 and 220 Å). Central cross sections of the reconstructed 3D map taken perpendicular to the icosahedral 3-, 5-, and 2-fold axes show well-resolved densities in the inner shell compared to fuzzier densities at larger radii (Figure S1B), consistent with the protruding capsomers exhibiting high flexibility [7], [15]. All secondary structural elements in the VP60 S domain were clearly resolved and, in some regions, densities corresponding to residue side chains were evident (Figure 1D). Compared to reconstructions of the RHDV VLP at 8 Å [15] and the native virion at 11 Å [7], the present result represents the most detailed view of the RHDV capsid structure and this, along with results from X-ray crystallography, enabled us to build a reliable, pseudo-atomic model.
As shown previously [7], the RHDV capsid has an overall spherical shape, with a maximum outer diameter of 44 nm and an inner chamber with a diameter of 28 nm (Figure 1C). The asymmetric unit of the RHDV capsid consists of three, quasi-equivalent VP60 subunits (A, B and C) arranged with T = 3 icosahedral symmetry. The 180 VP60 subunits that comprise the capsid are organized as 90 dimers, each of which appears as an arch-like capsomer. Thirty C/C capsomers are located at the icosahedral two-fold symmetry axes and the remaining 60 A/B capsomers are located at pseudo (“local”) two-fold axes. Three A/B and three C/C dimers are positioned in alternate fashion around each icosahedral three-fold axis to form pseudo-six-fold arrangements, and five A/B dimers encircle each five-fold axis. Together, these capsomers produce a contiguous shell and 32 cup-shaped, surface depressions, the latter of which are a characteristic feature of the structure of all caliciviruses [25].
RHDV VP60 is subdivided into three domains, NTA (the N-terminal arm, a.a. 1–65), S (the shell, a.a. 66–229), P (the protrusion, a.a. 238–579) and a short hinge (a.a. 230–237) that connects S and P (Figure 2A). The S domain together with the NTA domain (a.a. 1–230) was cloned and expressed in E.coli, purified, and crystallized in space group C2. We solved the crystal structure of the S domain by molecular replacement and refined it to a resolution limit of 2.0 Å with final Rwork and Rfree values of 20.0% and 24.1%, respectively (Table 1). The NTA domain could not be traced owing to lack of electron density, though SDS-PAGE analysis of crystals did not exhibit any obvious protein degradation. This indicates that the NTA domain is inherently quite flexible in crystals. The S domain of RHDV shares high sequence homology with the S domains of other caliciviruses (Figure S2A) and folds into a canonical, eight-stranded, BIDG-CHEF β-barrel [26] (Figure 2B). The structure of the RHDV S domain superimposes quite closely with the corresponding S domains of FCV, SMSV, and NV (Figure S2B). The root mean squared deviations (r.m.s.d) of the Cα coordinates of the RHDV S domain compared to each of these three viruses are 1.51 Å (149 Cα), 1.41 Å (150 Cα), and 1.32 Å (153 Cα), respectively, suggesting that the structures of the inner shells of all caliciviruses are highly conserved.
The fragment (a.a. 228–579) that includes the entire VP60 P domain was expressed in a baculovirus system, purified, and formed crystals that belong to space group P212121. Its crystal structure (Figure 2C) was determined by molecular replacement, with the capsomer portion of the RHDV cryoEM density map used for initial phasing. This structure was refined to a resolution of 2.0 Å with final Rwork and Rfree values of 19.9% and 23.2%, respectively (Table 1 and Figure S3). The asymmetric unit of the crystal contains a dimer of P domains. The P domain of RHDV, like in other caliciviruses [12], consists of sub-domains P1 (a.a. 238–286, 450–466, 484–579) and P2 (a.a. 287–449 and 467–483) (Figure 2A, C and D). The P1 sub-domain of RHDV has a conserved fold compared to caliciviruses FCV, SMSV, and NV, with r.m.s.d values for the Cα coordinates of 1.53 Å (144 Cα), 1.49 Å (145 Cα), and 2.14 Å (134 Cα), respectively (Figure 2E). The P2 sub-domain has a predominant β-barrel core comprised of six anti-parallel β strands (β6-β7-β9-β5-β3-β11) folded in a Greek-key topology and a two-stranded β sheet (β12–β16), which are connected by seven loops (L1–L7) of various lengths and surrounded by two short helices (η3 and η4) (Figure 2C and D). The P2 sub-domains of RHDV, NV, SMSV, and FCV exhibit no obvious sequence homology (Figure S4), and the Cα coordinate r.m.s.d between the P2 sub-domain of RHDV and that of NV, SMSV, and FCV are 3.00 Å (38 Cα), 2.68 Å (123 Cα), and 4.32 Å (107 Cα), respectively. Although they share a consensus β-barrel core, the loop regions differ significantly (Figure 2E) and are expected to be a primary determinant of the host range for each particular virus.
The crystal structures of the S and P domains of VP60 were docked into the high-resolution cryoEM map to construct a pseudo-atomic model of the complete RHDV capsid. With the exception of a few loops, the S domain fit quite well into the density map (Figure 3A). Despite the absence of density for the NTA domain in the crystal structure of the NTA-S recombinant molecule (Figure 2B), a difference map computed by subtracting the fitted S domain model from the RHDV virion cryoEM map enabled us to build an ab initio model of the NTA domain (residues 30–65) (Figure 3A). At each three-fold axis of the virion, three A/B and three C/C dimers pack in alternate fashion via their S domains and clear densities at the interface of each dimer show that each NTA domain folds onto its cognate S domain (Figure 3A, S5A and B). The NTA domains of the B and C monomers form a network of interactions with a plug-like density (formed by residues 1–30) surrounding the three-fold axis (Figure 3A and B) as was also described previously [16]. Contacts formed by the NTA domains in the inner shell of the virion confirm the importance of this domain for virion assembly, which concurs with previous truncation [27] and insertion studies [16].
The folding back of NTA onto the S domain of the same VP60 subunit in RHDV is similar to that seen in NV [12], but differs from that in SMSV, where the NTA domain extends away from the cognate S domain to interact only with the S domain in an adjacent subunit [14]. The cryoEM density map of RHDV showed that the protruding regions of the A/B and C/C dimers only interact between the P2 sub-domains (Figure 3C and D), which is consistent with the crystal structure of SMSV [14]. However, the NV crystal structure shows that these dimers include P1-P1 as well as P2-P2 interactions [12].
Following initial rigid-body docking of the crystal structures of the S and P domains into the RHDV cryoEM map along with the modeled NTA segments, MDFF procedures [21] were used to build a complete, pseudo-atomic model of the capsid (Figure 3E). The refined model fits the cryoEM map very well for both P and S domains with apparently good consistence (Figure 3C and D, S5C and D). Furthermore, comparison of the MDFF-refined model with the initial rigid-body-fit model, showed that the local cross correlation coefficient between the atomic model and the cryoEM map improved from 0.473 to 0.634 (Table S1). The r.m.s.d between the initial model and MDFF-refined model is 2.45 Å. In particular, the local cross correlation coefficient for the S domain improved from 0.452 (before MDFF) to 0.673 (after MDFF). MDFF not only improved the fitting in the loop region around the 3-fold axis, but also closed the gaps between B and C subunits at the interface (Figure S5E and F). The improvements in local cross correlation coefficients for other domains are given in Table S1. Structural comparisons among the A, B and C monomers of the MDFF-refined model, when aligned to the P domains, revealed that large conformational changes accompany relative movements and rotations of the S domain with respect to the P domain (Figure 3F). The complete, pseudo-atomic model of the RHDV capsid exhibits the classic calicivirus features: an inner shell formed by 180 S domains and 90 protrusions formed by dimeric arrangements of the P domains (Figure 3E and Movie S1).
Next, we compared our current structural model of RHDV with the previously reported backbone model derived from the 8.0 Å VLP cryo-reconstruction and homology modeling [16]. Comparison of the three quasi-equivalent monomers (A, B and C) in the two models (Figure S6) showed that the relative positions of the P and S domains correspond closely to each other, but that nevertheless two significant differences are found. First, the NTA domain in the previous model extends to interact with the S domain in an adjacent subunit, whereas the current model shows instead that the NTA domain folds back onto its cognate S domain. Second, except for the β-barrel core motif, the loop regions in the P2 sub-domains differ completely between the two models. As a result, our higher resolution cryoEM map of the RHDV capsid (especially in the inner shell) and the crystal structure of the P domain together provide more accurate structural details about the NTA domain and P2 sub-domain. This detail lays a foundation for understanding how RHDV interacts with its hosts and how the virus displays a specific antigenic epitope.
The first step of viral entry in NV and RHDV infections involves recognition of HBGAs [18], [19]. Crystal structures of NV variants V387 and V207, bound with HBGAs, revealed that the binding sites in NV are located at the outer surface of the arch-like P dimers with both P domains contributing to the formation of the binding interfaces [18], [28]. Because the structure of the RHDV P domain bound with HBGAs is currently not available, we selected the crystal structure of NV variant V207 complexed with the non-secretor HBGA Lewis y (Ley) tetrasaccharide as a model (PDB code 3PUN) [18] to compare with our atomic model of RHDV VP60 (Figure 4A). The crystal structure of the NV V207 P dimer was superimposed onto the C/C capsomer of RHDV by aligning to one of the subunits. The relative positions of the two subunits within the dimer differ slightly between the NV V207 and RHDV models. The binding site of the Ley tetrasaccharide in the P dimer of NV V207 corresponds to loop L6 or L2 in the P domain of RHDV VP60 (Figure 4A). A surface representation of the NV P dimer shows that Ley tetrasaccharides bind to the outer portion of the dimeric interface between P domains (Figure 4B). However, this interface is completely different in RHDV (Figure 4C), and therefore, the RHDV capsomer likely utilizes distinct binding sites for HBGAs.
Though genetic diversity among RHDV isolates is far lower than that among isolates of other caliciviruses, it has been suggested that all current RHDV isolates could be assigned to one of six genetic groups and the binding specificities of HBGAs for those genetic groups have been the subject of intensive investigation recently [20]. We performed a multi-sequence alignment of VP60 among these six groups and found that seven regions of high variation (V1 to V7) distinguish these groups (Figure 5A). These regions all occur on the P2 sub-domain (Figure 5B and S7). Most significantly, these regions correspond to loops L1 to L7 in the P2 sub-domain (Figure 2 and 5). Thus, in addition to the antigenic variation contributed by these loop regions, at least some and perhaps all of these loops may give rise to different HBGA binding specificities. A relationship between variation regions and receptor binding specificity was also gleaned from the cryoEM structure of FCV bound with its receptor, fJAM-1 (feline junctional adhesion molecule 1) [11]. In addition, we found three cavities on the outer surface of the RHDV capsomer (labeled C1, C2 and C3 in Figure 5B), one or more of which might contribute to HBGA binding. Whether these are true binding sites awaits investigation by mutagenesis experiments.
Variation region V1 is a contiguous stretch of mostly hydrophilic residues on loop L1 (a.a. 304–314) (Figure 5A) and is highly flexible in crystals as evidenced by high crystallographic B-factors (Figure S8A and B). Given that L1 is the most exposed loop on the surface of the RHDV capsomer and that it lies juxtaposed to three putative HBGA binding pockets (Figure 5B), this loop is hypothesized to be a primary determinant of RHDV host interaction such that it represents an effective epitope in RHDV. Also, the sequence of this loop constitutes the most diverse region in VP60 in RHDV isolates (Figure S8C) and suggests that this sequence plays a critical role in defining RHDV antigenicity.
To test our hypothesis, we designed two peptides, NJ85 (a.a. 300–318) and NJ85Δ (missing 4 residues N308ATN311 of the loop L1 on the most exposed position of the capsomer), derived from the VP60 protein of the RHDV NJ85 isolate strain. Each peptide was synthesized with an N-terminally-labeled fluorescent isothiocyanate (FITC) and then used as a reagent to analyze receptor-binding activity in rabbit hepatocytes, primary splenocytes, and kidney (RK13) cells from healthy male, New Zealand white rabbits. Both peptides bound to the surfaces of the hepatocytes and primary splenocytes, but neither one bound to RK13 cells (Figure 6A). This suggests that the hepatocyte and splenocyte cells express receptors capable of binding both peptides, but that rabbit kidney cells do not, which concurs with previous studies on the specific tissue distributions of RHDV [29], [30]. This binding assay also suggested that at least one of the RHDV-host interaction sites resides at the top surface of the capsomer (i.e., in the loop). However, the four residues (308–311) at the top part of the capsomer, which vary the most across isolates (Figure S8C), unexpectedly did not affect interactions between RHDV and its host.
To explore whether the protruding loop L1 of the capsomer can function as an antigenic site and induce an effective host immune response, we coupled the NJ85 and NJ85Δ peptides with keyhole limpet hemocyanin (KLH) (KLH-NJ85 and KLH-NJ85Δ) and used these two constructs to immunize rabbits. Western blot and ELISA analyses showed that antibody titer induced by KLH-NJ85 is about ten-fold higher than that induced by KLH-NJ85Δ (Figure 6B and S9). The RHDV hemagglutination inhibition assay revealed that the inhibition titers of the serum raised by those two peptides were 1∶64 for KLH-NJ85 and 1∶32 for KLH-NJ85Δ, respectively (Figure S10). As a result, although the highly exposed four residues (a.a.308–311) at the top of the capsomer are not required for host cell interaction, they do elicit a strong immunological response from the host.
We next investigated whether antibodies raised by those two peptides could neutralize RHDV and protect rabbits. Fifteen rabbits were divided into three groups of five. The sera that were raised by KLH-NJ85 and KLH-NJ85Δ were diluted 32-fold. An aliquot of each (800 µL) was mixed with 256 hemagglutination units of RHDV and incubated at 37°C for 1 hour, and then the mixture was given to rabbits intranasally. The serum from specific pathogen free rabbits was used as a negative control. In the negative control group, one of the five rabbits died within 24 hours after inoculation, three of the five rabbits died within 72 hours, and the fifth succumbed within 96 hours. In the groups of rabbits inoculated with sera raised by KLH-NJ85 and KLH-NJ85Δ, the rabbits were continuously housed and monitored every 24 hours for 10 days and all ten rabbits survived (Figure 6C and Table S2).
Furthermore, virus challenge with RHDV displayed 100% immune protection in the two groups of rabbits vaccinated separately with KLH-NJ85 and KLH-NJ85Δ (Figure 6D). In a control group that was vaccinated in parallel with KLH, two of the five rabbits died within 48 hours after challenge whereas the other three succumbed within 72 hours. Each virus challenge experiment was repeated four times and yielded consensus results (Table S3). These experiments demonstrated that loop L1 in the P2 sub-domain of VP60 forms an epitope on RHDV, and peptides derived from this loop are sufficient to stimulate rabbits to produce antibodies that immunize them against RHDV infection.
It is noteworthy that previous structural studies of a norovirus/Fab complex suggested the two loops (A′-B′ and E′-F′) in the P2 sub-domain of MNV to contact antibody [10], [31]. When we superimposed the crystal structures of the RHDV and MNV P2 sub-domains, we found that the L1 and L5 loops of RHDV correspond, respectively, to the two loops in MNV (Figure S11). Consequently, our results with RHDV concur with those of at least one other calicivirus.
It is unclear why peptides NJ85Δ and NJ85 provide equal protection from RHDV challenge when titers of the sera elicited by them differ. Hence, we performed an immunological assay to determine the expression levels of cytokines in the sera raised by KLH-NJ85 and KLH-NJ85Δ. Four cytokines (IL 2, IFN γ, IL 6, and IL 10) were detected by ELISA. Specific pathogen free rabbit serum was used as a negative control. Except for IL 6, expression levels of IL 2, IFN γ, and IL 10 in sera raised in response to challenges by both KLH-NJ85 and KLH-NJ85Δ were higher than the negative control (Figure S12). It was known that high levels of IL 2 proliferate activated T cells and high levels of IFN γ activate macrophages, neutrophils, and NK cells, and then promote cell-mediated immunity for antiviral effects [32]. Also, high levels of IL10 promote B-cell proliferation and antibody responses [32]. As a result, though NJ85Δ and NJ85 lead to different titers of antibodies, both are able to stimulate similar levels of cytokine expression and activate a cell-immune response that allows rabbits to resist challenges from lethal doses of RHDV.
In this study, we used cryoEM methods to reconstruct the structure of the RHDV capsid to an overall estimated resolution limit of 6.5 Å (5.5 Å in the shell domain) and solved the crystal structures of the S and P domains of the RHDV VP60 protein both at 2.0 Å resolution. A model of the NTA domain of VP60 was built based on the near atomic resolution cryoEM map of the RHDV inner shell. A complete pseudo-atomic model of the RHDV capsid was then built by docking all the domain structures into the cryoEM map followed by MDFF refinement [21]. Structural comparison revealed a specific P2 sub-domain of RHDV in which RHDV isolates differ most and this variation contributes to HBGA binding specificity. The most exposed surface loop, L1 (a.a. 300–318), which exhibits high sequence variation among isolates, was probed to test its ability to interact with host tissue cells and to stimulate neutralizing antibodies. Cell- and animal-based experiments with synthetic peptides derived from this loop provided strong evidence that the loop is involved in virus-host interactions and stimulates production of high-titer antibody that can protect rabbits from RHDV infection.
Animal experiments were approved by the Harbin Veterinary Research Institute of the Chinese Academy of Agricultural Sciences. All procedures were conducted in accordance with animal ethics guidelines and approved protocols. The Animal Ethics Committee approval number was Heilongjiang-SYXK 2006-032.
RHDV (HYD isolate strain) was prepared from the livers of infected rabbits. These were cut into small pieces (∼5×5×5 mm3) and homogenized with a glass pestle in PBS buffer (8 mM Na2HPO4, 1.5 mM KH2PO4, 2.7 mM KCl, 137 mM NaCl, pH7.4) kept between 0 and 4°C. Tissue suspensions were centrifuged for 20 min at 4,000 g. An equal volume of chloroform was added to the supernatant and the mixture was shaken vigorously by hand for 15 seconds, followed by incubation for 2∼3 min at 4°C and centrifugation at 12,000 g for 15 min at 4°C. The chloroform phase was discarded and the above steps (shaking, incubation, and centrifugation) were repeated four times. The aqueous phase was then filtered through a 0.22 µm pore-size filter and overlaid into a discontinuous sucrose gradient (30%, 40%, 50%, 60%). The gradient with the clarified liver homogenate was centrifuged at 350,000 g for 80 min at 18°C in a Beckman L8-80M centrifuge with a 75 Ti rotor. Precipitant was collected and dissolved in PBS and the final sample for cryoEM studies was purified through a 25% sucrose cushion by ultracentrifugation at 145,000 g for 3 hr (Rotor Ti-75, Beckman). The resulting pellet was resuspended in TNE buffer (50 mM Tris, 50 mM NaCl, 5 mM EDTA) and fast frozen in liquid nitrogen for storage before it was used for cryoEM studies.
Small aliquots (∼3.5 µL) of purified RHDV samples were applied to holey grids (GiG) and blotted for 3 sec in a chamber at 100% humidity using an FEI Vitrobot Mark IV and then quick plunged into liquid ethane cooled by liquid nitrogen. Images were recorded with a Gatan UltraScan4000 (model 895) 16-megapixel CCD in an FEI Titan Krios cryo-electron microscope operated at 300 keV, at a calibrated magnification of 160770 (corresponds to a pixel size of 0.933 Å at the specimen), and an electron dose of ∼20 e/Å2 for each micrograph. A total of 1,100 cryoEM micrographs of RHDV were recorded. The defocus and astigmatism of each micrograph were estimated with CTFFIND3 [33] and corrected using the “applyctf” routine of EMAN [34]. Image processing and 3D reconstruction were performed using EMAN [34], with Spider [35], [36] scripts embedded for correspondence analysis (CORAN) of each image class, which was wrapped in the Appion package [37]. The 3D reconstruction was computed from ∼36,000 individually boxed virus particle images. The final reconstructed density map was further sharpened by application of an amplitude correction algorithm in the program BFACTOR [38] with a negative B-factor 1/(300 Å2). CryoEM maps were segmented, displayed, and fitted with atomic models using UCSF Chimera [39]. All illustrations of structures were rendered using either UCSF Chimera or PyMol [40].
The fragment (a.a.228–579) covering the entire P domain of the RHDV VP60 protein was cloned into pFastEL-3G vector (from Dr. Fei Sun's lab). This construct, fused with a GST tag and a precision protease digestion site at the N-terminus, was expressed in Sf21 insect cells. After GST-column (GE Healthcare) affinity purification, Prescission Protease (GE Healthcare) digestion, anion exchange by Resource Q (GE Healthcare), and gel filtration by Superdex 75 (GE Healthcare) on a BioLogic DuoFlow system (Bio-Rad), the recombinant protein was isolated at high purity (>98%). The purified sample was buffered at pH 8.0 in 50 mM Tris-HCl, 150 mM NaCl and concentrated to 3.0 mg/ml for crystallization. We used the hanging drop, vapor diffusion method to obtain brick-shaped crystals of the P domain at 289 K in the presence of 0.1 M sodium acetate, 1.1 M succinic acid, pH 5.5 and 1.0% PEG2000MME.
The contiguous NTA and S domains (a.a. 1–230) genes of the RHDV VP60 protein were cloned into the pEXS-DH vector [41] and expressed with an N-terminal 8×His tag in E.coli (BL21). This construct was purified using a Ni-NTA affinity column (GE Healthcare), anion exchange chromatography using a Resource Q column (GE Healthcare), and gel filtration using a Superdex 75 column (GE Healthcare) on a BioLogic DuoFlow system (Bio-Rad). The purified protein was buffered at pH 8.0 in 50 mM Tris-HCl and concentrated to 5.0 mg/ml for crystallization. Brick-shaped crystals were obtained via hanging drop, vapor diffusion at 289 K in the presence of 0.2 M MgCl2, 0.1 M HEPES-Na, pH 7.0 and 30% PEG400.
X-ray diffraction data sets of the crystals of the P and S domains were collected to 2.0 Å at the beam line BL17U (Shanghai Synchrotron Radiation Facility, SSRF) and the beamline BL17A (Photo Factory, Japan), respectively. All diffraction data were processed and scaled using HKL2000 [42]. Two copies of the S domain constitute each asymmetric unit of the crystal with a solvent content of 34.7%. The crystal structure of the S domain was solved by molecular replacement with PHASER [43] using the VP60 S domain structure from SMSV (PDB code: 2GH8) as the initial phasing model. The structure of the RHDV S domain was built manually in COOT [44] and refined using REFMAC5 [45]. The stereochemistry of the final model was evaluated by PROCHECK [46].
The determination of the P domain crystal structure was not straightforward because molecular replacement failed to yield a correct set of phases when the crystal structure of the P domain of SMSV (PDB code: 2GH8) was used as a phasing model. Instead, the cryoEM map of the RHDV virion served as a reliable initial model; the structure of SMSV P domain was fitted into the cryoEM map and modified manually by deleting the regions outside the map in COOT [44]. This EM map-based model was used as an initial model to run molecular replacement using PHASER [43]. The solution with the highest translational likelihood gain (89.33) and Z-score (3.6) was selected for further phasing. Only diffraction data up to 3.0 Å were used for phasing as this process led to a more continuous density map compared to the map that was obtained using the complete set of diffraction data. Initial phasing yielded a clear density map for P1, but not for P2. Density in both these sub-domains was gradually improved by imposing non-crystallographic symmetry (NCS) without phase extension, and further improved by changing some residues to Ser/Thr to fit the apparent density, during several rounds of refinement by REFMAC5 [45]. Subsequently, all diffraction data to 2.0 Å were used for further phasing and refinement. Automatic model building was performed by ARP/WARP [47] and 96 out of 714 total residues could be built correctly with side chains, and this guided the building of the complete model manually in COOT [44]. The final structure of the P domain was refined to 2.0 Å in REFMAC5 [45], and its stereochemistry was evaluated by PROCHECK [46] with 94.0% of the residues in most favored regions, 5.0% in allowed regions, and 0.8% in generously allowed regions.
Statistics for the data collection, processing, and structure refinement for both the P and S domains are summarized in Table 1.
Molecular dynamics flexible fitting (MDFF) is a computational method that employs molecular dynamics simulation to fit atomic models into cryo-EM density maps [21], [48] and has been successfully applied recently [49], [50], [51]. The initial atomic model of VP60 was obtained by combining the NTA structure derived from cryoEM density and the crystal structures of the P and S domains. Missing loops were modeled using MODELLER [52]. After rigid body docking into the cryoEM map, proteins were solvated in a box of water molecules with 150 mM NaCl in VMD [53], using 17 Å of padding in all directions. Counter ions were added to neutralize the simulated system, which was bounded by a cubic box of dimension 460 Å and contained 9,891,665 atoms. Simulations were performed with NAMD 2.9 [54], using the CHARMM27 force field with CMAP corrections [55], [56].
Two peptides, NJ85 (G300SASYSGNNATNVLQFWYA318) and NJ85Δ (G300SASYSG306N311VLQFWYA318), based on the VP60 sequence of RHDV NJ/China/1985 isolate strain (GeneBank accession number: AY269825) were synthesized, labeled with FITC and conjugated onto KLH, respectively, by using the commercial service from ChinaPeptides.
Viable fresh rabbit hepatocytes, primary splenocytes, and kidney epithelial cells RK13 (ATCC CCL-37) [57] were harvested from a healthy male, New Zealand white rabbit by using the standard collagenase perfusion technique [58] and maintained at 37°C and 5% CO2 in a humidified incubator.
Based on a previous protocol [59], adherent RK13 cells, hepatocytes, and splenocytes were fixed with 30% carbinol for 30 min at room temperature. The FITC conjugated peptides (FITC-NJ85 and FITC-NJ85Δ), dissolved in phosphate buffered saline (PBS) with 10% FBS (Fetal Bovine Serum) and 3% BSA (Bovine Serum Albumin), were respectively added to different cells at a final concentration 30 ug/mL and incubated for 1 hr at room temperature. Cells were washed three times with PBS containing 0.3% BSA and 0.1% Triton-X100. The interactions between the two FITC-conjugated peptides and the three types of rabbit tissue cells were imaged with an SP5 confocal microscope (Leica Microsystems, Heidelberg, Germany). Confocal stacks were combined with Image J [60] to construct the three dimensional image.
Healthy male, New Zealand white rabbits were subcutaneously immunized with 1 mg NJ85-KLH and NJ85Δ-KLH, respectively, in Freund's complete adjuvant. Further vaccinations were performed on days 14 and 21 with 1 mg of each antigen in Freund's incomplete adjuvant. Finally, rabbit sera were collected on day 28 after the initial immunization inoculation.
Antibody titres were assessed by ELISA. Briefly, RHDV virus (HYD isolate strain) (100 µl, 1 µg/ml, incubated overnight at 4°C) were used to capture antibodies in the sera (incubated for 1 h at 37°C), which were then detected with 100 µl horseradish peroxidase-conjugated goat anti-rabbit IgG (Jingmei Biotech) per well (diluted 1∶5000 in PBS containing 0.5% Tween 20 and 10% FBS), followed by 100 µl 3,3′,5,5′-Tetramethylbenzidine (TMB) Liquid Substrate (Sigma) per well for 30 min at room temperature in the dark. End-point titers were defined as the highest plasma dilution that resulted in an absorbance value (A450) two times higher than that of non-immune plasma with a cut-off value of 0.05. Data are presented as log10 values.
For Western blot experiments, RHDV viruses (HYD isolate strain) were fractionated by SDS–PAGE on a 10% gel and blotted onto Nitrocellulose Transfer Membrane (Whatman) using a semidry electro-transfer system (Amersham Biosciences). Analysis of sera was carried out by probing with anti-KLH-NJ85 and anti-KLH-NJ85Δ raised in rabbits at a dilution of 1∶1000. The reaction was detected by horseradish peroxidase-conjugated anti-mouse IgG antibody (rabbit) and visualized by enhanced chemiluminescence. The relative densities of bands were analyzed and integrated with Image J [60].
All experimental data are expressed as means ± SD and were analyzed by a t-test using the SPSS 10.0 statistical software. Probability values of <0.05 were considered to be statistically significant.
Hemagglutination (HA) of RHDV in the liver homogenates was tested according to Capucci [61]. The reaction was performed at room temperature for 30 min in PBS (pH 7.4). Two-fold serial dilution of the virus was added in a 96-well, V bottom microplate with 25 µL for each well. Then, a 1% suspension of human type O red cells was added to a final volume of 50 µL. The highest dilution of virus that caused complete hemagglutination of red cells was considered as the end point (Figure S10A).
Hemagglutination inhibition titers of the sera were tested as described [62]. After inactivation at 56°C for 30 min, sera were diluted two-fold serially from 1∶2 to 1∶512 respectively into PBS and added together with 8 HA units of RHDV antigen (1∶2048 dilution) into a 96-well V bottom microplate with 25 µL in each well. The plate was incubated for 1 hour at room temperature. Then, 25 µL of 1% suspension of human type O red cells was added into each well and incubated for 30 min. The highest sera dilution that caused complete inhibition was considered as the end point. Specific pathogen free rabbit serum was used as a negative control.
Healthy male, New Zealand white rabbits weighing between 3.0 and 3.5 kg were divided into three groups (n = 5 in each group) and raised in individual ventilated cages in a bio-safety level 3 enhanced containment laboratory approved by the Chinese Ministry of Agriculture. One group was subcutaneously immunized with 1 mg KLH-NJ85 in Freund's complete adjuvant, one group with KLH-NJ85Δ and the rest with KLH as a negative control. After immunization, those rabbits in each group were challenged intranasally with 256 hemagglutination titer of RHDV [63] and continuously housed and monitored every 24 hours for investigation of survival rate.
The isolate strain used for cryoEM study and virus challenge experiments was HYD isolate strain (GeneBank accession number: JF412629). The P and S domain of RHDV VP60 was cloned from the JX/CHA/97 isolate strain (GeneBank accession number: DQ205345). The two peptides, NJ85 and NJ85Δ, were synthesized according to the VP60 sequence of RHDV from NJ/China/1985 isolate strain (GeneBank accession number: AY269825).
The coordinates of the crystal structures of the RHDV VP60 S and P domains are deposited in the Protein Data Bank (PDB) with accession numbers 4EJR and 4EGT, respectively. The cryoEM map of the RHDV virion is deposited in the Electron Microscopy Data Bank (EMDB) with accession number EMD-5410 and its corresponding pseudo-atomic model is deposited in the PDB with accession number 3J1P.
|
10.1371/journal.ppat.1000262 | Distinct Roles of Plasmodium Rhomboid 1 in Parasite Development and Malaria Pathogenesis | Invasion of host cells by the malaria parasite involves recognition and interaction with cell-surface receptors. A wide variety of parasite surface proteins participate in this process, most of which are specific to the parasite's particular invasive form. Upon entry, the parasite has to dissociate itself from the host-cell receptors. One mechanism by which it does so is by shedding its surface ligands using specific enzymes. Rhomboid belongs to a family of serine proteases that cleave cell-surface proteins within their transmembrane domains. Here we identify and partially characterize a Plasmodium berghei rhomboid protease (PbROM1) that plays distinct roles during parasite development. PbROM1 localizes to the surface of sporozoites after salivary gland invasion. In blood stage merozoites, PbROM1 localizes to the apical end where proteins involved in invasion are also present. Our genetic analysis suggests that PbROM1 functions in the invasive stages of parasite development. Whereas wild-type P. berghei is lethal to mice, animals infected with PbROM1 null mutants clear the parasites efficiently and develop long-lasting protective immunity. The results indicate that P. berghei Rhomboid 1 plays a nonessential but important role during parasite development and identify rhomboid proteases as potential targets for disease control.
| Malaria is one of the major infectious diseases and is responsible for the death of more than a million people, mostly children under the age of five. Plasmodium, the causative agent of malaria, is transmitted by female Anopheles mosquitoes. Successful development of the parasite requires efficient recognition, attachment, and invasion of host cells. Several parasite cell-surface molecules have been implicated in these processes and may require proteolytic processing in order for the parasite to complete invasion. Rhomboid family proteins are serine proteases that cleave within the transmembrane region of their substrates. Here, we use a genetic approach to study the function of Plasmodium berghei rhomboid 1 (PbROM1). PbROM1 is expressed in both vertebrate and mosquito stages of parasite development, and the protein is present in secretory organelles that contain other parasite molecules required for invasion. We find that PbROM1 is required for efficient infection of both the mosquito and the vertebrate host. Interestingly, we also find that mice infected with ROM1(−) parasites clear the infection efficiently and are protected upon subsequent wild-type parasite challenge. Our study suggests a role for PbROM1 throughout parasite development and identifies ROM1 as a target for disease intervention.
| For successful development and transmission, Plasmodium has to invade multiple cell types both in the mammalian host and in the mosquito vector. Much of our knowledge about the molecular mechanisms of invasion comes from the study of P. falciparum merozoite invasion of red blood cells (RBCs). RBC invasion involves an initial attachment followed by re-orientation and entry of the parasite into the host cell [1]. There are two main classes of parasite surface molecules, the GPI-anchored proteins such as the merozoite surface protein family (MSP) [2] and transmembrane domain-containing proteins such as AMA1 [3],[4], erythrocyte binding-like family (EBL) [5],[6] and reticulocyte binding-like family proteins (RBL) [7],[8]. A few host-cell receptors to which these ligands bind have been identified [9]–[12].
In the mosquito, motility plays an important role in ookinete and sporozoite invasion. Motile ookinetes form within the mosquito blood meal and invade the midgut epithelium. After exiting on the basal side facing the hemocoel they differentiate into sessile oocysts [13]. Subsequently, sporozoites released from mature oocysts invade the salivary glands from where they are delivered to the vertebrate host by a mosquito bite. These sporozoites travel through the blood stream until they reach the liver, where they invade and infect hepatocytes. All three invasive forms (ookinetes, sporozoites in the mosquito and sporozoites in the mammalian host) utilize the same actin-based motor for entry into the host cell. Thrombospondin-related anonymous protein (TRAP) family homologues constitute one class of protein required for motility and host cell invasion [14]–[16]. The extracellular domains of TRAP interact with host-cell receptors, while the cytoplasmic tail links to the actin-myosin cytoskeleton [17]. As the parasite glides, the parasite surface ligand-receptor complexes translocate towards the posterior end. Dissociation of these interactions by proteolytic processing is thought to be important, as this enables the parasite to move forward [18]–[20]. In another Apicomplexan parasite-Toxoplasma-the TRAP homologue MIC2 is cleaved within its transmembrane domain releasing the receptor-binding domain from the parasite surface [18] and Plasmodium merozoite TRAP (MTRP) also appears to be cleaved in a similar manner [16].
Rhomboid-family (ROM) proteins are serine proteases that cleave their substrates within their membrane domain [21],[22]. Multiple rhomboid-family proteins have been identified in the genomes of Plasmodium and Toxoplasma [23]. Cleavage requires the presence of helix-destabilizing residues within the membrane domain of substrates [24]. Indeed, Apicomplexan surface proteins such as EBL and RBL proteins, AMA1, TRAP and their homologues contain such helix-destabilizing residues [23]. Assays in cultured mammalian cells identified possible substrates for both Toxoplasma and Plasmodium falciparum rhomboid proteins [25],[26]. Toxoplasma ROM5 localizes to the posterior end of the parasite and can cleave MIC2 within its transmembrane domain [25],[27]. Plasmodium does not have a ROM5 homologue but ROM4 is able to cleave EBA175 [28], an EBL family protein involved in binding to erythrocytes [10]. Processing of EBA175 within its membrane domain appears to be essential for parasite invasion [28].
Here we report on experiments investigating the role of Plasmodium berghei rhomboid 1 (PbROM1) during parasite development in the vertebrate host and the mosquito vector. Our data suggests a role for PbROM1 throughout Plasmodium development and indicate a role in invasion of host cells. We also find that a null PbROM1 mutant is efficiently cleared from mice and that these animals are protected from a subsequent lethal challenge of wild-type P. berghei. These findings identify a unique target for interfering with both disease causing and disease transmitting forms of the parasite.
Parasite maintenance and mosquito infections were performed as described previously [29]. We used Anopheles stephensi mosquitoes, Plasmodium berghei ANKA 2.34 parasites and female Swiss Webster mice in all our studies.
Antibodies were raised in rabbit against the N-terminal 52 amino acids of PbROM1 expressed in bacteria as a fusion protein using the pBAD expression system (Invitrogen). P. berghei schizonts, merozoites and sporozoites were fixed in ice-cold methanol and incubated for 1 h with the anti-PbROM1 antibody diluted 1∶500. Midgut and salivary gland sporozoites were obtained by gently homogenizing the infected tissues and centrifuging to remove cell debris. A anti-AMA1 monoclonal antibody (28G2) that recognizes the highly conserved cytoplasmic tail [30] was also used to label schizonts and merozoites, while a anti-CSP monoclonal antibody (3D11) that recognizes the repeat region [31] was used to label midgut and salivary gland sporozoites. Slides were then incubated for 1 h with Alexa Fluor 488-conjugated anti-rabbit IgG and rhodamine-conjugated anti-mouse or anti-rat IgG secondary antibodies. After washing, images were visualized in a Leica upright fluorescent microscope with a 100× objective and images were captured with a SPOT camera.
Sporozoite-infected salivary glands were fixed in 4% paraformaldehyde (Electron Microscopy Sciences, PA) in 0.25 M HEPES (pH7.4) for 1 h at room temperature, followed by 8% paraformaldehyde in the same buffer overnight at 4°C. The fixed glands were permeabilized, frozen and sectioned as previously described [32]. Sections were immunolabeled with rabbit anti-PbROM1 antibodies (1∶20 in PBS/1% fish skin gelatin), then with anti-rabbit IgG, followed by 10 nm protein A-gold particles (Department of Cell Biology, Medical School, Utrecht University, the Netherlands) before examination with a Philips CM120 Electron Microscope (Eindhoven, the Netherlands) under 80 kV.
For targeted disruption of the PbROM1 gene, a disruption plasmid was constructed by PCR amplification with primers, PbROM1(−)F-5′CCATACATTAGCAGAGTATAGGGA3′ and PbROM1(−)R-5′ACTTGCAC CCACTTTTATTGTAC3′ using P. berghei genomic DNA as template. Cloning into the P. berghei transfection vector [33] resulted in plasmid pROM1. This plasmid was linearized at the unique NdeI site and transfected into P.berghei schizonts as described [34]. To confirm disruption of the PbROM1 gene, integration-specific PCR was performed using specific primer combinations, P1-5′CGAGCAACAATGTCTGAC3′, P2-5′GAGTTCATTTTACACAATCC3′ and P3-5′TAATACGACTCACTATAGGGAGA3′. Disruption was also confirmed by RT-PCR using primers PbROM1F-5′TTATTACGGAGTGTTTCTTC3′ and PbROM1R-5′CGGAGAAATACATAGATTA3′ P.berghei circumsporozoite gene primers CSF-5′GTACCATTTTAGTTGTAGCGTC3′ and CSR-5′CATCGGCAAGTAATCTGTTG3′ were used as positive control.
The ability of the parasites to differentiate into gametocytes and form male gametes (exflagellation) was assessed as described previously [35]. An. stephensi mosquitoes were fed on infected mice and the ability of the disruptant parasites to form ookinetes (24 h) and oocysts (day 15) was examined microscopically. To assess ookinete numbers, individual midguts were dissected 24 h after feeding. Ookinete numbers were calculated after examining a Giemsa-stained smeared preparation of the midgut contents and counting both ookinetes and red blood cells. We assumed that each mosquito ingested 2 µl [36] and that mouse blood has 4×109 RBCs/ml [37]. Mature oocysts were counted on day 15 by direct light microscopic examination of dissected midguts. Sporozoites were isolated from midgut oocysts and salivary glands and counted on day 25–26 using a hemocytometer.
Sporozoites isolated from salivary glands were incubated for 15 minutes at 37°C in chamber slides coated with BSA. The supernatant was gently aspirated and sproozoite trails were fixed with 4% paraformaldehyde. The trails were visualized by labeling them with anti-CSP (mAb 3D11) antibody and rhodamine conjugated anti-mouse secondary antibody.
Sporozoites were isolated from salivary glands on ice and partially purified by passing through glass wool to remove mosquito debris. Protease inhibitors N-tosyl-L-lysine chloromethyl ketone (TLCK, 20 mM stock in water) and phenylmethylsulfonyl fluoride (PMSF 100 mM stock in ethanol) were obtained from Sigma. 30000 sporozoites were incubated at 4°C or 37°C in the presence or absence of protease inhibitors for 1h. EDTA was used to rule out nonspecific processing by metalloproteases. Parasite lysates were run on a SDS-PAGE and transferred onto PVDF membrane. These were probed with anti-TRAP antibodies that recognize the repeat region of the protein, followed by peroxidase-conjugated secondary antibody. To determine TRAP processing in PbROM1(−) parasites, 30000 wild-type and PbROM1(−)sporozoites were analyzed by SDS-PAGE as mentioned above.
Sporozoites isolated from salivary glands were counted using a hemocytometer and mice were injected intravenously with 150, 1000 or 10000 sporozoites. Infection efficiency was assayed by monitoring the pre-patent period of blood stage infection after sporozoite injection. Prepatent period is the time elapsed between mouse infection and when the first infected red blood cell (RBC) was observed upon examination of at least 25,000 RBCs. For quantifying efficiency of liver infection, mice were injected intravenously with either 103 wild-type or 103 PbROM1(−) sporozoites. Animals were sacrificed 36–40 h after sporozoite injection and total RNA was prepared using Trizol reagent. P. berghei 18S rRNA was quantified using primers (PbrRNA1-5′TGGGAGATTGGTTTTGACGT TTATGT3′ and PbrRNA2-5′ AAGCATTAAATAAAGCGAATACATCCTTAC3′) as described [38] and the results were normalized using mouse GAPDH. Results from 4 mice per group are expressed as mean±s.d. of rRNA copy number.
Mice were infected with either PbROM1(−) sporozoites or infected RBCs as described above. Parasitemia was checked every day until at least 30 days after the last PbROM1(−) parasites was detected. To confirm complete parasite clearance, 3×107 RBCs from these animals were injected into naïve mice and these animals were observed for 30 days to ensure that no infection resulted. After complete remission, the PbROM1(−) infected mice were challenged by intravenous (iv) or intraperitoneal (ip) injection of 105 wild-type P. berghei iRBCs. A second challenge was performed either 33 days or 7 months after the first challenge and a third 9 months after the first. Parasitemia was followed as described above. Protection is defined as the number of animals that survive the challenge.
Plasmodium berghei ROM1 (PbROM1) was initially identified in a subtractive hybridization screen for genes expressed during parasite development in the mosquito [29]. PbROM1 encodes a protein predicted to have seven transmembrane domains carrying a conserved, membrane-embedded Asparagine, Glycine-X-Serine and Histidine “rhomboid” motif (Figure S1). At least seven rhomboid genes were identified in the genome of various Plasmodium species [23]. Though PbROM1 homologues are highly conserved among rodent (92% identity) and human malaria species (55% identity), sequence identity among rhomboid genes of a given species is very limited (<20%, data not shown). This points to independent evolution of different rhomboid genes and suggests that each rhomboid protein plays distinct functions in the parasite life cycle.
Microarray analysis indicates that PfROM1 is expressed in both mosquito and vertebrate forms of the parasite [39]. We have produced an antibody to the first 52-amino acids of PbROM1 and used it to investigate protein expression and subcellular localization. The protein is expressed in both blood- and mosquito-stage parasites. PbROM1 protein has a punctate distribution in segmented (mature) schizonts and localizes to the apical end of free merozoites (Figure 1A). A number of organelles such as rhoptries and micronemes are found in the apical tip of the merozoite. These organelles secrete parasite proteins involved in host recognition and invasion. AMA1 (apical membrane antigen 1) is a micronemal protein required for invasion of RBCs and is also found on the surface of merozoites (Figure 1A). Immunoelectron microscopy confirmed the apical localization of PbROM1 and >85% of the gold label were found in micronemes (Figure 2A and 2B). PbROM1 expression is limited to schizonts and free merozoites and is not detectable in ring or trophozoite stages (data not shown). This is in agreement with the microarray analysis of P. berghei asexual stages in which PbROM1 is induced only in mature schizonts [40].
In mosquito stages, the PbROM1 transcript was initially identified among RNAs from mosquito midguts infected with mature oocysts [29]. Despite this, little or no protein was detected in sporozoites from these oocysts (Figure 1B). In contrast, PbROM1 protein is detected in sporozoites after invasion of mosquito salivary glands (Figure 1B). Immuno-electron microscopy more precisely localized PbROM1 in such sporozoites (Figure 2C and 2D). The protein is present along the entire length of the sporozoite both on the surface as well as in micronemes. We examined 65 parasite cryosections to quantify the distribution of PbROM1 in different cellular locations. Most of the gold particles were present on the sporozoite plasma membrane (76.4%) and in the micronemal membrane (17.7%) while the remaining particles were located over other parasite organelles (3.3%) and the mosquito salivary duct (2.6%).
To gain insights on PbROM1 function we disrupted the gene by homologous recombination and investigated the effects of gene loss on parasite development. Gene disruption was achieved by inserting a DNA fragment encoding a drug resistance marker into the open reading frame of PbROM1 (Figure 3A). Gene disruption was confirmed by insertion-specific PCR that identifies the disrupted locus from the wild-type locus (Figure 3B). In addition, disruption was confirmed by the absence of the transcript in PbROM1(−) sporozoites (Figure 3C).
We examined the possible function of PbROM1 in ookinetes by feeding PbROM1(−) parasites to mosquitoes. Ookinete efficiency of midgut invasion was assessed by counting the resulting number of oocysts. Disruption of the PbROM1 gene did not affect ookinete formation (Figure 4A and Table S1). However, in 6/7 experiments we found strong reduction in oocyst numbers (Figure 4B and Tables S2 and S3). These results suggest that loss of PbROM1 function impairs the ability of ookinetes to form oocysts. Subsequent development of PbROM1(−) parasites appears to be normal. The number of sporozoites formed by PbROM1(−) oocysts was similar to wild-type oocysts and no differences of salivary gland invasion could be detected (Figure 4C and Table S4). This result is consistent with the apparent lack of ROM1 protein expression in midgut sporozoites (Figure 1B).
To investigate whether PbROM1 plays a role in liver infection, we injected mice intravenously with an equal number of WT and PbROM1(−) sporozoites. The efficiency of infection was dose dependent and mice infected with PbROM1(−) parasites showed a consistent delay in the pre-patent period by one day or more compared to mice infected with wild-type sporozoites (Table 1). Efficiency of infection was also assessed by quantifying parasite loads in livers infected with equal numbers of mutant or wild-type sporozoites. Livers of mice infected with the mutant sporozoite had a 68% lower parasite load compared with mice infected with wild-type sporozoites (Figure 4D). This suggests that PbROM1 is required for efficient hepatocyte infection.
To determine if the defect observed in hepatocyte infection is due to a defect in motility, we performed a sporozoite gliding assay. PbROM1(−) sporozoites are motile as observed by circumsporozoite protein trails on glass slides (Figure 5C). PbTRAP, the parasite adhesin essential for gliding motility [41], is proteolytically processed by a serine protease (Figure 5A [42],[43]). This processing appears to occur independent of ROM1 (Figure 5B). This suggests that the reduction in parasite numbers may not be due to impairment in motility but rather a defect in invasion and/or a subsequent defect in development.
Parasitemia develops slower in animals infected with PbROM1(−) parasites compared to WT infected animals (Figure 6A and 6B). This phenotype is observed in animals infected by injection of sporozoites (Figure 6A) as well as when bypassing liver invasion by injecting infected RBCs (iRBCs) (Figure 6B). This slow-growth phenotype is specific to PbROM1 disruptants as another rhomboid (ROM3) disruptant and an oocyst capsule protein disruptant [44] have growth kinetics similar to wild-type parasites (data not shown).
Mice infected with PbROM1(−) parasites survive better than those infected with WT parasites (Figure 6C). Animals infected with PbROM1(−) parasites reach peak parasitemia of >35%, similar to WT parasites. At such high parasitemia, animals infected with WT parasites succumb to the infection. On the other hand, more than 80% of animals infected with PbROM1(−) parasites survive and eventually clear the parasites from their blood stream.
Mice that had cleared PbROM1(−) parasites from their bloodstream were challenged by intravenous injection of 105 WT iRBCs at least 30 d after the last circulating parasite was detected. Peak parasitemia in 12/14 mice after WT challenge ranged between 0.004%–2.6% (Figure 6D). Importantly, all the animals were able to successfully clear the wild-type parasites (Table 2). This protective immunity lasts for at least 7–9 months after the initial PbROM1(−) parasite exposure (Table 2). It is possible that the reduced RBC invasion efficiency of PbROM1(−) merozoites may trigger this protective immune response.
Invasion requires the specific recognition and attachment of parasite surface ligands to host cell receptors and subsequent processing of the bound ligands to facilitate detachment and entry into the host cell. This can be achieved by proteolytic processing of protein ectodomains [19] or in some cases by processing within the protein's transmembrane domain [18]. Plasmodium AMA1, EBL, RBL and TRAP proteins function in host-cell interaction and all have potential rhomboid cleavage sites within their predicted transmembrane domains. Recent studies using an in vitro mammalian cell-based assay indicate that Plasmodium ROM1 and ROM4 are able to cleave AMA1, EBL, RBL and TRAP members within their membrane-spanning domains [26],[28]. This suggests an important function for rhomboid proteins in invasion of host cells. In the present study we undertook a genetic approach to investigate the role of Plasmodium berghei rhomboid 1 (PbROM1) during the parasite development in the mammalian host and the mosquito vector.
Microarray analysis of P. falciparum genes identified PfROM1 as being expressed in both the mosquito and the asexual forms of the parasite [39]. Similarly, P. berghei ROM1 is also expressed in the mosquito and in its mammalian host [29],[40]. In agreement with the mRNA expression data, we find PbROM1 protein to be expressed in schizonts, in free merozoites and in sporozoites after salivary gland invasion. Though PbROM1 transcripts can be found in ookinetes (Figure S2), we could not detect the protein by indirect immunofluorescence. This may be due to the low abundance of the protein in this parasite form. The difference in PbROM1 protein expression between midgut and salivary gland sporozoites suggests post-transcriptional gene regulation. Incompletely spliced PbROM1 transcripts can be found in mature oocysts and sporozoites isolated from these oocysts (Figure S2). Furthermore, the ROM1 mRNA may be translationally regulated. Post-transcriptional regulation has been observed for a number of genes, especially in the sexual stages and plays an important role in Plasmodium development [29],[45],[46].
Our genetic analysis indicates that PbROM1 functions in both the vertebrate and mosquito stages. This is based on the observation that PbROM1(−) ookinetes form fewer oocysts, sporozoites isolated from infected mosquitoes infect the mouse liver less efficiently and the growth kinetics of the asexual forms is significantly delayed. Hence the phenotype of PbROM1(−) parasites points to ROM1 roles during cell invasion. However, a role in intracellular development cannot be formally excluded. We believe this to be less likely for several reasons. First, the mutant parasites fully complete development after invasion of the mosquito midgut epithelium, mouse liver and mouse RBCs. Second, WT and ROM1(−) ookinetes (Table S1), sporozoites (Table S2) and blood-stage merozoites (data not shown) develop equally well. Third, the ROM1 protein localizes to merozoite and sporozoite micronemes (an organelle that secrete proteins involved in invasion), in addition to the sporozoite surface. Together, these observations point to a role for ROM1 in host cell invasion.
Mice infected with PbROM1(−) parasites survive longer and are able to clear the infection efficiently. Those that clear the infection develop long-lasting immunity against a subsequent lethal wild-type P. berghei challenge. The immunity developed by PbROM1(−)-infected mice could be a result of slower infection, which provides the animal with an opportunity to mount a better immune response. Another interesting possibility is that parasite proteins normally processed by PbROM1 during invasion modulate the immune response. The absence or reduced levels of these cleaved proteins would allow the animals to develop immunity against the parasite. Interestingly, the Toxoplasma gondii ROM1 orthologue has also been shown to be required for efficient growth and invasion of host cells [47]. In addition to its role in invasion, TgROM1 also appears to play a role in intracellular replication as they form fewer parasites within the parasitophorous vacuole [47]. However, PbROM1 does not appear to play a significant role in the development neither of sporozoites within oocysts (Table S4) nor of merozoites within schizonts (data not shown). However, a role for PbROM1 in parasite replication in the mouse liver cannot be excluded. The observed differences between Plasmodium and Toxoplasma could represent a species-specific difference of ROM1 function.
Even though PbROM1(−) parasites are defective in multiple invasive stages, they do complete their life cycle successfully in both the vertebrate and invertebrate hosts. It is possible that in PbROM1(−) parasites, impairment of proteolytic processing only delays parasite invasion. Alternatively or in addition, other rhomboid proteins and/or proteases may take over the function of PbROM1, albeit with lower efficiency. There is precedent for such redundant function from in vitro data suggesting that some substrates are cleaved well by either PfROM1 or PfROM4, while other substrates are cleaved by both enzymes, albeit at different efficiencies [26]. A number of candidate substrates for PbROM1 such as AMA1 have been identified using mammalian cell-based assays [26]. However, these would have to be validated by in vivo experiments and factors such as spatial and temporal regulation of the protease and its substrate(s) are also expected to play a role. Our results suggest that PbTRAP, the parasite adhesin required for sporozoite motility, is cleaved by a serine protease. The protease inhibitors used does not necessarily inhibit only TRAP processing, but would be expected to inhibit several other serine proteases. However, the assay specifically measures only TRAP processing. TRAP is processed in the absence of ROM1 suggesting that it might not be a substrate. Alternatively, as discussed above, TRAP processing in ROM1(−) parasites could be due to functional redundancy. Data from in vitro processing assays suggest that this is unlikely because ROM4 but not ROM1 was able to cleave TRAP [26].
In conclusion, this study points to distinct roles for Plasmodium berghei ROM1 throughout parasite development. The lack of an effective vaccine is attributed to the high degree of antigenic variation [48] and the ability of the parasite to switch invasion pathways [49]–[52]. On the other hand, a common phenomenon in the different invasion pathways could be the need for processing and release of the adhesins. For instance, processing of EBA175 within the membrane domain is essential for invasion [28]. As suggested by our genetic analysis, targeting rhomboid proteins offers an attractive new approach to the control of malaria.
|
10.1371/journal.ppat.1002134 | Illumination of Parainfluenza Virus Infection and Transmission in Living Animals Reveals a Tissue-Specific Dichotomy | The parainfluenza viruses (PIVs) are highly contagious respiratory paramyxoviruses and a leading cause of lower respiratory tract (LRT) disease. Since no vaccines or antivirals exist, non-pharmaceutical interventions are the only means of control for these pathogens. Here we used bioluminescence imaging to visualize the spatial and temporal progression of murine PIV1 (Sendai virus) infection in living mice after intranasal inoculation or exposure by contact. A non-attenuated luciferase reporter virus (rSeV-luc(M-F*)) that expressed high levels of luciferase yet was phenotypically similar to wild-type Sendai virus in vitro and in vivo was generated to allow visualization. After direct intranasal inoculation, we unexpectedly observed that the upper respiratory tract (URT) and trachea supported robust infection under conditions that result in little infection or pathology in the lungs including a low inoculum of virus, an attenuated virus, and strains of mice genetically resistant to lung infection. The high permissivity of the URT and trachea to infection resulted in 100% transmission to naïve contact recipients, even after low-dose (70 PFU) inoculation of genetically resistant BALB/c donor mice. The timing of transmission was consistent with the timing of high viral titers in the URT and trachea of donor animals but was independent of the levels of infection in the lungs of donors. The data therefore reveals a disconnect between transmissibility, which is associated with infection in the URT, and pathogenesis, which arises from infection in the lungs and the immune response. Natural infection after transmission was universally robust in the URT and trachea yet limited in the lungs, inducing protective immunity without weight loss even in genetically susceptible 129/SvJ mice. Overall, these results reveal a dichotomy between PIV infection in the URT and trachea versus the lungs and define a new model for studies of pathogenesis, development of live virus vaccines, and testing of antiviral therapies.
| Human parainfluenza viruses (HPIVs) are a leading cause of pediatric hospitalization for lower respiratory tract infection, yet it is unknown why primary infection typically induces immunity without causing severe pathology. To study the determinants of PIV spread within the respiratory tracts of living animals, we developed a model for non-invasive imaging of living mice infected with Sendai virus, the murine counterpart of HPIV1. This system allowed us to measure the temporal and spatial dynamics of paramyxovirus infection throughout the respiratory tracts of living animals after direct inoculation or transmission. We found that the upper respiratory tract and trachea were highly permissive to infection, even under conditions that limit lower respiratory infection and pathogenesis. The timing of transmission coincided with high virus growth in the upper respiratory tracts and trachea of donor mice independent of the extent of infection in the lungs. After transmission, infection spread preferentially in the upper respiratory tract and trachea, inducing protective immunity without weight loss. Our work reveals a disconnect between Sendai virus transmissibility and pathogenicity, and the experimental model developed here will be instrumental in studying PIV pathogenesis.
| The parainfluenza viruses (PIVs) are non-segmented, negative-strand RNA viruses of the family Paramyxoviridae. The paramyxoviruses include not only the PIVs but also a number of other important human pathogens transmitted via the respiratory route such as human respiratory syncytial virus (HRSV), metapneumovirus, measles virus, and mumps virus [1], [2]. The human PIVs (HPIVs) consist of four serotypes (HPIV1-4), are a common cause of upper respiratory tract (URT) infections, and are a leading cause of lower respiratory tract (LRT) disease in infants and children [3]. The HPIVs are efficiently transmitted by direct contact and exposure to nasopharyngeal secretions [4], and nearly all children are infected with HPIV3 by age 2 and with HPIV1 and HPIV2 by age 5 [5], [6]. No licensed anti-PIV vaccines or drugs are available, and therefore non-pharmaceutical interventions are currently the only means of control. In view of these facts, an understanding of how PIV infection spreads within the respiratory tract, promotes pathogenesis, elicits immunity, and is transmitted to naïve hosts would greatly advance the development of novel vaccines and therapeutics.
Experimental studies of HPIV infection in tissue culture and animal models have helped reveal basic replication mechanisms and evaluate preclinical vaccine candidates [7]–[9]. However, knowledge about the spread of PIV infection in individual, living animals that are fully susceptible to PIV-associated disease would allow more thorough investigations of PIV virus-host interactions and transmission. Mice are poorly permissive to infection by the HPIVs, and HPIV infection in cotton rats, hamsters, guinea pigs, and ferrets is usually asymptomatic with minimal or undetectable pathology [1]. As a result, a number of studies have used Sendai virus (SeV) infection of mice as a model to investigate pathogenesis in an experimental setting [10], [11]. SeV is the murine counterpart of HPIV1, the leading cause of laryngotracheobronchitis (pediatric croup) [12]. SeV and HPIV1 have 78% amino-acid sequence identity [13], elicit cross-protective immunity [14]–[16], and have similar tissue tropism and epidemiology [1], [11]. Moreover, SeV shows promise as a Jennerian vaccine for HPIV1 [17] and as a vaccine vector for HRSV, HPIV3, and HPIV2 [18]–[20].
Although SeV and the HPIVs were first isolated in the 1950s and have been studied for more than 50 years [1], fundamental aspects of PIV infection and immunity that remain unknown are directly relevant to our understanding of pathogenesis and transmission. For example, the spatial and temporal spread of natural infection in the respiratory tract after SeV transmission remains poorly understood because of the ambiguous results (marked inter-animal variability and error) of classical experiments measuring virus titers in sacrificed mice [21], [22]. It is also unknown how HPIV and SeV infection after transmission often results in immunity without causing severe pathology. The contribution of LRT infection to transmission is unknown. Finally, while infection of the lungs and the concomitant immune response are clearly associated with disease severity [1], [11], [23], [24], many questions remain about the contribution of infection in the URT and trachea to clinical outcome and protective immunity [25], [26]. For example, we are unaware of any reports of studies investigating the effect of the dose of virus inoculum, the replicative fitness of the virus, or the genetic susceptibility of the host on the growth and clearance of SeV in the URT and trachea.
To measure the dynamics of PIV infection in living animals, we generated three luciferase-expressing SeVs that allow non-invasive in vivo bioluminescence imaging in mice. Analogous systems have previously been reported for DNA and positive-strand RNA viruses [27] but have been elusive for negative-strand RNA viruses, largely due to virus attenuation [28] or genetic instability resulting from reporter gene insertion [29]. We considered SeV an ideal candidate for non-invasive imaging because (i) foreign-gene expression by paramyxovirus vectors is usually genetically stable [30], (ii) in vivo imaging of a non-replicating SeV in intact mice has been successfully demonstrated [31] and (iii) the match of SeV and the murine host allows pathogenesis studies [11]. The reporter virus rSeV-luc(M-F*) described here was found to express high levels of luciferase yet replicate and promote disease in mice similar to wild-type (WT) virus. We imaged the dynamics of SeV infection in living, intact mice after direct inoculation and after contact transmission, varying both the virus dose and mouse strain. Unexpectedly, we observed a dichotomous tissue tropism in which the URT and trachea supported robust virus growth, efficient transmission, and protective immunity even under conditions that resulted in little infection in the lungs.
To develop a model in which PIV infection could be visualized non-invasively in living, intact mice, we generated three recombinant Sendai viruses (rSeVs) in which a firefly luciferase reporter gene was inserted into the P-M, M-F, and F-HN gene junctions, respectively, of SeV (Figure 1A, Figure S1). Insertion of the firefly luciferase gene and gene junction into the SeV genome was expected to unacceptably decrease downstream viral gene expression and, consequently, virus replication [32]. To generate a luciferase-expressing SeV expected to suffer little or no attenuation, the rSeV-luc(M-F*) virus was constructed to contain both the luciferase reporter gene and a more efficient transcription start sequence AGGGTGAAAG upstream of the F gene (Figure S1). Therefore, the attenuating effects of reporter gene insertion could be counteracted by optimization of the naturally inefficient gene start sequence upstream of the F gene [33]. For the rSeV-luc(P-M) and rSeV-luc(F-HN) constructs, in which the luciferase gene was inserted into the P-M and F-HN gene junctions, respectively, the natural transcription start sequence upstream of the F gene was left intact (Figure S1).
Multiple-step growth curves were measured at 33°C and 37°C in LLC-MK2 cells that had been infected at a multiplicity of infection (MOI) of 0.01 PFU/cell (Figure 1B). Titers of rSeV-luc(M-F*) and rSeV-luc(F-HN) were similar at both temperatures and similar to SeV WT, showing that these two luciferase-expressing viruses were not substantially attenuated or temperature restricted. In contrast, the rSeV-luc(P-M) virus showed reduced growth kinetics at 33°C and grew even more slowly at 37°C. To compare luciferase reporter gene expression by the recombinant SeVs, we measured in vitro luciferase activity in LLC-MK2 cell lysates (MOI 5 PFU/cell) (Figure 1C). Upstream insertion of luciferase in rSeV-luc(P-M) resulted in greater luciferase activity than did downstream insertion in rSeV-luc(F-HN), consistent with the results of previous studies of SeVs using secreted alkaline phosphatase as the reporter gene [32]. Luciferase expression by rSeV-luc(M-F*) exceeded that of rSeV-luc(P-M) within 6 h post-infection (p.i.), showing that the enhanced gene start sequence engineered into the M-F* virus (Figure S1) increased reporter-gene transcription at later time points, perhaps due to greater downstream transcription of the L polymerase gene. To determine how the reporter gene insertions might have altered SeV protein expression, LLC-MK2 cells were infected at an MOI of 5 PFU/cell and lysates were subjected to radioimmunoprecipitation and SDS-PAGE analysis. Low levels of expression of the M, F, HN and presumably L proteins by the rSeV-luc(P-M) virus (Figure S2A) most likely contributed to the attenuation of this virus. Viral protein expression by rSeV-luc(M-F*) and rSeV-luc(F-HN) was sufficient to generate virions with WT-like compositions (Figure S2C), consistent with the in vitro growth of these two reporter viruses to levels similar to that of WT virus (Figure 1B).
An ideal luciferase-reporter virus for non-invasive bioluminescence imaging and pathogenesis studies would express high levels of luciferase without altering virus replication and disease severity. To assess the virulence of the luciferase-expressing SeVs, 129/SvJ mice were anesthetized with isoflurane and inoculated intranasally with 30 µl containing 7,000 PFU of virus. This method of inoculation delivers ∼1/3 of the volume to the nasopharynx and ∼1/2 to the lungs [34]. Infection with SeV WT, rSeV-luc(M-F*), and rSeV-luc(F-HN) resulted in a mean weight loss of ∼25% and mean mortality rates of 80% (Figure 1D,E). Thus, these two luciferase-expressing viruses remained fully virulent at a dose of 7,000 PFU. In contrast, 129/SvJ mice inoculated with 7,000 PFU of the attenuated rSeV-luc(P-M) virus experienced only 12% weight loss and no mortality. All mice inoculated with 70,000 or 700,000 PFU of rSeV-luc(P-M) also survived (data not shown), further demonstrating that the attenuated rSeV-luc(P-M) virus is avirulent.
Acute viral pneumonia by SeV induces high levels of lymphocyte infiltration that show a peak in bronchoalveolar lavage fluid (BALF) at ∼10 d p.i. [35]. To compare lymphocyte influx caused by the luciferase-expressing viruses and WT virus, we sacrificed 129/SvJ mice that had been infected with 7,000 PFU at 10 d p.i. and collected BALF. Similarly large total numbers of lymphocytes, CD4+ T-lymphocytes, and CD8+ T-lymphocytes were detected in BALF after infection with WT, rSeV-luc(M-F*), or rSeV-luc(F-HN) virus (Figure 1F; Figure S3A–B), whereas mice inoculated with the attenuated rSeV-luc(P-M) had total lymphocyte counts only 10% as high. To determine the extents to which the reporter viruses elicited SeV- or luciferase-binding antibodies, ELISA assays were performed on sera collected at 10 d p.i. The anti-SeV antibody titers elicited by all three rSeVs were similar to that induced by WT virus (Figure 1G). The three reporter viruses also induced similar anti-luciferase antibody titers (Figure S3C). Thus, despite being attenuated and avirulent, rSeV-luc(P-M) elicited a robust antibody response. rSeV-luc(M-F*), which induced WT-like morbidity and mortality while expressing high levels of luciferase, was identified as the best suited surrogate for WT virus for use in subsequent bioluminescence imaging experiments.
In studies to determine whether non-invasive bioluminescence accurately reflected in vivo infection, 129/SvJ mice were intranasally inoculated with 7,000 PFU, imaged with a Xenogen IVIS instrument, and immediately euthanized. Respiratory tissues were promptly collected for ex vivo measurement of luminescence and viral titers. As in previous studies in immunocompetent mice [36], [37], viral titers and bioluminescence were limited to the respiratory tract. As shown in Figure S4, in vivo bioluminescence intensity levels in living animals were well correlated with ex vivo luminescence (R2 0.878) and with viral titers in the nasopharynx (R2 0.864), trachea (R2 0.915), and lungs (R2 0.961). The correspondence of these data validates the technique as a means of noninvasive measurement of infection in vivo. To determine whether the luciferase-reporter genes were genetically stable in the three rSeVs, we recovered lung tissues from 129/SvJ mice inoculated with 7,000 PFU of virus at 7 d p.i., homogenized the samples, and conducted plaque assays in LLC-MK2 cells. Five plaques of each of the three luciferase-expressing viruses were picked, amplified by one round of replication in eggs, RT-PCR transcribed, and sequenced. All of the individual viral clones contained the luciferase insert, which had no amino acid mutations, and expressed luciferase when grown in LLC-MK2 cells.
We next measured the kinetics and tropism of bioluminescence in living 129/SvJ mice and compared the results to conventionally measured viral titers in dissected tissues (Figures 2 and 3). Just as rSeV-luc(M-F*) and rSeV-luc(F-HN) had in vitro replication rates and in vivo pathogenicity similar to those of WT virus, they also had WT-like titers in the nasal turbinates, trachea, and lungs. In the nasal turbinates, high virus titers (>105 PFU) were detected by day 2 p.i. and were maintained until day 9 p.i., after which rapid clearance occurred (Figure 3B). Between days 2 and 9 p.i., high levels of in vivo bioluminescence were similarly observed in the nasopharynx (>108 photons/s) of 129/SvJ mice infected with rSeV-luc(M-F); bioluminescence peaked at about 5 d p.i. (Figure 3A). In the lungs, the titers of all three luciferase-expressing viruses and of WT SeV peaked by day 5 p.i. and fell to low levels by day 9 p.i. Infection with the attenuated rSeV-luc(P-M) resulted in peak lung titers of ∼104 PFU (approximately 5% of the WT titer) at day 5 p.i. (Figure 3D). Similarly low levels of rSeV-luc(P-M) bioluminescence were observed in the lungs (Figure 3A), consistent with the attenuated and avirulent virus phenotype. On the other hand, rSeV-luc(P-M) reached high peak titers (∼105 PFU, similar to the WT titer) in the nasal turbinates at 7 d p.i. (Figure 3C), and high levels of bioluminescence were observed in the nasopharynx between days 3 and 7 p.i. (Figure 3A).
The high permissivity of the URT and trachea to infection by the attenuated rSeV-luc(P-M) virus was unexpected. We next investigated whether these tissues were also highly permissive to infection by the WT-like virus rSeV-luc(M-F*) at a low inoculating dose. Our preliminary studies showed that the 50% mouse infectious dose (MID50) of rSeV-luc(M-F*) was 9 PFU and that a 70-PFU dose resulted in 100% infection, similar to results obtained for WT SeV in mice [38] and HPIV1 in humans [39]. We inoculated 129/SvJ mice intranasally with 70, 700, or 7,000 PFU of rSeV-luc(M-F*) in equal 30 µl volumes and then measured bioluminescence and viral titers. After inoculation with 70 PFU, viral titers and bioluminescence in the lungs were ∼10% of that induced by a 7,000-PFU dose (Figure 4A,B), and weight loss was far less (Figure 4C). In contrast, infection in the nasopharynx and trachea after a 70-PFU inoculation was delayed ∼1 d compared to 7,000-PFU inoculation, reached a similar level by ∼5 d p.i. (Figure 4A,B), and induced relatively high titers of SeV-specific antibodies (>105) (Figure 4D). Therefore, low-dose inoculation of the WT-like rSeV-luc(M-F*) virus resulted in preferential infection of the URT and trachea, inducing a robust antibody response without causing much weight loss.
While it is known that 129/SvJ and DBA/2 mice are highly susceptible to lung infection by SeV and BALB/c and C57BL/6 mice are highly resistant [40]–[43], the effect of host genetics on SeV replication in the URT and trachea has not been reported. Therefore, we measured the in vivo dynamics of SeV infection in 129/SvJ, DBA/2, C57BL/6, and BALB/c strains of mice that had been intranasally inoculated with 7,000 PFU of rSeV-luc(M-F*). As expected from previous studies, the extent of pulmonary infection and weight loss correlated with each other and followed the trend C57BL/6<BALB/c<<DBA/2<129/SvJ (Figure 4). In contrast, similarly high levels of bioluminescence were observed in the URT and trachea in all four strains of mice. The titers of rSeV-luc(M-F*) in BALB/c mice correlated with bioluminescence in intact mice (Figure S5A), as they were in 129/SvJ mice. Therefore, use of the bioluminescence technique to measure respiratory tract infection in living mice was validated in both 129/SvJ and BALB/c strains.
SeV, the HPIVs, and HRSV are thought to be transmitted primarily via contact with respiratory secretions as opposed to long-range transmission of these secretions as small-particle aerosols [21], [22], [24], [44], [45]. It has also been shown that growth of SeV [21] and influenza virus [46] in the URT promotes transmission. Two fundamental questions about PIV transmission that have long remained unanswered are (i) how growth of virus in the lungs of donors influences transmission and (ii) how infection spreads in the respiratory tracts of contact animals after transmission. To address these fundamental questions about SeV transmission, we inoculated BALB/c or 129/SvJ donor mice with 70 or 7,000 PFU of rSeV-luc(M-F*) and then at 1 d p.i. we placed 1 donor mouse in a “clean” cage with 3 naïve contact mice. We measured bioluminescence daily in inoculated and contact mice until primary infection was cleared. We then collected sera on day 60, challenged the mice with 7,000 PFU of rSeV-luc(M-F*) on day 63, and subsequently imaged the mice daily to detect reinfection (Figures 5 and 6). Transmission to every naïve contact mouse was observed by nasopharyngeal bioluminescence and seroconversion, including resistant BALB/c mice exposed to donor animals inoculated at the lower dose of 70 PFU (Figure 5). The timing of transmission was not influenced by the extent of lung infection in donors, as lung titers were ∼10 times lower in BALB/c than in 129/SvJ donor mice after 7,000-PFU inoculation (Figure S5C), yet the timing of transmission (time between detection in inoculated animals and contact animals) was similar (3.3 and 3.4 days, respectively) (Figure 6F). While LRT infection occurred in both strains of mice and may have contributed to transmission, the primary determinant of transmission appeared to be virus shedding in the URT and trachea. For example, both high-titer (>105 PFU) shedding in the nasal cavities and trachea of 129/SvJ donor mice (Figure 4A,B) and contact transmission (Figure 6E,F) occurred ∼1 day earlier after 7,000-PFU inoculation than after 70-PFU inoculation.
Under all four conditions tested (129/SvJ or BALB/c donor mice infected with 70 or 7,000 PFU of virus), the tropism and magnitude of infection in contact animals was similar to that observed after direct intranasal inoculation with 70 PFU of rSeV-luc(M-F*). After contact transmission, bioluminescence was first observed in the nasopharynx and then spread to the trachea and lungs an average of 0.8 and 1.0 days later, respectively (Figure S6A–D). Robust infection was observed in the nasopharynx and trachea after transmission (Figure 6A–D, Figure S6E–H). In contrast, low levels of infection in the lungs were observed after transmission, consistent with low weight loss (Figure 6G–H). In all four groups of mice, SeV-specific antibody titers on day 60 were similarly high (∼106) and all animals were protected from challenge on day 63 (Figure 5). After challenge, a low level of bioluminescence (<106 photons/s), but no weight loss, was detected in only 1 of the 30 contact mice; this animal had shown the lowest level of bioluminescence on days 5–12 after primary infection (Figure 5B, solid black circles). As this animal also had the lowest level of SeV-specific antibodies at day 60 before challenge, a threshold level of infection may be required to induce the highest levels of protective immunity. Overall, SeV infection after transmission was observed to be sufficiently robust in the URT and trachea, yet sufficiently limited in the lungs, to induce protective immunity without causing severe pathogenesis.
In this study, we generated and used luciferase-reporter viruses to study the kinetics of SeV infection in living mice after direct inoculation or contact transmission. WT SeV virus and the luciferase-expressing virus rSeV-luc(M-F*) had a similar replication rate in vivo and elicited similar levels of weight loss, mortality, bronchoalveolar lymphocyte influx, and serum antibody titers. Both susceptible (129/SvJ) and resistant (BALB/c) strains of mice were intranasally inoculated with 70- and 7,000-PFU doses of rSeV-luc(M-F*), and the spread of infection was measured by both in vivo bioluminescence in intact mice and ex vivo virus titers in the tissues of sacrificed animals. The consequences of infection in the URT and trachea were found to be distinct from those of infection in the lungs. Unexpectedly, under all conditions tested, including 70-PFU inoculation of resistant BALB/c mice, the URT and trachea supported robust SeV growth, efficient contact transmission, and protective immunity independently of the extent of infection in the lungs. In contrast, the extent of infection in the lungs varied with the virus dose and mouse strain and was highly correlated with weight loss and mortality. Overall, the results reported here reveal a tissue-specific dichotomy in the respiratory tract in which robust infection in the URT and trachea supports efficient transmission while the extent of infection in the lungs and the host response determines disease severity.
Here we describe for the first time the development of a non-invasive bioluminescence imaging system to visualize negative-strand RNA virus infection throughout the bodies of living animals, using the respiratory paramyxovirus SeV as a model. The development of a non-attenuated paramyxovirus that expresses sufficiently high levels of a reporter gene to allow non-invasive imaging of small animals has been challenging because of the polarized transcription mechanism of these non-segmented negative-strand RNA viruses [2]. A significant advance described here is the generation of the rSeV-luc(M-F*) virus, in which the attenuating effect of reporter-gene insertion [32] is counteracted by enhancement of the naturally occurring but suboptimal start sequence upstream of the F gene [33]. Expression of the F gene, a virulence factor [47], [48], is also downregulated by HPIV1 [49], HPIV3 [50], PIV5 [51], measles virus [52] and canine distemper virus (CDV) [47] by readthrough transcription or long untranslated regions. Therefore, we predict that other WT-like reporter paramyxoviruses that express high levels of luciferase can be engineered by inserting the reporter gene into the M-F junction and maintaining F gene expression through compensating mutations. Reporter gene expression without attenuation of SeV has also been achieved by construction of a bicistronic gene that contains an internal ribosome entry site [53], although it is not yet clear whether this approach yields sufficient luciferase expression to allow non-invasive imaging of in vivo infection.
The use of the luciferase reporter gene in the present work enabled the measurement of infection throughout the entire respiratory tracts of intact animals, allowing us to measure the spread and clearance of infection after direct inoculation or transmission. eGFP-expressing reporter viruses have also been used to study the dynamics of CDV infection in ferrets [54], [55] and measles virus infection in monkeys [56], [57]. The eGFP reporter gene provides the advantage of allowing the tropism of infection to be studied at the cellular level in dissected tissues. Moreover, eGFP-expressing viruses can also be used to quantify and type infected cells in the peripheral blood, skin, and mouths of living animals. eGFP-expressing HPIV3 and Sendai viruses have been used to study the cellular tropism of PIV infection in well differentiated, primary epithelial cultures. In the case of HPIV3, infection was found to be restricted to ciliated epithelial cells and to cause little cytopathology [58]. In contrast, SeV was found to infect ciliated and non-ciliated cells, but not goblet cells, and was observed to induce ciliostasis, cell sloughing, apoptosis, and cellular degeneration [59]. It is unknown whether cell-free virus or cell-associated virus is associated with SeV transmission.
A major finding reported here is that the efficiency and timing of SeV transmission are independent of the extent of pulmonary infection, clinical symptoms, and host genetics. HPIV1 transmission from asymptomatic human donors has also been observed in an experimental setting [39] and is consistent with epidemiological observations for PIV outbreaks in general [23], [24]. These observations suggest that LRT infection and the severity of clinical symptoms are poor predictors of transmission potential in surveillance and infection control efforts. As in previous work [21], [38], we observed that SeV transmission coincides with high-titer virus growth in the URT and is remarkably efficient because of the high infectivity of the virus (e.g., the MID50 of SeV is <10 PFU). HPIV1, HPIV3, and HRSV are similarly highly infectious and also transmit predominantly by direct contact or indirect exposure to nasal secretions [44], [45], [60]–[62]. In the absence of an available prophylactic drug for uninfected individuals in high-risk groups (e.g., premature infants and the immunocompromised), the results described here suggest that efforts to control PIV infection should focus on reducing URT shedding from infected individuals, disinfecting contaminated surfaces, and hand washing. In contrast to infection control, which would be best served by limiting URT infection, therapeutic antivirals would be better targeted to the LRT to control clinical manifestations of PIV-associated disease.
Genetic factors have been identified that modulate viral susceptibility and disease severity in humans [63]–[65] and in the lungs of mice [40], [42], [43], [66]–[70]. Our results show for the first time that genetic factors limiting virus growth in the lungs of resistant BALB/c mice, compared to susceptible 129/SvJ mice, do not limit robust virus growth in the URT and trachea and, consequently, do not limit transmission. Furthermore, BALB/c and 129/SvJ mice showed a similarly high extent of infection in the URT and trachea and a similarly low extent of infection in the lungs after exposure to cagemates inoculated with high or low virus doses. This finding shows that host genetics do not play a major role in SeV transmission, at least in these strains of mice. These observations reinforce our inference that transmission and pathogenesis are independent consequences of URT versus LRT infection, respectively. Additional experiments are needed to delineate the mechanisms responsible for the higher permissivity of the URT and trachea than of the lungs to SeV infection. Possible mechanisms include the site of inoculation in the nasal cavity, lower temperature in the URT, tissue-specific differences in virus replication and innate immunity, and antiviral mechanisms, such as surfactant proteins in the lungs. Reduced replication in the lungs may be associated with lower levels of the secreted tryptase Clara, which is required for cleavage of the F protein to allow viral entry [71], [72].
Asymptomatic infection that promotes immunity and transmission represents a balanced relationship that benefits both the virus and the host. Such has been the case in several enzootic (clinically unapparent) epidemics of SeV in which subclinical infections were maintained in mouse and hamster colonies for years with no increase in pathogenicity, causing apparent disease only occasionally in suckling and old animals [73], [74]. These epidemiological observations are reminiscent of the low virulence yet high transmissibility of the reverse-genetics engineered SeV described here, which was derived from a modified Enders strain that had been passaged in embryonated chicken eggs. Our results show that the level of virus replication in the lungs affects neither the timing nor the efficiency of transmission; thus, SeV replication in the lungs may offer no selective advantage. Instead, we suggest the following mechanism for symbiotic virus-host interplay in enzootic epidemics of SeV: natural infection after transmission is sufficiently limited in the lungs to avoid clinical signs of disease yet is sufficiently robust in the nasopharynx and trachea to promote efficient transmission and induce protective immunity.
Epizootic (clinically apparent) outbreaks of SeV have caused morbidity and high rates of mortality in mouse colonies [75]–[77]. Two closely related, highly pathogenic field isolates of SeV are the Ohita and Hamamatsu strains [78], [79]. While inoculation with only a few PFU of unpassaged Hamamatsu strain SeV results in mortality in mice, the MLD50 of the virus was attenuated by as much as 400-fold after 50 passages in eggs [80]. When the highly pathogenic Ohita and Hamamatsu strains were adapted to LLC-MK2 cells and chicken eggs, they were found to have selected for mutations in the C protein and untranslated leader region, respectively, which increase replication in cultured cells but attenuate replication and pathogenesis in the lungs of mice [81]–[83]. The bioluminescence imaging system described here would be useful in determining whether the mutations that attenuate replication in the lungs also attenuate replication in the URT and trachea, thereby reducing transmission, or whether they actually promote sustained transmission by supporting nasal and tracheal shedding of virus while reducing pathogenesis in the lungs. Such experiments may also reveal whether our observations about the spread and transmission of the egg-adapted SeV extend to unpassaged, highly pathogenic field isolates.
SeV is a promising Jennerian vaccine against HPIV1 [1], [13], and recombinant SeV vaccine vectors containing an envelope gene from HRSV, HPIV3, or HPIV2 inserted into the F-HN gene junction have been shown to elicit both B- and T-cell responses that lead to protection from challenge in small-animal models [18]–[20]. While SeV is pathogenic in mice, an ongoing clinical trial has demonstrated SeV to be well tolerated in humans [17]. In non-human primates, SeV has been shown to protect against HPIV1 challenge with no associated adverse events [15], [84]. This result is likely due in part to the sensitivity of SeV to human IFN-mediated innate immunity [85]. As SeV is developed further as a vaccine vector, the luciferase-expressing SeVs, imaging system, and methods described here will be useful in investigating the effect of vaccine dose, volume, and position of foreign antigen insertion in the SeV genome on tissue-specific vector growth and the immune response in small animal models. Of course, replacing the luciferase reporter gene in SeV with a vaccine antigen could alter in vivo replication of the vector. For example, three different recombinant HPIV3 vectors expressing HPIV1 HN, HPIV2 HN, or measles virus HA inserted into the P-M gene junction were found to replicate to different levels in hamsters [86].
In summary, we have described the development of the non-attenuated reporter virus rSeV-luc(M-F*), which can be used to quantify tissue-specific SeV infection in living mice. Our results reveal how infection by SeV spreads in individual, living animals after direct intranasal inoculation and after transmission. Importantly, infection in the URT and trachea were found to be associated with contact transmission while infection in the lungs was found to be associated with pathogenesis. The imaging tools developed here will provide a method to study the effect of viral factors, host genetics, host age, immune status, environmental conditions, and inoculation mode on the dynamics of infection and transmission. For example, infection can be tracked non-invasively in WT and knockout mice before immune responses are measured ex vivo and then interpreted in light of the preceding infection. Methods similar to those reported here could also be developed to image infection by other paramyxoviruses in small-animal models. Overall, our model system and results suggest tissue-targeted approaches to PIV infection control and vaccine development, while our non-invasive bioluminescence imaging technique is expected to advance the preclinical testing of candidate vaccine vectors and experimental therapies.
All animal studies were approved by the Animal Care and Use Committee of St. Jude Children's Research Hospital and were performed in compliance with relevant institutional policies, the Association for the Accreditation of Laboratory Animal Care guidelines, the National Institutes of Health regulations, and local, state, and federal laws.
Monolayer cultures of LLC-MK2 cells were grown in Dulbecco's minimal essential medium (DMEM) supplemented with 10% fetal bovine serum, 1% L-glutamine, 1% penicillin, and 1% streptomycin at 37°C, 5% CO2.
Unique NotI recognition sites were cloned into the P-M, M-F, and F-HN intergenic junctions of an Enders-based pSeV viral genome plasmid using cloning sites described previously [32]. The firefly luciferase gene was amplified by PCR using the pGL3 Basic vector (Promega) and a pair of AscI tagged primers, subcloned into a shuttle plasmid containing a SeV intergenic junction and flanking NotI restriction sites [32], and then subcloned into the unique NotI site of each of the pSeV viral genome plasmids. Within the pSeV-luc(M-F*) plasmid, the start signal upstream of the F protein was changed from AGGGATAAAG to AGGGTGAAAG by using the QuikChange Site-Directed Mutagenesis Kit (Stratagene Corp). The rSeVs were rescued from the pSeV genome plasmids as described previously [20].
rSeV-infected LLC-MK2 cells (MOI, 5 PFU/cell) were incubated at 33°C, 5% CO2, and lysates were collected at various times p.i.. Luciferase assays were performed using the Luciferase Assay System (Promega) and expression was measured on an automated luminometer (Turner Biosystems, Inc.) as described previously [87].
Virus titers from multistep growth curves (MOI of 0.01 PFU/cell) and homogenized tissues were determined by plaque titration in LLC-MK2 cells as described previously [48]. Eight week-old female 129/SvJ mice or BALB/c mice (Jackson Laboratories) were anesthetized with isoflurane (Baxter Health Care Corp.) and inoculated intranasally (i.n.) with 30 µl of PBS or PBS containing virus. Animals were monitored daily for weight loss, morbidity, and mortality. Before imaging, mice were injected intraperitoneally with luciferin (Xenogen Corp) at a dose of 150 mg/kg of body weight and anesthetized with isoflurane for 5 min. In vivo images were acquired with a Xenogen IVIS CCD camera system (Caliper Life Sciences) and analyzed with Living Image 3.2 software (Caliper Life Sciences) using an exposure of 60s, 30s, or 5s (binning 4; f/stop 1). Pseudocolor images (representative of bioluminescence) of mice were displayed using a binning of 4 on a colorimetric scale ranging from 1×106 to 1×109 surface radiance (photons/s/cm2/steradian), which is defined as the number of photons that leave a cm2 of tissue and radiate into a solid angle of one steradian. To quantify bioluminescence, regions of interest (ROI) were defined manually and graphed data were expressed as total flux (photons/s), which is defined as the radiance within each pixel summed over the ROI area (cm2)×4π. For experiments shown in Figures 1D, 1E, 2, 3, 4E, 4F, and S4, mice were anesthetized by IP injection of 300 µl 2,2,2-Tribromoethanol (300 mg/kg) and chest hair was removed by shaving and application of a depilatory cream 3 d before inoculation.
Sera and BALF were collected from euthanized animals on day 10 or day 60 p.i.. BALF samples (3 ml) were centrifuged to collect cellular material and plated in a tissue culture dish for 1 h at 37°C to remove adherent cells. Suspension cells were harvested, total lymphocytes were counted microscopically, and red blood cells were lysed. For analyses by flow cytometry, cells were stained with FITC-conjugated anti-CD4 (RM4-4) and PE-conjugated anti–CD8b (53-5.8) antibodies (BD Biosciences Pharmingen). Lymphocytes were gated based on forward and side scatter, and the percentages of CD4+ and CD8+ T cell populations were measured within this gate. ELISAs were used to measure the levels of SeV-specific or luciferase-specific antibodies present in the sera. Briefly, 96-well plates were coated overnight with disrupted, purified SeV (10 µg/ml) or firefly luciferase (1 µg/ml, Abcam). Plates were blocked with PBS containing 1% BSA and then incubated with 10-fold serially diluted serum samples. After incubation, plates were washed, incubated with HRP-Goat anti mouse IgG (Southern Biotechnologies) and then washed further. To quantify levels of antibodies, TMB substrate (Kirkegaard and Perry Laboratories) was added to the wells followed by stop solution and absorbance was read at a wavelength of 450 nm. GraphPad Prism non-linear regression software was used to calculate antibody titers.
Donor animals were inoculated intranasally with 30 µL of rSeV-luc(M-F*) and were individually placed into cages containing 3 naïve contact mice at 24 h p.i. Bioluminescence was monitored daily until it remained consistently at background levels (∼15 days). Sera were collected on day 60 so that SeV-specific antibody levels could be measured as described above. On day 63, mice were challenged with 7000 PFU rSeV-luc(M-F*) administered i.n. and bioluminescence was measured daily.
|
10.1371/journal.pcbi.1003178 | Sensory Information and Encounter Rates of Interacting Species | Most motile organisms use sensory cues when searching for resources, mates, or prey. The searcher measures sensory data and adjusts its search behavior based on those data. Yet, classical models of species encounter rates assume that searchers move independently of their targets. This assumption leads to the familiar mass action-like encounter rate kinetics typically used in modeling species interactions. Here we show that this common approach can mischaracterize encounter rate kinetics if searchers use sensory information to search actively for targets. We use the example of predator-prey interactions to illustrate that predators capable of long-distance directional sensing can encounter prey at a rate proportional to prey density to the power (where is the dimension of the environment) when prey density is low. Similar anomalous encounter rate functions emerge even when predators pursue prey using only noisy, directionless signals. Thus, in both the high-information extreme of long-distance directional sensing, and the low-information extreme of noisy non-directional sensing, encounter rate kinetics differ qualitatively from those derived by classic theory of species interactions. Using a standard model of predator-prey population dynamics, we show that the new encounter rate kinetics derived here can change the outcome of species interactions. Our results demonstrate how the use of sensory information can alter the rates and outcomes of physical interactions in biological systems.
| Encounters between individual organisms are an essential part of biology; in many sexually reproducing species, males and females must encounter one another in order to mate, pollinators must find flowers, and predators must locate prey before capturing and consuming them. Many species accomplish these tasks by actively searching for their targets using sensory information. Despite this, classical mathematical models used to predict the rate of encounters between searchers and their targets assume that searchers make movement decisions without using sensory information. Here we develop a mathematical framework for incorporating sensory information into searcher movement behavior to study how sensory response changes the relationship between encounter rate and target density. By comparing searchers that use sensory information to those that do not, we show that sensory response not only increases encounter rate, but that it also changes the form of the relationship between encounter rate and target density. By using sensory information, predators encounter prey at a rate that is less sensitive to changes in prey density when prey density is low. Our results demonstrate a strong connection between the usage of sensory information and the encounter rates that are so critical to survival and reproduction in nature.
| Classical models of species interactions assume that encounters between individuals are governed by a process akin to mass-action; individuals move along random linear trajectories and encounter one another when they come within a critical distance [1], [2]. Under these assumptions, a searcher such as a predator or pollinator will encounter its targets at a rate that scales linearly with target density [1],[3]. The form of the function relating encounter rate to target density is essential; it affects both the dynamics and outcome of species interactions. The form of this relationship at low target density, in particular, strongly influences ecological and evolutionary dynamics in systems of interacting species by determining the degree to which a searcher can deplete limiting resources (e.g., [4]).
Recent work has extended the study of encounter rates to consider searchers that encounter targets probabilistically, destroy targets after encounters, search intermittently, and follow trajectories that are not linear [5]–[7]. Under a variety of circumstances, these models too predict linear scaling (for a list of conditions, see [6]). Yet, a vital assumption both of older and newer models is that the searcher moves independently of the locations of its targets. In the context of predator-prey interactions, this implies for instance that predators do not alter their movement behavior in response to sensory cues emitted by their prey.
Of course, the assumption that searchers move independently of targets is often made for mathematical convenience. The question is whether models that rely on this assumption capture the salient features of encounter rate kinetics in nature. Empirical studies have shown that inhibiting particular sensory modalities such as chemosensing or flow sensing can dramatically decrease search performance (e.g., [8]), and that sensory cues appear to influence both small-scale [9] and large-scale [10],[11] search behavior. While such studies more rigorously confirm the intuition that using sensory data should improve search performance, little is known about how sensing can influence the functional form of the relationship between encounter rate and target density. Here, we show that sensory response can have a dominant effect on the rate of encounters between searchers and their targets, not only by increasing mean encounter rate, but also by qualitatively changing the form of the relationship between encounter rate and target density.
Below we adopt the language and intuition associated with a predator searching for prey. We assume that the predator samples its environment for sensory cues passively emitted by prey, and adjusts its movement behavior according to explicit mathematical models presented here. This approach builds on a recently developed framework for modeling search decision-making [12] to model the flow of sensory information from prey to predators. We consider three scenarios: (1) perfect sensing and response: the predator can ascertain the precise locations of prey from the sensory data it receives and responds optimally, (2) imperfect sensing and response: the predator detects noisy scent signals emitted by prey and modulates its movement behavior in response, and (3) purely random search: the predator does not use sensory information to guide its movement decisions. We choose (1) and (2) in such a way that they represent upper and lower bounds, respectively, on the acquisition and use of information about prey positions. Our central finding is that there is a systematic shift away from a linear encounter rate function at both of these bounds, suggesting that the collection and use of sensory data may fundamentally alter encounter rate kinetics. We discuss the role of information in governing predator-prey encounter rates, but note that our general methodology could be applied to rates of encounters in other types of ecological interactions (e.g., between mates, competitors, mutualists). We propose that the linear encounter rate models that are typically used to model interaction rates may not correctly capture encounter rate kinetics at low target densities. If this is indeed the case, the most commonly used models of coupled population dynamics, food webs, competition, immune function, and many other systems, may mischaracterize the outcome and dynamics of species interactions.
Studies of biological search typically describe how the type of movement behavior used by a searching organism affects the time needed to encounter its first target , or the rate of target encounters . For consistency with past work, we define as the prey encounter rate of a single predator (e.g., prey per [predator hour], [6]). We assume that predator density is low enough that does not depend on the density of predators, and instead, depends only on the density of prey . We define two encounter rate functions: the mean first encounter rate , and the mean encounter rate after encounters . The latter is often referred to as the encounter rate associated with destructive search [6],[7], emphasizing that the activity of the searcher alters the target landscape. In past studies, the non-destructive search rate is often defined in terms of random variable which represents the time required to find the first target. The empirical first encounter rate is then defined to be where indicates an average over many searches.
To illustrate how sensory information can affect encounter rates, we consider an idealized model of a searching predator in a two-dimensional environment (we discuss search in three dimensions in Text S1). We assume that the predator moves at a constant speed and that prey do not move, at least for the duration of the predator's search. This approximation applies to a wide variety of realistic predator-prey interactions (e.g., large terrestrial carnivores searching for grazing prey). Moreover, incorporating prey movement behavior significantly complicates analysis, for example by introducing the need for a game theoretic formulation of the searcher-target interaction [13]. In the following sections, we further assume that prey density is low, and that handling time is therefore negligible relative to search time. As in past approaches, the predator divides its search into two phases: a scanning phase and a movement phase [12],[14]. This intermittency reflects the observed tradeoff between locomotion and perceptual acuity (e.g., [15]), and the intermittent nature of sampling through major sensory modalities [16]. During the scanning phase, the predator collects sensory data , and encounters any prey within a radius with probability one. During the movement phase, the predator moves a distance at an angle . After moving, the predator re-enters the scanning phase and this process is repeated. Thus, the predator's movements consist of a set of movements with scanning phases in between. The process the predator uses to determine and constitutes its search strategy.
When prey density is low, a predator will typically detect little or no signal and proceed with the information that the target is not likely to be nearby. When prey density is high, an increasing fraction of the landscape is covered by regions that are within the encounter radius of a prey item. In this way, our modeling framework naturally captures a predator's behavioral transition between these two regimes. For example, if prey are distributed according to a Poisson process, a simple Poisson thinning argument shows that the probability that a given location is not within the encounter radius of a prey item is given by . Thus, as prey density increases, there is a seamless transition from limited to perfect information about where the nearest prey is located.
In addition to the analytical results described above, we used search simulations to compare the behavior of a predator that moves according to a purely random strategy to a predator with imperfect sensing and response. In both cases, we assume that the intrinsic movement behavior is described by a symmetric two-dimensional Pareto distribution. Because of the symmetry we can separately draw the turn angle and the move length , where is the density of a Pareto random variable,(3) is a minimum move length, and is a parameter that determines whether the walk is superdiffusive (superdiffusive for ). We use a Pareto distribution with a power law tail to model intrinsic movement behavior because it has been argued that such a distribution may have evolved as a statistical movement strategy for locating resources when sensory data are not useful [19]. In Text S1, we show that our qualitative results hold when predators move according to a diffusive random walk. As shown in past work [12], the use of sensory data to make movement decisions dominates intrinsic movement behavior so that the distinction between diffusive and superdiffusive intrinsic strategies becomes relatively unimportant.
In each simulation, we placed a single predator in a prey periodic environment (i.e., environment was a torus) and populated the environment with a Poisson number of prey with mean 600. The size of the environment was then scaled to achieve the desired prey density. In each scanning phase, was sampled from a Poisson distribution with mean given by Equation (S2) (see Text S1) summed over all prey. In each simulation, the searcher was positioned at a random location and allowed to move through the environment until it came within a distance of of a prey item during its scanning phase. We designated this an encounter and the prey was destroyed. For each strategy, we performed 1,000 simulations and recorded the time until the desired number of prey encounters was achieved. Predators were assumed to travel at a constant speed of one body length per second, which is a realistic speed for foraging predators.
We performed two sets of simulations. In the first, prey positions were generated using a Poisson point process. We then recorded the time required for the predator to encounter the first prey and used this to compute encounter rate . This is consistent with a scenario in which predators search for and capture a single prey item, and then cease to forage for a period of time, during which prey redistribute themselves in the environment. When predators encounter and destroy multiple prey in succession, they can create local zones of prey depletion. To determine whether the scaling of encounter rate is sensitive to such a local depletion effect, we allowed predators to encounter and destroy 32 prey items. We then computed , where was the mean time required to encounter of the prey present on the environment. We chose as a compromise between maximizing the number of targets encountered in a single search, while ensuring that this local depletion did not substantially change global target density. For reasons of computational efficiency, we wished to limit the mean number of targets in the environment to 600, and therefore chose of targets as a reasonable compromise. Finally, to determine whether the scaling of the encounter rate depends on the distribution of targets, we generated prey distributions according to a highly clustered point process that we will call a preferential attachment model and repeated simulations to compute and for . Briefly, prey were generated by drawing from a Poisson distribution with mean 600. The size of the environment was then scaled to achieve the desired prey density. A fraction of the prey were chosen to act as seed points and placed uniformly at random on the space. The remaining prey were each assigned as daughters to one of the seed points iteratively with probability , where is the number of daughters around the ith seed point. Positions of daughters were assigned uniformly within a circle of radius around the seed point, where was chosen so that all clusters had the same local prey density.
Our primary goal was to characterize the form of the encounter rate function in the low prey density regime. We simulated predators exploring environments with prey densities ranging from 0.5–100 prey per squared predator body lengths. All simulations were performed using Matlab. The lower limit of the prey density range was chosen based on realistic low prey densities encountered by predators in nature. For example, Serengeti lions experience densities of ungulate prey that can be as low as 0.3–10 prey per squared predator body lengths; snow leopards experience densities of their primary prey, blue sheep, as low as 7 prey per squared predator body lengths; and northern hawk owls capture and consume small rodents with densities as low as 40 prey per squared predator body lengths [24]–[26].
As in previous investigations (e.g., [5]), we expected that would be a linear function of for the purely random predator. On the other hand, as shown above, the predator with perfect sensing and response has an encounter rate function with several scaling regimes in the range of densities that interest us: one in which encounter rate is proportional to , and one in which encounter rate is proportional to . To accommodate these functional forms, we assumed that locally, the encounter rate can be described by a power function of the form . This allows for both linear and sublinear scaling. To determine whether simulated predators had multiple scaling regimes, we fitted (i) a single power function, (ii) a segmented function with two distinct scaling regimes, and (iii) a segmented function with three distinct scaling regimes. Prior to fitting, we log-transformed density and encounter rate data from search simulations. We used a recently developed statistical method for simultaneously estimating both the break points between distinct scaling regimes and the scaling exponents in each regime [27]. Briefly, this technique allows one to fit a piecewise regression model in which the locations of the break points in the piecewise function are fitted parameters in the model. This technique is particularly useful when fitting functions to data when there is no a priori knowledge of the precise transitions between regimes. We compared the fits models with a single regime to models with multiple regimes using . We computed as the AIC value of the model with one scaling regime minus the AIC value of the best fitting model with two or three scaling regimes. Statistical analyses were conducted using the Segmented package [28] in R [29].
There is a dramatic difference between movement patterns of predators that use sensory data and those that do not. As is evident from Fig. 2, predators with imperfect sensing and response concentrate scanning effort near prey (Fig. 2A), whereas purely random predators scan roughly uniformly over the environment (Fig. 2B). In the study of animal search, concentration of effort near targets is known as area-restricted search (ARS). ARS is a canonical feature of search behavior in nature [30]. It is interesting that this seemingly complex behavior can emerge from an extremely reduced sensing and decision making process like the one modeled here [12].
Signal-modulated predators perform ARS because they move short distances between scans when they receive strong sensory signals and move long distances when they measure weak signals [12]. This behavior improves search efficiency, but perhaps more importantly, it leads to a qualitatively different relationship between the encounter rate of signal-modulated predators and their prey (Fig. 3A). As expected from past work on random search [5],[6], purely random predators encounter prey at a rate that scales nearly linearly with across all prey densities. The encounter rate of signal-modulated predators, on the other hand, is strongly nonlinear in (compare Fig. 3A yellow points to blue triangles). In particular, at low but realistic prey densities (Fig. 3A blue curve), the encounter rate of signal-modulated predators changes sublinearly with changing prey density. This anomalous scaling makes the encounter rate of signal-modulated predators more robust with respect to changes in prey density (see Implications for coupled population dynamics below).
Predators that used a purely random search strategy encountered prey at a rate that was nearly proportional to prey density (Fig. 3A, yellow circles; ; ). This near-linear scaling held when prey were clustered and also when predators encountered and destroyed multiple prey per search (). The encounter rate function did not show evidence of multiple scaling regimes ( in both clustered and uniform environments with and ). This result mirrors that of particle collision models and other random-walk-based models of organismal search, which all predict that encounter rate is proportional to target density when target density is low.
Across all densities studied, predators that use sensory data to make movement decisions encounter prey at a higher rate than predators that do not use sensory cues (Fig. 3A, B). Indeed, at low and intermediate densities, signal-modulated predators encounter prey at a rate that can be 5–14 times higher than the encounter rate of random predators (Fig. 3B).
As prey density increases, the encounter rate of signal-modulated predators increases nonlinearly and clearly displays multiple scaling regimes (Fig. 3A, blue triangles; AIC single regime minus AIC three regimes = 682). At the lowest densities, encounter rates increased linearly or superlinearly with prey density. For the particular parameter values explored here, there is a transition to a second scaling regime at however, the exact transition depends on the length scale of scent detection (Fig. S1). In the second, intermediate regime, which covers low but realistic prey densities, signal-modulated predators encounter prey at a rate proportional to , where . The value of the scaling exponent , is close the square-root scaling exhibited by the searcher with perfect sensing response. For higher densities, data indicated a third regime, in which encounter rate increased superlinearly with prey density () corresponding to conditions in which prey are plentiful and predators do not need to search. The qualitative form of the encounter rate function of signal-modulated predators in a uniform prey environment was preserved when prey were highly clustered, and when predators encountered and destroyed multiple prey items in a single search. Fig. 4 shows that the mean encounter rate after encounters exhibited near-linear regimes at relatively high and low densities, and sublinear regimes at intermediate densities ( in sublinear regime).
In addition to engaging in area-retricted search, signal-modulated predators successfully locate nearby prey more frequently than purely random predators (Fig. 5A, upper diagram), which often wander away from nearby prey without encountering them (Fig. 5A, lower diagram). To examine this more rigorously, we isolated all occasions in which a predator came within a distance of a prey and defined these as proximity events. A proximity event ends when the predator encounters a prey item, or moves to a location that is at least from any prey. We computed the fraction of proximity events that resulted in encounters and defined this as the empirical encounter probability. We chose because at that distance, predators have a probability of only 0.05 of measuring a non-zero scent signal in s meaning that a predator beyond this distance is very unlikely to receive any further scent information from prey. Fig. 5A shows that the empirical encounter probability of signal modulated predators (Fig. 5A blue points) is higher for all prey densities, and approaches 1 for prey densities above 10, indicating that signal-modulated predators do not miss nearby targets when density is high. Purely random predators miss nearby prey even as prey density approaches 100 (as approaches 100, the typical distance between adjacent prey approaches the encounter radius body lengths). At low density, encounter probabilities of both types of predator approach constant values. For the signal-modulated predator, this value is 0.17, similar to the value of 0.23 predicted for a Brownian searcher with constant diffusivity (see Text S1). Fig. 5B shows that this minimum encounter probability is roughly three times higher for signal modulated predators than for purely random predators.
Our results demonstrate that the use of sensory information alters encounter rate kinetics, both at the extreme of perfect information and decision-making, and at the other extreme of minimal sensing and rudimentary decision-making. In studies of coupled population dynamics, the encounter rate function is a central component of the functional response, the relationship that couples prey and predator populations. Given the anomalous scaling of encounter rate shown by predators that use sensory information to make movement decisions, a natural question is whether such predator search behavior might affect coupled population dynamics. Here, we explore this question.
Predator-prey dynamics can be modeled by the following system of equations:(4)which is a generalized version of the Rosenzweig-MacArthur model [31]. The variables and represent the prey and predator densities (number per squared predator body lengths) respectively. The intrinsic growth rate of the prey population is given by the parameter , and prey growth is limited by the carrying capacity, , in the absence of predators. The predator population is assumed to die at rate and the parameter is a measure of predator energy conversion efficiency. All time rates are on a per day scale.
To relate our encounter rate findings to coupled population dynamics, we must translate the encounter rate into a long-term functional response . This is necessary because search and reproduction take place on distinct time scales (e.g., hours versus years, respectively). We consider a scenario that reflects the general theme of the work presented here: a regime where prey are sparse, but rare encounters are sufficient to sustain a predator. We assume that predators undertake a succession of hunting expeditions each day. The predator engages in hunts per day, each lasting a period before the predator relents. The hunt is assumed to end if a prey is captured. Therefore the number of consumption events in a given day, which we denote can be written as the sum of indicator functions which take the value one or zero depending on whether the corresponding search expedition is successful. The success probabilities depend on the encounter rate and the respective search durations . Adopting the simplest assumption for the search distribution, we take each . Then, treating the encounter process as a Poisson process with rate parameter , it follows that, given the value of , the success probability of the jth hunt is . Using Wald's equation, a simple calculation reveals the following form for the functional response:Note that this form closely resembles the Holling type II functional response [32]. Note also that it is increasing in prey density, concave down, and satisfies
We now proceed to study the effect of the form of on the outcome of predator-prey dynamics. The mathematical structure of (4) is more readily apparent if we introduce the functions and to generalize the autonomous growth function and functional response, respectively. The function should be zero when the prey density is equal to zero or , differentiable, and concave down everywhere. We take , and writing we have(5)We see that if then when . Moreover, if there exists a coexistence fixed point with and , where(6)Under suitable conditions, which biologically plausible forms of and will generally satisfy, this fixed point is unique. In Text S1 we analyze the stability of the coexistence fixed point for systems with encounter rates of the form where . We show that there is a critical value such that if the coexistence fixed point is stable. Otherwise, it is unstable; however, numerical studies indicate the presence of a stable limit cycle that contains the fixed point. Notably, for values of that lead to very low values of , the coexistence fixed point is unstable. This is true for all models considered here, including those with a linear encounter rate function. This fundamental instability is due to the nonlinear nature of the Holling Type II form we use to translate the encounter rate into functional response.
To explore how the form of the encounter rate function affects the outcome of coupled population dynamics, we parameterize the population model described above for the sparse prey regime (Fig. 6). This analysis demonstrates several differences between population dynamics involving sensory predators and more traditional models that assume a linear encounter rate function. First, predators that use sensory data deplete prey to lower densities than predators that search randomly. The ability to deplete prey to low levels is a critical trait in ecological dynamics; for example, R* theory posits that a species' competitive ability is determined by its ability to deplete resources and persist when resources are rare [4]. Fig. 6A shows the steady state density of prey as a function of the ratio of predator conversion efficiency to predator mortality rate . Both signal-modulated predators and predators with perfect sensing and decision-making reduce prey density to lower levels than do purely random predators (Fig. 6A, blue and cyan curves are below yellow curve). To demonstrate the extent to which the nonlinearity in the encounter rate function of signal-modulated predators contributes to this pattern, we added a “linearized signal-modulated encounter rate” (Fig. 6A orange curve), which matches the signal-modulated predator for prey densities above 20, but remains linear for lower densities. Signal-modulated predators (Fig. 6A, blue curve) reduce prey density well below that of the linearized analogue (Fig. 6A, orange curve) illustrating the substantial effect of the nonlinearity in the encounter rate function. We note that throughout the depicted regime, the coexistence fixed point is unstable, but with a containing stable limit cycle.
A second important observation is that, for a given value of the parameter , signal-modulated predators persist at lower prey carrying capacity than purely random predators, and predator density of signal-modulated predators is less sensitive to variation in . Fig. 6B shows steady state density of predators (dashed lines) and prey (solid lines) as a function of prey carrying capacity. Steady-state prey density does not depend on carrying capacity until becomes so low that predators no longer persist. Below this point, prey steady state density is equal to . Steady state density of purely random predators rapidly decays to zero below . Steady state density of signal-modulated predators, on the other hand, remains insensitive to carrying capacity for greater than approximately 2.
Our results demonstrate that the use of information about the position of targets fundamentally alters the relationship between encounter rates and target density. Not only do predators that use sensory information encounter prey more often, but they are less sensitive to changes in prey density. This is true even when sensory cues contain a minimal amount of information about target locations, and searchers do not remember past signals. This increased robustness provides an ecological mechanism through which sensory response may allow predators to cope with fluctuations in prey density. Moreover, it can alter coupled population dynamics. These findings are robust to a range of assumptions about target distribution, capture behavior, and the length over which searchers detect scent signals (Text S1).
Reaching a general understanding of the effect of sensory data on species encounter rates is challenging. Searching organisms collect a wide variety of sensory data and biologists do not know, in general, how they use these data to make decisions [17]. Here, we have taken the approach of studying two extreme cases of the collection and use of sensory data. In the extreme of perfect sensing and response, predators encounter prey at a rate proportional to prey density to the power at low prey density (where is the dimension of the search environment), rather than exhibiting the linear scaling predicted by models of purely random search. At the opposite extreme, when we perturb purely random search behavior by introducing a very limited capacity for sensing and decision-making based on a noisy, directionless signal, the encounter rate function immediately departs from the linearity expected when predators move without using information [2],[5],[6]. This observation has immediate implications for classical population models, where the encounter rate informs the functional response. The functional responses that are most commonly used in predator-prey models are either linear in prey density (e.g., Holling type I), or nonlinear and concave down at high densities to incorporate effects of satiation and handling time when prey density is high (e.g., Holling type II). These forms are linear when prey density is low. By contrast, we argue that in the important regime where prey are rare and predators must search for them, the functional response is nonlinear and concave down. This change in the form of the functional response makes predators less sensitive to changes in prey density and can alter the outcome of predator-prey interactions by allowing predators to deplete prey to lower levels and persist with prey over a broader range of prey carrying capacity.
Our work suggests several ways to better integrate experiments with models of encounter rates. For example, we suggest that encounter rate and functional response of predators should be nonlinear at low prey densities. Yet, most experimental studies of encounter rates and functional response measure rates at high prey density, where handling time and predator satiation determine the shape of the rate function (but see [33],[34], which show concave down encounter rate functions in hunting fish and birds as we predict). Data from carefully designed experiments are needed to determine the most appropriate forms of encounter rate functions and functional responses at low prey density. Distinguishing “high” from “low” prey density is not arbitrary; rather, high and low density regimes are determined by the length scale of predator-prey encounters (e.g., predator striking distance) and by the length scale of the propagation of sensory signals. Future experimental work should evaluate the scaling of encounter rate with when the typical distances between prey are similar to or greater than the distance at which predators can acquire sensory cues from prey.
Finally, we note that nonlinearity of the encounter rate function depends on the ratio of the length scale of sensory signal transmission to the length scale at which encounters occur. When predators can only detect prey that are very nearby (i.e. detection distance/encounter distance ), sensory information does not strongly affect search performance [12], and mass action kinetics may provide a reasonable description of encounter rate kinetics. For example, in predator-prey interactions at low Reynold's number, cruising predators may still use sensory information to make movement decisions, yet the length scales associated with prey detections can be very short (e.g., less than one predator body length) relative to the distances between adjacent prey [35].
Our results show that introducing a response to even relatively information-poor, noisy sensory signals qualitatively alters the relationship between predator-prey encounter rate and prey density in many biologically plausible scenarios. Behaviors such as area-restricted search emerge naturally from our model of search behavior, even in the absence of signal gradients, complex signal processing, and memory of past signal and target encounters [12]. The framework we introduce here can be used to understand the connection between information and the encounter rates that are so critical to many core concepts in biology.
|
10.1371/journal.pcbi.1000658 | Acute Effects of Sex Steroid Hormones on Susceptibility to Cardiac Arrhythmias: A Simulation Study | Acute effects of sex steroid hormones likely contribute to the observation that post-pubescent males have shorter QT intervals than females. However, the specific role for hormones in modulating cardiac electrophysiological parameters and arrhythmia vulnerability is unclear. Here we use a computational modeling approach to incorporate experimentally measured effects of physiological concentrations of testosterone, estrogen and progesterone on cardiac ion channel targets. We then study the hormone effects on ventricular cell and tissue dynamics comprised of Faber-Rudy computational models. The “female” model predicts changes in action potential duration (APD) at different stages of the menstrual cycle that are consistent with clinically observed QT interval fluctuations. The “male” model predicts shortening of APD and QT interval at physiological testosterone concentrations. The model suggests increased susceptibility to drug-induced arrhythmia when estradiol levels are high, while testosterone and progesterone are apparently protective. Simulations predict the effects of sex steroid hormones on clinically observed QT intervals and reveal mechanisms of estrogen-mediated susceptibility to prolongation of QT interval. The simulations also indicate that acute effects of estrogen are not alone sufficient to cause arrhythmia triggers and explain the increased risk of females to Torsades de Pointes. Our results suggest that acute effects of sex steroid hormones on cardiac ion channels are sufficient to account for some aspects of gender specific susceptibility to long-QT linked arrhythmias.
| It is well known that female gender is an independent risk factor for some types of cardiac arrhythmias. However, it has been difficult to determine how much of a role physiological concentrations of circulating sex steroid hormones play in gender linked arrhythmia susceptibility because the cardiac system is so extraordinarily complex. Here we employ a computational strategy, based on experimental measurements, to tease out the individual contributions of estrogen, progesterone and testosterone on cardiac electrical behavior and then make predictions about their effects in combination and in the presence of drugs. The computational models convincingly reproduce observed fluctuations of QT intervals (as recorded on the ECG (electrocardiogram), the QT interval reflects the time period between ventricular excitation and relaxation) through the menstrual cycle in females and effects of testosterone on ECG parameters. Our simulations also predict that testosterone and progesterone are protective against drug-induced arrhythmias, while estrogen likely exacerbates the breakdown of normal cardiac electrical activity in the presence of QT-prolonging drugs.
| In the past decade, studies have suggested that female gender is an independent risk factor for long-QT (LQT) dependent cardiac arrhythmias [1]–[3]. Since the differences in QT intervals in males and females appear from the time of puberty [4],[5], sex steroid hormone effects on cardiac repolarization have been implicated. Clinical studies have found no difference in QT interval in male and female children, but shorter QT intervals in men versus women under age 50 [4]. The international Long QT syndrome (LQTS) registry 1998 reported that females had higher risk of a first cardiac event between 15 and 40 years [6]. Moreover, clinical findings observed that more than 68% of drug-induced torsade de pointes (TdP) occur in women [7]–[9].
It is known that one way that sex steroid hormones cause functional physiological changes is via transcriptional regulation. Sex hormones may bind to sex hormone receptors and then translocate into the nucleus. In the nucleus, a ligand-bound sex hormone receptor acts a transcription factor by binding to the promoter region of genes containing a hormone responsive element (HRE), leading to regulation of gene expression. For example, in the heart, lipocalin-type prostaglandli D synthase (L-PDGS) has been found to be transcriptionally upregulated by estradiol and estrogen receptor (ER) [10]. This genomic action requires several hours before the effects can be observed. In addition to the genomic pathway, sex steroid hormones may induce a rapid activation of mitogen-activated protein kinase (MAPK) leading to transcription factor activation [11],[12] as well as activation of membrane bound endothelial nitric oxide synthase (eNOS) [13],[14].
Interestingly, recent studies have demonstrated that sex steroid hormones may also act acutely and rapidly modulate cardiac ion channel activity directly via a PI3K/Akt/eNOS pathway [15]–[17]. Testosterone induced phosphorylation of the Ser/Thr kinase Akt and eNOS leads to NO synthase 3 (NOS3) activation and production of nitric oxide (NO) [15]. NO leads to s-nitrosylation of cysteine residues on the channel underlying the slow delayed rectifier K+ current (IKs) [17]. L-type Ca2+ current (ICa,L) is conversely suppressed by NO via a cGMP dependent pathway. Regulation of IKs and ICa,L by testosterone is dose-dependent [15] and leads to shortening of action potential duration (APD) [15] and QT intervals [18]–[20]. In adult men, the serum testosterone level is reported to be 10 to 35 nM [21], however circulating levels of testosterone begin to decline in men as young as 40 [22]. QT intervals are shorter in adult men than in adult women until around the age of 50 [4], suggesting a likely role for circulating testosterone.
In females, progesterone fluctuates through the menstrual cycle. The reported serum progesterone level is 2.5 nM in the follicular phase and 40.6 nM in the luteal phase [23]. It was recently shown by Nakamura et al. that progesterone increases IKs current through the NO production pathways and prevents cAMP-enhancement of ICa,L [16].
The apparent result of acute effects of progesterone and testosterone is to shorten ventricular repolarization and diminish incidence of arrhythmias [15],[16],[20],[24]. Recently, experiments have suggested protective effects of testosterone against arrhythmia. In vivo experiments show that orchiectomized male rabbits treated with dihydrotestosterone (DHT) had shorter QT interval and APD90 compared to non-DHT treated rabbits [18],[20]. Also, experiments in testosterone treated female animals have shown that DHT reduces drug-induced arrhythmia by dofetilide [24].
The acute effects of estradiol result in suppression of human ether-a-go-go-related gene (hERG) underlying the rapid delayed rectifier current (IKr) by directly binding to the channel, altering channel kinetics and reducing current [25]. Kurokawa and co-workers showed that 17β-estradiol (E2) increases the channel rate of closure (deactivation) and lessens repolarizing current. They also showed that in the presence of E2, hERG is more sensitive to block by drugs. The group proposed that aromatic centroid of E2 may be responsible for increasing the sensitivity of hERG block by E4031 via interaction with the aromatic side chain of Phe656 and aromatic rings of the hERG blocker. Because 1) the concentration of E2 is not constant through the menstrual cycle, but rather fluctuates from the peak follicular phase serum E2 level of 1 nM to 0.7 nM in the luteal phase, and 2) E2 has dramatic effects on sensitivity to hERG block within this range, it stands to reason that susceptibility to drug-induced arrhythmia by hERG block may vary through the menstrual cycle.
Although studies have shown that female hormones estradiol and progesterone have opposite effects on cardiac repolarization: E2 prolongs QT intervals, and progesterone reduces QT interval [16],[25],[26], the question of whether normal hormonal fluctuations are sufficient to account for variability in QT during the menstrual cycle in not known. Neither are the effects of physiological concentrations of hormones on arrhythmia susceptibility well understood. Some studies do report that dynamic fluctuations in QT intervals during the menstrual cycle are related to changes in susceptibility to TdP risk [27],[28]. Other studies in postmenopausal women also suggest the importance of female hormones as estrogen hormone replacement therapy prolongs QT intervals and increases arrhythmia risk [26],[29],[30]. Other data have not found marked fluctuation in QT interval during specific phases of the menstrual cycle [28],[31],[32]. Burke et al., (1997) found that the corrected QT (QTc) interval does not significantly change through menstrual cycle in pre-menopausal women; however, QTc is reduced in the luteal phase after autonomic blockade [31]. A study of drug-induced QT prolongation during the menstrual cycle observed that QTc did not vary during the menstrual cycle, but QTc shortening was more pronounced in the luteal phase with ibutilide application [28]. Nonetheless, both the clinical and experimental data suggest that women have both longer QT intervals than men and are more likely to develop long-QT dependent arrhythmias and TdP arrhythmias [9],[28]. Women are especially susceptible to increased arrhythmia risk in response to QT-prolongation drugs [9],[28],[33],[34].
It is a major challenge to specifically determine the relationship between sex steroid hormones and arrhythmia susceptibility in males and females since the cardiac system is extraordinarily complex. In order to attribute risk to a particular parameter, in this case physiologically relevant concentrations of sex steroid hormones, the specific effect must be studied in isolation without other perturbations to the system. This is the strength of the computational approach that we employ. In the present study, we focus on acute effects of sex steroid hormones on cardiac ion channel targets. We use guinea pig models that incorporate the effects of hormones measured experimentally from guinea pig, and then can test these changes specifically within the complex cellular and tissue milieu.
Importantly, we use the model to make predictions about the effects of physiological concentrations of sex steroid hormones on gender specific cardiac physiology parameters and arrhythmia susceptibility. Some recent experimental studies investigating functional effects of sex hormones on cardiac function have utilized hormone concentrations in the micromolar range that is orders of magnitude higher than the nanomolar physiological circulating concentration of E2 [35]. This is a critical consideration because micromolar concentrations of E2 are apparently cardioprotective via effects on L-type Ca2+ current (ICa,L). Although high hormone concentrations may be relevant during phases such as pregnancy, a recent study showed that E2 at 1 nM did not have significant effects on IKs or ICa,L [25]. Our model simulations reproduce observed fluctuations of QT through the menstrual cycle in females in both cell and tissue-level. Simulations also predict that effects of testosterone and progesterone on ion channels hasten repolarization and protect from drug-induced arrhythmias.
To investigate the acute effects of sex steroid hormone on cardiac electrophysiology and arrhythmia susceptibility, we developed a computational model that mimicked the conditions employed experimentally so that we could directly validate our model by comparison to experimental measurements. Experiments were conducted in isolated ventricular myocytes from Langendorff-perfused adult female guinea pigs, so that they were free of endogenous neuronal and hormonal effects. The isolated cells were then incubated with human physiological concentrations of hormones for 10 min. and the effects of hormones on cardiac ion channels were measured. A range of cardiac ion channels were screened for functional changes induced by sex steroid hormones, but acute effects of progesterone were found only to modify IKs [16] while testosterone primarily increased IKs and inhibited ICa,L [15]; acute E2 treatment only significantly suppressed IKr current [25]. We utilized the experimentally observed effects of physiological concentrations of sex-steroid hormones in adult women and men and incorporated these functional changes into our computational models (described in detail in Supplemental Text S1).
Experiments [25] show that E2 primarily affects the conductance of IKr, and has a minor, but measurable and significant effect on slowing channel activation kinetics. To simulate the experimentally observed IKr current reduction by E2 (Figure 1A – right), we scaled the IKr conductance and incorporated the minor effects of E2 on the voltage dependence (not shown) of IKr in the model (Figure 1A – left). E2 at 1 nM reduced IKr tail current in a dose-dependent manner, but did not affect the time course of tail current decay (Figure 1A).
Unlike the direct effects of E2 on IKr, progesterone modulates the IKs through non-genomic activation of eNOS. We used experimental data [16] (Figure 1B – left traces) to scale the conductance of ionic currents in the model to incorporate effects of progesterone on IKs. Progesterone-induced IKs enhancement is concentration-dependent as shown in Figure 1B. Experimentally recorded and simulated dose-response curves for progesterone effects on IKs tail current amplitude is shown in Supplemental Figure S1. IKs current was simulated with different concentrations corresponding to progesterone concentrations at various points in the menstrual cycle (0 nM – control case, 2.5 nM – follicular phase, 40.6 nM – luteal phase and 100 nM - maximal experimental concentration) during a voltage pulse from −40 mV to +50 mV. Note that the effect of progesterone on IKs is nearly saturated at a concentration of 40.6 nM, corresponding to the peak value during the luteal phase of the menstrual cycle (indicated by the near overlay of the 100 nM curve).
Like progesterone, testosterone modifies cardiac ion channels comprising IKs and ICa,L via eNOS production of NO. We used the same method as above to incorporate experimental ratios of control conductance for testosterone. Dose-dependent effects of testosterone on IKs enhancement and ICa,L suppression are shown in Figure 1C for experiments (top) and simulated currents (lower panels). Simulated IKs and ICa,L are compared to experimentally recorded guinea pig IKs and ICa,L using the same protocol. Cells were depolarized to test potential +50 mV for 3.5 seconds and then repolarized to −40 mV to record IKs. ICa,L was experimentally recorded during a voltage step from −40 mV to 0 mV. Testosterone strongly enhances IKs current (Figure 1C – left traces) at 10 nM while high concentrations of testosterone (300 nM) markedly suppress ICa,L (Figure 1C – right traces).
Like humans, many studies have demonstrated that female guinea pigs have slower repolarization than male guinea pigs [1],[36]. To examine the contribution of sex-steroid hormones on the ventricular action potential duration (APD), we included the effects of E2, progesterone and testosterone on membrane currents and simulated action potentials (APs) in three cell types. Figure 2 shows APs for the 50th beat at 1000 ms pacing rate in M cells. Simulated APs of epicardial and endocardial cells are described in Supplemental Figure S2.
E2-induced IKr suppression contributes to APD prolongation in a dose-dependent manner (Figure 2A). A low concentration of E2 (0.1 nM), corresponding to the early follicular phase of the menstrual cycle, has slight effects on APD compared with control case (from 233 to 235 ms — 0.86% prolongation). However, a concentration of E2 corresponding to the late follicular phase of the menstrual cycle (prior to ovulation) (1.0 nM) prolonged APD (250 ms) by 7.3% (Figure 2A). This value is in good agreement with the observed APD prolongation in guinea pig myocytes in patch-clamp experiments with E2 incubation (11±1%) [25]. Figure 2B shows that progesterone reduced APD in a concentration-dependent manner (222 ms — 4.7% reduction at 2.5 nM corresponding to the follicular phase; 212 ms — 9.0% reduction at 40.6 nM, corresponding to the luteal phase), which agrees with patch-clamp experimental data (6.3% reduction at 40.6 nM) [16].
To investigate the combined effects of E2 and progesterone as they fluctuate during the normal menstrual cycle on the cardiac action potential, we used clinically measured concentrations of hormones at three discrete phases of the menstrual cycle (early follicular, late follicular and luteal). During the early follicular stage, E2 = 0.1 nM, progesterone = 2.5 nM, during the late follicular stage, E2 = 1.0 nM, progesterone = 2.5 nM and during the luteal stage, E2 = 0.7 nM, progesterone = 40.6 nM [23]. As see in Figure 2C, the simulations predict longer APD in the late follicular phase (233 ms) than in the early follicular (223 ms — 4.3% reduction). Simulations predict shortest APD in the luteal phase (218 ms — 6.4% reduction), consistent with experimental observations (≈11% shortening) [36].
We also simulated changes in APD at two physiological concentrations of testosterone (10 nM and 35 nM) shown in Figure 2D, which reflect the normal low and high ranges found in post-pubescent pre-senescent males [21]. The simulations predict marked APD shortening by 10.7% (208 ms) and 15.9% (196 ms) at 10 and 35 nM testosterone, respectively.
We next computed the effects of sex-steroid hormones in a one-dimensional strand of coupled M cells (results from other cell types are shown in Supplemental Figure S2) to determine the effects of hormones in an electrotonically coupled system (Figure 3). We also computed spatial gradients of depolarization and repolarization to generate a pseudo ECG electrogram (Figure 3B). APs were initiated via a stimulus applied to the first cell and then propagated from top to bottom along the 1 cm fiber. Figure 3A show that the first cell fired first and then repolarized first.
The effects of E2 on IKr leads to dose-dependent APD prolongation in the simulated tissue (Figure 3A), and results in a longer QT interval in the presence of 1 nM (7.7% prolongation) from 260 ms (Figure 3A-i — 0 nM sex-steroid hormone) to 280 ms as seen in Figure 3B (top panel). Also, the simulations clearly show progesterone shortened APD in a dose-dependent manner (3A-iv 2.5 nM, and 3A-v 40.6 nM). The corresponding computed electrograms from the fibers in Figure 3B (lower panel) demonstrates the progesterone-induced QT interval reduction from 260 ms (control case) to 250 ms (3.8% — iv) and 240 ms (7.7% — v).
A recent clinical study has observed that the QT intervals fluctuate during the menstrual cycle, suggesting that progesterone may reverse effects of the estrogen-induced QT prolongation [27]. Figure 4A represents the results of simulations in a 1D cable at combined hormone concentrations observed during various phases of the menstrual cycle. Simulations show a QT interval reduction of 10 and 20 ms in the luteal phase compared to the early and late follicular phases, respectively (Figure 4B — top panel), which agree with the clinically observed QT shortening (≈10 ms shortening in the luteal phase compared to the follicular phase) [27]. The models demonstrate that despite the presence of E2 (0.7 nM) during the luteal phase, high progesterone (40.6 nM) results in luteal phase shortening of APD and a QT interval (on the pseudo-ECG) reduction of 4% (from early follicular phase) and 7.7% (from late follicular phase).
The experimental study from Liu et al. suggested the QT intervals were significantly shorter (11.3%) in male than in female rabbits [37]. In Figure 4A-iv and Figure 4A-v, our simulations show the effects of testosterone on APD in simulated one-dimensional tissue. The model predicts that testosterone-induced faster repolarization and caused QT interval reduction to 230 ms (11.5% shortening — case iv) and 220 ms (15.3% shortening — case v) compared with the late follicular phase (260 ms) in Figure 4B. We also ran these simulations in the presence of 10 nM and 35 nM testosterone and 0.1 nM E2, which is estimated as the average circulating concentration of E2 in men [28] (shown in Supplemental Figure S3). In the presence of E2, QT intervals increase by 10 ms, corresponding to 7.7% (10 nM) and 11.5% (35 nM) shortening compared to the late follicular phase in females.
Experimental evidence suggests that in the presence of physiological concentrations of E2, the potency of IKr block by drugs is increased [25]. This finding may explain, in part, the increased susceptibility of females to drug-induced arrhythmias [8],[9]. Hence, we next tested the effect of E2 on IKr suppression induced by the IKr channel blocker E-4031 and investigated the effects of female hormones on drug-induced arrhythmia susceptibility. Experimental results [25] shown in Figure 5A (top) illustrate that E2 (1 nM) considerably increased the suppression of hERG by E-4031 (light gray line). However DHT did not greatly change the drug-induced inhibition of hERG current (dark gray line). We then obtained measured ratios of IKr conductance in the presence of E-4031 and E2 or DHT from the experimental data and used these values to simulate dose-dependence curves for IKr suppression by E-4031 (control — black line) and after addition of 1 nM E2 (light gray line) and DHT 3 nM (dark gray line) (Figure 5A — lower panel). In Figure 5B (top panel), we show a simulation of a one-dimensional strand of coupled M cells (100 cells) in the late follicular phase during E-4031 treatment, where the model predicts the most dramatic APD and QT interval prolongation. At 10 nM E-4031, the simulated tissue-level APD is shorter with testosterone application (250 ms — 3 nM) compared with APD in the presence of female hormones (280 ms — E2 = 1.0 nM, progesterone = 2.5 nM) as seen in Figure 5B. The pseudo ECG (5B — lower traces) shows that QT interval is substantially longer in the late follicular phase (case i) than with testosterone treatment (case ii).
The exact mechanism of TdP induction is unclear, but it is thought that pause-induced early afterdepolarizations (EADs) can underlie TdP initiation [38],[39]. Hence, we performed a series of simulations to investigate pause-dependent LQT syndrome and its association with arrhythmia susceptibility in the presence of male and female hormones. Single M cells were paced for 10 beats of BCL at 1000 ms (s1) followed by a premature beat (s2) with varying s1–s2 intervals and then a long pause of varying duration as indicated. Our simulations show no EADs (APD>450 ms) occurred during the late follicular phase with no drug application (Figure 6A — left panel) or with the application of E-4031 in the presence of testosterone 3 nM (middle) during a short-long pacing protocol. However in the absence of sex-steroid hormones, EADs were generated by addition of 10 nM E-4031 when the pause interval was very long (>2500 ms) (right panel).
In Figure 6B, we investigated the short-long pacing induced EAD by E-4031 in the late follicular phase, where the concentration of E2 is highest, after pacing at three basic cycle length (500, 750 and 1000 ms). This pacing sequence triggered EADs over a wide range of pauses in all three conditions. APDs of the s3 (post pause) beat are notably lengthened with increasing basic cycle lengths from 500 ms to 1000 ms (Figure 6B — left panel to right panel). Severe EADs were induced at 1000 ms pacing length with a pause greater than 1500 ms (6B — right panel). The point in Figure 6B (right) indicates an EAD that was triggered following a pause of 1500 ms and s1–s2 interval of 810 ms during baseline pacing length of 1000 ms. We have carried out the simulations in a coupled one-dimensional M-cell tissue (6B — lower panels) using the same protocol, and observed propagation of the EAD in the tissue. These simulations suggest that 3 nM testosterone is sufficient to prevent EAD development in the presence of E-4031 10 nM. However, in females, during the late follicular phase of the menstrual cycle, the increased concentration of estrogen appears to exacerbate drug-induced TdP arrhythmias.
Finally, to test the potential for E2-exacerbated EADs to trigger reentrant arrhythmia activity in 2D heterogeneous tissue, we carried out a series of simulations with varying combinations of hormones and/or drug application. The simulated tissue was stimulated along one edge, and a point stimulus was applied to induce an ectopic beat during a short-long-short sequence described in Supplemental Text S1. Figure 7 shows the results of simulations in four cases at indicated time points. In the absence of hormones or drugs, an initiated wave propagates in all directions, and no reentry occurs (first row). The same behavior is observed following drug application alone (E-4031) and with testosterone application alone (DHT 10 nM). However, when E2 (1 nM) is present (bottom row), the M-cell region is preferentially prolonged (due to the effect of E2 on the background of less repolarizing current that defines this region), which prevents the wavefront from crossing the refractory M-cell region. Instead, the wave propagates leftward at first - until the M-cell region repolarizes, and allows the wave to first cross the M-cell region and then slowly turn to the right. The slowly traveling wavefront (Na+ channels are only partially recovered following the prolonged action potential initiated by previous stimulus) begins a cycle of reentry – turning around and continuing to propagate on the wake of the preceding wave (Figure 7 – bottom panel).
In Figure 8A (top), the simulations suggest no reentrant activity during the late follicular phase of the menstrual cycle (progesterone 2.5 nM and 1 nM E2). However, when 10 nM E-4031 is applied during the late follicular phase, a spiral wave is readily induced (Figure 8A – middle). We also tested the effects of male hormone (testosterone) in the presence of E-4031. Figure 8A (bottom) shows that testosterone 3nM with 10 nM E-4031 did not trigger reentry activity.
Interestingly, the induction of a spiral wave in the presence of E-4031 during the late follicular phase of the menstrual cycle is a robust phenomenon. Reentry was introduced in this condition when the ectopic stimulus was applied in the subendocardial or subepicardial region – although not in the M-cell region (not shown). The position of the stimulus is also not critical. Figure 8B shows the effect of a point stimulus applied in the middle of endocardial tissue, leading to the initiation of a pair of counter-rotating spiral waves (Supplemental Figure S4 – protocol 2).
Here, we demonstrate the acute effects of sex steroid hormones in model cells and tissue, from physiological blood concentration to channel interaction, to their effects on APD and tissue dynamics. We used a computational approach to examine the role for acute application of sex steroid hormones on susceptibility to cardiac arrhythmias. The benefit of this approach is that it allows us to investigate the consequences of hormones on cardiac ion channels in isolation, so that observed changes can be specifically attributed to them.
We simulated the acute effects of sex steroid hormones on cardiac cell and tissue dynamics and on fluctuations of QT interval. It has been shown that progesterone enhances IKs, which counterbalances the IKr reduction by E2. Because estrogen and progesterone dominate in the different phases of menstrual cycle, simulations show that during the late follicular phase (prior to ovulation) of the menstrual cycle, QT interval is longer than in the luteal phase when progesterone is increased, which is consistent with the clinical observation by Nakagawa et al [27]. Notably, the fluctuations in QT interval during the menstrual cycle predicted by our model are within a relatively narrow range of 20 ms, which approximates the clinically assessed standard deviation in pooled QT intervals for a patient population assessed at each phase of the menstrual cycle [31]. One explanation is that such an analysis is unlikely to be sensitive enough to observe significant individual differences in QT intervals as they are fluctuating throughout the menstrual cycle, since biological variability between patients may be larger than fluctuations in individual patients.
Here we demonstrated that increasing testosterone reduced the APD and QT interval in a dose-dependent manner (Figure 2 and 4) by enhancing IKs and inhibiting ICa,L current. Moreover, differences in APD become more pronounced between E2 treatment and testosterone treatment when cycle length ≥800 ms - shown in Supplemental Figure S5. Taken all together, these results suggest that sex hormones influence cardiac repolarization in a dose- and cycle length-dependent manner. This is consistent with experimental studies of gender-related differences on cycle length-dependent QT [2],[37].
Since clinical findings suggest female gender is an independent risk factor for TdP arrhythmias and previous experimental studies have shown that E-4031 induced greater prolongation in E2-treated than in DHT-treated animals [2],[37],[40], we investigated the potential for E2 to exacerbate and testosterone to ameliorate arrhythmias in the presence of IKr block. We used simulations to probe these effects and ask if the presence of these two hormones at physiological concentrations play a key role on gender differences in drug-induced LQTS. We incorporated the experimentally measured combined effects of E2 and E-4031 on IKr and then simulated them on cardiac tissue dynamics. Since drug-induced TdP is often observed following a short-long-short pacing sequence in clinical studies [38],[41],[42], we explored tissue dynamics using such a protocol. Although we did not observe the development of arrhythmogenic EADs during the late follicular phase of the menstrual cycle (when E2 concentration is at its peak), addition of the IKr blocker E-4031 resulted in EAD formation in the late follicular phase for a wide range of pacing protocols (Figure 6).
The model simulations also suggested that E-4031 treatment in late follicular phase could lead to initiation of spiral wave reentrant arrhythmias (Figure 8). These simulations imply that at certain phases of the menstrual cycle, elevated levels of E2 may put females at risk for drug-induced arrhythmias – in particular by agents that bind to the promiscuous drug target hERG. Furthermore, we demonstrated that progesterone has a protective effect against E2-induced LQT syndrome (Figure 7 and 8). Spiral waves were not initiated in the presence of low concentrations of progesterone (2.5 nM – late follicular phase). This suggests that progesterone may play an important role in protecting against arrhythmia in females. Unlike the apparent pro-arrhythmic effects of E2 in the presence of Ikr block, testosterone was shown in our simulations to prevent E-4031 induced EADs during a short-long-short pacing protocol (Figure 6 and 8).
In the present study, our computational investigation demonstrates the acute effects of progesterone, estradiol and testosterone on cardiac ion channels that are critical for the rate of cardiac repolarization and resultant QT intervals. The models successfully simulates the effects of progesterone and testosterone on cardiac IKs and ICa,L. These two hormones hasten repolarization, albeit to different extents, and reduce QT interval and susceptibility to LQT-linked arrhythmias. On the contrary, E2 increased QT interval and propensity for TdP arrhythmias by reducing repolarizing current via IKr.
There are several limitations to this study that must be noted. First, we deliberately focused on the effects of acute application of physiological concentrations of sex steroid hormones here, but this means that we have neglected the effects of chronic hormone application. Several studies have suggested that chronic exposure to sex hormones may alter the response of the tissue to acute application of sex hormones [35], alter expression of ICaL (in a species dependent manner) [43]–[47] and may cause structural remodeling of the myocardium [45]. We ran simulations incorporating the measured differences by Verkerk [43] in human ICaL between males and females (since the goal was not to examine guinea pig sex differences). The results of these simulations are in Supplemental Figure S6. As expected, the additional Ca2+ current in the female prominently exacerbated the APD prolongation associated with E2. These effects may even be expected to increase in human females where primary repolarizing K+ currents, especially IKs, are apparently less prominent than in guinea pig [48]. However, progesterone effects on IKs in human may offset some of the observed increase in ICaL in human females compared to human males (described above). This issue should be addressed in future studies when the human data are more complete. The guinea pig model is also lacking transient outward K+ currents (Ito) – a subset of channels that have not yet shown to be affected by acute application of sex steroid hormones.
In summary, our study suggests that a computational approach to investigating effects of physiologically circulating hormones can be useful to test and predict their contribution to gender differences in cardiac arrhythmia susceptibility. Moreover, the findings from our model simulations suggest the potential utility of progesterone as a therapeutic agent for inherited and acquired forms of Long-QT Syndrome and that progestin-only contraceptives be given special consideration for their potential amelioration of LQT risk among pre-menopausal women. Finally, the link between estrogen containing hormone replacement therapy among post-menopausal women and increased incidence of adverse cardiac events needs to be investigated in the context of acute hormone effects on ion channels.
The guinea pig cardiac cell model was chosen in this study because we used the guinea pig ventricular myocytes experimental results reported in Ref. 11, 12, and 21. We modified the Faber-Rudy cardiac cell model [49]. The IKr channel was replaced with Markov model of wild type based on Clancy and Rudy so that E2 effects on activation gating could be readily incorporated. Although the level of detail in terms of description of gating by the IKr Markov model or the H-H IKr model is the same, the key difference is that the Markov model takes into account the property of coupling between discrete states, while H-H presumes independence between gating processes. This difference is only relevant in the setting of perturbation to one gating process – precisely what is observed when E2 is present, where activation gating is exhibits a small positive shift in voltage dependence [50]. All the simulation code was in C/C++ and run on Mac Pro 3.0 GHz 8-Core computers. The time step was set to 0.0005 ms during AP upstroke, otherwise the time step was 0.01 ms. Numerical results were visualized using MATLAB R2007a by The Math Works, Inc. Details of computational models and simulation protocols can be found in Supplemental Text S1.
|
10.1371/journal.pbio.1001330 | Species Interactions Alter Evolutionary Responses to a Novel Environment | Studies of evolutionary responses to novel environments typically consider single species or perhaps pairs of interacting species. However, all organisms co-occur with many other species, resulting in evolutionary dynamics that might not match those predicted using single species approaches. Recent theories predict that species interactions in diverse systems can influence how component species evolve in response to environmental change. In turn, evolution might have consequences for ecosystem functioning. We used experimental communities of five bacterial species to show that species interactions have a major impact on adaptation to a novel environment in the laboratory. Species in communities diverged in their use of resources compared with the same species in monocultures and evolved to use waste products generated by other species. This generally led to a trade-off between adaptation to the abiotic and biotic components of the environment, such that species evolving in communities had lower growth rates when assayed in the absence of other species. Based on growth assays and on nuclear magnetic resonance (NMR) spectroscopy of resource use, all species evolved more in communities than they did in monocultures. The evolutionary changes had significant repercussions for the functioning of these experimental ecosystems: communities reassembled from isolates that had evolved in polyculture were more productive than those reassembled from isolates that had evolved in monoculture. Our results show that the way in which species adapt to new environments depends critically on the biotic environment of co-occurring species. Moreover, predicting how functioning of complex ecosystems will respond to an environmental change requires knowing how species interactions will evolve.
| Understanding how species adapt to new environments is important both for evolutionary theory and for predicting and managing ecosystem responses to changing environments. However, most research into adaptation to new environments has considered species in isolation. Whether results from these systems apply to more realistically diverse groups of species remains unclear. We exposed five species of bacteria, collected from pools around the roots of beech trees, to a novel laboratory environment either in isolation or in species mixtures for approximately 70 generations. We found that each species evolved more in diverse species mixtures than it did when cultured in isolation. Moreover, species diverged in their use of resources and how they used the waste products of other species. These changes meant that the community of bacteria that evolved together used more of the available resources and were thereby more productive than the same group of species that evolved in isolation. Our findings show that species interactions can have a major effect on evolutionary dynamics, which can in turn influence ecosystem functioning.
| Understanding how species adapt to novel environments is an important task both for understanding the dynamics of living systems and for predicting biotic responses to anthropogenic changes in the natural environment [1]–[3]. However, most studies of evolutionary adaptation consider single species in isolation. Although this approach is useful for uncovering genetic mechanisms, virtually all species co-occur with many other species. Faced with a new abiotic environment, communities might respond by evolution of component species, but ecological changes in species' abundances and distributions can also occur. If ecological interactions such as competition affect evolutionary responses [4],[5], then results from single species studies might not accurately predict evolutionary dynamics in diverse assemblages.
Although there has been growing interest in how evolution affects ecological dynamics [6]–[10], most studies have still considered single species or pairs of interacting species. In addition, the question of how ecological interactions affect evolutionary responses to novel abiotic environments has received even less attention [11],[12]. If ecological interactions among species are weak, then evolutionary changes should be the same as those predicted in single species studies. However, if species use overlapping resources or otherwise interact, the extent and type of evolutionary responses might differ from those predicted if the same set of species each adapted to the new abiotic conditions in isolation [13],[14].
Several mechanisms might influence evolutionary dynamics in mixtures of species. First, species in diverse communities might have their resource use restricted by competitors, lowering effective population sizes and therefore reducing the rate at which beneficial mutations arise and the species adapts to a novel environment [5],[15]. In this scenario, species in communities should adapt to the new environment as they would in isolation, but the rate of adaptation would be reduced. Second, if trait variation among species exceeds variation within species, a new abiotic environment might act on the relative abundance of different species (ecological sorting) rather than on genetic variation within species [4],[5]. In models of this mechanism, pre-adapted species increase in abundance at the expense of less well-adapted species and the average amount of evolution in surviving species is typically reduced compared to responses of the same species in monoculture (although in rare scenarios the amount of evolution can increase [4]). Third, there might be a trade-off between adaptation to biotic and abiotic components of the environment [16]. Such trade-offs might result from the production of costly adaptations involved in species interactions such as defences [17],[18] or from selective interference between adaptations to the biotic and abiotic environment [19]. In this case, species that evolved in communities should be less well adapted to the abiotic environment than if they adapted in isolation and vice versa. In the most extreme case, species might adapt to use resources generated by other species [20], in which case they will evolve entirely different resource use depending on whether other species are present.
These mechanisms could change both the magnitude and direction of evolutionary change in communities compared to predictions from single species studies. However, evidence for an effect of diversity is currently scarce. Experiments have shown that diversity can inhibit evolution; for example, Brockhurst et al. [21] showed that niche occupation restricts adaptive radiation of a single bacterial strain. Similarly, Collins [16] found that diversity limits adaptation to elevated CO2 in algae and Perron et al. [22] showed that diversity limits the evolution of multi-drug resistance (although this effect was alleviated by horizontal transfer of resistance mutations). However, these studies considered genetic diversity within species rather than species diversity within communities. There is abundant evidence that coevolution drives fast evolution between species with strong ecological interactions [23],[24] and that pairwise coevolution can change the direction of evolution compared to adaptation in isolation [14]. Furthermore, studies of diffuse coevolution have shown that adaptation of a focal species to particular interacting species, such as insect herbivores, is influenced by interactions with other species, such as vertebrate herbivores [25]–[28]. For example, character displacement of limnetic and benthic species pairs of sticklebacks only occurred in lakes with low species diversity of other fish [29]. To the best of our knowledge, however, the evolution of interactions among multiple species in a community has not been investigated using an experimental evolution approach.
Evolutionary dynamics in diverse systems will have important consequences for ecosystem functioning in altered environments. Ecosystem functions such as decomposition and productivity emerge from the degree to which species are adapted to their biotic and abiotic environments [30]–[32]. Following a change in the environment, ecosystem functioning might be disrupted either because the species abundances change or because component species fail to adapt to the new environmental optimum. Alternatively, coevolution among species might act to enhance ecosystem properties, for example if species evolve complementary resource use and thereby increase ecosystem productivity [33]. Understanding of these processes is needed to predict how ecosystem functioning will respond to environmental changes over evolutionary timescales.
Here, we test whether species diversity influences environmental adaptation and ecosystem functioning using naturally co-occurring decomposer bacteria from temporary pools around the roots of beech trees (Fagus sylvatica), which have previously been used successfully for experimental ecology [34],[35]. We chose five species of bacteria differing in colony colour and shape so that each species could be isolated from species mixtures (Tables S1 and S2). Sequencing of 16S rDNA showed that isolates belong to five different families (Table S1). We refer to them as species since they represent genetically and phenotypically distinct clusters that co-occurred naturally. Monocultures of each species and polycultures containing all five species were allowed to adapt to laboratory conditions by regular serial transfer on beech-leaf extract (Figure 1). Laboratory conditions represent a new environment and differ from wild tree-holes in several ways: tree-holes receive a larger quantity and variety of resources, are spatially complex, and have an unpredictable input of water and leaves, whereas laboratory cultures experienced regular dilution with uniform medium in a shaken container. Growth assays were used to determine evolutionary responses. We predicted that species should adapt to laboratory conditions by evolving faster growth rates on the beech tea medium, but that the presence of other species might change the direction and extent of adaptation by one of the mechanisms outlined above.
To measure species interactions and changes in resource use, our approach was to grow one species on beech tea, then to filter-sterilize the medium and to assay the growth of a second species on the “used” beech tea. If the second species used similar resources to the first (i.e., if their niches overlapped), the second species should grow less well on “used” beech tea than on “unused” beech tea because its resources would have been consumed. If the two species were specialized on different resources (i.e., occupied different niches), the second species should grow equally well on “used” and “unused” tea. Finally, if the second species used resources produced by the first (called facilitation or cross-feeding [36]), the second species should grow better on “used” tea than on “unused” tea. While this method does not provide direct information on competitive interactions in mixtures, it provides a tractable and reproducible measure of changes in resource use of each species during evolution. Because other types of interaction, such as direct inhibition by bacteriocides [37], might also affect growth rates, we also used nuclear magnetic resonance (NMR) spectroscopic profiling of “used” and “unused” tea to investigate changes in resource use directly.
Finally, we tested whether adaptation to the presence of other species affected productivity (rate of production of CO2) by reassembling communities with different evolutionary histories using isolates that either evolved in monoculture or co-evolved in the same polyculture. If adaptation increased community productivity, we expected communities reassembled with isolates that evolved in polycultures to be more productive than those reassembled with isolates that had evolved in monoculture.
Although able to grow on beech tea in the lab at the start of the experiment, one species (E) dwindled to low cell densities during the evolution experiment (Table S3) and was excluded from growth assays and subsequent experiments because it failed to re-grow from frozen cultures. Species A to D were recoverable in all treatments and were used for subsequent experiments. Across species, final isolates that evolved in monoculture grew on average faster than ancestral isolates of the same species on unused beech tea (dark bars, first and second rows, Figure 2), consistent with the prediction that they adapted to laboratory conditions of serial dilution in beech tea medium by increasing growth rates on this medium. The effect was significant in species B, C and D, which grew between 47% and 120% faster after evolving in monoculture compared to their ancestral isolates. Growth rates of evolved monoculture isolates of species A were not significantly different from its ancestral isolate. Note that phenotypic plasticity and parental effects can be discounted as explanations for differences among treatments. In all our assays, frozen isolates were first grown in beech tea medium for 4 d ( = 4 to 6 generations, Table S3), and then an aliquot was taken from these cultures to start the assay cultures. Differences in phenotypes between treatments were therefore maintained after several generations of growth in identical environments and cannot be readily explained by phenotypically plastic responses.
Isolates of species A, B, and C that evolved in polyculture grew significantly slower on unused beech tea than their corresponding ancestral isolates and than the isolates that evolved in monoculture (Figure 2). Growth rates were 87% to 100% slower after evolving in polyculture compared to the corresponding ancestral isolates. This is consistent with the existence of a trade-off between adaptation in the presence of other species and adaptation in the absence of other species; when evolving in the presence of other species, isolates of A, B, and C nearly lost the ability to grow on unused beech tea. In contrast, the polyculture isolate of species D grew significantly faster on unused beech tea than either ancestral or monoculture isolates. By adapting in the presence of the other species, species D evolved to grow at a faster rate on beech tea when assayed with other species absent. This result is not readily predicted by the general theories outlined in the introduction and is discussed further below.
Reduced growth of ancestral isolates on beech tea previously used by other ancestral isolates showed that species had generally negative interactions (Figures 3A, S1), as predicted if species used overlapping resources. The exception was species D, whose growth was not reduced on tea previously used by other species even though tea used by species D reduced the growth of other species (arrows towards species D on Figure 3A). This result might indicate that species D used a greater range of resources than the other species, but which included the resources used by the other species. Growth of monoculture isolates on tea previously used by monoculture isolates of other species showed that negative interactions among species were reinforced: now species D also grew significantly slower on tea previously used by other species (Figure 3B). These results would be expected if species in monoculture converged to use a more similar set of resources.
Species interactions evolved to be more positive between polyculture isolates than between ancestral or monoculture isolates (Figure 3C). Species B and C evolved in polyculture to grow significantly faster on tea previously used by other species than on unused beech tea (Figures 3C and S1). Thus, interactions shifted to facilitation as predicted if species adapted to use resources being produced by other species as waste products of metabolism. Polyculture isolates of D remained negatively affected on substrate used by other species, although less so than their monoculture isolates (Figure S1, relative growth rate on used tea versus unused tea: in monoculture, 0.24±0.05, and in polyculture, 0.77±0.07), indicating that species D also adapted to the presence of other species. Polyculture isolates of species A grew poorly on all substrates (Figure S1), but again the interactions were significantly less negative than between ancestral and between monoculture isolates.
Forty-three separate resonances (i.e., peaks) were distinguished and integrated from the NMR spectra (Figures S2 and S3). Variation in the net use and production of peaks in the spectra across ancestral, monoculture, and polyculture isolates of each species confirmed that resource use evolved in each of the species in ways that matched findings from the growth assays (Figures 4, S2, and S3). Considering the multivariate space of resource use and production across all compounds, polyculture isolates displayed greater differences from ancestral isolates than did monoculture isolates (across species, mean and standard error of Euclidean distance between paired ancestral isolates and monoculture isolates = 1.20±0.26; mean and standard error of distance between ancestral isolates and polyculture isolates = 2.42±0.37, p = 0.003, Monte Carlo simulation). Moreover, although species evolved, if anything, to have marginally more similar resource use in monoculture (not significantly so, p = 0.36, Monte Carlo simulation), patterns of resource use and production diverged significantly between species in polycultures (p = 0.010, Monte Carlo simulation; mean and standard error of Euclidean distance between species: ancestral isolates = 2.27±0.01; monoculture isolates = 1.98±0.01; polyculture isolates = 3.41±0.01). Together these results show that species' use of NMR-visible carbon substrates in the beech tea evolved more in polyculture treatments than in monoculture treatments and did so in a way to increase the differences in metabolism between species and thereby to reduce negative interactions between them.
Principal components analysis identified the main axes of variation in net use or production of these compounds across ancestral, monoculture, and polyculture isolates of each species (Figure 4). The first principal component distinguished isolates based on the degree to which they used glucose, choline, formate, and succinate to produce pyruvate (Figure S4). The second principal component distinguished isolates based on whether they used up or produced acetate, formate, and lactate. Notable changes in polyculture isolates were as follows: species A evolved to produce 96% more acetate and to produce rather than use formate; species B evolved to use up to 84% more choline, formate, and lactate and to use rather than produce succinate; species C evolved to use rather than produce acetate; and species D evolved to produce rather than to use lactate and to use rather than produce acetate (Figure S3).
These observed changes indicate possible cases of cross-feeding evolving in polycultures, which might explain the positive interactions observed in growth assays. For example, species D evolved to produce lactate in polycultures and species B to use it. To test whether species generally evolved increased use of other species' waste products in polycultures, we quantified the amounts of substrates produced by each species grown on beech tea and the amounts of the same substrates that were used by a subsequent species grown on the “used” beech tea (Figure S5). On average across species, polyculture isolates displayed significantly increased use of substrates (i.e., a more negative change in the amounts of the substrate on the y-axis of Figure 5) that were produced in increased amounts by other species (a more positive change in the amount of substrates on the x-axis of Figure 5, Pearson's correlation, r = −0.74, p<0.0001), relative to ancestral isolates. Moreover, although monoculture isolates were also able to use waste products generated by other monoculture isolates, the correlation between increased production and increased use (relative to ancestral isolates) was significantly weaker (Pearson's correlation r = −0.20, p = 0.03; significant interaction between slope and treatment, linear model results in Figure 5). Polyculture isolates therefore appear to have evolved greater use of waste products generated by polyculture isolates of other species.
Communities were reassembled to contain one isolate of each of the four surviving species. Communities reassembled using isolates that evolved in polycultures displayed significantly higher productivity, measured as CO2 production rate, than communities reassembled using isolates that evolved in monoculture (Figure 6). Adaptation to the biotic environment of co-occurring species therefore increased community productivity.
Our results show that species interactions had a major impact on how species adapted to the new environment in the laboratory. In all four surviving species, the magnitude of evolution in terms of changes in growth rate on beech tea medium and changes in use of NMR-visible resources was significantly greater in polycultures than in monocultures. Moreover, species diverged in resource use in polycultures compared to monocultures and ancestral isolates. This provides experimental evidence for a classic scenario of character displacement reducing the overlap of resources used by interacting species [38]. Furthermore, not only were negative interactions reduced, but species also adapted to use waste products of other species in polycultures, leading to positive interactions between some pairs of species. Together, these changes led to increased productivity of the entire community. By evolving to use different resources, and to metabolise waste products of other species, the species collectively decomposed substrates in the beech tea more effectively. Similar results have been observed for cross-feeding ecotypes evolving during monoculture experiments [33],[39]; here we show that cross-feeding also evolves readily between distantly related species of bacteria.
The effect of species interactions on evolution varied among species. In three species, A, B, and C, there was a trade-off between adaptation to the laboratory environment in the presence of other species and adaptation in the absence of other species: polyculture isolates grew less well when assayed in isolation than did monoculture isolates. In species B and C this occurred because they adapted to use waste products generated by other species, which was demonstrated both by their increased growth on medium previously used by other species and by their increased use of waste products from other species. In species A it occurred because this species changed to use different carbon sources than the other species: its interactions became less negative in the polyculture treatment than between ancestral or monoculture isolates (but not positive) and it used more trehalose and less glucose and lactate (Figure S3).
In contrast, species D displayed a positive effect of diversity on its adaptation to abiotic conditions: the polyculture isolate had enhanced growth rate when assayed on its own compared to either the ancestral or monoculture isolates. There is no evidence that species D polyculture isolates evolved to use any of the NMR-visible resources more effectively than any other isolates. We therefore hypothesize that polyculture isolates of species D evolved increased use of complex carbon sources that cannot be distinguished by NMR. One clue supporting this hypothesis is that polyculture D produced large amounts of lactate and was the only isolate to do so and without correlated negative change in any other compound. We suggest therefore that species D could be producing lactate from metabolism of compounds not distinguishable by NMR—for example, macromolecular structures such as mixtures of proteins. None of the general theories outlined in the introduction readily explain why species D should enhance its ability to grow on its own after evolving in polyculture. However, in rare circumstances in the niche simulation model by de Mazancourt et al. [4], competition among species could “push” one species to evolve into a wider range of niches than it would do so when in the absence of competitors. The observation that species D has shifted away from its ancestral and monoculture isolates in resource use and away from the polyculture isolates of other species is consistent with this possibility (Figure 4).
Despite differences in response among the species, in all cases the effects of diversity arose because co-adaptation between species altered their ability to grow in an environment free of other species. The other mechanisms outlined in the introduction cannot explain our results. Effective population sizes were generally lower in polycultures (Table S3), but still exceeded 106 in all surviving species, and polyculture isolates did not adapt more slowly than monocultures. Instead, co-adaptation with other species rendered species A, B, and C even less well adapted to the abiotic environment in the absence of other species than their ancestors, and species D better adapted. Similarly, our results do not reflect the damping of evolutionary responses by ecological sorting, because species' use of NMR-visible compounds in fact evolved more in polycultures than in monocultures. Species E might have dwindled to low numbers in polycultures because of one of these two mechanisms (Table S3), but in any case it failed to sustain large populations during the experiment even in monocultures.
The NMR results show that changes in resource use can explain observed changes in interactions and productivity (see also [40]). It remains possible that other interactions could be operating among these species as well, but which remained undetected by our assays. Some of the metabolites generated by species could have had toxic effects on other isolates, and some of the observed metabolic changes could have been to reduce toxic effects rather than increase resource use. Also, bacteria are known to produce signalling molecules that can have inter-specific effects—for example, antimicrobial properties [37] or positive effects such as stimulating enzyme production [41]. In principle, these could have caused some of the changes in growth rates we observed in interaction assays and they would be interesting traits to investigate in future studies. However, typical signalling molecules such as quorum sensing compounds are usually not produced at high enough concentrations for detection by NMR [42], and therefore the changes observed here reflect changes in resource use rather than changes in signalling. Because the NMR results match inferences from the growth assay results, it is more parsimonious to conclude that changing resource use is the dominant mechanism explaining our findings.
Our results provide among the first experimental evidence supporting recent theories that species interactions in diverse communities affect evolutionary responses to an environmental change. The way in which species adapted to new conditions in the laboratory when in monoculture—the setting assumed for many evolutionary theories and experiments—provided little information on the outcome of evolution in the diverse community. Co-occurring species modified the environment by generating new resources, and thereby altered the selection pressures on other species and how they used the available resources. Other experiments have reported that genetic diversity inhibited adaptation to the environment [16],[21] but have not investigated whether adaptation to the biotic environment of co-occurring species changed how species adapt to a new abiotic environment. If the processes we observed here are common in other communities, including multicellular eukaryotes over longer timescales, then attempts to understand evolutionary dynamics in the wild must take into account the biotic environment of co-occurring species [13],[43].
As well as being important for understanding evolutionary dynamics, our experiments show that evolutionary interactions had important consequences for ecosystem-level functions. Co-adaptation for approximately 70 generations—not an unrealistic timescale for responses of annual eukaryotic organisms to predicted changes over the next hundred years—acted to enhance community productivity, through the evolution of complementary use of resources. Niche complementarity and facilitation are known to be important determinants of community productivity [44],[45], and our results add to growing evidence from microbial systems that niche evolution can exert a strong influence on productivity [10],[46]. Recent work has shown that biofilms derived from a single clone of Burkholderia cenocepacia evolved cross-feeding morphotypes that together had enhanced productivity compared to the morphotypes grown alone [33]: our study demonstrates similar processes operating between phylogenetically distinct species. It remains to be determined whether adaptation generally acts to enhance ecosystem productivity [47],[48], but if so, it will be an important process to consider in predicting the impacts of current environmental changes on ecosystem services. Ecosystem functions such as decomposition rate might be reduced by local extinction of species providing important functions, but it is important to know whether evolution of surviving species will restore (as found here) or further disrupt those functions.
Our communities were far less diverse and far simpler than natural communities. A single tree-hole likely contains thousands of bacterial species, including anaerobes and many other functional groups excluded by our isolation protocol. Even the comparatively depauperate community of multicellular eukaryotes in tree-holes would typically contain many more than four or five species [49]. A major goal for future research is to understand whether our findings scale to natural ecosystems and how other ecological mechanisms such as predation affect evolutionary outcomes in diverse communities. Strong interactions have been demonstrated between bacteria and their phages in natural settings [50], but reciprocal co-adaptation between bacterial species might be rare compared to adapting to the general biotic environment because of the large number of potential pairwise interactions among species [51]. Another important process in natural communities is immigration, which can add variants (new genotypes or species) that might swamp evolutionary responses [52]. Understanding how natural assemblages respond to new environments, such as those caused by global warming, ocean acidification, or pollution, depends critically on understanding the balance between ecological and evolutionary responses of the kind we demonstrate here.
Bacteria were isolated from single colonies from temporary pools formed by the roots of a beech tree at Silwood Park, Berkshire, United Kingdom, in November 2008 (Text S1). BLAST and Ribosomal Database Project [53] matches and photographs of colonies of each species are provided in Tables S1 and S2. Species A and E belong to families Sphingobacteriaceae and Flavobacteriaceae, respectively (both in the phylum Bacteroidetes); species B and C belong to families Enterobacteriaceae and Pseudomonadaceae, respectively (both in the class Gammaproteobacteria of the phylum Proteobacteria); and species D belongs to the family Sphingomonadaceae (in the class Alphaproteobacteria of the phylum Proteobacteria). Note that our isolation protocol means that all our bacteria are expected to be aerobic heterotrophs. Isolates were grown on beech-leaf tea prepared by autoclaving 50 g of autumn fall beech leaves in 500 ml of water and diluting the filtrate 32-fold [34].
Fifteen replicates of each species in monoculture and of each five-species community were set up following the protocol in Figure 1 and Text S1. The tubes were incubated at 25°C and shaken at 100 rpm. Every 3 and 4 d, 100 µl from each microcosm was transferred to 2 ml of fresh media for a total of 15 serial dilutions over 8 wk. Cell densities prior to transfer were estimated by colony counts on R2A agar. Bacteria were isolated from final cultures by plating on R2A agar, selecting single colonies, and re-suspending them in 1 ml of 1/32× beech tea. Isolates were stored at −84°C for use in subsequent assays.
Growth assays were performed in 1 ml of 32× beech tea in 24-well plates inoculated with 250 µl of bacteria from a liquid culture grown up for 4 d from stored frozen isolates. The plates were kept at 25°C for 4 d without shaking and growth measured daily using OD600. Readings were subtracted from negative controls of sterile medium placed on each column of the plate. Nine replicates were used for each Species×Treatment combination. “Used” beech tea was prepared by inoculating 14 ml of beech tea with 200 µl of single bacterial species and allowing growth at 25°C for 14 d. The first and second isolate used for each assay always belonged to the same treatment—that is, both ancestral, both monoculture, or both polyculture isolates. Substrates were then filter sterilized using a 0.2 µm membrane to remove bacterial cells and leave any unused nutrients in the substrate. Sterility was confirmed by plating on agar. Growth was measured as described for growth assays on unused beech tea for nine replicates of each Species×Substrate×Treatment combination.
Samples of unused beech tea, tea used previously by one isolate, and tea used previously by one isolate and then a second isolate (as described in the previous section) were analysed using proton NMR. Because of the low concentration of carbon substrates in the beech tea, 5 ml of each sample were lyophilized and resuspended in 550 µl 90% 2H2O (superscript numbers are atomic weights; i.e., 1H2O is “normal” water and 2H2O is deuterated) containing 1 mmol l−1 3-(trimethylsilyl)propane-1-sulfonic acid (DSS) and, 5 mmol l−1 sodium azide. The 2H2O provided a field frequency lock for the spectrometer and the DSS served as an internal chemical shift reference. Spectra were acquired on a Bruker 800 US2 NMR spectrometer (Bruker BioSpin), with a magnetic field strength of 18.8 T and resulting 1H resonance frequency of 800 MHz, equipped with a 5-mm cryogenic probe. Spectra were acquired following the approach given in [54]. Briefly, a one-dimensional NOESY pulse sequence was used for water suppression; data were acquired into 64 k data points over a spectral width of 12 kHz, with eight dummy scans and 256 scans per sample. Spectra were phased in iNMR 3.6 (Mestrelab) and exported to Matlab 2010b (Mathworks) for further analysis. Distinct peaks were integrated and baseline-corrected using in-house scripts and assigned where possible using in-house databases. One resonance with a singlet at chemical shift δ = 3.22 ppm was assigned as choline; a COSY spectrum of the unused medium showed a cross-peak at δ 4.05/3.52 ppm, as would be expected for the methylene protons of choline (although the resonances were too low intensity to be visible in the 1D spectra). To measure resource use or production, we calculated the size of each peak in medium obtained after the growth of an isolate minus the size of the peak in the medium before the species had grown on it. Positive values indicate net production of a compound and negative values indicate net consumption. We used correlation tests to identify correlated peaks with r>0.95, which might indicate multiple peaks derived from the same compound. Contaminant peaks derived from methanol and acetonitrile were removed from the dataset. Variation in resource use and production across isolates was explored using principal components analysis of unscaled variances implemented with the prcomp() function in R [55]: we used unscaled rather than scaled variances to focus on compounds showing larger changes in their absolute concentrations.
MicroResp kits were used to measure community respiration. Respired CO2 results in a change in colour of cresol red indicator dye suspended above each well of a 96-well plate. Ten replicates were used per treatment in a single plate and the experiment was repeated in triplicate. Each well contained 840 µl of 1/32× beech tea and 40 µl of each species from a stock culture of standard density. The plate was sealed and the change in optical density (OD) at 570 nm of the indicator gel measured after 6 h as recommended by the manufacturers [56]. The change in OD of blank wells (filled with 1 ml 1/32× beech tea) was used to account for the base level of CO2 in the vials. The rate of CO2 respiration per ml of culture medium was calculated using the formula provided in the MicroResp manual [56].
To calibrate OD600 in terms of cell density per ml of culture medium [57], we performed serial dilution and colony counts of stock cultures of isolates of each species from each treatment. We fitted a linear model with log (colony count)/ml as the response variable and species, treatment, and OD600 as explanatory variables, including interaction terms. The model simplified to retain species and OD600, but no interaction terms (i.e., different intercept for calibration line for each species, but same slopes, F4,67 = 32.9, p<0.0001, r2 = 0.64, Figure S8). The fitted lines were used to calibrate in units of log(number of cells) per ml. We used linear mixed effects models of repeated measures of cell density over time to compare growth of bacteria among treatments and species in the growth assays (Text S1). To report the direction and effect size of differences among treatments, we used the rate of change in density over the first 48 h as a simple measure of VMAX—that is, the maximum rate of growth from low densities (Figure S6). Analysis of variance (ANOVA) and Tukey's Honest Significant Difference tests were used to identify significant contrasts between particular treatments of interest. There was no evidence of different evolutionary trends in carrying capacity of isolates (i.e., using density at 96 h) as opposed to growth rate (Figure S1 versus Figure S7). To test for significant differences in NMR profiles between treatments, we used Monte Carlo simulation tests shuffling profiles randomly among species and treatments. The Euclidean distance between samples was recorded, and the mean distance between both evolved treatments in turn and ancestral isolates was used to measure the amount of evolution, and the mean distance between each species within a treatment was used to measure the amount of divergence in resource use among species. Observed values were compared to randomised values from 10,000 random permutations. Two-tailed tests were used.
|
10.1371/journal.pgen.1008400 | Evolution of the Auxin Response Factors from charophyte ancestors | Auxin is a major developmental regulator in plants and the acquisition of a transcriptional response to auxin likely contributed to developmental innovations at the time of water-to-land transition. Auxin Response Factors (ARFs) Transcription Factors (TFs) that mediate auxin-dependent transcriptional changes are divided into A, B and C evolutive classes in land plants. The origin and nature of the first ARF proteins in algae is still debated. Here, we identify the most ‘ancient’ ARF homologue to date in the early divergent charophyte algae Chlorokybus atmophyticus, CaARF. Structural modelling combined with biochemical studies showed that CaARF already shares many features with modern ARFs: it is capable of oligomerization, interacts with the TOPLESS co-repressor and specifically binds Auxin Response Elements as dimer. In addition, CaARF possesses a DNA-binding specificity that differs from class A and B ARFs and that was maintained in class C ARF along plants evolution. Phylogenetic evidence together with CaARF biochemical properties indicate that the different classes of ARFs likely arose from an ancestral proto-ARF protein with class C-like features. The foundation of auxin signalling would have thus happened from a pre-existing hormone-independent transcriptional regulation together with the emergence of a functional hormone perception complex.
| Plants transition from water to land was determining for the history of our planet, since it led to atmospheric and soil condition changes that promoted the appearance of other life forms. This transition initiated around 1 billion years ago from a Charophyte algae lineage that acquired features allowing it to adapt to the very different terrestrial conditions. Land plants coordinate their development with external stimuli through signalling mechanisms triggered by plant hormones. Therefore, evolution of these molecules and their signalling pathways likely played an important role in the aquatic to terrestrial move. In this manuscript we study the origin of auxin signalling, a plant hormone implicated in all plant developmental steps. Our studies suggest that out of the three families of proteins originally proposed to trigger auxin signalling in land plants, only one existed in Charophyte ancestors as a likely transcriptional repressor independent of auxin. We show that despite millions of years of evolution, this family of proteins has conserved its biochemical and structural properties that are found today in land plants. The results presented here provide an insight on how hormone signalling pathways could have evolved by co-opting a pre-existing hormone-independent transcriptional regulatory mechanism.
| Charophytes diverged from chlorophyte algae more than a billion years ago (y.a.) and led to land plants emergence around 450 million y.a. [2–6]. “Early divergent” clades display a range of body complexity going from unicellular algae in Mesostigmatophyceae and Chlorokybophyceae, to multicellular filaments in Klebsormidiophyceae (Fig 1; S1 Fig) [7,8]. “Late divergent” clades include Charophyceae and Coleochaetophyceae that share features with land plants (S1 Fig), [9,10] but also Zygnematophyceae, that despite their simple structure are considered sisters to land plants according to recent phylogenetic studies [11,12].
Given the importance of the phytohormone auxin in plant development, the acquisition of its signalling pathway allowing for auxin-dependent changes in transcription is thought to have been a milestone in the water-to-land transition [2]. In land plants, this signalling pathway, called the Nuclear Auxin Pathway (NAP), relies on three main protein families: TIR1/AFB (Transport Inhibitor Response 1/Auxin Signalling F-box) co-receptors, Aux/IAA transcriptional repressors (Auxin/Indole-3-Acetic Acid Protein) and ARF (Auxin Response Factors) Transcription Factors (TFs) [13,14]. ARFs have been classified into three evolutive classes, A, B and C. Class A includes activator ARFs whereas classes B and C contain repressor ARFs [1]. ARF interaction with DNA is mediated by their B3 domain (B3ARF). Such domain is also present in ABI3 (Abscisic Acid insensitive 3, B3ABI3) and RAV (Related to ABI/VP1, B3RAV) plant TFs but with different DNA binding specificities [15,16]. In the ARF family, the B3 domain is embedded in a larger N-terminal DNA Binding Domain (DBD) that includes a Dimerization Domain (DD). As dimers, ARFs bind double AuxREs (Auxin Response Elements) sites arranged in three possible orientations: Direct Repeat (DR), Everted Repeat (ER) and Inverted Repeat (IR) (S2 Fig) [2,17–19].
In charophyte algae and the bryophyte Marchantia polymorpha, the B3RAV and B3ARF domains are often associated with a C-terminal PB1 oligomerization domain, a landmark of most ARF TFs in higher plants but that was lost by RAV TFs from tracheophytes [2,20]. This shared B3 + PB1 domain composition led to the initial hypothesis that ARF could have arisen from RAV [21]. In the NAP, the PB1 domain mediates ARF homo-oligomerization and hetero-oligomerization with Aux/IAAs [22]. Under low auxin concentrations, Aux/IAAs bind activator ARFs through their PB1 domain [23–26] and recruit TOPLESS (TPL)/TOPLESS-RELATED co-repressors, leading to the formation of a repressor complex on regulatory sequences of auxin-responsive genes [27]. When auxin levels increase, the hormone-mediated interaction between Aux/IAA and TIR1/AFB leads to Aux/IAA proteasomal degradation, unlocking activator ARFs and inducing transcription [28,29]. Most class B and C ARF members have limited interaction capacities with Aux/IAAs [30–32] and are proposed to regulate auxin transcriptional responses in an auxin-independent manner, possibly by competitive binding with class A ARFs on DNA regulatory sequences [33,34]. Since some of class B and C ARFs can interact directly with TPL, formation of co-repressor complexes was proposed as another possible mechanism for transcriptional repression of auxin target genes [34–36].
The NAP was established at the beginning of land plants history. In the bryophyte M. polymorpha for example, the 3 families of NAP proteins are present (with one member of each ARF class) as well as the TPL co-repressor [37,38]. Recent studies showed the existence of two ARF subfamilies in charophytes, class C and class A/B [20], but the absence of functional TIR1/AFB and Aux/IAAs suggested that a fully functional NAP did not exist before land plants [2,20,37,39–41]. How these ancestral components evolved to form the land plants NAP remains an open question. Through the structural, biochemical and phylogenetic characterisation of a proto-ARF from an early divergent charophyte we set a scenario of how the co-option of ancestral mechanisms of transcriptional control possibly led to the evolution of hormone signalling pathways in plants.
To understand the evolution of ARFs, we first characterized the biochemical properties of proto-ARFs and closely related proto-RAVs from early divergent charophytes. We searched for B3 homologues in charophyte transcripts databases (OneKp and Marchantia.info) [12,44] and classified them as B3RAV or B3ARF, depending on the residues signature of their predicted DBDs (S1 Table) [45]. B3RAV domains were frequently associated with an APETALA2 (AP2) domain and/or PB1 domains in the basal charophyte Mesostigma viride and all later clades (Fig 1; S2 Table). M. viride also has an ARF homologue (GBSK01006108.1) devoid of a PB1 domain [2]. Its DBD was reliably modelled as an ARF (100% confidence with AtARF1 [46,47]), but it lacks most residues involved in the interaction with AuxREs (S3 Fig) and thus does not qualify as a functional ARF. The proto-ARF of the earliest diverging green Charophyte algae with predicted functional B3ARF and PB1 domains was found in C. atmophyticus. Other ARF homologues were also present in all later diverging clades (Fig 1; S3 Fig).
We determined the properties of “ancestral” RAV and ARF proteins, focusing on K. nitens proto-RAV (containing predicted AP2, B3RAV and PB1) (KnRAV, kfl00094_0070) and C. atmophyticus proto-ARF (CaARF, AZZW-2021616). The predicted B3 domains of KnRAV and CaARF display the signature residues typical of B3RAV and B3ARF, respectively (S1 Table; S3 and S4 Figs) suggesting that their divergent DNA binding specificities were already established in charophytes. To test this hypothesis, we characterized the binding of their DBD against the canonical DNA binding sites identified in angiosperms for ABI3, RAV and ARF TFs. KnRAV specifically bound the AP2/B3RAV bipartite element described for Arabidopsis thaliana RAV TFs (Fig 2A) [48]. CaARF interacted strongly with double AuxRE sites (DR or ER, Fig 2B) but not with a single AuxRE site suggesting that the DBD of CaARF binds DR and ER motifs as a dimer without the help of the Middle Region (MR) and the PB1 domain. Altogether, these results confirm that RAV and ARF DNA binding preferences were established in basal charophytes and maintained along evolution.
Next, we studied the oligomerization capacity of their PB1 domain. Based on AtARF5 PB1 structure [23,47], the PB1 domains of KnRAV and CaARF were modelled as type I/II PB1 with electrostatic oligomerization potential (Fig 2C and 2D). Molecular weight determination of KnRAV-PB1 and CaARF-PB1 by Size Exclusion Chromatography combined with Multi-Angle-Light Scattering (SEC-MALLS) experimentally validated that both domains form oligomeric complexes (Fig 2E and 2F) but with a lower oligomeric potential than AtARF5-PB1 (S3 Table). Charophycean algae therefore appear to possess proto-RAV and proto-ARF proteins with oligomerization potential and diverging DNA binding specificities (Fig 2A and 2B; S3 and S4 Figs).
To further characterize the biophysical properties of proto-ARFs, we determined the predicted structure of CaARF DBD and showed that it was reliably modeled (99% confidence; Phyre 2) with AtARF1 and AtARF5 DBDs [46,47] except for an additional disordered region in CaARF present within the DD (Fig 3A). Similar disordered regions were found as a characteristic feature of all class C ARFs (Fig 3B and 3C; S3 and S7 Figs). In agreement with this, our phylogenetic studies position CaARF within clade C (S5 Fig). Such insertions are expected to modify class C DNA binding compared to A and B ARFs. We tested this hypothesis using ER motifs with different spacing (ER4-9). Unlike Arabidopsis AtARF2 (class B) and AtARF5 (class A) that largely prefer ER7/8 motifs (Fig 3D and 3E), CaARF showed promiscuous binding to ER4-9 but did not interact with a single AuxRE motif (Fig 3D–3F) confirming its interaction with ER motifs as a dimer. Arabidopsis class C AtARF10 behaves similarly to CaARF (Fig 3G). This shows that CaARF has a relaxed DNA specificity allowing binding to ER binding sites with various distances between the monomeric sites and that this specificity was maintained in class C ARF along plants evolution. The presence of a specific disordered region (Fig 3A–3C; S3 and S7 Figs) in class C ARF DBDs suggests a possible role in their relaxed specificity, that remains to be tested.
As mentioned before, certain land plants ARF proteins have the capacity to interact directly or indirectly with the TPL co-repressor [35,36,50]. We wondered when in evolution this interaction was first established. Direct TPL-recruitment usually involves two different amino acid regions in the Middle Region (MR) of repressor ARFs: the EAR-motif (ERF-associated Amphiphilic Repression motif with LxLxL sequence or its LxLxPP variant) and the BRD domain (B3 Repression Domain with the K/RLFG sequence) [35,36], the BRD domain also being found in RAV proteins. CaARF-MR presents two potential repression regions with an EAR-like motif (LPLLPS, similar to LxLxPP) and a BRD domain (KLFG). Since TPL EAR-interacting-region (TPL N-terminal, TPL-N) is extremely conserved between charophytes and land plants [49,51] (Fig 3H; S8 Fig; S4 Table), we used A. thaliana TPL-N (AtTPL202) to assay the TPL/CaARF interaction. CaARF interacted with AtTPL202 in co-purification assays and this interaction was lost with AtTPL202-F74A, mutated in the hydrophobic EAR peptide binding groove (Fig 3I) [49]. Moreover, mutations in CaARF KLFG (CaARF-L523S/F524S) or LPLLPS (CaARF-L492A/L493A) weakened the interaction with AtTPL202, indicating that both sites might participate to TPL-N recruitment. The binding of the BRD domain of CaARF differs from that of the RAV1 of A. thaliana which interacts with the C-terminal part of TPL [52], suggesting different TPL recruitment mechanisms for these two protein families. The presence of similar TPL-recruitment sequences in proto-ARFs of different charophytes clades ARFs (S5 Table) suggests that they might also interact with TPL.
The present biochemical characterization of CaARF, a proto-ARF from an “early divergent” charophyte, identifies this protein as class C ARF, in agreement with our phylogenetic analyses (S5 Fig). Mutte et al. (2018) proposed the existence of two ARF classes in “late divergent” charophytes, C and A/B, deriving from a common ancestor that diverged in an ancient charophyte clade [20]. Based on phylogenetic analyses showing that class C ARF is sister to classes A and B, and on the identification of a M. viride sequence classified as a class A/B, Flores-Sandoval et al. (2018) proposed a similar scenario where the divergence between classes A and B and class C occurred prior to the diversification of extant streptophytes [40]. This plausible scenario, built before the identification of class C ARFs in “early divergent” charophytes, is based on an unusual M. viride sequence that does not exhibit the conserved ARF DNA binding residues (S3 Fig), and implies repeated loss of class A/B ARFs from Chlorokybophyceae to Coleochaetophyceae (S6 Fig). Further identification of class C ARFs in the “early divergent” charophytes (Klebsormidiophyceae [2] and Chlorokybophyceae (this work)) and the presence of both classes C and A/B in the “late divergent” C. orbicularis suggest a second and more parsimonious scenario in which class A/B ARF members come from an ancestral proto-ARF, belonging to class C or class C-like that existed before the emergence of “late divergent” charophytes (S6 Fig). This hypothesis implies only a few class C ARF gene losses in some Klebsormidiophyceae, Coleochaetophyceae and Zygnematophyceae species. Still, all these scenarios need to be taken with caution as they are based on transcriptomic datasets and could be challenged when genomic sequences become available.
When comparing C and A/B clades we found a disordered region within the predicted DD of ancestral and land plants clade C ARFs that is not present in clade A/B neither in land plants clades A and B. We speculate that during the duplication event leading to A/B emergence from clade C, the loss of this disordered sequence occurred. The DNA interaction experiments presented in this manuscript suggest that this event might have contributed to the acquisition of a more restricted DNA specificity of class A and B ARFs for ER motifs.
Apart from the similar behaviour observed for CaARF and AtARF10 when binding to DNA, ancestral clade C ARFs already presented PB1 oligomerization potential and interaction with the co-repressor TPL. The conservation of these properties along evolution is consistent with experiments conducted on Marchantia showing partial complementation of the loss of function MpARF3 by class C AtARF10 [40]. Moreover, these biochemical facts are instructive on several aspects of the evolution of the NAP in plants. First, proto-ARFs being able to interact with AuxREs supports that the NAP could have co-opted sets of genes already regulated by ARFs in charophytes, as suggested in other studies [2,20,39,40]. In this context, the emergence of the A/B clade with a different DNA binding behaviour could have allowed to target a more specific set of genes. Second, proto-ARF interaction with TPL provides functional evidence for a role for class C ARFs as transcriptional repressors. Putative TPL interaction motifs are also present in proto-RAV and most proto-ARFs across charophytes, which includes class A/B ARFs. The capacity to recruit TPL co-repressors could thus be an ancestral property of RAV and ARF TFs.
From these observations, we propose ARFs recruitment of co-repressor complexes to AuxREs promoter elements as a primitive and conserved mechanism predating the NAP. The absence of a functional TIR/AFB-Aux/IAA co-receptor [2,20,41] indicates that this primitive system was auxin-independent. These observations are consistent with a series of experiments in Marchantia showing that auxin-responsive genes show similar transcriptional responses in WT and MpARF3 mutants [20,40]. Alongside the diversification of ARF DNA binding specificity, emergence of the auxin perception complex in the first land plants turned ARFs-regulated genes into auxin-responsive genes through ARF-Aux/IAA-TIR/AFB interactions evolution (Fig 4).
Our work thus allowed proposing a scenario where the evolution of the binding specificity of an ancestral TF together with the emergence of a functional hormone perception complex create a hormone signalling pathway. This scenario offers a better understanding of how hormone signalling pathways can evolve from pre-existing mechanisms of transcriptional regulation independent of any hormone signalling.
Potential homologs of the NAP components were searched by sequence homology to the corresponding NAP proteins from M. polymorpha. Blasts were done using different databases: OneKp, PlantTFDB and Marchantia.info. Due to the lack of proteomic data in charophyte organisms, we carried out tblastn. Each transcript was then translated using Expasy Translate tool. Sequences resulting from this search were classed using protein sequence alignments and phylogenetical studies. Protein sequences alignments were done with Multialin (http://multalin.toulouse.inra.fr/multalin/) and ESPrit (http://espript.ibcp.fr/ESPript/ESPript/) online tools. Phylogenetic analyses were conducted using predicted DBDs from charophyte proto-ARFs and DBDs belonging to A. thaliana and M. polymorpha ARFs. Phylogenies were done with MEGA and Phylogeny.fr software using Maximum likelihood algorithm.
Protein structure modelling was done with Phyre2 online tool [47]. Three-dimensional structures were visualized with PyMOL software (www.pymol.org).
cDNA sequences coding for potential ancestors and the corresponding mutants were constructed as synthetic DNA (Thermofisher). KnRAV and CaARF (full-length, fragments (CaARF-DBD (residues 1–421), CaARF-PB1 (residues 644–750), KnRAV-DBD (residues 256–523), KnRAV-PB1 (residues 724–798)) or mutants) coding sequences were cloned into pETM40 plasmid (EMBL) that contains a MBP-tag in the N-terminal region except for PB1 domains from both proteins that were cloned into pETM11 (EMBL) that confers a N-terminal His-tag.
KnRAV and CaARF specific domains were isolated by PCR from synthetic cDNA sequences (S6 Table). Full-length ARF2, ARF5 and ARF10 were cloned into pHMGWA vectors (Addgene) containing N-terminal His-MBP-His tags.
All proteins were expressed in Escherichia coli BL21 strain. Bacteria cultures were grown with the appropriate antibiotics at 37°C until they achieved an OD600nm of 0.6–0.9. Protein expression was induced with isopropyl-β-D-1-thyogalactopiranoside (IPTG) at a final concentration of 400 μM at 18°C overnight. Bacteria cultures were centrifuged, and the pellets were resuspended and sonicated in the buffers indicated in S7 Table.
After centrifugation, soluble fractions of KnRAV, KnRAV-DBD, CaARF, CaARF-DBD and CaARF mutants were loaded on Dextrin-Sepharose (GE Healthcare) column previously equilibrated in buffer A (S7 Table). After column washing, proteins were eluted in buffer A with maltose 10 mM (S6 Table).
PB1 domains of KnRAV and CaARF as well as full-length proteins ARF2, ARF5 and ARF10 were purified on Nickel-Sepharose (GE Healthcare) columns previously equilibrated in the appropriate buffers (S6 Table). After protein binding, columns were washed with 30 mM imidazole to remove all proteins non-specifically bound to the column. Proteins were eluted in the corresponding buffer containing 300 mM imidazole (S6 Table). His-tags of PB1 domains were cleaved by TEV protease (5% w/w) overnight at 4°C followed by incubation at 20°C for 2 h for SEC-MALLS experiments.
AtTPL202 and mutants were purified as explained in Martin-Arevalillo et al., 2017 [49]. Following purification step, all proteins were dialyzed for 15 h at 4°C in their purification buffers, frozen in liquid nitrogen and conserved at -80°C until used.
DNA probes were artificially designed based on the DNA binding site for each TF (S8 Table) (Eurofins). Oligonucleotides for the sense strand were designed with an overhanging G in 5’ that allows the labelling of the DNA (S8 Table). Annealing of the oligonucleotides and Cy5-labelling of the probes were performed as described in Stigliani et al.,(2019) [19]. Electrophoretic Mobility Shift Assays (EMSA), were done on native 2% agarose gels prepared with TBE buffer 0.5X. Gels were pre-run in TBE buffer 0.5X at 90 V for 90 min at 4°C. Protein-DNA mixes contained Salmon and Herring Sperm competitor DNA (final concentration 0.07 mg/ml) and labelled DNA (final concentration 20 nM) in the interaction buffer (20 mM HEPES pH 7.8; 50 mM KCl; 100 mM Tris-HCl pH 8.0, 2.5% glycerol; 1 mM DTT). Mixes were incubated in darkness for 30 min at 4°C and next loaded in the gels. Gels were run for 1 hour at 90 V at 4°C in TBE 0.5X. DNA-protein interactions were visualized with Cy5-exposition filter (Biorad ChemiDoc MP Imaging System).
For protein-protein interaction analyses, complexes between potential interaction partners were first formed by mixing MBP-tagged CaARF (wt or mutants) (90 μg) with His-tagged AtTPL202 (and mutants) (70 μg) in CAPS 20 mM pH 9.6; Tris-HCl 100 mM pH 8; NaCl 50 mM; TCEP 1 mM buffer for 1 h at 4°C. Complexes formed were fixed through the MBP tag to Dextrin-Sepharose columns previously equilibrated with CAPS 20 mM pH 9.6; Tris-HCl 100 mM pH 8; NaCl 50 mM; TCEP 0.1 mM buffer. After incubation of the complexes with Dextrin-Sepharose for 30 min at 4°C, nonspecific interactions were removed by a washing step with the same buffer. Protein complexes were eluted with 200 μl of the same buffer containing 10 mM of maltose. MBP was used as control for unspecific interactions. The eluted fractions were analysed by SDS-page polyacrylamide gel electrophoresis 12%.
Molecular weights were determined by Size-Exclusion Chromatography-Multi Angle Light Scattering (SEC-MALLS) on an analytical Superdex-S200 increase (GE Healthcare) connected to an in-line MALLS spectrometer (DAWN HELEOS II, Wyatt Instruments). Analytical size exclusion chromatography was performed at 25°C at a rate of 0.5 mL/min for untagged PB1 domains resulting from TEV cleavage. Untagged KnRAV-PB1 MW determination was carried out in CAPS 100 mM pH 9.6; TCEP 1mM buffer, whereas Tris-HCl 20 mM pH 8; TCEP 1 mM was used for untagged CaARF-PB1 and AtARF5. The refractive index measured with in-line refractive index detector (Optirex, Wyatt Instruments) was used to follow the differential refractive index relative to the solvent. Molecular masses calculation was done with the Debye model using ASTRA version 5.3.4.20 (Wyatt Instruments) and a theoretical dn/dc value of 0.185 mL/g.
|
10.1371/journal.pgen.1003895 | ALS-Associated FUS Mutations Result in Compromised FUS Alternative Splicing and Autoregulation | The gene encoding a DNA/RNA binding protein FUS/TLS is frequently mutated in amyotrophic lateral sclerosis (ALS). Mutations commonly affect its carboxy-terminal nuclear localization signal, resulting in varying deficiencies of FUS nuclear localization and abnormal cytoplasmic accumulation. Increasing evidence suggests deficiencies in FUS nuclear function may contribute to neuron degeneration. Here we report a novel FUS autoregulatory mechanism and its deficiency in ALS-associated mutants. Using FUS CLIP-seq, we identified significant FUS binding to a highly conserved region of exon 7 and the flanking introns of its own pre-mRNAs. We demonstrated that FUS is a repressor of exon 7 splicing and that the exon 7-skipped splice variant is subject to nonsense-mediated decay (NMD). Overexpression of FUS led to the repression of exon 7 splicing and a reduction of endogenous FUS protein. Conversely, the repression of exon 7 was reduced by knockdown of FUS protein, and moreover, it was rescued by expression of EGFP-FUS. This dynamic regulation of alternative splicing describes a novel mechanism of FUS autoregulation. Given that ALS-associated FUS mutants are deficient in nuclear localization, we examined whether cells expressing these mutants would be deficient in repressing exon 7 splicing. We showed that FUS harbouring R521G, R522G or ΔExon15 mutation (minor, moderate or severe cytoplasmic localization, respectively) directly correlated with respectively increasing deficiencies in both exon 7 repression and autoregulation of its own protein levels. These data suggest that compromised FUS autoregulation can directly exacerbate the pathogenic accumulation of cytoplasmic FUS protein in ALS. We showed that exon 7 skipping can be induced by antisense oligonucleotides targeting its flanking splice sites, indicating the potential to alleviate abnormal cytoplasmic FUS accumulation in ALS. Taken together, FUS autoregulation by alternative splicing provides insight into a molecular mechanism by which FUS-regulated pre-mRNA processing can impact a significant number of targets important to neurodegeneration.
| FUS/TLS is a frequently mutated gene in amyotrophic lateral sclerosis (ALS). ALS, also known as Lou Gehrig's disease, is characterized by a progressive degeneration of motor neurons. The abnormal cytoplasmic accumulation of mutant FUS protein is a characteristic pathology of ALS; however, recent evidence increasingly suggests deficiencies in FUS nuclear function may also contribute to neurodegeneration in ALS. Here we report a novel autoregulatory mechanism of FUS by alternative splicing and nonsense mediated decay (NMD). We show FUS binds to exon 7 and flanking introns of its own pre-mRNAs. This results in exon skipping, inducing a reading frame shift and subsequent degradation of the splice variants. As such, this mechanism provides a feedback loop that controls the homeostasis of FUS protein levels. This balance is disrupted in ALS-associated FUS mutants, which are deficient in nuclear localization and FUS-dependent alternative splicing. As a result, the abnormal accumulation of mutant FUS protein in ALS neurons goes unchecked and uncontrolled. Our study provides novel insight into the molecular mechanism by which FUS regulates gene expression and new understanding of the role of FUS in disease at the molecular level. This may lead to new potential therapeutic targets for the treatment of ALS.
| Amyotrophic lateral sclerosis (ALS) is a neuronal degenerative disorder caused by progressive loss of motor neurons in brain and spinal cord, leading to paralysis and death [1]. FUS is a frequently mutated gene in ALS (combining familial and sporadic ALS), in addition to C9ORF72, SOD1 and TDP-43 [1]–[3]. Most ALS-associated FUS mutations are within the nuclear localization signal (NLS) in the carboxyl terminus [2], [4], [5], resulting in increased cytoplasmic FUS localization [6], [7]. The abnormal cytoplasmic aggregation of FUS mutants in neuron and glial cells is a pathological hallmark of ALS and some cases of frontotemporal lobar degeneration (FTLD) [8]–[10]. It's noteworthy that there is a correlation between the observed cytoplasmic FUS accumulation and the age of ALS onset, with the more cytoplasmic FUS accumulation the earlier age of disease onset [8], [11]–[13]. Several studies suggest that cytoplasmic accumulation of FUS mutant protein can lead to direct cytoplasmic cytotoxicity or may indirectly result in the loss of FUS function in the nucleus. Studies in yeast models demonstrated that expression of ALS-associated FUS mutants can lead to protein aggregation and cytotoxicity that recapitulate FUS proteinopathy [14]. Investigations in some Drosophila, C. elegans and rat models showed that expression of ALS-associated FUS mutants can lead to motor neuron dysfunction and neurodegeneration [15]–[17]. However, some Drosophila and zebrafish models support that the loss of FUS function can lead to behavioral and structural defects of motor neurons [18], [19]. Exactly how the loss of FUS nuclear function and/or the gain of cytoplasmic cytotoxicity contribute to neurodegeneration at the molecular level is still unknown.
FUS is predominantly a nuclear protein [20], and binds both DNA and RNA [21], [22]. FUS is involved in multiple steps of RNA metabolism including transcription, pre-mRNA splicing and mRNA transport for site specific translation [23]–[25]. The alteration of FUS-regulated RNA processing is a proposed key event in ALS pathogenesis, given that RNA binding proteins and splicing misregulation are linked to neurological diseases [8], [26], [27]. To understand the normal function of FUS in RNA processing, it is essential to identify FUS RNA targets. Recently a large number of FUS RNA targets in various cell lines and neural tissues were identified by CLIP-seq (cross-linking and immunoprecipitation, followed by high-throughput sequencing), a method to purify protein-RNA complexes coupled with deep sequencing [28]–[32]. The challenge now is to begin to understand what the biological significance of FUS-regulated RNA processing is, and how these processes are altered in FUS mutants and may therefore contribute to ALS pathogenesis.
Our CLIP-seq data in HeLa cells show that FUS binding is enriched in introns flanking cassette exons of pre-mRNAs. Among the identified FUS-binding cassette exons and their flanking introns, the most highly enriched target is exon 7 and flanking introns of FUS pre-mRNA itself. Here we demonstrate that FUS is a repressor of exon 7 and that the exon 7-skipped splice variants of FUS are subject to nonsense-mediated decay (NMD). Overexpression of FUS leads to repression of exon 7 splicing and predictably a reduction of endogenous FUS protein levels. Conversely, knockdown of FUS protein reduces the repression of exon 7. Moreover, the reduction of exon 7 repression can be rescued by the expression of EGFP-FUS. Taken together, these studies show that FUS dynamically autoregulates its own protein levels by directly modulating the alternative splicing of exon 7. Furthermore, our data show that ALS-associated FUS mutants are deficient in nuclear localization, exon 7 repression and autoregulation of its own protein levels. We propose that the compromised FUS autoregulation in ALS forms a feed-forward loop, exacerbating the abnormal cytoplasmic FUS accumulation, and as such, provides a molecular mechanism that can potentially contribute to ALS pathogenesis.
To identify RNA targets of FUS, we performed FUS CLIP-seq in HeLa cells. Western blot and autoradiography showed successful immunoprecipitation of FUS protein and FUS-RNA complexes (Figure 1A, 1B and 1C). Sequencing of FUS CLIP RNA yielded 1,879,212 non-redundant reads mapped to human genome GRCh37, with the majority (1,305,507 reads) to pre-mRNAs (Figure S1). Using the peak-finding algorithm CisGenome (www.biostat.jhsph.edu/~hji/cisgenome/) [33], we identified 1928 FUS CLIP clusters (sites containing significantly enriched overlapping FUS CLIP tags) corresponding to 1149 target genes (Table S1) in HeLa cells. FUS RNA targets identified by our CLIP-seq were compared with those previously identified by other CLIP-seq [28]–[30], [32], PAR-CLIP [31] and RIP-chip [34] (Figure S2). HeLa and HEK293 cells [31] share 845 common target genes (Figure S2, Table S2), accounting for 74% of all identified targets in HeLa cells. Gene Ontology (GO) biological process (BP) analysis of these 845 genes showed an enrichment of genes regulating gene expression and transcription. Analysis of various CLIP-seq datasets of mouse brains or neurons [28]–[30], [32] identified 508 common genes, which are enriched for genes regulating cell adhesion, synaptic transmission, glutamate signaling pathways and nervous system development. 120 genes are common to all the datasets analyzed, and show an enrichment of genes regulating cell motion and protein dephosphorylation. Taken together, our analyses revealed cell-type specific and common RNA targets of FUS.
Seventy five percent of our FUS CLIP clusters were located within introns (Figure 1D), consistent with previous reports [28]–[32]. To address the function of FUS in alternative splicing, we analyzed the association between FUS CLIP clusters and known alternative splicing events. Using the UCSC Known AltEvent database as a reference, we scored a FUS CLIP cluster as associated with an alternative splicing event if the CLIP cluster overlapped the alternative splicing event itself or overlapped its immediate flanking introns and/or exons, as described previously [35]. Our analysis identified “cassette exon” as the top category of alternative splicing events associated with FUS CLIP clusters (Figure S3). FUS CLIP clusters are associated with 206 cassette exons in total (Figure S3). To identify FUS binding regions flanking cassette exons, we used 87 FUS-associated cassette exons that are flanked by constitutive exons (Table S3) to generate a normalized complexity map as previously described [36]. We found that FUS binding was enriched in the flanking introns, particularly proximal to splice sites flanking the cassette exons (Figure 1E). The peak at 5′ splice sites, within 100 nucleotides (nt) downstream of the cassette exons, showed the highest enrichment of FUS CLIP tags. The peak proximal to 3′ splice sites was about 150 nt upstream of the cassette exons, instead of immediately upstream of 3′ splice sites (less than 50 nt), as previously described [31]. We also observed a peak at about 400 nt downstream of the cassette exons and a peak at about 300 nt downstream of the upstream constitutive exons. The locations of all these four peaks in our complexity map were also detected as statistically significant FUS binding sites in the complexity map from Lagier-Tourenne et al. [30]. Comparing our FUS complexity map with all the other reports [28]–[30], it is consistent that in general FUS binding is enriched in the intronic regions 500 nt upstream or downstream of cassette exons or constitutive exons flanking cassette exons, although the exact nucleotide positions are not identical in different studies. FUS-RNA binding may be both position and sequence dependent.
We next analyzed the sequences of FUS CLIP clusters associated with cassette exons and their flanking introns for possible de novo consensus RNA-binding motifs using the HOMER algorithm [37]. Analysis of the CLIP clusters within each individual peak on the complexity map (Figure 1E) did not identify any significant common consensus motifs (data not shown). Individually, the highest FUS-binding peak at the 5′ splice sites downstream of cassette exons did show an enrichment of CAGGUU (2.6 fold, P = 0.001) (Figure S4); however, this is expected as CAGGUU is very similar to the human 5′ splice site consensus sequences MAG|GURAGU (M is A or C and R is A or G) [38].
To assess whether genes encoding FUS-associated cassette exons can be clustered into functional groups, we analyzed the Gene Ontology (GO) biological process (BP) terms and KEGG pathways using DAVID Bioinformatics Resources 6.7 [39]. Our results showed that the most enriched GO BP terms were regulation of transcription, RNA metabolic process and neurogenesis (Table S4). The most enriched KEGG pathways were Wnt, adherens junction and Notch signaling pathways (Table S5).
Out of the 87 cassette exons enriched with FUS CLIP clusters, the top candidate was exon 7 and flanking introns 6 and 7 of the pre-mRNAs of FUS itself (Table S3). This CLIP cluster was also in the top 10 of all 1928 FUS CLIP clusters identified, as ranked by fold enrichment. The number of FUS CLIP tags within exon 7 and its flanking introns were 8.1 fold higher than the control mouse IgG CLIP tags (FDR = 0.043), as determined by the peak finding algorithm CisGenome [33] (Figure 2A). The region encompassing FUS intron6-exon7-intron7 is ∼3 kb and highly conserved in 38 vertebrate species (Figure 2B). Human and mouse DNA sequences share 77% identity within this region, while the average similarity throughout other introns of FUS is 40%. Sub-regions with highly enriched CLIP tags (over 100 overlapping CLIP tags in the center) were used for de novo consensus RNA motif analysis. Analysis of all the CLIP tags within these selected regions using the Homer algorithm revealed that GU or GGU containing sequences are statistically enriched over background (all pre-mRNA sequences) (Figure S5). This is consistent with previous reports that GGUG, GUGGU, or GGU containing RNA sequences are potential FUS binding motifs [29], [30], [40].
To validate the interaction between FUS and its own pre-mRNA, we performed anti-FUS immunoprecipitation in HeLa cells, followed by RNA purification, reverse transcription (RT) and PCR using primers specific to the introns flanking exon 7. Our results showed that indeed FUS interacted in vivo with its own pre-mRNAs at the region of exon 7 and its flanking introns (Figure 2C), compared to the constitutive exon 5 that was not enriched in FUS immunoprecipitates. Of note, FUS-exon7 complex (Figure 2C, lane 5) was immunoprecipitated even in the absence of UV crosslinking and under high salt wash conditions (containing 750 mM NaCl), suggesting a strong association of FUS protein with exon 7 of its own pre-mRNAs.
Skipping of FUS exon 7 results in an open reading frame shift and introduces a premature stop codon in exon 8. The exon 7-skipped FUS transcripts (NCBI RefSeq, NR_028388.2; Ensemble, ENST00000566605) are predicted to be subject to NMD. To detect the exon 7-skipped variant, we treated cells with cycloheximide (CHX) for 6 hours (h), which inhibits translation and thereby NMD [41]. FUS splice variants were assessed by reverse transcription (RT) and PCR using primers in exon 6 and exon 8. Indeed, we observed that exon 7-skipped splice variants were present in a variety of human and mouse cell lines including human cervical cancer cells HeLa, embryonic kidney cells HEK293, neuroblastoma cells SH-SY5Y, and mouse motor neuron cells NSC-34 (Figure 2D). Although the ratio of exon 7-skipped variants varies in different cell lines, they were all increased after CHX treatment, suggesting these variants undergo NMD. The PCR products of exon 7-skipped variants in HEK293 cells were confirmed by cloning and sequencing. At 6 h post CHX treatment, no significant changes were detected at the FUS protein level in all the cells tested (Figure S6).
We assessed the splicing of FUS exon 7 in the context of the splicing reporter human β globulin minigene (Figure 3A) by RT-PCR [42]. In the pDUP-FUS-E7L (Long) construct, we cloned FUS exon 7 and 2.8 kb of flanking introns 6 and 7 to encompass the entire conserved region enriched with FUS CLIP tags. In the pDUP-FUS-E7S (Short) construct, we cloned FUS exon 7 with about 300 bp of each flanking intron to assess the effects of distal intronic regions without compromising exon 7 splice sites. The pDUP-FUS-E5 construct containing FUS exon 5 and its flanking introns was used as a control, since exon 5 is a constitutive exon with no enrichment of FUS binding. The level of exon 7 inclusion is around 50% in the context of pDUP-FUS-E7L reporter transfected into HEK293 cells (Figure 3B, lane 1), similar to the endogenous FUS transcript. Interestingly, exon 7 inclusion level was lower in the pDUP-FUS-E7S reporter (Figure 3B, lane 3), suggesting additional regulatory elements in intron 6 and intron 7. This may explain why the entire region spanning intron6-exon7-intron7 is highly conserved.
To assess whether the splicing of exon 7 is affected by FUS protein levels, a gain of function assay with the EGFP-FUS plasmids and a loss of function assay with FUS siRNA were performed in HEK293 cells. Our results showed that the repression of exon 7 was enhanced significantly in both pDUP-FUS-E7L (from 48.3%±0.6% to 93.1%±2.7%, mean ± SD, n = 3) and pDUP-FUS-E7S reporters (from 74.9%±1.3% to 91.0%±0.6%) when EGFP-FUS was expressed (Figure 3B, 3C). As a control, the splicing of exon 5 in the pDUP-FUS-E5 reporter was not affected by FUS overexpression. Conversely, the level of the exon 7-skipped products decreased strikingly in the pDUP-FUS-E7L reporter (from 48.7%±4.5% to 3.7%±1.4%) when endogenous FUS protein was reduced by siRNA (Figures 3D, 3E). The exon 7-skipped splice variant in pDUP-FUS-E7S reporter was also decreased but to a lesser extent, from 64.8%±4.8% to 46.0%±4.1%, which suggests that more regulatory elements in the entire intron6-exon7-intron7 region are required to control exon 7 alternative splicing.
To further test the dependence of exon 7 splicing on FUS protein levels, a rescue assay with the pDUP-FUS-E7L reporter was performed by knocking down endogenous FUS using siRNAs that target the 3′ UTR of FUS pre-mRNAs, followed by expressing EGFP-FUS (Figure 4A, 4B). The level of exon 7-skipped splice variants was reduced from 67.7%±0.6% (lane 1) to 9.5%±3.8% (lane 3) at 48 h post siFUS treatment, and recovered to 82.0%±0.6% (lane 9) after introduction of EGFP-FUS for 24 h (Figure 4C). This assay strongly supports that FUS is a repressor of its own exon 7. Moreover, it also demonstrated that EGFP-FUS is as competent as the endogenous FUS to repress exon 7 splicing.
To determine whether FUS protein levels affect exon 7 splicing of endogenous FUS transcripts, semi-quantitative PCR using radiolabeled primers was performed to examine the FUS splice variants after siRNA knockdown. To prevent NMD of the exon 7-skipped FUS transcripts, we treated cells with cycloheximide (CHX), which allowed visualization of increases of exon 7 skipping in endogenous transcripts (Figure 5A, comparing bottom bands in lanes 4 and 5 with those in lanes 1 and 2). It is important to note that the siRNA targets both FUS splice variants, but the difference in the reduction of each splice variant relative to its corresponding mock transfection control is quantifiable and informative [43]. The level of exon 7-skipped variants (bottom band, lane 6) was reduced to 15% of the mock transfection (bottom band, lane 4), while the level of the exon 7-included variant (top band, lane 6) was only reduced to 50% of the mock transfection control (top band, lane 4) upon siRNA knockdown of FUS and CHX treatment (Figure 5A). A lesser reduction in exon 7-included variants than in exon 7-skipped variants was also observed without CHX treatment (lane 3). This result is consistent with the splicing reporter minigene assay, indicating that FUS is a repressor of exon 7 and reduced FUS protein levels results in less exon 7 repression. Western blot analysis confirmed siRNA knockdown of endogenous FUS protein (Figure 5B). Expression of splicing factors SF2 and hnRNPA1 were unaffected (Figure 5B), suggesting changes in FUS splicing were unlikely the result of an indirect mechanism.
Taken together, we have demonstrated that FUS is a key repressor of its own exon 7 in both splicing reporter assays and endogenous FUS splicing assays.
We showed that FUS repressed its exon 7 splicing and that the resultant exon 7-skipped transcripts were degraded by NMD and cannot be translated to protein. This observation led us to hypothesize that FUS can autoregulate its own protein levels by regulating the alternative splicing of exon 7. If our hypothesis is correct, we predict that exogenous expression of FUS will downregulate endogenous FUS protein levels by promoting exon 7 skipping and consequently NMD. Western blot analysis using FUS antibody detected both endogenous FUS and EGFP-FUS. The results showed that the endogenous FUS protein level was decreased by about 50% with transient expression of EGFP-FUS in HEK293 cells (Figure 5C). The endogenous FUS mRNA level was measured by quantitative RT-PCR (qRT-PCR) using primers annealing to the 3′ UTR of endogenous FUS transcripts but not the coding sequence of the EGFP-FUS transcripts. There was a slight reduction of endogenous FUS mRNA levels in the EGFP-FUS expressing cells, but no statistical significance was detected (Figure S7), suggesting the observed reduction in FUS protein levels occurs mainly at the post transcriptional level. Our finding of FUS autoregulation is also consistent with the observation in a FUS transgenic mouse model that the endogenous mouse FUS protein was reduced following overexpression of human FUS [44].
The majority of ALS-associated FUS mutations occur within the region coding for the nuclear localization signal, resulting in cytoplasmic retention of FUS [5] and inferred loss of FUS function in the nucleus. We propose that FUS autoregulation is deficient in ALS mutants due to the alteration of their cellular localization, which results in compromised FUS-dependent splicing regulation. To test this hypothesis, we made EGFP-FUS constructs with the ALS-associated mutations R521G, R522G and ΔE15 (deletion of last 12 amino acids in the C-terminus), which respectively correlates with minor, moderate, and severe cytoplasmic accumulation of FUS, as reported by others both in ALS patients and in cell culture systems [6], [8], [9], [11]. As a control, an RNA binding incompetent mutant EGFP-FUS RRM 4F-L was made by mutating four phenylalanine (F) to leucine (L) (F305L, F341L, F359L, and F368L) in the RNA recognition motif (RRM) [45]. We tested the effects of these mutants on FUS cellular localization, exon 7 splicing and autoregulation in HEK293 cells.
Consistent with previous reports [6], [8], [9], [11], we observed predominant nuclear localization of wildtype FUS, minor cytoplasmic and mainly nuclear localization of R521G, moderate cytoplasmic accumulation and aggregation of R522G, and severe cytoplasmic aggregation with much less nuclear localization of the FUS ΔE15 mutant, following transient transfection in HEK293 cells (Figure 6A) and mouse motor neuron cells NSC-34 (Figure S8). The RRM mutant, like the wildtype FUS protein, was predominantly localized in the nucleus.
To test the function of the FUS mutants in exon 7 alternative splicing, the splicing reporter minigene pDUP-FUS-E7L together with either wildtype or mutant EGFP-FUS plasmids were transfected into HEK293 cells (Figure 6B). Expression of wildtype EGFP-FUS protein resulted in repression of exon 7 as expected, with an increase of the exon7-skipped products from 40.4%±1.7% to 89.3%±3.0% (mean ± SD, n = 3). Expression of the ALS-associated mutants, compared to the wildtype FUS, resulted in significantly compromised repression of exon 7 (P≤0.05, n = 3). The more cytoplasmic localization of the ALS mutants, the less exon 7 repression, with exon 7 skipping ratio of 87.6%±2.8% for R521G, 70.6%±2.4% for R522G and 33.3%±10.1% for ΔE15 mutant. Expression of the EGFP-FUS RRM 4F-L mutant only resulted in a mild reduction of exon 7 repression with a ratio of exon 7 skipping of 82.1%±4.3%, which was statistically different from the EGFP-FUS wildtype protein (P≤0.05, n = 3). While the RRM 4F-L mutant was reported incompetent to bind RNA [45], its expression did increase the nuclear concentration of FUS protein, which may result in more endogenous FUS binding the intron6-exon7-intron7 region in the reporter minigene and repression of the exon 7 splicing. This may explain the slight reduction of exon 7 repression when the RRM 4F-L mutant was expressed. Our data suggest that both the nuclear concentration of FUS protein and RNA binding are critical for the regulation of FUS exon 7.
Consistent with the splicing data, western blot analysis confirmed that increased exon 7 skipping led to a decrease in endogenous FUS protein levels by 55.5%, when EGFP-FUS wildtype protein was expressed (Figure 6C). Conversely expression of various EGFP-FUS ALS mutants, which showed reduced nuclear localization and exon 7 skipping, resulted in less reduction of the endogenous FUS protein (Figure 6C). Expression of the ΔE15 mutant that is predominantly localized in the cytoplasm only downregulated the endogenous FUS protein by 14.6%. This mild downregulation of FUS protein is consistent with little increase of exon 7 skipping observed in the splicing assay (Figure 6B). We also confirmed in human neuroblastoma cells SH-SY5Y that ALS-associated FUS mutants were deficient in regulating exon 7 repression in the context of the pDUP splicing reporter minigene, and that the deficiency correlated with the extent of cytoplasmic localization of the mutants (Figure S9).
We observed that wildtype endogenous FUS protein was co-localized with the cytoplasmic aggregates of FUS ΔE15 mutant in both HEK293 cells (Figure 6D) and mouse motor neuron cells NSC-34 (Figure S8), using an antibody which detects only the endogenous FUS protein but not the ΔE15 mutant by recognizing a C-terminal FUS epitope. These data suggest that the FUS mutants may sequester the wildtype FUS protein in the cytoplasm and further contribute to the aggregation. This is consistent with a recent report that GFP-FUS ALS mutant is co-localized with MYC-FUS wildtype protein in the cytoplasmic aggregates [46]. We report here the localization of the endogenous FUS protein in the cytoplasmic aggregates of ALS-associated FUS mutants.
Taken together, our data suggest that the severity of FUS cytoplasmic accumulation correlates with the deficiency of exon 7 repression and autoregulation of FUS protein levels. The deficiency of ALS-associated FUS mutants in alternative splicing and autoregulation may exacerbate the cytoplasmic accumulation of ALS-associated FUS mutants.
Our data showed FUS autoregulation is deficient in cells expressing ALS-associated FUS mutants. Moreover, this deficiency increases with the relative severity of the cytoplasmic accumulation of individual FUS mutants, which would be expected to exacerbate the rate of cytoplasmic accumulation and FUS proteinopathy. Use of splicing-modulating antisense oligonucleotides (ASOs) is a therapeutic strategy of great potential to treat diseases arising from splicing defects [47]–[49]. We rationalized that ASOs promoting FUS exon 7 skipping should mimic the repression of exon 7 by FUS and thereby have the potential to restore the deficient FUS autoregulation in patients with ALS-associated FUS mutations. We designed FUS-ASOs to target the junction of intron 6 and exon 7. FUS-ASOs were synthesized using 2′-O-methyl-oligoribonucleotides with phosphorothioate linkages to increase ASO stability and then tested with the pDUP-FUS-E7L minigene in HEK293 cells. Our results showed that FUS-ASOs induced repression of exon 7 in a dose dependent manner (Figure 7). This suggests a possibility that deficient FUS autoregulation can be therapeutically restored to reduce or alleviate the extent of abnormal FUS cytoplasmic accumulation occurring in ALS patients with FUS mutations.
Here we report a novel autoregulatory mechanism of FUS by alternative splicing and NMD. The model shown in Figure 8A illustrates FUS autoregulation as a feedback loop to control the homeostasis of FUS protein levels. High levels of FUS protein lead to increased FUS binding to exon 7 and its flanking introns, promoting exon 7 skipping and NMD to reduce excessive FUS protein. Low levels of FUS protein would favor exon 7 inclusion, resulting in increased FUS protein production. Alternative splicing-mediated NMD and highly conserved intronic sequences represent an emerging common mechanism utilized by RNA binding proteins (RBPs) to maintain their homeostasis [43], [50], [51]. FUS now joins this increasing list of autoregulated RBPs, including PTB, hnRNP L, Nova and TDP-43 [43], [50]–[52]. FUS regulates many aspects of gene expression including transcription, alternative splicing and RNA transportation [23]–[25]. Dynamic regulation and conserved targets suggest it is important to keep these functional activities of FUS in tight control, and that FUS likely has a co-factor role in coordinating them. For example, loss of FUS can cause genomic instability and developmental defects in mouse, Drosophila and Zebrafish [18], [19], [53]. Conversely, high levels of FUS are associated with cancer and ALS, and moreover, are known genetic determinants of these diseases. Overexpression of FUS is observed in liposarcoma and leukemia with FUS translocations [54], [55]. Aberrant accumulation of FUS mutant protein is a characteristic pathology of FUS-associated ALS [8], [9]. Depletion of FUS in the mouse nervous system affects the abundance or the splicing of about 1000 mRNAs [30], suggesting that maintaining equilibrated FUS protein levels is critical for RNA processing.
In ALS, another frequently mutated gene TDP-43 is also an RNA binding protein that autoregulates its own protein levels [52], [56]. The direct mechanism of TDP-43 autoregulation is different from what we report here for FUS. TDP-43 binds to the 3′ UTR of its own pre-mRNA to trigger either NMD [56] or exosome-dependent degradation [52]. To our knowledge, we are the first to report a FUS autoregulatory mechanism through alternative splicing and NMD. Autoregulation of both FUS and TDP-43 by post transcriptional mechanisms suggests their functional activities are tightly controlled and that unbalancing of this regulation may underpin molecular mechanisms that promote neurodegeneration in ALS. Mice and rats expressing TDP-43 without the autoregulatory sequence developed more severe neurodegeneration than those expressing autoregulated wildtype or ALS-linked TDP-43 mutants, strongly suggesting deficient TDP-43 autoregulation contributes to neurodegeneration [57]. This is also likely the case for FUS autoregulation, which needs to be experimentally tested in rodent models. Interestingly, in both FUS-associated ALS and cancer, loss of heterozygosity of the FUS gene is never observed. This suggests, at least genetically, that while compromised FUS autoregulation contributes to the initiating or driving events resulting from FUS mutations in these diseases, the activity of the wild-type FUS allele is required or selected to maintain this pathological state. In this regard, it is important to understand how both mutant and wild-type FUS activities may contribute to the progression of ALS and cancer.
Compromised FUS homeostasis by autoregulation is expected in cells harbouring ALS-associated mutations (Figure 8B). The majority of ALS-associated mutations are located in the nuclear-localization signal (NLS) of FUS, resulting in both a cytoplasmic retention of FUS mutants and a reduction of FUS protein levels in the nucleus. The reduction of nuclear-localized FUS leads to the reduction of FUS exon 7 repression, which in turn likely induces production of more exon 7-included transcripts for translation, thereby driving elevated protein synthesis of FUS. In our model of FUS autoregulation, a deficiency of FUS in the nucleus would favour a feed-forward mechanism, and thereby actually exacerbate the abnormal cytoplasmic accumulation of FUS mutants. Our observation that endogenous wildtype FUS protein was co-localized with the cytoplasmic aggregates of FUS mutants in both HEK293 cells and NSC-34 motor neuron cells suggests that the FUS mutants may further sequester the wildtype FUS protein in the cytoplasm and form more aggregates. In ALS, FUS cytoplasmic accumulation is a progressive process and increases with disease duration [58]. The deficient FUS autoregulation may lead to long term detrimental effects, and could be part of the mechanism underlying age-dependent neurodegeneration and death of neurons with ALS-associated FUS mutants.
Indeed, a genotype-phenotype relation between different FUS mutations and FUS cytoplasmic accumulation or the age of ALS onset is observed [8], [11]–[13]. The stronger the NLS mutation (severity of cytoplasmic retention), the earlier the age of ALS onset. The three FUS mutants we constructed, R521G, R522G and ΔE15 (last 12 amino acids truncation) represent minor, moderate and severe cytoplasmic accumulation, respectively. The reported mean age of disease onset is 43 for R521G, 28.5 for R522G and 18 for R495X (last 32 amino acid truncation) in the later generation [8], [11]. Here, we demonstrated that in HEK293 cells and SH-SY5Y cells expressing these same R521G, R522G and ΔE15 FUS mutants, increased cytoplasmic localization of FUS directly correlated with increased deficiencies of exon 7 skipping and FUS autoregulation. We speculate that FUS exon 7-skipped splice variants are reduced in the tissues or cell lines derived from ALS patients with FUS mutations, which can be further experimentally tested.
Regulated splicing of exon 7 is a good model to examine FUS-dependent alternative splicing in detail, since it is one of the most significant FUS CLIP clusters identified in our FUS CLIP-seq. Our finding is also consistent with the report that FUS binds to highly conserved introns of genes encoding RNA binding proteins [32]. FUS exon 7 and its flanking introns were also identified previously as RNA targets of FUS and of TDP-43 respectively from FUS CLIP-seq and TDP-43 CLIP-seq of mouse brains [30], [56], but the functional significance of this implicated region was not experimentally tested. Interestingly, the prime molecular target of two genetic determinants of ALS converges on the same highly conserved FUS alternative exon and its flanking introns. This makes a compelling argument that the processing of FUS pre-mRNA specifically, and by extension, the role of FUS in alternative exon splicing in general, is an important molecular determinant of ALS. Lagier-Tourenne et al. proposed that FUS binding to FUS intron6-exon7-intron7 may result in retention of intron 7 to make a shorter FUS transcript with an alternative 3′ UTR [30]. We noticed that both NCBI RefSeq database (NR_028388.2) and Ensemble database (ENST00000566605) annotated a FUS splice variant without exon 7, which is predicted to undergo NMD. We experimentally validated that the exon 7-skipped variant of FUS was expressed in multiple human and mouse cell lines and that the steady state levels of the exon 7-skipped variant was increased after inhibition of NMD. Furthermore, we demonstrated that FUS is a repressor of its own exon 7 by splicing assays of both splicing reporter minigenes and endogenous FUS pre-mRNAs.
FUS-regulated alternative splicing of cassette exons is not just limited to its own exon 7. We found FUS CLIP clusters were significantly associated with alternative splicing events of cassette exons. Our normalized complexity map of 87 FUS-associated cassette exon events revealed that FUS CLIP clusters were enriched in the introns flanking cassette exons, proximal to upstream 3′ splice sites and downstream 5′ splice sites, with the highest peak overlapping the downstream 5′ splice sites. FUS binding proximal to 5′ splice sites suggests that FUS may be associated with the assembly of spliceosome at 5′ splice sites, consistent with previous reports that FUS, as well as the related family member TAF15, are in the U1 snRNP (small nuclear ribonucleoprotein) complex [59], [60]. FUS binding proximal to 3′ splice sites suggests it may also affect the spliceosome assembly at 3′ splice sites; this functional significance is yet to be determined [31]. Activation or repression of cassette exon splicing can be dependent on RNA binding positions, as suggested in Nova [36] and FOX2 [61] CLIP-seq data, which showed that Nova and FOX2 binding proximal to 3′ splice sites repressed cassette exons, while conversely binding proximal to 5′ splice sites promoted cassette exons. However, splicing factors such as hnRNP A1 and PSF binding to introns proximal to 5′ splice sites can also repress cassette exons [62], [63]; which suggests that activation or repression of cassette exon splicing is likely more complex. We experimentally demonstrated in this paper that FUS is a repressor of its own exon 7. However, it does not rule out the possibility that FUS may activate other cassette exons. The repression or activation of a given cassette exon by FUS in a tissue specific manner might be controlled by different signaling pathways and/or cell type specific splicing factors involved in the complex in a spatio-temporal manner.
Sequence motif analysis of FUS CLIP clusters in our data set did not identify a significant common consensus motif (data not shown). We found variable motifs throughout the four binding peaks in the complexity map (data not shown), suggesting limited or context-dependent FUS binding specificity, consistent with previous reports [28], [29], [31], [32]. Analysis of CLIP clusters within the region encompassed by the highest binding peak in the normalized complexity map (5′ splice site downstream of cassette exons) did identify an enrichment of a CAGGUU motif, which is similar to the human 5′ splice site consensus sequence MAG|GURAGU [38]. While this might be expected, it is interesting to note that GGU is the most common FUS binding site consensus sequence. This was originally reported in a SELEX assay (GGUG) [40], and subsequently by CLIP-seq analysis from Lagier-Tourenne et al. (GUGGU) [30] and Rogelj et al. (GGU) [29]. A detailed examination of consensus RNA motifs within FUS intron 6 and intron 7 did reveal that GU or GGU containing sequences were statistically enriched, which provides FUS intron6-exon7-intron7 as a model for further experimental determination of FUS binding sites.
FUS CLIP-seq data and RNA-seq data revealed a wide range of pre-mRNAs as candidate targets of FUS [28]–[32], [64]. However, the biological significance of FUS-regulated RNA processing is only now being examined. Here we demonstrated that FUS-regulated cassette exon splicing of its own pre-mRNA leads to NMD, suggesting that FUS-regulated alternative splicing may be a common post transcriptional mechanism for the regulation of gene expression. This function of FUS in regulating cassette exon splicing is likely conserved in different tissues and species, since FUS-associated cassette exons were also observed in human and mouse neural tissues [28]–[30]. Our evidence demonstrating FUS is a splicing repressor of its exon 7 strongly supports the hypothesis that alternative splicing of many other neuronal- and disease-associated genes containing FUS-targeted cassette exons may also be regulated by FUS.
In conclusion, our study uncovers an autoregulatory mechanism of FUS expression through alternative splicing and NMD, and demonstrates that its function in splicing regulation is deficient in ALS-associated FUS mutants. This study addresses a biological significance of FUS-regulated alternative splicing, and its potential relevance to ALS pathogenesis. Furthermore, our findings have important implications for the development of new therapeutic approaches to target alternative splicing in treating ALS. Taking advantage of our findings, splicing-modulating antisense oligonucleotides can be developed to induce exon 7 skipping and produce the FUS splice variants undergoing NMD. This may be a promising strategy to reduce the abnormal FUS cytoplasmic accumulation in ALS. Moreover, FUS autoregulation by alternative splicing provides insight into a molecular mechanism by which FUS-regulated pre-mRNA processing can impact a significant number of targets important to neurodegeneration.
Human cervical cancer cell line HeLa (ATCC, CCL-2) and human embryonic kidney cell line HEK293 (ATCC, CRL-1573) were grown in DMEM with 10% bovine growth serum (HyClone, SH30541.03). Human neuroblastoma cell line SH-SY5Y (gift from Dr. Louise Simard) was cultured in MEM-F12 medium (1∶1) supplemented with 10% fetal bovine serum (GIBCO, 12483). Mouse motor neuron cell line NSC-34 (gift from Dr. Louise Simard) was cultured in DMEM with 10% fetal bovine serum.
CLIP-seq was performed as described [65]. Briefly, HeLa cells on 15 cm dishes were UV crosslinked in vivo at 400 mJ/cm2 using Stratalinker (Stratagene 1800). Cell lysates were prepared in lysis buffer [65], followed with partial RNase A (Sigma, R6513) digestion at different final concentrations ranging from 0.001 µg/ml to 0.1 µg/ml. FUS-RNA complexes were immunoprecipitated using mouse monoclonal anti-FUS antibody 10F7 pre-bound with protein G agarose beads (Pierce, 22851). 10F7 is a mouse monoclonal antibody previously developed in the Hicks laboratory by immunizing BALB/c mice with GST (glutathione-s-transferase)-FUS fusion protein. Various clones were screened using western blotting, immunocytochemistry, and flow cytometry, and the 10F7 antibody performed best for all three methods. This antibody recognizes amino acids 34–51 of FUS. Immunoprecipitation using mouse IgG (Sigma, I5381) prebound with protein G agarose beads was performed in parallel as a control. While FUS-RNA complexes were still bound to beads, the 5′ end of CLIP RNA was radiolabeled with [γ-32P] ATP, and the 3′ end of CLIP RNA was ligated to a 3′ RNA linker. The radiolabeled FUS-RNA complex was separated onto a 10% (w/v) Bis-Tris gel (Novex NuPAGE), transferred to a nitrocellulose membrane (Bio-Rad, 162-0115), and exposed to X-ray film (Amersham Hyperfilm MP) for autoradiography. The appropriate shifted FUS-RNA bands were cut out of the nitrocellulose membrane and subject to protein digestion using proteinase K (Roche, 3115879001). RNA was recovered using phenol chloroform extraction and sodium acetate, ethanol-isopropanol (1∶1) precipitation. The recovered RNA was further ligated to a 5′ RNA linker, and subject to DNase (Promega, M6101) digestion and RNA recovery again. The final RNA product was reverse transcribed to cDNA using linker specific primers and subject to deep sequencing using the Illumina Genome Analyzer II (single end, 72 bp) at the Center of Applied Genomics in Toronto (TCAG).
Sequences of unique CLIP tags were mapped to the human genome (GRCh37) by BlastN [66] after trimming the CLIP linker sequences and removing duplicate CLIP tags. A peak finder algorithm CisGenome (www.biostat.jhsph.edu/~hji/cisgenome/) [33] was used to define CLIP clusters with significant enrichment (FDR≤0.05; FUS CLIP vs control mouse IgG CLIP). The RNA targets from our FUS CLIP-seq data in HeLa were compared to the RNA targets previously identified by other FUS CLIP-seq, PAR-CLIP and RIP-ChIP in different tissues and cells [28]–[32], [34]. If the gene list was not reported in the paper, the raw data of deposited fastq files [28]–[30] were retrieved from ENA (The European Nucleotide Archive, http://www.ebi.ac.uk/ena/) and uploaded to Galaxy (http://main.g2.bx.psu.edu/) [67] for sequence analysis. The CLIP-seq reads were mapped to mouse genome (mm9) using Bowtie (version 1.1.2) [68] built in Galaxy, with parameters reported in the corresponding paper. The same peak finder algorithm CisGenome was used to identify the CLIP clusters in all the datasets. Mouse genes were converted to the HGNC-approved human gene symbols (http://www.genenames.org/) to facilitate the comparison of different datasets. To identify the overlapping and the non-overlapping genes between different datasets, lists of the genes from the datasets were loaded into Venn, a web-based Venn diagram program, or Venn Diagram Plotter, a PC-based Venn diagram program (http://omics.pnl.gov/).
To identify our FUS CLIP clusters overlapping alternative splicing events, the genomic coordinates of CLIP clusters were searched against the UCSC Known AltEvent database (hg 19) [69] as described previously [35], [70]. Out of all 206 FUS-associated cassette exons, 87 cassette exons which are flanked by constitutive exons were used to generate a normalized complexity map as previously described [36]. FUS CLIP tags within 500 nucleotides upstream and/or downstream of these cassette exons and flanking exons were mapped. Control was an average of 100 sets of normalized complexity of 87 constitutive exons randomly selected from genes expressed in HeLa cells as determined by RNA-seq [71].
Analysis of de novo consensus RNA motif enrichment was performed using findMotifsGenome perl script of the Homer software [37] with parameters of 5 or 6 bases for motif length, 42 for target size, and –RNA option. Gene Ontology (GO) analysis was performed using DAVID Bioinformatics Resources 6.7 (http://david.abcc.ncifcrf.gov/) [39] or Enrichr (http://amp.pharm.mssm.edu/Enrichr/index.html) [72].
Immunoprecipitation of FUS protein was performed using mouse monoclonal anti-FUS antibody (10F7). FUS-bound RNA was recovered using phenol-chloroform extraction and sodium acetate, ethanol–isopropanol (1∶1) precipitation, as described previously [65]. DNA was removed by DNase treatment (Ambion, AM1906). The recovered RNA was reverse transcribed to cDNA using Superscript III (Invitrogen) and amplified using Phusion Hot Start Polymerase (NEB). Primers used for amplifying the region flanking FUS exon 7: 5′-ACAACCTTTTGTAGCCGTTGGAAG-3′ (forward), 5′-CTTTCTGGAGGTGGTTCTGGACAC-3′ (reverse). Primers used for amplifying the region flanking FUS exon 5: 5′-TCCCTAGTTACGGTAGCAGTTCTC-3′ (forward), 5′-GCTGCAGACAAAGCTGAAGACATC-3′ (reverse). PCR products were resolved on a 2% agarose gel and visualized by ethidium bromide staining.
Cycloheximide (CHX; Sigma) was added to cell culture medium at the final concentration of 100 µg/ml to inhibit translation and thereby nonsense mediated decay (NMD). At 6 h post CHX treatment, cytoplasmic RNA was extracted using RNeasy kit (QIAGEN) as per manufacture's recommendation. Reverse transcription (RT) was performed using Superscript III reverse transcriptase (Invitrogen). Radiolabeled PCR was performed to amplify FUS exon 7 splice using Phusion Hot Start DNA Polymerase (NEB). Primers were designed to anneal to exon 6 and exon 8. Primers: 5′-AGTGGTGGCTATGAACCCAGAGGT-3′ (forward), 5′-AGTCATGACGTGATCCTTGGTCCC-3′ (reverse). The reverse primer was labeled with [γ-32P] ATP using T4 PNK (NEB). PCR products were resolved on a 6% polyacrylamide/8M urea denaturing gel. The gel was dried, exposed to a phosphorimager plate (Kodak). The images of radioactivity signals were captured by a phosphorimager (Bio-Rad, Personal FX). The density of the radioactive bands was quantified using the Image J program v1.44p (NIH, Bethesda, MD, USA, http://rsbweb.nih.gov/ij/).
FUS cDNA was amplified from human fetal liver pAct2 cDNA library (Clontech). EGFP-FUS expression construct was made by subcloning the open reading frame of FUS cDNA (RefSeq: NM_004960.3) into the BglII and KpnI sites of pEGFP-C1 plasmid (Clone Tech). Mutagenesis of ALS-associated mutations (R521G, R522G) and deletion (ΔE15) were performed using Quickchange Lighting Site-directed Mutagenesis Kit as per manufacturer's recommendation (Stratagene). The EGFP-FUS RNA recognition motif (RRM) mutant 4F-L (F305L, F341L, F359L, and F368L) was generated using the QuikChange Lightning Multi-Site-Directed Mutagenesis kit as per manufacturer's recommendation (Stratagene). EGFP-FUS and mutant constructs were verified by DNA sequencing.
EGFP or EGFP-FUS (wildtype or mutant) plasmids were transiently transfected into HEK293 cells, or NSC-34 cells using Lipofectamine 2000 (Invitrogen) reagent as per manufacturer's recommendation. Cells were fixed with 4% paraformaldehyde. To detect the endogenous FUS protein, cells were incubated with the primary antibody rabbit anti-FUS (Bethyl Laboratories, BL1355) and the secondary antibody Alexa Fluor 568 or Alexa Fluor 488 donkey anti-rabbit antibody (Invitrogen). Nuclei were counter stained with DAPI or NucRed Dead 647 dye (Invitrogen). The localization of EGFP-FUS, mutants, and endogenous FUS protein was imaged using an Olympus FV500 confocal microscope and analyzed with Fluoview software version 4.3 (Olympus). Line sequential scanning was applied to avoid the potential bleed-through of fluorescence.
Splicing reporter minigene pDUP-FUS constructs: FUS exon 7 and its flanking introns were amplified using Phusion Hot Start Flex Polymerase (NEB) from human genomic DNA extracted from HEK293 cells. The sequence of interest was subcloned between the ApaI and BglII sites of the splicing reporter minigene pDUP175 plasmid [42]. Three reporters were made. The pDUP-FUS-E7L (Long) construct contains FUS exon 7, and 1453 bp upstream and 1355 bp downstream of the flanking introns. The pDUP-FUS-E7S (Short) construct contains FUS exon 7, 292 bp upstream and 321 bp downstream of the flanking introns. The pDUP-FUS-E5 construct is a control construct containing the sequence of FUS exon 5 and its flanking introns.
Splicing reporter minigene assay with FUS overexpression: 0.5 µg of pDUP reporter and 2 µg of EGFP-FUS plasmid or EGFP-FUS mutants were transfected into HEK293 cells using Lipofectamine 2000 (Invitrogen) reagent as per manufacturer's recommendation. At 48 h post transfection, cytoplasmic RNA was purified using RNeasy kit (QIAGEN). RT-PCR was performed to assess the splicing of FUS exon 7 using Superscript III reverse transcriptase (Invitrogen) and Phusion Hot Start DNA Polymerase (NEB). Primer sequences: 5′-CTCAAACAGACACCATGCATGG-3′ (forward) and 5′-CAAAGGACTCAAAGAACCTCTG-3′ (reverse). PCR products were resolved on a 3% agarose gel with ethidium bromide staining, and imaged using Fusion FX imager (Vilber Lourmat). The intensity of PCR bands was quantified using ImageJ software as described above.
Splicing reporter minigene assay with siRNA knockdown of FUS: 0.5 µg of the pDUP-FUS-E7L reporter and 20 nM (final concentration) of FUS siRNA (Dharmacon ON-TARGETplus SMARTpool) were transfected into HEK293 cells using Lipofectamine 2000 reagent (Invitrogen) as per manufacturer's recommendation. At 48 h post transfection, cytoplasmic RNA was purified; and splice variants were analyzed by RT-PCR and gel electrophoresis, as described above.
Rescue assay: Endogenous FUS protein was knocked down by transfecting HEK293 cells with a custom designed siRNA targeting 3′ UTR of human FUS (siFUS). siFUS sequence: 5′-UAUAGUUACAAUUACAUAGUCCGACAC-3′ (IDT, DsiRNA). The siRNA was transfected using Lipofectamine 2000 (Invitrogen) as per manufacturer's recommendation. At 48 h post transfection of siRNA, cells were retransfected with pDUP-FUS-E7L plasmid alone or together with EGFP or EGFP-FUS plasmid to rescue the FUS protein levels. At 24 h post re-transfection, cytoplasmic RNA was isolated for analysis of splice variants by RT-PCR, as described above.
Rabbit anti-FUS antibody (Bethyl Laboratories, BL1355) or mouse anti-FUS antibody (10F7), rabbit anti-Actin antibody (Sigma, A2066), mouse anti-SF2 (Santa Cruz, SC73026) and mouse anti-hnRNP A1 (ImmuQuest, IQ205) were used for western blot analysis. Western blot was developed using ECL prime reagent (Amersham) and imaged with ChemiDoc MP imaging system (Bio-Rad). The protein band intensity was quantified using the ImageJ program v1.44p (NIH, Bethesda, MD, USA, http://rsbweb.nih.gov/ij/). Actin was used to normalize the loading of protein amount.
The endogenous human FUS transcripts were measured by quantitative RT-PCR with qPCR SYBR Green mix (Fermentas), using a real-time PCR system (Bio-Rad, CFX96). Primers were designed to anneal to the 3′ UTR region of the endogenous FUS transcript. Primer sequences: 5′-CCAATTCCTGATCACCCAAGGGTTT-3′ (forward), 5′-TGGGCAGGGTAATCTGAACAGGAA-3′ (reverse).
2′-O-methyl-oligoribonucleotides with phosphorothioate linkages were synthesized and purified by Trilink Biotechnologies, Inc. (San Diego, CA). FUS-ASOs target the splice junction spanning intron 6 and exon 7. FUS-ASO sequences: 5′-GUCACUUCCGCUGGAGAAGA-3′. Control ASOs (Ctrl-ASOs) are ASOs targeting the SRA gene [73]. ASOs alone (final concentration 2 nM, 10 nM and 50 nM) or ASOs together with pDUP-FUS-E7L reporters (0.5 µg) was transfected into HEK293 cells using Lipofectamine 2000 reagent (Invitrogen) as per manufacturer's recommendation. The splicing of exon 7 in the reporter was assessed by RT-PCR, as described above.
|
10.1371/journal.pcbi.1003024 | Reinforcement Learning Using a Continuous Time Actor-Critic Framework with Spiking Neurons | Animals repeat rewarded behaviors, but the physiological basis of reward-based learning has only been partially elucidated. On one hand, experimental evidence shows that the neuromodulator dopamine carries information about rewards and affects synaptic plasticity. On the other hand, the theory of reinforcement learning provides a framework for reward-based learning. Recent models of reward-modulated spike-timing-dependent plasticity have made first steps towards bridging the gap between the two approaches, but faced two problems. First, reinforcement learning is typically formulated in a discrete framework, ill-adapted to the description of natural situations. Second, biologically plausible models of reward-modulated spike-timing-dependent plasticity require precise calculation of the reward prediction error, yet it remains to be shown how this can be computed by neurons. Here we propose a solution to these problems by extending the continuous temporal difference (TD) learning of Doya (2000) to the case of spiking neurons in an actor-critic network operating in continuous time, and with continuous state and action representations. In our model, the critic learns to predict expected future rewards in real time. Its activity, together with actual rewards, conditions the delivery of a neuromodulatory TD signal to itself and to the actor, which is responsible for action choice. In simulations, we show that such an architecture can solve a Morris water-maze-like navigation task, in a number of trials consistent with reported animal performance. We also use our model to solve the acrobot and the cartpole problems, two complex motor control tasks. Our model provides a plausible way of computing reward prediction error in the brain. Moreover, the analytically derived learning rule is consistent with experimental evidence for dopamine-modulated spike-timing-dependent plasticity.
| As every dog owner knows, animals repeat behaviors that earn them rewards. But what is the brain machinery that underlies this reward-based learning? Experimental research points to plasticity of the synaptic connections between neurons, with an important role played by the neuromodulator dopamine, but the exact way synaptic activity and neuromodulation interact during learning is not precisely understood. Here we propose a model explaining how reward signals might interplay with synaptic plasticity, and use the model to solve a simulated maze navigation task. Our model extends an idea from the theory of reinforcement learning: one group of neurons form an “actor,” responsible for choosing the direction of motion of the animal. Another group of neurons, the “critic,” whose role is to predict the rewards the actor will gain, uses the mismatch between actual and expected reward to teach the synapses feeding both groups. Our learning agent learns to reliably navigate its maze to find the reward. Remarkably, the synaptic learning rule that we derive from theoretical considerations is similar to previous rules based on experimental evidence.
| Many instances of animal behavior learning such as path finding in foraging, or – a more artificial example – navigating the Morris water-maze, can be interpreted as exploration and trial-and-error learning. In both examples, the behavior eventually learned by the animal is the one that led to high reward. These can be appetite rewards (i.e., food) or more indirect rewards, such as the relief of finding the platform in the water-maze.
Important progress has been made in understanding how learning of such behaviors takes place in the mammalian brain. On one hand, the framework of reinforcement learning [1] provides a theory and algorithms for learning with sparse rewarding events. A particularly attractive formulation of reinforcement learning is temporal difference (TD) learning [2]. In the standard setting, this theory assumes that an agent moves between states in its environment by choosing appropriate actions in discrete time steps. Rewards are given in certain conjunctions of states and actions, and the agent's aim is to choose its actions so as to maximize the amount of reward it receives. Several algorithms have been developed to solve this standard formulation of the problem, and some of these have been used with spiking neural systems. These include REINFORCE [3], [4] and partially observable Markov decision processes [5], [6], in case the agent has incomplete knowledge of its state.
On the other hand, experiments show that dopamine, a neurotransmitter associated with pleasure, is released in the brain when reward, or a reward-predicting event, occurs [7]. Dopamine has been shown to modulate the induction of plasticity in timing non-specific protocols [8]–[11]. Dopamine has also recently been shown to modulate spike-timing-dependent plasticity (STDP), although the exact spike-timing and dopamine requirements for induction of long-term potentiation (LTP) and long-term depression (LTD) are still unclear [12]–[14].
A crucial problem in linking biological neural networks and reinforcement learning is that typical formulations of reinforcement learning rely on discrete descriptions of states, actions and time, while spiking neurons evolve naturally in continuous time and biologically plausible “time-steps” are difficult to envision. Earlier studies suggested that an external reset [15] or theta oscillations [16] might be involved, but no evidence exists to support this and it is not clear why evolution would favor slower decision steps over a continuous decision mechanism. Indeed biological decision making is often modeled by an integrative process in continuous time [17], where the actual decision is triggered when the integrated value reaches a threshold.
In this study, we propose a way to narrow the conceptual gap between reinforcement learning models and the family of spike-timing-dependent synaptic learning rules by using continuous representations of state, actions and time, and by deriving biologically plausible synaptic learning rules. More precisely, we use a variation of the Actor-Critic architecture [1], [18] for TD learning. Starting from the continuous TD formulation by Doya [19], we derive reward-modulated STDP learning rules which enable a network of model spiking neurons to efficiently solve navigation and motor control tasks, with continuous state, action and time representations. This can be seen as an extension of earlier works [20], [21] to continuous actions, continuous time and spiking neurons. We show that such a system has a performance on par with that of real animals and that it offers new insight into synaptic plasticity under the influence of neuromodulators such as dopamine.
How do animals learn to find their way through a maze? What kind of neural circuits underlie such learning and computation and what synaptic plasticity rules do they rely on? We address these questions by studying how a simulated animal (or agent) could solve a navigation task, akin to the Morris water-maze. Our agent has to navigate through a maze, looking for a (hidden) platform that triggers reward delivery and the end of the trial. We assume that our agent can rely on place cells [22] for a representation of its current position in the maze (Figure 1).
Temporal difference learning methods provide a theory explaining how an agent should interact with its environment to maximize the rewards it receives. TD learning is built on the formalism of Markov decision processes. In what follows, we reformulate the framework of Markov decision process in continuous time, state and action, before we turn to the actor-critic neural network and the learning rule we used to solve the maze task.
Let us consider a learning agent navigating through the maze. We can describe its position at time as , corresponding to a continuous version of the state in the standard reinforcement learning framework. The temporal evolution of the state is governed by the agent's action , according to(1)where describes the dynamics of the environment. Throughout this paper we use the dot notation to designate the derivative of a term with respect to time.
We model place cells as simple spiking processes (inhomogeneous Poisson, see Models) that fire only when the agent approaches their respective center. The centers are arranged on a grid, uniformly covering the surface of the maze.
Reward is dispensed to the agent in the form of a reward rate . A localized reward at a single position would correspond to the limit , where denotes the Dirac -function. However, since any realistic reward (e.g., a piece of chocolate or the hidden platform in the water-maze) has a finite extent, we prefer to work with a temporally extended reward. In our model, rewards are attributed based on spatially precise events, but their delivery is temporally extended (see Models). The agent is rewarded for reaching the goal platform and punished (negative reward) for running into walls.
The agent follows a policy which determines the probability that an action is taken in the state (2)The general aim of the agent is to find the policy that ensures the highest reward return in the long run.
Several algorithms have been proposed to solve the discrete version of the reinforcement problem problem described above, such as Q-Learning [23] or Sarsa [24]. Both of these use a representation of the future rewards in form of -values for each state-action pair. The -values are then used both to evaluate the current policy (evaluation problem) and to choose the next action (control problem). As we show in Models, -values lead to difficulties when one wishes to move to a continuous representation while preserving biological plausibility. Instead, here we use an approach dubbed “Actor-Critic” [1], [8], [21], where the agent is separated in two parts: the control problem is solved by an actor and the evaluation problem is solved by a critic (Figure 1).
The rest of the Results section is structured as follows. First we have a look at the TD formalism in continuous time. Next, we show how spiking neurons can implement a critic, to represent and learn the expected future rewards. Third, we discuss a spiking neuron actor, and how it can represent and learn a policy. Finally, simulation results show that our network successfully learns the simulated task.
The goal of a reinforcement learning agent is to maximize its future rewards. Following Doya [10], we define the continuous-time value function as(3)where the brackets represent the expectation over all future trajectories and future action choices , dependent on the policy . The parameter represents the reward discount time constant, analogous to the discount factor of discrete reinforcement learning. Its effect is to make rewards in the near future more attractive than distant ones. Typical values of for a task such as the water-maze task would be on the order of a few seconds. Eq. 3 represents the total quantity of discounted reward that an agent in position at time and following policy can expect. The policy should be chosen such that is maximized for all locations . Taking the derivative of Eq. 3 with respect to time yields the self-consistency equation [19](4)
Calculating requires knowledge of the reward function and of the environment dynamics (Eq 1). These are, however, unknown to the agent. Typically, the best an agent can do is to maintain a parametric estimator of the “true” value function . This estimator being imperfect, it is not guaranteed to satisfy Eq. 4. Instead, the temporal difference error is defined as the mismatch in the self-consistency,(5)This is analog to the discrete TD error [1], [19](6)where the reward discount factor plays a role similar to the reward discount time constant . More precisely, for short steps , [19].
An estimator can be said to be a good approximation to if the TD error is close to zero for all . This suggests a simple way to learn a value function estimator: by a gradient descent on the squared TD error in the following way(7)where is a learning rate parameter and is the set of parameters (synaptic weights) that control the estimator of the value function. This approach, dubbed residual gradient [19], [25], [26], yields a learning rule that is formally correct, but in our case suffers from a noise bias, as shown in Models.
Instead, we use a different learning rule, suggested for the discrete case by Sutton and Barto [1]. Translated in a continuous framework, the aim of their optimization approach is that the value function approximation should match the true value function . This is equivalent to minimizing an objective function(8)A gradient descent learning rule on yields(9)Of course, because is unknown, this is not a particularly useful learning rule. On the other hand, using Eq. 4, this becomes(10)where we merged into the learning rate without loss of generality. In the last step, we replaced the real value function derivative with its estimate, i.e., , and then used the definition of from Eq. 5.
The substitution of by in Eq. 10 is an approximation, and there is in general no guarantee that the two values are similar. However the form of the resulting learning rule suggests it goes in the direction of reducing the TD error . For example, if is positive at time , updating the parameters in the direction suggested by Eq. 10, will increase the value of , and thus decrease .
In [19], a heuristic shortcut was used to go directly from the residual gradient (Eq. 7) to Eq. 10. As noted by Doya [19], the form of the learning rule in Eq. 10 is a continuous version of the discrete [1], [27] with function approximation (here with ). This has been shown to converge with probability 1 [28], [29], even in the case of infinite (but countable) state space. This must be the case also for arbitrarily small time steps (such as the finite steps usually used in computer simulations of a continuous system [19]), and thus it seems reasonable to expect that the continuous version also converges under reasonable assumptions, even though to date no proof exists.
An important problem in reinforcement learning is the concept of temporal credit assignment, i.e., how to propagate information about rewards back in time. In the framework of TD learning, this means propagating the TD error at time so that the value function at earlier times is updated in consequence. The learning rule Eq. 10 does not by itself offer a solution to this problem, because the expression of explicitly refers only to and at time . Therefore does not convey information about other times and minimizing does not a priori affect values and . This is in contrast to the discrete version of the TD error (Eq. 6), where the expression of explicitly links to and thus the TD error is back-propagated during subsequent learning trials.
If, however, one assumes that the value function is continuous and continuously differentiable, changing the values of and implies changing the values of these functions in a finite vicinity of . This is in particular the case if one uses a parametric form for , in the form of a weighted mixture of smooth kernels (as we do here, see next section). Therefore, the conjunction of a function approximation of the value function in the form of a linear combination of smooth kernels ensures that the TD error is propagated in time in the continuous case, allowing the temporal credit assignment problem to be solved.
We now take the above derivation a step further by assuming that the value function estimation is performed by a spiking neuron with firing rate . A natural way of doing this is(11)where is the value corresponding to no spiking activity and is a scaling factor with units of [reward units]×s. A choice of enables negative values , despite the fact that the rate is always positive. We call this neuron a critic neuron, because its role is to maintain an estimate of the value function .
Several aspects should be discussed at this point. Firstly, since the value function in Eq. 11 must depend on the state of the agent, we must assume that the neuron receives some meaningful synaptic input about the state of the agent. In the following we make the assumption that this input is feed-forward from the place cells to the (spiking) critic neuron.
Secondly, while the value function is in theory a function only of the state at time , a spiking neuron implementation (such as the simplified model we use here, see Models) will reflect the recent past, in a manner determined by the shape of the excitatory postsynaptic potentials (EPSP) it receives. This is a limitation shared by all neural circuits processing sensory input with finite synaptic delays. In the rest of this study, we assume that the evolution of the state of the agent is slow compared to the width of an EPSP. In that limit, the firing rate of a critic neuron at time actually reflects the position of the agent at that time.
Thirdly, the firing rate of a single spike-firing neuron is itself a vague concept and multiple definitions are possible. Let's start from its spike train (where is the set of the neuron's spike times and is the Dirac delta, not to be confused with the TD signal). The expectation is a statistical average of the neuron's firing over many repetitions. It is the theoretically favored definition of the firing rate, but in practice it is not available in single trials in a biologically plausible setting. Instead, a common workaround is to use a temporal average, for example by filtering the spike train with a kernel (12)Essentially, this amounts to a trade-off between temporal accuracy and smoothness of the rate function, of which extreme cases are respectively the spike train (extreme temporal accuracy) and a simple spike count over a long time window with smooth borders (no temporal information, extreme smoothness). In choosing a kernel , it should hold that , so that each spike is counted once, and one often wishes the kernel to be causal (), so that the current firing rate is fully determined by past spike times and independent of future spikes.
Another common approximation for the firing rate of a neuron consists in replacing the statistical average by a population average, over many neurons encoding the same value. Provided they are statistically independent of each other (for example if the neurons are not directly connected), averaging their responses over a single trial is equivalent to averaging the responses of a single neuron over the same number of trials.
Here we combine temporal and population averaging, redefining the value function as an average firing rate of neurons(13)where the instantaneous firing rate of neuron is defined by Eq. 12, using its spike train and a kernel defined by(14)This kernel rises with a time constant and decays to 0 with time constant . One advantage of the definition of Eq. 12 is that the derivative of the firing rate of neuron with respect to time is simply(15)so that computing the derivative of the firing rate is simply a matter of filtering the spike train with the derivative of the kernel. This way, the TD error of Eq. 5 can be expressed as(16)where, again, denotes the spike train of neuron in the pool of critic neurons.
Suppose that feed-forward weights lead from a state-representation neuron to neuron in the population of critic neurons. Can the critic neurons learn to approximate the value function by changing the synaptic weights? An answer to this question is obtained by combining Eq. 10 with Eqs 13 and 16, leading to a weights update(17)where is the time course of an EPSP and is the spike train of the presynaptic neuron , restricted to the spikes posterior to the last spike time of postsynaptic neuron . For simplicity, we merged all constants into a new learning rate . A more formal derivation can be found in Models.
Let us now have a closer look at the shape of the learning rule suggested by Eq. 17. The effective learning rate is given by a parameter . The rest of the learning rule consists of a product of two terms. The first one is the TD error term , which is the same for all synapses , and can thus be considered as a global factor, possibly transmitted by one or more neuromodulators (Figure 1). This neuromodulator broadcasts information about inconsistency between the reward and the value function encoded by the population of critic neurons to all neurons in the network. The second term is synapse-specific and reflects the coincidence of EPSPs caused by presynaptic spikes of neuron with the postsynaptic spikes of neuron . The postsynaptic term is a consequence of the exponential non-linearity used in the neuron model (see Models). This coincidence, “Hebbian” term is in turn filtered through the kernel which corresponds to the effect of a postsynaptic spike on . It reflects the responsibility of the synapse in the recent value function. Together these two terms form a three-factor rule, where the pre- and postsynaptic activities combine with the global signal to modify synaptic strengths (Figure 2A, top). Because it has, roughly, the form of “TD error signalHebbian LTP”, we call this learning rule TD-LTP.
We would like to point out the similarity of the TD-LTP learning rule to a reward-modulated spike-timing-dependent plasticity rule we call R-STDP [6], [16], [30]–[32]. In R-STDP, the effects of classic STDP [33]–[36] are stored into an exponentially decaying, medium term (time constant ), synapse-specific memory, called an eligibility trace. This trace is only imprinted into the actual synaptic weights when a global, neuromodulatory success signal is sent to the synapses. In R-STDP, the neuromodulatory signal is the reward minus a baseline, i.e., . It was shown [32] that for R-STDP to maximize reward, the baseline must precisely match the mean (or expected) reward. In this sense, is a reward prediction error signal; a system to compute this signal is needed. Since the TD error is also a reward prediction error signal, it seems natural to use instead of . This turns the reward-modulated learning rule R-STDP into a TD error-modulated TD-STDP rule (Figure 2A, bottom). In this form, TD-STDP is very similar to TD-LTP. The major difference between the two is the influence of post-before-pre spike pairings on the learning rule: while these are ignored in TD-LTP, they cause a negative contribution to the coincidence detection in TD-STDP.
The filtering kernel , which was introduced to filter the spike trains into differentiable firing rates serves a role similar to the eligibility trace in R-STDP, and also in the discrete TD() [1]. As noted in the previous section, this is the consequence of the combination of a smooth parametric function approximation of the value function (each critic spike contributes a shape to ) and the form of the learning rule from Eq. 10. The filtering kernel is crucial to back-propagation of the TD error, and thus to the solving of the temporal credit assignment problem.
Having shown how spiking neurons can represent and learn the value function, we next test these results through simulations. However, in the actor-critic framework, the actor and the critic learn in collaboration, making it hard to disentangle the effects of learning in either of the two. To isolate learning by the critic and disregard potential problems of the actor, we temporarily sidestep this difficulty by using a forced action setup. We transform the water-maze into a linear track, and “clamp” the action choice to a value which leads the agent straight to the reward. In other words, the actor neurons are not simulated, see Figure 2B, and the agent simply “runs” to the goal. Upon reaching it at time , a reward is delivered and the trial ends.
Figure 2C shows the value function over color-coded trials (from blue to red) as learned by a critic using the learning rule we described above. On the first run (dark blue trace), the critic neurons are naive about the reward and therefore represent a (noisy version of a) zero value function. Upon reaching the goal, the TD error (Figure 2D) matches the reward time course, . According to the learning rule in Eq. 17, this causes strengthening of those synapses that underwent pre-post activity recently before the reward (with “recent” defined by the kernel). This is visible already at the second trial, when the value just before reward becomes positive.
In the next trials, this effect repeats, until the TD error vanishes. Suppose that, in a specific trials, reward starts at the time when the agent has reached the goal. According to the definition of the TD error, for all times the -value is self consistent only if — or equivalently . The gray dashed line in Figure 2C shows the time course of the theoretical value function; over many repetitions the colored traces, representing the value function in the different trials, move closer and closer to the theoretical value. The black line in Figure 2C represents the average value function over 20 late trials, after learning has converged: it nicely matches the theoretical value.
An interesting point that appears in Figure 2C is the clearly visible back-propagation of information about the reward expressed in the shape of the value function. In the first trials, the value function rises only for a short time just prior to the reward time. This causes, in the following trial, a TD error at earlier times. As trials proceed, synaptic weights corresponding to even earlier times increase. After trials in Figure 2C, the value function roughly matches the theoretical value just prior to , but not earlier. In subsequent trials, the point of mismatch is pushed back in time.
This back-propagation phenomenon is a signature of TD learning algorithms. Two things should be noted here. Firstly, the speed with which the back-propagation occurs is governed by the shape of the kernel in the Hebbian part of the learning rule. It plays a role equivalent to the eligibility trace in reinforcement learning: it “flags” a synapse after it underwent pre-before-post activity with a decaying trace, a trace that is only consolidated into a weight change when a global confirmation signal arrives. This “eligibility trace” role of is distinct from its original role in the term, where it is used to smooth the spiking activity of the critic neurons (Eq. 12). As such, one might be tempted to change the decay time constant of the term in the learning rule so as to control back-propagation speed, while keeping the “other” of the signal fixed. In separate simulations (not shown), we found that such an ad-hoc approach did not lead to a gain in learning performance.
Secondly, we know by construction that this back-propagation of the reward information is driven by the TD error signal . However, visual inspection of Figure 2D, which shows the traces corresponding to the experiment in Figure 2C, does not reveal any clear back-propagation of the TD error. For , a large peak mirroring the reward signal (gray dashed line) is visible in the early traces (blue lines) and recedes quickly as the value function correctly learns to expect the reward. For , the is dominated by fast noise, masking any back-propagation of the error signal, even though the fact that the value function is learned properly shows it is indeed present and effective. One might speculate that if a biological system was using such a TD error learning system with spiking neuron, and if an experimenter was to record a handful of critic neurons he would be at great pain to measure any significant TD error back-propagation. This is a possible explanation for the fact that no back-propagation signal has been observed in experiments.
We have already discussed the structural similarity of a TD-modulated version of the R-STDP rule [6], [30], [31] with TD-LTP. Simulations of the linear track experiment with the TD-STDP rule show that it behaves similarly to our learning rule (data not shown), i.e., the difference between the two rules (the post-before-pre part of the coincidence detection window, see Figure 2A) does not appear to play a crucial role in this case.
We have seen above that spiking neurons in the “critic” population can learn to represent the expected rewards. We next ask how a spiking neuron agent chooses its actions so as to maximize the reward.
In the classical description of reinforcement learning, actions, like states and time, are discrete. While discrete actions can occur, for example when a laboratory animal has to choose which lever to press, most motor actions, such as hand reaching or locomotion in space, are more naturally described by continuous variables. Even though an animal only has a finite number of neurons, neural coding schemes such as population vector coding [37] allow a discrete number of neurons to code for a continuum of actions.
We follow the population coding approach and define the actor as a group of spiking neurons (Figure 3A), each coding for a different direction of motion. Like the critic neurons, these actor neurons receive connections from place cells, representing the current position of the agent. The spike trains generated by these neurons are filtered to produce a smooth firing rate, which is then multiplied by each neuron's preferred direction (see Models for all calculation details). We finally sum these vectors to obtain the actual agent action at that particular time. To ensure a clear choice of actions, we use a -winner-take-all lateral connectivity scheme: each neuron excites the neurons with similar tuning and inhibits all other neurons (Figure 3B). We manually adjusted the connection strength so that there was always a single “bump” of neurons active. An example of the activity in the pool of actor neurons and the corresponding action readout over a (successful) trial is given in Figure 3C. The corresponding maze trajectory is shown in Figure 3D.
In reinforcement learning, a successful agent has to balance exploration of unvisited states and actions in the search for new rewards, and exploitation of previously successful strategies. In our network, the exploration/exploitation balance is the result of the bump dynamics. To see this, let us consider a naive agent, characterized by uniform connections from the place cells to the actor neurons. For this agent, the bump first forms at random and then drifts without preference in the action space. This corresponds to random action choices, or full exploration. After the agent has been rewarded for reaching the goal, synaptic weights linking particular place cells to a particular action will be strengthened. This will increase the probability that the bump forms for that action the next time over. Thus the action choice will become more deterministic, and the agent will exploit the knowledge it has acquired over previous trials.
Here, we propose to use the same learning rule for the actor neurons' synapses as for those of the critic neurons. The reason is the following. Let us look at the case where : the critic is signaling that the recent sequence of actions taken by the agent has caused an unexpected reward. This means that the association between the action neurons that have recently been active and the state neurons whose input they have received should be strengthened so that the same action is more likely to be taken again in the next occurrence of that state. In the contrary case of a negative reinforcement signal, the connectivity to recently active action neurons should be weakened so that recently taken action are less likely to be taken again, leaving the way to, hopefully, better alternatives. This is similar to the way in which the synapses from the state input to the critic neurons should be strengthened or weakened, depending on their pre- and postsynaptic activities. This suggests that the action neurons should use the same synaptic learning rule as the one in Eq. 17, with now denoting the activity of the action neurons, but the signal still driven by the critic activity. This is biologically plausible and consistent with our assumption that is communicated by a neuromodulator, which broadcasts information over a large fraction of the brain.
There are two critical effects of our -winner-take-all lateral connectivity scheme. Firstly, it ensures that only neurons coding for similar actions can be active at the same time. Because of the Hebbian part of the learning rule, this means that only those which are directly responsible for the action choice are subject to reinforcement, positive or negative. Secondly, by forcing the activity of the action neurons to take the shape of a group of similarly tuned neurons, it effectively causes generalization across actions: neurons coding for actions similar to the one chosen will also be active, and thus will also be given credit for the outcome of the action [16]. This is similar to the way the actor learns in non-neural actor-critic algorithms [18], [19], where only actions actually taken are credited by the learning rule. Thus, although an infinite number of actions are possible at each position, the agent does not have to explore every single one of them (an infinitely long task!) to learn the right strategy.
The fact that both the actor and the critic use the same learning rule is in contrast with the original formulation of the actor-critic network of Barto et al. [18], where the critic learning rule is of the form “TD error×presynaptic activity”. As discussed above, the “TD error×Hebbian LTP” form of the critic learning rule Eq. 17 used here is a result of the exponential non-linearity used in the neuron model. Using the same learning rule for the critic and the actor has the interesting property that a single biological plasticity mechanism has to be postulated to explain learning in both structures.
In the Morris water-maze, a rat or a mouse swims in an opaque-water pool, in search of a submerged platform. It is assumed that the animal is mildly inconvenienced by the water, and is actively seeking refuge on the platform, the reaching of which it experiences as a positive (rewarding) event. In our simulated navigation task, the learning agent (modeling the animal) is randomly placed at one out of four possible starting locations and moves in the two-dimensional space representing the pool (Figure 4A). Its goal is to reach the goal area ( of the total area) which triggers the delivery of a reward signal and the end of the trial. Because the attractor dynamics in the pool of actor neurons make it natural for the agent to follow a straight line, we made the problem harder by surrounding the goal with a U-shaped obstacle so that from three out of four starting positions, the agent has to turn at least once to reach the target. Obstacles in the maze cause punishment (negative reward) when touched. Similar to what is customary in animal experiments, unsuccessful trials were interrupted (without reward delivery) when they exceeded a maximum duration .
During a trial, the synapses continually update their efficacies according to the learning rule, Eq. 17. When a trial ends, we simulate the animal being picked up from the pool by suppressing all place cell activity. This results in a quick fading away of all neural activity, causing the filtered Hebbian term in the learning rule to vanish and learning to effectively stop. After an inter-trial interval of 3s, the agent was positioned in a new random position, starting a new trial.
Figure 4B shows color-coded trajectories for a typical simulated agent. The naive agent spends most of the early trials (blue traces) learning to avoid walls and obstacles. The agent then encounters the goal, first at random through exploration, then repeatedly through reinforcement of the successful trajectories. Later trials (yellow to red traces) show that the agent mostly exploits stereotypical trajectories it has learned to reach the target.
We can get interesting insight into what was learned during the trials shown in Figure 4B by examining the weight of the synapses from the place cells to actor or critic neurons. Figure 4C shows the input strength to critic neurons as a color map for every possible position of the agent. This is in effect a “value map”: the value the agent attributes to each position in the maze. In the same graph, the synaptic weights to the actor neurons are illustrated by a vector field representing a “policy preference map”. It is only a preference map, not a real policy map because the input from the place cells (represented by the arrows) compete with the lateral dynamics of the actor network, which is history-dependent (not represented).
The value and policy maps that were learned are experience-dependent and unique to each agent: the agent shown in Figure 4B and C first discovered how to reach the target from the “north” (N) starting position. It then discovered how to get to the N position from starting positions E and W, and finally to get to W from S. It has not however discovered the way from S to E. For that reason the value it attributes to the SE quarter is lower than to the symmetrically equivalent quarter SW. Similarly the policy in the SE quarter is essentially undefined, whereas the policy in the SW quarter clearly points in the correct direction.
Figure 4D shows the distribution of latency – the time it takes to reach the goal – as a function of trials, for 100 agents. Trials of naive agents end after an average of (trials were interrupted after ). This value quickly decreases for agents using the TD-LTP learning rule (green), as they learn to reach the reward reliably in about trials.
We previously remarked that the TD-LTP rule of Eq. 17 is similar to TD-STDP, the TD-modulated version of the R-STDP rule [6], [30], [31], at least in form. To see whether they are also similar in effect, in our context, we simulated agents using the TD-STDP learning rule (for both critic and actor synapses). The blue line in Figure 4D show that the performance was only slightly worse than that of the TD-LTP rule, confirming our finding on the linear track that both rules are functionally equivalent.
Policy gradient methods [5] follow a very different approach to reinforcement learning to TD methods. A policy gradient method for spiking neurons is R-max [4], [6], [32], [38], [39]. In short, R-max works by calculating the covariance between Hebbian pre-before-post activity and reward. Because this calculation relies on averaging those values over many trials, R-max is an inherently slow rule, typically learning on hundreds or thousands of trials. One would therefore expect that it can't match the speed of learning of TD-LTP or TD-STDP. Another difference of R-max with the other learning rules studied is that it does not need a critic [32]. Therefore we simulated an agent using R-max that only had an actor, and replaced the TD-signal by the reward, . The red line of Figure 4 show that, as expected, R-max agents learn much slower than previously simulated agent, if at all: learning is actually so slow, consistent with the usual timescales for that learning rule, that it can't be seen in the graph because this would require much longer simulations.
One might object that using the R-max rule without a critic is unfair, and that it might benefit from a translation into a R-max rule with R = TD, by replacing the reward term by the error, as we did for R-STDP. But this overlooks two points. Firstly, such a “TD-max” rule could not be used to learn the critic: by construction, it would tend to maximize the TD error, which is the opposite of what the critic has to achieve. Secondly, even if one were to use a different rule (e.g. TD-LTP) to learn the critic, this would not solve the slow timescale problem. We experimented with agents using a “TD-max” actor while keeping TD-LTP for the critic, but could not find notable improvement over agents with an R-max actor (data not shown).
Having shown that our actor-critic system could learn a navigation task, we now address a task that requires higher temporal accuracy and higher dimensional state spaces. We focus on the acrobot swing-up task, a standard control task in the reinforcement control literature. Here, the goal is to lift the outermost tip of a double pendulum under the influence of gravity above a certain level, using only a weak torque at the joint (Figure 5A). The problem is similar to that of a gymnast hanging below an horizontal bar: her hands rotate freely around the bar, and the only way to induce motion is by twist of her hips. While a strong athlete might be able to lift her legs above her head in a single motion, our acrobot is too weak to manage this. Instead, the successful strategy consists in moving the legs back and forth to start a swinging motion, building up energy, until the legs reach the sufficient height.
The position of the acrobot is fully described by two angles, and (see Figure 5A). However, the swinging motion required to solve the task means that even in the same angular position, different actions (torque) might be required, depending on whether the system is currently swinging to the left or to the right. For this reason, the angular velocities and are also important variables. Together, these four variables represent the state of the agent, the four-dimensional equivalent of the x–y coordinates in the navigation task. Just as in the water-maze case, place cells firing rates were tuned to specific points in the 4-dimensional space.
Again similar to the maze navigation, the choice of the action (in this case the torque exerted on the pendulum joint) is encoded by the population vector of the actor neurons. The only two differences to the actor in the water-maze are that (i) the action is described by a single scalar and (ii) the action neuron attractor network is not on a closed ring anymore, but rather an open segment, encoding torques in the range .
Several factors make the acrobot task harder than the water-maze navigation task. First, the state space is larger, with four dimensions against two. Because the number of place cells we use to represent the state of the agent grows exponentially with the dimension of the state space, this is a critical point. A larger number of place cells means that each is visited less often by the agent, making learning slower. At even higher dimensions, at some point the place cells approach is expected to fail. However, we want to show that it can still succeed in four dimensions.
A second difficulty arises from the faster dynamics of the acrobot system with respect to the neural network dynamics. Although in simulations we are in full control of the timescales of both the navigation and acrobot dynamics, we wish to keep them in range with what might naturally occur for animals. As such the acrobot model requires fast control, with precision on the order of 100ms. Finally, the acrobot exhibits complex dynamics, chaotic in the control-less case. Whereas the optimal strategy for the navigation task consists in choosing an action (i.e., a direction) and sticking to it, solving the acrobot task requires precisely timed actions to successfully swing the pendulum out of its gravity well.
In spite of these difficulties, our actor-critic network using the TD-LTP learning rule is able to solve the acrobot task, as Figure 5B shows. We compared the performance to a near-optimal trajectory [40]: although our agents are typically twice as slow to reach the goal, they still learn reasonable solutions to the problem. Because the agents start with mildly random initial synaptic weights (see Models) and are subject to stochasticity, their history, and thus their performance, vary; the best agents have performance approaching that of the optimal controller (blue trace in Figure 5B).
We next try our spiking neuron actor-critic network on a harder control task, the cartpole swing-up problem [19]. This is a more difficult extension of cartpole balancing, a standard task in machine learning [18], [41]. Here, a pole is attached to a wheeled cart, itself free to move on a rail of limited length. The pole can swing freely around its axle (it doesn't collide with the rail). The goal is to swing the pole upright, and, ideally, to keep it in that position for as long as possible. The only control that can be exerted on the system is a force on the cart (Figure 6A). As in the acrobot task, four variables are needed to describe the system: the position of the cart, its velocity , and the angle and angular velocity of the pole. We define a successful trial as a trial where the pole was kept upright () for more than 10 s, out of a maximum trial length of . A trial is interrupted and the agent is punished for either hitting the edges of the rail () or “over-rotating” (). Agents are rewarded (or punished) with a reward rate .
The cartpole task is significantly harder than the acrobot task and the navigation task. In the two latter ones, the agent only has to reach a certain region of the state space (the platform in the maze, or a certain height for the acrobot) to be rewarded and to cause the end of the trial. In contrast, the agent controlling the cartpole system must reach the region of the state space corresponding to the pole being upright (an unstable manifold), and must learn to fight adverse dynamics to stay in that position.
For this reason learning to successfully control the cartpole system takes a large number of trials. In Figure 6B, we show the number of successful trials as a function of trial number. It takes the “median agent” (black line) on the order of 3500 trials to achieve 100 successful trials. This is slightly worse but on the same order of magnitude as the (non-neural) actor-critic of [19], which needs trials to reach that performance.
The evolution of average reward by trial (Figure 6C) shows that agents start with a phase of relatively quick progression (inset), corresponding to the agents learning to avoid the immediate hazard of running into the edges of the rail. This is followed by slower learning, as the agents learn to swing and control the pole better and better. To ease the long learning process we resorted to variable learning rates for both the actor and critic on the cartpole task: we used the average recent rewards obtained to choose the learning rate (see Models). More precisely, when the reward was low, agents used a large learning rate, but when performance improved, the agents were able to learn finer control strategies with a small learning rate. Eventually agents manage fine control and easily recover from unstable situations (Figure 6D). Detailed analysis of the simulation results showed that our learning agents suffered from noise in the actor part of the network, hampering the fine control needed to keep the pole upright. For example, the agent in Figure 6D has learned how to recover from a falling pole (top and middle plots) but will occasionally take more time than strictly necessary to bring the pole to a vertical standstill (bottom plot). The additional spike firing noise in our spiking neuron implementation could potentially explain the performance difference with the actor-critic in [19].
In this paper, we studied reward-modulated spike-timing-dependent learning rules, and the neural networks in which they can be used. We derived a spike-timing-dependent learning rule for an actor-critic network and showed that it can solve a water-maze type learning task, as well as acrobot and cartpole swing-up tasks that both require mastering a difficult control problem. The derived learning rule is of high biological plausibility and resembles the family of R-STDP rules previously studied.
Throughout this study we tried to keep a balance between model simplicity and biological plausibility. Our network model is meant to be as simple and general as possible for an actor-critic architecture. We don't want to map it to a particular brain structure, but candidate mappings have already been proposed [42], [43]. Although they do not describe particular brain areas, most components of our network resemble brain structures. Our place cells are very close to – and indeed inspired by – hippocampal place cells [22]. Here we assume that the information encoded in place cells is available to the rest of the brain. Actor neurons, tuned to a particular action and linked to the animal level action through population vector coding are similar to classical models of motor or pre-motor cortices [37]. So-called “ramp” neurons of the ventral striatum have long been regarded as plausible candidates for critic neurons: their ramp activity in the approach of rewards matches that of the theoretical critic. If one compares experimental data (for example Figure 7A, adapted from van der Meer and Redish [44]) and the activity of a typical critic neuron (Figure 7B), the resemblance is striking. The prime neuromodulatory candidate to transmit the global TD error signal to the synapses is dopamine: dopaminergic neurons have long been known to exhibit TD-like activity patterns [7], [45].
A problem of representing the TD error by dopamine concentration is that while the theoretically defined error signal can be positive as well as negative, dopamine concentration values [DA] are naturally bound to positive values [46]. This could be circumvented by positing a non-linear relation between the two values (e.g., ) at the price of sensitivity changes over the range. Even a simpler, piecewise linear scheme (where is the baseline dopamine concentration) would be sufficient, because learning works as long as the sign of the TD error is correct.
Another possibility would be for the TD error to be carried in the positive range by dopamine, and in the negative range by some other neuromodulator. Serotonin, which appears to play a role similar to negative TD errors in reversal learning [47], is a candidate. On the other hand this role of serotonin is seriously challenged by experimental recordings of the activity of dorsal raphe serotonin neurons during learning tasks [48], [49], which fail to show activity patterns corresponding to an inverse TD signal.
One of the aspects of our actor-critic model that was not implemented directly by spiking neurons but algorithmically, is the computation of the TD signal which depends on the reward, the value function and its derivative. In our model, this computation is crucial to the functioning of the whole. Addition and subtraction of the reward and the value function could be done through concurrent excitatory and inhibitory input onto a group of neurons. Similarly, the derivative of the value function could be done by direct excitation by a signal and delayed (for example by a an extra synapse) inhibition by the same signal (see example in Figure 7C). It remains to be seen whether such a circuit can effectively be used to compute a useful TD error. At any rate, connections from the the ventral striatum (putative critic) to the substantia nigra pars compacta (putative TD signal sender) show many excitatory and inhibitory pathways, in particular through the globus pallidus, which could have precisely this function [50].
A crucial limitation of our approach is that we rely on relatively low-dimensional state and action representations. Because both use similar tuning/place cells representations, the number of neurons to represent these spaces has to grow exponentially with the number of dimensions, an example of the curse of dimensionality. While we show that we can still successfully solve problems with four-dimensional state description, this approach is bound to fail sooner or later, as dimensionality increases. Instead, the solution probably lies in “smart” pre-processing of the state space, to delineate useful and reward-relevant low dimensional manifolds on which place cells could be tuned. Indeed, the representation by place cells can be learned from visual input with thousands of “retinal” pixels, using standard unsupervised Hebbian learning rules [20], [51], [52].
Moreover, TD-LTP is derived with the assumption of sparse neural coding, with neurons having narrow tuning curves. This is in contrast to covariance-based learning rules [53], such as R-max [4], [6], [38], [39] which can, in theory, work with any coding scheme, albeit at the price of learning orders of magnitude slower.
Although a number of experimental studies exist [11]–[14], [54] targeting the relation between STDP and dopamine neuromodulation, one is at pain to draw precise conclusions as to how these two mechanism interplay in the brain. As such, it is hard to extract a precise learning rule from the experimental data. On the other hand, we can examine our TD-LTP learning rule in the light of experimental findings and see whether they match, i.e., whether a biological synapse implementing TD-LTP would produce the observed results.
Experiments combining various forms of dopamine or dopamine receptor manipulation with high-frequency stimulation protocols at the cortico-striatal synapses provide evidence of an interaction between dopamine and synaptic plasticity [8]–[11]. While these experiments are too coarse to resolve the spike-timing dependence, they form a picture of the dopamine dependence: it appears that at high concentration the effect of dopamine paired with high-frequency stimulation is the induction of long-term potentiation (LTP), while at lower concentrations, long-term depression (LTD) is observed. At a middle “baseline” concentration, no change is observed. This picture is consistent with TD-LTP or TD-STDP if one assumes a relation between the dopamine concentration and the TD error.
The major difference between TD-LTP and TD-STDP is the behavior of the rule on post-before-pre spike pairings. While TD-LTP ignores these, TD-STDP causes LTD (resp. LTP) for positive (resp. negative) neuromodulation. Importantly this doesn't seem to play a large role for the learning capability of the rule, i.e., the pre-before-post is the only crucial part. This is interesting in the light of the study by Zhang et al. [13] on hippocampal synapses, that finds that extracellular dopamine puffs reverse the post-before-pre side of the learning window, while strengthening the pre-before-post side. This is compatible with the fact that polarity of the post-before-pre side of the learning window is not crucial to reward-based learning, and might serve another function.
One result predicted by both TD-LTP and TD-STDP and that has not, to our knowledge, been observed experimentally, is the sign reversal of the pre-before-post under negative reward-prediction-error signals. This could be a result of the experimental challenges required to lower dopamine concentrations without reaching pathological levels of dopamine depression. However high-frequency stimulation-based experiments show that a reversal of the global polarity of long-term plasticity indeed happens [8], [11]. Moreover, a study by Seol et al. [54] of STDP induction protocols under different (unfortunately not dopaminergic) neuromodulators shows that both sides of the STDP learning window can be altered in both polarity and strength. This shows that a sign change of the effect of the pre-then-post spike-pairings is at least within reach of the synaptic molecular machinery.
Another prediction that stems from the present work is the existence of eligibility traces, closing the temporal gap between the fast time requirements of STDP and delayed rewards. The concept of eligibility traces is well explored in reinforcement learning [1], [5], [55], [56], and has previously been proposed for reward-modulated STDP rules [6], [30]. Although our derivation of TD-LTP reaches an eligibility trace by a different path (filtering of the spike train signal, rather than explicitly solving the temporal credit assignment problem), the result is functionally the same. In particular, the time scales of the eligibility traces we propose, on the order of hundreds of milliseconds, are of the same magnitude as those proposed in models of reward-modulated STDP [6], [30]. Direct experimental evidence of eligibility traces still lacks, but they are reminiscent of the synaptic tagging mechanism [57]. Mathematical models of tagging [58], using molecular cascades with varying timescales, provide an example of how eligibility traces could be implemented physiologically.
One interesting result of our study, is the fact that although our TD signal properly “teaches” the critic neurons the value function and back-propagates the reward information to more distant points, it is difficult to see the back-propagation in the time course of the TD signal itself. The reason for this is that the signal is drowned in rapid fluctuations. If one were to record a single neuron representing the TD error, it would probably be impossible to reconstruct the noiseless signal, except with an extremely high number of repetitions under the same conditions. This might be an explanation for the fact that the studies by Schultz and colleagues (e.g., [45]) repeatedly fail to show back-propagation of the TD error, even though dopamine neurons seem to encode such a signal.
In this study, TD-STDP (and TD-LTP) is used in a “gated-Hebbian” way: a synapse between A and B should be potentiated if it witnessed pre-before-post pairings and the TD signal following later is positive. This is fundamentally different from the role of the reward-modulated version of that learning rule (R-STDP) in [32], where it is used to do covariance-based learning: a synapse between A and B should be potentiated if it witnesses positive correlation between pre-before-post pairings and a success signal, on average. One consequence of this is the timescale of learning: while TD-based learning takes tens of trials, covariance based learning typically requires hundreds or thousands of trials. The other side of the coin is that covariance-based learning is independent of the neural coding scheme, while TD-based learning requires neural tuning curves to map the relevant features prior to learning. The fact that the mathematical structure of the learning rule (i.e., a three-factor rule where the third factor “modulates” the effect of pre-post coincidences [59]) is the same in both cases is remarkable, and one can see the advantage that the brain might have had to evolve such a multifunctional tool — a sort of “Swiss army knife” of synaptic plasticity.
For the actor and critic neurons we simulated a simplified spike response model (, [60]). This model is a stochastic variant of the leaky integrate-and-fire neuron, with the membrane potential of neuron of given by(18)where is the efficacy of the synapse from neuron to neuron , is the set of firing times of neuron , is the membrane time constant, scales the refractory effect, is the Heaviside step function and is the last spike of neuron prior to . The EPSP is described by the time course(19)where is the synaptic rise time and is a scaling constant, and is the membrane time constant, as in Eq. 18. Given the membrane potential , spike firing in the is an inhomogeneous Poisson process: at every moment the neuron has a probability of emitting a spike, according to an instantaneous firing rate(20)where , and are constants consistent with experimental values [61]. In the limit , the becomes a deterministic leaky integrate-and-fire neuron.
The Morris water-maze pool is modeled by a two-dimensional plane delimited by a square wall. The position of the agent on the plane obeys(21)When the agent is within boundaries it moves with speed , as defined by the actor neurons' activity (Eq.29). Whenever the agent encounters a wall, it instantly “bounces” back a distance along unitary vector , which points inward, perpendicular to the obstacle surface. Every “bumping” against a wall is accompanied by a punishing, negative reward delivery (see reward delivery dynamics below).
We used two variants of the navigation task. The linear track is a narrow rectangle of size centered around the origin, featuring a single starting position in and a wide goal area () on the opposite side. Because the goal of this setup is to study critic learning, the action is clamped to a fixed value , so that the agent runs toward the goal at a fixed speed.
The second variant is the navigation maze with obstacle. It consists of a square area of size centered around the origin, with four starting positions at . The goal area is a circle of radius centered in the middle of the maze. The goal is surrounded on three sides by a U-shaped obstacle (width of each segment: 2, length: 10).
In both variants, place cells centers are disposed on a grid (blue dots on Figure 1), with spacing coinciding with the width of the place fields. The outermost centers lie a distance outside the maze boundaries. This ensures a smooth coverage of the whole state space. In the case of the maze, the place cell grid consists of centers. For the linear track setup, the grid has centers.
Trials start with the agent's position being randomly chosen from one out of four possible starting positions. The place cells, indexed by , are inhomogeneous Poisson processes. After a trial starts, the place cells' instantaneous firing rates are updated to(22)where is a constant regulating the activity of the place cells, is the place cells separation distance and the are the place cells centers. The presynaptic activity in the place cells generates activity in the post-synaptic neurons of the critic and the actor with a small delay caused by the rise time of EPSPs.
The value function is calculated according to Eqs 12 and 13, with parameters and . Because is delayed by the rise time of the kernel, at the start of a trial the TD error is subject to large, boundary effect transients. To cancel these artifacts, we clamp the TD error to , for the first of each trial. We use a reward discount time constant .
The goal of the agent is to reach the circular area which represents the submerged platform of the water-maze. When the agent reaches this platform, a positive reward is delivered, the trial ends and the agent is put in a so-called “neutral state”, which models the removal of the animal from the experiment area. The effects of this is (i) the place cells corresponding to the maze become silent, presumably replaced by other (not modeled) place cells, and (ii) the expectation of the animal becomes neutral, and therefore its value function goes to zero. So at the end of a trial, we turn off place cell activity (), and the value function is no longer given by Eq. 13, but decays exponentially to 0 with time constant from its value at the time of the end of the trial. Importantly, synaptic plasticity continues after the end of the trial, so that the effect of affects the synaptic weight even though its delivery takes place in the neutral state. Additionally, a trial can end without the platform being reached: if a trial exceeds the time limit , it is declared a failed trial, and interrupted with the agent put in the neutral state, just as in the successful case, but without reward being delivered.
According to Eq. 3, rewards are given to the agent as a reward rate. This reflects the fact that “natural” rewards, and reward consumption, are spread over time, rather than point-like events. So we translate absolute rewards () to a reward rate (), calculated as the difference of two decaying “traces” obeying dynamics(23)i.e.,(24)At most times, the reward is close to 0. Reward is delivered only when some event (goal reached or collision against an obstacle) occurs. The delivery of a reward happens through instantaneous update of the traces(25)The resulting effect is a subsequent positive excursion of , with rise time and fall time , which, integrated over time, amounts to .
In the acrobot task, the position of the pendulum is described by two angles: is the angle between the first segment of the pendulum and the vertical, and is the angle between the second segment and an imaginary prolongation of the first (Figure 5A). When , the pendulum hangs down. Critical to solving the task are also the angular velocities and . As in the maze navigation case, place cells tuned to specific centers are used to represent the state of the acrobot. We transform the angular velocities , . This allows a fine resolution over small velocities, while maintaining a representation of higher velocities with a small number of place cells. The state is represented by the four variables .
The place cells centers are disposed on a 4-dimensional grid defined by indexes , such that with(26)This yields a total of centers. The activity of a place cell with center is defined by(27)where is a function returning the difference between two angles modulo in the range and the place cell widths to correspond to the grid spacing as in Eq. 26.
The acrobot dynamics obeys the following equations [1]:Here, , , and are convenience variables, is the torque applied to the joint, are the lengths of the segments, of mass , with moments of inertia and lengths to the centers of mass , under the influence of gravity . All dimensions except time are unit-less.
The goal is for the tip of the acrobot to reach a height over the axis, i.e., fulfill the condition . Once this happens, or the maximum trial time is reached, the trial ends. To entice the acrobot to do something, we give an ongoing punishment to the agent for not reaching the reward, to be compared with the reward received at the goal. As in the water-maze case, we use a reward discount time constant .
Due to the larger number of place cells, we use less critic and actor neurons than in the maze case, respectively and , to reduce the number of synapses and the computational load.
To compare the performance of our agent against an “optimal” strategy, we use the direct search method [40]. The main idea behind the method is to search for the sequence of action that will maximize the system's total energy, with knowledge of the acrobot dynamics. To make the search computationally tractable, a few simplifications are made: actions are limited to the alternative , actions are only taken in steps of 100 ms, only a window of the next 10 steps is considered at a time, and the number of action switch in each window is limited to 2. Thus only 55 action sequences have to be examined, and the sequence that maximizes the total energy reached over the window, or reaches the goal height the sooner, is selected. The first action in that sequence is chosen as the action for the next step and the whole procedure is repeated with the window shifted by one step. The goal height reaching latency found with this method was 7.66s (red line in Figure 5B).
The position of the cartpole system is described by the cart position , the cart velocity , the angle of the pole with the vertical ( corresponds to the pole pointing upwards) and the angular velocity ; these form the state vector . Similar to the acrobot, the place cells for the cartpole problem are regularly disposed on a four-dimensional grid of cells. The location of a place cell with index is at location with(28)The activity of a place cell is defined in a way analog to Eq. 27. The variance of the gaussian place fields is diagonal , where corresponds to the grid spacing in dimension .
The dynamics of the cartpole are [62]:Here a = v is the acceleration of the cart, is half the pole's length, and are coefficients of friction of the cart on the track and of pole rotation respectively. The cart, with mass , and the pole, with mass , are subject to the acceleration of gravity . As in the acrobot case, all dimensions except time are unit-less.
Following [19], the agent is rewarded continuously depending on the current height of the pole with , and the reward discount time constant is set to . If the cart runs off its rail () or over-rotates () the trial is ended and a negative reward is given. A trial ends without reward after . When a new trial starts, the position of the system is initialized with a random and .
In population vector coding, each actor neuron “votes” for its preferred action in the action space, by firing an action potential. An action vector is obtained by averaging the product of the instantaneous firing rate (see Eq. 12) and the action vector of each neuron, i.e.(29)where is defined as(30)with filterwith and being filtering time constants. The term in Eq. 29 is a normalization term. In the case of the navigation task (two-dimensional action), it is equal to the number of actor neurons, . In the cases of the acrobot and the cartpole task (scalar action), .
We enforce a N-winner-takes-all mechanism on the action neurons by imposing “lateral” connectivity between the action neurons: action neurons coding for similar actions excite each other, while they inhibit the neurons coding for dissimilar actions. The synaptic weight between two action neurons and is(32)where is a lateral connectivity function. This is zero for , peaks for and monotonously decreases towards 0 as and diverge. is a normalization constant. The parameters and regulating the recurrent connections were manually tuned: the lateral connectivity has to be strong enough so that there is always exactly one “bump” of similarly tuned neurons active whenever the action neurons receive some excitation from the place cells, but not so strong that it completely dominates the feed-forward input from the place cells.
The preferred vectors of the action neurons and the function are dependent on the learning task. In the case of the maze navigation task, the preferred action vectors are where is a constant representing the agent velocity per rate unit and , for . The function was chosen as(33)with .
In the case of the acrobot and cartpole tasks, the action vectors are . For the acrobot represents the maximum torque that the agent can exert and for the cartpole task is the maximum force on the cart. The lateral connectivity function in both cases was chosen as(34)with . Additionally, we algorithmically constrain the torque exerted by the agent to the domain . This models the limited strength of the agent's “muscles”.
In R-STDP [6], [30]–[32], the effects of classic STDP are modulated by a neuromodulatory signal , where is a constant baseline. We transformed the reward-modulated R-STDP into the TD-modulated rule TD-STDP by replacing the with . The TD-STDP rule can be written as(35)where the STDP learning window isThe eligibility trace kernel is the result of an exponential decay, i.e., , with time constant . The positive constants and govern the size of the pre-before-post and post-before-pre parts of the learning window respectively, and the time constants and determine their timing requirement.
R-max [4], [6], [32], [38] is a reward-modulated learning rule derived from policy gradient principles [5]. It can be written as(36)where is the instantaneous firing rate of neuron , as defined in Eq. 20.
Initial values of the synaptic weights to both critic and actor were randomly drawn from a normal distribution with mean and standard deviation . These values ensured an initial value function and reasonable action neuron activity prior to learning.
For all learning rules, synaptic weights were algorithmically constrained to the range , to avoid negative or runaway weights. Learning rate values were manually adjusted (one value for actor and another one for critic synapses) to the value that yielded the best performance (as measured by the number of trials completed in 2.000s of simulated time). These values for the navigation and acrobot tasks are printed in Table 1. For the cartpole task, somewhat faster learning was achieved by using a variable learning rate(37)for the critic, where is a running average of past reward rates , computed by filtering with an exponential window with time constant 50s. The actor learning rate was .
All simulations were ran using Euler's method with time-step , except for the acrobot and cartpole dynamics, simulated using 4th order Runge-Kutta with .
In this section we calculate the term , needed to derive Eq. 17. Using Eqs 12–13, and focusing on the synaptic weight from to , we find(38)where we used the fact that ρi′(t) is independent of for . The derivative of the spike train is ill-defined: in our stochastic neuron model, the spike train itself is independent of the synaptic weights. It is only the probability of the spike train actually being emitted by the neuron that depends on the weights. Therefore we replace with , the expected value of the spike train conditional on the input . This yields(39)where the sum is over all possible spike trains and is the probability density of the spike train being equal to . The probability density of that spike train , lasting from to , being produced by an SRM0 neuron receiving inputs is [38](40)where is the membrane potential (Eq. 18) and we have used Eq. 20. Combining Eqs 39 and 40 yields(41)The integration reflects the fact that the probability of a spike being emitted by the neuron at time is dependent not only on recent presynaptic spikes, but also on the time of the last spike of neuron , which in turn depends on its whole history.
It is not clear that, in our context, this history dependence is a desirable outcome. Two devices already take the spike train history into account. Firstly, the definition of the value function in the TD framework is conditional only on the current state, and not the long-term history. (This stems from the Markov decision process at the root of TD.) Secondly, the filtering of the spike train by already ensures that the short-term history is remembered, making the integral over the history redundant.
For these reasons, we choose to neglect the neuron's history, and to perform the following substitution(42)i.e., we take the last spike time of neuron as given, and we ask how does the mean spiking at time vary as a function of the synaptic weight . Therefore we have(43)where we have used the definition of the neuron's firing rate, Eq. 20, and is the Dirac distribution. Using Eqs 18 yields(44)where is the spike train of neuron culled to times posterior to the spikes of neuron , i.e., , with denoting the Heaviside step function. Wrapping up the steps from Eqs 38 and 42–44, we finally have(45)
In the Results section we derive a learning rule starting from Eq. 10. We also suggest that starting from a gradient descent on the squared TD error (Eq.17) should yield a valid learning rule. Here we derive such a learning rule. Combining Eq. 10, the definition of the TD error (Eq. 5) and the result of the previous section (Eq. 45), we find(46)where is the spike train of presynaptic neuron . This learning rule has the same general form as the TD-LTP rule (Eq. 17): a “Hebbian” pre-before-post coincidence term is first temporally filtered, and then multiplied by the TD error with a term (Figure 8A). The difference lies in the extra in the filter, which comes from a term. As Figure 8 suggests, the term largely dominates over . This is the consequence of our choice of a long discount time constant () with a short () kernel.
Here we show, both analytically and in simulations, that the squared TD gradient learning rule of Eq. 46 suffers from a noise bias problem. This arises from the noise in the individual neurons estimating the value function, and is serious enough to prevent learning. To see this, we start by decomposing the spike train of a neuron into a mean and a noise term, i.e.(47)where we have defined , with the brackets denoting expectation, i.e., averaging over all possible outcomes of critic neurons activity conditioned on the presynaptic neural activity . With this definition, we can rewrite Eq. 46 as(48)where the error has been spelled out explicitly (Eqs 5, 13 and 12). Eq. 48 suggests that quadratic terms in the noise might play a role in the learning rule. Indeed, distributivity and use of the facts and for gives(49)Here we have defined the autocorrelation of the noise terms , as well as for brevity.
The first term in the right-hand side of Eq. 49 is analog to Eq. 46, with replacing , and replacing . In effect this is a “mean” version of the learning rule: this is what one would get by replacing the stochastic spiking neurons in the model by analog, noiseless units with a similar exponential activation function.
The second term arises from the correlation of neuron noise in the TD term and the Hebbian component of the learning rule. This term is a function of the autocorrelation function of the postsynaptic neuron. This carries only indirect information about the postsynaptic firing (and thus the current value function ) and no information about the reward . For this reason, we conjecture that this second element is a potentially problematic term, which we refer to as the “nuisance” term. This hypothesis is confirmed by linear track simulations using the learning rule Eq. 46, shown in Figure 8B. These indicate that the learning rule is unable to learn the task, contrary to TD-LTP (Figure 8C, same as Figure 2B). More precisely, the value functions learned by the squared TD gradient rule suffer from a negative “drag” term.
We next try to identify this negative “drag” with the nuisance term. Although there's no closed form expression for , one can use the statistics of a Poisson process as a first order approximation. In that case ( is the Dirac distribution) and Eq. 49 becomes(50)
The last term on the right-hand side of Eq. 50 implies that, on average, each presynaptic spike in neuron causes the synaptic weight to depress by a fixed amount. This quantity increases with the variance of the noise process, in this case the inhomogeneous Poisson process that drives the neuron, and inversely to the number of critic neurons. The time course of the presynaptic spike effect is ruled by , which is plotted in the top panel of Figure 8D. The aggregate nuisance effect on of a single presynaptic spike is proportional to the integral of over time.
In Figure 8E, we explore the dependence of the nuisance term on in numerical simulations. Eq. 50 suggests that the mean learning rule term should obey a relationship of the form(51)Here is the result of the “useful” part of the learning rule, and contains all the other dependencies of the nuisance term. We tested the dependency by simulating agents with variable numbers of critic neurons in a linear track scenario. The setup was similar to that of Figure 2, except that the weights were frozen, i.e., we collected the value of the learning rule at each time step, but we didn't actually update the weights. The mean learning rule outcome for 200s of simulations are plotted in Figure 8E as crosses, against the number of critic neurons. The black line shows a fit of the data by Eq. 51: both are in good agreement.
From Eq. 50, we see that the nuisance term also depends on the variance of the noise process. It is difficult to control the variance of our spiking neurons' noise process without also altering their firing rate and thus the result of the learning rule. To circumvent this difficulty, we turned to a rate model, where the single critic neuron's firing rate was(52)where is a constant, the place cells rates are defined in Eq. 22 and is a white noise process. Similar to the steps above, a gradient descent on yields a learning rule of the form(53)Due to the noise component in , the learning rule suffers from the same noise-driven nuisance as the spiking version. This depends on the noise's variance , so that the mean weight change obeys(54)where . In Figure 8F, we use the rate-based model and rule in the same “frozen weights” linear track scenario as in Figure 8E. This time we looked at how the mean weight change varied as a function of the noise variance. Again, the data is well matched by a fit with Eq. 54 (black line), suggesting that the nuisance term behaves as expected.
In the preceding section we found that a noise correlation nuisance in the squared TD gradient learning rule causes it to be ineffective. However, the same actually should apply to the TD-LTP rule. Indeed, if we repeat the steps above leading to Eq. 50 for the learning rule TD-LTP, we get(55)
The only difference is the time course of the nuisance term, which is for the squared TD gradient rule versus for TD-LTP. Figure 8D shows a plot of both expressions: because the TD-LTP expression is much smaller, these are plotted on different axes. As noted before, the integral of the nuisance is proportional to these time courses (shown on Figure 8D). The term for TD-LTP is more than three orders of magnitude smaller than that of the square TD gradient rule.
In Figure 8G and H, we repeat the experiments of Figure 8E and F, respectively. These show that the TD-LTP learning rule also suffers from a nuisance term, but that it is orders of magnitude smaller than for the squared TD gradient rule. As shown by Figure 8C and in the Results section, this nuisance is not sufficient to prevent TD-LTP from properly learning the value function .
In the Results section, we claim that -values based algorithms, such as Sarsa [24] and Q-Learning [23] are difficult to extend to continuous time in a neural network setting. Here we develop this argument.
In the discrete Sarsa algorithm, the agent maintains an estimation of the state-action -values. For an agent following the policy , starting at time step in state and executing action , this is defined as the discounted sum over future rewards :(56)Here is a discount factor, and and represent the future states and actions visited by the agent under policy . To learn -values approximations to the real , Sarsa suggests the following update rule at time step :(57)where the TD error is defined as(58)
If one were to propose a continuous time version of Sarsa, one would start by redefining the state-action value function to continuous time t, similar to the value function of Eq. 3(59)Here now plays the role of the discount factor . As we did for Eq. 5, we define the TD error on the -value by taking the derivative of Eq. 59(60)To calculate the TD error, one therefore needs to combine the three terms in Eq. 60. We assume the reward is given by the environment. Typically [16], [20], neural networks implementations of -values based reinforcement learning consist of a number “action cells” neurons , each tuned to a specific action and rate-coding for the state-action values(61)where is neuron 's firing rate. In that case, reading out the value is thus simply a matter of reading the activity of the neuron coding for the action selected at time .
Reading out the temporal derivative is harder to do in that context, because the currently chosen action is evolving all the time. For small , we can approximate(62)where we also used Eq. 61 and identified the action neuron tuned to action .
The difficulty that arises in evaluating Eq. 62 is the following. It requires a system that can keep track of the two recent actions and , identify the relevant neurons and , and calculate a difference of their firing rates. This is hard to envision in a biologically plausible setting. The use of an actor-critic architecture solves this problem by having a single population coding for the state-based value at all times.
|
10.1371/journal.pcbi.0030075 | Coping with Viral Diversity in HIV Vaccine Design | The ability of human immunodeficiency virus type 1 (HIV-1) to develop high levels of genetic diversity, and thereby acquire mutations to escape immune pressures, contributes to the difficulties in producing a vaccine. Possibly no single HIV-1 sequence can induce sufficiently broad immunity to protect against a wide variety of infectious strains, or block mutational escape pathways available to the virus after infection. The authors describe the generation of HIV-1 immunogens that minimizes the phylogenetic distance of viral strains throughout the known viral population (the center of tree [COT]) and then extend the COT immunogen by addition of a composite sequence that includes high-frequency variable sites preserved in their native contexts. The resulting COT+ antigens compress the variation found in many independent HIV-1 isolates into lengths suitable for vaccine immunogens. It is possible to capture 62% of the variation found in the Nef protein and 82% of the variation in the Gag protein into immunogens of three gene lengths. The authors put forward immunogen designs that maximize representation of the diverse antigenic features present in a spectrum of HIV-1 strains. These immunogens should elicit immune responses against high-frequency viral strains as well as against most mutant forms of the virus.
| The ability of human immunodeficiency virus type 1 (HIV-1) to acquire mutations that preserve virus viability yet evade immune responses contributes to the current failure in producing a vaccine. We describe the generation of candidate HIV-1 immunogens that include multiple forms of variable elements of the virus including some that retain colinearity with the virus and thus are expected to retain protein function. These antigens compress the variation found in many viral strains into lengths suitable for vaccine immunogens. For example, we can capture 62% of the variation found in the Nef protein and 82% of the variation in the Gag protein into immunogens of three gene lengths. We put forward immunogen designs that maximize representation of the diverse antigenic features present in a spectrum of HIV-1 strains. These immunogens should elicit immune responses against high frequency viral strains as well as against most mutant forms of the virus.
| The failure of AIDS vaccine efforts in the past 20-plus years is thought to be due, in part, to the enormous viral antigenic diversity found within and among patients with human immunodeficiency virus type 1 (HIV-1) infection. However, until recently, relatively little effort had been devoted to choosing particular viral variant sequences or designing sequences to include within vaccines [1,2]. There were early attempts to design vaccines by concatenating commonly recognized T cell and antibody epitopes [3], but these did not produce a viable vaccine candidate. New methods of combining epitopes are being explored in vaccine design, including production of pseudoprotein strings of T cell epitopes [4], and the synthetic scrambled antigen vaccine (SAVINE) [5], which employs consensus overlapping peptide sets from HIV-1 proteins scrambled together. Focusing on the use of whole viral protein sequences, natural strains (NSs) as well as consensus (CON) sequences are being used as a means to minimize the abrogating effect of antigenic diversity in vaccine antigens [2,6,7], as are the inferred most recent common ancestors (MRCA, or ANC) [6,8–10] of targeted virus populations defined as sequences that reside at the basal node of the set of in-group sequences in a phylogenetic tree reconstruction [11]. HIV-1 env sequences representing both the CON and ANC have been prepared and studied, but neither has generated exceptionally broad humoral immune reactivity in initial small animal studies [7,12].
In an effort to develop antigens that capture both the summary of circulating variation found in CON estimates, and the coupling of mutations generated with inferred ANC sequences, we have developed an alternative computational method that reconstructs the ancestral state sequence at the center of tree (COT) ([13] and Rolland M, Jensen MA, Nickle DC, Learn GH, Heath L, et al., unpublished data). The COT sequence explicitly minimizes genetic distance, as does the CON, and because it is derived from a phylogenetic tree, it embodies the most likely mutational coupling relationships found in the ANC. Despite these efforts, it may be that no single unit-length antigen, including any NS, CON, ANC, or COT, will encompass sufficient antigenicity to elicit protective immune responses against a broad array of viruses [7,12], as will be required of an AIDS vaccine. This led us to hypothesize that we would need more than one antigenic sequence, or greater than one gene length of the antigen, to elicit protection against the broad antigenic diversity encountered in natural infections. However, cocktails of large numbers of native, full-length NS antigens would quickly become unmanageably complex for practical use as vaccines.
Here, we propose a means to cope with HIV-1 diversity by engineering vaccine antigen constructs to include short protein sequences present at high frequencies in natural viral populations. Currently, this method is explicitly directed toward developing CD8+ cytotoxic T lymphocyte (CTL) responses, which are critical to controlling viremia during infection [14–17]. Because the cumulative strength of the CTL-mediated immune response depends on the presence of recognizable epitopes (often approximately nine amino acids in length) in the target proteins, it is logical to seek to maximize epitope coverage within a vaccine antigen. However, although substantial, our current catalog of known CTL epitopes appears to be woefully incomplete [18], hence our strategy relies on the universe of HIV sequences and not solely on known epitope content. Thus, here we will define coverage as the sum of the frequencies of all nine amino acid segments (9mers) where the frequency is derived from random independent HIV-1 subtype B isolates found in the vaccine construct. As our epitope catalog increases and our knowledge of protein degradation, CTL epitope binding, and HLA presentation is expanded, this epitope-specific data can be integrated into the measure of coverage (e.g., by weighting epitope frequencies in accordance to their relative “importance” when computing coverage). In this study, we applied our method to Nef because it is highly variable and is potentially very difficult to design an immunogen against, and to Gag because it is immunologically important yet more conserved. We considered subtype B sequences because more immunological information is available about this subtype than any other. This clearly makes the vaccine construct described here as region-specific because of the biogeographic nature of the distribution of viral subtypes across the globe [19]. However, our purpose is to illustrate and demonstrate that this method has promise at producing a vaccine against highly variable infectious agents such as HIV.
Vaccination with all known viral sequences would capture all known viral sequence variation, but realistic vaccine constructs might at best include several sequence lengths, each length containing major variants for immune presentation. To quantify variant representation and rationally choose the included variation on this basis, Jojic and colleagues have proposed a method based on machine-learning for the compression of sequence variation into a sequence of minimal length (the “epitome”; [20,21]). Below, we describe an alternative, more transparent algorithm also designed to attain optimized sequence coverage over a fixed-length antigen. We refer to the constructs generated by our method as COT+ because they consist of COT antigens augmented by the addition of high-frequency 9mers. We demonstrate the performance of our approach on the highly variable and epitope-rich viral Nef protein as well the epitope-rich major structural protein, Gag. The algorithm consists of five steps applied to a sample of viral nucleotide sequences, each isolated from a separate patient. We started with all publicly available nef and gag gene sequences from HIV-1 subtype B [22]. By excluding sequences with more than two stop codons and with large indels, and including only independent single sequences from a given individual to avoid sampling bias, we obtained a 169-sequence dataset for Gag; the Nef data set was also constrained to 169 sequences for comparative purposes (Table 1 includes the GenBank IDs of all sequences used). The algorithm, however, can rapidly process datasets with thousands of sequences when such datasets become available.
(1) A COT sequence is calculated as described ([13 and Rolland M, Jensen MA, Nickle DC, Learn GH, Heath L, et al., unpublished data) from a phylogenetic tree that captures the relationships among genes in the sample using maximum likelihood methods [23]. Briefly, from aligned sequences we estimate a maximum likelihood tree under a HKY + Γ + I model of evolution in PAUP*v4beta10 [24]. The resulting tree is re-rooted at the point that describes the least-squares distance to all the tips on the phylogeny (the COT node). We then infer the maximum likelihood state using the same model of evolution as above.
(2) A table of unique 9mer peptides [20,21] with their corresponding frequencies (the 9mer distribution) is constructed from translated protein sequences. To illustrate this, note that if our sample contained N identical sequences of length L each, but every 9mer in the COT peptide library was unique, then each peptide would be at equal frequency
. On the other hand, if every sequence were different from all others, to the extent that no 9mer was represented twice, the frequency for each peptide would equal
. Actual samples will yield an intermediate distribution that can be exploited for vaccine design (see Figure 1). We used this distribution to compute “coverage”; that is, as we select candidate fragments to be included in the potential vaccine, we will select only those fragments that are highly represented under the 9mer curve.
(3) Unique or rare 9mers, which by definition are unlikely to be common in circulating viral strains, are likely to derive from low-fitness variants [25,26] and, because of their low frequency, have low probability of being incorporated in our vaccine constructs. Specifically, we calculate the frequency of all observed mutations at each site, and revert any mutation with a frequency below a fixed “smoothing” threshold, M, to the corresponding character in the COT sequence. All 9mers present in the COT sequence are then removed from the 9mer distributions before proceeding to the next step.
(4) Given a fixed window size F (ranging from 9 to L, where L = the length of the protein sequence [we start with 9 because that is the size of the peptide that is most often found to encode epitope sequences] and a stride parameter S [ranging from 1 to L, the length of the protein]), we generate all sequence fragments from the sampled sequence by iteratively shifting the frame S residues at a time. We then compute the coverage for each sequence fragment not already present in the COT sequence, and append the sequence fragment to the COT string, compressing with possible overlap to yield a COT+ molecule with the highest ratio of coverage per length. Specifically, fragments are chosen by their level of coverage and whether or not they have differences with respect to the COT sequences. The highest coverage fragments are chosen first, with subsequent fragments with lower coverage being chosen subsequently. This process is repeated until the sequence of desired length is derived. The length of the COT+ sequence is arbitrarily chosen by taking into account plasmid size limitations for producing and delivering an antigen construct and the amount of variability that can be efficiently incorporated as the length is extended, which in turn depends on the variability found in circulating strains that have been sampled for a particular gene. We note that it is possible to arrange the order in which sequence fragments are added to COT+ to maximize the overlap of consecutive fragments, thereby further compressing the antigen.
(5) The values of window size F, stride step S, and smoothing threshold M are varied to achieve maximum coverage (Figure 2A and 2B).
We compared our constructs of various lengths to randomly drawn sequences from the curated dataset of 169 sequences using the optimal values for F and S. We generated COT+ for both Gag and Nef at ever-increasing unit protein lengths until we reached 100% coverage. For comparison, we concatenated randomly sampled protein sequences 100 times at ever-increasing unit protein lengths from both Gag and Nef and measured 9mer coverage across the same gene lengths (Figure 3A and 3B). We chose protein unit length for our comparison, but COT+ can be derived for any partial unit protein length desired.
To ensure that we were not overestimating the coverage of our constructs due to the finite size of our dataset, we repeated our approach using 10-fold cross-validation. We partitioned the data into ten sets, and for each we estimated COT+ from the remaining 90% of the data and then measured its coverage of the sequences in the chosen set. Thus, given that our assessment of coverage is on a set of sequences not seen in training, we yield an estimated lower bound on the coverage we would obtain for a larger population. We report this lower bound as a percentage of similarity to the estimated upper-bound COT+, derived from training and testing on all 169 sequences for both Gag and Nef. This study is geared to understand the effect of sample size on the on the COT+ estimation and to show that we are not overfitting the estimations.
Although the list of known HIV-specific CD8 T-cell epitopes is far from complete [18], we sought to determine how well our 9mer coverage-based constructs identified known epitopes. To this end, we obtained all available HIV CTL epitopes from the Los Alamos National Laboratory (LANL) HIV immunology database [27] and counted the perfect matches to our constructs. Because many true epitopes are listed multiple times and larger peptides are reported frequently where the true epitope is embedded, we curated the database to remove any larger epitope that had a smaller embedded known epitope with the same supertype HLA response pattern, and removed any duplicates.
We inferred COT sequences from databases of Gag and Nef protein sequences from HIV-1 subtype B from 169 independently infected individuals, and then added frequently observed variant 9mer peptides to create COT+ sequences. The frequencies of unique 9-mer peptides are shown in Figure 1. We find that maximal coverage occurs when the window size, F, is 17, the stride length, S, is 1, and when smoothing M is 0 (Figure 2A and 2B). One possible reason for why an S value equal to 1 leads to the highest coverage is that it gives every amino acid in the sequences a chance to be in every possible position in a high-scoring peptide. Counterintuitive to this is the observation that S values greater than 1 do not get penalized with big drops in 9mer coverage. We think the explanation for this observation has to do with the fact that even with S larger than 1, every amino acid in the sequences is still considered when building a construct. This is exemplified by the fact that the biggest drops in 9mer coverage come when S is larger than F, because it is in this parameter space that some amino acids have the probability of not being considered at all in the resulting construct.
Adding peptides to generate a three-gene-length COT+ construct achieved 82% 9mer coverage for Gag and 62% for Nef, whereas an antigen constructed from several random concatenated database sequences [22] needed to achieve the same level of coverage required ten gene lengths for Gag and approximately 11 for Nef (Figure 3A and 3B). When COT+ is compared with 100 constructs of the same length obtained by concatenating randomly selected sequences from the Los Alamos National Laboratory database [22], the COT+ estimate had a higher level of coverage in every case (randomization test, p < .01) for both Gag and Nef. The flattening of the curves in Figure 3A and 3B suggests that after the COT+ construct has grown past a few gene lengths, the benefit of adding more length is dramatically reduced. For example, the extension of the COT+ construct from one to three gene lengths results in a 16% increase in coverage for Gag and a 13% increase in coverage for Nef. However, extending COT+ from three to five gene lengths yields only 5% additional coverage for both Gag and Nef. The COT+ sequence reaches 100% coverage at 33 gene lengths for Gag and 67 gene lengths for Nef, while the randomly sampled sets reach 100% coverage only after all 169 sequences are included. The latter observation is due to the fact that many of the mutations found in HIV are private (i.e., found only within the lineage infecting a particular person).
When applied to small datasets, our algorithm generates COT+ constructs with high coverage. An extreme example is making a three-gene construct from just three genes in the training set. In this scenario, we can trivially achieve 100% coverage. The larger the training set, the lower the coverage in a three-gene-length vaccine construct. A 10-fold cross-validation study was therefore designed to determine the effects of sample size on our COT+ constructs. Specifically, at three protein lengths, the cross-validated coverage of Gag and Nef are 96% and 93%, respectively. This suggests that for both proteins these inferences are generalizable across HIV-1 subtype B and that adding more sequence data into the training dataset would add very little to these estimations. That is to say, 10% of the original 169 sequences produce estimations of the COT+ that are highly consistent with the estimations from the entire dataset, supporting the notion that there is a saturation effect and that adding sequences beyond the 169 will not give rise to better estimations.
Assessing the inclusion of functional CTL epitopes in our constructs is problematic. The majority of the known CTL epitopes were mapped using peptides derived from a limited number of HIV strains (e.g., laboratory-adapted strains and consensus sequences). The CTL database is also incomplete (e.g., a recent study that used a subset of autologous peptides from a single patient enabled recognition of 28% more epitopes in the virus than were previously reported [18]), and it is unclear whether characterized epitopes form an unbiased sample of naturally occurring antigenic peptides. It is also necessary that the epitope be presented in the proper context of adjacent amino acids for efficient immunoproteasome cleavage. We therefore assessed the overall size of the peptides needed to obtain maximal coverage of included 9mers. As shown in Figure 2A and 2B, maximal coverage of both the Gag and Nef datasets was obtained with a window size of 17 amino acids and a stride of one amino acid and no smoothing required (see Methods). Hence, we are able to construct immunogens that preserve much of the extended local amino acid environment of the epitope without sacrificing coverage. This enhances the likelihood that the desired peptide epitope will be properly cleaved by cellular proteases and presented efficiently on HLA molecules.
Next, we assessed the inclusion of known CTL epitopes in our constructs by comparing the number of known HIV-1 Nef and Gag epitopes [27] contained in the three-gene-length COT+ constructs to that of 1,000 combinations of three randomly selected database sequences (Figure 4). Sequences from the viral strains used to map these epitopes were excluded from the randomization study. Although our algorithm does not attempt to explicitly enrich for known CTL epitopes, the number of known epitopes in COT+ is significantly higher than in a random three-gene construct (p < 0.001) for both Gag and Nef. This suggests that COT+ provides a substantial boost in the number of epitopes shared between the immunogen and a random circulating database variant, and thus may have enhanced potential as an immunogen.
COT+ constructs provide a means to extensively compress epitope variation into an immunogen of minimal size. Much of the known variation of both the relatively conserved HIV-1 Gag gene and the quite variable Nef gene can be successfully compressed into COT+ constructs of a few gene lengths. Little increases in variation coverage are noted, however, beyond three to four gene lengths. Coverage grows with length approximately in a y = mlog(x) + b form where y is coverage and x is length of the construct. The difference between COT+ construct of Gag and Nef can be broken down into these terms. The coverage intercept parameter b is higher for Gag constructs than for Nef simply because Gag is a more conserved protein than Nef. However, the parameter m is larger for Nef than it is for Gag because the benefits of 9mer compression on coverage are higher with constructs made from variable proteins.
Our COT+ generation algorithm is a rapid, computationally efficient heuristic approximation, though it is not guaranteed to find the antigen that achieves maximal epitope coverage for a fixed length. More computationally intensive approaches, such as genetic algorithm searches or approximate solutions to the classic Traveling Salesman problem (see http://mathworld.wolfram.com/TravelingSalesmanProblem.html), could also be brought to bear on the problem of antigen design. Surprisingly, selecting the high-frequency 9mers alone and appending them to the COT sequence does poorly in terms of total coverage (unpublished data). This observation is due to the fact that many of the 9mers do not overlap, and therefore the fragments cannot be efficiently joined. By going back and selecting high-coverage peptide windows from the original data, we obtain better compression in the vaccine construct leading to higher coverage constructs for the same length.
It is a reasonable assumption that the retention of native protein structures might be advantageous in generating CTL epitopes, since epitopic peptides are generated in vivo by protein degradation within infected cells. Nef and Gag COT clearly adopt a native structure, as they retain biological activity (Rolland M, Jensen MA, Nickle DC, Learn GH, Heath L, et al., unpublished data). However, the extended COT+ component of antigens generated in the manner proposed here does not preserve a sequence that is necessarily collinear with the native gene over the second and third gene lengths (Figure 5A). Hence, we have also considered additional means of optimizing immunogen structures that also preserve native structure. First, we can assemble high-frequency variable elements in a pattern collinear with the native gene, with some segments redundant with COT to retain collinearity (Figure 5B). We can also use NS sequences in combination with the COT sequence to optimize coverage (Figure 5C). We can also do very well in generating coverage by exclusive use of NS sequences that maximize 9mer coverage (Figure 5D). Although it is not guaranteed, these additional constructs (Figure 5B–5D) should have biologically acceptable tertiary structures. The COT+ approach captures more of the 9mer distribution and more of the known CTL epitopes than any of the potential constructs presented here. Applying high-frequency peptides onto COT to create a collinear pattern provides the second highest level of diversity and epitope enrichment, but the use of COT plus two NSs is not beneficial relative to judicious choice of three NSs. Last, it should be noted that all of these methods substantially exceed the coverage afforded by the use of a single strain as a vaccine.
Immunodominance gives rise to a rank order of immune responses to specific epitopes [28], and the underlying biological mechanisms giving rise to these rank orders are poorly understood. The antigen designs we report here do not take immunodominance into account. One can argue that the combination of epitopes we have derived could elicit an immunodominant response that does not reflect what is found in circulating HIV strains and hence could be a poor choice for vaccine design. However, since the strings of peptides in our immunogen design are captured by their frequency in the circulating viral population, we surmise that these antigens have epitopes that are shared across many potential challenge strains and could thus lead to potentially broad immune response. However, immunodominance rank order patterns can be partially illuminated by expressing epitopes from different vaccine vectors [29–31]. By vaccinating with different combinations of vectors encoding a single or more antigens, they found that using separate vectors elicited broader CD8+ T cell responses. Because COT+ is directed towards capturing high-frequency fragments from a variable protein, it is well-suited to being expressed as segments on separate vectors. The COT+ algorithm can be generalized to produce sets of immunogens that can take advantage of this phenomenon.
COT+ constructs are able to capture significantly more known epitopes and potential antigen variability than much longer constructs composed by combining circulating strains. Considering the substantial expense and difficulty involved in production and testing of candidate vaccines, careful crafting of potential antigens using computational methods, including that shown here, may be beneficial. Furthermore, this approach is applicable not only to HIV vaccine design, but to the design of vaccines targeting any pathogen capable of rapid escape from immune recognition. |
10.1371/journal.ppat.1006522 | Cucumber mosaic virus coat protein modulates the accumulation of 2b protein and antiviral silencing that causes symptom recovery in planta | Shoot apical meristems (SAM) are resistant to most plant viruses due to RNA silencing, which is restrained by viral suppressors of RNA silencing (VSRs) to facilitate transient viral invasion of the SAM. In many cases chronic symptoms and long-term virus recovery occur, but the underlying mechanisms are poorly understood. Here, we found that wild-type Cucumber mosaic virus (CMVWT) invaded the SAM transiently, but was subsequently eliminated from the meristems. Unexpectedly, a CMV mutant, designated CMVRA that harbors an alanine substitution in the N-terminal arginine-rich region of the coat protein (CP) persistently invaded the SAM and resulted in visible reductions in apical dominance. Notably, the CMVWT virus elicited more potent antiviral silencing than CMVRA in newly emerging leaves of infected plants. However, both viruses caused severe symptoms with minimal antiviral silencing effects in the Arabidopsis mutants lacking host RNA-DEPENDENT RNA POLYMERASE 6 (RDR6) or SUPPRESSOR OF GENE SILENCING 3 (SGS3), indicating that CMVWT induced host RDR6/SGS3-dependent antiviral silencing. We also showed that reduced accumulation of the 2b protein is elicited in the CMVWT infection and consequently rescues potent antiviral RNA silencing. Indeed, co-infiltration assays showed that the suppression of posttranscriptional gene silencing mediated by 2b is more severely compromised by co-expression of CPWT than by CPRA. We further demonstrated that CPWT had high RNA binding activity leading to translation inhibition in wheat germ systems, and CPWT was associated with SGS3 into punctate granules in vivo. Thus, we propose that the RNAs bound and protected by CPWT possibly serve as templates of RDR6/SGS3 complexes for siRNA amplification. Together, these findings suggest that the CMV CP acts as a central hub that modulates antiviral silencing and VSRs activity, and mediates viral self-attenuation and long-term symptom recovery.
| In many virus-infected plants, the development of viral symptoms on the upper leaves gradually decline, until finally the top leaves appear normal and become resistant to secondary infection. Many documented cases suggest that symptom recovery is accompanied with antiviral RNA silencing. Most plant viruses encode viral suppressors of RNA silencing (VSRs) that can facilitate transient viral entry into meristems by blocking host RNA silencing defenses. However, the mechanisms of the following longer-term viral exclusion and modulation of VSRs functions remain elusive. Our studies with a substitution mutation in the CP gene demonstrate that the CMV CP has a negative role in SAM invasion. The studies suggest that during late infections, increasing CP concentrations induce potent RDR6/SGS3-dependent antiviral silencing by down-regulating accumulation of the 2b protein and induction of efficient siRNA amplification. Here, we propose a new evolutionary strategy in which the CMV CP has a role mediating viral self-attenuation and long-term symptom recovery.
| RNA silencing is a well-established plant antiviral response triggered by viral double-stranded RNAs or highly structured single-stranded RNAs in host plants. Host Dicer-like (DCL) enzymes cleave both RNA types into 21 to 24 nucleotide (nt) small interfering RNAs (siRNAs) that are subsequently sorted into Argonaute-containing RNA-induced silencing complexes (RISC) to guide specific cleavage of the cognate viral RNAs [1–7]. Some cleavage products serve as templates for host RNA-directed RNA polymerase (RDR) 1, or RDR6, to synthesize abundant de novo dsRNAs that are processed by DCLs into secondary siRNAs that enhance antiviral RNA silencing [5, 7–11]. In addition, a plant-specific RNA binding protein, Suppressor of Gene Silencing 3 (SGS3), is required for siRNA amplification through forming complexes with RDR6 [12]. RNA silencing, as a major defense mechanism, occurs in all virus-infected tissues and is an extensive feature in newly emerging tissues [11, 13–15]. Recent studies have revealed that symptom recovery from viral infection is generally concomitant with induction of RNA silencing in the shoot apices of infected plants and this depends in part on RDR activity. For example, Potato virus X (PVX), Turnip crinkle virus (TCV), and Potato spindle tuber viroid (PSTVd) transiently invade the meristems of plant mutants defective in host RDR6 [11, 13, 14].
As a counter-defense against host RNA silencing, many plant viruses have evolved VSRs to block various RNA silencing steps [16–18]. Some VSRs are also viral pathogenicity factors during systemic infections. For instance, the 2b protein of CMV and the 16K protein of Tobacco rattle virus (TRV) facilitate shoot apical meristem (SAM) invasion by blocking antiviral RNA silencing [19–21]. Ectopic expression of VSRs, like the potyvirus HC-Pro protein, enhance viral RNA accumulation of two distinct nepoviruses and prevent symptom recovery [22, 23]. Nonetheless, viral meristem invasion is transient and is followed by long-term meristem exclusion [19, 20]. Hence, the mechanisms of long-term recovery from transient SAM invasion remain to be elucidated.
CMV is the type virus of the genus Cucumovirus in the family Bromoviridae. The CMV genome is composed of three positive-stranded RNAs [24]. RNA1 and RNA2 encode 1a and 2a proteins, respectively, which comprise the viral RNA-dependent RNA polymerase subunits [24]. RNA2-derived subgenomic RNA4A encodes 2b protein, a well-known VSR and determinant factor in viral virulence and systemic infection [5, 25, 26]. RNA3 and its subgenomic RNA4 encode movement protein (MP) and coat protein (CP) that are required for viral cell-to-cell and systemic movements, respectively [24].
The CMV CP is a multifunctional factor that has roles in viral systemic movement, host range and aphid transmission [27–31]. For the Pepo and MY17 CMV strains, CP-mediated cell-to-cell movement is implicated in SAM invasion of host plants [32]. In addition, the CMV CP is an aphid transmission determinant that mediates viral spread between host plants [28, 31]. These properties and studies of different CMV strains suggest that CMV CPs are key host range determinants [30]. Previous studies have also shown that some amino acid residues of CMV CP contribute to various symptoms induction. For instance, CMV pepper strain mutants, in which proline 129 is replaced by 19 other amino acids, induce various systemic symptoms in plants [33]. Moreover, CMV CP amino acid 129 also determines viral invasion of the SAM in tobacco plants [32]. Although CMV CPs have been extensively studied as factors involved in positive regulation of viral spread and symptom induction, their negative roles in SAM infections have not been described.
In the current study, we found that the N-terminal R-rich region (R13RRRPRR19) of the CMV CP has a negative role in persistent viral SAM invasion. Our data indicate that elevated expression levels of the CMV CP induces potent antiviral RNA silencing by down-regulating the accumulation level of 2b VSRs and inducing siRNA amplification. Thus, we propose a novel self-attenuation mechanism, in which the CMV CP antagonizes the suppression effects of 2b protein and plays a pivotal role in regulating compatible interactions between CMV and host plants to prevent viral over-accumulation and persistent viral invasion of the SAM.
Plant virus-encoded CPs mainly participate in encapsidation and movement [34–38], and are increasingly appreciated as an important regulator of viral RNA replication and translation that are associated with CP RNA binding affinity [15, 39, 40]. Protein sequence analyses revealed that the N-terminal region (R13RRRPRR19) of the CMV CP is enriched in basic and positively charged amino acid residues that contribute to functional RNA binding activities. To explore the requirements of this R-rich region in viral infection, the basic amino acid residues were substituted by alanine (R13-19: A), and the resulted mutant was designated as CPRA (Fig 1A). To compare the functions of CPWT and CPRA in the context of viral sequences, we first developed an Agrobacterium tumefaciens-mediated CMV infection system in N. benthamiana plants. The cDNAs of the three CMV genomic RNAs were engineered into pCass4-Rz to generate pCass-RNA1, -RNA2, and -RNA3 (Fig 1A). The CPWT- and CPRA-containing viruses are named CMVWT and CMVRA, respectively. Infections were carried out by co-infiltration of equal concentrations of A. tumefaciens EHA105 mixtures harboring pCass-RNA1, -RNA2, and -RNA3 into N. benthamiana leaves.
We first explored whether CMVRA could form normal viral particles during viral infections. The CMVWT and CMVRA-infected leaves were homogenized for virions purification, and the purified virions were observed by transmission electron microscopy, which showed that CMVWT and CMVRA formed viral particles with similar appearance (S1A Fig). Nonetheless, the CMVRA particles were less stable than CMVWT virions in an RNase assay (S1B Fig). These results indicate that the CP R-rich motif is not essential for virion assembly or systemic infection in N. benthamiana plants.
The susceptibility of N. benthamiana was examined to determine the function of the R-rich region in systemic infections. At 7 dpi, CMVWT induced mosaic symptoms in the fully expanded leaves, but only elicited limited or recovered symptoms in shoot apices of infected N. benthamiana plants (Fig 1B, right panel). In sharp contrast, CMVRA infections resulted in severely distorted newly emerging leaves (Fig 1B, middle panel). Subsequently, CMVRA-infected plants exhibited extremely short internodes and petioles at shoot apices to produce a rosette appearance combined with substantial stunting between 14 and 21 dpi (Fig 1C). At 42 dpi, CMVWT-infected plants developed mosaic symptoms in leaves present in the central parts of the stems, and exhibited substantial reductions in growth compared to mock-infected plants (Fig 1D). However, symptoms in the shoot apices of CMVWT-infected plants were modulated and the apices maintained apical dominance (Fig 1D). In contrast, all of the CMVRA-infected plants developed several lateral shoots without distinguishable primary stems (Fig 1D). These symptoms suggested that the shoot apices were severely infected and that apical dominance was disturbed leading to production of lateral bud outgrowths [41].
Since the 2b protein is a strong silencing suppressor [5, 25, 26], we next examined whether the persistent SAM invasion by CMVRA depends on the 2b protein. A 2b-deleted mutation (CMV-Δ2b) was engineered into pCass-RNA2 by point mutations as described previously [6], and co-infiltrated into N. benthamiana with pCass-RNA1 and wild-type pCass-RNA3 (CMVWT-Δ2b), or mutated pCass-RNA3-CPRA (CMVRA-Δ2b). CMVWT-Δ2b caused mild mosaic symptoms in the systemic leaves, whereas CMVRA-Δ2b did not induce any obvious symptoms (S2A Fig). To explore the replication of CMVWT-Δ2b and CMVRA-Δ2b at 7 dpi, the infiltrated leaves were sampled and the CP accumulation was detected by Western blotting at 7 dpi. Both CMVWT-Δ2b and CMVRA-Δ2b CPs had accumulated to similar levels (S2B Fig), suggesting that the 2b deletion did not affect virus proliferation in infiltrated leaves. In contrast, CMVWT-Δ2b CP was present in the upper uninfiltrated leaves at 7 dpi, but CMVRA-Δ2b CP was not detected (S2C Fig). RT-PCR was performed to detect viral RNA accumulation in the upper leaves with primers corresponding to the RNA3 CP region. CP specific bands were detected in plants inoculated with CMVWT-Δ2b, but not with CMVRA-Δ2b (S2D Fig). Collectively, these data demonstrate that CMV CPWT exerts a negative role in shoot apex infections, and 2b protein is required for systemic infection of the CMVRA mutant.
Because apical dominance was abolished in CMVRA infections, we proposed that CMVRA invaded meristems and altered the meristematic activity of the infected plants. To explore this possibility, longitudinal sections of the topmost flowers and shoot apices from mock-, CMVWT-, and CMVRA-infected plants were examined by in situ hybridization with digoxigenin-labeled CMV RNA3 probes. CMVWT invaded most shoot meristems and floral primordia by 7 dpi, but subsequently disappeared between 14 and 21 dpi, despite of some detectable signals below the SAM and floral primordia (Fig 2, middle panels). In contrast, CMVRA was abundant below the meristems at 7 dpi, and partially moved into the meristems by 14 dpi (Fig 2, right panels). CMVRA invaded the meristems of all the infected plants by 21 dpi (Fig 2, right panels), but as expected were absence in meristems from mock inoculated plants (Fig 2, left panels). These results demonstrate that CMVRA accumulated to high levels and resulted in severe stunting and abolished apical dominance, whereas CMVWT-infected plants recovered from meristem infection.
Because RNA silencing is a key antiviral mechanism in meristems infections [11, 13, 14, 19–21], we analyzed levels of viral RNA and siRNAs by Northern blotting at 7 dpi. The newly grown tissues were collected for the Northern blotting analysis of viral RNA and siRNA at 7 dpi. A markedly increased accumulation of viral RNA of CMVRA was detected in the newly grown tissues compared with those of CMVWT (Fig 3A, compare lanes 3, 4 with 5, 6). However, CMVRA RNA3-derived siRNAs (RNA3-vsiRNAs) were present at a much lower level than those of CMVWT (Fig 3B, compare lanes 3, 4 with 5, 6). To determine the relative accumulation levels of viral genomic RNA3 and RNA3-vsiRNAs, the hybridization signal intensities in three independent experiments were evaluated. The values of RNA3 and RNA3-vsiRNAs in CMVRA-infected plants were set as one unit. CMVWT RNA3 accumulated to about 30% of CMVRA RNA3 (Fig 3C, left panel, P-value < 0.001). However, the levels of CMVWT RNA3-vsiRNAs were 2.3 times those of CMVRA RNA3-vsiRNAs (Fig 3C, middle panel, P-value < 0.01). The relative RNA3-vsiRNAs/RNA3 ratio in the CMVWT-infected newly emerging tissues was six-fold higher than that of CMVRA-infected plants (Fig 3C, right panel, P-value < 0.01), indicating that CMVWT induces more potent antiviral silencing than CMVRA in the emerging tissues.
Collectively, CMVWT promotes the accumulation of vsiRNAs that inhibit virulence in the shoot apices of infected plants. In contrast, the lower amounts of CMVRA vsiRNAs reduced antiviral silencing and enabled CMVRA to persistently infect the newly emerging tissues.
To provide more detailed genetic analyses of host effects on CMV infection, we investigated the impacts of the CMV CP derivatives during infection of Arabidopsis thaliana. CMVWT induced visible disease symptoms in the fully expanded leaves of Columbia-0 (Col-0) wild-type plants at 21 dpi (Fig 4A and S3A Fig). These plants had similar bolting times as the mock inoculated plants and maintained apical dominance during late infection stages between 35- and 56- dpi (S3A Fig). Compared with CMVWT, CMVRA infection resulted increased numbers of distorted leaves in the newly emerging tissues (Fig 4A and S3A Fig), late bolting, and reduced apical dominance in the late infection stages (S3A Fig). Northern blotting results of viral RNA and vsiRNAs extracted from CMVWT- and CMVRA-infected systemic leaves at 21 dpi were consistent with those of infected N. benthamiana, showing that CMVRA infected plants accumulates higher levels of viral genomic RNA and less vsiRNAs compared with CMVWT infections in both plant species (Fig 4B and S3 Fig). Thus, CMVWT elicits more potent antiviral silencing than CMVRA in emerging tissues of infected A. thaliana, and is similar in this regard to infected N. benthamiana plants.
To explore whether CMVWT induction of potent antiviral silencing depends on host RDRs, we inoculated A. thaliana rdr6 and sgs3 mutants with CMVWT or CMVRA. The rdr6 and sgs3 mutants infected with CMVRA displayed severe symptoms similar to those of wild-type plants and whole plant development was restrained (Fig 4A and Fig 4C). The rdr6 and sgs3 mutant plants infected with CMVWT exhibited more severe disease symptoms in the newly emerging tissues compared with wild-type plants (Fig 4A and Fig 4C). In addition, Northern RNA hybridizations revealed that the accumulation level of viral RNA in CMVWT-infected wild-type plants was dramatically lower than those of infected rdr6 or sgs3 plants, whereas higher levels of vsiRNAs accumulated in infected wild-type plants than in rdr6 or sgs3 plants (Fig 4B, compare lanes 3, 5, and 7). In contrast, high levels of viral RNA and low levels of vsiRNAs were detected in the emerging tissues of Col-0, rdr6 and sgs3 mutants infected with CMVRA (Fig 4B, compare lanes 2, 4, and 6). Collectively, these results demonstrate that RDR6 and SGS3-dependent amplification of vsiRNAs is required for the CMVWT-induced potent antiviral RNA silencing in emerging tissues of CMVWT infected plants.
We next investigated the accumulation of the CP and 2b protein in emerging tissues of CMV infected Col-0, rdr6, and sgs3 mutants. Accumulation of CMVWT CP in the rdr6 and sgs3 mutants increased to level similar to those of CMVRA CP (Fig 4D, top panel). Notably, accumulation of CMVWT 2b protein was lower level in the rdr6 and sgs3 mutants compared with CMVRA infection, indicating that the accumulation of 2b protein was significantly down-regulated in CMVWT-infected plants (Fig 4D, middle panel). Collectively, these results demonstrate that the down-regulated 2b protein cannot efficiently inhibit RDR6/SGS3-dependent antiviral silencing that restricts elevated accumulation of CMVWT in emerging tissues.
Our findings have demonstrated that CPWT exerts negative effects during viral SAMs infection. To independently verify the antagonistic roles of CP and 2b, we next examined the effects of CPWT and CPRA on VSR activities of 2b by co-infiltration assays in N. benthamiana plants [42]. Green fluorescence occurring early during transient co-expression of GFP disappeared completely at 5 dpi, indicating that potent silencing was induced (Fig 5A). In contrast, high intensity of GFP fluorescence in the regions of leaves co-expressing 2b and GFP suggested that GFP silencing was efficiently suppressed by the 2b protein (Fig 5A). We next compared local GFP silencing suppression by 2b co-expressed with the empty vector (V), CPWT, or CPRA, respectively. The CPWT, unlike V or CPRA, significantly attenuated the 2b-mediated suppression of GFP silencing (Fig 5A). The observations were further verified by Western blotting showing that co-expression of CPWT markedly reduced the expression levels of GFP protein expression compared to co-expression of V and CPRA (Fig 5B, top panel). Simultaneously, the CP and 2b protein also accumulated to lower level during co-expression with CPWT than co-expression with V and CPRA (Fig 5B, middle and bottom panels).
To examine the silencing potency in infiltrated regions of N. benthamiana leaves, the accumulation of GFP mRNA and siRNAs was compared by Northern blotting analyses, and hybridization signal densities were measured to determine the relative accumulation of mRNA and siRNAs. The values from leaf samples co-expressing GFP, 2b and V were set as one unit. The GFP mRNA expression level in the sample co-expressing 2b protein and CPWT significantly decreased compared with those of V or CPRA (Fig 5C, top panel and RA1 values, compare lane 4 with lanes 3 and 5), but all patches accumulated similar levels of GFP-derived siRNAs (Fig 5C, middle panel and RA2 values, compare lane 4 with lanes 3 and 5). The relative accumulated levels of GFP siRNAs versus GFP mRNA (RA2/RA1) were compared. The GFP siRNAs/mRNA ratios in the patches co-expressing 2b and CPWT were at least four-fold higher than those co-expressing 2b with V or CPRA (Fig 5C, RA2/RA1 values, compare lane 4 with lanes 3 and 5), demonstrating that CPWT enhances the potency of GFP silencing.
During CMV infection, the accumulation of the CP increases gradually during viral propagation. Therefore, we wondered whether CP affected VSR activities of 2b in a dose-dependent manner. To answer this question, we compared the GFP fluorescence from the different patches infiltrated with a CP concentration gradient (OD600 = 0, 0.1, 0.2, 0.4, and 0.8) and 2b (OD600 = 0.2) (Fig 5D, middle panel). A low concentration of infiltrated CP (OD600 = 0.1) had negligible effects on 2b-mediated suppression of GFP silencing, whereas increasing CP concentrations gradually compromised the inhibitory effects as the increasing concentration of CP, and the highest CP concentration (OD600 = 0.8) significantly down-regulated the 2b suppression (Fig 5D, left panel). Additionally, Western blotting analyses revealed that the accumulation of GFP and 2b decreased in proportion to the increasing concentrations of co-expressed CP (Fig 5D, right panel). Thus, our results indicate that low CMV CP levels do not affect 2b suppressive activities, whereas highly abundant CMV CP concentrations down-regulate 2b protein accumulation and attenuate 2b-mediated silencing suppression. The R-rich region of CP is highly conserved in many subgroup I and subgroup II CMV strains, as well as in Tomato aspermy virus (TAV) (S4A Fig), and all the CPs exert negative effects on 2b-mediated suppression of RNA silencing (S4 Fig).
In the field, synergistic viral diseases are usually caused by interactions of different viruses that result in dramatically increased viruses titer and symptom induction, which are mainly dependent on VSRs activities [43]. With regard to the co-infection of CMV with other plant viruses in the field, we postulated that CMV CP compromises the suppression activity of other VSRs in co-infected plants. To test this hypothesis, we co-infiltrated CP with P19, P38, or HC-Pro in N. benthamiana leaves. In agreement with 2b, the suppressions mediated by the VSRs were attenuated by coinfiltration with CPWT, but this effect was not observed with the CPRA (S5A Fig). The results were also confirmed by Western blotting which revealed a substantial reduction of GFP accumulation in the patches co-expressing of VSRs and CPWT compared with VSRs and CPRA (S5B Fig). At the same time, the co-expression of CPWT rather than CPRA also decreased the accumulation of the VSRs (S5B Fig). We further demonstrated that the compromising effect of the CMV CP on P19-mediated suppression was also dependent on CPWT concentrations (S5C Fig). In conclusion, highly abundant CMV CP concentrations compromise various VSRs suppression activities in patch assays, implying that the CMV CP modulates the synergistic viral disease by regulating silencing interactions and VSRs in the co-infected plants.
The decreased accumulation of VSRs by high-abundant CMV CP might be due to the strong RNA binding activity of CPWT and resulting in translation inhibition. To test this possibility, we carried out North-Western blotting assays to compare the RNA binding affinity of CPWT and CPRA. GST-tagged CPWT, CPRA, and untagged GST were expressed and purified from E.coli expression systems. Different amounts (1 μg, 2 μg, and 4 μg) of GST-CPWT and GST-CPRA were separated in SDS-PAGE gels, transferred to nitrocellulose membranes, renatured, and exposed to digoxigenin-labeled CMV RNA4 or luciferase (Luc) mRNA. High amounts (4 μg) of GST served as a negative control and failed to binding the RNAs (Fig 6A, lane 7), confirming that the GST tag does not have RNA-binding affinity. The GST-CPWT bound much higher levels of CMV RNA4 and Luc mRNA than the GST-CPRA protein (Fig 6A, compare lanes 2, 4 and 6 with lanes 1, 3 and 5). Together, these results clearly indicate that the R-rich region is important for the high RNA binding affinity of CMV CPWT.
We further examined whether the high unspecific RNA binding activity of CPWT resulted in translation inhibition by comparing in vitro translation efficiency in the wheat germ system. The results showed that high concentrations of GST tag did not affect the translation of Luc mRNA (Fig 6B, black line). The highest concentration (66.7 μg/ml) of GST-CPRA had a~ 20% translation reduction compared with the empty control (P-value < 0.01) (Fig 6B, blue line), indicating that the weak RNA binding of the GST-CPRA protein partially affected mRNA translation. However, more dramatic reductions in the luciferase translation effects were observed as the GST-CPWT concentration was gradually increased, and finally the highest concentration (66.7 μg/ml), was ~1% of the empty control (P-value < 0.001) (Fig 6B, red line). These results are in agreement with previous studies, in which the Potato virus A (PVA) CP was found to be involved in inhibition of viral RNA translation [44, 45]. Therefore, it appears that the strong RNA-binding affinity and resulting translation inhibition by the CMV CPWT contributes to reduced VSRs accumulation in CMVWT virus infections.
In previous studies, high CP concentrations of several positive-sense RNA viruses repressed RNA translation and facilitated virion assembly [44, 46, 47]. Similarly, our results show that CMV CP also efficiently inhibits translation of Luc mRNA in wheat germ system (Fig 6B). In addition to initiating viral encapsidation and/or CMV ribonucleoprotein (RNP) formation, translation inhibition also contributes to elevated accumulation of aborted mRNA transcripts without translation that might be recognized as aberrant RNA by host RDR to initiate siRNAs amplification. SGS3, acting as co-factor of RDR6, mainly binds to and stabilizes RNA substrates to amplify secondary siRNAs [48–50].
To visualize potential CP–RDR6/SGS3 protein associations in living cells and their subcellular occurrence, we conducted bimolecular fluorescence complementation (BiFC) assays with Agrobacterial-infiltrated leaves of N. benthamiana. For this assay, the SGS3 and RuBisco proteins (Rub) were fused with the N-terminal half of sYFP, and the tagged CPWT, CPRA and Rub proteins were fused with the C-terminal half of sYFP. YFPC-CPWT and YFPC-CPRA could be associated with YFPN-SGS3 in the cytoplasm, and formed punctate granules that co-localized with RFP-tagged RDR6 (Fig 7A, top two panels). BiFC fluorescence was not detected in the Rub control samples that were co-expressed with CP or SGS3 (Fig 7A, three bottom panels). To further evaluate the CP–SGS3 protein association in vivo, we performed co-immunoprecipitation (co-IP) assays in planta. In these experiments, Flag-CPWT, Flag-CPRA proteins were co-expressed with GFP-SGS3 or GFP proteins, and extracts from infiltrated leaves were used in co-IP assays with anti-Flag beads. Both the Flag-CPWT and Flag-CPRA, immunoprecipitated GFP-SGS3 efficiently, but the GFP protein did not (Fig 7B). Nevertheless, we could not detect the interaction of CPs and SGS3 through a yeast-two-hybrid assay (S6 Fig), hence the association of CPs and SGS3 in N. benthamiana might be indirect. Collectively, both CPWT and CPRA appear to colocalize with SGS3 in punctate granules in vivo and RDR6 also is present in these granules. However, in contrast with CPRA, the CPWT protein has a high affinity with RNA to result in general inhibition of host RNA and viral RNA translation. This inhibition results in production of aberrant RNAs that in turn appear to be associated with SGS3 and RDR6 for siRNA amplification.
In summary, we conclude that the CMV CPWT protein RNA binding contributes indirectly to high potency RNA silencing that presumably results in reduced accumulation of the CMV 2b protein. Reductions in 2b VSRs activities in turn result in a cycle in which increased siRNAs production by RDR6/SGS3 dependent amplification leads to reductions in CMV RNAs, elevated virus attenuation, and protracted symptom recovery in newly emerging leaves of infected plants.
Symptom recovery represents an extreme virus attenuation effect, in which, infected plants initially develop sever leaf symptoms, but subsequently newly emerging leaves exhibit a drastically reduced virus accumulation due to induction of antiviral RNA silencing [15, 51, 52]. In previous studies, host RNA silencing effects and interactions with the CMV Pepo strain 2b protein were shown to be involved in transient appearance of CMV in meristems at 7 dpi, and decreases in virus concentration as new leaves emerged and disappearance in recovered tissues [19, 21]. In agreement with these results, we found that CMV Fny strain also infected SAM transiently and then was excluded from the shoot apices leading to symptom recovery (Fig 1 and Fig 4). Our studies unexpectedly found that a mutant harboring an alanine substitution in the N-terminal R-rich region of the CMV CP could persistently invade meristems and block the growth of apical shoots, implying that CMV CPWT facilitates long-term SAM exclusion at late infection stages (Fig 1 and Fig 2). The previous studies have shown that the amino acid 129 of the Pepo CMV CP affects the cell-to-cell movement and determines successful SAM invasion in tobacco plants at 6–8 dpi, but the shoot meristems recovered from the Pepo infection at 21 dpi [21, 32]. Combined with previous studies, our results indicate that CMV CP is not only required for successful SAM invasion at the early infection, but also modulates viral exclusion from SAM later in infection. Thus, we propose that the CMV CP plays a critical role in compatible interactions between CMV and host plants.
Plant viral CPs have a primary function involving viral genome encapsidation, but also have been implicated in viral translation and/or replication. At low concentrations, viral CPs usually facilitate RNA replication and/or translation, whereas at higher concentrations, they may inhibit these processes in favor of virion assembly [39, 40, 44–46]. For instance, the Brome mosaic virus (BMV) CP binds to an RNA element within the 5´UTRs of the viral genome and suppresses the translation of RNA replication proteins [46]. In addition, the Hepatitis C virus (HCV) core protein binds specifically to the internal ribosome entry site (IRES) in the 5´UTR of the viral genome [47, 53–55]. This binding requires positively-charged residues in the N-terminal portion of the core protein and results in suppression of translation of downstream genes. In addition, lysine-to-alanine mutations in the N-terminal region of Red clover necrotic mosaic dianthovirus (RCNMV) CP induced more severe symptoms than wild-type virus in N.benthamiana, indicating that the lysine-rich N-terminus of RCNMV CP modulates symptomatology, independently of its role in virion assembly [56]. In line with these examples, we have shown that the CMV CP has high unspecific RNA binding activities and inhibits the translation of Luc mRNA protein in the wheat germ system (Fig 7). The mutant CPRA harboring an alanine substitution in the N-terminal R-rich region was significantly compromised in translation inhibition, implying that the basic and positively charged amino acid residues are required for translation repression by the CMV CP (Fig 7). The high concentrations of CMV CP in the wheat germ system resemble CP concentrations at late stages of infection when the CP is among the most prominent proteins in the cell. Additional experiments are required to explore the binding regions of CMV CP in the mRNA and the viral genomic/subgenomic RNA interactions. Given the well-known functions of 2b as viral virulence determinants and silencing suppression [6, 9, 26, 57–63], we postulate that decreased accumulation of 2b protein by saturated CP directly or indirectly contributes to potent antiviral RNA silencing that resulting in virus exclusion from SAM and symptom recovery. However, we cannot exclude that other host and/or viral components possibly affected by the CP are involved in symptom recovery. For example, non-specifically binding of RNAs by the CPWT protein might disturb metabolic processes of the host and lead to low virus accumulation in the SAM. Another possibility is that unstable particles assembled by the CPRA mutant permit re-infections in the same cells and therefore lead to higher RNA accumulation. Future studies are anticipated to provide exciting insights into symptom recovery processes.
In the process of RNA silencing, host RDRs are required to initiate or amplify RNA silencing via dsRNA synthesis, and the substrates for dsRNA synthesis in vivo are aberrant RNA lacking a cap structure or poly(A) tails. Arabidopsis ETHYLENE-INSENSITIVE5 (EIN5)/ EXORIBONUCLEASE4 (XRN4) encodes a cytoplasmic 5′-3′ exoribonuclease that degrades RNA intermediates derived during mRNA decay and/or RISC slicing, and regulates RDR6-dependent production of siRNAs [64, 65]. Arabidopsis Super-Killer2 (SKI2) functions as a cytoplasmic SKI complex to unwind RNAs into the 3´to 5´exoribonuclease complex for decay, and acts as a repressor of endogenous PTGS [66]. Therefore, both 5´to 3´and 3´to 5´cytoplasmic RNA decay pathways function in RDR-dependent silencing. Here, we propose that the high binding activity of CPWT with RNA protects viral RNA intermediates from RNA decay, which increases the substrate concentration of RDR/SGS3 complex and subsequently improves host antiviral silencing. We consistently found that CMVWT-induced antiviral silencing is significantly compromised in emerging leaves of sgs3 and rdr6 Arabidopsis mutants compared with those of wild-type Arabidopsis plants (Fig 4).
As the result of our comparative analyses of CPWT and CPRA in the virus infection and GFP co-infiltration assays, we propose the schematic model shown in Fig 8. At early stages of infection, a low level of CP fails to efficiently inhibit 2b protein accumulation or to induce siRNA amplification, which facilitates high-speed replication of viral RNA. Later in infection, abundant CP binds to viral RNA to initiate virion packaging, inhibit RNA translation and facilitate ribonucleoprotein formation for viral movement, as is consistent with the CP functions of other plant viruses, such as BMV and PVA [44, 46]. Simultaneously, saturated CP results in decreased accumulation of 2b protein, which interferes with inhibition of host antiviral RNA silencing. In our experiments, the CP bound RNAs failed to participate in translation and the mRNAs likely were recognized as aberrant RNAs by the RDR/SGS3 complex, and used as substrates for siRNA amplification. Accordingly, the high amounts of CP found at late infection stages reduces synthesis and accumulation of the 2b protein and/or induces siRNA amplification to culminate viral clearance from the shoot apices of infected plants (Fig 8). By comparison, the much lower affinity of CPRA with viral RNA fail to reduce the accumulation of 2b protein efficiently, or induce siRNA amplification. Therefore, we propose that the VSR activities of 2b facilitate continuous infection of CMVRA in SAM regions. Although, it must be noted here that the proposed model is based only on N.benthamiana and Arabidopsis as the host plants, but since CMV infects more than 1000 species, our proposed model provides the basis for a variety of other experiments in a diverse assay of host plants.
Since a relative healthy host provides a better environment for multiplication within-host and between-host transmission, many plant viruses down-regulate viral virulence to avoid severe disease in order to promote the survival of their hosts [67]. For instance, the strong suppressor P0 suppressor encoded by poleroviruses does not accumulate to detectable levels because of suboptimal translation initiation, which leads to low suppressor activity and reduced viral pathogenicity [68]. The Tobacco mosaic virus movement protein also regulates the spread of RNA silencing to self-control viral propagation [69]. Here, our study has revealed a novel self-attenuation mechanism in which the suppression effects of the 2b protein are down-regulated by saturated CP.
Collectively, the roles of the interactions between RNA silencing and virus-encoded suppressors in the co-evolution of hosts and pathogens have been extensively investigated. Our results demonstrate that the CMV CP serves as a central hub to facilitate regulation of dynamically integrated connections between antiviral silencing and VSR activities.
N. benthamiana plants were grown in a growth room with a controlled environmental climate programmed for 16 hours (hrs) of light at 24°C and 8 hrs in the dark at 21°C. Seedlings with six to eight fully expanded leaves were used for virus inoculations. For Agrobacterium tumefactions constructions, the cDNAs of RNA1, RNA2, and RNA3, as well as RNA2-Δ2b and RNA3-CPRA were amplified, digested with Stu I and BamH I, introduced into pCass4-Rz, and transformed into A. tumefaciens strain EHA105 [70]. Equal amount of agrobacteria harboring CMV plasmid derivatives were mixed and infiltrated into N. benthamiana leaves as described previously [71]. All the experiments were repeated at least three times with reproducible results. Arabidopsis thaliana rdr6-15 (SAIL_617_H07) and sgs3-1 in Columbia (Col) ecotype were described previously [6]. After vernalized in the dark at 4°C, the seeds were transferred into a growth room with the condition of 10 hrs in light and 14 hrs in dark at 22°C. CMVWT and CMVRA virions propagated in N. benthamiana leaves were purified and used as a inocula at 100 μg/mL.
Shoot and floral apices of infected plants were collected from infected N. benthamiana, embedded in wax, sectioned, and in situ hybridized as described previously [14]. CMV RNA was detected with the digoxigenin (Roche Diagnostics GmbH) labelled vitro-transcribed RNA fragment corresponding to 3′ terminal 200 nucleotides of CMV RNA3, and then detected with antibody anti-digoxigenin conjugated to alkaline phosphatase (Roche Diagnostics GmbH) with nitroblue tetrazolium (NBT) and 5-Bromo-4-Chloro-3-Indolyl Phosphate (BCIP, Sigma). Stained samples were examined with a bright-field microscope (DP72, Olympus) for visualization and photography.
For transient protein expression in N. benthamiana leaves, CMV CPWT, CPRA and 2b cDNAs were introduced into the pGD binary vector [72]. Leaves of 4-week-old N. benthamiana plants were co-infiltrated with mixed Agrobacterium cultures harboring the positive sense GFP (sGFP) expression plasmid with different combinations of empty vector (V), 2b, and CP plasmids. At 5 dpi, GFP fluorescence in infiltrated leaves was recorded under a long wavelength UV lamp (UVP, California, USA) using a 600D Cannon digital camera [62]. Local suppression assays were independently performed at least three times with reproducible results.
The topmost infected leaves of 10 to 15 plants were pooled for RNA extraction with Trizol reagent according to the manufacturer instruction (Invitrogen, USA). As described previously [6], 5 μg and 10 μg total RNAs were used for detection of viral RNA and vsiRNAs, respectively. CMV genomic and sgRNAs cDNA detection probes from the 3′ terminal 240 nt of Fny-CMV RNA2 was randomly labeled with [α-32P] dCTP. VsiRNAs were detected by the labeled DNA oligonucleotides corresponding to CMV gRNA3 as described previously [6]. The upper uninoculated leaves inoculated with CMVWT-Δ2b or CMVRA-Δ2b were collected for RNA extraction and detection at 7 dpi. Total RNA was treated with RNase free-DNase I, and used as a template for first-strand cDNA synthesis with M-MLV reverse transcriptase (Promega, USA) as described previously [73]. Viral infections were monitored by RT-PCR using primers corresponding to the RNA3 CP region. Protein phosphatase 2A (PP2A) was used as an RT-PCR control. For local silencing suppression assay, 5 μg and 10 μg total RNA from infiltrated leaves were used for GFP mRNA and siRNA detections, respectively. The randomly-labeled cDNA probe corresponding to GF region (nt 1–400) of GFP cDNA was used for GFP mRNA detection. The GFP-derived siRNA was detected by [α-32P] UTP-labeled RNA probe corresponding to GF region of GFP.
Total proteins extracted from viral inoculated or agro-infiltrated leaf tissue with SDS buffer [100 mM Tris (pH 6.8), 20% glycerol, 4% SDS, and 0.2% bromophenol blue, 10% β-mercaptoethanol] were separated in SDS-PAGE gels, and transferred onto nitrocellulose membranes. Anti-GFP (1:1000), -CP (1:4000), and -2b (1:2000) polyclonal antibodies were used to detect accumulation of GFP, CP, and 2b, respectively. Then, goat anti-rabbit IgG horseradish peroxidase conjugate at a 1:3000 diluted was applied as the secondary antibody, and the membranes were incubated with Pierce ECL Plus chemiluminescent substrate before exposure to x-ray films.
Firstly, amplified CPWT and CPRA cDNAs were introduced into pGEX-KG vectors respectively, and the recombinant plasmids were transformed into BL21. After induction with 0.2mM isopropyl β-D-thiogalactoside(IPTG) at 18°C for 18 h, the resulting GST-tagged fusion proteins and GST tags were purified over Glutathione Sepharose 4B (GE Healthcare) affinity columns according to the manufacturer’s instructions. The RNA binding assays were performed via a described North-Western blot procedure as described [74]. Briefly, 5 μg of purified GST, GST-CPWT, or GST-CPRA were separated by 12.5% SDS-PAGE and transferred to nitrocellulose membranes. The membranes were incubated with renaturation buffer (50 mM Tris-HCl, pH 7.5, 0.1% TritonX-100, 10% glycerol, 0.1 mM ZnCl2 and 250 mM KCl) overnight at 4°C. Then, the membranes were transferred into binding buffer (10 mM Tris–HCl, pH 7.5, 1 mM EDTA, 100 mM NaCl, 0.05% Triton X-100, and 1X Denhardt’s reagent) containing digoxigenin-11-UTP-labelled (Roche) RNA probe corresponding to CMV RNA 4 or Luc RNA. The bound RNA was blotted with the anti-digoxigen conjugated alkaline phosphatase (1:3000 dilution, Roche) in a NBT/BCIP solution.
The in vitro translation assay were performed as described previously [75]. First, the full-length luciferase (Luc) cDNA was introduced into the pMD19-T vector. Then, the bacteriophage T7 promoter and a poly(A) tail were inserted at the Luc cDNA N- and C- termini, respectively, and the resulting Xba I-linearized plasmid was used as a template for in vitro transcription by the mMESSAGE T7 kit (Ambion, USA). Two micrograms (μg) of Luc mRNA and different concentrations of purified GST-CPWT, GST-CPRA or GST were translated in the Wheat Germ Extract Plus kit (Promega, USA) for 2 hours at 25°C. Then, luciferase activity of translated products was determined with a 20/20 luminometer (Promega, USA) as described previously [76].
Coimmunoprecipition of CMV CP and AtSGS3 was performed as described previously [77]. N. benthamiana plants were agroinfiltrated for transient expression of Flag-CPWT, Flag-CPRA with GFP-SGS3 or GFP and leaf tissues were homogenized in coimmunoprecipition buffer [10% glycerol, 25 mM Tris-HCl (PH7.5), 200 mM NaCl, 1 mM EDTA, 0.1% Tritonx-100, 2% PVP-40, 50 μM MG132, 10 mM DTT and cocktail]. After filtration and centrifugation, the supernatants were incubated with anti-Flag M2 affinity gel (Sigma) for 2 hrs followed by washing five times. The immunoprecipitated products were boiled with SDS buffer for Western blotting assays with corresponding antibodies.
BiFC assays were performed with minor modifications as described previously [78]. AtSGS3 and CPWT/RA cDNA fragments, and were the RuBisco control protein (Rub) were cloned into the BiFC vectors pSPYNE-35S and pSPYCE-35S, respectively. A. tumefaciens EHA105 strains containing the recombinant BiFC plasmids and the tomato bushy stunt virus P19 plasmid were co-infiltrated into N. benthamiana leaves at a final ratio of 0.5:0.5:0.3 (OD600) and epidermal cells of infiltrated leaves were observed for fluorescence analysis (YFP) at 2 dpi using confocal laser scanning microscopy (CLSM) (Olympus FV1000).
|
10.1371/journal.pgen.1000269 | A RAC/CDC-42–Independent GIT/PIX/PAK Signaling Pathway Mediates Cell Migration in C. elegans | P21 activated kinase (PAK), PAK interacting exchange factor (PIX), and G protein coupled receptor kinase interactor (GIT) compose a highly conserved signaling module controlling cell migrations, immune system signaling, and the formation of the mammalian nervous system. Traditionally, this signaling module is thought to facilitate the function of RAC and CDC-42 GTPases by allowing for the recruitment of a GTPase effector (PAK), a GTPase activator (PIX), and a scaffolding protein (GIT) as a regulated signaling unit to specific subcellular locations. Instead, we report here that this signaling module functions independently of RAC/CDC-42 GTPases in vivo to control the cell shape and migration of the distal tip cells (DTCs) during morphogenesis of the Caenorhabditis elegans gonad. In addition, this RAC/CDC-42–independent PAK pathway functions in parallel to a classical GTPase/PAK pathway to control the guidance aspect of DTC migration. Among the C. elegans PAKs, only PAK-1 functions in the GIT/PIX/PAK pathway independently of RAC/CDC42 GTPases, while both PAK-1 and MAX-2 are redundantly utilized in the GTPase/PAK pathway. Both RAC/CDC42–dependent and –independent PAK pathways function with the integrin receptors, suggesting that signaling through integrins can control the morphology, movement, and guidance of DTC through discrete pathways. Collectively, our results define a new signaling capacity for the GIT/PIX/PAK module that is likely to be conserved in vertebrates and demonstrate that PAK family members, which are redundantly utilized as GTPase effectors, can act non-redundantly in pathways independent of these GTPases.
| Cell migration is essential for the development and maintenance of metazoan tissue. A migrating cell must navigate through complex environments and properly interpret the signals present in its path. This cellular movement is accomplished through transduction of the signals into directed reorganization of the cellular structure. Among the most important molecules that orchestrate signals from the exterior of the cells into cellular movement are the small GTPases, which function in intracellular signal transduction cascades. We have studied the interactions between GTPases, their effectors, and the environmental signals during cellular migrations in C. elegans. We have found that while some GTPases do control the guidance of these migrating cells, a certain highly conserved complex of proteins thought to be involved in mediating GTPase signaling during cellular migrations in fact functions independently of these GTPases to specifically control the structure and movement of the migrating cells. These results have revealed an unexpected role of a well-known and highly conserved signaling complex, which is particularly important since members of this complex are associated with human mental retardation. Our results may imply that the disease phenotype is likely more complex than previously thought and may in fact occur from disruption of this novel pathway.
| The GIT/PIX/PAK signaling pathway is a highly conserved signaling module which controls cytoskeletal dynamics across metazoans. The functions of this signaling complex are diverse. In humans it controls the migrations of fibroblasts through modulation of adhesion complexes, and participates in T cell receptor signaling in the immune system. The GIT/PIX/PAK complex has also been shown to regulate neuronal plasticity and development in the nervous system [1]–[3]. The importance of this protein complex is further highlighted by the observation that in humans a loss of either PAK3 or αPIX leads to impaired function of the nervous system from nonsyndromic mental retardation [4],[5]. To further understand how this complex functions in a well-defined in vivo system, we have isolated the C. elegans orthologs of the GIT/PIX/PAK complex and studied their roles in the migrations of the gonad distal tip cells (DTCs).
PAKs are downstream effectors of RAC and CDC-42 GTPases [6]. RAC and CDC-42 are RAS superfamily GTPases of the RHO subtype and are known to control cytoskeletal dynamics through their function as molecular switches [7]. In the canonical GTPase/PAK pathway, an activated RAC or CDC-42 GTPase binds to PAK and stimulates the activation of PAK's kinase activity. Despite the importance of the canonical GTPase/PAK pathways it has become increasingly clear that PAKs can also function in non-canonical pathways independent of GTPases [8]. While studies in vertebrates have indicated the likely existence of GTPase-independent PAK activation pathways the mechanistic details, biological relevance and prevalence of these pathways remain poorly understood.
GIT and PIX have been shown to regulate cellular processes through PAKs in diverse model systems [2],[3],[9]. It is generally thought that GIT/PIX/PAK pathways utilize GTPases, as PIX contains a clear GEF (guanine exchange factor) domain for RAC and CDC-42 GTPases and all of these proteins control the same cellular processes. Recently two reports have indicated a possible GTPase-independent GIT/PIX/PAK signaling pathway is likely to exist. These studies found in vitro that PAK can be activated by PIX and GIT in the absence of a GTPase-PAK interaction. In the first of these studies it was shown that a guanine exchange factor (GEF) deficient PIX can activate PAK, while the second study demonstrated that the ARF GAP (ADP-ribosylation factor GTPase activating) domain of GIT can activate PAK [10],[11]. These two studies suggested that the GIT/PIX/PAK complex can function independent of GTPases but the possible in vivo function of this pathway remains unclear.
We find that in C. elegans the PAKs, RACs, CDC-42, GIT and PIX are all involved in gonad morphogenesis. During gonad development the DTC functions as a leader cell to direct its elongation [12]–[14]. The movement of the DTC is controlled by guidance molecules [12],[15],[16], as well as other factors that are associated with the formation and regulation of the extracellular matrix (ECM) [17]–[22]. However, little is known about the signaling pathways that transduce these environmental cues into directed cell movements. Here we define two distinct signaling pathways that control the guidance of the DTCs during gonad morphogenesis. One is a typical GTPase/PAK pathway that utilizes either PAK redundantly while the other is a GIT/PIX/PAK pathway that also controls the shape and migration of the DTCs. Remarkably we find that the highly conserved GIT/PIX/PAK complex is specific for one of the PAKs and functions in a novel RAC/CDC-42 independent manner during these processes.
While investigating the roles of the C. elegans PAKs we found that the two PAKs, pak-1 and max-2 are redundantly required for proper formation of the gonad. In wild type animals the two DTCs function as leader cells to guide the elongating gonads, which eventually form two bilaterally symmetric U shaped gonad arms (Figure 1A). The elongation of gonad during morphogenesis occurs in three phases (Figures 1A–C). In the first phase, the DTC leads the developing gonad away from a mid-body position along the ventral side of the animal. In the second phase the DTCs turn orthogonally and migrate towards the dorsal side of the animal. In the third and final phase, the two DTCs turn back and then migrate towards each other, reaching the vulva by the young adult stage. Throughout gonad elongation the DTCs exhibit a sharp tapering morphology such that they have a cone-like shape when viewed from the side (Figures 1B–C). In order to understand the role of PAK signaling pathways in gonad morphogenesis, we first examined individual pak mutants.
pak-1 mutants were found to display mild defects in DTC morphology and migration (Figures 1D–E and 2B). In pak-1 mutants the DTCs generally lacked the sharp tapering morphology of wild type DTCs and instead had a bloated or distended structure (morphology defect) (Figures 1D–E). The pak-1 mutant DTCs also often failed to migrate all the way to the vulva (migration defect) (Figure 2B). max-2 mutants did not exhibit any of these defects. To reveal redundancy between these genes we examined PAK double mutants for gonad defects. The pak-1 and max-2 mutants used are putative null alleles [23]. The pak-1(ok448) allele has a deletion that removes most of the kinase domain, results in a frame shift and introduces an early stop codon. The max-2 allele nv162 has a deletion that removes the start codon, the first 4 exons and does not contain another in frame start codon until midway through the kinase coding sequence. The max-2(cy2) allele contains a missense mutation resulting in a glycine to glutamate substitution at a conserved residue in the kinase domain. The pak-1;max-2 double mutants exhibit even more severe defects than pak-1 single mutants. In addition to morphology defects, the DTCs in pak-1;max-2 double mutants wandered during their migrations (guidance defect), failed to execute at least one of the turns and did not migrate completely to the vulva (Figures 1H–M). These results demonstrate a role for the PAKs in regulating DTC morphology, migration and guidance during gonad morphogenesis, and suggest that the two PAKs are only partially redundant, such that there is a role for PAK-1 in regulating DTC morphology and migration that MAX-2 does not fulfill.
PAKs are the best known RAC GTPase effectors. There are three rac genes in C. elegans: ced-10, mig-2 and rac-2/3 [24]. The RACs themselves are required for DTC guidance, and they are partially redundant with each other for gonad development [24],[25]. We therefore investigated whether the PAKs act with the RACs in DTC guidance. We made use of the following rac mutants: for mig-2 we utilized the putative null allele mig-2(mu28). As CED-10 is required for embryogenesis, we utilized the ced-10(n1993) allele which is expected to be a strong loss of function. Because of the presence of the gene duplication in rac-2/3 we utilized RNAi for the rac-2/3 loss of function analysis. As previously reported we observed characteristic extra turns during the last phase of the DTC migrations resulting from a loss of function in any of the racs (Figure 1N). We then examined double mutants of the two paks (pak-1 and max-2) with the racs. Mutations in max-2 did not enhance the DTC guidance defects of any of the racs (Figure 1N), indicating that MAX-2 works with the RACs in DTC guidance. In contrast, pak-1 mutants severely enhanced the DTC guidance defects of any of the rac mutants (Figure 1N), indicating that PAK-1 acts at least partly in parallel to the RAC GTPases.
To identify factors that may function with PAK-1 in the RAC-independent pathway, we examined genes that are known to interact with PAKs in other species. In this manner we identified orthologs of vertebrate PIX and GIT genes, which are referred to as pix-1 and git-1 respectively. PIX and GIT proteins are highly conserved among worms, flies, mice and humans (Figure S1). Utilizing putative promoter regions from the two genes to drive GFP expression in C. elegans we studied their expression patterns and found that both genes were expressed in the DTCs throughout the DTC migrations. (Figure S1). To begin to address the functions of pix-1 and git-1 in the DTCs we examined deletion mutants for gonad defects. The allele pix-1(gk416) has a deletion beginning 4 codons after the translational start, which removes the entire SH3 domain and is expected to result in a very early stop codon due to a frame shift. The nature of this deletion indicates that gk416 is a null allele. For git-1 we utilized git-1(tm1962) which contains a 484 bp genomic deletion (in frame) resulting in 133 amino acids of the protein being deleted including the second GIT domain. As this domain is required to bind PIX in fibroblasts [9], any functional protein generated in the mutant is expected to be unable to bind PIX-1. The tm1962 deletion likely results in a strong loss of function.
Similar to pak-1 mutants, the pix-1 and git-1 mutants exhibited the characteristic defects in DTC migration and DTC morphology (Figures 2A–D), but not the DTC guidance defects seen in the rac single mutants or in the pak-1;max-2 double mutants. We then tested whether pak-1, pix-1 and git-1 function together in a pathway by examining all possible double mutant combinations. Double mutant combinations of pak-1, pix-1 and git-1 did not enhance the DTC defects relative to the strongest single mutant (Figure 2H). Nor was there any statistical difference in DTC defects between the triple mutant and any of the single mutants (Figure 2H). Interestingly, the characteristic morphology defects found in the single, double or triple mutants of pak-1, pix-1 or git-1 could be observed in actively migrating DTCs (compare Figure 2I with 2J–L). Collectively these data indicate that GIT-1/PIX-1/PAK-1 signaling complex is required for the proper migration of the DTCs and the regulation of their cellular morphology and this pathway is not redundant with the classical RAC/PAK pathway.
In contrast, MAX-2 functions in parallel to PIX-1 and GIT-1 to mediate DTC guidance. In addition to the morphology and migration defects of the GIT/PIX/PAK pathway, double mutants of max-2 with any of the genes from this pathway (pak-1, pix-1 or git-1) also showed major guidance defects in all stages of gonad elongation (Figures 2E–H). These results indicate that PAK-1, PIX-1 and GIT-1 function in a redundant DTC guidance pathway in parallel to MAX-2. As MAX-2 works with the RAC GTPases this suggests that GIT-1 and PIX-1 may also function independent of the RACs. In support of this, pix-1 and git-1 mutants also profoundly enhance the guidance defects resulting from a loss of function in the rac genes (Figure 3). The gonadal defects seen in pix-1;rac and git-1;rac double mutants were similar to the pak-1;rac double mutants, which in turn were similar to the double mutants of the git-1/pix-1/pak-1 pathway with max-2.
In summary, our mutant analysis showed that any mutant in the GIT-1/PIX-1/PAK-1 pathway led to migration and morphology defects of the DTCs, while a loss of any of the racs (which work in a pathway with max-2) led to guidance defects of the DTCs. Double mutants between these pathways led to severe defects in DTC guidance. Taken together, these results indicate that there are at least two distinct PAK pathways controlling DTC guidance during gonad morphogenesis: one is a classical RAC/PAK pathway, in which both MAX-2 and PAK-1 are utilized. The other is RAC-independent PAK pathway, in which PAK-1 (but not MAX-2), PIX-1 and GIT-1 are utilized and this latter pathway is used non-redundantly to regulate DTC migration and morphology.
Since the GIT/PIX/PAK pathway functions independent of RAC GTPases, we next sought to explore whether the pathway functions independent of other GTPases. CDC-42 is also a RHO subfamily GTPase that has been shown to activate PAKs, and PIX is predicted to also be a GEF (Guanine Exchange Factor) for CDC-42. If the GIT/PIX/PAK pathway does function independent of CDC-42, knocking down CDC-42 would enhance the defects resulting from a loss of the GIT/PIX/PAK pathway. As CDC-42 is required for viability, we utilized tissue specific RNAi [26] in the post-embryonically born DTCs, to bypass the embryonic requirement for CDC-42. Double RNAi of cdc-42 and pix-1 caused much more profound defects than RNAi of either of them alone. However, RNAi of max-2 did not enhance the defects caused by RNAi of cdc-42 (Figure 4A). This data indicates that the GIT/PIX/PAK pathway may function independently of CDC-42. Interestingly the lack of enhancement with the cdc-42; max-2 double RNAi may suggest that MAX-2 works with CDC-42 during gonad elongation. However these negative results are less than definitive as RNAi causes a partial loss of function and simply may not cause enough of a knock down to generate any possible enhancement of the cdc-42 phenotype.
If PAK-1 functions independently of CDC-42 during DTC migrations, PAK-1 may not require conserved amino acids that allow it to bind GTPases. To test this we selectively altered PAK-1 at an amino acid in the GTPase binding domain (pak-1(S76P)) that in other systems has been shown to be required for binding to CDC-42 and that is likely to disrupt binding to all GTPases [11]. We found that both wild type and the mutant PAK-1 partially rescued the pak-1 gonad morphology defects (Figure 4B). This suggests that activation by CDC-42 is not necessary for the non redundant PAK-1 function in DTC morphology and migration. As an important control, we tested pak-1(S76P) in the guidance of motor axons, where we showed previously that PAK-1's function is RAC dependent [23]. As expected, injecting pak-1(S76P) failed to rescue the axon guidance defect in pak-1 mutant, while injecting wild-type pak-1 gene did (Figure 4C). This latter result also indicates that the mutated PAK-1 loses its ability to interact with RAC GTPases. Collectively our results demonstrate that the GIT-1/PIX-1/PAK-1 pathway functions at least partly independent of RAC and CDC-42 GTPases.
To gain insight into these distinct pathways controlling gonad morphogenesis, we used fluorophore tagged proteins to examine the subcellular localizations of the PAKs, PIX and GIT in migrating DTCs in vivo. These tagged proteins rescued the DTC defects when expressed in the DTCs of the respective mutants (Figure S2). We found that the tagged PAKs were diffusely present throughout the cytoplasm of the DTC during all stages of its migration (Figure 5A–B) suggesting that the PAKs function through a transient local activation mechanism. In contrast, both GIT-1::GFP and PIX-1::GFP localized to punctate structures in the DTC during its migrations (Figure 5C–D). These puncta were observed throughout the cytoplasm of the migrating DTC. In addition, we found that GIT-1::GFP and PIX-1::mRFP co-localize throughout all of the phases of the migrating DTC (Figure 5E–T). The extent of co-localization is nearly complete as there were few, if any, sites in the DTC where the RFP and GFP signals did not overlap (Figure 5Q–T). These results indicate that C. elegans GIT-1 and PIX-1 are likely to interact directly, as has been repeatedly observed for their orthologs in a variety of different systems [1],[2],[27]. The punctate pattern is highly reminiscent of the localization of GIT/PIX in other systems where they have been characterized as forming large multimeric complexes that are thought to be scaffolds for intracellular signaling [28]. Collectively our results suggest that the GIT/PIX complex locally activates PAK-1 from a reservoir of cytoplasmically localized inactive PAK-1.
PAKs, PIX, GIT and RACs have all been implicated in integrin-regulated processes in other model systems [9],[29]. To explore whether integrin signaling in the DTC is mediated by PAK signaling pathways, we first examined the phenotypes of integrin mutants by RNAi. Integrins function as heterodimers that consist of alpha and beta subunits. C. elegans genome contains two alpha (ina-1 and pat-2) and a single beta (pat-3) subunits. The integrins have previously been implicated in controlling DTC migration [15],[30],[31]. As all of the integrin genes are required for embryogenesis, we examined their function in DTCs with tissue specific RNAi. We found that a loss of function in any of the integrin genes led to similar defects as those we observed in the double PAK pathway mutants. The integrin mutants have both the severe migration and guidance defects of (rac/max-2);(pak-1/pix-1/git-1) double mutants as well as the bloated DTC morphology phenotype observed in mutants of the GIT-1/PIX-1/PAK-1 pathway (Figures 6A–E). This was also observed in the two available ina-1 hypomorphs gm39 and gm144 (data not shown). Unfortunately we were unable to generate double mutants of these hypomorphs with either of the paks, leading us to conclude that these double mutants may be unviable. Nevertheless, these data suggest that the integrins may function with both the GIT/PIX/PAK and the RAC/PAK signaling pathways.
To further determine whether the PAK signaling pathways function with the integrins, we made use of a PAT-3 beta-integrin interfering construct (beta tail) previously reported to disrupt integrin signaling in the DTCs [30]. If the PAK pathways function with the integrins, a loss of either PAK pathway may not lead to an enhancement of the defects resulting from inhibiting normal integrin signaling. However, if either of the PAK pathways functions independently of the integrins, a loss of that pathway should enhance the defects caused by inhibiting normal integrin signaling. As reported, we found that tissue-specific expression of the beta tail caused low penetrance defects in gonad morphogenesis (Figure 6F). When these transgenic lines were crossed into the triple pak-1;pix-1;git-1 mutants or were examined in a max-2 RNAi background there was no significant enhancement in the defects (Figure 6F). As a control, we also tested whether the beta tail would enhance the defects of a mutant in a pathway that is expected to function independently of integrins. We utilized the UNC-6/UNC-40/UNC-5 pathway which specifically controls the dorsal migrations of the DTCs [12],[13]. As has previously been reported, we found that loss of the unc-40 gene resulted in defects specifically in the dorso-ventral guidance of the DTCs (Figure 6F). RNAi of unc-40 in the beta tail transgenic background resulted in additive enhancement of the DTC defects. These results collectively suggest that both the RAC/PAK and the GIT/PIX/PAK pathways function with the integrins to control DTC morphology, migration and guidance.
During gonad morphogenesis, the distal tip cell (DTC) leads the elongating gonad over a long distance to reach its final destination. Several guidance and motility systems are known to facilitate the elongation of the gonad [32]. For example, a protease system that rearranges the ECM allowing motility (GON-1) and guidance (MIG-17) of the DTC are required for proper gonad elongation. Another is the UNC-6/UNC-40/UNC-5 system which specifically directs the dorsal (phase 2) turning of the gonad. Finally there is the integrin system, which controls multiple aspects of gonad elongation by coordinating the interactions between the ECM and the DTC [15],[30],[31]. We have extensively studied the signaling pathways inside the DTC that are regulated by PAKs during gonad morphogenesis, and have identified two distinct PAK signaling pathways that differentially control the morphology, migration and guidance of the DTC. Our analysis also suggests that these PAK pathways are regulated through integrin signaling during gonad elongation.
The two PAK signaling pathways are a classical RAC dependent PAK pathway and a RAC/CDC-42 independent GIT/PIX/PAK pathway. Both pathways function in the guidance of the migrating DTC, but only the latter is required for maintaining the DTC morphology during DTC migrations (Figure S3). What are the roles of PAK-1 and MAX-2 in these two separate pathways? Although our genetic analysis indicates that PAK-1 contributes significantly to the GIT/PIX/PAK signaling pathway, PAK-1 also likely functions in the RAC/CDC-42 dependent pathway. This conclusion comes from the observation that max-2 single mutants do not yield DTC guidance defects yet double pak-1;max-2 mutants have profound DTC guidance defects. Therefore a loss of max-2 is being compensated for by the presence of pak-1. However, we also find that double mutants of max-2;pix-1 or max-2;git-1 are profoundly defective in guidance even though there is still a functional PAK-1 present. These results suggest that PAK-1 by itself cannot completely compensate for MAX-2 in DTCs. One possible explanation is that pak-1 is only partially redundant with max-2, perhaps due to differential kinase specificity of MAX-2 and PAK-1 while acting as RAC effectors. An alternate interpretation is that the loss of a functional PAK-1/PIX-1/GIT-1 pathway sensitizes the system such that the entire RAC pathway must now remain intact. The latter is supported by our observation that the loss of any component of the PAK/PIX/GIT pathway causes major DTC guidance and migration defects when combined with the loss of any of the racs (Figures 1N and 3).
That git-1and pix-1function together with pak-1 in a genetic pathway in C. elegans strongly supports the notion that these genes have a conserved function across phyla. In addition to our results these proteins have been implicated as working together to regulate cellular processes in diverse model systems. Using genetic analysis in C. elegans we demonstrate that this highly conserved GIT/PIX/PAK pathway can function independent of RAC and CDC-42 GTPases. Interestingly, only PAK-1, but not MAX-2, is required, indicating that PAKs are not redundant for this pathway, demonstrating PAK specificity in a RAC/CDC-42 independent pathway. We also attempted to address whether all GTPases are not required in the GIT/PIX/PAK pathway. We generated a mutated PAK-1 that specifically disrupts its P21 binding domain and does not bind to any GTPase, and have found that this mutated PAK-1 can still partially rescue the DTC phenotype in pak-1 mutants. Our results suggest that perhaps the GIT/PIX/PAK pathway is independent of all GTPases. In addition, our genetic and cell localization studies suggest a model where the GIT/PIX complex is selectively activating PAK-1 through a direct interaction. This conclusion is supported by previous studies in fibroblasts that GIT can activate PAK in the absence of GTPase binding [11]. Furthermore it was recently shown that in T cells a GIT/PIX/PAK pathway functions in parallel to a pathway utilizing VAV (a RAC GEF) along with RAC and PAK [33]. Together, these results suggest that the GTPase-independent GIT/PIX/PAK signaling pathway is a conserved signaling pathway utilized for multiple cellular processes.
In addition to migration defects, the GIT/PIX/PAK pathway mutants exhibit abnormal DTC morphology. Both the migration and morphology phenotypes are consistent with a defect in adhesion to the ECM substrate or the failure to execute coordinated changes in the cytoskeleton. Failure to elongate the proper distance may indicate that the DTCs have difficulty in removing/recycling their contacts with the basal lamina, which could result in the DTCs stalling prior to their targeted final destination. The bloated cell morphology may also result from an adhesion defect. The mutant DTCs may not properly adhere to their substrate and therefore adopt a less organized morphology. Similar DTC phenotypes are also observed in integrin mutants. Regulation of integrin signaling has previously been attributed to the GIT/PIX/PAK pathway in migrating fibroblasts where they are involved dismantling the integrin associated adhesion complexes. Interestingly, orthologs of PAK-1, PIX-1, and GIT-1 are all known to be involved in turnover of focal adhesions [9],[34], and GIT has also been reported to cycle between several different locations including the focal adhesions and cytoplasmic structures [35]. Taken together, it is likely that the GIT/PIX/PAK pathway functions to control either the sorting or the stability of integrin based organization of the cytoskeleton of the migrating DTC.
Our genetic analysis indicates that the two distinct PAK signaling pathways are functioning with the integrins during gonad morphogenesis. First, the overall integrin mutant phenotypes are similar to the combination of mutants from the GIT/PIX/PAK pathway and the RAC/PAK pathway. Second, an interfering construct that is reported to perturb integrin signaling and does cause a gonad phenotype does not significantly enhance the defects of mutants from either of the PAK signaling pathways. Collectively these data support the model that the PAK pathways are all functioning with the integrins. Unfortunately due to the lack of a viable null mutant in any of the integrin subunits our results are less than definitive and there are caveats to our conclusions. First, phenotypic similarity just suggests that they control the same process and does not necessitate that they function together to control that process. Second, the interfering construct causes only weak defects. Because of this we tested whether the construct could enhance an unrelated pathway (UNC-6/UNC-5/UNC-40) and we found that it did enhance this pathway. This clear enhancement of an unrelated pathway strengthens the significance of the non-enhancement with the PAK pathways result and indicates that the interfering construct is likely to disrupt aspects of the integrin signaling pathways that are involved with the PAK signaling pathways. The simplest explanation of our results is therefore that the PAK pathways act with integrin signaling.
It is well known that the RACs are highly redundant for many processes. In C. elegans the RACs are only partly redundant. The specific DTC guidance defects in single rac mutants (an inappropriate reversal of direction in the final phase of migration) indicate that RAC GTPases are required in a non redundant manner at a specific stage in DTC guidance. It was previously reported that the RACs act with each other to inhibit this extra turn [24]. Such a lack of redundancy in the RAC GTPases may result from RAC specificity at the level of the RACs activator's (the GEFs) or at the level of the RAC effectors. Our results here do not address the redundancy of the RAC GTPases, but they do indicate that any such effector specificity is not occurring through the PAKs (PAK-1 and MAX-2). Instead our results indicate that PAKs are always redundant as RAC GTPase effectors. That is to say either PAK can be activated by any of the RACs. This model predicts that in the case where the RACs are non-redundant either PAK can act with any RAC therefore the PAKs will still be redundant with each other. Similarly if the RACs act together to mediate a pathway the PAKs can both act at either and both steps of the pathway and will still be redundant with each other.
Our conclusion that the two PAKs are completely redundant as RAC effectors comes from multiple lines of evidence. Previously we found that the both PAKs function completely with the RACs to mediate P cell migrations. That is they are completely redundant for this process. However in commissural motor neuron axon guidance max-2 has a phenotype alone while pak-1 does not, yet the double is extremely severe (they were partly redundant) [23]. Here we find that the converse relationship is true; pak-1 has a phenotype alone and the double is very severe. Collectively examining these situations we found that if the paks are completely redundant then the individual pak mutants do not enhance the individual rac mutants. If the paks are partly redundant then the PAK with the phenotype would enhance any of the racs while the other would not enhance any of them. Our model to account for this describes that the PAKs are redundant as RAC effectors but additional PAK activators exist that do not require RAC GTPases and they activate with specificity towards the PAKs. In P cell migrations there is no such activator, in axon guidance the activator is specific for MAX-2 and during gonad morphology the activator is likely the GIT/PIX complex and it is specific for PAK-1. It is easy to speculate how such a phenomenon could arise evolutionarily. First redundancy at the level of the highly utilized RAC effector pathway would be favorable; after all the RACs themselves are highly redundant and are so in most organisms. This would favor a gene duplication of the PAKs. New roles could then evolve for the PAKs that do not come at a cost of the RAC effector pathway. This would add to the signaling capacity of a cell yet allow it to retain the improved capacity for RAC signaling arising from the gene duplication.
Finally, it is worth noting that the movement of the DTC is distinctly different from the migration of many other migrating cells. Cell migrations are typically characterized by protrusion of filopodia and lamelopodia followed by invasion of the cytosol into these structures, steadily dragging the cell forward. In DTCs we do not observe front protrusion of membranous structures. Instead the migrating DTCs maintain an arrowhead shape during migration (Figure 2), suggesting that they are not moving through a normal fibroblast type mechanism. The DTCs while migrating are also capping a rapidly growing gonad and seem to be pushed from behind by the elongating gonad. Thus the movement of the DTCs is likely to be controlled by the directional secretion of the proteases [17],[21] as well as the regulation of its contacts with the ECM. Our studies indicate that integrin signaling through a novel GIT/ PIX/PAK pathway is important for maintaining the structural integrity and regulating the ECM contacts. Further studies will be necessary to elucidate how these signaling pathways inside the DTC coordinate all these and other factors to properly direct its movement during gonad morphogenesis.
Worm cultures were maintained with standard methods [36]. All newly characterized mutants were backcrossed at least five times to wild type prior to analysis. Mutant genotypes were confirmed by PCR or direct sequencing of PCR products or by confirmation of a known phenotype. For RNAi experiments, dsRNA was microinjected into the gonad of young adult animals [37]. The following RNAi clones, Ahringer Library Clones [38] unless otherwise specified, were utilized in this study: max-2 (II 8F19), pak-1 (C09B8.7 (open biosystems)), pix-1 (made from the YK clone YK447g6), ina-1 (III 4N10), pat-2 (III 4P15), pat-3 (III 1P02) and rac-2/3 (IV 7L24).
LG II: max-2(cy2), max-2(nv162); LG IV: ced-10(n1993), eri-1(mg366); LG V: rde-1(ne215); LG X: oxIs12[Punc-47::GFP, lin-15(+)], pak-1(ok448), pix-1(gk416), git-1(tm1962), mig-2(mu28).
To score distal tip cell (DTC) defects, we analyzed young adult hermaphrodites with completely formed vulvas that had yet to pass an oocyte through the spermatheca. For each animal, the anterior and posterior gonads were scored separately. A gonad was deemed to have a DTC defect if the DTC failed to make proper turns (guidance defect), if the DTC failed to reach the vulva (migration defect), or if the DTC had a bloated structure (morphology defect). Specifically, a DTC was deemed to have a guidance defect if it lacked the characteristic U Shape. A DTC was scored as having a migration defect if the DTC was greater than 24 micrometers away from reaching the midline of the vulva. A DTC was deemed to have a morphology defect if the cells diameter (as judged by the diameter of the distal most region of the gonad) was greater than 24 micrometers. The 24 micrometer distance in migration and morphology was chosen as we found that greater than 99% of wild-type animals' DTCs (n = 80) were within this range. For graphical representations these phenotypes were combined and displayed together as the percent of animals with abnormal gonads. The DD and VD commissural motor axon guidance defects were scored as previously described [23].
The allele pix-1(gk416) which was generated by the Vancouver branch of the C. elegans Gene Knockout Consortium has a deletion beginning 4 codons after the translational start, which removes the entire SH3 domain and is expected to result in a very early stop codon due to a frame shift. The allele can be followed by the primers 416.f1 gagatacaccccgcaaaaga, 416.f2 gggaaggaacacatgaagga (internal to deletion) and 416.r1 gccgatccacgttgtaaatc. For git-1 we have utilized the tm1962 allele generated by Shohei Mitani. git-1(tm1962) contains a 484 bp genomic deletion (in frame) resulting in 133 amino acids of the protein being deleted including the second GIT domain. As this domain is required to bind PIX in fibroblasts [9], any functional protein generated in the mutant is expected to be unable to bind PIX-1. The allele can be followed by the primers 1962.f1 ttctccgttgttttcccaag, 1962. f2 gcaccagtatccgaaccacccaa (internal to deletion) and 1962.r1 tagccaatggagatggcatc.
For the tissue specific RNAi experiments we expressed an rde-1 (cDNA) in an rde-1(ne219) mutant [26] resulting in a transgenic line (HJ229) that only has functional RNAi where rde-1 is expressed. To drive the expression of rde-1 we utilized the lag-2 promoter (5′ primer ctagacagtcagcggcccataag) up to but not including the start codon and fused this to a rde-1::unc-54 3′UTR PCR fragment generated from the pKK1253 plasmid (gift from Hiroshi Qadota).
Cloning of DNA and generation of transgenes were accomplished by standard techniques. In particular we made extensive use of PCR based gene fusion and subsequent cloning of PCR products into TOPO vectors (Invitrogen).
The Ppak-1::max-2::venus construct was constructed by fusing the 5′ region of pHJ102 [23] to the 3′ region of the partial cDNA clone Y38F1A.10::venus (A gift from Queelim Ch'ng). The resulting construct contained a full length max-2 cDNA under its own promoter fused to YFP (venus). We then fused the max-2 (cDNA)::venus region to a pak-1 promoter [23] to generate Ppak-1::max-2::YFP. To generate Ppak-1::pak-1::mRFP we generated a Ppak-1::pak-1(cDNA) minigene and fused it to the mRFP::UNC-54 (3′UTR) from Punc-25::mRFP [39] (A gift from Ken-Ichi Ogura). For the PIX-1 translational reporters, we utilized the partial cDNA yk447g6 and fused it to the 5′ pix-1 genomic region ending at the second exon (5′ primer: gccatggtagtaagagcattccg). This Ppix-1::pix-1 (cDNA) minigene was then fused to mRFP or GFP as described in the preceding and following text. To generate Pgit-1::git-1::GFP we utilized the yk1688c03 (Yuji Kohara) full length cDNA and fused it to its 5′ genomic region (5′ primer gggtgaacggtcacttgactaga) generating a Pgit-1::git-1 (cDNA) minigene. This was then fused to the GFP::UNC-54 (3′ UTR) from pPD95.75 (Fire Vector Kit) yielding Pgit-1::git-1::GFP. Lag-2 promoter regions used for DTC specific expression consisted of 2,790 bp of DNA 5′ to the ORF through the start codon (5′ primer acgtcttgtaaccccctcccacc).
For microscopy animals were mounted on 2% agarose pads with 5 mM sodium azide. Animals were scored by examination with microscopy at 400× on a Zeiss Axioplan II. Confocal images were captured with a Zeiss (Thornwood, NY) LSM 510 META laser-scanning confocal microscope. Images were analyzed using Zeiss META software version 3.2 SPZ.
|
10.1371/journal.pcbi.1005432 | Reconstruction and signal propagation analysis of the Syk signaling network in breast cancer cells | The ability to build in-depth cell signaling networks from vast experimental data is a key objective of computational biology. The spleen tyrosine kinase (Syk) protein, a well-characterized key player in immune cell signaling, was surprisingly first shown by our group to exhibit an onco-suppressive function in mammary epithelial cells and corroborated by many other studies, but the molecular mechanisms of this function remain largely unsolved. Based on existing proteomic data, we report here the generation of an interaction-based network of signaling pathways controlled by Syk in breast cancer cells. Pathway enrichment of the Syk targets previously identified by quantitative phospho-proteomics indicated that Syk is engaged in cell adhesion, motility, growth and death. Using the components and interactions of these pathways, we bootstrapped the reconstruction of a comprehensive network covering Syk signaling in breast cancer cells. To generate in silico hypotheses on Syk signaling propagation, we developed a method allowing to rank paths between Syk and its targets. We first annotated the network according to experimental datasets. We then combined shortest path computation with random walk processes to estimate the importance of individual interactions and selected biologically relevant pathways in the network. Molecular and cell biology experiments allowed to distinguish candidate mechanisms that underlie the impact of Syk on the regulation of cortactin and ezrin, both involved in actin-mediated cell adhesion and motility. The Syk network was further completed with the results of our biological validation experiments. The resulting Syk signaling sub-networks can be explored via an online visualization platform.
| The complex nature of cancer hampers traditional biological approaches to unravel its molecular mechanisms and develop targeted drug therapies. Cancer affects a number of “hallmark” cellular processes controlled by multiple signaling pathways. Our goal is to identify the pathways that negatively affect tumor development and progression. We established that the Syk protein tyrosine kinase exhibits a tumor-suppressive function in breast cancer. Large scale global biochemical analyses allowed to identify Syk targets in cancer cells, but their mechanisms and interrelationships remain unknown. Our main goal was to pinpoint a limited number of biologically realistic molecular “paths” from Syk to its effectors. We therefore developed a new methodology combining graph theoretical methods allowing to reveal the shortest “paths” between “nodes” in a graph including an approach that investigates also longer “paths”. Applied to the Syk network, this method allowed us to propose and validate new signaling axes relating Syk to major effectors of the cell adhesion and mobility that are crucial cancer hallmarks.
| Tyrosine phosphorylation of proteins acts as an efficient switch allowing to control key signaling pathways involved in cell proliferation, apoptosis, migration, and invasion, and is thus involved in oncogenesis. Understanding the functioning of such complex pathways is crucial for both fundamental research and clinical applications and relies on the ability to build in-depth network models from extensive global experimental data [1–7].
The non-receptor spleen tyrosine kinase Syk has for a long time been considered as a hematopoietic cell-specific signaling molecule. In these cells, Syk is involved in coupling activated immunoreceptors to downstream signaling events affecting cell proliferation, differentiation and survival [8]. We and others have discovered that Syk is also present in non-hematopoietic cells [9–12]. More precisely, its expression was found in mammary epithelial cells and low-tumorigenic breast cancer cell lines, whereas invasive and metastatic breast cancer cells lacked Syk expression [11]. In patient samples, Syk expression exhibits a gradual loss during breast cancer progression and the low Syk levels are correlated with an increased risk of metastasis [13,14]. In hematopoietic cells, Syk functions as an essential component of the signaling machinery of multiple immune receptors and adapter proteins that are, however, not expressed in non-hematopoietic cells. Unveiling the Syk signaling pathways and tumor suppressor mechanisms is a public health issue as pharmacological Syk inhibitors are being used in clinical trials for treating auto-immune diseases [15,16].
We and other groups performed quantitative phospho-proteomic studies, based on differential Syk expression or activity, in order to identify novel Syk signaling effectors in breast cancer cells [17–19]. These approaches, however, allowed only to establish a comprehensive list of direct and indirect Syk targets. In this study, we use the data produced in these investigations to reconstruct a Syk-based signaling network and to identify the intermediary pathways via which the signal propagates from Syk to its effectors in this network.
Phospho-proteomic studies provide data sets of phosphorylated proteins with a significant “fold change” in differential experiments. Comparably to the gene data sets in transcriptomic studies, these protein targets can be used to build networks by using comprehensive interaction databases (hypergeometric test [20]; GSEA [21]; DAVID [22]; enrichment maps [23]). On the one hand, the set of identified targets is incomplete and intermediate variables and interactions are needed to describe systems-level functioning. On the other hand, the set may be too substantial and contain spurious or inessential components. Consequently, in order to obtain comprehensive networks that expose new essential constituents, network reconstruction procedures should contain both enrichment and pruning steps.
Network pruning can be performed by sub-network extraction [7,24–27]. Three main approaches based on generic graph analysis methods have been used to extract sub-networks associated to a subset of its nodes: the classical shortest path [27–29], Steiner trees [7,25,26] and random walk processes [30]. Module identification using expression data to score sub-networks [31] is a closely-related method but its aim is to identify a number of significant small sub-networks rather than produce a simplified connected network.
Here we focus on the identification of signaling pathways from a single source to a collection of targets identified in phospho-proteomic studies. In this context, the classical shortest paths approach can be inaccurate and does not take into account alternative paths, while the k-shortest paths extension generates a collection of ranked alternative paths, but relies on well-separated weights between arcs to be effective [27]. Steiner trees enable the identification of the smallest set of edges allowing to connect a set of nodes: it may lead to longer individual paths but will reduce the number of interactions when considering the dataset as a whole, with the drawback of further reducing the number of alternative paths [25]. Suboptimal solutions of the Steiner tree problem were computed for a lymphoma network [32] but the functional significance of disparate solutions was not appraised. Finally, random walk processes have been used to estimate the probability of reaching network nodes by observing information flow propagation [30,33] The random walk approach can prioritize some proteins [30,34] but does not usually aim to identify paths from the source(s) to the target(s). Additionally, it can be inaccurate and fail to render specific features of the signaling pathways because it assumes that the flow of information through the network satisfies linear equations (Laplace equation on a graph) with transitions that are mainly guided by topology. Other turn-key tools have been developed, such as Ingenuity Pathways Analysis (IPA) [35]. IPA mainly involves gene-regulatory networks and request the signs of the regulations in order to score interactions and paths. The latter information is rarely available in relation with protein modification/interaction and phospho-proteomics data. Furthermore, IPA is a proprietary software with a private database. Despite their limitations, each of these methods provides valuable information and it is useful to combine several approaches with ad-hoc adjustments when analyzing networks in relation with specific data.
In this paper we assembled the interaction network of signaling pathways controlled by Syk in breast cancer cells, exploiting existing phospho-proteomic studies. To reconstruct and analyze the signaling network, we propose a novel methodology that combines the shortest paths methods with random walk processes. We defined interaction weights based on functional annotations and experimental datasets. Mainly, we pinpointed phosphorylation-based interactions leading directly to targets whose changes of phosphorylation are significant and interactions whose sources were identified targets. These weights define transition probabilities of a Markovian random walk on the network, which are further refined by replacing them with probability currents of the stationary Markov process. Thus, the random walk is not used for pruning directly, like in other implementations of this technique[30,33], but applied for weight re-evaluation. To produce a list of biologically relevant paths relating Syk to its direct and indirect targets, we then searched for near-shortest paths in the resulting network with refined weights. Our method combines the advantages of weighted shortest path methods that take into account the functional importance of the interactions with those of the random walk methods that propagate the information on the network and smoothen large weight differences that could incidentally occur. The increase in specificity obtained by combining several sub-network extraction methods has also been exploited for analyzing metabolic networks [36]. By network pruning, relatively dense networks are downscaled to a few biologically significant alternative paths from Syk to its targets. This helps to generate hypotheses about Syk signal propagation that, however, need to be experimentally validated.
We substantiated our in silico hypotheses with molecular and cell biology experiments, and identified two candidate mechanisms that support the impact of Syk on the regulation of cortactin and ezrin, two proteins involved in actin-based cell adhesion and motility. As a result of our biological validations, we propose a new Syk-Src-cortactin signaling axis, and a direct ezrin regulation by Syk phosphorylation. The Syk network was further corroborated with biological validation experiments and exploited to generate the sub-networks of the paths from Syk to its targets involved in (i) cell adhesion and motility, (ii) cell growth and death, (iii) immunity and inflammation and (iv) cell differentiation.
The proposed sub-network extraction method was applied to connect Syk, a source in the network, to its direct and indirect targets. The same method can be applied in numerous other studies to connect several sources amongst themselves and to their targets. This method reveals important targets and interactions, allows to generate hypotheses and test new interactions. The simplified network resulting from this method provides insights into biological processes controlled by Syk and provides a challenging access to more mechanistic modeling approaches.
Starting with the set of Syk-dependent targets identified by phosphoproteomics, we identified a network explaining the propagation of the signal from Syk to its targets. In order to generate this network we interrogated comprehensive pathways databases, extracted significantly enriched pathways and consistently merged them by avoiding duplicate interactions and nodes. The result of this assembly was a connected but very large network, too difficult to analyze and containing many unessential interactions. Here, we propose several methods allowing to prune this network and to extract significant paths from it. Several paths connecting Syk to functionally important targets were biologically validated.
To bootstrap the reconstruction of a comprehensive network of Syk downstream signaling in breast cancer cells, we analyzed tyrosine phospho-proteomic data acquired in two independent mass spectrometric studies using breast cancer cell lines with modified Syk catalytic activity or protein expression. On the one hand, we selected from our own study the proteins lost or gained in tyrosine-phosphorylated protein complexes of the Syk-positive MCF7 cells treated with a pharmacological Syk inhibitor (further referenced as the MCF7 dataset) [19]. On the other hand, we identified the proteins with modified tyrosine phosphorylation after exogenous Syk expression in the Syk-negative MDA-MB-231 cells (further referenced as the MDA231 dataset) [17]. Post-treatment procedures of the original phospho-proteomic data are detailed in the Materials and Methods section. From these two studies; we respectively selected 265 and 487 proteins as Syk targets (S1 and S2 Tables). Only 64 proteins were found in common, reflecting the complementarity of the two original phospho-proteomic studies. Indeed, the phospho-tyrosine enrichment prior to mass spectrometry as well as the experimental cell models used in those studies are different (see the Materials and methods section). We analyzed the two datasets separately to evaluate whether they point to distinct or similar cell signaling pathways.
We searched for enriched pathways in the lists of Syk targets, using pathways from the KEGG database [37]. We selected pathways which contain at least one of the proteins from the target lists and used a Fisher exact test to assess their enrichment. As we are only interested in overrepresented pathways, we removed the underrepresented ones: i.e; the ones for which the ratio “number of proteins from the target list per total number of proteins” is lower than the same ratio in the background list.
Within the two rather poorly overlapping datasets, we found among the most enriched KEGG pathways those related to cell-cell and cell-substrate adhesions, actin cytoskeleton regulation and apoptosis (S3 and S4 Tables). This observation is consistent with the reported role of Syk on cell adhesion, motility, proliferation and death in breast cancer cells [19,38–43]. Furthermore, the similarity in pathway enrichment, despite the limited overlap between the two datasets, indicates their relationship. We merged the original datasets and present here the results obtained for the network reconstruction.
We assembled and explored a prior-knowledge interaction network to extract candidate mechanisms underlying the datasets. While such networks are often assembled using complete pathway or interaction databases, here we focused on previously identified enriched pathways. Using pathways rather than individual interactions will enable the extension of sub-networks with relevant interactions in their neighborhood for further analysis. By restricting our full network to a subset of pathways, some coverage may be lost but it allows to reduce the amount of irrelevant interactions and to better assess the relevance of the identified pathways. Hereafter we show that the enriched pathway can be used to identify candidate mechanisms. We also extended our search by using the Pathway Commons database [44]. KEGG provides a set of well-established pathways, while Pathway Commons allows a higher coverage by integrating pathways from multiple sources (notably Reactome, Panther, and PID). To select the most relevant pathways guaranteeing to cover most identified targets, we filtered the pathways based on their enrichment p-value and selected those presenting a p-value lower than 0.1 (in this step we aim to be as complete as possible, selecting 83 pathways from KEGG and 419 from Pathway Commons). Furthermore, we included the pathways containing Syk targets not covered by significantly over-represented pathways from the same database. This allowed to integrate 41 additional pathways from KEGG and 9 from Pathway Commons (S5 and S6 Tables). In this merged network, each protein and interaction keeps track of the list of pathways in which it is involved. We also used the GO annotation to identify the proteins involved in processes in which Syk is implicated (cell adhesion and motility, cell growth and death, immunity and inflammation, cell differentiation). The resulting oriented and partly signed network comprises 6438 proteins and 62322 interactions, from 552 pathways (124 from KEGG, 428 from Pathway Commons), covering 350 of the 687 identified targets (75 are only found in KEGG, 125 in Pathway Commons and 150 in both). Among these 350 targets, 245 are reachable from Syk (steps 1–2 in Fig 1). This network contains 979 interactions between two targets identified in the datasets. A closer scrutiny of these interactions further highlights connections between the two datasets: 146 and 260 interactions are associating targets specific to the MCF7 and MDA231 datasets respectively, 253 involve at least one target shared by the two datasets and 320 connect targets specific to different datasets.
The paths connecting Syk to the identified targets in the reconstructed network describe possible mechanisms for its signal propagation. As a massive amount of alternative paths exist, it is crucial to identify the appropriate candidates amongst the mass by extracting the sub-networks that exhibit selected connections between Syk and a specific target (step 3 in Fig 1; S1 Fig).
We focused on the signal propagation from Syk to its targets involved in cell adhesion and motility, and in particular on cortactin and ezrin, two proteins that are differentially phosphorylated in a Syk-dependent manner and that functionally link the plasma membrane to the actin cytoskeleton [45,46]. We first considered a parsimonious approach to find the paths implicating the fewest nodes, and searched for these shortest paths using the classical Dijkstra algorithm [47].
We observed that the shortest paths between Syk and cortactin contain two intermediates, with three alternative proteins directly upstream of cortactin (Fig 2A). Src is the only phospho-tyrosine modifier (tyrosine kinases and phosphatases) amongst them, suggesting that the paths involving Src are more credible to explain the change in cortactin phosphorylation.
The shortest paths network linking Syk to ezrin included more proteins and contained also two intermediate nodes (Fig 2B). These paths were related with the classical regulation of ezrin by the membrane lipid PIP2 or its phosphorylation on serine/threonine residue(s) [48,49]. None of the nodes directly upstream of ezrin were phospho-tyrosine modifiers that could explain the effect of Syk on ezrin tyrosine phosphorylation. The sub-networks obtained for cortactin and ezrin illustrate the need for careful analysis of phospho-tyrosine modifiers to explain the phosphorylation-based modifications noticed in the MDA231 dataset.
To take into account the importance of phospho-tyrosine modifiers, we extended the GO annotation of network nodes with the corresponding terms (S7 Table) and modulated the length of the path using distances attached to the interactions: a path involving more steps associated to small distances can be selected over a short path with longer steps. We assigned distances giving the priority to paths linking a phospho-tyrosine modifier upstream of proteins experimentally identified as differentially phosphorylated. We also favored the inclusion of other experimentally identified proteins by reducing the distances of their outgoing interactions.
The introduction of these distances led to more realistic suggestions for the cortactin sub-network, in which Src is the only protein directly upstream of cortactin (paths via white nodes in Fig 2A). As shown in the dedicated Results subsection, biological validation confirmed the role of Src in mediating Syk signal propagation to cortactin. However, it did not enable the identification of significantly better suggestions for ezrin, suggesting a “missing link” (paths via white nodes in Fig 2B). To resolve this inconsistency, we decided to evaluate the possibility of a direct interaction between Syk and ezrin. These results are described below and prompted us to propose that Syk directly phosphorylates ezrin.
Taking into account our biological validation, we added the new protein interaction from Syk to ezrin to the Syk network. We also extended the list of Syk direct substrates by integrating the results of a third dataset that identified the peptides phosphorylated on tyrosine by Syk after an in vitro kinase reaction [18] (S2 Table; for details, see Materials and methods). Finally, we added the direct interactions between Syk, E-cadherin and alpha-catenin, both members of the same cell-cell adhesion complex that we previously identified as direct Syk substrates [19] (step 4 in Fig 1).
The sub-networks obtained with the weighted shortest-paths analysis still contain several equivalent paths which we would like to classify further by integrating a parameter based on the network topology. A random walk process provides a “reachability” score for each protein. By taking into account both a mix of network topology and the weighted interactions, this score highlights key proteins that are involved in multiple interesting and plausible candidate paths. To integrate this parameter into our analysis method, we used these scores to refine the distances associated to their outgoing edges, allowing the shortest-path approach to also include such key proteins. The weight refinement algorithm is fully described in the Methods section.
This methodology was applied to refine analysis of the Syk signal propagation to its targets involved in cell adhesion and motility. The size of this sub-network (e.g. number of nodes and edges) decreased after modulation of the length of the paths and even more after random walk refinement (Fig 3). This ranking property was appropriate to highlight the most likely paths regarding our biological and topological criteria. These observations were confirmed by the sub-networks linking Syk to its targets involved in Syk-related cell processes (S2–S4 Figs)
Nevertheless, we considered this method as too stringent and searched for near-shortest paths instead of strict shortest paths in order to generate a set of alternatives, selecting all paths for which the total distance is up to 20% higher than that of the shortest path. Using this setting for the analysis of the signal propagation from Syk to its targets involved in cell adhesion and motility, we retrieved a sub-network as large as the one obtained without random walk refinement. Moreover, the sets of interactions linking Syk to its targets involved in Syk-related cell processes were slightly distinct as compared to the sub-networks obtained after weighted shortest paths analysis, and after refinement with random walks, both allowing a 20% overflow (Fig 4 and S5–S7 Figs). This suggests that introducing the “reachability” parameter not only ranks the alternatives, but also selects novel elements that allow to generate hypotheses on Syk signal propagation.
Taken together, the purpose of Figs 3 and 4 is to illustrate the flexibility of our method. These two figures tell us that (i) taking into account the molecular biology parameters by attaching distances to interactions for shortest-paths analysis and (ii) taking into account the network topology by the random walk refinement, leads not only to shortest paths to be biologically validated in the first instance (Fig 3, network size decrease), but also to novel alternative paths that were absent in the unweighted shortest paths analysis, and to rank more realistically the set of paths.
Weighted shortest-path analysis from Syk to cortactin pointed to the Src tyrosine kinase as the phospho-tyrosine modifier that could account for the Syk impact on cortactin tyrosine phosphorylation (paths via white nodes in Fig 2A). This hierarchy was contradictory with previous studies describing the direct interaction of Syk and cortactin in breast cancer cells [18,42,50], and the impact of Src on Syk phosphorylation in colon cancer cells, opposite to our observations [51]. To test the ability of Src to drive the Syk signal propagation, we analyzed cortactin tyrosine phosphorylation, together with Syk and Src activity in cells treated with tyrosine kinase inhibitors. Cortactin phosphorylation and Src activity, evaluated by the phosphorylation of its tyrosine 418 residue, were decreased after cell pretreatment with Syk or Src pharmacological inhibitors (Fig 5A and 5B). Syk activity, evaluated by its auto-phosphorylation on the tyrosine 525/526 residues, was not affected by Src inhibitors, demonstrating the Syk-Src-cortactin hierarchy. In the MCF7 dataset, cortactin is poorly affected by Syk inhibition. As the quantitative measurement was obtained by calculating the median SILAC ratio of several peptides from cortactin, we decided to analyze the phosphorylation of its individual peptides. The quantity of the phosphotyrosine pTyr334 cortactin peptide was ~2 fold decreased by Syk inhibition (S8 Fig). Conversely, Tyr446 phosphorylation was not affected (S9 Fig). Those observations were consistent with the data from MDA231 dataset (Fig 5C). Taken together, our results indicate that the signal transmission from Syk to cortactin is mediated by the Src kinase.
None of the nodes directly upstream of ezrin were phospho-tyrosine modifiers that could explain the Syk impact on ezrin tyrosine phosphorylation (Fig 2B). To explore this more profoundly, we evaluated the possibility of a direct interaction between Syk and ezrin. Both proteins are localized in phospho-tyrosine enriched plasma membrane extensions (ruffles) of MDA-MB-231 breast cancer cells in which actin is dynamically reorganized (Fig 6A). Colocalization of Syk and ezrin was evaluated quantitatively (Fig 6B). Purified recombinant Syk was able to induce direct ezrin phosphorylation on tyrosine residue(s) in in vitro kinase assays (Fig 6C). Moreover, in an immune-complex in vitro kinase assay using endogenous or exogenously expressed FLAG tagged or GFP fusion-proteins, ezrin was phosphorylated dependent on the Syk catalytic activity (S10A and S10B Fig). It is worth noting that tyrosine phosphorylation of Syk is also enhanced in the presence of ezrin (compare lane 3 with lane 1) in both autoradiography and Western blot analyses which may have a biological explanation. Ezrin contains a canonical ITAM motif that has been shown to interact with Syk [52] and even play a role in Syk recruitment and activation by binding to its tandem SH2 domains [53]. A positive feedback loop may thus exist as binding of Syk is activated by its SH2 domain binding to bi-phosphorylated ITAM motifs but should be explored more deeply. Detailed analysis of the ezrin post-translational modifications by isoelectric focusing revealed that its in vitro phosphorylation by Syk induced a unique phosphorylation of one third of the total ezrin protein (S10C and S10D Fig). Mass spectrometric analysis demonstrated that ezrin is phosphorylated by Syk on a peptide containing the phosphorylated Tyr424 residue (S11 Fig). The same ezrin phospho-peptide was the only one identified in the Syk-positive breast cancer cells in the MDA231 dataset. Taken together, these results prompted us to propose the new protein interaction between Syk and ezrin, leading to ezrin phosphorylation on the Tyr424 residue(Fig 6D).
In this study, we constructed a network by which the Syk tyrosine kinase acts on its targets in breast cancer cell lines. This network is comprehensive enough to cover 350 targets, which represent 50% of the Syk targets identified in several phospho-proteomics experiments. The network was further reduced by sub-network extraction which made its analysis possible. The network enrichment and pruning steps were performed by a novel bioinformatics method that combines the shortest paths method with random walk processes. We thus propose a flexible tool, well-adapted for reconstruction and analysis of signaling networks applied on phospho-proteomic data. This method confirmed that Syk is engaged in several cancer-related pathways associated to cell growth and death, adhesion, motility, polarity and cytoskeleton regulation. In addition, it led to new biological findings concerning molecular paths in breast cancer cells linking the tyrosine kinase Syk and two targets involved in cell adhesion and motility: (i) the signaling axis Syk-Src-cortactin and (ii) the direct action of Syk on ezrin.
Using datasets extracted from two published complementary phospho-proteomic studies that identified Syk targets in breast cancer cells, we reconstructed a Syk-controled molecular network by integrating the components of signaling pathways enriched for Syk targets (S1 File). Two different breast cancer cell models were used: Syk-positive MCF7 cells treated or not with a pharmacological Syk inhibitor versus exogenous Syk expression in the Syk-negative MDA-MB-231 cells. This was expected to bring some heterogeneity in the molecular paths linking Syk to its targets. Nevertheless, lists of Syk targets emerging from the different datasets showed similarities in pathway enrichment. Moreover, in the Syk network subpart containing the reachable targets, we found that one third of the interactions involving two targets connect targets identified in different datasets, which justified combining them into a unique set despite the partial overlapping. This proportion is conserved after shortest paths analysis, suggesting that a number of mechanisms by which Syk activates its targets are common to the cell models used. Although the biochemical methods to enrich protein extracts before mass spectrometry analysis were distinct (enrichment of tyrosine phosphorylation dependent-protein complexes versus tyrosine-phosphorylated peptides), they produced complementary information. On the one hand, phospho-proteomic studies at the protein level identified not only proteins that are differentially phosphorylated but also their partners present in the protein complexes. Several conserved protein domains are involved in phospho-tyrosine-dependent protein-protein interaction. On the other hand, phospho-proteomic studies at the peptide level identified the differentially phosphorylated tyrosine residues, allowing to increase the precision of our experimental validations. Knowing the phosphosites provides insight in the functional consequences of the phosphorylation and the integration of this functional information allows to construct more coherent networks.
The introduction of distances was a crucial step in our approach that allowed the selection of realistic paths. The extra random walk step enabled the refinement of these distances, so the selection of too many paths having the same length was avoided. In contrast to methods proposing a unique solution (or too many alternatives), our method allows to downscale to a few biologically significant alternative paths. The simplified network obtained by sub-network extraction allowed us to generate computational hypothesis about Syk signal propagation that were subsequently validated.
As a result of our biological validations, we proposed a new molecular explanation of the impact of Syk on cortactin, a protein involved in cell adhesion and motility by its ability to regulate the cortical actin cytoskeleton. Previous studies described cortactin as a direct Syk substrate [42,50]. However, in the phospho-proteomic data exploited in our study, the cortactin Tyr421 residue directly phosphorylated by Syk in vitro [18] does not match the Tyr residues phosphorylated in a Syk-dependent manner in cellulo [17]. In our model, we proposed a new Syk-Src-cortactin signaling axis (paths via white nodes in Fig 2A). We demonstrated that cortactin tyrosine phosphorylation induced by Syk is dependent on Src catalytic activity. Conversely, inhibition of Src did not affect the Syk catalytic activity, but both are required to induce cortactin phosphorylation in breast cancer cells (Fig 5A and 5B). This tyrosine kinase hierarchy does not match previous observations showing that Src inhibition induces a decreased tyrosine phosphorylation of Syk [51]. This discrepancy could be explained by the distinct cell models (breast cancer versus colorectal cancer cells) and by the fact that Leroy and colleagues [52] analyzed global Syk phosphorylation rather than specific phosphorylation on the Tyr525/526 residues, which are located in the activation loop of the Syk kinase domain and are more relevant to detect changes in Syk catalytic activity [54]. Multiple tyrosine phosphorylation sites have been described within cortactin (for review, [46]http://www.phosphosite.org/). We confirmed that the phosphorylation of cortactin Tyr446 residue is not directy affected by Syk (Fig 5C). Conversely, the phosphorylation of the Tyr334 residue is Syk-dependent, but can, according to our model, be phosphorylated also by Src [55]. There is currently no information about the functional consequences of this residue’s phosphorylation. Finally, the Syk-Src-cortactin signaling axis we proposed is consistent with the positive impact of Syk on E-cadherin dependent cell-cell adhesion [19,42]. Src-dependent phosphorylation of cortactin is necessary to link the E-cadherin adherens junction complex to the actin cytoskeleton, that subsequently supports cell-cell contact formation [56,57].
Our initial computational hypotheses of molecular circuits linking Syk to ezrin did not explain its tyrosine phosphorylation, by the lack of a phospho-tyrosine modifier directly upstream of ezrin (Fig 2B). We demonstrated that Syk phosphorylates ezrin in a direct manner on its Tyr424 residue and that both proteins colocalize in the plasma membrane ruffles in breast cancer cells (Fig 6A and 6B). Previous studies described a Syk-dependent tyrosine phosphorylation of ezrin in B lymphocytes on its Tyr353 residue [58,59], a site that can be also phosphorylated by the epidermal growth factor receptor (EGFR) [60]. Phosphorylation of this residue leads to activation of the ezrin downstream signaling pathways as JNK or PI3K/Akt [59,61]. Two other ezrin tyrosine residues, Tyr146 and Tyr478, can be phosphorylated by Src (Tyr146, also being phosphorylated by EGFR), mediating cell scattering and stimulating motility [62,63]. We showed here that in breast cancer cells, Syk phosphorylates ezrin on the Tyr424 residue, rather than Tyr353 as it occurs in B lymphocytes, and could induce a signaling cascade different than the ones previously described for tyrosine phosphorylated ezrin. This novel molecular regulation of ezrin could help to explain the negative impact of Syk on epithelial cell motility.
By this study, we do provide to the cancer cell signaling community access to the Syk network and sub-networks of the paths from Syk to its targets involved in (i) cell adhesion and motility, (ii) cell growth and death, (iii) cell differentiation and (iv) immunity and inflammation (see Supporting Information). As Syk is involved in breast cancer suppression, it is not surprising that these cellular processes involved in cancer progression can be affected by Syk in breast cancer cells. Nonetheless, many of the Syk interactions described in pathway databases are extracted from molecular studies in cells of hematopoietic origin. More precisely, Syk signaling has been extensively studied in B lymphocytes where it is indispensable for immune cell differentiation. Which part of Syk signaling is shared between epithelial and hematopoietic cell types could be determined by Syk related phospho-proteomic experiments with a more comprehensive collection of cell models.
At this stage, our model suggests a number of plausible mechanisms linking Syk with cancer-related cellular processes. These can be used to generate more hypotheses, validation and provide valuable inputs for further developments. A crucial issue in Syk-related research is how to activate compensatory mechanisms maintaining tumor suppression signaling even when Syk is downregulated. The network we propose here is the first step towards addressing this question. To advance towards more refined mechanistic models the annotation of the interactions, for instance the sign, should be completed by integration of more data and by formal inference methods [64,65]. The careful consideration of possible feed-backs should also be considered in these developments. For instance, in tyrosine kinase signaling, many feed-back interactions involve phosphatase players whose role and significance are largely ignored even for very well studied pathways such as MAPK. New biological experiments are needed to unravel new players and interactions. Basic networks like the one resulting from our approach can be used for planning such experiments. As a simple example, if phosphatases are present upstream in the network one would want to test the effect of their inhibition on downstream proteins.
The bioinformatics method we used to reconstruct and prune the Syk signaling network can be used in other studies, whenever the focus is on finding candidate mechanisms explaining how signals propagate in large networks and how the network state-changes under perturbations. We consider that the combination of ad-hoc distances, random walk, and near-shortest paths provides good candidate mechanisms, by implementing the following requirements: (i) minimize the number of steps (shortest paths); (ii) fit with the data (ad-hoc distances); (iii) further favor intermediates which belong to multiple appropriate candidate paths (random walk); (iv) propose and rank multiple alternatives (near-shortest paths). This strategy can be generally applied to other studies of signaling networks using datasets based on distinct post-translational modifications, separately or combined. The open source code for the network reconstruction and extraction of relevant sub-networks has been made available for the computational biology community (S1 File and https://github.com/aurelien-naldi/NetworkReconstruct).
Uniprot ID mapping from uniprot.org/downloads (2015/07)
HGNC dataset from genenames.org/cgi-bin/statistics (2015/07)
GO ontology from geneontology.org/page/download-ontology (go-basic.obo, 2015/10)
GO annotation from geneontology.org/page/download-annotations (goa_human.gaf, 2015/10)
KEGG: www.kegg.jp, release 75 (2015/07)
Pathway commons:pathwaycommons.org release 7 (2015/03)
Proteins are identified by their Uniprot IDs, without the isoform postfix. We used Uniprot ID mapping files to associate KEGG and HGNC identifiers (transferred to the corresponding gene symbols) with these Uniprot IDs. When multiple Uniprot IDs are associated to the same KEGG or HGNC ID, they were grouped and a single ID is selected for the group, preferably a reviewed entry (allowing to map unreviewed Uniprot entries to the associated reviewed entry when possible).
The network is annotated based on the dataset, the pathways, and the GO annotation. Nodes and interactions keep track of the list of pathways in which they appear. Interactions types (phosphorylation, modification, regulation) and signs (+ or -) are transferred from the pathways when available. When the same interaction is described in several pathways, all types and a combination of the signs are conserved (an interaction described as positive in a pathway and negative in another is considered as unclear). We selected some groups of GO terms representing relevant processes and functions in this dataset, in particular cell adhesion and motility (GO:0048870, GO:0007155, GO:0034330, GO:0022610, GO:0060352, GO:0030030), cell growth and death (GO:0008283, GO:0007049, GO:0008219, GO:0019835, GO:0000920, GO:0007569, GO:0051301, GO:0060242), immunity and inflammation (GO:0002376, GO:0001906), and cell differentiation (GO:0030154, GO:0036166). We also annotated as phospho-tyrosine modifiers the components of the network with tyrosine kinases (GO:0004713) and tyrosine phosphatases (GO:0004725) GO terms and manually verified this list (S7 Table).
Many nodes in KEGG pathways represent groups of proteins, where the same protein can be part of several groups (with variable overlap). These groups are conserved in the merged network by introducing “group nodes” with bidirectional links to their members. Proteins can thus have interactions associated directly to them or through one or several groups. Such groups are “exploded” before the path search step described below.
Cytoscape 3.4 (http://www.cytoscape.org/) was used to generate figures and Cytoscape web was used to provide interactive access to the Syk network.
We searched for “near-shortest paths” between Syk and a list of targets, using an approach similar to the classical Dijkstra algorithm for shortest paths. We started by identifying the length of the shortest path for every node as in Dijkstra’s method (starting from the source node, we iteratively picked the closest new neighbor of all reachable nodes: at each step we obtained the best result for a new node, ending with the node with the longest of the shortest paths). In the Dijkstra algorithm, the shortest paths were then obtained by starting from the target nodes and going backward to the source by selecting the incoming edge(s) which can satisfy this best distance: i.e. the best distance of the current node is equal to the sum of that of the source and the distance of the edge. Here we define an acceptable extra distance to include additional nodes and edges during this backtracking step. Note that the resulting sub-network can contain paths that are longer than acceptable, but all selected nodes and edges are involved in at least one acceptable path. For example if (A,B,C) is the shortest path from A to C, and (A,I,B,C) and (A,B,J,C) are also acceptable, then the path (A,I,B,J,C) exists in the resulting sub-network despite being too long. In the resulting sub-network, the selected nodes and edges are annotated with the “overflow” needed to include them: i.e. the extra distance of the best path using them compared to the actual shortest path. Members of the shortest paths have no overflow.
To improve the identified paths, we selected edge weights based on the available annotations: “normal” edges have a distance of 5 (d = 5), we promoted edges coming out of identified proteins (d = 3), edges reaching an identified protein while coming out of a tyrosine kinase or phosphatase (d = 2) or combining these two conditions (d = 1). On the other end, we demoted edges reaching a target identified as differentially phosphorylated, but which did not come from a tyrosine kinase or phosphatase (d = 8), even if they come out of another identified protein (d = 6). Finally, edge distances are refined to integrate the results of the random walk estimation: they are multiplied by the inverse of the normalized score of their source node.
We adapted the netwalk implementation in R from the GUILD software [66]: http://sbi.imim.es/web/index.php/research/software/guildsoftware
Edge weights are based on the distances originally defined for the shortest paths search (reversed as a higher distance corresponds to a lower weight).
More precisely, let dij be the distance from a node i to its direct target j, previously defined for the shortest path search. Given a node i the set of all its direct targets (successors in the directed network) is denoted Succ(i). The random walk is defined by transition probabilities pij defined as:
pij=(1−p0)dij−1∑j∈Succ(i)dij−1,
where p0 is the return probability to the origin node Syk (chosen the same for all nodes).
The probabilities pij together with p0 added as last entry of each row are the entries of the stochastic matrix P (each row of this matrix sums to one). The equilibrium or limiting distribution of the random walk is a normalized row vector π satisfying the equation:
πP=π.
This distribution can be estimated by starting the random walk from any node and running it a sufficiently long time for equilibration. A finite, connected network with possibility of return to the Syk node from terminal nodes is ergodic guaranteeing the existence and uniqueness of the equilibrium distribution.
Nodes are scored by the values of the equilibrium probabilities πi. In order to eliminate biases created by topology a second simulation is performed where all edges have the same weight. The resulting scores in this second simulation are the equilibrium probabilities πi0. The two scores are then used to refine the distances as follows
d˜ij=dijπi0πi.
MCF7, MDA-MB-231 and COS7 cell lines were obtained from the ATCC and maintained in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) supplemented with 10% fetal calf serum (FCS, Eurobio). All cell cultures were carried out at 37°C using a 5% CO2 atmosphere. For cell stimulation studies, cell lines were stimulated with Sodium pervanadate (PV, premix of 1 mM H2O2 and 1 mM Na3VO4) and incubated for 15 min at 37°C. For the evaluation of the effect of the kinase inhibitors (all from Selleckchem), cells were incubated in medium for 2 hours with 2.5 mM R406, 5 mM PRT062607, 1 mM PP2, 0.5 mM AZD0530. Stock solutions for all those kinase inhibitors were prepared in dimethyl sulfoxide (Sigma, Hybri-Max grade), which is used as vehicle negative control
Cells were washed with ice-cold phosphate buffered saline solution and scrapped in 10 mM Tris-HCl (pH 7.4), 150 mM NaCl, 0.5 mM EDTA (Sigma), 1% Nonidet-P40 (Sigma), 0.5% sodium deoxycholate (Sigma), 1 mM Na3VO4 (Sigma), 50 mM NaF (Sigma) and a protease inhibitor cocktail (Sigma) at 4°C. After transfer to an Eppendorf tube and extensive vortexing, lysates were cleared by centrifugation at 10,000 rpm for 10 min at 4°C and supernatants diluted in 4x SDS-PAGE sample buffer. Immunoprecipitations were performed as described previously [44]. Protein samples were diluted in 4x SDS-PAGE Laemmli sample buffer, denatured by boiling for 5 min at 95°C in SDS-PAGE sample buffer, separated electrophoretically and transferred to polyvinylidene difluoride membranes. Membranes were blocked using 5% BSA in tris-buffered saline solution with Tween-20 detergent (TBS-T; 25 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.1% Tween-20) for 1 h and then incubated at 4°C with the appropriate primary antibodies diluted in blocking buffer. Those included a mix of two monoclonal antibodies to phospho-tyrosine (1:1 mix vol/vol mix of the 4G10 and PY20 hybridoma supernatants), monoclonal antibodies to the FLAG epitope (clone M2, Sigma), cortactin (clone 4F11; Millipore), Syk (clone 4D10, Santa Cruz) and alpha-tubulin (clone DM1A, Sigma), rabbit polyclonal antibodies to pTyr418 Src (Invitrogen), GFP (Chemokine), a home-made rabbit polyclonal antibody to the COOH-terminal domain of human ezrin [67], and a rabbit monoclonal antibody to pTyr525/526 Syk (Cell Signaling Technology). After three washes with TBS-T, the membrane was incubated with horseradish peroxidase-conjugated appropriate secondary antibody (1:5000, Jackson ImmunoResearch) for 1 h at room temperature. Immunoblots were revealed using a standard chemoluminescent method (ECL, Ozyme) and a Multi-application gel imaging system (PXi, Syngene). Membranes were optionally stripped with the Restore PLUS Western blot stripping buffer (Thermo Scientific) before a second immunoblotting. Immunoblot-derived signals were quantified using the ImageJ software (NIH) with three independent biological and technical replicates for each quantification. The signals were normalized on the lane corresponding to the total protein quantity loaded (immunoglobulin heavy chain in case of immunoprecipitation; Tubulin-α in case of whole cell lysate) and the unstimulated condition was arbitrarily set at 1.
Assays with proteins extracted from cell lysates, immunoprecipitations and Western blot analyses were performed as described previously [43]. Otherwise, recombinant GST-Syk (BPS Bioscience, San Diego, CA) and GST-ezrin (previously described in [68]) were used. In vitro kinase assays were performed as described [43]. For the two-dimensional electrophoresis analysis, proteins were precipitated for 2 h in two volumes of acetone at −20°C, and resuspended in 8 M urea, CHAPS 4%, and thiourea 2 M. We used 18 cm IPG-strips (Amersham Biosciences) with linear pH range of 3–10 for the first dimension. Proteins were loaded on the IPG-strips and run in a Multiphor II apparatus (Amersham Biosciences). After focusing, a second migration was performed in 10% SDS-PAGE gel and proteins were stained with silver nitrate (Amersham Biosciences).
MDA-MB-231 cells were transiently transfected with pDsRed-Syk [43] using Fugene 6 (Roche Applied Bioscience). Immunostaining procedures have been described before [43]. The following primary antibodies were used: monoclonal antibody 4G10 hybridoma supernatant (diluted 1:50 in TBS) and a home-made rabbit polyclonal antibody to the C-terminal domain of human ezrin [67]. The secondary antibodies used were goat-anti-mouse-Cy5 and donkey-anti-rabbit-FITC (Jackson ImmunoResearch Laboratories). Confocal images of immunostained cells were obtained as described [69]. For quantitative analysis of colocalization, we used the ImageJ software plug-ins (https://imagej.nih.gov/ij/) with the “Colocalization Finder” module (https://imagej.nih.gov/ij/plugins/colocalization-finder.html) to generate the merged picture of Syk and ezrin channels in which colocalized pixels are displayed in white, and with the “Coloc 2” module (http://imagej.net/Coloc_2) to generate the scatter plot of pixel intensities in Syk and ezrin channels and to compute the Pearson correlation of pixel intensities over space.
Statistical analyses were performed using the two-tailed Student’s t test for paired and unpaired data versus control values. Experimental values in this work are all given as mean and standard error of the mean (SEM). Results with a P value ≤ 0.05 were considered as statistically significant.
|
10.1371/journal.pntd.0004207 | Inflammation Caused by Praziquantel Treatment Depends on the Location of the Taenia solium Cysticercus in Porcine Neurocysticercosis | Neurocysticercosis (NCC), infection of the central nervous system by Taenia solium cysticerci, is a pleomorphic disease. Inflammation around cysticerci is the major cause of disease but is variably present. One factor modulating the inflammatory responses may be the location and characteristics of the brain tissue adjacent to cysticerci. We analyzed and compared the inflammatory responses to cysticerci located in the parenchyma to those in the meninges or cysticerci partially in contact with both the parenchyma and the meninges (corticomeningeal).
Histological specimens of brain cysticerci (n = 196) from 11 pigs naturally infected with Taenia solium cysticerci were used. Four pigs were sacrificed after 2 days and four after 5 days of a single dose of praziquantel; 3 pigs did not receive treatment. All pigs were intravenously injected with Evans Blue to assess disruption of the blood-brain barrier. The degree of inflammation was estimated by use of a histological score (ISC) based on the extent of the inflammation in the pericystic areas as assessed in an image composed of several photomicrographs taken at 40X amplification. Parenchymal cysticerci provoked a significantly greater level of pericystic inflammation (higher ISC) after antiparasitic treatment compared to meningeal and corticomeningeal cysticerci. ISC of meningeal cysticerci was not significantly affected by treatment. In corticomeningeal cysticerci, the increase in ISC score was correlated to the extent of the cysticercus adjacent to the brain parenchyma. Disruption of the blood-brain barrier was associated with treatment only in parenchymal tissue.
Inflammatory response to cysticerci located in the meninges was significantly decreased compared to parenchymal cysticerci. The suboptimal inflammatory response to cysticidal drugs may be the reason subarachnoid NCC is generally refractory to treatment compared to parenchymal NCC.
| The cystic larvae of the pork tapeworm Taenia solium may affect the human brain causing neurocysticercosis (NCC), a very frequent cause of neurological symptoms in developing countries. The clinical expression and response to treatment of human NCC are related to the location of cysticerci inside (intraparenchymal) or outside the brain parenchyma (extraparenchymal NCC). We used a naturally infected pig model to assess the characteristics of inflammation around brain cysticerci of parenchymal, meningeal and mixed locations. There were no major differences in inflammation without treatment. After antiparasitic treatment with praziquantel, inflammation around parenchymal brain cysticerci increased in comparison to meningeal located cysticerci. Cysticerci partially surrounded by both brain parenchyma and meninges showed increased inflammation in relation to the extent of the cysticercus in the brain parenchyma. The location of cysticerci within the brain is a factor that determines the extent and degree of the immune response following anticysticidal treatment. Similar changes may occur in treated human infections. Our work could contribute to explain the differences in response to antiparasitic treatment in different forms of human neurocysticercosis.
| Neurocysticercosis (NCC), infection of the central nervous system (CNS) by the larval stage (cysticercus) of the parasitic helminth Taenia solium, is a common disease in regions where pigs are raised and allowed to roam freely [1,2]. NCC is a major cause of epileptic seizures in developing countries and therefore a serious public health problem [2]. Seizures and other symptoms of NCC depend on the number, location and distribution of cysticerci, as well as the degree of brain inflammation and developmental stage of the parasite, giving rise to a wide variety of manifestations [3,4].
The adult tapeworm, which resides in the small intestine of a human carrier, produces embryonated eggs containing embryos called oncospheres. After ingestion of contaminated feces by pigs or the accidental ingestion by humans, the oncospheres hatch, make their way to the bloodstream and mostly develop into cysticerci in the muscle, brain and subcutaneous tissues [2,5]. In the brain, the distribution of cysticerci generally follows the distribution of blood [3]. Cysticerci can be lodged in the parenchyma of the brain, cerebellum or brainstem, ventricles, subarachnoid space and spine [2,3]. According to cysticercal location there are two main forms of NCC: parenchymal and extraparenchymal disease (in which cysticerci have ventricular and/or subarachnoid locations) [2,3].
The clinical presentation, pathophysiology and treatment differ depending on the location and stage of cysticerci, degree of inflammation and other variables. Parenchymal NCC is associated with seizures or focal neurological signs. This form of NCC is relatively easy to treat, and has a fairly good prognosis compared to extraparenchymal NCC [6,7], which commonly causes hydrocephalus and intracranial pressure, and is associated with a poor prognosis [8,9].
Histology and medical imaging (computed tomography and magnetic resonance) have been useful to study and compare inflammation in human [10,11] and porcine NCC [12,13,14], including the effects of anthelmintic treatment [15] or cysticerci distribution [16,17]. However information of the association between anthelmintic treatment, inflammation and cysticercal location is scarce.
The vessels in parenchymal and meningeal tissue differ in the constitutive proteins of the junctional complex, susceptibility to changes in permeability of the blood-brain barrier (BBB), expression of cytokines that regulate BBB function and the type and number of associated astrocytes [18,19,20]. For the three highly vascularized tissue membranes of the meninges, dura mater, arachnoid and pia mater [21], the permeability of vessels in the pia mater is the most susceptible to perturbation and most critical in relation to the integrity of the BBB [22]. These tissue-specific characteristics could explain the differences on inflammation around the cysticercus in experimental murine models [19,20] and thus be involved in the severity of the inflammatory reaction elicited when the parasite degenerates–naturally or after anthelmintic therapy [2]. Cysticerci can be totally embedded within parenchymal or meningeal tissue or partially adjacent to both the brain parenchyma and meninges (corticomeningeal).
This study used a naturally infected pig as disease-based model to quantify and compare the inflammatory response following cysticidal treatment to cysticerci located in the brain parenchyma, meninges or both tissues.
This was a cross-sectional study. The methodology including pig husbandry, injection of Evans blue, use and dosing of pigs with praziquantel, euthanasia, as well as extraction, fixation and histopathological analysis of brain cysticerci is described in previously published reports employing the T. solium naturally infected pig model [23,27]. Brains from 11 pigs from Huancayo (Andean region endemic for NCC), naturally infected with Taenia solium cysticerci were used. The age of the pigs was between 2 and 4 years and the weight range was 42 kg to 107 kg. The animals had been randomly allocated to different groups: 3 pigs did not receive any drug (untreated, D0) and 8 received a single oral dose of 100 mg/kg praziquantel (Saniquantel 10%, Montana SA, Peru) which results in consistent cysticercal damage and subsequent enhanced host inflammatory responses [15,23,27]. Of these, 4 were sacrificed after two days (D2) and 4 after five days of the anthelmintic treatment (D5). All pigs were injected with Evans blue (EB, 80 mg/Kg; Sigma-Aldrich, St. Louis, MO) two hours before euthanasia [23].
The primary study was performed at the Universidad Nacional Mayor de San Marcos, Lima, Peru, under the protocol “Evaluación de la permeabilidad vascular en cerebro y músculo de cerdos naturalmente infectados con Taenia solium”, with Dr. Armando Gonzalez as principal investigator. The study protocol was approved by the Animal Ethics and Wellbeing Committee of the University -CEBA (Constancia de autorización ética No. 006, November 2010) and comply with the National Institutes of Health/AALC guidelines [23].
Brains previously fixed in 10% neutral buffered formalin for 24 hours were cut in coronal 10–12 mm slices before collecting samples (Fig 1A). Biopsies with the parasite and adjacent brain tissue were taken, fixed in buffered formalin and embedded in paraffin. Serial 4-μm thick sections, cut following the same coronal orientation of the brain slices, were mounted on slides for histological analysis by hematoxylin–eosin (HE) and Masson’s Trichrome stains. Only complete cysticercus sections (including cysticercus wall and scolex) and surrounding inflammatory reaction were considered. Cysticerci that were no longer structurally recognizable, as well as degenerated cysticerci and cicatricial lesions were excluded [23]. Each cysticercus was represented by one section as explained below.
Microphotographs were taken with 40X magnification with a Carl Zeiss microscope with AxioVision software and each image was saved in the Joint Photographic Experts Group (.jpg) format. Individual 40X images of the studied slides were obtained by capturing images with a 10 to 20% overlap between contiguous fields; the software stitched the individual.jpg images together to form a single large image (“cysticercus map”).
Each cysticercus was classified into one of three location categories: parenchymal, meningeal, or corticomeningeal. Parenchymal cysticerci were totally surrounded by parenchymal tissue and meningeal cysticerci by meninges (Fig 1B). Corticomeningeal cysticerci had a parenchymal and a meningeal region (Fig 1B). The coronal orientation of the sections (Fig 1A) made any involvement with parenchymal tissue evident and thus the location of the cysticercus could not be mistaken; this allowed evaluation of only one section per cysticercus, mainly around its center. The variable CP (“contact with parenchyma”) refers to the proportion of the cysticercus in contact with parenchymal tissue relative to the total perimeter.
The immune infiltrate in each cysticercus was classified into four separate inflammatory stages (IS), IS1 to IS4 as previously described [27]. In IS1 (low inflammation) the parasites were surrounded only by a thin layer of collagen (fibrosis), in IS2 (moderate inflammation) there was a sparse inflammatory infiltrate in between the collagen fibers; in IS3 (strong inflammation), granuloma formation was evident (cellular infiltrate in layers, together with the fibrosis) and in IS4 (severe inflammation) the parasite was surrounded by an eosinophil-rich exuberant infiltrate with fibrosis and obvious degeneration of the vesicular membrane [24,25,26]. The inflammatory score composite (ISC) was calculated by determining the percentage of the perimeter of the cysticercus for each IS, multiplied by the numerical value of the IS (1 to 4). The total ISC for one cysticercus was the sum of all IS components, as per the formula [27]:
ISC=(IS1%x1)+(IS2%x2)+(IS3%x3)+(IS4%x4);maximum=400.
(1)
The methods detailing analysis of cysticercal damage were described previously [27]. The loss of identifiable structures at the cysticercus wall surface, tegument or subtegument layers was considered damage. The length in mm of the cysticercus wall that showed any degree of damage was measured using the AxioVision software and the percentage of this damage relative to the total perimeter of the cysticercus was expressed without units. [27,28]
Pigs were injected with the intravital stain Evans blue two hours before being euthanized as previously described [23]; extravasation of the dye into the host tissue around the cysticercus (capsule) is a measure of the presence and degree of disruption of the BBB [29,30]. Clear and blue staining of the inflammatory reaction around the cysticercus represented intact and disrupted BBB, respectively [23].
Inflammatory score composite (ISC) and cysticercal damage were continuous parameters; BBB disruption was treated as a dichotomic variable, and treatment groups were also categories. Associations according to cysticercal location were evaluated using the Mann-Whitney or Kruskal-Wallis tests (inflammation or cysticercal damage and treatment) and the Fisher’s exact test (treatment group and BBB disruption). A Gaussian family generalized linear model (GLM) with an identity link was used for multivariate analysis of corticomeningeal cysticerci to model the association between ISC and contact with the parenchyma (CP), adjusted by treatment group and the interaction between CP and treatment, considering each animal as a cluster. This final model was defined after testing each covariate (BBB disruption, cysticercal damage, treatment and CP) in a sequence of nested models with likelihood test. All statistical analyses were performed with a 95% significance level using the statistical software STATA (STATA Corp LP, College Station, TX) 11.0.
The number of cysticerci in each pig brain was variable. The range was 3–95 considering all cysticerci (371 before selection of samples) in all groups; total parasites per pig in each group were as follows: D0 (8, 14 and 88), D2 (11, 27, 51 and 95) and D5 (4, 6, 10 and 57). Pigs in the D2 group had greater number of cysticerci than untreated pigs (D0) and the D5 group (total of 102 versus 56 and 38, respectively). Parenchymal cysticerci were more frequent: 105 out of 196 total cysticerci (53.6%) compared to 38 (19.4%) meningeal cysticerci and 53 (27.0%) corticomeningeal cysticerci (Table 1).
Inflammatory stages (IS1-IS4) and morphology varied according to the cysticercus location (total or partial parenchyma or meninges). Parenchymal IS1 had a collagen layer of variable thickness, whereas in meninges this layer was typically very thin; inflammatory cells were absent in both kinds of tissue. IS2 represented an increase in the collagen layer and the appearance of scattered cells both for parenchyma and meninges. Areas of IS3 could be found in untreated pigs (D0), although these were focal and restricted to a very small fraction of the cysticercus perimeter (not enough to increase the overall ISC value, thus not leading to statistical differences between locations in this group (Table 2).
Some differences between parenchymal and meningeal reactions could be seen on IS3; fibrosis, multinucleated cells and inflamed vessels (i.e., surrounded by inflamed cells, mostly eosinophils and lymphocytes) were more abundant in parenchyma than in meninges, and astrogliosis was, by nature of their location, exclusive to parenchyma (Fig 2). In corticomeningeal cysticerci, each distinct region (parenchymal and meningeal) behaved as either parenchymal or meningeal tissue. Inflammatory stages were distributed similarly between both regions.
ISC in D0, D2 and D5 was compared according to cysticerci location using the Mann-Whitney test. In untreated brains (D0), median values of ISC were not statistically different between parenchymal, meningeal or corticomeningeal cysticerci, indicating that inflammatory stages (IS1 to IS4) were distributed similarly among the three types of cysticerci. On the contrary, in D2 and D5 parenchymal cysticerci were statistically different from corticomeningeal (p = 0.03 and 0.01, respectively) and meningeal cysticerci (p = 0.002 and 0.004, respectively; Table 2). Treatment also resulted in higher ISC values after 2 and 5 days in parenchymal cysticerci than in corticomeningeal and meningeal cysticerci. (ISC up to 400, which corresponded to larger extensions of the higher stages of inflammation). These changes were statistically significant for parenchymal cysticerci (p<0.001, Kruskall-Wallis test) but not for corticomeningeal (p = 0.206) nor meningeal cysticerci (p = 0.898; Table 2).
There was a trend towards more cysticercal damage 5 days after treatment in the three locations (Table 2). Comparing cysticerci locations, only at D2 parenchymal cysticerci showed significantly more damage than corticomeningeal cysticerci (p<0.001).
After treatment, the main histological changes were seen in parenchymal tissue: IS3 was clearly more frequent and greater in D2 and D5 cysticerci than in D0. Also, astrocytic gliosis associated to IS3 and IS4 was more intense in the treated groups than in the untreated but only in the parenchyma. While IS3 could be seen in meningeal tissue, it was considerably less frequent and not as widespread as in parenchyma, and only a few of the treated cysticerci showed an increase (Fig 2). In corticomeningeal cysticerci, inflammatory stages were distributed similarly between parenchymal and meningeal regions. Following treatment, the extent of higher grade IS3 inflammation around the cysticercus increased in parenchymal cysticerci but was only rarely seen in meningeal cysticerci, resulting in higher ISC values for parenchymal cysticerci (Fig 3A and 3B).
Anthelmintic treatment has been shown to increase the permeability of the BBB in pericystic tissue and this is related to inflammation [23]. This effect was observed in parenchymal cysticerci; the proportion of cysticerci showing disrupted BBB (blue staining) increased from 67% in the untreated group to 85% in D2 and 96% in D5.
To evaluate the relationship between treatment and the increase of the disruption of the BBB in meningeal, parenchymal and corticomeningeal cysticerci separately, we used the Fisher’s exact test. Only parenchymal cysticerci showed a statistically significant association between treatment and BBB disruption (p = 0.013; Fisher’s exact test). In corticomeningeal and meningeal cysticerci there were no apparent associations between treatment (D2 and D5) and an increased BBB disruption (Table 3).
Corticomeningeal cysticerci were used to confirm that the inflammatory effect of treatment occurred mainly or only on parenchymal tissue. After determining that these cysticerci had comparable values of CP at D0, D2 and D5, ISC was calculated separately for the parenchymal and meningeal regions (Table 4).
A bivariate analysis comparing ISC of these cysticerci according to treatment, measured separately for parenchyma and meninges, showed that inflammation in the meningeal region did not follow any discernible pattern of change after treatment, while inflammation in the parenchymal region increased and correlated to the proportion of the cysticercus in contact with the brain parenchyma. In a stratified analysis of ISC, using three separate GLM, D5 had the highest regression coefficient for CP, and with a significant p = 0.049, which indicates that inflammation at this time point of the treatment was significantly higher in cysticerci with more contact with parenchyma (Table 5).
The clinical expression and response to treatment in human NCC vary greatly according to whether the parasitic larvae are located in the brain parenchyma or subarachnoid space [3]. In human NCC parenchymal cysticerci are associated with seizures and respond better to antiparasitic treatment, whereas subarachnoid cysticerci commonly cause hydrocephalus and intracranial hypertension and their response to antiparasitic therapy is poor [5]. In order to study the effect of the location of cysticerci in the inflammation after treatment in pigs, we used histological specimens from naturally infected pigs to investigate the characteristics of inflammatory reaction around T. solium cysticerci according to their location in the brain parenchyma, meninges, or in both locations, and the effect of anthelmintic treatment on the above variables. The main finding is that exacerbation of pericystic inflammation on day 5 after praziquantel treatment is location-dependent: Inflammation was proportional to the extent of contact of the cysticercus with the brain parenchyma (Fig 1C). The relative lack of host response in the subarachnoid tissue following treatment could help explain the clinical observation that extraparenchymal cysticerci are commonly refractory to treatment.
Other studies in humans and pigs with NCC previously demonstrated that treatment with anthelmintic drugs increases inflammation in the brain parenchyma and cysticercal damage [11,15]. Our data demonstrate that without treatment this inflammatory response is similar between parenchymal, meningeal and corticomeningeal cysticerci, but significantly increases following treatment (day 2 and day 5) in parenchymal cysticerci compared with those in meningeal and corticomeningeal locations (Table 2).
Enhanced inflammatory responses in the parenchyma compared to the meninges may be due to differences in the vascular network within the tissues [18,21]. Parenchymal tissue has denser vascularization and more microvascularization, which would be expected to result in an increase in the number of inflamed vessels as well as a greater number of infiltrating cells in the parenchymal side of corticomenigeal cysticerci. For the same reason, fewer drugs, and in lesser amounts, may reach the meninges compared to the parenchyma, resulting in the observed unresponsiveness of extraparenchymal NCC.
Treatment was associated with BBB disruption only in parenchymal cysticerci. The proportion of Evans blue staining increased from 67% on D0 to 85% and 96% in D2 and D5, respectively, while there was no similar increase in the other locations (Table 3). Differences in BBB damage may also be explained by differences in the vasculature network as mentioned above, as well as differences in the junctional structure within vessels.
Antiparasitic treatment of human parenchymal NCC is frequently associated with an exacerbation of neurological symptoms, particularly seizures, which peak between the second and the fifth day after treatment onset [31]. Also, the increase in treatment related symptoms is frequently accompanied with the appearance of or increase in enhancement as seen using MRI imaging. These early side effects are very likely due to the induction of pericystic inflammation in parenchymal cysticerci. Likewise, in this report as well as in earlier studies we saw an exacerbation of inflammation and disruption of the BBB of parenchymal cysticerci during the same time frame as seen in humans. As for clinical manifestations in the animals during the five days of treatment, convulsions have so far not been reported nor were observed this time; evident signs of discomfort were also absent and it is not known if they would have been seen under constant observation.
The present study has certain limitations. Inflammation, location and damage to the parasite were evaluated using newly constructed score variables. These included the inflammatory score composite (ISC), based upon descriptions by Alvarez [24]; cysticercal damage, used previously by our group [27, 28] and degree of contact with brain parenchyma (CP), described in the present work. While we worked with small numbers of animals and could not properly account for intra-group correlation (cysticerci collected from a same pig can have similar features, which can create an intra-group correlation), we used robust standard errors and analyzed each pig as a cluster. Despite these limitations, our findings suggest a significant association between the degree of contact with the parenchyma and treatment induced inflammation. The enhanced inflammation in the parenchyma may in part also be due to decreased concentration of active drug in the meninges compared to the parenchyma. A major difference between the two locations is the quantity and unique character of the blood vessels. Because of the increased quantity of microvessels in the parenchyma, it may be that the concentration of cysticidal drug may be increased around parenchymally located cysticerci compared to those in the meninges. The consequence would be an increase in cysticercal damage and antigen release [32] inducing a greater degree of inflammation.
In summary, there are clear differences between the immune response profiles according to the location of each cysticercus in treated and untreated pigs. These differences likely contribute to known differences in treatment efficacy between parenchymal cysticerci and subarachnoid NCC. More thorough knowledge of the factors causing this differential response, including whether PZQ pharmacodynamics act differentially in pigs and in humans, should contribute to the understanding of NCC pathogenesis and its better management.
|
10.1371/journal.pgen.1006909 | Termination factor Rho: From the control of pervasive transcription to cell fate determination in Bacillus subtilis | In eukaryotes, RNA species originating from pervasive transcription are regulators of various cellular processes, from the expression of individual genes to the control of cellular development and oncogenesis. In prokaryotes, the function of pervasive transcription and its output on cell physiology is still unknown. Most bacteria possess termination factor Rho, which represses pervasive, mostly antisense, transcription. Here, we investigate the biological significance of Rho-controlled transcription in the Gram-positive model bacterium Bacillus subtilis. Rho inactivation strongly affected gene expression in B. subtilis, as assessed by transcriptome and proteome analysis of a rho–null mutant during exponential growth in rich medium. Subsequent physiological analyses demonstrated that a considerable part of Rho-controlled transcription is connected to balanced regulation of three mutually exclusive differentiation programs: cell motility, biofilm formation, and sporulation. In the absence of Rho, several up-regulated sense and antisense transcripts affect key structural and regulatory elements of these differentiation programs, thereby suppressing motility and biofilm formation and stimulating sporulation. We dissected how Rho is involved in the activity of the cell fate decision-making network, centered on the master regulator Spo0A. We also revealed a novel regulatory mechanism of Spo0A activation through Rho-dependent intragenic transcription termination of the protein kinase kinB gene. Altogether, our findings indicate that distinct Rho-controlled transcripts are functional and constitute a previously unknown built-in module for the control of cell differentiation in B. subtilis. In a broader context, our results highlight the recruitment of the termination factor Rho, for which the conserved biological role is probably to repress pervasive transcription, in highly integrated, bacterium-specific, regulatory networks.
| Bacillus subtilis is a widely used model to study cell differentiation in the bacterial world. This soil-dwelling bacterium can engage in several alternative developmental programs, which generate distinct cell types adapted to different lifestyles, to cope with its complex and changing natural environment. The underlying differentiation control mechanisms are interconnected and tightly regulated, because these physiological and morphological cellular states are mutually exclusive and a correct choice is crucial. Here, we describe a previously unrecognized mechanism that regulates cell fate decisions in B. subtilis. It is based on the elements of pervasive genome-wide transcription controlled by the termination factor Rho. Pervasive transcription originating from non-defined or cryptic signals is spread throughout bacterial transcriptomes, but its physiological role is not yet well understood. We show that the elements of the Rho-controlled transcriptome affect specific developmental programs in B. subtilis: cell motility, biofilm formation, and sporulation, by directly or indirectly targeting expression of the key factors of cellular differentiation. Our results illuminate how Rho plays a prominent role in complex and organism-specific regulatory networks by controlling pervasive transcription. These findings rank Rho among the global transcriptional regulators of B. subtilis and invite systematic exploration of its role in other microorganisms.
| Transcription provides the basis for cellular development and metabolism in all living organisms by allowing the expression of the information stored in the DNA sequence of the genes. A different type of transcription not associated with classical, clearly delineated, expression units was discovered nearly fifteen years ago [1]. The term “pervasive transcription” was coined for this non-canonical type of transcription, found in all kingdoms of life, because of its generally genome-wide distribution, initiation from unexpected, often non-defined, or cryptic signals, or its arising from transcriptional read-through at weak or factor-dependent terminators [2–6].
From its discovery, the phenomenon of pervasive transcription raised questions concerning the biological functions of the associated RNA species. Indeed, this potentially futile process could have deleterious effects on cell physiology by interfering with sense transcription or chromosome replication or by compromising genome stability or cellular energy status [5–7]. However, turning it completely off may be difficult and counterproductive from an evolutionary stand-point, since mutations can continuously create new transcription initiation sites or alter the termination of existing transcription units. Indeed, this process may provide raw material for the evolution of novel functional biomolecules [8]. Pervasive transcription may thus be the result of a tradeoff between evolutionary forces and the production of essentially nonfunctional transcripts that are neutral or even slightly deleterious in terms of organism fitness. However, extensive studies in eukaryotes have established pervasive transcription as a fundamental component of the regulatory circuits that notably increases the complexity of gene control [7, 9–11]. The produced non-coding RNAs (ncRNAs) are involved in a wide range of cellular processes, playing crucial roles in development, aging, disease, and the evolution of complex organisms [7, 12, 13].
Pervasive transcription has been found in various bacterial transcriptomes [14–21], but its physiological role is still unclear. At the same time, mechanisms preventing pervasive transcription in bacteria are well known [6]. The transcription termination factor Rho, an ATP-dependent RNA helicase-translocase responsible for the main factor-dependent termination pathway in bacteria, plays an important role in preventing pervasive transcription [22–27]. In contrast to intrinsic terminators, the sequence features required for the function of Rho are complex and poorly defined [23, 24]. Rho is nearly ubiquitous in bacterial genomes and the basic principles of Rho-dependent-termination are conserved across species, despite some structural differences between Rho proteins. Over the past decade, the importance of Rho in gene regulation and its conserved role in the enforcement of transcription-translation coupling, by interrupting transcription of untranslated mRNAs, has been substantiated by studies performed in several bacterial species [6, 25–27]. The major role of Rho in the suppression of pervasive, primarily antisense, transcription has been demonstrated for the Gram-negative and Gram-positive microorganisms Escherichia coli, Bacillus subtilis, Staphylococcus aureus, and Mycobacterium tuberculosis, under conditions of Rho depletion [17, 18, 20, 21]. Complete or even partial inactivation of Rho in these bacterial species causes widespread transcription originating from cryptic promoters and read-through of transcription terminators [17, 18, 20, 21]. The biological relevance of this Rho-controlled component of the bacterial transcriptome is poorly understood. In E. coli, Rho inactivation is lethal, and a single amino acid substitution can cause changes in sense transcript levels and altered cellular fitness in the presence of various nutrients [28]. However, the increase of antisense transcription, due to partial inhibition of Rho, was reported to not interfere with sense transcription or gene expression in E. coli [18]. Similarly, no correlation between the levels of sense and antisense transcripts has been detected in M. tuberculosis. However, Rho inactivation in this bacterium significantly affected gene expression and caused cell death in cultures and during infection [21]. In contrast, a negative relationship between sense and antisense transcripts was observed in an S. aureus rho mutant [20], suggesting that Rho-controlled asRNAs may influence gene expression. Nonetheless, the lack of Rho did not significantly modify the growth behavior of either B. subtilis or S. aureus cells under the growth conditions tested [17, 20].
An increasing number of reports support a role of individual ncRNAs and antisense RNAs (asRNAs) in the regulation of gene expression in bacteria [5, 16, 29–33], but it is still accepted that most pervasive transcription represents non-functional and relatively low-level transcriptional noise [18, 34]. However, noise, or random fluctuations in gene expression due to the stochastic character of cellular processes involving low copy number cellular components (e.g. transcription factors or mRNAs) [35, 36], is also an important component of the fundamental processes of development and cell fate decision making in living organisms, from bacteria to mammals, as well as viruses [35–37]. Nonetheless, the influence of pervasive transcription on the regulation of developmental programs has not been experimentally addressed in bacteria.
The Gram-positive, soil dwelling, bacterium B. subtilis is a model for studying phenotypic heterogeneity and transitional developmental programs in prokaryotes, as it can express distinct cell types associated with specific phenotypes: motility, production of lipopeptide surfactin, genetic competence, biofilm formation, protease production, and sporulation [37–40]. During exponential growth, one sub-population of B. subtilis cells can synthesize flagella and grow as individual motile cells. The motile state of B. subtilis populations is determined by the alternative sigma factor, SigD, which drives the expression of genes essential for the synthesis and regulation of the flagellar apparatus [40, 41]. The transition from motility to an alternative type of cellular growth within surface-associated communities, known as biofilms, involves the repression of flagellar genes and activation of genes essential for production of biofilm extracellular matrix composed of polysaccharides, protein fibers and nucleic acids [41–43]. Under conditions of limiting nutrients, a sub-population of B. subtilis cells can initiate a multistage differentiation program to form highly resistant endospores (spores) [44, 45]. The respective gene networks controlling these mutually exclusive developmental programs are interconnected and share common regulators and regulatory feedback loops, which prevent their simultaneous activation within a cell [41, 43, 44, 46, 47]. The key determinant in the regulation of biofilm formation and sporulation is the master regulator Spo0A, for which the activity depends on its gradually increasing phosphorylation state, determined by a multicomponent phosphorelay system [48–54]. When the concentration of phosphorylated Spo0A (Spo0A~P) is low, biofilm formation and sporulation are repressed by the transcriptional regulator of exponential growth, AbrB, and the biofilm-specific repressor, SinR [55, 56]. This negative control is removed at intermediate levels of Spo0A~P, which activates the biofilm formation program [47, 57–62]. Only cells expressing high levels of Spo0A~P can enter into sporulation [45, 58, 63]. At the same time, matrix production is blocked by high Spo0A~P [43, 56–58]. Thus, the level of Spo0A~P determines heterogeneity of the matrix and spore production in populations of B. subtilis.
Here, we investigated the impact of pervasive transcription on the physiology of B. subtilis cells taking advantage of the viability of B. subtilis rho-null mutant. Comparative transcriptome and proteome analyses of B. subtilis wild type (WT) and rho mutant (RM) strains revealed significant perturbation of the global gene expression landscape in the absence of Rho and highlighted potential alterations of the regulatory networks known to define cell fate in B. subtilis. Further functional studies demonstrated that at least three of the above-mentioned differentiation programs, motility, biofilm formation, and sporulation were altered in RM cells due to the loss of Rho-mediated control of pervasive transcription. We describe several mechanisms by which Rho directly or indirectly participates in the in fine regulation of cell fate decision-making. Rho-controlled transcription represents a new level of regulation of gene expression in the Gram positive bacterium B. subtilis and the termination factor Rho can be considered among the global transcriptional regulators.
We reanalyzed the dataset of genome-wide expression profiles previously established for B. subtilis 168 derivative strain 1012 and its isogenic rho mutant (RM) grown exponentially in rich medium as a starting point for dissecting the pathways by which the absence of Rho could affect B. subtilis physiology [17]. These tiling array data can be visualized on the B. subtilis expression data browser with expression profiles established for the BSB1 strain, a tryptophan-phototrophic (trp+) derivative of 168 (http://genome.jouy.inra.fr/cgi-bin/seb/index.py), [17].
The previous analysis of the RM strain was mainly focused on the detection of transcription outside of the transcribed regions (TRs) detected in the wild type strain (native TRs) [17]. The goal of the present reanalysis was to characterize global changes that affect functional regions (in particular mRNAs) and can lead to phenotypic variations. Therefore, we focused on the differential expression of sense and antisense expression levels aggregated according to a repertoire of 5,875 native TRs including mRNAs and ncRNAs identified in the BSB1 strain (WT) across 104 conditions. In this repertoire, 1,583 transcribed regions outside the Genbank-annotated genes were previously designated as S-segments and are numbered S1-S1583, according to their chromosomal position [17] (S1 Table).
Significant changes in expression (log2 RM/WT ≥ 1 or ≤ -1) consisted primarily in up-regulation of the antisense strand, in agreement with our initial observations [17]. The detected changes were decomposed into 456 up-regulations and 223 down-regulations on the sense strand (Fig 1A) and 1,446 up-regulations and 36 down-regulations on the antisense strand (Fig 1B). Many of these changes exceeded a four-fold threshold (log2 RM/WT ≥ 2 or ≤ -2): 162 up- and 38 down-regulations on the sense strand, and 613 up- and seven down-regulations on the antisense strand. S1 Table presents the detailed results of this re-analysis.
We used the non-domesticated NCIB 3610 strain for subsequent physiological analysis of rho mutants. This strain is a member of the 168-like group of strains originating from the Marburg ancestor and characterized by highly similar genome sequences [64]. Thus we also collected new transcriptome data to investigate the effect of rho deletion in the NCIB 3610 background. These new RNA-Seq-based data for the NCIB 3610 WT and NCIB 3610 RM strains (Materials and methods) are consistent with the previous data obtained by tiling array in the B. subtilis 1012 background. Differential expression analysis of the NCIB 3610 RM strain identified 1,029 up-regulations and 375 down-regulations of the sense strand, along with 2,115 up-regulations and 72 down-regulations of the anti-sense strand (S1 Table). Approximately 80% of the up-regulations identified in 1012 RM were also found in NCIB3610 RM; the correspondence between data sets was lower for the sense strand (≈35%), but still highly statistically significant (Fig 1A and 1B).
Expression changes on the sense strand can be divided into three main categories (Fig 2A): direct up-regulation downstream of Rho-dependent termination sites; indirect cis effects by which an up-regulated antisense transcription affects expression of the overlapping gene on the opposite strand; and indirect trans effects resulting from regulatory cascades initiated by direct effects. Disentangling these three types of effects can be difficult.
Based on B. subtilis 1012 tiling arrays, up-regulation was detected for the sense expression levels of 456 (7.8%) native TRs, including 181 S-segments and 275 protein coding genes. Most of these events were due to altered termination of transcription (direct cis effect) as previously reported [17] (Fig 2B and 2C). In total, 90 up-regulated S-segments and 197 up-regulated protein coding genes clearly result from imperfect termination (termination read-through and/or lack of termination; Fig 2D and 2E). However, 26.3% (120/456) of the detected up-regulated native TRs were associated with at least two-fold increased expression levels immediately after the corresponding promoters (Fig 2F and 2G) and thus likely result from indirect trans effects. Similarly, down-regulation of 163 of 223 (73.1%) native TRs was associated with significantly decreased expression levels (RM/WT log2 ≤ -1) immediately after promoters (Fig 2H), which could reflect either decreased RNA synthesis, due to indirect trans effects, or increased RNA degradation.
Although the tiling array analysis indicated that genes targeted in RM cells by anti-sense transcripts were down-regulated more often than would be expected at random, this result could not be entirely confirmed with the RNA-Seq data. Thus, it is still unclear whether the up-regulated antisense strands in B. subtilis cells contribute globally to down-regulation of the sense strands, in addition to indirect trans effects.
We obtained further insights into the potential physiological consequences of the altered expression profiles in the RM strains by analyzing the distribution of up- and down-regulated native TRs in terms of regulons and functional categories (sigma factor regulons as established in [17], other regulons, and functional categories, as designated in the SubtiWiki database [65]). The complete list of statistically significant associations based on the analyses of B. subtilis 1012 tiling array and NCIB 3610 RNA-seq data (Fisher exact test p-value ≤ 1e-4) is presented in S2 Table.
Over-representation of the SigD regulon (p-value 1e-59) among the down-regulated native TRs accounted for the strongest associations with known functional categories and regulons in both 1012 RM and NCIB 3610 RM strains. Most of the genes from functional category “motility and chemotaxis” were down-regulated in both RM strains, consistent with known functions of SigD. All genes from the Spo0A and CodY regulons, which were over-represented (p-value 6.59e-15 and 1.57e-17, respectively) among the down-regulated TRs in both RM strains also belong to the SigD regulon. PBSX prophage genes and genes controlled by the specific sigma factor Xpf were down-regulated in the 1012 RM but not the NCIB 3610 RM strain.
The up-regulated protein-coding genes and S-segments exhibited weaker biases towards specific regulons and functional categories. Nonetheless, we observed over-representation of genes from the SigK and SigG regulons (p-value 4e-9 and 9.7e-5, respectively), active during sporulation, and genes from the close functional category of “sporulation proteins” (p-value 1.4e-7) [65] (S2 Table). These genes were not expressed in WT cells during exponential growth in rich LB medium. SigA genes were statistically under-represented (p-value 2e-9), but still accounted for a majority of this set of up-regulated genes. Over-representation of the SigB regulon (p-value 3.8e-7) was detected in the NCIB 3610 RM but not 1012 RM strain.
We performed a comparative proteome analysis of the BSB1 WT and RM strains (Materials and methods) to complement the transcriptome data. Membrane and cytosolic fractions were prepared separately to maximize the chances of protein identification. Protein Abundance Index (PAI) values, calculated from mass spectrometry data, were used to compare the two proteomes [66]. In total, 1,619 proteins were identified (see S1 Table for the PAI of identified proteins and S3 Table for raw proteome data), corresponding to 38% of the protein-coding genes. The log2 PAIs of detected proteins correlated with the abundance of cognate mRNAs both in WT and RM strains (Pearson correlation coefficients > 0.60). We evaluated the effect of Rho inactivation on the proteome using the same two-fold cut-off as for the transcriptome: abundance increased for 157 proteins and decreased for 101 proteins in RM, with 85 proteins detected only in RM and 38 detected only in WT cells (S1 and S3 Tables). The proteome analysis confirmed the strong down-regulation of the SigD regulon in the absence of Rho: 42 of 44 SigD-controlled proteins detected in the WT proteome were under-represented in the RM proteome (Table 1). SigD protein itself was not detected in any proteome. The observed decrease of protein abundance in RM cells was variable and reached 26-fold in the case of the HemAT protein.
In summary, transcriptome and proteome analyses documented notable alterations of the genome expression landscape in RM cells during exponential growth in rich medium. They highlight unscheduled expression of a number of sporulation genes and down-regulation of the SigD regulon, which could reflect important physiological changes. This prompted us to more thoroughly examine the impact of Rho inactivation on cell behavior during corresponding developmental programs.
Analysis of the transcriptome and proteome data showed that SigD-controlled genes were significantly down-regulated in RM derivatives of 1012 and NCIB 3610 strains (Table 1 and S1 Table). Further examination showed that genes belonging to the SigD regulon, but primarily expressed from SigA promoters [17, 65], displayed either a weak (for example, log2 RM/WT ≤ −1.0 for the yjcP-yjcQ operon) or intermediate (for example, log2 RM/WT ≤ −1.5 for the fla-che operon) decrease in expression. In contrast, genes exclusively controlled by SigD were strongly down-regulated (for example, log2 RM/WT = −2.8 in 1012; log2 RM/WT = −2.1 in NCIB 3610 for the motA gene). The expression level of sigD itself was lower in both RM strains (log2 RM/WT = −0.91 in 1012; log2 RM/WT = −1.69 in NCIB 3610) (S1 Table).
The down-regulation of 11 SigD-controlled genes was associated with the presence of asRNAs expressed above the cut-off level (S4 Table). Four of these asRNAs have been previously detected and annotated in BSB1 WT: S1367, S1403, S125, and S829 [17]. Expression of the non-annotated asRNAs facing the yvyC-fliD-fliS-fliT-yvzG operon and cheV and flhO genes was specific to RM (can be visualized on http://genome.jouy.inra.fr/cgi-bin/seb/index.py).
Several phenotypes are known to be associated with expression of the SigD regulon of B. subtilis cells, in particular, motility, the capacity to synthesize flagella and to swim in liquid, or to swarm over a solid surface [67]. SigD is a key determinant of the phenotypic switch between motile and sessile states within the exponentially growing cell population [40]. Cells with a high level of SigD (SigD-ON) are motile and those with a low level of SigD (SigD-OFF) are sessile. We compared the undomesticated strain NCIB 3610 and its isogenic rho mutant to investigate the impact of Rho inactivation on the SigD-controlled motility phenotype, as laboratory strains of B. subtilis do not exhibit swarming motility [68]. Swarming of the NCIB 3610 RM strain was significantly impeded (Fig 3). We restored the wild type rho allele at the chromosome of the RM strain to prove that the observed deficiency was directly linked to Rho inactivation. Conversion of the NCIB 3610 RM derivative back to wild type (NCIB 3610 rho wt*) restored motility, showing that the motility-deficient phenotype of the NCIB 3610 RM strain was due to the deletion of rho (Fig 3A and 3B).
Complex mechanisms, including several regulatory feedback loops, control expression of the genes from the SigD regulon and sigD gene itself, thus determining the SigD-ON or SigD-OFF state [40]. In particular, the anti-sigma factor FlgM antagonizes SigD activity by direct binding to SigD and inhibition of its interaction with RNA polymerase [69]. Additionally, genes from the SigD regulon are negatively controlled by the SinR, SlrR, and SlrA transcription factors, acting as SinR-SlrR and SlrR-SlrA heterodimers or a SlrA/SinR/SlrR functional complex [70, 71]; global transcription regulator CodY [72]; stringent response regulator RelA [73]; and adaptive response regulator YmdB [74, 75]. Global regulator DegU acts as a repressor or activator of the SigD regulon, depending on its phosphorylation state [76]. Finally, SwrA and SwrB proteins positively control SigD [68, 77].
Expression of most of these regulatory genes was not significantly different between RM and WT cells, with the exception of lower expression of swrB (log2 RM/WT = −0.88 in 1012; log2 RM/WT = −1.61 in NCIB 3610) and higher expression of slrR (log2 RM/WT = 1.44 in 1012; log2 RM/WT = 3.16 in NCIB 3610) and slrA (under the two-fold cut-off level, log2 RM/WT = 0.967 in 1012; log2 RM/WT = 0.60 in NCIB 3610) genes. Increased expression of both slr genes was due to 3’ extensions: of the asRNA targeting the epsA-O operon for the slrR gene (see next section); and of the S1475 segment for the slrA gene (S2 Fig). Extension of the S1475 segment protects the slrA mRNA from enzymatic degradation [78]. An increase in slrA copy number down-regulates the fla/che operon containing the sigD gene and, consequently, the entire SigD regulon; this inhibition depends on active SlrR and SinR proteins [71, 78]. Thus, we tested whether the non-motile phenotype of RM cells was due to increased expression of slrR and/or slrA. Inactivation of the slrR gene, which disables both regulators [71], partially restored the motility of NCIB 3610 RM cells (Fig 3).
The observed homogenous down-regulation of genes exclusively transcribed from SigD promoters suggests direct inhibition of SigD activity. This prompted us to consider the possible implication of FlgM in the observed phenotype of the RM strain. The anti-SigD activity of FlgM is dose-dependent and transcriptionally and post-translationally controlled [79–81]. Transcriptome analysis did not reveal any changes of flgM expression in RM strains (S1 Table; http://genome.jouy.inra.fr/cgi-bin/seb/index.py). Post-translational regulation of FlgM is exerted via FlgM secretion from the cytoplasm by the flagellar export apparatus after assembly of the intermediate hook-basal body of flagellum [81]. Thus, SigD activity tightly correlates with the efficiency of FlgM secretion, which in turn depends on completion of the flagellar hook [67, 81, 82]. It is thus remarkable that expression of the flhO-flhP genes encoding the components required for hook completion was significantly lower in the RM strains (for the flhO gene, log2 RM/WT = −2.06 in 1012; log2 RM/WT = –1.62 in NCIB 3610). The decrease of flhO-flhP transcription correlates with the ~ 860 nucleotides (nt) long 3’-extension of the annotated S1403 asRNA (log2 RM/WT = 2.77 for the flhO asRNA in 1012, log2 RM/WT = 3.92 in NCIB 3610), (Fig 3C and S1 Fig). In the absence of Rho, transcription of S1403 extended through the RNA hairpin structure (ΔG = −16.6) [83] within the flhP gene (Fig 3C and 3D). The flhP asRNA, with the 3’-end matching the position of this hairpin, was detected by genome-wide 3’ end-mapping in the Rho-proficient B. subtilis PLBS802 strain [84]. In RM cells, the extended S1403 asRNA spreaded over the whole flhO gene, overlaped with the flhO-flhP promoter, and may have down-regulated expression of the flhO-flhP operon. This could impede the synthesis of flagellar hook, leading to FlgM accumulation in the cytoplasm and consequently, reduced expression of SigD-dependent genes, as observed for B. subtilis flhO and flhP mutants [82].
First, we tested this possibility by examining the contribution of FlgM to the motility-defective phenotype of the RM strain by inactivating the flgM gene. Deletion of flgM improved the motility of NCIB 3610 RM cells (S3 Fig). We next compensated the down-regulation of the flhO-flhP operon in RM cells by inserting a copy of the flhO-flhP operon expressed from its own promoter, into the amyE chromosomal locus of NCIB 3610 RM (NCIB 3610 RM amyE::PflhO-flhO-flhP). The expression of the flhO-flhP genes from the ectopic position improved the swarming motility of NCIB 3610 RM (Fig 3A and 3B). Subsequent inactivation of the slrR gene (NCIB 3610 RM slrR, amyE::PflhO-flhO-flhP) had an additive effect, but yet did not restore cell motility to the WT level (Fig 3A and 3B). This pinpoints the existence of additional factors that inhibit motility in RM cells.
Rho-controlled sense transcripts associated with slrR and slrA and the antisense transcript associated with flhO-flhP genes represent newly identified components of the regulatory network that control cell motility. These findings provide additional evidence that read-through of Rho-dependent terminators affecting the expression of downstream genes can propagate into regulatory networks and cause phenotypic changes.
The switch from the motile to sessile state in growing B. subtilis populations is associated with activation of an alternative developmental program, known as biofilm formation. Biofilms are multicellular aggregates assembled within a self-produced extracellular matrix. The main components of the biofilm matrix, exopolysaccharides (EPS) and amyloid-like protein fibers, are encoded by the 15-gene-long epsA-O operon and the tapA-sipW-tasA operon, respectively [85–87]. The global transcription regulator, Spo0A, indirectly controls the expression of matrix operons through the AbrB and SinI/SinR pathways [56, 88–92]. Expression of both operons is activated when Spo0A~P is present at low and/or intermediate levels and suppressed by high levels of Spo0A~P [47, 56, 88].
Motility genes are involved in the initial stages of air-liquid interface biofilm (pellicle) formation, but not in the development of architecturally complex colonies (colony biofilm) on an agar surface [93, 94]. Non-motile cells can also proceed to biofilm formation directly [39, 41, 61, 88, 95]. Thus, inactivation of Rho may affect the program of biofilm development due to altered SigD activity. In addition, up-regulation of slrR and slrA genes (S1 Table) could contribute to de-repression of the matrix operons and favor biofilm development in RM cells.
We investigated the consequences of Rho inactivation on biofilm formation by comparing the dynamics of pellicle formation by the NCIB 3610 WT and RM strains in biofilm-promoting MSgg medium, as well as their capacity to develop architecturally complex colonies on an agar surface. The NCIB 3610 WT strain formed thick, robust pellicles and exhibited complex colony architecture as described in the literature [96] (Fig 4A). In contrast, the isogenic RM strain formed only thin pellicles and flat unstructured colonies, which did not attain a phenotype similar to that of the wild type biofilms (Fig 4A). There were no differences in the biofilm phenotypes between the NCIB 3610 WT and rho-restored NCIB 3610 rho wt* strains (Fig 4A). This shows that the biofilm-deficient phenotype of the NCIB 3610 RM strain is primarily due to the deletion of rho, similarly to the motility defect.
The defective architecture of colonies formed by the NCIB 3610 RM strain suggested that inefficient biofilm formation was not solely due to down-regulation of the SigD-regulon [43, 93]. We thus examined biofilm formation by the strain NCIB 3610 RM amyE::PflhO-flhO-flhP. Indeed, ectopic expression of the flhO-flhP operon, which improved motility of the NCIB 3610 RM strain, did not improve biofilm formation (S4 Fig).
We next compared expression of the matrix genes between the WT and RM strains, using transcriptional fusions of the epsA and tapA promoters with the firefly luciferase (luc) gene [97], to gain insight into the impaired capacity of RM cells to form biofilms. These fusions were introduced at the native eps or tapA chromosomal loci of BSB1 WT and RM strains. We monitored luciferase activity during growth in liquid MSgg medium with constant aeration ([88], Materials and methods). We observed maximal expression of the eps-luc and tapA-luc fusions in WT cultures at the end of the exponential growth phase, in accordance with previously published data [88]. At the same time, both the eps and tapA promoters were significantly less active in the RM strain (Fig 4B and 4C), indicating inefficient de-repression of the matrix operons negatively controlled by SinR [56, 91]. We therefore expected that inactivation of SinR would restore biofilm formation by the RM strain. We examined biofilms formed by the NCIB 3610 sinR mutant and observed the formation of vigorous pellicles, colonies with an elevated surface, and increased production of mucoid substances, as reported previously [51, 56, 91] (Fig 4A). In contrast, the NCIB 3610 RM sinR strain formed fragile and shattered pellicles, resembling those of the NCIB 3610 eps mutants [43, 86, 98], and colonies which were morphologically different from the Rho-proficient NCIB 3610 sinR mutant (Fig 4A). Inactivation of AbrB, the second repressor of matrix operons in B. subtilis, did not restore biofilm formation by the NCIB 3610 RM or NCIB 3610 RM sinR mutant strains (S5 Fig). Therefore, relieving matrix operons of SinR- and AbrB-mediated repression is not sufficient to restore normal biofilm formation by the RM strain.
The transcriptome analysis revealed an additional factor that could potentially interfere with biofilm formation by B. subtilis RM cells. It is represented by an asRNA of ~15,740 nt, which starts near the 3’-end of the epsO gene, probably due to read through at an intrinsic terminator of the oppositely oriented yvfG gene, and overlaps the entire epsA-O operon (for epsC asRNA, log2 RM/WT = 3.01 in 1012 and log2 RM/WT = 4.79 in NCIB 3610; Fig 4D, S1 Table).
We tested whether the activity of eps asRNA contributes to the impaired biofilm formation of RM cells by blocking its synthesis. This was achieved by insertion of three Rho-independent transcription terminators within the epsO gene, with the active orientation blocking the synthesis of eps asRNA (Materials and methods, Fig 4D and 4E). RT-PCR confirmed that the synthesis of eps asRNA was abolished in the NCIB 3610 RM epsO:Ter strain (Fig 4F). Previously, the epsO gene was shown to be dispensable for pellicle formation [99]. Indeed, the NCIB 3610 derivative carrying the epsO:Ter insertion did not display any defect in biofilm formation (Fig 4A). The NCIB 3610 RM epsO:Ter strain had somewhat stronger pellicles and more complex colony biofilms than the parental RM strain (Fig 4A). Simultaneous prevention of antisense transcription and de-repression of the matrix operons in the NCIB 3610 RM sinR, epsO:Ter strain greatly improved development of pellicles and colony biofilms, which were similar to those formed by the NCIB 3610 sinR mutant (Fig 4A). These results demonstrate that Rho-controlled eps asRNA negatively affects EPS production.
Altogether, our results show that inactivation of Rho results in impaired biofilm formation. This phenomenon is mainly due to inefficient de-repression of both matrix operons and the negative effect of the eps-specific asRNA on the expression of the eps genes.
Comparative transcriptome and proteome analysis of B. subtilis WT and RM cells revealed that Rho inactivation led to perturbations of the multi-component phosphorelay system responsible for Spo0A phosphorylation [48, 49], (Table 2; S6 Fig).
Among the genes coding for five sensor histidine protein kinases (KinA—KinE), which are at the basis of the Spo0A phosphorelay, the kinB gene was strongly upregulated in RM cells (log2 RM/WT = 2.02; Table 2). We also detected the KinB protein in the RM, but not WT proteome, consistent with the transcriptome data. Analysis of the transcription profiles of WT cells during exponential growth revealed that kinB mRNA level was not constant across the gene but showed a marked down-shift at approximately one-third part of the open reading frame. In contrast, we did not observe this down-shift in RM cells (http://genome.jouy.inra.fr/cgi-bin/seb/index.py). The level of gene expression is aggregated into a single value computed as the median for probes within the transcription region [17]. Thus, the presence of the down-shift strongly reduces the value of kinB mRNA expression in the WT relative to RM cells. The mechanism of an intragenic down-shift within kinB will be discussed later. RNA levels of the other kinase genes were not significantly affected in the RM strain (Table 2). However, the KinA and KinE proteins were detected in the RM but not the WT proteome, whereas the KinC and KinD kinases were detected at slightly decreased levels in the RM proteome.
The transfer of the phosphoryl group from sensor kinases to Spo0A is catalyzed by two phosphotransferases, Spo0F and Spo0B [48]. Both spo0F and spo0B transcripts were up-regulated in RM (log2 RM/WT = 0.895 and 1.185, respectively). The amounts of Spo0F and Spo0B proteins were also increased in the RM proteome (1.5- and 2.9-fold, respectively), consistent with the transcriptome data.
In addition to the main phosphorelay components, the expression of several genes encoding accessory proteins was modified in the absence of Rho. Transcript level of the kbaA gene, which encodes a positive effector of KinB [100], was higher in RM cells (log2 RM/WT = 2.04) due to the 3’ extension of the upstream salA mRNA. Transcript levels of the sivA and sivB genes, encoding factors that negatively control the level of Spo0A~P through inhibition of KinA autophosphorylation [101], were lower in RM cells (log2 RM/WT = −1.22 and −0.97, respectively). For both genes, this was apparently due to lower activity of the corresponding promoters (indirect trans effect).
The genes of the rapA-phrA operon, encoding RapA phosphatase, which specifically dephosphorylates Spo0F~P, and its inhibitor, the PhrA peptide [102], were equally up-regulated in the RM strain (log2 RM/WT = 2.15 and 2.52, respectively). We also detected higher levels of RapA in the RM proteome, consistent with the transcriptome data. The up-regulation of rapA-phrA mRNA levels in RM cells was associated with the disappearance of the down-shift within the rapA transcript observed in WT cells, similar to kinB transcription. In contrast, RapB, the second phosphatase active on Spo0F~P, was down-regulated in the RM strain, apparently due to lower activity of the rapB promoter. Transcript levels of the yisI gene, encoding a phosphatase specific for Spo0A~P [103], increased in the RM strain (log2 RM/WT = 2.50). The remaining components of the B. subtilis phosphorelay system were not significantly affected by the lack of Rho (S1 and S3 Tables) and are not reported in Table 2.
We translationally fused KinA and KinB proteins with the SPA peptide and compared the levels of SPA-tagged proteins in WT and RM cells to assess expression of these main sensor kinases at different growth stages. RM cells contained higher levels of KinA and KinB proteins than WT cells during the exponential and stationary phases of growth in LB (Fig 5). The effect was more prominent for KinB, as no or very little protein was detected in WT cells grown in LB. The propagation of cells in sporulation-inducing DS medium stimulated the synthesis of both kinases with a prevailing effect in RM cells, especially for KinB (Fig 5).
Taken together, these results highlight important changes in the expression of the multi-component phosphorelay system controlling the phosphorylation state of Spo0A in RM cells. The gradual activation of Spo0A~P by sequential phosphorylation might be shifted towards higher phosphorylation levels in the absence of Rho, given the known functions of the up- and down-regulated genes in this process.
We sought to experimentally establish whether the observed changes of phosphorelay in RM cells results in modification of the phosphorylation level of Spo0A~P as this could contribute to their defect in biofilm formation. Indeed, a high level of Spo0A~P induces suppression of the matrix operons [47, 56, 88]. During the transition to stationary phase, accumulating Spo0A~P increases spo0A gene expression via several positive feedback loops [104] and a transcription switch from the SigA-dependent vegetative promoter to the SigH-controlled sporulation-specific promoter [105]. We postulated that a high level of Spo0A~P in RM cells would lead to detectable changes of spo0A expression. The real-time kinetics of spo0A expression was previously analyzed at a population-wide level using the luc reporter gene fused, in-frame, to the spo0A start codon at its natural locus [97]. The construction monitors the activity of both spo0A promoters while maintaining an intact copy of the spo0A gene [97]. We used this spo0A-luc fusion to compare spo0A expression between the WT and RM strains.
Initially, we analyzed the cells growing in biofilm-promoting MSgg medium, in which Spo0A~P accumulates to an intermediate level, stimulating de-repression of the matrix operons [58, 88]. We followed this event by simultaneous analysis of tapA-luc expression. The expression of spo0A in WT cells gradually increased during exponential growth and then remained relatively constant, producing a few weakly oscillating peaks (Fig 6A). At the end of exponential growth, one of the spo0A expression peaks coincided with activation of the tapA-sipW-tasA operon, indicating that the cells accumulated an appropriate Spo0A~P level. In RM cells, spo0A expression was lower during exponential growth than in WT cells, but exhibited a spike at the end of exponential growth (Fig 6A). Such a burst of spo0A activity might reflect activation of the sporulation-specific, SigH-dependent spo0A promoter by a high level of Spo0A~P [105]; the transition phase-specific SigH factor has been shown to be active in MSgg medium on other promoters [89]. Indeed, the spike of spo0A activity in RM cells correlated with decreased tapA-sipW-tasA expression, known to be suppressed by a high level of Spo0A~P [87, 106].
Next, we assessed spo0A expression in WT and RM cells grown in sporulation-inducing DS medium. In WT cells, spo0A expression was characterized by several pulses during the exponential and stationary growth phases, closely resembling previously reported spo0A expression kinetics (Fig 6B), [97]. A double-headed peak of spo0A expression observed at the moment of growth arrest has been previously shown to mark entry of the cells into sporulation (T0), as it coincides with activation of the early sporulation genes (Fig 6B), [97]. This peak would reflect a sporulation-inducing high threshold level of Spo0A~P [107]. Inactivation of Rho had no significant effect on spo0A promoter activity during exponential growth of RM cells but modified it at T0, when spo0A expression peaked at a higher level than in WT cells. This spike in the activity of the spo0A promoter at T0 was highly reproducible in RM cells and most likely resulted in a higher Spo0A~P level than in WT cells.
We further established an increase in the level of Spo0A~P in sporulating RM cells by following luciferase expression from the SigH-dependent promoter of the spoIIAA-AB-sigF operon, which is activated by a high level of Spo0A~P [57, 108]. In both WT and RM strains, spoIIA-luc induction coincided with pulses of spo0A activity, attributable to high Spo0A~P levels (Fig 6C). However, spoIIA-luc expression in the RM culture initiated about one hour earlier and was notably more efficient than in WT cells. This indicates that a sub-population of cells, in which Spo0A~P reached the required threshold to activate early sporulation genes, was higher in the RM culture.
We investigated whether the changes of spo0A activity in RM cells propagate further into the sporulation-specific cascade of gene expression by analyzing the expression of the gerE gene, which depends on the late mother cell-specific sigma factor, SigK, thus reflecting the final steps of sporulation [109]. The expression of gerE-luc in RM cells occurred within a narrow pulse starting ~1.5 hours earlier and reaching a ~10-fold higher maximum than in WT cells (Fig 6D; note different ordinates for the WT and RM gerE-luc expression curves). Such kinetics would account for more synchronous sporulation in the RM population, most probably due to efficient initiation of the process by high Spo0A~P.
In summary, different expression patterns of spo0A and Spo0A-regulated genes in WT and RM cells account for more efficient phosphorylation of Spo0A in the absence of Rho, both under biofilm- and sporulation-promoting growth conditions. In RM cells, the rapid increase of Spo0A~P to a high level in MSgg medium could inhibit matrix gene transcription and thus impair biofilm development, whereas in DS medium, higher Spo0A~P levels would trigger sporulation earlier and in a larger sub-population of cells.
We then assessed whether the effects of the Rho mutation on the expression of sporulation genes leads to more productive sporulation. The laboratory B. subtilis 168-related strains are sporulation-proficient and could thus be used for this analysis. We used the exhaustion method to induce sporulation. The first heat-resistant spores were detected in BSB1 RM cultures four hours after entry into sporulation (T4), and by T7, almost 100% of the RM cells had formed spores (Fig 7A). Less than 20% of the WT cells had produced spores by the same timepoint, reflecting the well-known dichotomy of sporulation in B. subtilis [104, 109, 110]. We performed the same experiment using other B. subtilis strains: non-domesticated NCIB 3610; TF8A, a phage-cured derivative of 168; and PY79, a laboratory prototroph strain genetically distant from 168 [64, 111]. Deletion of the rho gene accelerated sporulation in all genetic backgrounds and most of the RM cells produced mature spores by T7(Fig 7B).
We examined whether rho deletion can suppress sporulation defects of the kinA and kinB mutants to formally link more efficient sporulation by RM cells to increased activity of the Spo0A phosphorelay. KinA is commonly considered as the main sporulation kinase, as its inactivation severely inhibits this process. The inhibitory effect of kinB mutations is more variable and apparently depends on the genetic background of the cells [53, 112, 113]. We cultured the kinA and kinB mutants of BSB1 and PY79 under sporulation conditions and reproduced both reported trends: the kinA mutation strongly reduced sporulation in both strains, whereas the inhibitory effect of the kinB mutation was strong in BSB1 and weak in PY79 (Fig 7C and S7 Fig). We used BSB1 derivatives for further experiments to remain consistent with the gene expression analysis, although the efficiency of sporulation was generally lower in the BSB1 than PY79 background.
Inactivation of Rho improved the sporulation of BSB1 kinA and kinB mutants, although to different degrees: an increase of ~15 fold in RM kinA and ~2.5-fold in RM kinB strains relative to the BSB1 kinA and kinB mutants (Fig 7C). Similarly, the rho mutation preferentially rescued the sporulation defect of the PY79 kinA mutant, although the effect was relatively small (S7A Fig). Inactivation of Rho in the double kinA kinB mutants had no effect on the basal level of sporulation in either background (S7B Fig; data presented for the PY79 mutant derivatives). Altogether, the observed genetic interactions indicate that the stimulatory effect of rho deletion on sporulation mostly involves the KinB kinase, suggesting its increased role in the Spo0A phosphorelay system in RM cells.
We tested this hypothesis by analyzing the expression of the Spo0A~P-dependent spoIIAA-AB-sigF operon during sporulation of the kinase mutants. Expression of spoIIA-luc was similarly inhibited in BSB1 kinA and kinB mutants, indicating a strong decrease of Spo0A activity in the absence of either kinase (Fig 7D), corroborating the results of the sporulation assay in these strains. The expression of spoIIA in RM kinA cells was higher than the wild-type level, nearly reaching the maximum observed in RM cells (Fig 7D). However, spoIIA induction was delayed in RM kinA cells relative to RM cells (Fig 7D), which might underlie their different sporulation efficiencies (Fig 7C). Rho inactivation in the kinB mutant (RM kinB) also improved the expression of spoIIA, however it remained below the wild-type level (Fig 7D). The pattern of spoIIA expression in RM kinB cells thus correlates with their low sporulation efficiency. Altogether, these results indicate that the increased phosphorylation of Spo0A in the RM strain is mainly due to KinB.
As mentioned above, transcriptome analysis revealed an abrupt down-shift within the kinB gene in WT, but not RM cells, explaining the higher expression of KinB in the absence of Rho (Fig 8A). The internal down-shift of kinB transcription is also observed in cells depleted of the RNaseY, RNase J1, or RNase III ribonucleases [114, 115], excluding its formation by the action of these enzymes. Analysis of the kinB region using Petrin [17] and MFOLD software [83] did not reveal any putative secondary structures characteristic of intrinsic terminators. Altogether, these observations suggest that the internal down-shift of kinB transcription is due to the activity of an intragenic Rho-dependent terminator.
We used a plasmid-based system for the detection of transcription terminators [116, 117] to more firmly establish the role of Rho in termination within the kinB gene. Two DNA fragments containing the kinB promoter, translation initiation region (TIR), and differently sized 5’-terminal regions of the kinB orf were cloned in front of the cat gene, encoding chloramphenicol (Cm) acetyltransferase (Fig 8B). The transcript of the long fragment (first 417 bp of kinB orf; plasmid pKinB-Long) likely contained sequence features required for termination (estimated to be ~350 ribonucelotides downstream of the kinB start codon), whereas the transcript of the small fragment (first 157 bp of kinB orf; plasmid pKinB-Short) did not (Fig 8B). The truncated kinB orf ended with a stop codon in both plasmids to ensure that translation of the cat mRNA initiating from the kinB promoter depended on its own RBS. At the same time, a transcription terminator located upstream of the cat gene would decrease cat expression and consequently Cm-resistance [117]. As shown in Fig 8C, BSB1 WT and RM cells containing the pKinB-S plasmid were mostly resistant to the tested Cm concentrations. In contrast, the pKinB-L plasmid conferred similar Cm-resistance only to RM cells, whereas WT cells containing this plasmid were considerably more sensitive (Fig 8C). These results are consistent with Rho-dependent termination within the large kinB fragment cloned into pKinB-L.
According to the E. coli model, Rho loads onto nascent transcripts not protected by translating ribosomes [22–25]. Therefore, Rho-dependent intragenic termination is modulated by the efficiency of translation initiation [118]. Thus, the efficiency of Rho would depend on kinB translation if it has a direct role in transcription termination within kinB. We varied the translation initiation rate of kinB by replacing its TIR, containing an imperfect RBS sequence (RBS wt), by a TIR with a strong RBS (RBSm+) or a TIR with degenerated RBS (RBSm-; Fig 8D) [119]. The pKinB-S-RBSm+ and pKinB-S-RBSm- plasmids conferred similar levels of Cm-resistance in WT and RM cells as the pKinB-S plasmid (Fig 8E, data shown for WT cells). This demonstrated that modifications of the kinB TIR had no effect on translation of the cat mRNA. In contrast, WT cells carrying the pKinB-L-RBSm+ plasmid were considerably more resistant to Cm than cells carrying pKinB-L plasmid, whereas the absence of an active RBS in pKinB-L-RBSm- resulted in much lower Cm-resistance (Fig 8F). However, rho inactivation in cells bearing the pKinB-L-RBSm- plasmid significantly improved their resistance to the antibiotic and restored viability at high Cm concentrations (Fig 8G).
Altogether, these results confirm the major role of Rho in the premature termination of kinB transcription.
Thorough analysis of the kinB transcription profile in B. subtilis BSB1 across a database of 104 different growth conditions [17] revealed the presence of an intragenic down-shift of kinB expression in a substantial proportion of the dataset (S5 Table, Fig 9A and 9B). However, the expression level was remarkably similar between the 5’ and central segments of the kinB gene during the initial stages of sporulation, which correspond precisely to conditions in which kinB expression has been identified to be maximal (S5 Table). However, expression of the 5’ segment of kinB appeared to be relatively constant between non-sporulating cells (e.g. transition phase) and cells entering sporulation (e.g. S1; Fig 9B). This is in accordance with previous data, indicating that there is no specific regulation of the activity of the kinB promoter during sporulation [53]. Thus, intragenic transcription termination appears to be a mechanism for the control of kinB expression, which is weaker at the onset of sporulation.
The amount or activity of Rho should decrease during the early stages of sporulation if it regulates kinB expression. We assessed changes in the cellular level of Rho protein by constructing a Rho-SPA translational fusion, which retains regulatory activities of non-modified Rho (S8 Fig), and monitoring its expression in WT cells during growth in DS medium. Rho-SPA protein levels started to decrease after the mid-exponential growth phase and became barely detectable two hours after growth arrest (Fig 9C and 9D). This is in accordance with previous transcription analyses, showing that the level of rho expression is relatively high during exponential growth in rich medium and decreases during sporulation [17]. Previously, B. subtilis Rho was found to auto-regulate its expression by transcriptional attenuation at the Rho-dependent terminator(s) located within the leader region of rho mRNA [120]. Our results suggest that regulation of Rho expression during sporulation is more complex.
In conclusion, increased kinB expression due to the absence of intragenic transcription termination at early stages of sporulation appears to coincide with a decrease of Rho protein content. The mechanism(s) regulating Rho protein expression and/or stability during sporulation remain(s) to be elucidated.
The negative correlation between Rho content and kinB expression during sporulation in WT cells suggests that Rho termination activity may be dose-dependent. We tested this possibility by investigating the physiological effects of Rho overexpression using the middle-copy number plasmid pDG148Rho (hereafter pRho), which constitutively expresses Rho at a level which we estimate to be ~3-fold higher than normal (S9 Fig).
We started by testing the effect of Rho overproduction on the efficiency of the kinB intragenic Rho-dependent terminator. Introduction of the pRho plasmid into WT cells containing pKinB-L considerably decreased their Cm-resistance (Fig 10A, data shown for Cm 3μg/ml; compare with Fig 8C). In contrast, pRho did not affect Cm-resistance of WT cells carrying the pKinB-S plasmid, which does not contain the Rho-dependent terminator (Fig 10A). This indicated that the efficiency of the kinB intragenic transcription terminator increased with higher levels of Rho. Indeed, Western-blot analysis of the kinB-SPA translational fusion showed that cells grown in sporulation-inducing DS medium produced less KinB kinase in the presence of pRho than of a control vector (Fig 10B). Next, we assessed Spo0A phosphorylation and cell commitment to sporulation using the reporter spoIIA-luc fusion. Introduction of pRho into WT and RM cells decreased activity of the spoIIA promoter by ~3 fold relative to respective vector-containing controls (Fig 10C). These results suggest less efficient Spo0A phosphorylation when Rho is produced above its natural level. Of note, pRho-mediated inhibition of spoIIA activity was higher in the WT strain, which has the rho gene in the chromosome. The sporulation efficiency of the WT strain containing pRho was ~3 fold lower than that of control (Fig 10D), consistent with low Spo0A activation. A similar (~ 5-fold) inhibition of sporulation by pRho was observed in the BSB1 kinA mutant. In contrast, pRho had no significant effect on sporulation in BSB1 kinB cells, showing again that Rho preferentially targets kinB transcription (Fig 10D). Altogether, these results show that Rho overexpression inhibits sporulation, most likely by decreasing activity of the Spo0A phosphorelay system. One of the determinant factors of this inhibition is reinforced intragenic termination of kinB transcription.
Rho overexpression has the opposite effect of rho deletion, resulting in much less efficient sporulation. It is thus possible that Rho overexpression could stimulate the developmental programs for which Rho inactivation is inhibitory. We addressed this possibility by analyzing biofilm formation and swarming motility of strains containing pRho. Presence of the pRho plasmid rescued the negative biofilm phenotype of the NCIB 3610 RM strain, but had no global effect on naturally robust biofilms formed by NCIB 3610 WT (Fig 10E). Apparent insensitivity of biofilm formation to Rho overexpression in WT cells was expected due to known functional redundancy between phosphorelay kinases and the minor involvement of KinB in biofilm development under conditions of MSgg growth [85, 121]. However, the introduction of pRho not only suppressed the sessile phenotype of NCIB 3610 RM, but also increased swarming capacities of the parental swarming-proficient NCIB 3610 WT strain (Fig 10F and 10G). This higher-than-natural motility phenotype suggests that Rho over-production increases a subpopulation of WT cells with active SigD.
These experiments established that Rho inactivation and overexpression have opposite effects on B. subtilis physiology. This suggests that cells might be sensitive to intercellular variations of a naturally low level of Rho expression [120, 122].
Here, we report that pervasive transcription controlled by the termination factor Rho is an integrative element of the genomic regulatory networks that govern phenotypic heterogeneity and cell decision making in the Gram-positive model bacterium B. subtilis.
The transcriptome and proteome analyses presented here demonstrate the important effects of rho deletion on the B. subtilis gene expression landscape, encompassing approximately one-tenth of the known functional regions in a given growth condition. Prominent alterations of sense-strand expression in RM cells are caused by a combination of direct cis up-regulation due to transcription read-through of Rho-dependent terminators, and, in some cases, cis down-regulation of genes confronted with antisense transcription. These primary events propagate into regulatory networks and cause other changes that can be considered to be indirect trans effects. Recently, another transcription factor, NusA, was shown to regulate global gene expression in B. subtilis by controlling transcription read-through at suboptimal intrinsic terminators. Depletion of this essential protein caused a substantial increase of antisense transcription and misregulation of many genes mainly involved in DNA replication and DNA metabolism [84].
Physiological analyses of RM strains demonstrated that numerous changes in gene expression caused by Rho inactivation are not fortuitous. Indeed, they are related to biologically relevant phenotypes linked to three distinct B. subtilis cell fates: the synthesis of flagella leading to cell motility, matrix production underlying biofilm formation, and sporulation. These mutually exclusive developmental programs are controlled by complex regulatory networks, which are organized in a way that avoids their simultaneous activation in the same cell. Each program is characterized by a high level of phenotypic heterogeneity and bistable expression of a specific set of genes [37–40, 43, 44, 71]. The biological significance of Rho-controlled transcription is illustrated by the opposite phenotypes of RM and Rho over-expressing strains, corresponding to its high and low steady-state levels, respectively. RM cells are mostly sessile, exhibit low extracellular matrix production, and sporulate with high efficiency. In contrast, cells over-expressing Rho sporulate weakly, but are highly motile. These opposite phenotypes are determined by a specific architecture of the Rho-controlled transcriptome, of which the elements appear to be organized for the simultaneous stimulation of sporulation and repression of the principle alternative programs, once the control by Rho is removed.
Several Rho-controlled transcripts within the motility differentiation program act to down-regulate expression and activity of the motility-specific sigma factor SigD. In the absence of Rho, independent events of read-through transcription at Rho-dependent terminators directly up-regulate slrR and slrA genes which products negatively control sigD expression [59, 60–62, 71, 78]. Simultaneously, a Rho-dependent antisense transcript down-regulates the flhO-flhP operon, encoding components of the flagella export apparatus, which is essential for secretion of the anti-SigD factor FlgM [69, 81]. Altogether, these events disturb the self-reinforcing circuit of sigD expression and bias the SigD-ON/SigD-OFF bistable switch of motility towards a SigD-OFF state. Other factors contributing to motility are also affected by Rho inactivation, but were not analyzed in this study. For example, the flagellar chaperons FliS and FliD, encoded by the yvyC-fliD-fliS-fliT operon, were down-regulated in RM cells, probably as a result of the Rho-controlled asRNA targeting this locus (Table 1, S1 and S4 Tables).
Concomitant with the inhibition of flagellar synthesis, the absence of Rho severely alters the program of biofilm development. Both epsA-epsO and tapA-sipW-tasA operons encoding the main components of the extracellular matrix are down-regulated in RM cells. Weak expression of both matrix operons results from reinforcement of SinI/SinR-mediated repression of their promoters caused by a deregulated phosphorylation of Spo0A. Our analysis demonstrates that Spo0A~P rapidly accumulates beyond a level needed to induce matrix production and reaches a higher, sporulation-triggering, threshold, due to increased activity of Spo0A phosphorelay in RM cells. High Spo0A~P should inhibit expression of the SinI anti-repressor and, consequently, reinforce SinR-mediated repression of the matrix operons.
Besides this indirect trans effect on the expression of matrix operons, inactivation of Rho provokes the generation of a new antisense transcript, which spans across the entire epsA-epsO operon and, as shown here, contributes to the inhibition of biofilm formation in RM cells. The exact mechanism by which eps asRNA inhibits expression of the eps operon remains to be established, but it may directly interfere with activity of the eps promoter and/or the RNA switch located between the epsB and epsC genes, which allows transcription of the long eps operon [123], or act post-transcriptionally [30].
Additionally, we detected down-regulation of the skf operon (S1 Table; http://genome.jouy.inra.fr/cgi-bin/seb/index.py) involved in the production and release of the cannibalism toxin, Skf, known to stimulate biofilm development and delay sporulation [38, 124]. Thus, inefficient production of Skf might have also contributed to reduced biofilm formation by RM cells. Expression of the skfA-H operon is inhibited by a high level of Spo0A~P [43, 58], similarly to the matrix operons. In addition, highly expressed asRNA overlapping the entire skf operon was revealed in the RM strains (S1 Table). The putative role of the skf-specific asRNA in the regulation of the skfA-H operon merits further analysis.
All Rho-controlled transcripts detected within the Spo0A phosphorelay are sense and lead to up-regulation of the cognate genes when de-repressed in the absence of Rho. Aside from the rapA gene, other targets of Rho-controlled transcription encode positive factors of Spo0A phosphorylation: sensor kinase KinB, its positive effector KbaA, and the phosphotransferase Spo0B. Pervasive transcription, together with several positive feedback loops of spo0A expression, intensifies phosphorelay activity. As a result, RM cells engage in differentiation more efficiently and synchronously than WT cells, characterized by broadly heterogeneous Spo0A phosphorylation [104].
Key factors involved in the increase of phosphorelay activity in RM cells are the up-regulated KinB and, to a lesser extent, KinA kinases, which are known to trigger sporulation if over-expressed [49, 63]. In RM cells, kinA up-regulation is most likely due to increased Spo0A~P, acting within a positive auto-regulatory loop [104], and thus would be an indirect trans effect of Rho-controlled transcription. In contrast, up-regulation of kinB in RM cells is the direct result of a read-through at intragenic transcription terminator. We experimentally proved that the 5’-terminal segment of the kinB transcript contains sequence features required for Rho-dependent transcription termination. We showed that the efficiency of intragenic kinB termination depends on Rho availability and negatively correlates with the kinB translation initiation rate in a Rho-dependent manner. This anti-correlation is in accordance with the well-documented preferential activity of Rho at untranslated RNAs and its role in the control of transcription-translation coupling [23–25, 118].
The global transcriptional regulators, AbrB and CodY, and positive stringent control regulate the expression of KinB at the transcription initiation level [125–127]. Our results highlight a novel regulatory mechanism of kinB expression that acts through the premature termination of transcription. Inactivation of Rho and, consequently the lack of termination, generates a full-length kinB transcript, leading to an increased level of KinB, higher levels of Spo0A~P, and the activation of sporulation. In contrast, high Rho amounts strengthen intragenic termination of kinB, decrease cellular levels of KinB and phosphorelay activity, and lead to a weak-sporulation phenotype.
The intragenic termination of kinB in WT cells is less efficient at early stages of sporulation, correlating with decreased rho expression. Programmed decrease of Rho amount might be needed to compensate probable strengthening of the premature kinB termination due to decrease of the translation efficiency under nutrient-limiting conditions. This emphasizes the biological significance of Rho-mediated control of kinB expression. By premature termination of kinB transcription, Rho delays sporulation, giving cells the possibility to continue exploring the environment. Rho might also influence other developmental programs, as KinB kinase has been found to be involved in the control of sliding motility and biofilm formation under particular growth conditions [47, 128, 129]. Thus, Rho-mediated control of kinB expression appears to be an integral part of the deterministic regulation of B. subtilis development, mediated by Spo0A.
Recent study in E. coli has established that Rho acts as a global regulator of genes expression during the exponential growth by prematurely terminating transcription within the 5’ un-translated regions of hundreds of genes, including the global stress response sigma factor rpoS gene. Rho-mediated control of rpoS expression is modulated by the regulatory sRNAs and is relieved at the stationary phase of growth [130]. This and our analyses illustrate the diversity of strategies by which bacteria employ Rho-mediated transcription termination to adapt to the environmental and metabolic changes.
Our data suggest another potential role of Rho: that as a factor of phenotypic heterogeneity within B. subtilis population. Heterogeneity based on intrinsic fluctuations of gene expression provides potential flexibility to a genetically homogenous population to respond to environmental changes [131]. We showed that population heterogeneity is considerably reduced when Rho-controlled transcription levels are artificially high or low. This suggests that pervasive transcription in WT cells varies depending on the intracellular concentration of Rho. This was already suggested by our previous study, which noted that the length of sense and antisense 3’extensions generated by read-through transcription negatively correlates with rho expression [17].
B. subtilis Rho is a low abundant protein present at 0.5 to 4.8% of the level of RNA polymerase [17, 120, 122]. According to the E. coli paradigm, Rho should function as a hexamer [22]. This suggests that even minor variations of Rho levels might have substantial impact on the efficiency of pervasive transcription. The stochastic generation of pervasive, often antisense, transcription targeting gene expression via various mechanisms [29, 30, 132] may increase intercellular heterogeneity and phenotypic variation within isogenic B. subtilis population. A similar hypothesis was recently proposed following analysis of the Clostridium botulinum Rho protein expressed in heterologous systems [133].
Comparative analyses of closely related bacterial species have revealed both conservation of and significant differences between respective non-coding transcriptomes. It has thus been proposed that pervasive transcription represents a major element of inter-strain divergence, providing a potential for physiological adaptation [8, 134, 135]. Here we show that several transcripts regulated by Rho are similarly functional in different B. subtilis strains (BSB1, NCIB 3610, PY79). Moreover, the nucleotide sequences of the corresponding genomic regions are conserved among B. subtilis strains (≥ 99% identity), while the conservation level is lower in other Bacillus species (for example, the kinB and flhO-flhP genomic regions of B. subtilis and B. amyloliquefaciens present only 67%-73% of nucleotide identity, respectively). Thus, RNA features controlled by Rho might represent a specific trait fixed by evolution, at least in B. subtilis.
It is important to note that the Rho-dependent regulatory network in B. subtilis may be broader than it emerges from our analysis, as Rho-controlled transcripts expressed under other conditions and/or dependent on alternative sigma factors, may have escaped identification. Future studies in B. subtilis and other bacteria will help to understand how elements of the Rho-controlled pervasive transcription are recruited to achieve important regulatory functions.
Our results support the view that, in terms of gene regulation, transcription termination can be as important as the repression or activation of transcription initiation. They advance our understanding of the role of pervasive transcription in bacteria, considered for a long time as a “dark matter” of bacterial transcriptomes. Part of Rho-controlled transcription appears to constitute an integral module of the B. subtilis cell differentiation regulatory network instead of simply being non-functional transcriptional noise. This ranks termination factor Rho among the global regulators of B. subtilis.
B. subtilis strains used in the work are listed in S6 Table. When needed, cells contained plasmids as indicated in Results section. Cells were routinely grown in Luria-Bertani liquid or solidified (1.5% agar; Difco) medium at 37°C. Standard protocols were used for transformation of E.coli and B. subtilis competent cells. SPP1 phage was used for transduction of NCIB 3610 strains as described [136]. Biofilm formation was analyzed in liquid (for pellicles) or solid (for colony biofilms) MSgg medium [96]. Sporulation was analyzed in supplemented Difco Sporulation medium (Difco) [137]. When required for selection, antibiotics were added at following concentrations: 100 μg per ml of ampicillin, 60 μg per ml of spectinomycin, 0.5 μg per ml (for B. subtilis) and 90 μg per ml (for E. coli) of erythromycin, 3 μg per ml of phleomycin, 5 μg per ml of kanamycin, and 5 μg per ml of chloramphenicol.
B. subtilis strains and the plasmids used in the study are listed in S6 Table. The plasmids were constructed in E. coli TG1 strain. The used oligonucleotides are listed in S7 Table.
To repair NCIB 3610 RM (Δrho::phleo) strain back to the wild type rho allele, a near-by DNA fragment containing ywjH gene was amplified using oligonucleotides eb460 and eb461, digested by HindIII and BamHI endonucleases and cloned at pMutin4 plasmid [138]. The resulting plasmid was integrated by single crossover in the ywjH locus of BSB1 chromosome and subsequently transferred in NCIB 3610 RM with selection at erythromycin. NCIB 3610 RM ywjH::pMutin transductants were tested for sensitivity to phleomycin that indicated substitution of the rho deletion by the wild type rho allele. Phleomycin-sensitive clones were further selected for loss of the erythromycin-resistance that indicated excision of pMutin4 from the chromosome and restoration of the ywjH gene. Thus obtained NCIB 3610 rho wt* clones were controlled for integrity of the rho and ywjH wild type alleles by PCR.
To inactivate kinB gene, the internal part of the gene was amplified using oligonucleotides veb608 and veb610, digested with HindIII and EcoRI endonucleases and cloned at pMutin4 plasmid. The resulting plasmid was integrated by single crossover in the kinB locus of BSB1 chromosome leading to its disruption.
The translational fusions of the kinA, kinB and rho genes with the sequential peptide affinity (SPA) tag sequence were constructed for immunoblot analysis of the proteins. The 3’- terminal parts of the kinA and kinB genes were amplified using pairs of oligonucleotides eb625 and eb626 and veb606 and veb607, respectively. The amplified kinA and kinB DNA fragments were digested, respectively, by BamHI and NcoI and by Acc65I and NcoI endonucleases, and ligated with pMutin-SPA plasmid [139] cutted by BglII and NcoI for the kinA, and by Acc65I and NcoI for the kinB cloning. The rho-SPA fusion was constructed using ligation-independent cloning as described [140]. The 3’-terminal part of rho was amplified using oligonucleotides rhoSpa-Fwd and rhoSpa-Rev, treated with T4 DNA polymerase in the presence of dTTP and annealed to pMUTIN-LICSPA plasmid [141], linearized by AscI endonuclease and treated with T4 DNA polymerase in the presence of dATP. The annealing mixture was transformed into E.coli cells. The resulting plasmids with the kinA-, kinB- and rho-SPA fusions were transferred in BSB1 cells where they integrated into respective chromosomal loci by single crossover.
To express the flhO and flhP genes from ectopic position, the flhOP operon was PCR-amplified using oligonucleotides eb617 and eb618, digested by Acc65I and BamHI endonucleases and cloned onto integrative plasmid pSG1729 [142]. The resulting plasmid was integrated in the amyE locus of BSB1 chromosome by double crossover with selection of the chloramphenicol-resistant transformants, which lost amylase activity.
To suppress anti-sense transcription within the eps operon, the 5’-part of epsO gene was amplified using oligonucleotides veb680 and veb681, cutted by EcoRI endonuclease and cloned at pMutin4 plasmid between PacI site filled-in by T4 polymerase and EcoRI site. In the resulting plasmid, the 3’ end of the inserted fragment is flanked by three intrinsic transcription terminators of the vector [138]. The plasmid was integrated in the epsO locus of BSB1 chromosome by single crossover and subsequently transferred in NCIB 3610 WT and RM cells. In the mutant epsO:Ter allele, the inserted transcription terminators are oriented to block transcription of the eps asRNA initiated near the 3’- end of epsO gene.
To analyze genes expression, pUC18Cm-luc plasmid was used to construct genes transcriptional fusions with the butterfly luciferase gene luc [97]. The promoters of spoIIA, gerE, epsA and tapA genes were amplified together with the upstream chromosomal fragments of ~1 Kbp, using the corresponding pairs of oligonucleotides listed in S7 Table. The fragments containing spoIIA and gerE promoters were digested by HindIII and BamHI endonucleases and cloned at pUC18Cm-luc. The luc fusions with epsA and tapA promoters were obtained by the assembly Gibson’s procedure with linear vector amplified with oligonucleotides F- pUC18-luc and R- pUC18-luc [143] (S7 Table). The obtained plasmids were used to transform B. subtilis where they integrated by single crossover at chromosomal loci of the targeted genes. This event reconstructs natural regulatory region of gene upstream the fusion and an intact copy of gene downstream.
To inactivate the anti-SigD factor FlgM without inducing polar effects on the downstream genes, a marker-less in-frame flgM deletion was constructed by a two-step integration-excision method similar to the previously described [144]. Two chromosomal fragments of ~1 Kbp were amplified using oppositely oriented and partially over-lapping oligonucleotides veb687 and veb688, which match close to the extremities of flgM gene, and their counterpart primers veb686 and veb689, respectively. The amplified fragments were joined by PCR using primers veb686 and veb689 and cloned between the BamHI and SalI sites at the thermo-sensitive plasmid pMAD [145]. The resulting plasmid contains a fragment of B. subtilis chromosome with in-frame deletion of the most part of flgM gene (flgMΔ63). The flgMΔ63 structure was controlled by sequencing. The plasmid was transformed in BSB1 cells with selection for erythromycin-resistance at non-permissive for replication temperature 37°C. This led to plasmid insertion into the chromosome by single crossover and duplication of yvyF-csrA locus, which copies contained the wild type or flgMΔ63 alleles. The duplicated region was transferred in NCIB 3610 RM cells, which were further cultivated at permissive 30°C without erythromycin to stimulate excision of the yvyF-csrA duplicate from the chromosome. The resulting clones were controlled for the presence of flgMΔ63 allele by PCR using primers veb686 and veb689.
For Rho overproduction, rho gene was amplified using oligonucleotides eb423 and eb424; digested by NheI and BamHI endonucleases and cloned at pDG148 [146] plasmid between the XbaI and BamHI sites. The resulting plasmid pRho is deleted for lacI repressor gene and expresses Rho constitutively from Pspac promoter. To construct the control vector pDG148Δlac, the XbaI-BamHI double-cutted pDG148 was treated by T4 DNA polymerase and self-ligated. To estimate Rho production from Pspac promoter, pRho plasmid was modified to express a SPA-tagged protein. The 3’-terminal part of the chromosomal rho-SPA fusion (see above) was amplified using oligonucleotides eb423 and pdg148-rev, which matches the sequence behind the SPA-tag, and digested by XhoI endonuclease. The PCR product was cloned at pRho plasmid between BamHI site filled-in by T4 DNA polymerase and XhoI site.
To analyze presence of the Rho-dependent terminator within kinB gene, two plasmids were constructed which contain kinB promoter and differently sized 5’-terminal parts of the gene. The kinB gene and the upstream 650 bp region were amplified using oligonucleotides veb610 and veb611. The fragment was digested either by SmaI and NsiI or by SmaI and MboI endonucleases and the fragments of 390 and 650 bp, respectively, were gel-purified and cloned at the terminator-screening vector pGKV210 [116] digested, respectively, by SmaI and PstI or by SmaI and BamHI endonucleases. The resulting plasmids pKinB-S(hort) and pKinB-L(ong) were transformed into B. subtilis cells with selection at erythromycin. Both plasmids contain the promoter of kinB gene and the first 157 and 417 bp of kinB orf, respectively.
To modify efficiency of kinB translation initiation at pKinB-L and pKinB-S plasmids, the whole pKinB-S was amplified using the oligonucleotides veb678 and veb679 or veb679 and veb690 designed to substitute natural kinB ribosome binding site (RBS) by a stronger or a weaker, respectively. PCR products were treated by DpnI endonuclease, to degrade template DNA, 5’-phosphorylated by T4 polynucleotide kinase, self-ligated and transformed in E. coli cells with selection at erythromycin. The resulting plasmids were used as templates for amplification of the modified kinB fragments using the oligonuclotides veb676 and eb407. PCR products were controlled by sequencing, digested by EcoRI and NarI endonucleases and cloned at similarly digested pKinB-L or pKinB-S. The resulting derivative plasmids contain either canonic GGAGGA (RBSm+) or a weak TGATAA (RBSm-) RBSs (Fig 8D).
Tiling array data obtained with a strand-specific resolution of 22 bp for exponential growth in LB [17] were reanalyzed. This re-analysis used the same signal processing and gene-level aggregation procedures as the initial study but differed by the normalization and differential expression analysis. Briefly, the raw log2-transformed hybridization signal was smoothed with an algorithm that accounts for probe-specific biases and changes in expression levels between adjacent regions that can take a form of abrupt shifts and more continuous drifts [17]. Then, a whole genome transcription profile was aggregated into sense and antisense gene level data by computing the median of the smoothed signal on the sense and antisense strand of a repertoire of native expression segments, i.e. detected as transcribed in the wild-type in one out of 269 hybridized RNA samples intended to capture the diversity of the lifestyles of the bacterium [17].
To allow precise between-sample comparison of expression levels on the both sense and antisense strands of the native expression segments, the quantile normalizing transformation fitted on aggregated sense strand levels was also applied to the antisense strand levels and to the smoothed transcription profiles as described [20]. Statistical comparison of the 3 biological replicates for RM and WT relied on moderated t-statistics computed with the functions “lmFit” and “eBayes” of R package “limma” [147]. Control the False Discovery Rate relied on q-values obtained with R package “fdrtool” [148]. Sense strand and antisense strand levels were considered simultaneously in these analyses.
The same statistical procedure served here to examined the expression levels immediately downstream a repertoire of 3242 transcriptional up-shifts encompassing promoters of most genes [17]. We considered that an up-regulation was detected at a given promoter if the downstream smoothed-normalized signal exhibited differential expression according to the specified amplitude (log2 RM/WT ≥ 1) and false discovery rate (q-value ≤ 0.01) cut-offs and if the downstream transcription level was at least twice higher than the upstream level in the 3 biological replicates for RM (indicating activity of this promoter as opposed to transcription from an upstream promoter). Reciprocally, down-regulation was considered detected when log2 RM/WT ≤ -1, q-value ≤ 0.01, and the downstream transcription level was at least twice higher than the upstream level in the 3 biological replicates for WT.
RNA was extracted from B. subtilis WT and RM derivative strains grown in LB or MSgg medium at 37°C under vigorous agitation up to an OD600nm ~0.5. RNA preparation and DNase treatment were done as described [17]. Quality and quantity of RNA samples were analyzed on Bioanalyzer (Agilent, CA). For analysis of the antisense transcription of flhOP and eps operons by RT-PCR, cDNA was synthesized using flhO, epsL and 16S rRNA specific oligonucleotides eb700, eb706 and eb715, respectively, (S7 Table) and 50 ng of total RNA as a template in reaction with SuperScriptIV Reverse Transcriptase (Invitrogen) according to the supplied protocol, and treated with RNaseH (Invitrogen) for 20 min at 37°C. To amplify internal DNA fragments, PCR (35 cycles) was performed by Thermo Scientific DreamTaq DNA Polymerase (ThermoFicher) with the oligonucleotides pairs specific for flhO (eb705 and eb702), epsK (eb708 and eb710) and 16S rRNA genes (eb715 and eb716) (S7 Table).
RNA samples of B. subtilis BSB1, NCIB3610, and the corresponding rho-deletion mutants were prepared as described in the previous section. Transcriptome analysis was performed by Transcriptome and EpiGenome platform (Pasteur Institute, France). Briefly, RNA samples have been submitted first to ribosomal RNA depletion using the RIBOZero rRNA removal kit Bacteria (Illumina, San Diego, California). Purified fraction was then treated for library preparation using the Truseq Stranded mRNA sample preparation kit (Illumina, San Diego, California) according to manufacturer’s instruction. Fragmented RNA samples were randomly primed for reverse transcription followed by second-strand synthesis to create double-stranded cDNA fragments. No end repair step was necessary. An adenine was added to the 3'-end and specific Illumina adapters were ligated. Ligation products were submitted to PCR amplification. Sequencing was performed on the Illumina Hiseq2500 platform to generate single-end 65 bp reads bearing strand specificity. Reads were trimmed based on sequencing quality using Sickle (v1.200) and mapped on AL009126.3 reference genome assembly using Bowtie2 (2.2.6; options "-N 1 -L 16 -R 4") [149] before read-count aggregation on the sense and antisense strand of each transcribed region (annotated genes and S-segments) with Htseq-count (0.6.0; standard options). Raw sequencing data and aggregated counts have been deposited in GEO (GEO submission number GSE94303). Experiments were made in duplicates for B. subtilis NCIB3610 to allow statistical differential expression analysis. RPKM normalization [150] served for a first level of exploratory analysis. Differential expression analysis of B. subtilis NCIB3610 relied on R library “DESeq2” [151] and associated "median ratio method" normalization procedure. Normalization relied on a control set of 1152 always well expressed sense regions a priori less impacted by low-level transcriptional read-through typical of the rho-deletion mutants. These were ranking in the 25% highest density of mapped reads in each of the four NCIB3610 samples. DESeq2 p-values were converted into q-values using R library “fdrtool” [148]. While the initial differential expression analysis relied on the four NCIB3610 samples, we also performed another, more discriminative, differential expression analysis excluding one of the parental NCIB3610 sample which exhibited anti-sense transcription levels markedly higher than other RNA-Seq (B. subtilis BSB1 and NCIB3610) and tiling array (B. subtilis 1012) samples of the parental strains probably because of more advanced growth status.
B. subtilis 168 derivative strains were grown in LB medium at 37°C under vigorous agitation up to an OD600nm ~0.6. Cells were harvested by centrifugation (6,000g for 10 min at 4°C), washed once with 50 mL of buffer A (10 mM Tris-Cl pH 7.5, 150 mM NaCl) before being centrifuged again. The cell pellets were frozen in liquid nitrogen and kept at -80°C. Cell pellets were thawed on ice and resuspended with 5 mL of buffer A, and disrupted by French press (pressure 2.7 MPa). Unbroken cells were removed by centrifugation at 15,000 RPM, and the supernatants were centrifuged at 100,000g for 1 hour at 4°C. The resulting supernatants were kept as the cytosolic fraction. The pellets were then washed twice with cold buffer A, and centrifuged twice at 100,000g for 1 hour at 4°C. The pellets were re-suspended in TE (20mM Tris, 2 mM EDTA) and considered as the membrane fraction. All experiments were carried out in duplicate. Membrane and cytosolic protein concentrations were measured using the Bradford method (Bio-Rad kit). Membrane and cytosolic samples were treated differently before separation by electrophoresis. Samples corresponding to the membrane fractions were mixed with a loading buffer containing 125 mM Tris-Cl pH 6.8, 20% glycerol, 10% SDS and 0.1% bromophenol blue, and left overnight at room temperature. Equal amounts of cytosolic proteins for each sample were treated with a classic Laemmli loading buffer and boiled for 5 min. Samples were then loaded on a 10% Bis-Tris polyacrylamide NuPAGE gel (Invitrogen) and the electrophoresis was left running at 100V for 15 min. The gel was then stained with Bio-Safe Coomassie G-250 Stain (Bio-Rad). After distaining the bands of 2 mm-wide along the protein migration lane were cut off and used as samples for the identification of the proteins by mass spectrometry. The gel pieces for each sample were washed twice with 0.2% TFA-50% acetonitrile, reduced by 10 mM DTT for an hour at 56°C, alkylated by 50 mM iodoacetamide for 1 hour at room temperature into darkness. Sequencing grade modified trypsin (Promega) diluted in 25 mM NH4HCO3 was added for 18 hours at 37°C. Tryptic peptides were recovered by washing the gel pieces twice with 0.2% TFA-50% acetonitrile, once with 100% acetonitrile and the supernatants were evaporated to dryness. The peptides were then re-suspended in 25 μL of pre-column loading buffer (0.05% trifluoroacetic acid (TFA) and 5% acetonitrile (ACN) in H2O), prior to LC-MS/MS analysis. Mass spectrometry was performed on the PAPPSO platform (MICALIS, INRA, France, http://pappso.inra.fr/). Protein identification was performed with X!Tandem software (DB: X!tandem version 2013.09.01.1) against a protein database of B. subtilis as well as a proteomic contaminant database (for details of the parameters used, see S2 Table). For quantification of the proteins, we used the number of spectra obtained during protein identification by mass spectrometry. The number of spectra is admitted to be proportional to the abundance of a given protein. For each protein, we calculated the relative abundance factor (PAI) as described in [66]. The PAI estimates the relative abundance of a protein and is calculated as the number of identified spectra divided by the number of theoretical peptides of the protein (theoretical peptide number corresponds to the number of peptides resulting from the theoretical digestion of the protein by trypsin and that are visible in mass spectrometry [i.e. having a mass ranging between 800 and 2,500 D.]). The PAI were log2-transformed after adding a pseudo count of 0.1 which corresponded approximately to quantile 10% of the PAI distributions.
Analysis of the promoters’ activity using translational fusions with luciferase was performed as described by [97] with minor modifications. Cells were grown in LB medium to mid-exponential phase (optical density OD600 0,4–0,5 with NovaspecII Visible Spectrophotometer, Pharmacia Biotech), after which cultures were centrifuged and resuspended in fresh DS or MSgg media to obtain OD600 1,0. The pre-cultures were next diluted in respective media to OD600 0.025. The starter cultures were distributed by 200μl in a 96-well black plate (Corning) and D-lucefirin (PerkinElmer) was added to each well to final concentration 1.5 mg/mL. The cultures were incubated with agitation at 37°C in PerkinElmer Envision 2104 Multilabel Reader (PerkinElmer) equipped with an enhanced sensitivity photomultiplier for luminometry (data presented in Figs 4, 6 and 10) or in Synergy 2 Multi-mode microplate reader (BioTek Instruments; data presented in Fig 7 and S7 Fig). Relative Luminescence Units (RLU) and OD600 were measured at 5 min intervals. Each fusion-containing strain was analyzed at least three times. Each experiment included four independent cultures of each strain.
Swarming and swimming motility tests were performed using NCIB 3610 strain and its derivatives as described by [68, 152] with some modifications. The fresh plates (9cm) were prepared from liquid LB medium (Difco) fortified by agar (Invitrogen, Life Technologies) at 0.3% or 0.7% concentration for swimming and swarming tests, respectively, and dried in a laminar flow hood for 15 minutes. B. subtilis cells were grown to an OD600 0.5 (Biochrom Libra S11 Visible Spectrophotometter, Biochrom), 2 ml of cells were pelleted and gently resuspended in 100 μl of phosphate-buffered saline (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4) containing 0.5% of India ink (Higgins). For motility assay, 5μl of cells were directly spotted on the plate and dried in a laminar flow hood for two minutes. Plates were incubated at 37°C and the extent of swimming or swarming was noted at defined time intervals. The images were acquired with the ChemiDoc MP system (BioRad) and treated using ImageLab 5.0 software (BioRad) after 5 or 20 hours of incubation for swimming or swarming motility tests, respectively. For each strain, from three to five independent cultures were analyzed in parallel during each experiment. At least four independent experiments were performed.
Overnight bacterial cultures were diluted 200-times in fresh LB medium and grown at 37°C with agitation to an OD600 ∼0.6. For the colony assay, 2μl of culture was spotted onto MSgg agar plate (1.5% agar, Invitrogen) and incubated at 30°C for 72h. For pellicle assay, 2μl of culture was added to 2ml of MSgg medium in a well of 24-well sterile microtiter plate (Evergreen Scientific). The plates were incubated without agitation at 30°C for 72h. Photographs were acquired with the Samsung Galaxy Tab E–SM-T560. For each strain, four independent cultures were analyzed in parallel during each experiment. At least five independent experiments were performed.
For sporulation assay, cells were diluted in LB in a way to obtain the exponentially growing cultures after over-night incubation at 28°C. The pre-cultures were diluted in pre-warmed liquid DS medium at OD600 0.025 and incubated at 37°C. The growth rates were the same for all strains. Starting from OD600 1.5 (taken as T0) cultures were analyzed for the presence of spores at the indicated time. Samples were split in two and one part was heated at 75°C for 15 min; heated and unheated cultures were plated in sequential dilutions at LB-agar plates and incubated for 36 h at 37°C. The percentage of spores was calculated as the ratio of colony forming units in heated and unheated samples. Each experiment included three independent isogenic cultures. Four independent experiments were performed to establish sporulation efficiency of each strain.
The crude cell extracts were prepared using Vibracell 72408 sonicator (Bioblock scientific). Bradford assay was used to determine total protein concentration in each extract. Equal amounts of total proteins were separated by SDS-PAGE (12% polyacrylamide). The SPA-tagged Rho, KinA and KinB proteins were visualized using the primary mouse ANTI-FLAG M2 monoclonal antibodies (Sigma-Aldrich; dilution 1:5,000) and the secondary goat peroxidase-coupled anti-mouse IgG antibodies (Sigma-Aldrich; dilution 1:20,000). MreB and Mbl proteins used as controls for samples equilibrium were visualized using primary rat anti-MreB and rabbit anti-Mbl antibodies (a gift of X. Henry, dilution 1:10,000) and the secondary peroxidase-coupled anti-rat and anti-rabbit antibodies A9037 and A0545, respectively (Sigma-Aldrich; dilution 1:10,000). Three independent experiments were performed, and a representative result is shown.
|
10.1371/journal.pcbi.1004219 | MAGMA: Generalized Gene-Set Analysis of GWAS Data | By aggregating data for complex traits in a biologically meaningful way, gene and gene-set analysis constitute a valuable addition to single-marker analysis. However, although various methods for gene and gene-set analysis currently exist, they generally suffer from a number of issues. Statistical power for most methods is strongly affected by linkage disequilibrium between markers, multi-marker associations are often hard to detect, and the reliance on permutation to compute p-values tends to make the analysis computationally very expensive. To address these issues we have developed MAGMA, a novel tool for gene and gene-set analysis. The gene analysis is based on a multiple regression model, to provide better statistical performance. The gene-set analysis is built as a separate layer around the gene analysis for additional flexibility. This gene-set analysis also uses a regression structure to allow generalization to analysis of continuous properties of genes and simultaneous analysis of multiple gene sets and other gene properties. Simulations and an analysis of Crohn’s Disease data are used to evaluate the performance of MAGMA and to compare it to a number of other gene and gene-set analysis tools. The results show that MAGMA has significantly more power than other tools for both the gene and the gene-set analysis, identifying more genes and gene sets associated with Crohn’s Disease while maintaining a correct type 1 error rate. Moreover, the MAGMA analysis of the Crohn’s Disease data was found to be considerably faster as well.
| Gene and gene-set analysis are statistical methods for analysing multiple genetic markers simultaneously to determine their joint effect. These methods can be used when the effects of individual markers is too weak to detect, which is a common problem when studying polygenic traits. Moreover, gene-set analysis can provide additional insight into functional and biological mechanisms underlying the genetic component of a trait. Although a number of methods for gene and gene-set analysis are available however, they generally suffer from various statistical issues and can be very time-consuming to run. We have therefore developed a new method called MAGMA to address these issues, and have compared it to a number of existing tools. Our results show that MAGMA detects more associated genes and gene-sets than other methods, and is also considerably faster. The way the method is set up also makes it highly flexible. This makes it suitable as a basis for more general statistical analyses aimed at investigating more complex research questions.
| In the past decade, genome-wide association studies (GWAS) have successfully identified new genetic variants for a wide variety of phenotypes [1]. However, despite growing sample sizes, the genetic variants discovered by GWAS generally account for only a fraction of the total heritability of a phenotype [2,3]. More than anything, GWAS has shown that many phenotypes, such as height [4], schizophrenia [5] and BMI [6] are highly polygenic and influenced by thousands of genetic variants with small individual effects, requiring very large sample sizes to detect them.
Gene and gene-set analysis have been suggested as potentially more powerful alternatives to the typical single-SNP analyses performed in GWAS [7]. In gene analysis, genetic marker data is aggregated to the level of whole genes, testing the joint association of all markers in the gene with the phenotype. Similarly, in gene-set analysis individual genes are aggregated to groups of genes sharing certain biological, functional or other characteristics. Such aggregation has the advantage of considerably reducing the number of tests that need to be performed, and makes it possible to detect effects consisting of multiple weaker associations that would otherwise be missed. Moreover, gene-set analysis can provide insight into the involvement of specific biological pathways or cellular functions in the genetic etiology of a phenotype. Gene-set analysis methods can be subdivided into self-contained and competitive analysis, with the self-contained type testing whether the gene set contains any association at all, and the competitive type testing whether the association in the gene set is greater than in other genes [7].
Various methods for gene and gene-set analysis are currently available [7–13]. However, one concern with most existing methods is that they first summarize associations per marker before aggregating them to genes or gene sets. As demonstrated by Moskvina et al. this makes the statistical power strongly dependent on local linkage disequilibrium (LD) [14], and also reduces power to detect associations dependent on multiple markers.
An additional concern is that current gene-set analysis methods generally use a permutation-based approach. These are often very computationally demanding, and since no parametric model is used it is often not made explicit which null hypothesis is being evaluated and what assumptions are made. This makes it more difficult to determine the properties of the analysis such as how the significance of a gene set relates to the significance of its constituent genes or whether the analysis corrects for a polygenic architecture. This complicates the interpretation of results and hampers comparison between results of different gene-set analysis methods.
To address such issues we have developed MAGMA (Multi-marker Analysis of GenoMic Annotation), a fast and flexible tool for gene and gene-set analysis of GWAS genotype data. MAGMA’s gene analysis uses a multiple regression approach to properly incorporate LD between markers and to detect multi-marker effects. The gene-set analysis is built as a distinct layer around this gene analysis, providing the flexibility to independently change and expand both the gene and the gene-set analysis. Both self-contained and competitive gene-set analyses are implemented using a gene-level regression model. This regression approach offers a generalized framework which can also analyse continuous gene properties such as gene expression levels as well as conditional analyses of gene sets and other gene properties, and which can be extended to allow joint and interaction analysis of multiple gene sets and other gene properties as well. More traditional gene analysis models are also implemented, for comparison and to provide analysis of SNP summary statistics.
To evaluate the performance of MAGMA we have applied it to the Wellcome Trust Case-Control Consortium (WTCCC) Crohn’s Disease (CD) GWAS data-set [15], using the MSigDB Canonical Pathways [16] for the gene-set analysis. Simulation studies were performed to verify type 1 error rates for MAGMA. The CD data set was then analysed using MAGMA and with five commonly used other tools for gene and gene-set analyses, specifically VEGAS [17], PLINK [8], ALIGATOR [9], INRICH [10] and MAGENTA [12]. The results show that MAGMA has greater statistical power than the other methods, while also being considerably faster.
The gene-set analysis is divided into two distinct and largely independent parts. In the first part a gene analysis is performed to quantify the degree of association each gene has with the phenotype. In addition the correlations between genes are estimated. These correlations reflect the LD between genes, and are needed in order to compensate for the dependencies between genes during the gene-set analysis. The gene p-values and gene correlation matrix are then used in the second part to perform the actual gene-set analysis.
The advantage of decoupling these two parts of the analysis in this manner is that each can be changed independently from the other, simplifying the development of changes and extensions to either part of the model. Moreover, since the second part only uses the output from the first part and not the raw genotype data they do not need to be performed at the same time or place, making it much more straightforward to perform multiple gene-set analyses on the same data or to analyse multiple data sets across a large-scale collaboration.
The gene analysis in MAGMA is based on a multiple linear principal components regression [18] model, using an F-test to compute the gene p-value. This model first projects the SNP matrix for a gene onto its principal components (PC), pruning away PCs with very small eigenvalues, and then uses those PCs as predictors for the phenotype in the linear regression model. This improves power by removing redundant parameters, and guarantees that the model is identifiable in the presence of highly collinear SNPs. By default only 0.1% of the variance in the SNP data matrix is pruned away.
With
Xg*
the matrix of PCs, Y the phenotype and W an optional matrix of covariates the model can thus be written as
Y=α0g1→+Xg*αg+Wβg+εg
, where the parameter vector αg represents the genetic effect, βg the effect of the optional covariates, α0g the intercept and εg the vector of residuals. The F-test uses the null-hypothesis H0:
αg=0→
of no effect of gene g on the phenotype Y, conditional on all covariates.
This choice of gene analysis model is motivated by a balance of statistical and practical concerns. This multiple regression model ensures that LD between SNPs is fully accounted for. It also offers the flexibility to accommodate additional covariates and interaction terms as needed without changing the model. At the same time, since the F-test has a known asymptotic sampling distribution the gene p-values take very little time to compute, making the gene analysis much faster than permutation-based alternatives.
The linear regression model is also applied when Y is a binary phenotype. Although this violates some assumptions of the F-test, comparison of the F-test p-values with p-values based on permutation of the F-statistic shows that the F-test remains accurate (see ‘Supplemental Methods—Implementation Details’). MAGMA therefore uses the asymptotic F-test p-values by default, though it also offers an option to compute permutation-based p-values using an adaptive permutation procedure. In addition, comparison with logistic regression models shows that the results of the linear model are effectively equivalent to that of the more conventional logistic regression model, but without the computational cost.
To perform the gene-set analysis, for each gene g the gene p-value pg computed with the gene analysis is converted to a Z-value zg = Φ−1(1 – pg), where Φ−1 is the probit function. This yields a roughly normally distributed variable Z with elements zg that reflects the strength of the association each gene has with the phenotype, with higher values corresponding to stronger associations.
Self-contained gene-set analysis tests whether the genes in a gene-set are jointly associated with the phenotype of interest. As such, using this variable Z a very simple intercept-only linear regression model can now be formulated for each gene set s of the form
Zs=β01→+εs
, where Zs is the subvector of Z corresponding to the genes in s. Evaluating β0 = 0 against the alternative β0 > 0 yields a self-contained test, since under the self-contained null hypothesis that none of the genes is associated with the phenotype zg has a standard normal distribution for every gene g.
Competitive gene-set analysis tests whether the genes in a gene-set are more strongly associated with the phenotype of interest than other genes. To test this within the regression framework the model is first expanded to include all genes in the data. A binary indicator variable Ss with elements sg is then defined, with sg = 1 for each gene g in gene set s and 0 otherwise. Adding Ss as a predictor of Z yields the model
Z=β0s1→+Ssβs+ε
. The parameter βs in this model reflects the difference in association between genes in the gene set and genes outside the gene set, and consequently testing the null hypothesis βs = 0 against the one-sided alternative βs > 0 provides a competitive test. Note that this is equivalent to a one-sided two-sample t-test comparing the mean association of gene-set genes with the mean association of genes not in the gene-set. Similarly, the self-contained analysis is equivalent to a one-sided single-sample t-test comparing the mean association of gene-set genes to 0.
It should be clear that in this framework, the gene-set analysis models are a specific instance of a more general gene-level regression model of the form
Z=β01→+C1β1+C2β2+…+ε
. The variables C1, C2, …, in this generalized gene-set analysis model can reflect any gene property, from the binary indicators used for the competitive gene-set analysis to continuous variables such as gene size and expression levels. Any transformations of, and interactions between, such gene properties can also be added. This generalized gene-set analysis model thus allows for testing of conditional, joint and interaction effects of any combination of gene sets and other gene properties. In practice, the competitive gene-set analysis implemented in MAGMA in fact uses such a generalized model by default, performing a conditional test of βs corrected for the potentially confounding effects of gene size, gene density and (if applicable, e.g. in meta-analysis) difference in underlying sample size, if such effects are present. This is achieved by adding these variables, as well as the log of these variables, as covariates to the gene-level regression model. The gene density is defined as the ratio of effective gene size to the total number of SNPs in the gene, with the effective gene size in turn defined as the number of principal components that remain after pruning.
One complication that arises in this gene-level regression framework is that the standard linear regression model assumes that the error terms have independent normal distributions, i.e.
ε~MVN(0→,σ2I)
. However, due to LD, neighbouring genes will generally be correlated, violating this assumption. This issue can be addressed by using Generalized Least Squares approach instead, and assuming that
ε~MVN(0→,σ2R)
. In MAGMA, the required gene-gene correlation matrix R is approximated by using the correlations between the model sum of squares (SSM) of each pair of genes from the gene analysis multiple regression model, under their joint null hypothesis of no association. These correlations are a function of the correlations between the SNPs in each pair of genes and thus provide a good reflection of the LD, and since they have a convenient closed-form solution they are easy to compute (see also ‘Supplemental Methods—Implementation Details’). Note that for the self-contained analysis, the submatrix Rs corresponding to only the genes in the gene set is used instead of R. In addition, since the self-contained null hypothesis guarantees that all zg have a standard normal distribution, the error variance σ2 can be set to 1.
Since raw genotype data may not always be available for analysis, MAGMA also provides more traditional SNP-wise gene analysis models of the type implemented in PLINK and VEGAS. These SNP-wise models first analyse the individual SNPs in a gene and combine the resulting SNP p-values into a gene test-statistic, and can thus be used even when only the SNP p-values are available. Although evaluation of the gene test-statistic does require an estimate of the LD between SNPs in the gene, estimates based on reference data with similar ancestry as the data the SNP p-values were computed from has been shown to yield accurate results [17,19].
Two types of gene test statistics have been implemented in MAGMA: the mean of the χ2 statistic for the SNPs in a gene, and the top χ2 statistic among the SNPs in a gene. For the mean χ2 statistic, a gene p-value is then obtained by using a known approximation of the sampling distribution [20,21]. For the top χ2 statistic such an approximation is not available, and therefore an adaptive permutation procedure is used to obtain an empirical gene p-value. A random phenotype is first generated for the reference data, drawing from the standard normal distribution. This is then permuted, and for each permutation the top χ2 statistic is computed for every gene. The empirical p-value for a gene is then computed as the proportion of permuted top χ2 statistics for that gene that are higher than its observed top χ2 statistic. The required number of permutations is determined adaptively for each gene during the analysis, to increase computational efficiency. Further details can be found in ‘Supplemental Methods—SNP-wise gene analysis’.
The MAGMA SNP-wise models can also be used to analyse raw genotype data, in which case the raw genotype data takes the place of the reference data and the SNP p-values are computed internally. Gene-set analysis based on these SNP-wise models proceeds in the same way as the gene-set analysis based on the multiple regression gene analysis model. The gene p-values resulting from the analysis are converted to Z-values in the same way to serve as input for the gene-set analysis. Similarly, the gene-gene correlation matrix R is obtained using the same formula as with the multiple regression model, but using the reference data to compute it.
A number of additional features has been implemented in MAGMA, more fully described in ‘Supplemental Methods—Extensions’. Gene analysis can be expanded with a gene-environment interaction component, which can subsequently be carried over to the gene-set analysis. Options for aggregation of rare variants and for fixed-effects meta-analysis for both gene and gene-set analysis are also available. Efficient SNP to gene annotation and a batch mode for parallel processing are provided to simplify the overall analysis process. MAGMA is distributed as a standalone application using a command-line interface. The C++ source code is also made available, under an open source license. MAGMA can be downloaded from http://ctglab.nl/software/magma.
To evaluate the performance of MAGMA, the WTCCC Crohn’s Disease (CD) GWAS data [15] in conjunction with both WTCCC control samples was used. The data was cleaned according to the protocol described by Anderson [22], resulting in a sample of 1,694 cases and 2,917 controls with data for 403,227 SNPs. The European samples from the 1,000 Genomes data [23] and the HapMap 3 data [24] were used as reference data sets for the summary statistics gene analysis.
SNPs were annotated to genes based on dbSNP version 135 SNP locations and NCBI 37.3 gene definitions. For the main analyses only SNPs located between a gene’s transcription start and stop sites were annotated to that gene, yielding 13,172 protein-coding genes containing at least one SNP in the CD data. An additional annotation using a 10 kilobase window around each gene was made, yielding 16,970 genes, to determine the effect of using a window on relative performance. These two gene annotations were used for all analyses, to ensure that differences in default annotation settings did not cloud the comparison between tools. The 1,320 Canonical Pathways from the MSigDB database [16] were used for the gene-set analysis. The relatively large number of gene sets and the fact that the MSigDB Canonical Pathways are drawn from a number of different gene-set databases ensures a wide variety of gene sets, which should prevent the results from being too dependent on the choice of gene-set database.
The MAGMA gene analysis was performed on the raw CD data using the PC regression model (MAGMA-main). Gene analyses with VEGAS and PLINK were performed using the mean SNP statistic for VEGAS and both the mean SNP statistic (PLINK-avg) and the top SNP statistic (PLINK-top) for PLINK. Pruning in PLINK was turned off for these analyses. An additional PLINK analysis using the mean SNP statistic with pruning set to its default (PLINK-prune) was performed as well.
To facilitate the comparison, several additional SNP-wise gene-set analyses were performed in MAGMA with test-statistics matching those of PLINK-avg, PLINK-top and VEGAS: mean χ2 (MAGMA-mean) and top χ2 (MAGMA-top) on the raw CD data to match the two PLINK analyses, and mean χ2 using CD SNP p-values and with either HapMap reference data (MAGMA-pval) to match VEGAS or with 1,000 Genomes reference data (MAGMA-pval-1K). The SNP summary statistics used for VEGAS and MAGMA-pval were computed using PLINK ‘--assoc’.
Gene-set analysis for MAGMA was performed based on the PC regression gene analysis model (MAGMA-main) as well as on the SNP-wise model with 1,000 Genomes reference data (MAGMA-pval-1K). Several other analyses were performed for comparison: PLINK self-contained gene-set analysis without pruning (PLINK-avg) and with pruning (PLINK-prune), as well as ALIGATOR, INRICH and MAGENTA competitive gene-set analysis. PLINK operates on raw genotype data, whereas all three competitive methods require only SNP p-values as input. No correction for stratification was used in any of the analyses except when explicitly specified. An overview of all analyses is given in Table 1.
Simulation was used to assess the type 1 error rates, using permutations of the CD phenotype to obtain a global null distribution of no associated SNPs (see ‘Supplemental Methods—Simulation Studies’ for details). For the gene analysis, type 1 error rates were found to be controlled at the nominal level of 0.050 for the PC regression model, the summary statistics analysis model, as well as the SNP-wise models (Table S1 in S2 File).
The type 1 error rates for the gene-set analysis were also found to be well controlled for both the self-contained and competitive test (Table S2 in S2 File). For the competitive test an additional simulation using a polygenic null model was performed, with effects explaining a combined 50% of the phenotypic variance assigned to randomly selected SNPs. This polygenic type 1 error rate was also well controlled. The type 1 error rates for the self-contained analysis under the polygenic null model are also shown. These are considerably inflated because self-contained gene-set analysis by its definition is not designed to correct for polygenicity, illustrating the risk of performing self-contained analysis on polygenic phenotypes.
The results of the gene analyses of the CD data are summarized in Table 2, which shows the number of significant genes at a number of different p-value thresholds. Since the Type 1 error rates have been shown to be properly controlled these results can serve as a good indicator of the relative power of the different methods, and compared to simulation-based power estimates this has the advantage that no assumptions about the genetic causal model. From Table 2 it is clear that whereas the power of all the other methods is very similar, the MAGMA-main model shows a clear advantage over the rest. After Bonferroni correction, MAGMA-main found a total of 10 genome-wide significant genes, including the well-known CD genes NOD2, ATG16L1 and IL23R [25,26]. This also indicates that although MAGMA can perform analysis of summary statistics, raw data analysis should always be preferred if possible.
Specific implementation issues can be ruled out as the cause of the power difference since the PLINK and VEGAS analyses yield results highly similar to their matched MAGMA models (S9 Fig), and using the pruning option in PLINK also has little effect on the overall results. This means that the difference must be due to the difference in the methods and test-statistics themselves. Comparing the MAGMA implementations of these models in Fig 1, the mean χ2 and top χ2 approaches are shown to produce very similar p-values. Moreover, the plots reveal that the superior power of the MAGMA-main model does not arise from consistently lower gene p-values, but rather from a small set of genes with low p-values for MAGMA-main that are simply not picked up by the other approaches. This is likely to be related to the way LD between SNPs is handled, as that is one of the key differences between the multiple regression model of MAGMA-main and all the others. A post-hoc power simulation indeed indicates that multi-marker effects with weak marginals are the most probable explanation (see ‘Supplemental Methods—Simulation Studies’).
To increase the generalizability of these findings, two variations on the CD analyses were performed for MAGMA-main, MAGMA-mean and MAGMA-top. First, the analyses were repeated with 10 principal components computed from the whole data set as covariates to correct for possible stratification. The results are shown in Table 2 and S10 Fig. There is shown to be only very limited stratification, and although the power does decrease somewhat MAGMA-main’s power advantage is maintained. The analyses were also repeated with the gene annotation extended to include a 10 kilobase window around each gene, with the comparison in S11 Fig showing a considerable impact on the results. However, although this suggests that the choice of window can strongly affect the results of a gene analysis Table 2 shows that the relative power stays the same, with MAGMA-main again maintaining its superior power.
As with the gene analysis, the results of the CD analysis (Table 3 and Fig 2) can again serve as a gauge of the relative power of the different gene-set analysis methods. For the self-contained gene-set analysis this comparison is straightforward with MAGMA showing considerably more power than the two PLINK analyses. For the most part MAGMA’s power advantage can be explained by the difference in the underlying gene model, given the superior power of the PC regression model over the SNP-wise model used by PLINK shown before. Differences in how the genes are combined may also play a role however since, in contrast to PLINK, MAGMA weighs genes equally rather than by the number of SNPs in them and explicitly takes correlations between genes into account. Of note is also that PLINK-prune does considerably better than PLINK-avg, and that its p-values are somewhat more strongly correlated with those of the MAGMA analysis (Fig 2). An additional summary statistics analysis (MAGMA-pval-1K) on SNP p-values and using 1,000 Genomes reference data was also performed. This showed less power than PLINK even though it uses the same model at the gene level, suggesting that the difference is due to how the genes are aggregated to gene-sets. One of the key differences in this regard is that PLINK gives larger genes greater weight whereas MAGMA weighs them equally. As such a likely explanation is that the PLINK results are partially driven by a smaller number of large genes, though constructing the intermediate models to verify this is beyond the scope of this paper.
The comparison of competitive methods is somewhat more complicated, due to the fact that ALIGATOR, INRICH and MAGENTA all use discretization using a p-value cut-off. This cut-off needs to be specified by the user and has no obvious default value, although for MAGENTA the 5th percentile cut-off is suggested as the most optimal [12]. For ALIGATOR and INRICH the analysis was therefore performed at four different cut-offs (0.0001, 0.001, 0.005, 0.01), and for MAGENTA at two (5th and 1st percentile).
Of the four tools, only MAGMA and INRICH yield significant results after multiple testing correction (Tables 3 and 4). As with the self-contained gene-set analysis, power for the MAGMA analysis is better when using raw data rather than SNP p-values as input, though both yield one significant gene set. For INRICH the results are strongly dependent on the SNP p-value cut-off used, with three significant gene sets at the 0.0001 cut-off but none at the higher ones, further emphasizing the problem of choosing the correct cut-off. It should also be noted that the p-values have not been corrected for the fact that the gene-sets have been analysed under four different thresholds, and thus might not fall below the significance threshold if they were.
Looking at the results in more detail (Fig 3) also suggests that the differences in results are not merely due to a difference in power. The concordance between methods is poor, with only MAGENTA and ALIGATOR showing a reasonable correlation in results. Moreover, there is considerable discordance between different p-values cut-offs for the same methods as well (Fig 4). This suggests that the different methods, or methods at different p-value cut-offs, are sensitive to distinctly different kinds of gene set associations. In particular, MAGMA and the other three methods at higher p-value cut-offs would be expected to respond best to gene-sets containing a larger number of somewhat associated genes. Conversely, at lower p-value cut-offs the latter three should become more sensitive to gene-sets containing a small number of more strongly associated genes. This is exemplified by the INRICH analysis. At the 0.0001 cut-off only quite strongly associated genes are counted as relevant, but as there are only 42 such genes overall the three gene sets (containing either 26 or 29 genes) become significant despite each containing only three relevant genes.
Aside from differences between methods, Table 3 also shows a clear difference between self-contained and competitive gene-set analysis. This is not a difference in power, but rather a difference of null hypothesis. Competitive tests attempt to correct for the baseline level of association present in the data and accordingly have a much more general null hypothesis. The impact of this difference in hypothesis can be illustrated by comparing the MAGMA self-contained and competitive analyses, since they are performed in the same framework. Whereas the self-contained analysis detects 39 gene sets that show association with the phenotype, the competitive analysis detects only one of those 39. For the remaining 38 gene sets, there is no evidence in the data that the associations in those gene sets are any stronger than would be expected by chance given the polygenic nature of CD. The gene-set that remains is the Regulation of AMPK via LKB1 (REACTOME) set. For two additional gene sets, Cell Adhesion Molecules (KEGG) and ECM-receptor Interaction (KEGG), the competitive p-value also drops below the significance threshold (Table 4 and S12 Fig) if the correction for gene size and gene density is turned off. This suggests that these gene sets do in fact contain significantly elevated levels of association, but that this is partially caused by confounding effects of the size and density of the genes they contain. Given the strength of the confounding effect it is evident that gene-set analyses should always be corrected for these and other potential confounders, to avoid false positive results. Full results for the analyses can be found in Table S5 in S2 File.
All analyses were performed on the Genetic Cluster Computer, which is part of the Dutch Lisa Cluster. In terms of computational performance MAGMA is shown to have a considerable advantage over the other methods (Table 5) for both gene and gene-set analysis. The most marked difference is between MAGMA and PLINK, the only one of the alternative methods using raw data input. However, the raw data analysis in MAGMA outperforms the summary statistics methods as well. Although INRICH and ALIGATOR show comparable computation times at their lowest SNP p-value cut-off, the need to repeat the analysis at multiple cut-offs means the total analysis for both takes considerably longer.
The low MAGMA computation times are largely due to the choice of statistical model. Since the statistical tests used have known asymptotic sampling distributions the need for computationally demanding permutation or simulation schemes is avoided. Note however that the permutation-based SNP-wise analyses in MAGMA also show very reasonable computation times. These results demonstrate that, given efficient implementation, there is no computational reason to prefer analysis of summary statistics over raw data analysis, even when using permutation.
We have developed MAGMA, a fast and flexible method for performing gene and gene-set analysis in a two-tiered parametric framework. Comparison with a number of other, frequently used methods shows that MAGMA has better power for gene analysis as well as for both self-contained and competitive gene-set analysis. An important factor in this is the multiple regression model used in the gene analysis, which is better able to incorporate the LD between SNPs than other methods. Because of its two-layer structure, this improvement in power at the gene-level subsequently carries over to the gene-set analysis.
MAGMA was also found to be generally much faster than other methods, even methods that used only summary statistics rather than raw data. This is primarily due to the choice of statistical model, which did not require the kind of computationally expensive permutation or sampling procedures used in the other methods. However, even the permutation-based SNP-wise models implemented in MAGMA outperformed their equivalents in other software and yielded very reasonable computation times.
Although MAGMA showed better power than other tools for both the self-contained and competitive gene-set analysis, these comparisons also revealed considerable differences between the methods. This was most pronounced for the competitive gene-set analysis, with even results for individual methods showing significant variability based on the choice of cut-off. At present no comprehensive evaluation of the differences between existing gene-set analysis methods exists, leaving the causes and implications of these difference unclear. It is beyond the scope of this paper to perform such an evaluation, but the degree of discordance between most methods strongly suggests a need for future research in this direction. An additional caveat is that it is unknown to what extent the observed differences in power between methods may depend on the specific genetic architecture of Crohn’s diseases, and as such generalizing the results to other genetic architectures must be done with caution.
The framework for MAGMA is built with future extensions in mind. Because of the two-tiered structure of the gene-set analysis, alternative gene analysis models are straightforward to implement and are automatically available for use in the gene-set analysis. Similarly, the linear regression structure used to implement the gene-set analysis offers a high degree of extensibility. At present it enables analysis of continuous gene-level covariates as well as conditional analysis of gene-sets correcting for possible confounders, and the analysis of the CD data demonstrates that correction for confounders such as gene size and gene density is indeed strongly advised. The model is easily generalized to much more general gene-level linear regression models to allow for simultaneous analysis of multiple covariates and gene-sets, opening up a wide range of new testable hypotheses.
|
10.1371/journal.pcbi.1004762 | Estimation of Directed Effective Connectivity from fMRI Functional Connectivity Hints at Asymmetries of Cortical Connectome | The brain exhibits complex spatio-temporal patterns of activity. This phenomenon is governed by an interplay between the internal neural dynamics of cortical areas and their connectivity. Uncovering this complex relationship has raised much interest, both for theory and the interpretation of experimental data (e.g., fMRI recordings) using dynamical models. Here we focus on the so-called inverse problem: the inference of network parameters in a cortical model to reproduce empirically observed activity. Although it has received a lot of interest, recovering directed connectivity for large networks has been rather unsuccessful so far. The present study specifically addresses this point for a noise-diffusion network model. We develop a Lyapunov optimization that iteratively tunes the network connectivity in order to reproduce second-order moments of the node activity, or functional connectivity. We show theoretically and numerically that the use of covariances with both zero and non-zero time shifts is the key to infer directed connectivity. The first main theoretical finding is that an accurate estimation of the underlying network connectivity requires that the time shift for covariances is matched with the time constant of the dynamical system. In addition to the network connectivity, we also adjust the intrinsic noise received by each network node. The framework is applied to experimental fMRI data recorded for subjects at rest. Diffusion-weighted MRI data provide an estimate of anatomical connections, which is incorporated to constrain the cortical model. The empirical covariance structure is reproduced faithfully, especially its temporal component (i.e., time-shifted covariances) in addition to the spatial component that is usually the focus of studies. We find that the cortical interactions, referred to as effective connectivity, in the tuned model are not reciprocal. In particular, hubs are either receptors or feeders: they do not exhibit both strong incoming and outgoing connections. Our results sets a quantitative ground to explore the propagation of activity in the cortex.
| The study of interactions between different cortical regions at rest or during a task has considerably developed in the past decades thanks to progress in non-invasive imaging techniques, such as fMRI, EEG and MEG. These techniques have revealed that distant cortical areas exhibit specific correlated activity during the resting state, also called functional connectivity (FC). Moreover, recent studies have highlighted the possible role of white-matter projections between cortical regions in shaping these activity patterns. This structural connectivity (SC) can be estimated using MRI, which measures the probability for two areas to be connected via the density of neural fibers. However, this does not provide the strengths of dynamical interactions. Many methods have thus been developed to estimate the connectivity between neural populations in the cortex that is hypothesized to shape FC. The strengths of these dynamical interactions are called effective connectivity (EC). We use a cortical model that combines information from Diffusion-weighted MRI (dwMRI) and fMRI in order to estimate EC. We demonstrate theoretically that directed C can be inferred using time-shifted covariances. The key point of our method is the use of temporal information from FC at the scale of the whole network. Applying our model on experimental fMRI data at rest, we estimate the asymmetry of intracortical connectivity. Obtaining an accurate EC estimate is essential to analyze its graph properties, such as hubs. In particular, directed connectivity links to the asymmetry between input and output EC strengths of each node, which characterizes feeder and receiver hubs in the cortical network.
| Patterns of neural activity at the scale of the whole cortex can be quantified by the correlations between the cortical regions. This so-called functional connectivity (FC) is predicted to some extent by the anatomical synaptic pathways in the white matter, or structural connectivity (SC) [1]. However, SC measures the neural fiber density and is not sufficient to fully explain the structure of FC, which also depends on the dynamics of network nodes. The dynamical interactions between cortical areas is captured by the ‘effective connectivity’ (EC), a concept that has emerged over years following progress in imaging techniques and related modeling [2–5]. In network models, EC is the key to understand the propagation of information, which links the network structure to function. Importantly, EC is model-dependent, which makes the comparison between studies non trivial and has raised much debate recently [6–8]. On the biological ground, EC accounts for mechanisms that determine the synaptic strength (e.g., types and concentration of neurotransmitters), as well as dynamic properties such as neural excitability that may vary with the local activity level and thus depend on the inputs to the network. This means that EC may differ from the anatomical SC obtained from dwMRI. FC and EC classically relate to data and models related to imaging techniques: fMRI, EEG and MEG. Beyond the analysis of these experimental data, uncovering the relationship between connectivity and activity has attracted much interest recently, with a particular focus for non-trivial connectivity topologies that mimic those observed in the biology [9–12]. Within this field, the inference of the network connectivity from empirically observed activity is an active line of research [1, 13–16] and is the purpose of the present study.
In order to infer the connectivity from empirical observations, one needs to define observables of the activity that the network model has to reproduce. Designing adequate and efficient estimation procedures is as important as the properties of the model itself. Over the past years, many techniques have been developed for various dynamical models: for oscillator networks based on synchrony [17] and covariances [18, 19]; structural equation models (SEMs) [20, 21]; statistical methods for generalized linear models (GLMs) [13, 22]; Granger causality in spiking networks with nonlinear dynamics [14, 23]; transfer entropy for an abstract model of neural activity observed via calcium imaging [24]; spike-time covariances in networks of Poisson neurons [25]; covariances for multivariate Ornstein-Uhlenbeck processes [16] with recent developments to estimate directed connectivity [26]; for the dynamic causal model (DCM) [5, 15]. As a general rule, three factors limit the accuracy of connectivity estimation: the size of the network, the amount of empirically observed activity and nonlinearities in the mapping between the connectivity and the activity observables. Moreover, the directionality of interactions between nodes is harder to estimate than the existence of an interaction between two nodes.
Here we do not focus on the detection of connections, which often involves statistical tests, but we want to estimate the ranking of individual weights in networks of about a hundred of nodes. In Methods, we develop the theory to recover from observed covariances (taken as FC) the network connectivity as well as the intrinsic variability experienced by individual nodes in a noise-diffusion (ND) network. More precisely, we perform a Lyapunov optimization (LO) during which the network parameters are tuned iteratively such that the network reproduces two covariance matrices set as ‘objectives’, or goals. This allows for the use of constraints on the network parameters, such as enforcing partial connectivity. The model is a multivariate Ornstein-Uhlenbeck process, where the activity fluctuations are caused by the intrinsic noise and shaped by the recurrent connections to generate the spatio-temporal covariance pattern. Compared to previous studies [18], we focus on the situation where the network effect is strong, meaning that FC significantly differs from EC: even unconnected or weakly connected nodes can be strongly correlated. This situation was considered for symmetric EC [27, 28]. As reviewed in [15] for the DCM model, many methods only consider connections individually for the observables; in contrast, partial correlations and multivariate autoregressive models (MVAR) consider the entire network activity and give the best performance for detecting undirected connections in that study. Our method also relies on the network covariances as a whole to estimate the connectivity. The performance of the estimation procedure for artificial networks is the subject of the first half of Results.
Following previous studies [1, 3, 5, 27, 29], we apply our theory to the estimation of cortico-cortical connectivity from fMRI data. Obtaining quantitative estimates for the connectivity is important to gain insight on the information flow within the cortex [30]. In our ND model, EC coincides with the recurrent connectivity. In many models applied to fMRI data such as the DCM [5] and dynamic mean-field (DMF) model [1], the hemodynamic response function (HRF) transforms the neural activity into the BOLD signal, introducing nonlinearity in the mapping EC-FC. Our ND model does not explicitly incorporate hemodynamics, so the estimated EC describes the interactions between cortical regions via the proxy of the BOLD signals. Therefore, we test our method on the more elaborate DMF model that incorporates a HRF [27]. Specifically, we verify that the spatio-temporal covariances in the DMF model with time shifts of seconds convey information about the underlying neural connectivity, which is required for our method to be successfully applied to fMRI data. Then, we check the agreement of the EC estimated using the ND model coincides with the neural connectivity of the DMF model. Finally we examine the consistency of our results when varying optimization parameters, in particular the time shift for the objective covariance matrices. In several previous models of the whole cortex [12, 27, 28, 31], EC corresponds to a scaled version of SC matrix obtained from diffusion tensor imaging, which is quasi symmetric. The corresponding fitting procedures often rely on zero-time-shift correlations, which implies that the directionality of cortical connections can hardly be estimated. In contrast, another direction of research has investigated directed interactions, but only for a small number of cortical areas [2–5, 7, 20, 32]. By recovering the strengths of individual connections and intrinsic variability for the whole cortex divided in 68 areas, we aim to combine the advantages of those studies.
We present the theory used to go back and forth between the network parameters and the observables:
The variables of importance are described in Table 1.
We consider a network of interconnected neural populations, as schematized in Fig 1A. The matrix C in Fig 1B represents the connection strengths for such a network of size 50. In our model, each node receives noise that propagates due to the recurrent connectivity. The activity in population i is described by the variable x i t, where t denotes the time. An example for the activity of two nodes is displayed in the left graph of Fig 1C. We analyze the neural dynamics up to the second-order fluctuations, assuming stationarity of the whole stochastic process. The means x ¯ i and covariances Q i j τ of variables x i t are defined as:
x¯:=〈xit〉,Qijτ:=〈(xit−x¯i)(xjt+τ−x¯j)〉,
(1)
where the angular brackets correspond to the average over the randomness due to the noise. The empirical covariances are evaluated from the discrete time series x i n , t with the corresponding time shifts from the activity of nsim simulated sessions with T samples each:
Q ^ i j τ = 1 n sim T ∑ 1 ≤ n ≤ n sim 0 ≤ t < T ( x i n , t - x ^ i n ) ( x j n , t + τ - x ^ j n ) , (2)
where the means x ^ i n = ∑ 0 ≤ t < T x i n , t / T are evaluated for each session. The stationarity hypothesis implies that averaging Q ^ i j τ over for a sufficiently long period gives the probabilistic mean Q i j τ. For the time series in Fig 1C, we obtain the Q ^ 0 and Q ^ τ matrices in Fig 1D. By definition, the matrix Qτ is symmetric for τ = 0, but its asymmetry increases with τ, as shown by lighter plotted dots and fitting curves in Fig 1F.
The node activity is governed by the following coupled ordinary differential stochastic equations:
d x i t = - x i t τ x + ∑ k ≠ i C i k x k t + e i d t + d B i t . (3)
Each x i t experiences an exponential decay with time constant τx and is excited by other x k t whose contributions are scaled by the recurrent weights Cij. Fluctuations are generated by the Gaussian white noise d B i t that has variance (σi)2; formally, B i t is a Wiener process scaled by a factor σi. We forbid self-connections, i.e., Cii = 0, although this is not crucial. In practice and without loss of generality, the input ei is homogeneous for all nodes.
Following previous work [16, 27, 33], we derive the well-known self-consistency equations for the mean x ¯ and covariances Qτ in Eq (1), assuming stationarity of the process. When the system has a stable fixed point, it is given by the zeros of the linear matrix system:
J x ¯ + e = 0 , (4)
where xt indicates the activity vector whose entries are x i t for 1 ≤ i ≤ N; e is the homogeneous vector of inputs ei. We have denoted the Jacobian of Eq (3) by
J i j = - δ i j τ x + C i j , (5)
where the Kronecker delta δij = 1 if i = j and 0 otherwise. The system has a single fixed point provided J is invertible.
Using Ito’s formula for the derivation of ( x i t - x ¯ i ) ( x j t + τ - x ¯ j ) with respect to t, we obtain the Lyapunov equation for the steady-state of the second-order fluctuations with zero time shift:
J Q 0 + Q 0 J † + Σ = 0 . (6)
The noise matrix Σ is diagonal with terms ( σ i ) 2 = δ i j 〈 d B i t d B j t 〉 and † is the matrix transpose operator.
As done for the variable t, the derivation of ( x i t - x ¯ i ) ( x j t + τ - x ¯ j ) with respect to τ yields an equation satisfied by the time-shifted covariance Qτ:
Q τ = Q 0 expm J † τ , (7)
where expm is the matrix exponential. When J is dominated by the diagonal elements, the autocovariances Q i i τ are close to an exponential decay with time constant τx, as illustrated in Fig 1E with a log y-axis by the straight black line that coincices with the mean in blue up to τ = 2.5 s.
For a model with given connectivity matrix C and intrinsic noise matrix Σ, the theoretical zero-time-shift covariance matrix Q0 is the unique solution of Eq (6) and can be evaluated using the long-known Bartels-Stewart algorithm [16]. Then, the model Qτ can be calculated from Q0 and J using Eq (7) for any given τ.
The inverse problem consists in finding an estimate of the connectivity matrix C such that the model reproduces the two objective matrices Q ^ 0 and Q ^ τ for a given τ > 0, e.g., given by empirical observations. The relation in Eq (7) allows for the calculation of J from Q ^ 0 and an occurrence of Q ^ τ for a given τ > 0:
J = 1 τ logm Q ^ 0 - 1 Q ^ τ † , (8)
where logm is the matrix logarithm [34]. In our case, the non-diagonal elements of the Jacobian directly give the connectivity weight Cij = Jij. Note that the diagonal terms of J can provide an estimation of τx, so far as we assume Cii = 0. Then, the noise matrix can be evaluated from the Lyapunov equation Eq (6):
Σ = - J Q ^ 0 - Q ^ 0 J † . (9)
Due to the matrix logarithm, this method is very sensitive to the noise in Q ^ 0 and Q ^ τ: it can give non-real numbers for J, hence C. For this reason, we develop the optimization method in the next section.
As will be shown later in Results, the direct method does not work well with noisy empirical Q ^ 0 / τ. We thus propose an alternative estimation method, where C is tuned iteratively such that the model reproduces the covariance observables, as illustrated in Fig 1H. Considering the noise matrix Σ to be known for now, we optimize C in order to reduce the model error
V ( C ) = ∑ m , n ( Q m n 0 - Q ^ m n 0 ) 2 + ∑ m , n ( Q m n τ - Q ^ m n τ ) 2 , (13)
for a pair of objective covariance matrices Q ^ 0 and Q ^ τ with a given τ > 0; in Results, it is referred to as τest. Being the sum of two matrix distances, the Lyapunov function V is positive definite and becomes zero only when the two covariance matrices are equal to the objective counterparts. Eq (8) ensures the unicity of the connectivity C for a given pair of covariances matrices, when the solution exists.
Starting from zero weights initially, each optimization step aims to reduce V. To do so, we calculate the Jacobian J in Eq (5) for the current connectivity C, then the model matrices Q0 and Qτ using Eqs (6) and (7) as we know Σ. We want to update C to obtain the following desired changes for the model covariances Q0 and Qτ:
ΔQmn0=ϵC(Q^mn0−Qmn0) ,ΔQmnτ=ϵC(Q^mnτ−Qmnτ) ,
(14)
where the optimization rate ϵC is a some small positive number. To evaluate the Jacobian update ΔJ that corresponds to ΔQ0 and ΔQτ, we consider the equivalent of Eq (8) for the theoretical covariance matrices, namely J = logm(X)†/τ with X = (Q0)−1 Qτ. We perform an implicit differentiation with respect to J and X, yielding
ΔJ=1τ(ΔX X−1)†=1τ { [ (Q0)−1ΔQ0(Q0)−1Qτ+(Q0)−1ΔQτ ] [ (Q0)−1Qτ ]−1 }†=1τ [ (Q0)−1 (ΔQ0+ΔQτexpm(−J†τ)) ]† .
(15)
Finally, the desired connectivity update is simply for all i ≠ j Δ C i j = Δ J i j , (16)
and zero for the diagonal elements.
To obtain the rhs of the first line in Eq (15), we have assumed that matrices X and ΔX commute. For the LO procedure, this implies the existence of a path of matrix increments corresponding to ΔJ such that the commutation requirement is satisfied at each step, in order to reach the global optimum. Assuming X to be arbitrary, the subspace of commuting matrices in which ΔX is constrained is a linear subspace of dimension n(n−1)/2 with n the number of nodes, to be compared to the dimension n2 of the space of X. Although this subspace is not dense, it is sufficiently rich to hope that suitable paths exist and the optimum can be approached; this will be verified using numerical simulation.
This optimization process may also not reach the global optimum when Q ^ τ and Q ^ 0 do not correspond to a real solution for C, such as noisy empirical covariances. We will thus verify the estimation performance numerically, using the known theoretical covariance matrices obtained from Eqs (6) and (7) to check for a given connectivity matrix C.
So far, we have considered Σ to be known and fixed during the optimization. Now we extend the theory to optimize Σ in parallel with C. We focus on the case where Σ is diagonal with elements Σii = (σi)2 that are the variances of the noise experienced by each network node. The key to adjust each Σii is to reproduce the corresponding diagonal element Q ^ i i 0. To do so, the Σii are simply modified at each optimization step based on the difference between the variance of the ith node of the model and its objective counterpart:
Δ Σ i i = ϵ Σ Q ^ i i 0 - Q i i 0 , (20)
where the update rate is ϵΣ.
The implications of tuning the noise matrix Σ will be explored in a dedicated section of Results. Intuitively, it can be seen from Eq (6) that Σ affects the model covariance Q0, hence Qτ via Eq (7); see Methods for reference. Therefore, assuming an erroneous Σ for the model implies incorrect desired updates ΔQ0 and ΔQτ, leading to a wrong update ΔC.
We also compare the LO method with a heuristic optimization, where existing weights are increased or decreased depending on the Qτ difference between the model and objective for the corresponding matrix element for a given τ Δ C i j = ϵ heur Q ^ i j 0 - Q i j 0 + Q ^ i j τ - Q i j τ . (21)
This method used previously [12] is adapted here to spatio-temporal covariances.
Unless stated otherwise, we use the parameters in Table 2 and Euler’s approach to simulate Eq (3) numerically. The code is written in python using the packages numpy and scipy, which incorporate a version of the Bartels-Stewart algorithm. The topology for the cluster-hub networks as in Fig 1A consists of two groups of 30% and 60% of the N nodes. Each group is connected recurrently with probability pcon, with a weight randomly chosen in [10, 100]% of cmax. We exclude self connections, although this is not crucial. The two groups are connected via hubs with a higher probability equal to 1.3pcon and a weight in the same range as other connections. Hubs do not have connections between each other. For random networks, all pairs of nodes have the probability pcon to be connected, with a random weight in [10, 100]% of cmax. Simulations of duration T with identical parameters but distinct initial conditions are repeated nsim times in order to evaluate the empirical Q.
We also compare our ND model with the DMF model [27] that simulates a network of neural population with AMPA and NMDA conductances, whose synaptic activity determines the BOLD signal via a hemodynamic nonlinear filter [29]. The equations regulating the synaptic activity variable Si with 1 ≤ i ≤ N for the DMF are
dSidt=−SiτNMDA+β(1−Si)H(ui)ui=J(wEESi+G∑kCikSk)+I0H(x)=ax+b1+exp[−c(ax+b)] . (22)
The HRF model relies on the synaptic activity Si to calculate the intermediate variables si, fi, νi and qi for each area, which give the BOLD signal Bi:
dsidt=Si−κsi−γ(fi−1)dfidt=sidνidt=fi−νi1/ατHdqidt=1τH[ fiρ[1−(1−ρ)1/fi]−qiνi1/α−1 ]Bi=V0*[7ρ(1−qi)+2(1−qi/νi)+(2ρ−0.2)(1−νi)] .
(23)
All parameters are recapitulated in Table 3.
Resting-state BOLD signal time series and dwMRI data were acquired for 25 healthy subjects aged between 18 and 39 years (mean 26.7), including 12 females and 13 males. A detailed description of the data acquisition can be found in [35], which we briefly summarize here. Subjects were asked to stay awake and keep their eyes closed, during which joint MRI-EEG recordings were performed using a 3 Tesla Siemens Trim Trio MR scanner and a 12-channel Siemens head coil [36, 37]. Acquired time series are spatially aggregated for voxels corresponding to the same area according to the parcellation. Only fMRI data are used in the present study: the BOLD signals were recorded for about 20 minutes (661 time points taken every 2 s). The BOLD signals are high-passed filtered above 0.01 Hz. Then the FC matrices correspond to covariances between all 68 regions, which are calculated using Eq (2) taking each of the 25 subjects as a separate session. Time shifts are multiples of the temporal resolution after preprocessing (2 s). For the most part, we use for FC the average covariance matrix over all subjects; a comparison with optimization based on FC for individual subjects is also presented in ‘Robustness of LO applied to experimental data. For each participant, dwMRI was used to evaluate the white-matter intracortical connectivity. Tractography is performed using corrections for motion, eddy currents, crossing pathways and the size of regions. Processing steps for MRI data include 1) preprocessing of T1-weighted scans, cortical reconstruction, tessellation and parcellation, 2) transformation of anatomical masks to diffusion space, 3) processing of diffusion data, 4) transformation of anatomical masks to fMRI space, 5) Processing of fMRI data. The dwMRI matrix used in the application to experimental data is the average over all subjects, which is thresholded to determine weights to tune in the optimization.
The paper makes a secondary use of experimental data already published [35]. The experiments have been approved by the Ethic Committee of Charite University (Berlin).
The goal of the present study is to tune a dynamic model of cortical activity to reproduce empirical FC obtained from fMRI data. Here FC corresponds to the spatio-temporal covariances Q ^ 0 / τ. The optimized network parameters are the matrices of recurrent connectivity C and noise Σ, the latter being diagonal. We perform the Lyapunov optimization (LO) developed in Methods to find the parameters that minimize the error between two model Q0/τ matrices and their empirical counterparts Q ^ 0 / τ, as illustrated in Fig 1H. The application to fMRI data is presented in the last two subsections of Results, where the tuned C is an estimate of directed intracortical EC. In particular, the asymmetry of EC relates to the non-reciprocality of cortical interactions and to the difference between the strengths of incoming and outgoing connections for each cortical area.
The first part of Results concerns artificial network models, on which we test the LO procedure. We examine the influence of the choice for the time shift τest related to the objective Q ^ τ upon the performance. We also show that it is necessary to tune the noise Σ received by the network nodes in order to accurately estimate the recurrent connectivity C. Our LO procedure is compared to two alternative to estimate the strengths of individual connections in a network, namely the ‘direct’ and ‘heuristic’ methods. Before applying our method to experimental data, we also investigate the mapping between EC and FC in a network model equipped with hemodynamics to generate the BOLD signal from neural activity. Specifically, we verify that time-shifted BOLD covariances convey information about the underlying connectivity between neural populations.
The network model consists of interconnected neural populations that experience intrinsic noise and excite each other. Details about the mathematical formalism and model parameters are provided in Methods. For artificial networks, we mainly consider the topology in Fig 1A, where two groups of interconnected populations are linked via hubs. An example of the connectivity matrix is displayed in Fig 1B, where the matrix element indexed by ij corresponds to the connection from area j to i. Although hubs (cyan) are connected to both groups and not between each other (as indicated red arrow), they seem to make one with the large group (light green) in Q0 and Qτ (see the blue arrow in Fig 1D). In contrast, the small group (dark green in Fig 1A and 1B) does not appear clearly as it exhibits rather low covariances. The reason behind this choice is to check whether the method can recover the correct topology from the Q0/τ values that significantly deviate from the underlying connectivity strengths in C.
Our estimation method uses the spatio-temporal information in the covariance Q ^ of the entire network to recover the directed weights in C, implicitly relying on the asymmetric matrix Q ^ τ for τ > 0. In general, Qτ is not symmetric for τ > 0 for an arbitrary C: Fig 1F shows that the asymmetry of Qτ linearly scales on average with that of C, with a slope that increases with τ. We measure the asymmetry using the following index
asym M = 0 . 5 ∑ i ≠ j | M i j - M j i | ∑ i ≠ j | M i j | (24)
for a matrix M. It ranges from 0 for symmetric matrices to 1 for antisymmetric matrices. Large weights with no or weak reciprocal connection mainly contribute to the index. The relationship in Fig 1F can be understood looking at the expression for Qτ in Eq (7), which amounts to Q0 multiplied by the matrix exponential expm(J† τ). The Jacobian J† is the transpose of the Jacobian matrix defined in Eq (5) and has the same off-diagonal elements as C, so asymJ = asymC. For small τ, Qτ can be approximated by Q0 + Q0 J† τ, which explains why asymQτ is smaller than asymC. The variability of the Qτ asymmetry compared to the small slopes of the curves makes it difficult to infer the asymmetry of C. This further motivates our method to estimate accurately C from Q.
We focus on the case where the recurrent weights are sufficiently large to generate a strong network effect: the mapping between C and Q is not trivially linear and many large values in Q0 correspond to nodes that are only weakly or not connected. Here, hubs in the network in Fig 1A are not connected, but have strong variances in Fig 1G and covariances between each other due to their numerous connections.
The optimization procedure is summarized in Fig 1H: C is iteratively modified such that the model covariance matrices Q0 and Qτ converge towards the objective covariance matrices Q ^ 0 and Q ^ τ, for a chosen τ = τest > 0. In the first place we assume Σ to be known. Starting from a initial matrix C with all weights equal to zero, LO calculate a C update at each step such that it reduces the Lyapunov function V in Eq (13); V is the sum of the matrix distance between Q0 and Q ^ 0 on the one hand, and that between Qτ and Q ^ τ on the other hand. To measure the goodness of fit of C or Q, we define the model ‘error’ as the normalized distance for a matrix M and its objective Mobj:
d ( M , M obj ) : = ∑ ( i , j ) ( M i j - M i j obj ) 2 ∑ ( i , j ) ( M i j obj ) 2 , (25)
which are involved in V.
Firstly we verify that LO recovers the correct connectivity for theoretical objectives Q ^ 0 / τ calculated with Eqs (6) and (7) with the original C and Σ. A typical example for the evolution of the C and Q errors is illustrated in Fig 2A for τest = τx = 1 s. While the Q0/τ errors exhibit a plateau, the C error keeps on decreasing. Fig 2B shows the faster convergence for Q than C for three stages of the optimization. The residual C error is very small; it may remain non-zero when the commutativity assumption necessary to derive Eq (15) is not satisfied.
For the same original C, we now simulate the network and calculate the empirical objectives Q ^ 0 / τ using Eq (2) for 50 independent repetitions of 300 s each with timestep 0.05 s (i.e., 3 × 105 data points). In Fig 2C, the Q0/τ errors first drop, then rebound and stabilize. Interestingly, the C error continues to decrease even though the Q error worsens. Because of the noise in the empirical Q ^ 0 / τ, there is no real solution C to Eq (7) and the residual error does not vanish. In Fig 2D, the model C gives a better fit for the original C after 5000 optimization steps than for the minimum for the Q error. LO also allows for the use of constraints on the weights in C. Non-negativity is imposed here, although it does not affect the results significantly.
Our method relies on information in time-shifted covariances to infer the connectivity of the underlying network. In the context of cortical models of BOLD activity, EC classically denotes the connections between neural populations, whose activity is transformed into BOLD signals via a hemodynamic response function (HRF). Before applying our method to fMRI experimental data, we need to test whether BOLD time series as obtained from the HRF convey information in their time-shifted measures about EC.
Therefore we compare our noise-diffusion (ND) network with the dynamic mean-field (DMF) model based on AMPA and NMDA dynamics [27] equipped with the Balloon/Windkessel model for the HRF [29]. As illustrated in Fig 6A, the synaptic input activity of the DMF (ui in Methods) is transformed by the HRF to generate the BOLD signal. We consider the two topologies in Fig 6B: a randomly connected network and a network whose connections correspond to the SC obtained from dwMRI (see Methods). The SC connectivity has 32% density corresponding to the largest dwMRI values, as will be used in the next sections for the ND model applied to fMRI data. The model parameters are taken from previous publications [27, 29] and recapitulated in Table 3. In Fig 6C, we simulate 50 configurations of the DMF+HRF model with the two topologies (‘DMF/rnd’ and ‘DMF/SC’, respectively). For each, we choose weights randomly and calculate the empirical autocovariances of the BOLD signals. We find significant values up to time shifts equal to 3 s, which then drop. Interestingly, the slope in the log plot for the smaller time shifts is close to that found in experimental data (black line) up to 2 s of time shift. Then we estimate the similarity between the neural covariances at a short timescale and the FC given by the BOLD covariances, using the Pearson correlation coefficient between the corresponding matrix elements. As shown by the Pearson correlations in Fig 6D (left panel), the match between the neural and BOLD covariances is good, especially when EC is constrained by SC. This means that, although the HRF involves nonlinearities, the DMF model preserves the ranking of the spatio-temporal covariances is preserved between neural and BOLD activity, as illustrated in Fig 6E. In particular, the time-shifted BOLD covariance matrices are asymmetric.
Now we use the LO procedure to estimate the original EC from the BOLD FC with τest = 1 s. The right panel in Fig 6D indicates the Pearson correlation between the original and estimated connectivity. For the DMF model, the agreement corresponds to 0.6 and 0.65 for the random and SC topologies, respectively. This means a fair recovery of the ranking between the original EC weights. For the DMF/SC networks where the existing connections are “known” (32% density), LO only tunes the corresponding weights in C.
Finally, we find that the variability of the estimated ECs decreases with the strength of the neural connection. This suggests that strong estimated EC values can be trusted based on a threshold on the variability of the estimations (standard deviation divided by the mean). Together, these results support the applicability of our method to fMRI data and directly estimate the EC strengths from the average FC.
Now we use the ND model to reproduce the cortical FC obtained from fMRI recordings during rest. In contrast to previous studies [12, 27], FC involves both zero-time-shift and time-shifted covariances of the BOLD signals for the 68 cortical regions here. Following the results in Fig 6, we denote by EC the connectivity C in our model, even though the network activity directly models the BOLD signals without a HRF. The data set corresponds to 25 subjects aged from 20 to 39 years. The empirical covariances are calculated using Eq (2) for τ that are multiples of 2 s, which is the temporal resolution of the BOLD time series. Fig 7A displays the objective FC matrices Q ^ 0 and Q ^ τ with τ = 4 s, which are averages over all subjects. The autocorrelograms of all 68 nodes are displayed in the inset of Fig 7B. The main panel shows the same curves with a log-scale for the y-axis: they seem quasi straight lines indicating exponential decays, as observed for artificial networks in Fig 1E. Therefore, we estimate from the mean BOLD autocovariance over all nodes (red curve of Fig 7B) the time constant τx = 5.3 s, which we use to calibrate the model. We use a single τx for all nodes, because the distribution of individual time constants over all nodes is narrow, as shown in Fig 7C.
The direct method in Eq (8) does not work here: the reason is that the matrix logarithm gives complex values with large imaginary parts. This motivated the development of the LO procedure. Moreover, our approach allows for the use of constraints on the weights in C: only a subset of all possible cortico-cortical connections can be tuned. The mask in Fig 7D represents the anatomical cortical connectome obtained from thresholding the dwMRI data averaged over all subjects: connections with large SC values are optimized, other weights are kept equal to zero. The density of existing connections is 32% in Fig 7D.
The blue curve in Fig 7E represents the evolution of the Q error during LO with τest = 4 s: it first decreases and then “explodes”. This instability corresponds to a transition to excessively high activity due to too strong feedback. The black curve show the normalized rate of change for C, whose minimum is close to that of the Q error. For the experimental data, the minimum of the black curve comes before the minimum of the Q error, but the rate of change of C remains close to its minimum until the explosion. In the following, we choose the estimated EC in Fig 7F and reconstructed FC as the Q and C for the minimum of the Q error. The match between the model and empirical Q matrices is illustrated in Fig 7G. Importantly, the match is similar for the Q corresponding to tuned connections (blue dots) and absent connections (black crosses). The fit of variances corresponding to the diagonal elements of Q (cyan dots) is very good.
Fig 7H indicates the poor match between the empirical FC and the symmetric part of estimated C: large values in Q ^ 0 arise from strong network effect even though the EC weight is low.
Fig 10A shows the very weak match between the C and the SC corresponding to the dwMRI data averaged over all subjects, with a Pearson correlation coefficient equal to 0.06. The estimated EC is thus structurally very different from a scaled version of SC. As dwMRI reflects the density of cortico-cortical white-matter fibers, this suggests that the efficacies of these connections are determined by other factors than their size, such as types of neurotransmitters, concentration of synaptic receptors and excitability of cortical areas.
The estimated EC matrix in Fig 7F is not symmetric, meaning that cortical interactions are not reciprocal. This information is important as dwMRI data estimate the density of axonal fibers in the white matter, but do not recover the direction of those fibers. SC is by construction quasi symmetric, as is the case in Fig 7D. The matrix asymmetry for C relates to the reciprocity of intracortical connections and can be seen in the difference between incoming and outgoing strengths in Fig 10B. No area has both large incoming and outgoing weights, meaning that hubs act either as receptors or feeders.
In Fig 10C, the results are mapped onto the cortical surface. Cortical feeders with the largest outgoing weights are the left and right fusiform, middle temporal and superior temporal gyri, as well as the pre- and postcentral gyri in the left hemisphere. Cortical receivers with the largest incoming weights are the left and right precuneus, lateral occipital and superior parietal gyri, as well as the left isthmus of the cingulate gyrus. We also find that the following areas exhibit the largest values for Σ, synonymous with strong intrinsic variability: both left and right lingual gyri, pericalcarine cortices and frontal poles, as well as right cuneus and transverse temporal gyrus, and left pars orbitalis. This suggests the propagation of spontaneous activity, mainly from visual cortices and the prefrontal area.
We have shown in a noise-diffusion network model how the directed connectivity C can be retrieved from empirical covariances Q ^. The key is to take into account the temporal information in covariances Q ^ τ for non-zero time shifts τ > 0. Our proposed method gives a better fit of all Q ^ τ for τ ≥ 0, not only Q ^ 0 that is often considered alone as an objective or goal in fitting procedures. Our theoretical study demonstrates two crucial requirements in order to recover the original C in the considered noise-diffusion networks: the time shift τest corresponding to Q ^ τ must be matched with the time constant of the dynamical system τx estimated from the data (Fig 3); it is also necessary to adjust the diagonal matrix Σ that relates to spontaneous fluctuations experienced by each network node (Fig 4).
Our method provides an estimation of the asymmetry of intracortical connections (EC) from fMRI data combined with anatomical information from dwMRI. This is to our knowledge the first of its kind for the human connectome at the scale of the whole cortex. In addition to intracortical EC, our method also estimates via Σ the intrinsic variability of each cortical area, which is then shaped by EC to generate FC. This estimation relies on information in the BOLD spatio-temporal covariances, which convey information about the underlying neural connectivity (Fig 6). Our results suggest that the EC cortical hubs are either receivers or feeders, but not both (Fig 10).
It is known that zero-time-shift covariances are not sufficient to retrieve directed connectivity, but only its symmetric part [16, 25, 28]. Information-based methods able to estimate directionality such as likelihood maximization [13], Granger causality [14, 23] and transfer entropy [24] also use temporal information of the observed activity. In minimizing the matrix distance between the model Q0/τ and objective Q ^ 0 / τ, instead of considering connections independently, the LO procedure captures network effects due to the recurrent feedback. Here we do not perform a stochastic gradient descent using many samples of the observed activity, but a deterministic optimization based on the Q ^ 0 / τ averaged over the whole observation period (or several simulation sessions). It follows that the optimization is quick: a few minutes for 104 optimization steps with a network of 50 nodes and a given τest on a recent desktop computer.
To obtain the best performance, we have shown that the time shift τest corresponding to Q ^ τ used in LO and the time constant of the dynamical system τx should be matched. As shown in Fig 3E, poor estimation for large τest arises from the inaccuracy in empirical Q ^ τ; for small τest, LO itself is unstable (see dashed curve in Fig 3F). Fig 5C shows that the method performance is not strongly affected by the network topology or the connectivity asymmetry, but worsens with the network size and becomes better with more observations used to calculate the empirical objective Q ^ 0 / τ. This is in line with previous results [25]. In addition to C, the intrinsic noise Σ received by the network nodes must be tuned to obtain a correct C estimation (Fig 4). Here we use a heuristic optimization for a diagonal Σ in Eq (20); the present framework should be extended to take into account correlated noise instead of white noise.
The direct method in Eq (8) to estimate C has been used previously with statistical tests to estimate the existence of connections from observed activity [16, 26]. As shown in Fig 5D, it does not work well for the level of noise in empirical observations Q ^ 0 / τ considered here. This motivated the development of our LO procedure in Methods. Here we focus on the estimation of connection weights, i.e., their ranking; the detection of connections using statistical methods can be based on the estimated C from LO, but this is left for further work. Nevertheless, detection should be based on an as-good-as-possible estimated ranking of connections weights, which can be measured by the Pearson correlation. Our method for Ornstein-Uhlenbeck processes also bears similarities with MVAR [3], but we enforce additional constraints on the connectivity.
Recent studies [12, 28] have also used greedy algorithms to optimize symmetric C relying on Q ^ 0 using more elaborate network models. Those procedures update C step-by-step according to various measures such as the Pearson correlation between all matrix elements of zero-time-shift correlations. Here we have transposed this heuristic method to reproduce Q ^ τ in addition to Q ^ 0, see Eq (21). Although the resulting Q0/τ fit is close to perfect, the C estimation remains poor in Fig 5D. Taking the network effect correctly into account via LO is important to recover the original C, as compared to tuning connections individually based on the corresponding Q0/τ value.
More generally, the problem with inferring C lies in the definition of observables or objective functions to constrain models without ambiguity: to a set of network parameters should correspond only a single value of the observable (here a pair Q ^ 0 + Q ^ τ). In [25], the minimization of matrix L1 norm for sparse networks was used to reduce the indetermination in using Q ^ 0 only. Beyond our results based on noise-diffusion processes, we expect that directed C can be recovered for other network models such as Hawkes processes (a.k.a. Poisson neurons) or binary neurons using second-order moments with non-zero time shift (here covariances). This is supported by recent results that demonstrate how the covariance structures are formally related across these neural models [38].
The present framework appears well suited to model activity as continuous signals; for spike trains generated by networks of GLM or Poisson neurons with simple temporal filters, it remains to be seen whether our fast tuning procedure can be adapted. We have only considered resting-state activity, but the procedure may also be extended to the case of multiple stimulation-response pairings. In particular, the external input e may be adjusted [16], in addition to C and Σ.
The goal of our model-based approach is to reproduce the resting-state FC obtained from fMRI. Although the ND model is not new, we propose a novel ‘LO’ procedure to tune the model parameters with suitable observables, namely time-shifted covariance matrices as FC. The estimated connectivity of the ND model relates to the intracortical EC, whose properties can then be analyzed [1]. For instance, EC can be searched for hubs, communities and similar features [30]. Here we have focused on hubs (Fig 10) and our results suggest that activity propagates from the visual, auditory and prefrontal areas. We find among the listening hubs the precuneus and superior parietal gyrus that belong to the default-mode network. These networks are usually found in resting-state FC and our results shed a new light on the architecture that shapes the activity propagation between them.
The ND model was previously used together with hemodynamics in order to reproduce similar fMRI data [12, 27]. As was demonstrated for that study for zero time shifts, the BOLD covariances with time shifts of the order of seconds also convey information about the interactions that determine the neural dynamics (Fig 6). This is in line with recent results showing that the BOLD time series convey information about cognitive processing for similar time lags [39, 40]. In our model, the EC asymmetry generates lags in covariances. Moreover, the LO procedure can estimate the EC ranking this neural connectivity from the BOLD covariances provided the ranking is preserved between the neural and the BOLD covariances for those more elaborate models involving a HRF (Fig 6). This supports the application of the ND model without HRF to model directly the BOLD data. In particular, the ND model seems well adapted to fMRI data at the considered parcellation of about 100 nodes: the BOLD autocovariances are close to exponential decays (i.e., straight lines for the log y-axis), as shown by the comparison between the ND model in Fig 1E and our experimental data for time shifts up to 8 s in Fig 7B. The corresponding time constants are rather homogeneous over all regions in Fig 7C, so we use a single τx = 5.3 s to calibrate the ND model, which corresponds to typical values for HRF [41]. In contrast, the autocovariances for the DMF+HRF in Fig 6C show a drop for τ ≥ 4 s that is not observed in the experimental data. We keep in mind that the precise relationship between the fluctuations of the BOLD signal and neural activity is still under debate [41–43]. Numerous studies such as those about repetition priming and suppression [44] have shown how changes in fMRI signals reflects those in neural activity, such as synchronization of the latter at a much shorter timescale. Moreover, fMRI is one of the few non-invasive methods to evaluate processing in the human brain (of course in vivo) and has many clinical applications [45], irrespective of its precise link to neural activity. This further supports efforts to develop generative models of BOLD and methods to better interpret these fMRI data.
An underlying assumption of our approach is that the individual variances of BOLD signals are meaningful [46]. This motivates the use of covariances to fit our model, rather than correlations that are often used [1]. The results in Fig 7 give a FC Pearson correlation coefficient (sometimes called predictive power) larger than 0.6 for both Q0 and Q4. In this sense, our study improves previous results [12, 31]. Importantly, the EC and Σ estimated by LO is surprisingly stable with respect to the choice of time shift for the considered experimental data. We find consistent results for a broad range of τest from 2 to 8 s, with a Pearson correlation coefficient larger than 0.9 in Fig 8. Choosing denser connectivity for EC improves the FC fit, but does not significantly change the EC ranking as shown in Fig 9A and 9B for results from 25% to 45% density. As an additional check, we have also shown that EC estimated from individual FCs coincides with EC obtained from the mean FC over all 25 subjects. In comparison, the heuristic and direct methods in Figs 8E and 9B and 9C give far more inconsistent results. Moreover, the heuristic method only uses part of the FC matrices when EC is sparse: it tends to overfit the existing connections compared to absent connections. A recent study [15] compared methods to estimate directed cortical interactions in a generative model of BOLD activity including hemodynamics: among those, Patel’s τ[47] gave the best performance, but that study was limited to sparse connectivity. Here we consider the FC for the whole cortex, which gives coupled constraints for a reasonably large network. Incorporating more areas limits the number of unknown contributions to the node activities, strengthening the estimation accuracy. This was observed for partial correlations in a similar manner for undirected connections [15].
Several previous studies used a scaled version of the SC matrix as EC [12, 28, 31]. As shown in Fig 8F, a symmetric EC does not satisfactorily fit time-shifted FCs. Furthermore, Fig 10A shows the weak match between dwMRI and our estimated EC values. This suggests that the fiber densities as measured by dwMRI may not be a good predictor for the dynamic interactions between cortical areas, but only for the skeleton of the cortical connectome. In addition, those previous studies used interhemispherical connections to improve the FC fit of the models. Here these connections also appear to be fairly strong in the estimated EC in Fig 7F, as shown by the two secondary diagonals in the top-left and bottom-right quadrant of the matrix. In any case, large covariances are obtained even for unconnected areas because of the strong overall feedback in the network, in line with results supporting that the cortex is in a state close to criticality [27]. As the dynamics of each cortical region has no nonlinearity in our model, large EC values are the only origin for the strong network effect, which induces the non-trivial mapping between EC and FC in Fig 7H.
As shown in Fig 1F for artificial networks, the C asymmetry is reflected in Qτ for τ > 0. In our experimental data, the corresponding FCτ (blue line in Fig 9C) is not symmetric and relates to propagating activity [39]. In the ND network, the noise received by each area is embodied by Σ, which is then shaped by EC to generate FC. Here the LO procedure extracts the spatio-temporal FC information to estimate the EC asymmetry. The information about asymmetry in intracortical connections is thus important and complements the SC obtained from dwMRI, which is symmetric. Compared to previous studies that investigate the directionality of cortico-cortical connections [2, 4, 7], the novelty is that we estimate this property at the scale of the whole cortex. As all nodes have the same activation function, EC values indicate the relative interaction strengths between areas. What matters in our EC analysis is the ranking of EC weights: the difference between a low EC value and an absent connection is not so important here. Here we have estimated the asymmetry of cortico-cortical EC to be equal to 0.35 in Fig 8C. This is larger than macaque’s COCOMAC asymmetry that gives 0.14 for the same index [48, 49], but note that nonlinearity in the model dynamics would affect the precise EC values; what is important here is that EC is significantly asymmetric. This translates to an imbalance between incoming and outgoing EC weights. Fig 10B suggests that hubs are either feeders or receivers, but do not have both strong incoming or outgoing connections. As a conclusion, our work sets a suitable ground to study both local and global properties of the whole cortical connectivity, which will give insight in the underlying neural processing.
|
10.1371/journal.pntd.0001794 | Global Gene Expression Analysis of the Zoonotic Parasite Trichinella spiralis Revealed Novel Genes in Host Parasite Interaction | Trichinellosis is a typical food-borne zoonotic disease which is epidemic worldwide and the nematode Trichinella spiralis is the main pathogen. The life cycle of T. spiralis contains three developmental stages, i.e. adult worms, new borne larva (new borne L1 larva) and muscular larva (infective L1 larva). Stage-specific gene expression in the parasites has been investigated with various immunological and cDNA cloning approaches, whereas the genome-wide transcriptome and expression features of the parasite have been largely unknown. The availability of the genome sequence information of T. spiralis has made it possible to deeply dissect parasite biology in association with global gene expression and pathogenesis.
In this study, we analyzed the global gene expression patterns in the three developmental stages of T. spiralis using digital gene expression (DGE) analysis. Almost 15 million sequence tags were generated with the Illumina RNA-seq technology, producing expression data for more than 9,000 genes, covering 65% of the genome. The transcriptome analysis revealed thousands of differentially expressed genes within the genome, and importantly, a panel of genes encoding functional proteins associated with parasite invasion and immuno-modulation were identified. More than 45% of the genes were found to be transcribed from both strands, indicating the importance of RNA-mediated gene regulation in the development of the parasite. Further, based on gene ontological analysis, over 3000 genes were functionally categorized and biological pathways in the three life cycle stage were elucidated.
The global transcriptome of T. spiralis in three developmental stages has been profiled, and most gene activity in the genome was found to be developmentally regulated. Many metabolic and biological pathways have been revealed. The findings of the differential expression of several protein families facilitate understanding of the molecular mechanisms of parasite biology and the pathological aspects of trichinellosis.
| Trichinellosis of human and other mammals was caused through the ingestion of the parasite Trichinella sparilis in contaminated meat. It is a typical zoonotic disease that affects more than 10 million people world-wide. Parasites of the genus Trichinella are unique intracellular pathogens. Adult Trichinella parasites directly release newborn larvae which invade striated muscle cells and causes diseases. In this study, we profiled the global transcriptome in the three developmental stages of T. spiralis. The transcriptomic analysis revealed the global gene expression patterns from newborn larval stage through muscle larval stage to adults. Thousands of genes with stage-specific transcriptional patterns were described and novel genes involving host-parasite interaction were identified. More than 45% of the protein-coding genes showed evidence of transcription from both sense and antisense strands which suggests the importance of RNA-mediated gene regulation in the parasite. This study presents a first deep analysis of the transcriptome of T. spiralis, providing insight information of the parasite biology.
| Thichinella spiralis is often referred as one of the largest intracellular parasite that cause trichinellosis with an estimation of more than 10 million people infected world-wide [1], [2]. Like many other food-borne zoonotic parasites, T. spiralis exists in several life-cycle stages. The complex life cycle of T. spiralis is completed in two niches, the intra-multicellular niche in intestinal epithelium (adults, Ad) and the intracellular niche in the skeletal muscle fibers (muscle larvae, ML). After being ingested with infected muscle tissue, the ML are released and revived in the small intestine, which invade the epithelial layer where they mature, mate and produce the newborn larvae (NBL). The NBL migrate through the lymphatic and blood vessels, invade striated muscle cells and develop into the ML, which is infective to the next host [3]. Thus, unlike other nematodes, T. spiralis are ovoviviparous [4], [5], which makes them evolutionary divergent from other nematodes previously analyzed by genomic approaches [6], especially the well-characterized free-living worm Caenorhabditis elegans. C. elegans is the first multicellular organism, of which the genome has been sequenced and phylogenetically classified in Clade V [7]. T. spiralis is a member of clade I that diverged early in the evolution of the Nematoda, with remarkably different biological and molecular characteristics from other nematodes [6].
Previous analysis of gene expression of T. spiralis at different developmental stages has mainly been carried out by sequencing of expression sequence tag (EST) [6]. However, the fragmentary data of ESTs are insufficient for a full understanding of the parasite biology. Recent study indicated that T. spiralis has a much smaller genome with an estimation of 64 Mb in nucleic DNA, coding for around 15,808 proteins [8]. The availability of the genomic information has made it possible for a deep dissection of the parasite's basic biology. In recent years, next-generation sequencing (NGS) techniques have dramatically improved the efficiency and the speed of gene discovery [9], [10]. NGS technology generates millions of short sequence reads from a single instrumental run which can be effectively assembled and employed for gene discovery and comparison of gene expression patterns. Further, NGS allows for a detection of genes with very low expression levels. It has been frequently used to characterize specific gene families or genetic pathways.
In this study, we compared gene expression variations among the three developmental stages of T. spiralis using next generation sequencing technology. Thousands of genes, especially the genes involving in host/parasite interactions and parasite development were identified. The data will facilitate discovery of potential vaccine and drug targets of T. spiralis.
Muscle larvae (ML) of T. spiralis (strain ISS534) were obtained from rats at 35 days post infection by digestion of minced skeletal muscle according to the previously described method [11], [12]. The adult worms and newborn larvae were collected as described previously [12], [13]. The study of using laboratory animals was reviewed and approved by the Ethical Committee of Jilin University affiliated to the Provincial Animal Health Committee, Jilin Province, China (Ethical Clearance number IZ-2009-008). All animal work was conducted according to the guidelines of the Chinese Law of Animal Protection (Section 6).
Total RNA of T. spiralis (Ad, NBL and ML) was purified using Trizol reagent (Invitrogen, CA, USA) according to the manufacturer's instructions. RNAs were dissolved in Diethylpyrocarbonate (DEPC)-treated water and treated with DNase I (Invitrogen, CA, USA). Total RNA was quantified by measuring the absorbance at 260 nm with a Nanodrop 1000 machine (Thermo Scientific CA, USA).
The DGE libraries were prepared by using the Illumina gene expression sample preparation kit [14]. Briefly, 6 µg of total RNA from each preparation was treated with Oligo-(dT) conjugated magnetic beads to purify mRNA. Double-strand cDNA was synthesized guided by the Oligo-(dT) as a primer and digested with the endonuclease NlaIII that recognizes the CATG sites. The Illumina adaptor 1, containing a MmeI restriction site, was added to the cDNAs attached to the magnetic beads, which was further digested with MmeI. Following MmeI digestion and dephosphorylation, cDNA fragments were purified and the Illumina adaptor 2 was ligated to the 3′ends of the tags to create tag library with different adapters at both ends. After15 cycles of linear PCR amplification, the 95 bp fragments were purified from 6% TBE PAGE gels and attached to the Illumina sequencing chip for sequencing.
After removal of low quality and adaptor sequences, the clean 21 bp tag sequences containing CATG were mapped onto the reference genome sequences, allowing for no more than one nucleotide mismatch. To compare the differences in gene expression in three DGE libraries, the tag frequency was statistically analyzed according to the method FDR (False Discovery Rate), a method used to determine the threshold of P-value in multiple tests [15]. A FDR<0.001 and an absolute value of the log2Radio>1 was used as the threshold to judge significant differences in gene expression. KEGG Ontology (KO) of the transcripts was identified trough blasting the KEGG database. Gene Ontology (GO) analysis of all differentially expressed genes was performed by searching in the GO database. The enriched p-values of KO and GO were calculated according to the hypergeometric test [16]:In this equation, N means the number of genes with GO/KO annotation, n means the number of differentially expressed genes in N, M means the number of genes in each GO/KO term, and m represents the number of differentially expressed gene in each GO/KO term. For GO enrichment analysis, all P-values were treated with Bonferroni correction. We selected a corrected p-value,0.05 as a threshold to determine significant enrichment of the gene sets. In contrast, for KO enrichment analysis, we used a FDR,0.05 as a threshold to determine significant enrichment of the gene sets. WEGO was employed to make a GO classification [17].
Phylogenetic tree was built for DNase II protein families using PHYLIP (version 3.69; [18]) after aligning the family members with CLUSTAL X (version 2.1). And a neighbor joining tree was generated using PHYLIP-NEIGHBOR. Then, the phylogenetic tree was visualized and edited using the Tree Figure Drawing Tool - FigTree (version 1.3.1).
Total RNA of T. spiralis (muscle larvae, adult worms and newborn larvae) was extracted using Trizol reagent (Invitrogen, CA, USA) and treated with Dnase I (Invitrogen, CA, USA). The RNAs were dissolved in Diethylpyrocarbonate (DEPC)-treated water and reverse transcribed with 200 U SuperScript™ III Reverse Transcriptase (Invitrogen) according to the manufacturer's instructions. The specific primers were listed in File S1 as forward and reverse primers based on the stage-specifically expressed genes. The GAPDH gene was used as an endogenous reference. The qPCR was performed using the SYBR Kit (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's protocol using an Applied Biosystems 7500 detection system. The relative expression was analyzed using the SDS1.4 software (Applied Biosystems, Foster City, USA).
To generate digital gene expression signatures of T. spiralis at different developmental stages, DGE libraries were generated from the three developmental stages of the parasite, and sequenced using Solexa (Illumina) high through-put technology. A total of 5,289,863 tags from muscular larvae (ML), 5,214,135 tags from adult worms (Ad), 5,055,659 tags from newborn larvae (NBL) were obtained (Table 1). After filtering the low quality tags and adaptor sequences, the total number of clean tags in ML, Ad and NBL were 5,077,645 (96.0% of total tags), 5,003,105 (96.0% of total tags), 4,849,883 (96.0% of total tags), respectively (Table 1). The sequence data has been submitted to the GEO website (ftp://ftp-private.ncbi.nlm.nih.gov/fasp/) with an accession number of GSE39151.
Heterogeneity and redundancy are two significant characteristics of gene expression in the three developmental stages of T. spiralis, which have previously been observed in other metazoans [19]. The distribution of clean tags in the three libraries shows a consistent pattern, with most of the tags coming from highly expressed genes. The percentage of distinct tags with high counts dropped dramatically and the distinct tags with more than 100 copies accounted less than 10%. However, more than 75% of total clean tags have an account above 100 (File S2).
The clean tags were then mapped onto the draft genome of T. spiralis (ftp://ftp.ncbi.nlm.nih.gov/genbank/wgs/wgs.ABIR.1.gbff.gz) [8] and the numbers of tags that could be mapped onto genes with no more than one base pair mismatch in Ad, ML, NBL were 2,670,150, 2,187,559 and 2,254,426, respectively (Table 1). In total, around 10,000 genes were identified from the three libraries, accounted for approximately 65% of genes in the annotated genome [8] (File S3).
To identify genes that differentially expressed in the three developmental stages, gene expression variations were analyzed by pair-wise (Ad versus ML, NBL versus Ad, and NBL versus ML) comparison of the sequence tags. A number of genes were found differentially expressed between the developmental stages (Figure 1 and File S4). The number of differentially expressed genes between NBL and Ad was more than that between Ad and ML. And the number of up-regulated genes expressed in NBL was more than that in the other two stages. Apart from the differentially expressed genes, a number of stage-specific genes were identified and the number of stage-specific genes expressed in NBL was twice as much as that of Ad or ML (Figure 2).
Among the differentially expressed genes, genes coding for families of proteases such as astacin protease, serine protease, and DNase II (Table 2, and File S4 and S5) were more prominent. Serine protease and DNase II constitute the two excretory-secretory (E-S) protein families involved in host-parasite interactions in trichinellosis [6], [20]. The two gene families showed obviously stage-specific variations in expression in three developmental stages. 47 differentially expressed DNase II family genes were identified and most of these genes were up-regulated in NBL compared to the other stages (Figure 3, Table 2 and File S6). In contrast, only 6 DNase II genes were up-regulate in ML compared to the other stages (Table 2 and File S6). Further, most of these genes have homology with 27 previously identified DNase II homologues (File S6) [20]. Tsp_11476 and Tsp_12138 showed very high expression in NBL, while Tsp_06568 was detected in rather high abundance in Ad; and Tsp_00874 and Tsp_00875 were mainly expressed in ML (Figure 4A). Another interesting gene family is that encode serine protease. Contrast to DNase II family, most of serine protease genes were up-regulated in Ad compared to the other stages. In the three developmental stages of the parasite, a large number of serine protease family genes showed stage-specific expression, especially Tsp_00436, Tsp_15812, Tsp_07356, Tsp_07750 and Tsp_14046 (Figure 4B and Table 2).
Apart from the DNase II and serine protease families, genes encoding zinc metalloprotease, serine protease inhibitor (serpin), the heat shock protein (HSP), macrophage migration inhibitory factor (MIF) and antioxidant enzymes were also identified. Zinc metalloprotease is high homologous with nematode astacin protease. Genes coding for serpin and zinc metalloprotease showed a similar stage-specific expression pattern with up-regulation in NBL rather than the other stages, especially the genes like tsp_00173, tsp_01570, tsp_06688, tsp_09479, tsp_04481, tsp_01304, tsp_00804 and tsp_03942. Superoxide dismutase (SOD) and glutathione perxidase are two important antioxidant enzymes which protect the parasite from reactive oxygen species. The genes encoding these enzymes also showed stage-specific expression. The genes tsp_01933 (SOD) and tsp_06126 (SOD) showed very high expression in NBL, while tsp_11103 (SOD) and tsp_02268 (glutathione perxidase) were detected in rather high abundance in ML. Tsp_06335 encoding MIF and tsp_11249 encoding cystatin were mainly expressed in Ad. The gene encoding HSP 70 (Tsp_06317) was found up-regulated in ML (Table 2 and File S7).
The enzymes involved in metabolisms showed obviously stage-specific in transcription pattern. Phosphofructokinase (tsp_05639), enolase (tsp_09466) and Pyruvate Kinase (tsp_08030) were up-regulated in NBL. Whereas, the expression of Tsp_08363 encoding the major hexokinase isoenzyme showed no significant differences. Tsp_05267 encoded the minor hexokinase isoenzyme and was mainly expressed in Ad and ML. Lactate dehydrogenase (tsp_08060) and phosphoenolpyruvate carboxykinase (tsp_007989), the key enzymes in anaerobic metabolism, were up-regulated in NBL rather than the other stages. Whereas Two genes (tsp_03114, tsp_08643) encoding subunits of pyruvate dehydrogenase, which associated with glycolysis in the citric acid cycle via conversion of pyruvate to acetyl-CoA were mainly expressed in Ad and ML. Citrate synthase (tsp_01728) and isocitrate dehydrogenase (tsp_05617) were up-regulated in Ad. Another isocitrate dehydrogenase (tsp_06181) was up-regulated in ML (Table 2 and File S7).
In order to verify the genes that were actually differentially expressed in the three developmental stages, the expression of 16 genes respectively coding for DNase II (Tsp_11476, Tsp_12138, Tsp_06568, Tsp_00874 and Tsp_00875), serine protease (Tsp_00436, Tsp_15812, Tsp_07356, Tsp_07750 and Tsp_14046), and 6 genes respectively encoding heat shock protein A (Tsp_06317), macrophage migration inhibitory factor (MIF) (Tsp_06335), cystatin (Tsp_11249), systeine-glycine (Tsp_01806) and two genes with unknown functions (Tsp_05189 and Tsp_11467) were analyzed by quantitative real-time PCR. The q-PCR results confirmed the data obtained in the sequencing analysis (Figure 4 A, B and C).
Among the transcripts of the 9,969 genes identified, 35% (3,496) could be assigned into one or more GO categories which were consistent with previous studies [8]. The remaining uncharacterized genes are likely to fulfill novel functions. In each of the three main categories (molecular function, cellular component and biological process) of the GO classification, ‘binding and catalytic activity’, ‘cell and cell part’ and ‘cellular process and metabolic process’ are dominant (Figure 5 and File S8).
In KO classification, 156 differentially expressed genes were significantly enriched in Ad versus ML, 408 genes in NBL versus ML and 395 genes in NBL versus Ad. Most of these genes encode proteins participating in ‘metabolism’, ‘environmental information processing’, ‘cellular processes’ and ‘organism systems’, i.e. glycolysis/gluconeogenesis, oocyst meiosis, calcium signaling pathway, vascular smooth muscle, insulin signaling pathway and lysosome (File S9). The comparison between Ad and the other two stages revealed that most of the expression of genes related to oocyst meiosis was up-regulated in Ad, while most of the genes correlated to glycolysis/gluconeogenesis, calcium signaling pathway and vascular smooth muscle contraction were up-regulated in NBL instead of Ad and ML.
One of the advantages of DGE technique is that it could reveal transcripts with strand specificity [14]. We thus analyzed the sense and antisense transcripts of the transcriptome obtained. Of the 9,969 genes identified in this study, approximately 70% showed evidence of transcription in both orientations. The antisense transcripts obtained in Ad, ML and NBL were 6,303, 6,690, 7,033, respectively (File S10). While the transcripts from both strands obtained in Ad, ML and NBL were 5,269, 5,657 and 6,058, respectively. Among these genes, 2,682, 2,818 and 3,105 genes had tags corresponding to sense strands more than that from antisense strands in Ad, ML and NBL. Further, 2,476, 2,662 and 2,867 genes had more tags from antisense strands than that from sense strands in Ad ML and NBL stages (Table 3 and File S10).
The availability of the draft genomic sequence of T. spiralis has made it possible to deeply investigate the global gene expression and gene regulation mechanisms in the development of the parasite. By comparing the DGE libraries obtained from three developmental stages of T. spiralis, we have identified a large number of functionally important genes with differential expression features.
First, like in other organisms with multiple developmental stages, the genome of T. spiralis is developmentally regulated. More than 70% of the genes in the genome were found activated in all three developmental stages, of which between 5–11% of genes are preferentially expressed at different stages. In general, more transcripts were identified in the new borne larvae, suggesting that genes at this stage are more active (Figure 1, 2, Table 1–2 and File S2). Unlike adult worms, the new-borne larvae will need to invade intestine epithelium and establish new niches in the tissue, thus the parasite need a different biological arsenal from the adult stage for adaptation and development in the host.
Secondly, functional analysis of the transcripts revealed a large number of genes encoding excretory-secretory proteins, especially families of parasite-derived DNase II and proteases. Previous studies have indicated that families of serine proteases and DNase II of T. spiralis were important parasitic components in host/parasite interactions and modulation of host immune responses [20]–[24]. Serine proteases have been proved to be critical for invasion of the mammalian host cells by trypanosoma cruzi and steinernema carpocapsae [25]–[28]. Here, we found 30 genes coding for homologous serine proteases were differentially expressed in the development of the parasite (Table 2 and Files S4 and S5). With the identification of the stage-specific expression of these proteases, it is now possible to further explore their importance in parasitization and as target for intervention such as vaccine development.
DNase II belongs to the acidic endonuclease family. They are essential nucleases nucleases that exist in non-metazoan, fulfilling a variety of functions from digesting DNA of apoptotic cell corpses and dietary DNA to modulating host immune responses [24]. They possess a histidine-rich domain which is believed to be the functional core of the protein family. Though there is more than one copy of DNase II genes in many organisms, the number of DNase II genes in T. spiralis is an exception with 47 copies identified in the strain studied. Phylogenetic analysis showed that these DNases exist in different subgroups (Figure 3). In light of the findings of their expression variation in different developmental stages, it is postulated that different variants fulfill different functionality. Recent studies suggested that DNase II in T. spiralis may function as self-protective molecules through modulation of host innate immune responses by cleaving DNA from apoptotic host cells [20], [29]. Deep investigation of functional specialty of the DNase II family members will be able to reveal mechanisms of host-parasite interaction and pathogenesis of trichinellosis.
Apart from the above protease and DNase II families, genes encoding heat shock protein 70 (HSP), macrophage migration inhibitory factor (MIF), systeine-glycine protein and cystatin were also found to be expressed in stage-specific manner. These molecules have been previously reported to be important in host/parasite interactions [30]–[35]. It has been proved that the parasites produce HSPs upon heat or oxidative stress that are related to resistance to harsh environment changes and are therefore beneficial to their survival [30]. MIF is a cytokine ubiquitous in mammals. However, many pathogens have genes encoding MIF homologues and the function of pathogen-derived MIF is believed to modulate host immune responses [35]. Thus it is not a surprise that T. spiralis also possesses this gene. Cystatins is a kind of cysteine protease inhibitor and has been reported as an important immuno-modulatory factor that contributes to the immune evasion strategies of the parasites [31].
Zinc-dependent metalloprotease of T. spiralis showed significant homology to the astacin metalloprotease family of Caenorhabditis elegan. Astacin metalloprotease have diverse functions including hydrolysis of extracellular matrix components such as type I collagen [36]. Since the parasites need to degrade the fibrinogen and plasminogen when they invade the epithelial and muscul cells (Ad and ML) or migrate to the small mesenteric veins (NBL), zinc-dependent metalloproteinases might be one of the effecter molecules in the process of tissue invasion. Serpins are a large protein family and can inactivate proteinases by forming complexes with serine proteinase. It has been reported that serpins could inhibit the host immune response and protects many parasites to evade the host immune defense [37]–[39]. Among the developmental stages, NBL is most fragile but is more exposed to the host immune system. That may explain why the genes encoding the zinc-dependent metalloprotease and serpin were up-regulated in NBL. Theses enzymes can help NBL for a more efficient penetration of host defensive barriers and, in the meantime, avoid the immune attack.
SOD and glutathione perxidase are two main important antioxidants. SOD can catalyze the conversion of superoxide anion into hydrogen peroxide and molecular oxygen, while glutathione perxidases catalyze the reduction of hydrogen peroxide along with oxidation of glutathione to glutathione disulfide. Secretion of antioxidant enzymes is believed to protect the parasite from reactive oxygen species which arise from host phagocytes [40]. Like DNase II and serine protease families, a number of genes encoding SOD and glutathione perxidases were identified and some of them showed different expression pattern in various developmental stages of T. spiralis. It is suggested that the DNase II, serine protease, SOD and glutathione perxidase were all contributed to self-protection of the parasite during invasion.
Thirdly, functional annotation of differentially expressed genes has partially revealed biological processes in different developmental stages of the parasite (Figure 5 and File S9). For example, oocyst meiosis is directly involved in the reproduction of the parasites. Therefore, most of differentially expressed genes related to oocyst meiosis were up-regulated in Ad compared to the other two stages, which was likely due to the preparation for the production of eggs. On the other hand, glycolysis/gluconeogenesis, calcium signaling pathway and vascular smooth muscle contraction pathways are mainly involved in energy metabolism, development and structure of tissue regeneration. Thus genes associated with these pathways were found up-regulated in NBL instead of Ad and ML due to the rapid growth.
Previous study indicated that there is a shift from anaerobic to aerobic metabolism in the developmental stage from ML to Ad [41]–[43], whereas no related information about the metabolism of NBL is available. In this study, we analyzed the differential expression pattern of the genes encoding the key enzymes involved in energy metabolisms. Phosphofructokinase, enolase and Pyruvate Kinase are essential enzymes in glycolysis. Genes coding for the three enzymes were up-regulated in NBL which indicated that energy metabolism were more activated in the NBL stage. Lactate dehydrogenase is a critical enzyme in anaerobic metabolism which can catalyze the conversion of pyruvate into lactate. Another key enzyme in anaerobic metabolism is phosphoenolpyruvate carboxykinase which can catalyze the conversion of phosphoenolpyruvate (PEP) into oxaloacetate and the reverse citric acid cycle. The genes encoding the two enzymes were found mainly expressed in NBL. Thus it is likely that T. spiralis adapted mainly an anaerobic metabolism in NBL stage. On the other hand, the expression of genes encoding pyruvate dehydrogenase, citrate synthase and isocitrate dehydrogenase, critical enzymes involved in citric acid cycle in aerobic metabolism, were up-regulated in Ad. No obvious significant differences in the expression of genes involved in the citric acid cycle and anaerobic metabolism between ML and Ad was observed.
Lastly, recent studies have revealed widespread expression of complementary sense-antisense transcript pairs by transcriptome sequencing [14], [44]–[47] and antisense-mediated gene regulation in developmental processes of different organisms have been proposed [48]–[50]. More than 7000 of the protein-coding genes were found bi-directionally transcribed in T. spiralis, accounted for approximately 70% of the identified genes in this study and 45% of all estimated genes in the genome, which was similar to that observed in S. japonicum, mice and human [14], [44], [45], [51]. To our knowledge, this is the first study to observe such abundant antisense transcripts in T. spiralis. Since the antisense transcripts were also polyadenylated, antisense RNAs can encode proteins [45], [51], [52]. However, most antisense transcriptions in the mammalian genome were found to be non-protein-coding RNAs [45], [49], [51]. Though the function of antisense transcripts has not been well understood, functional validation studies indicated that antisense transcripts were a heteromorphous group with common features and may modulate the expression level of the sense transcripts or influence the sense mRNA processing [53]. The mechanism of such regulation remains unknown; however, a mechanism of gene regulation by natural antisense transcripts (NAT) derived endogenous siRNAs (endo-siRNAs) has recently emerged. Endogenous siRNAs derived from transposable elements, NAT and long hairpin RNAs have been identified in Drosophila and mouse and S. japonicum. Endo-siRNAs can silence homologous transcripts by RNA interference (RNAi). Therefore, it is likely that natural antisense transcripts as sources of endo-siRNAs possess similar regulatory function in T. spiralis as in other organisms [13].
In summary, approximately 65% of genes in T. spiralis genome were identified and a large number of functionally interesting genes were discovered and analyzed in the three developmental stages of the parasite through high through-put RNA sequencing techniques. More than 45% of the protein-coding genes showed evidence of transcription from both sense and antisense strands. The data from this study has paved a way for deep investigation of the parasite biology and host parasite interaction.
|
10.1371/journal.pgen.1001341 | Macoilin, a Conserved Nervous System–Specific ER Membrane Protein That Regulates Neuronal Excitability | Genome sequence comparisons have highlighted many novel gene families that are conserved across animal phyla but whose biological function is unknown. Here, we functionally characterize a member of one such family, the macoilins. Macoilins are characterized by several highly conserved predicted transmembrane domains towards the N-terminus and by coiled-coil regions C-terminally. They are found throughout Eumetazoa but not in other organisms. Mutants for the single Caenorhabditis elegans macoilin, maco-1, exhibit a constellation of behavioral phenotypes, including defects in aggregation, O2 responses, and swimming. MACO-1 protein is expressed broadly and specifically in the nervous system and localizes to the rough endoplasmic reticulum; it is excluded from dendrites and axons. Apart from subtle synapse defects, nervous system development appears wild-type in maco-1 mutants. However, maco-1 animals are resistant to the cholinesterase inhibitor aldicarb and sensitive to levamisole, suggesting pre-synaptic defects. Using in vivo imaging, we show that macoilin is required to evoke Ca2+ transients, at least in some neurons: in maco-1 mutants the O2-sensing neuron PQR is unable to generate a Ca2+ response to a rise in O2. By genetically disrupting neurotransmission, we show that pre-synaptic input is not necessary for PQR to respond to O2, indicating that the response is mediated by cell-intrinsic sensory transduction and amplification. Disrupting the sodium leak channels NCA-1/NCA-2, or the N-,P/Q,R-type voltage-gated Ca2+ channels, also fails to disrupt Ca2+ responses in the PQR cell body to O2 stimuli. By contrast, mutations in egl-19, which encodes the only Caenorhabditis elegans L-type voltage-gated Ca2+ channel α1 subunit, recapitulate the Ca2+ response defect we see in maco-1 mutants, although we do not see defects in localization of EGL-19. Together, our data suggest that macoilin acts in the ER to regulate assembly or traffic of ion channels or ion channel regulators.
| The human genome project has given us a catalog of the genes that make a human; however, the function of about 40% of these genes remains elusive. Many of these mysterious genes have relatives in simpler organisms like worms and flies, where their function can be studied much more easily than in a mammal. Here, we investigate one such family of genes, called macoilins, using the worm C. elegans. We show that worm macoilin, like mouse macoilin, is expressed widely but specifically in nerve cells. We create worms in which the macoilin gene is defective and show that, although they retain a nervous system that looks normal, they have behavioral defects. We show that these behavioral defects reflect an inability of nerves to signal efficiently. Nerve signalling relies on calcium channels and the defect of macoilin mutants resembles that of animals defective in a particular calcium channel component. We find that in nerve cells the macoilin protein resides specifically in the “factory” that assembles nerve signalling molecules, including calcium channels. Our results suggest that macoilin either directly helps assemble an ion channel or is needed to make a channel regulator. Our work in worms provides a blueprint to investigate the function of macoilins in mammals.
| One of the most striking innovations in Metazoa is a nervous system with specialized nerve cells, pre- and post-synaptic structures, and associated signaling molecules. Neuronal signaling depends on complexes of multipass transmembrane proteins such as ion channels and G-protein-coupled receptors. Over the past few years several studies have identified specialized molecular machines in the endoplasmic reticulum by which particular complexes are assembled with appropriate stoichiometries and trafficked to their destination [1]. The emerging picture is that neurons have a highly specialized endoplasmic reticulum (ER), allowing channels to undergo quality control prior to export. However the identity of such maturation complexes remains unclear except for a handful of channels.
The extensive intracellular membrane system that makes up the ER varies, depending on cell type, but two domains, the rough and smooth ER (RER and SER), can usually be distinguished. The RER is studded with ribosomes and mediates translocation of secretory proteins across the membrane and insertion of membrane proteins. The SER is implicated in lipid synthesis and regulation of Ca2+ storage and signalling. Whereas most ER proteins are found in both RER and SER, a subset of proteins involved in translocation of newly synthesized proteins across the ER membrane is highly concentrated in the RER [2], [3]. In C. elegans neurons, RER proteins are concentrated in the cell body and excluded from dendrites and axons, whereas general ER proteins are found in both cell body and neurites [4]. Electron microscopy confirms that ribosomes and RER are abundant in the cell body of C. elegans neurons but rare in neurites, whereas smooth ER-like structures can be seen in axons and dendrites as well as the cell body.
Over the last decade, genome sequencing projects have provided gene catalogs for animals representing a spectrum of metazoan phyla, including Placozoa [5], Cnidaria [6], Echinodermata [7], Annelida (http://genome.jgi-psf.org/Capca1/Capca1.home.html) and Chordata [8]. Genome-wide comparisons have identified human genes that are conserved across these animal phyla, and highlighted their shared structural features. However the biological function of many of these conserved gene families remains mysterious. Genetic studies in model organisms such as Drosophila and C. elegans have provided a powerful way to functionally characterize novel conserved genes. This is exemplified by discovery of ion channel families, e.g. TRP [9], axon guidance pathways (UNC-6/netrin; UNC-40/DCC; ROBO [10]) and molecules involved in synaptic release (e.g. UNC-13 [11] and UNC-18 [12]). In all these cases genetic studies in flies or worms were recapitulated in mammals and catalyzed subsequent vertebrate work.
Here, we functionally characterize, for the first time, a member of a conserved family of proteins called macoilins. We find a macoilin gene in all available genome sequences of animals, from placozoa to man, but not in yeast or Dictyostelium. C. elegans macoilin, like mouse macoilin [13], is expressed throughout the nervous system. In C. elegans expression begins embryonically at the time neurons are born, and persists to adulthood. Using antibodies and compartment specific markers we show that C. elegans macoilin is localized to the rough endoplasmic reticulum and is excluded from neurites. We identify multiple C. elegans mutants of macoilin and show that these have altered behavior, but normal development of the nervous system. Using Ca2+ imaging, we show that macoilin is required for cell-intrinsic neuronal excitability in the O2-sensing neuron PQR. This phenotype is mirrored in animals defective in EGL-19, the sole C. elegans L-type voltage gated ion channel (L-VGCC) alpha1 subunit. Our data suggest that macoilin is involved in assembly or traffic of ion channels or ion channel regulators.
N2, the laboratory wild-type strain of C. elegans, feeds in isolation; however most wild-collected strains of this species feed in groups [14]–[15]. N2 animals fail to aggregate because of a gain-of-function mutation in the neuropeptide receptor npr-1: if this receptor is knocked out they aggregate strongly [14][16]. To define genes that promote aggregation we mutagenized npr-1(null) animals and sought non-aggregating mutants. One complementation group we identified comprised three recessive alleles, db1, db9 and db129, and defined a gene we called maco-1 (for macoilin-1 see below). All three maco-1 mutations strongly suppressed aggregation and bordering behaviors (Figure 1A) and disrupted the ability of npr-1 animals to switch between roaming widely and dwelling locally according to ambient O2 levels (Figure 1B) [17]. maco-1; npr-1 mutants were healthy and displayed strong attraction to diacetyl and benzaldehyde (Figure S1A, S1B), odorants that are detected by the AWA and AWC olfactory neurons respectively [18]. They also strongly avoided high osmotic potential, a response mediated by ASH nociceptive neurons (Figure S1C) [19]. Like N2 animals, maco-1 single mutants did not aggregate and only showed weak bordering behavior (Figure S1D). Closer examination, however, revealed additional behavioral phenotypes associated with maco-1 mutations. We quantified these phenotypes in the presence of the N2 allele of npr-1, since this is the standard genetic background. First, harsh touch to the head elicited significantly longer reversals in maco-1 mutants compared to N2 controls (Figure 1C). Second, maco-1 mutants exhibited swimming defects [20] characterized by a decrease in the frequency of body bends (Figure 1D), an increase in the amplitude of body bends (data not shown), and coiling (Figure 1E). maco-1 mutants also showed increased coiling when on an agar substrate (data not shown). Interestingly, the maco-1 locomotory defects increased with age (Figure S1E, S1F). Third, whereas N2 animals suppress egg-laying in the absence of food [21], this inhibition was partly relieved in db9 and db129 mutant animals (Figure 1F). These subtle but pleiotropic defects of maco-1 mutants suggest deficits in multiple neural circuits.
We mapped maco-1 to a 30 kb interval on Chromosome I, close to D2092.5, a previously uncharacterized gene. DNA sequencing revealed that all three maco-1 alleles disrupted D2092.5 (Figure 2A). The db1 allele modified the splice donor site of intron 9 (G6024A); db9 changed the arginine codon at position 534 to a stop codon and is predicted to truncate the MACO-1 protein (Figure 2D); the db129 allele was associated with a mutation in the splice acceptor site of intron 2 (G595A). Together, these data suggest that maco-1 corresponds to D2092.5. This was confirmed by transgenic rescue of the maco-1 mutant phenotypes with a wild-type D2092.5 transgene (Figure S2A–S2D). None of our maco-1 alleles were unambiguously null mutants. However the premature stop codon associated with the db9 allele would be expected to cause nonsense mediated degradation of maco-1 mRNA, as well as truncating half the MACO-1 protein, and therefore to be a strong loss-of-function mutation. Consistent with this, the phenotypes of maco-1(db9)/maco-1(db9) and maco-1(db9)/qDf16 were similar (Figure S1G); qDf16 is a large deletion that spans the maco-1 interval (see Materials and Methods).
cDNA analyses indicated that the D2092.5 gene can encode two splice isoforms by alternate splicing at exon 5 (Figure 2A). The proteins encoded by the resulting mRNAs were 908 (MACO-1a) and 897 (MACO-1b) amino acids long. Blast searches identified these proteins as homologues of vertebrate macoilins. Reciprocally, searching the C. elegans genome with vertebrate macoilins identified only one homologue, D2092.5 (Figure 2D). At least one macoilin can be found in every animal genome sequenced so far (Figure 2B, 2C). In some fish lineages but not in other vertebrates, a second MACO homologue can be found (MACO-2); MACO-2 probably arose during the genome duplication event that is thought to have occurred in teleost fish. Despite this ubiquity, little is known about this protein family. The only macoilin previously investigated is the mouse homologue, highlighted because it is expressed differentially between wild-type mice and reeler mutants [22],[13]. In situ hybridization indicates that mouse macoilin mRNA is highly expressed in all neuronal differentiation fields from embryonic stage E12.5 to birth [13]. After birth (PG10), expression decreases but remains associated with some neurons such as cerebellar granule cells, olfactory mitral and granule cells, pyramidal neurons in the hippocampus, and granule cells in dentate gyrus [13]. No significant expression of mouse macoilin has been reported outside the nervous system.
Comparison of macoilin sequences across phyla highlighted several common features (Figure 2D). All MACO-1 homologs were predicted to have at least three transmembrane domains towards the N-terminus and at least two coiled-coil domains towards the C-terminus. The exact number of transmembrane and coiled-coil domains varied depending on the prediction algorithm and species (data not shown). At the sequence level the transmembrane and coiled-coil domains were the most conserved parts of the protein (Figure 2D).
To determine where maco-1 was expressed, we created transgenic C. elegans that co-expressed maco-1 and gfp coding sequences as a polycistronic message from the maco-1 promoter. These animals first showed GFP fluorescence 6 hours after the first cell division (Figure S3A). By late embryogenesis and in the L1 larva, fluorescence was visible throughout the nervous system (Figure S3A). In adults, the GFP signal remained pan-neuronal, with very occasional expression in other tissues (Figure S3B). Thus, C. elegans MACO-1, like its mouse homologue, is expressed widely and almost exclusively in the nervous system.
To study endogenous MACO-1, we raised polyclonal antibodies against the C-terminus of both its protein isoforms. N2 worms stained with the antibody showed bright expression in most or all C. elegans neurons (Figure 3A–3C and Figure S4A). In contrast, maco-1(db9) mutants which have a nonsense mutation that truncates MACO-1 upstream of the epitope sequence showed no signal in the nervous system (Figure 3D, 3E, and Figure S4B, S4C). Interestingly, MACO-1 antibody staining was restricted to the neuronal cell body: no staining was observed in dendrites, in the synapse-rich axon bundles that make up the nerve ring, or in the axons comprising the ventral and dorsal cords (Figure 3A, 3B). To confirm that MACO-1 was absent from synapses, we co-stained worms expressing the synaptic marker SNB-1-GFP in GABAergic neurons (juIs1) with anti-MACO-1 and anti-GFP antibodies: the two markers did not show co-localization (Figure 4A–4D). Moreover, worms co-expressing juIs1 and psnb-1::maco-1-mcherry showed that transgenic MACO-1-cherry was also restricted to the cell body of neurons, overlapping only with the SNB-1-GFP signal in the cell body of GABAergic neurons (Figure 4N). Both transgenic and endogenous MACO-1 was also found restricted to neuronal cell bodies at embryonic developmental stages (Figure S5).
The striking localization of MACO-1 to the cell body raised the possibility that it resides in a specific membrane compartment. To investigate this we created different transgenic lines that expressed organelle-specific markers in a subset of neurons, using the glr-1 promoter (glr = glutamate receptor) (Figure 4E–4M; Figure S6). The markers used (a kind gift of M. M. Rolls, Pennsylvania State University) were phosphatidylinositol synthase (PIS) for the general endoplasmic reticulum (ER), translocating chain-associating membrane protein (TRAM) for rough ER, Emerin for the nuclear envelope, and Mannosidase (MANS) for the Golgi. All markers were tagged at the N-terminus with YFP. Co-immunostaining of MACO-1 and YFP revealed partial co-localization between MACO-1 and the ER general marker, YFP-PIS (Figure 4E–4I). Within the ER, MACO-1 was further co-localized with a marker restricted to the rough ER, YFP-TRAM, and nuclear envelope, YFP-Emerin (Figure 4J-4M and Figure S6A–S6D). Similar ER localization was seen in embryonic stages (Figure S6). No significant co-localization was observed with the Golgi-specific marker, YFP-MANS (Figure S6E–S6H; however the cell bodies of C. elegans neurons are small (2 microns), making it difficult to exclude the possibility that there is a low amount of MACO-1 in Golgi. These co-localization experiments suggest that MACO-1 predominantly resides in the ER of neurons, in particular in rough ER and nuclear envelope. We observed no gross morphological defects in maco-1 worms in any of the sub-cellular compartments expressing the YFP tagged constructs described above (data not shown).
The broad neuronal expression patterns of C. elegans and mouse macoilins suggest that this protein family has a general role in the development or function of the nervous system. To elucidate this role, we first examined the anatomy of the nervous system in maco-1(db9) mutants using neuron-specific GFP reporters. We examined mechanosensory neurons, chemosensory neurons, and GABAergic motor neurons. We detected no overt abnormalities in the cell bodies, axons, dendrites or cilia of any neuron we examined (Figure S7A–S7C and data not shown). We next asked if maco-1 regulates neuronal polarity (i.e. the placement of synapses) or axonal trafficking of synaptic vesicles. To test this we visualized synaptic vesicles in live animals using a fluorescently-tagged synaptic vesicle marker, synaptobrevin-GFP (SNB-1::GFP). We saw no defects either in the GABAergic DD motor neurons or in the URX O2-sensing neurons (Figure S7D–S7G and data not shown). These results suggest that MACO-1 is not required for correct establishment or maintenance of neuron polarity.
Precursor vesicles containing synaptic proteins are generated at the cell body and transported to synapses by microtubule-based motor proteins [23]. In C. elegans, this transport requires the KIF1A kinesin homologue unc-104 [23]. In unc-104 mutants tagged synaptobrevin expressed from the punc-25::snb-1::gfp transgene is retained in the cell bodies of the DD and VD motor neurons [24]. Macoilins have been proposed to function in axonal traffic [13]; however, in adult maco-1(db9) worms expressing the punc-25::SNB-1::GFP marker, the tagged synaptobrevin was still localized along the ventral and dorsal nerve cords (Figure 6A and 6C). This suggests that MACO-1 is not essential for transport of synaptic vesicles.
C. elegans synaptic function can be assayed by studying responses to the acetylcholinesterase inhibitor aldicarb and the acetylcholine receptor agonist levamisole [20], [25]. Aldicarb inhibits acetylcholinesterase (AChE), leading to accumulation of acetylcholine at the neuromuscular junction (NMJ), overstimulation of acetylcholine receptors, and paralysis of wild-type animals. Mutants defective in synaptic release have reduced acetylcholine accumulation and are therefore resistant to aldicarb [20]. However these mutants retain sensitivity to levamisole, which directly activates post-synaptic acetylcholine receptors. In contrast, mutants defective in postsynaptic responses to ACh are resistant to both aldicarb and levamisole [26]. maco-1(db9) mutants were resistant to aldicarb but sensitive to levamisole (Figure 5A, 5B), suggesting they have presynaptic defects in neurotransmission.
The aldicarb resistance of maco-1 mutants prompted us to examine synapse structure in these animals more closely. GABAergic type D motor neurons form neuromuscular junctions (NMJs) with ventral and dorsal body wall muscles [27]. We visualized the presynaptic terminals of these neurons using the punc-25::snb-1::gfp transgene juls1 [28]. Wild-type animals bearing juIs1 have SNB-1::GFP puncta of uniform shape and size distributed evenly along the ventral and dorsal nerve cord. These puncta correspond to the presynaptic termini of the 13 VD and 6 DD neurons, respectively. We measured puncta size and number along a 100 µm section of the ventral nerve cord (Figure 6). In wild-type animals, average puncta area in the ventral nerve cord was 1.58±0.13 µm2, with an average of 23.28±0.79 puncta per 100 µm (n = 22 animals). maco-1 mutants had fewer puncta that tended to be larger: the average size in the ventral nerve cord was 2.62±0.54 µm2 with an average of 19.87±1.18 µm puncta per 100 µm (n = 23) (Figure 6A–6D). These data suggest that maco-1 influences pre-synaptic structure.
We next investigated whether maco-1 mutants exhibit active zone defects, using SYD-2::GFP [29] and UNC-10::GFP [30] as markers (Figure 6E–6L). Both fusion proteins were expressed in the GABAergic VD and DD motorneurones from the unc-25 promoter. Wild-type animals carrying the punc-25::syd-2::gfp transgene hpIs3 have regularly-sized and spaced puncta along the dorsal and ventral nerve cords, with an average punctal area of 0.32±0.017 µm2 and on average 36.78±1.96 (n = 17) puncta per 100 µm. In maco-1(db9) adult animals the number of SYD-2::GFP puncta along the dorsal nerve cord was increased but their size was similar to wild-type (Figure 6I–6L). The average area of puncta in maco-1 mutants was 0.31±0.014 µm2, with an average of 45.44±1.51 (n = 23) puncta per 100 µm. Wild-type animals carrying the punc-25::unc-10::gfp transgene hpIs61 have uniformly shaped and evenly distributed puncta along the ventral and dorsal cord. In maco-1(db9) adult animals, there was a reduced number of puncta but they were larger than in wild-type animals in the ventral nerve cord. The average punctum area in the ventral nerve cord for wild-type animals was 0.55±0.035 µm2, with an average of 35.16±0.99 (n = 13) puncta per 100 µm. Average puncta size in maco-1 mutants was 0.85±0.12 µm2, and the average number of puncta was 28.12±2.29 (n = 12) per 100 µm (Figure 6E–6H). Together these data suggest that loss of maco-1 alters the structure of the synaptic active zone, but the effects are subtle.
We also examined the periactive zone, using the marker RPM-1::GFP [31]. This region just surrounds active zones and has been proposed to regulate synapse growth [32]. We found a slight increase in the size and number of puncta in maco-1 mutants (0.75±0.068 µm2 and 37.24±3.376, n = 10), compared to wild-type animals (0.63±0.036 µm2 and 33.09±2.7, n = 16). However, these differences were not significant (p = 0.1 and p = 0.34, respectively), suggesting that the periactive zone was not disorganized in maco-1 mutants.
Synapse development can be influenced by neural activity [33]. This prompted us to explore if the subtle synaptic defects in maco-1 reflected altered neuronal excitability. We focused our analyses on the O2-sensing neuron PQR, since a subset of the phenotypes of maco-1 mutants resembled those associated with loss of O2 sensing neurons [34]. To image Ca2+ transients we used the cameleon reporter YC3.60 [35] expressed from the gcy-32 promoter [15]. Baseline Ca2+ levels in 7% O2 were similar in wild type and maco-1 mutants, suggesting that PQR neurons were not chronically depolarized in maco-1 mutants. However the Ca2+ rise seen in wild type when O2 is raised to 21% was absent in most maco-1 mutant animals (Figure 7A). These data suggest that maco-1 is required for efficient activation of PQR neurons in response to a rise in O2.
To explore this further we first asked if pre-synaptic input was required for PQR neurons to respond to O2 stimuli. Null mutations in unc-13 or unc-31 CAPS, which disrupt release of synaptic vesicles and dense core vesicles respectively, did not significantly alter Ca2+ transients in PQR to a 7 – 21% O2 upstep or a 21 to 7% downstep (data not shown). This suggests that Ca2+ fluxes in PQR reflect cell-intrinsic responses to the O2 stimuli, and that loss of maco-1 disrupts either primary sensory transduction of ambient O2 or amplification of the sensory potential. O2-stimulated Ca2+ influx in PQR requires the atypical soluble guanylate cyclases GCY-35 and GCY-36. These soluble guanylate cyclases are themselves O2 sensors and activate a cGMP-gated ion channel [36]-[37]. Consistent with this, in a separate study we have shown that a rise in O2 stimulates a rise in cGMP in PQR neurons (A.C. and M.dB, in preparation). Mutations in maco-1 did not alter PQR cGMP responses to an O2 stimulus, suggesting that O2 sensing by GCY-35/36 was unaffected (A.C. and M.dB, in preparation).
Previous work has shown that tax-4, which encodes a cGMP-gated cation channel alpha subunit is required for the O2-evoked Ca2+ transients in PQR [15]. To explore how depolarization evoked by the cGMP channel is amplified and leads to Ca2+ influx in the cell body we imaged PQR responses to O2 stimuli in animals defective in various ion channels. The C. elegans genome does not appear to encode voltage-gated sodium channels. Instead, electrical signals are thought to propagate via voltage-gated Ca2+ channels and cation leak channels [38], [39]. C. elegans encodes 3 voltage gated Ca2+ channel α1 subunits: egl-19 (CaV1, L-type), unc-2 (CaV2, N-, P/Q, R-type) and cca-1 (CaV3, T-type) [40]. It also encodes 2 homologs of the vertebrate cation leak channel NALCN that regulates neuronal excitability [38]. Animals mutant for the UNC-2 P/Q-like voltage-gated Ca2+ channel (VGCC) [41], the T-type channel CCA-1 [42], or double mutant for the NCA-1and NCA-2 NALCN-like leak channels [38] showed overtly wild-type Ca2+ transients in the cell body of PQR in response to a 7 to 21% O2 shift (Figure 7C and data not shown). This is consistent with previous results in other neurons that suggest these channels contribute to Ca2+ signals at synapses and axons, but are not essential for Ca2+ changes in the cell body [38], [39]. In contrast, animals with partial loss-of-function mutations in the EGL-19 L-type VGCC showed frequent failure of cell body Ca2+ transients (Figure 7B). L-type VGCCs have previously been shown to contribute to dendritic Ca2+ currents both in C. elegans [39] and vertebrates [43]. Consistent with these imaging results, egl-19(ad1006); npr-1(ad609) double mutants animals failed to aggregate.
Together, our Ca2+ imaging results suggest that MACO-1 acts in the endoplasmic reticulum to promote assembly and/or traffic of either a cGMP-gated cation channel that contains the TAX-4 alpha subunit, or of an L-type Ca2+ channel containing the EGL-19 α1 subunit, or of another as yet unknown regulator that modulates O2-evoked Ca2+ entry into PQR. To investigate the first possibility we made transgenic animals that expressed a functional GFP-tagged TAX-4 protein in the AQR, PQR and URX neurons, and compared the localization of this channel subunit in npr-1 and maco-1; npr-1 mutant animals. We saw enrichment of TAX-4-GFP in the sensory endings of the O2-sensing neurons, as expected for a sensory transduction channel (Figure S8). We also observed TAX-4-GFP in the cell body and on axons and dendrites. However we found no effect of loss-of-function mutations in maco-1 on this distribution pattern (Figure S8 and data not shown). These data suggest maco-1 is not required for TAX-4 to be exported from the ER, although they do not rule out a potential role in the function of a TAX-4-containing channel. Next, we transgenically expressed EGL-19 protein that is N-terminally tagged with GFP from its endogenous promoter, and examined its localization in wild type and maco-1 mutant animals. As expected, GFP-EGL-19 was expressed very broadly, and both in muscles and neurons (Figure S9) [44]. In neurons GFP-EGL-19 was enriched in sensory endings and in cell bodies. However we did not see any striking defects in the EGL-19 localization pattern in maco-1 mutants (Figure S9). This does not rule out that MACO-1 modulates the function of an EGL-19-containing channel, but it does suggest that if it has a role it involves only a subset of EGL-19-containing channels; alternatively maco-1 regulates function of other, as yet unknown, ion channels.
Macoilins are a conserved family of multipass transmembrane proteins whose function has been mysterious. Members of the family can be found in eukaryotes that have a recognizable nervous system, from placozoa to humans, but not in yeast or Dictyostelium. Macoilins are expressed broadly but specifically in the nervous system. C. elegans macoilin is absent from neurites and is localized to the RER suggesting that it is involved in folding, assembly, or traffic of secreted or transmembrane proteins. The structure of macoilin contains two conserved regions: an N-terminal part that includes multiple transmembrane domains, and a C-terminal region that has coiled coil domains; the transmembrane domains are the most highly conserved parts of the protein. This combination of structural motifs is reminiscent of that of RIC-3 and its orthologues, which are implicated in assembly and traffic of nicotinic acetylcholine receptors in C. elegans [45] and of nicotinic acetylcholine receptors and 5-HT3 receptors in vertebrates [46]–[47]. Like macoilin, RIC-3 is expressed broadly in the nervous system and is an ER membrane protein with a coiled-coil region towards the C-terminus. Macoilin mutants exhibit defects in cell intrinsic neuronal excitability, not only in PQR (this study) but also in other neurons (see associated paper by Miyara et al.). Previous work has reported that neural activity levels regulate the morphology of certain synaptic connections in C. elegans [33]; the synaptic morphology defects of maco-1 mutants could therefore reflect loss of neuronal excitability. A simple hypothesis is that macoilin acts in the endoplasmic reticulum of neurons to promote the folding, assembly, or traffic of ion channels or ion channel regulators that control excitability of neurons. What might these targets be? Since baseline Ca2+ levels are normal in maco-1 mutants we do not think loss of macoilin disrupts function of ion pumps that keep neurons hyperpolarized. Instead our data point towards compromised signal transduction or signal amplification downstream of the GCY-35/GCY-36 O2-sensing soluble guanylate cyclases. As far as we can tell the cGMP-gated ion channel that transduces the O2-evoked cGMP rise in PQR, and which includes the TAX-4 α subunit, is appropriately expressed and localized in maco-1 mutants, although we cannot exclude the possibility that its function is somehow compromised. cGMP channels are expressed in only a small subset of C. elegans neurons [48] and some of these are clearly functional in maco-1 mutants (e.g. AWC); by contrast MACO-1 is expressed throughout the nervous system, not only in C. elegans but also in mouse.
The L-type VGCC α1 subunit EGL-19 is also required for O2-evoked responses in PQR, and is expressed widely both in the nervous system and in muscle (this work; [44]. Loss-of-function mutants of egl-19 have much more severe phenotypes than maco-1 mutants: egl-19(null) mutants die as embryos. This discrepancy in phenotype makes it unlikely that MACO-1 is critical for function of all EGL-19 containing channels. Consistent with this, mutations in maco-1 do not appear to disrupt localization of GFP- EGL-19 either in muscle or in neurons. However it remains possible that MACO-1 regulates assembly of particular subtypes of EGL-19–containing channels. L-type VGCC are composed of multiple subunits and it is the precise combination of subunits that determines the channel's regulatory properties; additionally egl-19 mRNA itself is alternatively spliced close to its C-terminus, in a region implicated in Ca2+ feedback regulation [49]. An alternative scenario is that MACO-1 regulates an as yet undiscovered pathway that helps amplify the depolarization initiated by cGMP-gated ion channel activation. Identifying proteins that interact with macoilin or mutants that recapitulate the maco-1 phenotypes will allow these hypotheses to be tested to help further unravel the function of this novel family of nervous system proteins.
Strains used were maintained as described previously [50] and are listed in Text S1). db1 and db9 were isolated as suppressors of aggregation from a screen of 20,000 haploid genomes; details of the screen will be described elsewhere. db129 was isolated in a non-complementation screen using the db1 allele. The db1 mutation was mapped to a 30 kb interval at the centre of Chromosome I between the SNP markers in cosmids F48A9 and D2092 using a combination of three-factor mapping and SNP genotyping [51].
PSI-Blast was used to search for Macoilin protein sequences, using human Macoilin as probe, at the NCBI (www.ncbi.nlm.nih.gov), Joint Genome Institute (www.jgi.doe.gov), ENSEMBL (www.ensembl.org), and the Sanger Institute (www.sanger.ac.uk). From approximately 100 sequences retrieved (e-value > e-10), a subset was obtained after removing splice variants, and redundant sequences. The amino acid sequences were aligned using various programs run under the umbrella of the M-Coffee server [52]. The multiple alignment was visually inspected and curated using BioEdit [53] (Figure S10). Un-rooted phylogenetic trees were generated using a Neighbour-Joining method [54]; the robustness of the nodes was verified with 10000 bootstrap replicates using the program Phylo-Win [55].
The cDNA sequence of maco-1 was obtained by sequencing clone yk1296a05 from the Kohara collection and by using RT-PCR. Briefly, N2 mixed stage animals were extracted with Trizol, and 1–5 µg of purified total RNA reverse transcribed using an oligo-d(T) primer and SUPER RTase at 42 °C for 1 h. Primers specific for maco-1 exons and the SL1 spliced leader sequence were used to amplify maco-1 cDNA, and the PCR products sequenced. The maco-1 expression construct was generated using the Gateway system (Invitrogen) [56]. The Destination vector included 4 kb of the sequence upstream of the maco-1 start site but omitted 318 bp between the trans-splice site and the initiation codon. The Entry vector places maco-1 cDNA plus 9 bp of the sequence upstream in an artificial operon with gfp [34]. This construct was sequenced and injected at 50 ng/µl with lin-15(+) as the co-injection marker.
Transgenic rescue: The fosmid WRM0640bE08, containing D2092.5, was injected into the strain AX129, maco-1(db9);npr-1(ad609) at a concentration of 2 ng/µl, with 50 ng/µl of punc-122::gfp as a co-injection marker. Further transgenic rescue experiments were carried out using a PCR amplified genomic DNA fragment containing the D2092.5 gene, including 4 kb upstream of the initiation codon and 1 kb after the stop codon. This PCR product was injected into the AX59, maco-1(db9); npr-1(ad609) strain at a concentration of 2 ng/µl, using pmyo-2::gfp as a co-injection marker (4 ng/µl) and 1 kb-ladder (96 ng/µl) as carrier.
Sub-cellular markers (a kind gift of Melissa M. Rolls, Penn State University) were used as described [4]. The plasmids used were C24F3.1a (pglr-1::yfp-TRAM), Y46G5a.5 (pglr-1::yfp-PIS), F558H1.1 (pglr-1::yfp-MANS) and M01D7.6 (pglr-1::yfp-EMERIN). These plasmids were individually injected at 4 ng/µl into AX206, lin-15(n765ts) animals with a lin-15(+) co-injection marker (40 ng/µl). All primer sequences used are available upon request.
The pgcy-32::tax-4-gfp transgene was made using Gateway; tax-4 cDNA was tagged at the 3'end with gfp and injected (at 10 ng/ul) with a lin-15(+) co-injection marker (40 ng/ul) into npr-1(ad609) lin-15(n765ts) animals. A fosmid containing the full-length egl-19 gene was modified by recombineering so as to express N-terminally GFP tagged EGL-19 from its endogenous control sequences. The recombineered fosmid was co-injected at 5 ng /ul with a pgcy-32::mcherry co-injection marker (25 ng/ul) and carrier DNA (DNA ladder at 70 ng/ul).
Mix-staged worms were stained following the modified Ruvkun and Finney method [57]. Primary antibodies were used at dilutions of 1/50 and 1/500 for mouse monoclonal anti-GFP antibody (clones 7.1 and 13.1; Roche, Germany) and rabbit polyclonal anti-MCL, respectively. Antibodies were incubated at 4 °C for 16 hrs with gentle mixing. The secondary antibodies Alexa Fluor 546 nm goat anti-rat IgG (H+L) (Invitrogen, UK) and AlexaFluor 488 nm goat anti-mouse (Invitrogen, UK) were used at a final dilution of 1/500 (4 mg/ml) and 1/250 (10 mg/ml), respectively; DAPI was added to a final concentration of 5 mM. After an incubation of 2 hrs at room temperature, worms were thoroughly washed with AbB Buffer (the details of buffer composition can be found in the Text S1), mounted in agarose and imaged.
Live animals were anaesthetized with 10 mM sodium azide, mounted on 2 % agarose pads, and examined under epifluorescence using a Zeiss Axioskop fluorescent microscope. Confocal images were taken using a Radiance Plus Confocal Scanning System (Bio-Rad). The images were processed and analyzed with LaserSharp2000 software (Bio-Rad). The different GFP markers were visualized in the different backgrounds as described previously [29],[58]. Measurements of GFP puncta were performed on confocal images. Briefly, confocal images were projected into a single plane using the maximum projection method and exported as a tiff file with a scale bar. Fluorescence intensity, number of puncta, total fluorescence and punctum area were measured in ImageJ. These numbers were exported to Microsoft Excel for statistical analyses using Student's two-tailed t-test.
Aggregation assays were done as previously described [14]. Egg-laying assays followed [59] with the following modifications. Worms were synchronized and young adults picked to unseeded plates to remove adhering food before transfer to plates seeded with 50 µl of E. coli OP50 or to un-seeded plates. Rings of 100 µl of 4M D-Fructose were painted on no-food plates to trap animals. Worms were left for one hour, then removed and eggs counted. Plates in which worms could not be found were discarded (around 10% in plates without food). Harsh touch was assayed by poking animals with a platinum wire pick. To analyze swimming defects, single worms were transferred to M9 media and left to equilibrate for a minute; head swings were counted during 10 second intervals; for coiling we counted the number of times the worm's nose touched the mid body in one minute. The results of the behavioral assays were analyzed using a two-tailed t-test. Aldicarb and levamisole assays were done as described [25]. Briefly, sensitivity to 1 mM aldicarb (Chem Services) or 0.4 mM levamisole was determined by assaying the time course of the onset of paralysis following acute exposure of a population of animals to these drugs. In each experiment, 25 worms were placed on drug plates and prodded every 10 min over a 2 h period to determine if they retained the ability to move. Worms that failed to respond at all to the harsh touch were classified as paralyzed. Each experiment was repeated five times.
Ca2+ responses of PQR neurons to O2 stimuli were imaged as described previously (Persson et al 2009) on an inverted microscope (Axiovert, Zeiss), using a 40× C Apochromat lens and Metamorph acquisition software (Molecular Devices). To measure Ca2+ we used the ratiometric FRET sensor YC3.60 [35]. Briefly, worms were glued to agarose pads using Nexaband glue (WPI Inc) and placed under the stem of a Y-chamber microfluidic device. Photobleaching was limited by using a 2.0 optical-density filter and a shutter to limit exposure time to 100 ms per frame. An excitation filter (Chroma) restricted illumination to the cyan channel. A beam splitter (Optical Insights) was used to seperate the cyan and yellow emission light. The ratio of the background-subtracted fluorescence in the CFP and YFP channels was calculated with Jmalyze [60]. Fluorescence ratio (YFP/CFP) plots and measurements of mean baseline ratios and mean peak ratios were made in Matlab (The MathWorks). Movies were captured at 2 frames per second. Average Ca2+ traces were compiled from at least six recordings per condition made across two or more days.
Whenever the data fitted a normal distribution (p<0.05, Kolmogorov-Smirnov) a two-sample (unranked) t-test was used. For non-normally distributed data, the non-parametric Kolmogorov-Smirnov (K-S) -test was used.
|
10.1371/journal.pntd.0006619 | Animal influence on water, sanitation and hygiene measures for zoonosis control at the household level: A systematic literature review | Neglected zoonotic diseases (NZDs) have a significant impact on the livelihoods of the world’s poorest populations, which often lack access to basic services. Water, sanitation and hygiene (WASH) programmes are included among the key strategies for achieving the World Health Organization’s 2020 Roadmap for Implementation for control of Neglected Tropical Diseases (NTDs). There exists a lack of knowledge regarding the effect of animals on the effectiveness of WASH measures.
This review looked to identify how animal presence in the household influences the effectiveness of water, hygiene and sanitation measures for zoonotic disease control in low and middle income countries; to identify gaps of knowledge regarding this topic based on the amount and type of studies looking at this particular interaction.
Studies from three databases (Medline, Web of Science and Global Health) were screened through various stages. Selected articles were required to show burden of one or more zoonotic diseases, an animal component and a WASH component. Selected articles were analysed. A narrative synthesis was chosen for the review.
Only two studies out of 7588 met the inclusion criteria. The studies exemplified how direct or indirect contact between animals and humans within the household can influence the effectiveness of WASH interventions. The analysis also shows the challenges faced by the scientific community to isolate and depict this particular interaction.
The dearth of studies examining animal-WASH interactions is explained by the difficulties associated with studying environmental interventions and the lack of collaboration between the WASH and Veterinary Public Health research communities. Further tailored research under a holistic One Health approach will be required in order to meet the goals set in the NTDs Roadmap and the 2030 Agenda for Sustainable Development.
| Neglected Tropical Diseases (NTDs) affect the health and economies of populations globally. Many of these diseases are zoonotic, occurring as a consequence of the interaction between humans and animals, particularly at the household level in low- and middle-income countries. Based on the WHO Global Strategy to accelerate and sustain progress on NTDs, including zoonoses, through improvement in sanitation, hygiene and water, this review identifies existing published studies examining the interaction between water, sanitation and hygiene elements, animals and zoonosis transmission within the household. Only two out of 7588 studies screened met the criteria. They showed the relevance of animal influence in the effectiveness of WASH measures, as well as the difficulties of designing studies that look at this particular interaction. A synthesis of several studies analysed in the second selection stage of the review shows a significant relationship between animal and WASH factors for disease transmission. It also shows certain contradictions regarding the importance of key risk factors for some diseases across studies. It is therefore crucial to carry out further studies showing the interaction between animals and water, hygiene and sanitation measures within the household to improve these control measures and reduce zoonotic neglected tropical disease transmission.
| Neglected tropical diseases (NTDs) are a group of communicable diseases estimated to affect over a billion people globally, particularly those with least economic resources, access to health care, good nutrition, clean water and sanitation facilities; the weak political influence of affected groups as well as the complex nature of these diseases has resulted historically in a lack of attention and resources, precipitating the use of the term “neglected”[1]. This has been acknowledged by the World Health Organisation (WHO) and a global Roadmap was released in 2012 to focus on reducing the burden of 17 NTDs. This “Roadmap for Implementation” [2] includes five ‘key strategies to combat NTDs by 2020’ of which one aims to improve veterinary public health at the human–animal interface, and another emphasises the provision of safe and clean sources of water and effective sanitation infrastructure, and ensuring appropriate hygiene practices (WASH) [3]. The Roadmap, together with the 2015 WHO global strategy on WASH and NTDs [4], espouses a holistic approach to disease control and elimination.
The new global development framework enshrined in the Global Goals of the United Nations’ 2030 Agenda for Sustainable Development [5] sets out a One-Health approach to poverty, inequalities, health and the environment, in contrast with the siloed structure of the previous Millennium Development Goals (MDGs), whose agenda ended in 2015. Global Goal 3 within this agenda sets ambitious targets for improving health and wellbeing, including NTDs, and acknowledges the importance of addressing social and environmental determinants of health [6]. A One Health approach that addresses the animal-human interface and defines disease control strategies that enhance livelihoods and reduce poverty can contribute to the achievement of the Global Goals, but also represents a departure from current prevailing practices. Further knowledge on effective programming approaches is therefore urgently needed.
Several of the NTDs are zoonotic diseases—infections transmitted between animals and humans, and are therefore referred to as Neglected Zoonotic Diseases (NZDs). These include cysticercosis, rabies, echinococcosis, foodborne trematodiases, zoonotic African trypanosomiasis and schistosomiasis. Several of these are related to WASH elements in terms of prevention and/or treatment. Other diseases recognised by WHO in its “Research Priorities for Zoonoses and Marginalized Infections” include toxoplasmosis, cryptosporidiosis and bacterial zoonoses, for which improved sanitation has proven effective in reducing transmission [3]. The global burden of these zoonotic diseases is considerable. Cystic echinococcosis causes, on average, the loss of 2 million annual disability-adjusted life years (DALYs), with associated costs rising up to US$ 3 billion for human treatment and livestock industry losses [7]. Taenia solium, the causal agent of taeniasis and cysticercosis, is responsible for an estimated cost of 2.8 million DALYs globally [8]. Mortality due to cysticercosis in humans increased by 58% between 1990 and 2010 [9], and the disease is estimated to affect over 50 million people globally, causing up to 30% of all epilepsy cases [10]. Zoonoses are estimated to contribute to up to 10% of the total DALYs lost, and 26% of DALYs lost due to infectious diseases in low income countries [11]. Zoonoses affect human health directly, but by affecting animal health, they can also cause important economic losses and limitations for affected rural communities that depend on animals for working fields, transportation, as a source of protein and as a source of income when sold in local markets [12]. For example, cysticercosis has been reported to cause $12,6 million in annual losses in Cameroon [13], $150 million in India [14] and 18.6 to 34.2 million US dollars in East Cape, South Africa [15].
These zoonotic diseases are neglected due to the relatively low mortality associated with them, their tendency to affect predominantly poor and marginalised populations, and the complex, intersectoral measures required to control them, which include community infrastructure and capacity building, health promotion programmes, improved diagnostics and treatment, vaccination and prevention programmes and policy adaptation at local, regional, national and international level [11]. Zoonotic pathogens have complex life cycles that commonly include different phases in human hosts, animal hosts and the environment before completion. Overlooking one or more of these three elements facilitates the perpetuation of the cycle, and with it, reinfection. A One Health approach to controlling zoonotic transmission is needed, considering animals, people and the environment in a comprehensive approach to public health. Since zoonoses are influenced directly and indirectly by multiple factors, focusing solely on transmission routes wrongfully overlooks socio-cultural, economic, anthropological and ecological elements that may affect transmission as well as delivery of control programmes.
The need for intersectoral control measures is especially evident in low income countries [16], where the rural population accounts for an average of 69% of the total [17]. Not only do poor, rural communities have fewer resources and less access to healthcare, they also possess less political influence and power than other population groups to demand services and resources from government authorities [18–20]. A One Health approach helps create resilient solutions for disease transmission by setting measures that can be implemented in the long term by community and government action, meeting the objectives for sustainability set by the Sustainable Development Goals [21]. In poor, rural settings, smallholder animal production of indigenous species of pigs, poultry and ruminants is dominant [22], and hence human and animal interaction within the household is more common in these settings, requiring special attention to this interaction in the control of zoonotic diseases [23]. However, given the dependence of rural households on animals as a major source of livelihood and as an alternate source of income in emergencies, certain measures that may support disease control objectives may not be feasible in practice [24]. For example, pig-corralling is recommended as a main method for control of cysticercosis, and hence programmes may be put in place to improve this practice amongst farmers [25]. However, for many households and communities in middle-low income countries, this is not economically feasible [26], since this would require the family to assume the added cost of feeding the pigs, instead of allowing the animals to forage for themselves [27]. Similarly, protecting water sources from animal access prevents contamination of water for human use with animal faeces and secretions. However, the need to provide livestock and humans with sufficient clean water from a protected source poses a challenge for many communities [28].
A One Health approach can help identify such multi-factorial elements and avoid omitting valuable programme components, including human, environmental and animal factors. Human behaviour factors such as conflict, migration and socio-cultural practices, shape disease patterns, due to relocation, high human density and reduced hygiene levels [29]. Similarly, economic and agricultural development will reshape the land and demands of society, changing animal farming and animal product consumption practices, increasing the risk of food-borne disease transmission and zoonotic influenza [30]. An example of an animal factor to consider is how wildlife reservoirs can help perpetuate infective cycles within local livestock. This poses a great challenge for zoonotic disease control in pastoral communities due to the difficulty of limiting direct and indirect interaction between wildlife and livestock species [30, 31]. Additionally, ecological factors like climate change and deforestation have a direct impact on the distribution of vector-borne diseases by altering the habitats of the vector and reservoir species, as well as allowing vectors to sustain their life cycle in new areas due to a rise in average temperatures, leading to emergence and re-emergence of these diseases in new parts of the world [30, 32]. Another example of One Health approaches helping to tackle ecological problems can be found in the reuse of animal excreta as crop manure, as incorrect use can lead directly to disease transmission through contact and clothes and indirectly through water contamination [33]. Use of animal excreta as crop manure can also alter the chemical properties of the soil, endangering the environmental sustainability of the area, and subsequently increasing the exposure of humans and animals to contaminated sources of infection [33]. Authors like Nguyen-Viet, Zinsstag and Charron propose an integration method as a solution for optimising the use of human and animal excreta as manure, by combining cross-sectoral knowledge and stakeholder engagement under a One Health framework [33, 34]. Such a framework enables the implementation of sustainable control strategies for NZDs in countries where economic resources are scarce.
Water, sanitation and hygiene (WASH) programmes can plausibly contribute to control of zoonotic disease given the knowledge about pathogen transmission cycles, through provision of sanitation infrastructure that safely removes human and animal faecal waste from the human environment, provision of clean water sources, and improvement of hygiene practices at the community and household level [4]. The WHO WASH and NTDs strategy is a step towards developing collaboration between WASH and NTDs programmes, both of which reference integration of control measures, but do not offer specific guidance or methods of monitoring on collaboration between the sectors [4]. However, the much needed guidance to encourage a One Health approach through engagement of other sectors such as agriculture and veterinary public health is not included in the remit of the WASH and NTDs strategy [5, 35]. The positive relationship between WASH programmes and reduction of NTDs incidence has been proven, yet many of these programmes still lack the multifactorial approach needed to cover the impact of other elements that affect disease transmission [36], such as animal presence within the household and human-animal interaction. Because of this, there are limitations to understanding why WASH programmes may not result in the expected disease control outcomes and how they can be optimized. No systematic research has been done to date on the impact of demand-side sanitation programmes on NZDs transmission [3].
Although the evidence base on the interaction of animals with sub-standard sanitation facilities is weak, it is plausible that the presence of free-roaming household animals alongside conditions of open defecation or poor containment of faeces can contribute to intensified disease transmission [37]. As mentioned in the WHO WASH and NTDs Strategy [4], and as several authors argue [36, 38–40], it is necessary to gather more information regarding WASH-related interventions and disease burden reduction. This is particularly relevant for zoonotic diseases, as, out of the existing reviews relating to WASH and disease burden, few focus specifically on zoonotic diseases. Those that do, often disregard the presence of animals in the household and its impact on the effect of WASH interventions on zoonotic disease. There is need to identify these linkages and knowledge gaps that require further study. The aim of this work was to conduct a systematic review to identify the existing published data, on how the presence of animals in the household impacts the efficacy of WASH interventions for zoonotic disease control.
The objectives of this review were: to identify how animal presence in the household influences the effectiveness of water, hygiene and sanitation measures for zoonotic disease control in low and middle income countries; to identify gaps of knowledge regarding this topic based on the amount and type of studies looking at this particular interaction.
A review protocol was designed to inform and direct the review steps before conducting the systematic review. The protocol was designed based on the guidelines given by “CRD’s guidance for undertaking reviews in health care” and the “WHO Handbook for Guideline Development” [41, 42], as well as example systematic review protocols found in various academic sources, approved by peer academic experts. The complete protocol can be found in Text S1.
Three databases were used: Medline, Web of Science and Global Health. These were chosen based on other systematic reviews conducted in the area of sanitation, hygiene and NTDs [43–45], and on expert academic advice solicited by the authors. The three databases were systematically searched for publications dating 1980 to 30th April 2016.
The search terms relative to WASH were chosen based on other WASH literature reviews and scientific articles. Animal terms were selected based on literature and expert advice, including those species most likely to interact with humans within the household, in low- and middle-income countries. The terms were then divided into four pools:
The terms amongst pools were combined by the Boolean operator “OR”, while those between pools were combined by the Boolean operator “AND”.
Diseases chosen for the terms were based on the list of neglected zoonotic diseases described in the WHO NTDs Roadmap [2]. The results obtained were sorted by “author” in descending order. Studies were selected through a three-stage process, first by title and abstract screening, then by full text analysis, based on the selection criteria for each stage, and finally by a quality control checklist. References were managed with the use of reference management software EndNote X7.
For the first stage, title and abstract screening, studies were included if the abstract mentioned a zoonotic disease term together with a WASH term, if a full text version was available and if the article was published in English or Spanish. Studies not meeting these requirements, and review articles, were excluded.
The full text versions of studies selected in this first stage were retrieved and analysed for further selection. In this second stage, articles that did not quantify burden of disease in human or animal populations, did not analyse the role of animals in zoonosis transmission in relation to WASH measures, or did not meet the requirements of the quality check described in the protocol, were excluded from the review. The type of study and its design were not deemed to be crucial inclusion/exclusion criteria, due to a low number expectancy of final study retrieval.
Studies selected for the last stage of the systematic review were analysed using a quality checklist based on the guidelines for public health studies from the National Institute for Health and Clinical Excellence [49].
Articles included in the full text review were subjected to data extraction based on the protocol, with special attention to the study population regarding burden of disease, the diagnostic method used, the WASH measures in place, description of animal presence within the household, and the statistical analysis approach taken by the study. Due to the consideration of various types of studies in the inclusion criteria and the expected low count of final studies making the last selection, pooling was not deemed possible. Therefore, a narrative approach was chosen for addressing data synthesis. Zoonotic diseases in which WASH measures play a relevant role in control were included in the analysis and synthesis of the results, as long as the selected study included it in its own analysis, even if said diseases were not considered to be neglected by inclusion in the WHO reference list.
Seven thousand five hundred and eighty-eight (n = 7588) studies where obtained after introducing the search terms into the three databases (Fig 1). Screening of titles and abstracts retrieved a total of 80 studies (n = 80) meeting the inclusion criteria for the first stage of the review: 46 from Medline, 28 from Global Health, and six from Web of Science. Of these 80, 13 were duplicates and three were unable to be retrieve in full-text form and were therefore discarded. The total number of articles selected for the next stage of the review was 64.
Full text for the remaining 64 articles was obtained, analysed and considered for review inclusion. After data extraction and analysis, two articles [50, 51] were identified that quantified the burden of disease in humans or animals and analysed the role of animals in zoonosis transmission in relation to WASH measures, hence meeting the final inclusion criteria as set out in the protocol. Due to the low count of studies included in the final review, the 64 articles analysed in this phase were summarised in the form of tables that show the research tendencies when addressing WASH and NZDs. The complete list with the main data extracted from each one can be found in Table 1, including location, type of study, number of participants in the study, disease of interest, diagnostic test used to address presence of disease, WASH and animal component studied, the type of statistical method used for the analysis, and a summary of the results of the study.
More than half of the studies (29) focused on cysticercosis, while 12 focused on toxoplasmosis (Table 2). Humans appear as the most studied species, with 36 studies looking at human burden of disease, while pigs were second with 26 citations. Fifty one out of 64 were designed as cross-sectional studies, 46 of these establishing a prevalence value through a serological test and combining it with a questionnaire for associated risk factors. Table 3 shows the study count for each of the categories for water, hygiene and sanitation components, and the proportion of studies that included one, two, or the three types is shown in Fig 2. Three studies had at least one factor in each of the categories.
The summarised data suggests the existence of a relationship between NZD epidemiology and the contact of humans and animals in the household, generally showing a negative impact of animal presence on WASH measures or an enhanced negative effect of animal presence on the impact of poor WASH conditions. In the case of cysticercosis, studies show contradictory results regarding the impact of WASH measures and animal presence on disease prevalence.
Due to the small number of studies that were selected based on the criteria, the outcome of the quality control check was not considered for further exclusion.
The study by Holt et al. (2016) was designed as a cross-sectional study examining prevalence of hepatitis E virus (HEV), Japanese encephalitis virus and Trichinella spiralis in both humans and pigs, as well as Taenia spp. solely in humans in two provinces of Lao PDR, with a multiple correspondence analysis and a hierarchical clustering of several components deemed relevant to disease transmission. Three clusters were identified: one referential (cluster 1) with the best sanitation and lowest pig contact; cluster 2, with moderate sanitation levels and slaughtering of pigs as the main source of animal contact; and cluster 3, with lower sanitation levels and a relative higher rate of free-roaming pigs. The risk of human infection, measured through Odds Ratio (OR), for each of the diseases and clusters when compared to cluster 1 are shown in Table 4. HEV had a very similar OR for risk of infection between clusters 2 and 3, despite the superior WASH conditions of cluster 2. For Taenia spp. and Cysticercosis, risk of infection proved higher in cluster 3 than cluster 2, but with a significant increased risk of infection in cluster 2 compared to the control, despite solid practices of hand washing and water boiling amongst the population. Finally, Japanese encephalitis showed an increased risk of infection in cluster 2 over cluster 3, despite better WASH conditions. Data regarding pig seropositivity was not clustered and WASH factors were not found to be significant in T. spiralis and HEV infection.
The other study (Bulaya et al. 2015) was a comparative study pre- and post- community-led total sanitation (CLTS) intervention for porcine cysticercosis control, identifying prevalence performing an Ag-ELISA test. There was no randomization in village selection or house selection, and instead selected based on village characteristics and willingness to participate, respectively. The prevalence pre-intervention was 13.5%, (6.8–20.1, 95% C.I.), compared to a value of 16.4% (12–20.8, 95% C.I.) post-intervention, although this increase was deemed non-significant by the author. After the intervention, latrine presence improved from 67.2% to 83.1%, with the percentage of free-roaming pigs changing from an 89.8% to a 30.3% of them free roaming, 43.8% partially free roaming and 25.8% penned. Home slaughter of pigs increased from 49.15% baseline to 80.90% post-intervention. Despite the improvement in latrine presence, animal husbandry was not improved enough to avoid direct and indirect contact between animals and humans within the household.
This review showed examples of the way animal-human interaction can affect the effectiveness of WASH interventions for zoonosis control. Importantly, it also highlighted the dearth of studies looking specifically at this interaction. After the search retrieved 7588 articles for this review, 64 were selected in the first screening, of which only 2 were selected for the final review after the second screening. This outcome is likely due to the sectoral focus of the studies. Traditionally, research groups investigating the effectiveness of WASH interventions focus on human factors as positive or negative influences. Similarly, the Veterinary Public Health community focuses more on animal-related factors and disease-transmission routes. The interaction between these two aspects is a research and programming ‘blind spot’, as was demonstrated by this review, and needs to be addressed with further intersectoral research studies.
As noted by Zinsstag in 2015 [33], a study in Vietnam showed how a One Health approach for WASH programmes integrates all factors into one framework. This helps identify the relationship between the factors, while exposing the missing links and the areas in need for further research, of which the main one stated is “the boundaries of the sanitation problem”. Sanitation and hygiene programmes have proven effective in reducing NTD burden in numerous studies, as backed by various systematic reviews [43–45]. However, effective, full-coverage implementation of control programmes considering both human and animal sanitation aspects can be challenging in practice. As described by Guilman et al. in 2012 [26], some communities may not have sufficient resources to change their animal farming system to one that limits animal-human contact. In other cases, the community may actually benefit economically from this new farming system [114], but as long as the population believes this is not the case, no change will be embraced by the community [115]. This reinforces the importance of accompanying these type of logistic measures with strong education and hygiene promotion campaigns that involve the community and show the importance and benefits of adopting them.
The study by Holt et al. [51] compared Odds Ratio of infection in several pig zoonoses between different sanitation and pig contact factors. For HEV, lower levels of sanitation, as described in the results section, proved to be a risk factor for virus presence, without significant differences between these lower levels specifically. However, increased contact with pigs, particularly through handling and slaughtering, proved significant in its influence on the effectiveness of WASH measures in disease control, as the cluster with moderate sanitation and close pig contact had equal risk of infection as the cluster with poorer sanitation. Pig contact has been described as a risk factor for HEV transmission previously [116], but according to this study, pig corralling impede their access to the household would not make a significant difference in disease transmission as long as the animals are still being slaughtered at home, due to direct human contact with pig blood. In the case of Trichinella, socioeconomic status acted as a confounder, since the main risk factor is pork consumption [117, 118], which in this study was associated with higher status due to availability and affordability cost, as are good sanitation and hygiene conditions. In the case of JEV, the cluster with higher direct contact with pigs showed a higher risk of infection, despite better sanitation and hygiene conditions, showing an example of how animal contact can severely hinder the effectiveness of WASH measures. This could be due to its vector-borne nature, which correlates to two factors of this particular cluster: unprotected water sources, which facilitates breeding areas for Culex spp.; hygiene practices, latrine use or corralling measures would not make a significant impact in its transmission unless done optimally, avoiding contamination of water that could facilitate Culex spp. reproduction. Regarding Taenia solium and cysticercosis, the cluster with higher rates of free-roaming pigs and open defecation showed the highest risk of infection, as expected. However, the high risk of infection presented by the cluster with moderate WASH and close contact with pigs shows how the latter can affect the effectiveness of the former.
During the selection process of this review, several studies (Table 1) were screened and later revisited, for further insights on the impact of animals on WASH interventions. Some showed presence, usage or condition of latrines and free roaming of pigs to be significant risk factors in disease transmission [84, 119, 120], but others had non-significant results [107], rather identifying the source of water for consumption and its quality as a risk factor. In contrast, Nkouawa et al. in 2015 [87] identified that despite having a non-potable (unsafe) water source, disease transmission was reduced by improving hygienic practices and corralling pigs. The study by Holt et al. [51] provided robust results on relative impact of animal and WASH factors, meeting the criteria for selection stated in the protocol of the review. However, future studies should ideally be designed in a way that focuses on isolating the influence of animal factors on the effectiveness of WASH measures. This is particularly difficult to achieve given the circumstances of the communities in which these studies need to be conducted: as noted by Schmidt et al. in 2014 [121], designing impact studies on water, sanitation and hygiene and retrieving significant results is a recurrent challenge for the scientific community: Randomised controlled trials are rarely free from bias, while observational studies usually lack a large enough study population or result significance [121]. Additionally, performing randomised controlled trials in the optimal representative geographical areas is logistically and economically challenging. Another factor to take into account is time, since marketing and promotion campaigns can take several years to have a significant effect, deeming any study that withholds investment in WASH services for such an extended period of time unethical [121].
A relevant limiting factor to assess the efficiency of any WASH programme implementation is the correct use, design and upkeep of sanitation facilities. Several studies show that although latrines were present in the community, they were not consistently used for defecation by all household members or kept in a sufficiently hygienic state [84, 85]. The incorrect use of latrines is often associated with socio-cultural and psychological factors, as identified by Thys in 2015 [122], such as a sense of reduced privacy, latrines being too close to the village, comfort of use or trust in its efficacy and need of use. Lack of ownership of the need for latrine construction and lack of ongoing support for maintenance and improvement can undermine potential health benefits of basic latrines.
The study by Bulaya et al. in 2015 [50], showed that despite the CLTS intervention resulting in increased latrine presence, net increase in latrine usage and improved pig husbandry, prevalence of disease in pigs increased slightly after the intervention. The study did not specify whether the newly built latrines resulted in safe separation of humans and animals from human faeces. Achieving that level of detail in the analysis is an objective for future studies. Although deemed non-significant, the 95% C.I. shows almost no change in prevalence from pre to post intervention. This was attributed by the authors to infected members of the community still practising open defecation due to lack of resources for latrine construction. Not corralling the totality of the pig population, therefore allowing for interaction of animals and humans within the household, could be the explanation as to why the increase in latrine presence had no effect in decreasing porcine cysticercosis. Free roaming of pigs has been identified as a risk factor for porcine cysticercosis by some of the studies screened before review inclusion [69, 75] but was found to be non-significantly others [72]. Similarly, the presence of latrines can be significant [72, 73] or non-significant [69] for disease prevalence in pigs, depending on the study, reinforcing the findings by Bulaya et al. (2015). As previously mentioned, low latrine usage has been described as a risk factor for disease transmission [59, 84, 85] but also as a recurrent sociocultural problem, since many members of the community do not use latrines on a consistent basis for a variety of reasons [59, 115, 122], or do not keep the latrines in a suitable condition for them to effectively reduce disease transmission [84, 115, 120]. However, poor programme design, lack of follow up or disputes between NGOs and community leaders on logistics, provisions and payments can be a cause for poor latrine construction and maintenance [123]. This reinforces the suggestion made by Bulaya et al.[50] of the importance of continued hygiene promotion programmes and access to sanitation hardware options in order to ensure the complete effectiveness of sanitation or animal husbandry improvement programmes.
As an example of a multifactorial approach to disease transmission control, prevalence of Schistosomiasis was significantly reduced in three studies in China [70, 102, 124] by implementing a complete WASH programme with sanitation facilities and hygiene educational programmes, reducing the indirect contact of animals and humans through water and reducing the population of the host snail species for Schistosoma. However, programmes that alter animal husbandry in drastic ways such as changing free-roaming farming systems into stabling farming systems, also alter the local economy of the community [125]. In the case of cysticercosis, the penning of pigs is not always possible in certain communities given the resulting increased costs of feed and infrastructure [125]. Substantial investment and economic compensation to farmers and households would therefore be required to maintain and sustain these programmes consistently over time [126].
In the case of toxoplasmosis, principal and consistent risk factors for infection identified throughout the literature, include unsafe water source, inadequate hygienic conditions of the household and cat presence in the household or the vicinity, and were common to human [52, 66] or animal [55, 58] infection. While providing clean water sources and creating appropriate hygienic conditions decreases the burden of disease, avoiding the presence of cats within the household could potentially increase the presence of rodents in many communities that use cats as the sole method of rodent control. A study showed how, when combined, the presence of cats and dogs in an area significantly reduced the local rodent population [127], however, more research should be conducted to clarify the impact of cat population control on rodent-transmitted diseases in rural communities.
The review protocol was designed to include animal-focused studies as well as human-focused studies to ensure a One Health approach to zoonotic disease transmission. Particularly for NZDs, interrupting sustained transmission requires a multifactorial approach considering both zoonotic and anthroponotic transmission paths. Reducing animal burden of disease has a direct effect on human prevalence of disease and vice versa [128], and therefore WASH programmes applied equally to human and animal populations are likely to provide better results than a human-centred approach. The review identified the lack of studies looking at the importance of animal influence in WASH programmes, exposing the existent lack of knowledge in the matter. Further research and programme design need to focus further on animal impact and isolating the study of animal components in the efficiency of WASH control programmes. One of the limitations of the review was the non-inclusion of rodent species in the study. Although rodents are acknowledged to be a source of NZD transmission within the household, they were deemed to overreach the scope and feasibility of this review: on one hand because the review focused in farmed animals kept by the household owners; on the other hand because thorough control of rodent activity in the household is difficult and less reliable than that of farmed animals, mainly due to the complex biological and ecological characteristics of each local rodent species [129, 130]. The initial literature review was conducted for fulfilment of an MSc with one student. All three co-authors advised on the approach to be taken and made revisions to the literature. Throughout the writing of the literature there was input from all authors who also held regular review meetings. To further optimise the systematic review, a second reviewer would have performed the search and selection and compared results. Also, had a longer period of time been available, more databases could have been screened, although the final count of studies would most likely be low, since the tendency identified in the review is that of a very low percentage of studies looking specifically at animal influence in WASH measures efficacy. The time constraints were due to the timelines of the MSc. However, all authors had additional input to the manuscript. Whilst the initial literature review was conducted by one student, the manuscript has been prepared after revisions by all authors with additional literature added after further reviews. This has been rewritten to reflect the input following the initial MSc project.
This systematic review demonstrated the relevance of human-animal interaction within the household for the effectiveness of WASH measures for control of NZDs. It also shows the significant lack of specific studies tending to the effect of animals on WASH programmes’ effectiveness for zoonotic disease control. Several examples exist in the literature describing prevalence of zoonotic disease and associated risk factors, yet, in the majority of cases, their design fails to assess the specific influence of animal presence in WASH interventions. Further research should be undertaken regarding the influence of animals in WASH programmes, ideally isolating the sanitation component and studying different levels of animal interaction and exposure within the household. Attention to animal burden together with human burden of disease would allow for better understanding and optimisation of WASH programme effectiveness on both disease control and broader development objectives. There exists an evident lack of direct coordination between WHO’s WASH and NTDs official programmes. Further developing of a research agenda around the animal-sanitation-disease link can help set out clear actions on which disease control programmes can be based.
|
10.1371/journal.pgen.1006129 | Misregulation of Alternative Splicing in a Mouse Model of Rett Syndrome | Mutations in the human MECP2 gene cause Rett syndrome (RTT), a severe neurodevelopmental disorder that predominantly affects girls. Despite decades of work, the molecular function of MeCP2 is not fully understood. Here we report a systematic identification of MeCP2-interacting proteins in the mouse brain. In addition to transcription regulators, we found that MeCP2 physically interacts with several modulators of RNA splicing, including LEDGF and DHX9. These interactions are disrupted by RTT causing mutations, suggesting that they may play a role in RTT pathogenesis. Consistent with the idea, deep RNA sequencing revealed misregulation of hundreds of splicing events in the cortex of Mecp2 knockout mice. To reveal the functional consequence of altered RNA splicing due to the loss of MeCP2, we focused on the regulation of the splicing of the flip/flop exon of Gria2 and other AMPAR genes. We found a significant splicing shift in the flip/flop exon toward the flop inclusion, leading to a faster decay in the AMPAR gated current and altered synaptic transmission. In summary, our study identified direct physical interaction between MeCP2 and splicing factors, a novel MeCP2 target gene, and established functional connection between a specific RNA splicing change and synaptic phenotypes in RTT mice. These results not only help our understanding of the molecular function of MeCP2, but also reveal potential drug targets for future therapies.
| Rett syndrome (RTT) is a debilitating neurodevelopmental disorder with no cure or effective treatment. To fully understand the disease mechanism and develop therapies, it is necessary to study all aspects of the molecular function of methyl-CpG binding protein 2 (MeCP2), mutations in which have been identified as the genetic cause of RTT. Over the years, MeCP2 has been shown to maintain DNA methylation, regulate transcription and chromatin structure, control microRNA processing, and modulate RNA splicing. Among these known functions, the role of MeCP2 in modulating RNA splicing is less well understood. We took several unbiased approaches to investigate the how MeCP2 may regulate splicing, what splicing changes are caused by the loss of MeCP2, and what functional consequences are caused by altered splicing. We discovered that MeCP2 interacts with splicing factors to regulated the splicing of glutamate receptor genes, which mediate the vast majority of excitatory synaptic transmission in the brain; and linked the altered splicing of glutamate receptor genes to specific synaptic changes in a RTT mouse model. Our findings not only advance the understanding of RTT disease mechanism, but also reveal a potential drug target for future development of therapies.
| Rett syndrome (RTT) is a progressive neurodevelopmental disorder that predominantly affects females[1, 2]. Classic RTT patients develop normally in the first 6–18 months, and then undergo a rapid regression of higher brain functions that eventually leads to the loss of speech and purposeful hand movement, microcephaly, dementia, ataxia and seizure[3]. Mutations in the human X-linked methyl-CpG-binding protein 2 (MECP2) gene are responsible for over 90% of classic RTT cases[4]. MeCP2 is abundantly expressed in the mammalian central nervous system (CNS) and binds to methylated CpG site throughout the genome[5]. Despite decades of work, the underlying molecular mechanism of how mutations of MECP2 lead to RTT is not fully understood.
In order to reveal the RTT disease mechanism, it is necessary to study the molecular function of MeCP2. Previous research on the molecular function of MeCP2 has focused on the localization of MeCP2 in the nucleus and the proteins that physically interact with MeCP2. At the microscopic level, MeCP2 appears to be colocalized with heterochromatin and thus is hypothesized to induce large-scale chromatin reorganization during terminal differentiation[6]. At the genomic level, MeCP2 can bind to unmethylated DNA[7], methylated cytosine[5], and hydroxymethylated cytosine[8], and may preferentially modulate the expression of long genes[9]. In parallel to research on MeCP2 localization, many proteins have been identified to physically interact with MeCP2. Based on the known functions of identified MeCP2-interacting proteins, previous studies have suggested a role for MeCP2 in maintaining DNA methylation[10], regulating transcription[11–16], chromatin structure[17–22], and RNA processing[23–25]. Future effort to combine the insights from the two approaches described above may allow more detailed understanding of the regulation of each of these specific protein-protein interactions across the entire genome, as well as the relevance of each interaction to RTT disease pathogenesis.
Misregulation of RNA alternative splicing has been implicated in a number of neurological disorders, which can be classified into two categories: cis-acting splicing disorder and trans-acting disorder[26]. Cis-acting disorder is caused by mutations that affect splicing of the mutant gene itself and therefore the function of that gene. An example of this type is frontotemporal dementia with Parkinsonism linked to chromosome 17 (FTDP-17), in which mutations in the MAPT (Tau) gene alter the function of Tau by increasing the exon 10 containing isoform[27]. In contrast, trans-acting disorders are caused by the loss of function of genes with regulatory roles in RNA splicing. For instance, the loss of survival motor neuron protein 1 (SMN1) function affects biogenesis of small nuclear RNA (snRNA) and lead to widespread splicing changes in spinal muscular atrophy (SMA)[28]. Relevant to RTT, the Zoghbi lab has identified RNA-dependent interaction between MeCP2 and Y box-binding protein 1 (YB1) in cultured cells, and further reported many altered RNA splicing events in a mouse model of RTT[23]. However, it is not clear if the splicing alterations are indeed dependent on the MeCP2/YB1 interaction and no link has been discovered between any gene-specific splicing change and specific neuronal phenotypes in RTT. Therefore the mechanistic and functional links between MeCP2, splicing regulation, and RTT phenotypes remain elusive.
To facilitate systematic identification of MeCP2-interacting proteins in the brain, we created a knockin mouse line (Mecp2-Flag) that expresses Flag-tagged MeCP2 from the endogenous Mecp2 locus[29]. This unique tool gives us two main advantages. First, it ensures that the MeCP2-Flag protein is expressed at the physiological level, so that non-specific protein-protein interactions caused by the overexpression of MeCP2-Flag is minimized. Second, it allows us to use a highly efficient anti-Flag antibody in the co-immunoprecipitation. The choice of antibody is not a trivial issue, because in the past, different anti-MeCP2 antibodies in co-immunoprecipitation experiments have generated conflicting results in the identification of MeCP2-interacting proteins[30, 31]. Mass spectrometry analysis of proteins co-immunoprecipitated by the anti-Flag antibody from the nuclear extract prepared from the Mecp2-Flag mouse brains showed that MeCP2 interacted with multiple splicing factors. Some of these physical interactions were disrupted by RTT-causing mutations in MeCP2. Furthermore, ChIP-seq analysis of revealed MeCP2 occupancy at exon/intron injections, which provides additional support of the role for MeCP2 in modulating alternative splicing. Consistent with previous findings, hundreds of splicing events were found to be misregulated in the cortex of Mecp2 knockout (KO) mice. More importantly, a specific splicing change in the Mecp2 KO cortex-a shift in the balance between the flip and flop exon in the AMPA receptor (AMPAR) genes was causally linked to synaptic phenotypes of faster desensitization kinetics of AMPAR-gated current and altered synaptic transmission. Together, our findings substantiate the role of MeCP2 in regulating alternative splicing of RNA by revealing direct physical interaction between MeCP2 and multiple splicing factors, association of MeCP2 at exon/intron junction, and providing the first functional link between a specific splicing alteration and synaptic phenotypes in RTT mice.
To facilitate identification of MeCP2-interacting proteins in the mouse brain, we generated the Mecp2-Flag knockin mouse line that expresses Flag-tagged wild-type MeCP2 from the endogenous locus. We purified nuclei from the brains of male Mecp2-Flag mice, prepared nuclear extract and performed co-immunoprecipitation (co-IP) using the anti-Flag antibody. Eluted protein sample was then subjected to protein identification by mass spectrometry. Forty-eight proteins were identified using highly stringent statistical filters (S1 Table). Identified proteins included previously known MeCP2-interacting transcriptional regulators and chromatin remodeling proteins, such as HDAC1 and components of the SWI/SNF complex. Consistently, gene ontology (GO) analysis showed that proteins identified by co-IP/MS were enriched with GO terms of chromatin organization, chromatin modification and regulation of transcription. Interestingly, these proteins were also enriched for RNA splicing (Fig 1a). MeCP2 has been implicated in regulating splicing, but its role in pre-mRNA splicing has not been studied extensively. Therefore we decided to focus our study on the interaction between MeCP2 and several splicing factors.
To validate the physical interaction between MeCP2 and splicing factors, we performed anti-Flag co-IP in nuclear extract from the Mecp2-Flag knockin mouse brain and probed it with antibodies against TDP-43, LEDGF, DHX9, FUS, hnRNP H, and hnRNP F, respectively. Western blot results showed that TDP-43, LEDGF, DHX9 and FUS were co-immunoprecipitated with MeCP2, whereas hnRNP H and hnRNP F were not (Fig 1b). Next, we performed reverse co-IP and detected MeCP2 in immunoprecipitate of anti-TDP-43, anti-LEDGF, anti-DHX9, anti-FUS, and anti-hnRNPH+F (Fig 1c), further confirming that MeCP2 physically interacts with these proteins in the mouse brain. In addition, the interaction between MeCP2 and splicing factors were not sensitive to Benzonase treatment, which digest and remove all nucleic acids, suggesting that these interactions were most likely direct interactions independent of either DNA or RNA (S1 Fig).
The interaction between MeCP2 and LEDGF has been previously reported in cancer cells[32], but not in the brain. MeCP2 interacts with the N-terminal PWWP-CR1 domain of LEDGF, but which domain of MeCP2 that LEDGF binds to is not defined. DHX9 has recently been revealed as an interacting partner of MeCP2[33], but the interaction domain is not known either. To examine which domain of MeCP2 is required for interaction with LEDGF and DHX9, we expressed HA-tagged MeCP2 with different deletions (Fig 2a) and Myc-tagged full-length LEDGF/p52, LEDGF/p75 and DHX9, respectively, in HEK293 cells. Co-IP with anti-HA antibody followed by Western blot with anti-Myc antibody showed that deletion of amino acids 163–380 of the MeCP2 protein significantly reduced the interaction between MeCP2 and LEDGF/p52, LEDGF/p75 or DHX9 (Fig 2b and 2c; S2a Fig). Reverse co-IP with anti-Myc antibody followed by Western blot with anti-HA antibody also demonstrated that amino acids 163–380 of the MeCP2 protein was required for interaction between MeCP2 and LEDGF/p52 (S2b Fig). Collectively, these results strongly suggested that the transcription repression domain (TRD) of MeCP2 is essential for the interaction of MeCP2 with RNA binding proteins.
Several RTT disease causing mutations locate in the region of amino acids 163–380 (R168X, R255X, R270X and R294X) and may disrupt the TRD domain, therefore we asked whether these mutations affect the interaction between MeCP2 and LEDGF or DHX9. To test this, we co-transfected MeCP2 constructs encoding MeCP2 WT, MeCP2R168X, MeCP2R255X, MeCP2R270X, and MeCP2R294X, respectively, with LEDGF/p52 or DHX9 in HEK293 cells. Co-IP assay showed that interaction between LEDGF/p52 and MeCP2R168X, MeCP2R255X, and MeCP2R270X was significantly impaired (Fig 2d and 2e). Interestingly, the interaction between MeCP2R294X (retaining a large fraction of TRD) and LEDGF/p52 was not significantly different from that between wild type MeCP2 and LEDGF/p52, suggesting that amino acids 270–294 of MeCP2 are required for its binding to LEDGF/p52 (Fig 2d and 2e). Similarly, we found that MeCP2R168X and MeCP2R255X interacted poorly with DHX9, while MeCP2R270X and MeCP2R294X had intact binding capability, indicating that amino acids 255–270 of MeCP2 are required for its binding to DHX9 (Fig 2f and 2g).
The newly identified interactions between MeCP2 and multiple splicing factors prompted us to determine whether there are widespread RNA splicing changes upon loss of MeCP2. We conducted high-throughput sequencing of RNA (RNA-Seq) from the cortex of wild type and Mecp2 knockout (KO) mice. As a measure of the quality of the RNA-Seq data, we first examined whether our data reflect transcriptional changes consistent with previous findings. We examined transcriptional changes in our RNA-Seq data by applying a negative binomial model in edgeR[34]. Recently, a meta-analysis of transcriptional changes across multiple brain regions in Mecp2 KO or overexpression (OE) mouse identified 466 MeCP2-repressed genes based on high degree of consistency (log2FC > 0 in KO or log2FC < 0 in OE in at least 7 out of 8 datasets; FC: fold change)[9]. Of these genes, 315 genes (~68%) were also found to be up-regulated (log2FC[KO/WT] > 0) in our analysis result, suggesting significant overlap between transcriptional changes identified in our study and previous studies (S2 Table). In addition, we selected seven previously known misregulated genes in Mecp2 KO[9, 35] as well as six novel differentially expressed gene identified by our study for further validation. qRT-PCR results show that all of them show similar changes as observed in our RNA-Seq data (Pearson’s r = 0.95) (S3 Fig). Taken together, these data indicate that our RNA-Seq data are robust in identifying transcriptional changes.
Next, we applied the Mixture of Isoforms (MISO)[36] algorithm to the RNA-Seq data and identified 263 alternative splicing (AS) events that were significantly changed in the cortex of Mecp2 KO mice using a stringent filter (S3 Table; see Methods for detail). Loss of MeCP2 affects various types of AS events, including skipped exon (SE), mutually exclusive exons (MXE), retained intron (RI), alternative 5’ ss exon (A5E), and alternative 3’ ss exon (A3E) (Fig 3a). Subsequent analysis indicated that although more RI or MXE events had slightly reduced percent spliced in (PSI) value, SE, A5E and A3E events, which in total represented the majority of events, had similar number of events with increased or decreased PSI (Fig 3b). These data suggest the loss of MeCP2 affects alternative splicing in both directions, which is similar to the knockdown or overexpression of a typical splicing factor[37, 38]. Additionally, functional enrichment analysis using DAVID showed that genes with splicing changes were enriched with splice variant, alternative splicing, phosphoprotein, cell junction, compositionally biased region (Ser-rich) and plasma membrane part (S4a Fig). Interestingly, gene expression analysis on the 232 genes associated with splicing changes revealed that the majority of them have similar total expression level between WT and Mecp2 KO (only 15 genes show larger than 1.25-fold change and only one shows larger than 1.5-fold change) (Fig 3c), suggesting that MeCP2-mediated transcriptional regulation and splicing regulation are independent of each other. To validate the splicing changes, we performed qRT-PCR with isoform specific primers to evaluate 20 SE events. We observed consistent changes in 13 genes as identified by MISO (65%), including 6 events with decreased PSI and 7 events with increased PSI in Mecp2 KO (S5 Fig). The overall validation rate from our study of using biological replicates of tissue is comparable to the success rate using cell lines in two recent studies (74% and 71%, respectively)[39, 40], when more than 20 events were selected for validation.
To generalize our observation that loss of MeCP2 leads to global splicing changes, we analyzed RNA-Seq data generated from Mecp2 KO hypothalamus and visual cortex in two recent studies[9, 35], respectively. Using a cutoff of |ΔPSI| ≥ 5% and Bayes factor≥1, 482 and 719 SE events were identified by MISO to be changed in Mecp2 KO hypothalamus and visual cortex, respectively. 150 of the 482 SE events identified in the Mecp2 KO hypothalamus and 171 of the 719 SE events identified in the Mecp2 KO visual cortex were also found in our study (S4–S6 Tables). We focused our meta-analysis on SE events because this is the best-annotated category of alternative splicing events in the mouse genome. In summary, the large number of alternative splicing changes in independent RNA-seq data sets and the significant overlap between data sets generated from different brain regions of different lines of Mecp2 KO mice at different ages are consistent with the notion that loss of MeCP2 results in global splicing alterations.
To further study whether MeCP2 may be directly involved in modulating splicing, we examined MeCP2 occupancy across the genome. ChIP-Seq analysis was performed using the anti-Flag antibody on chromatin prepared from the cortex of the Mecp2-Flag knockin mice. 20,652 high confidence MeCP2 ChIP-seq peaks were identified (S7 Table, see Methods for detailed description on ChIP-seq analysis and quality control statistics.). Based on statistical ranking and robustness of primer design, 5 of the identified peaks were selected for independent validation (highlighted in S7 Table). ChIP-qPCR on a separate cohort of Mecp2-Flag mice detected significant occupancy of MeCP2 at the genomic locations corresponding to these 5 peaks relative to Gapdh promoter (S7 Table). To gain an overall picture of MeCP2 distribution across the genome relative to genes, we examined the MeCP2 ChIP-seq signal in the 2kb region immediately upstream of all transcriptional start sites (TSS), the region from TSS to the transcription end sites (TES), and the 2kb region immediately downstream of TES across the genome. This analysis revealed that MeCP2 occupancy was depleted at promoters (~0 read count per million mapped reads [ChIP minus Input] from TSS to -2,000bp. In contrast, MeCP2 binding is enriched in the gene body (0.05–0.33 read count per million mapped reads [ChIP minus Input] from TSS to TES, S4b Fig). To assess the correlation between MeCP2 ChIP-seq signal and DNA methylation, we calculated the average percentage of mCG and mCH across all of our MeCP2 ChIP-seq peaks using previously published whole genome base-resolution methylation data in mouse cortex [42], and found that the percentage of mCG in MeCP2 ChIP-seq peaks is slightly higher than genome average (~83% vs 78%), and the percentage of mCH in MeCP2 ChIP-seq peaks is significantly higher than genome average (2.78% vs 1.30%), suggesting MeCP2 ChIP signal is correlated with mCH and mCG across the genome. These results are consistent with several previous studies that demonstrated that MeCP2 occupancy tracks DNA methylation across the genome[35, 43, 44]. Moreover, the average GC content in MeCP2 ChIP-seq peaks is ~53.8%, significantly higher the genome average of 42% [45]. The correlation between MeCP2 occupancy and GC content is consistent with findings reported in earlier this year [44]. Interestingly, gene ontology analysis found that genes with MeCP2 ChIP-seq peak(s) were enriched with GO terms of alternative splicing (Fig 3d). Indeed, alignment of MeCP2 ChIP-seq reads with the 5’ and 3’ ends of exons revealed a significant enrichment of MeCP2 ChIP-seq peaks around the exon/intron boundary and over exons (Fig 3e). Consistently, significant enrichment of hmC and mCG signals were also found at intron/exon boundary and on exons, while a modest enrichment of mCH signal was observed at the 3’ end of exons (S4c–S4e Fig). Taken together, the physical interaction between MeCP2 and splicing factors, the widespread changes in RNA splicing, and the enriched MeCP2 occupancy around exon/intron boundary are consistent with each other and strongly suggest that MeCP2 could play an important role in regulating alternative splicing.
Gria2 is a major component of the AMPA receptor (AMPAR), which mediates the vast majority of fast synaptic transmission in the CNS. Two electrophysiologically distinct isoforms for Gria2 are generated by a mutually exclusive splicing event of the Gria2 pre-mRNA. Depending of the usage of either the flip or the flop exon, Gria2 pre-mRNA can be spliced into either the flip or the flop isoform. Our RNA-Seq data revealed that ~ 51% of all Gria2 transcripts contained the flip exon in wild type mice. In contrast, only ~ 28% Gria2 transcripts included the flip exon in the Mecp2 KO mice (Fig 4a). qRT-PCR analysis in a separate cohort of animals confirmed a shift of flip/flop ratio toward a flop dominant state in Mecp2 KO mice, while the total expression level of the Gria2 gene remained unchanged (Fig 4b and 4c). The reduction of flip/flop ratio in Mecp2 KO mice are not likely due to delayed development of the brain because the flip isoform is more abundant during early brain development and the flop isoform gradually increases to a comparable level of the flip isoform toward adulthood[46].
Since alternative splicing of flip/flop exons is a common feature in all AMPAR genes, we asked whether similar changes also occurred in the Gria1, Gria3, and Gria4 genes. Quantification result showed that flip/flop ratio of Gria1, Gria3, and Gria4 genes was significantly reduced in the cortex of Mecp2 KO mice (Fig 4d), implicating a biased usage of flop exon in the mature transcripts of all AMPAR genes. Importantly, the total mRNA level of Gria1 was unchanged and only subtle trend of decreasing Gria3 and Gria4 mRNA level was observed in Mecp2 KO mice (Fig 4e).
Interestingly, analysis of RNA-Seq data from visual cortex and hypothalamus of Mecp2 KO mice also showed that percentage of flip isoform is significantly decreased (S8 Fig). Note that these two studies used the Bird allele (Mecp2tm1.1Bird) and our data was generated from the Jaenisch allele (Mecp2tm1.1jae). The consistent flip/flop splicing changes across different brain regions from different knockout mouse lines suggested that reduction of flip/flop ratio is a common defect due solely to loss of MeCP2. More importantly, we also found that the percentage of flip isoform in the hypothalamus of Mecp2 OE mice was significantly increased, which is opposite to the changes in Mecp2 KO (S8 Fig). Together, these results strongly suggest that MeCP2 directly modulates the regulation of Gria2 flip/flop splicing.
Finally, we tested whether splicing alteration of AMPAR genes also occurs in the cortex of heterozygous female Mecp2-/+ mice. Although not as drastic as that observed in Mecp2 KO male mice, Mecp2-/+ mice also displayed a significant reduction of flip/flop ratio in Gria1, Gria2, Gria3, and Gria4 (Fig 4f). Similar to Mecp2 KO male mice, Mecp2-/+ mice had unchanged total mRNA level in all four AMPAR genes (Fig 4g). These data suggest similar change in flip/flop usage may exist in female RTT patients.
To determine how loss of MeCP2 affects the splicing of flip/flop exon, we focused on the Gria2 gene to explore the potential involvement of several recent models of splicing regulation. Modulation of PolII elongation rate has been proposed as one model of how epigenetic mechanisms influence splicing. Slow PolII elongation rate allows longer time for spliceosome to assemble and hence increase the chance of the alternative exon being included in the mature transcript[47]. A recent study suggested that MeCP2 is enriched in particular alternative exons and facilitates exon inclusion by pausing PolII in cultured cells[48]. We set out to test whether MeCP2 regulates Gria2 flip/flop splicing through similar mechanism in the brain. Chromatin immunoprecipitation (ChIP) followed by qRT-PCR showed a significant enrichment of MeCP2 on the flip and flop exons of Gria2 gene (Fig 5a). However, no significant difference in PolII occupancy on the flip and flop exons between the wild type and Mecp2 KO mice was found by PolII ChIP (Fig 5b), suggesting the involvement of a PolII-independent mechanism underlying the flip/flop splicing change in Mecp2 KO brain. Another interesting epigenetic model for alternative splicing regulation is that histone modification can be bound by adaptor proteins which in turn recruit specific splicing factor to alternative exons[47]. It has been previously shown that trimethylated histone H3 lysine 36 (H3K36me3) is enriched on exons and can be bound by LEDGF, which recruits splicing factors such as SRSF1 to regulate splicing[49]. Although significant LEDGF occupancy was detected on the flip and flop exons (Fig 5c), no significant difference in the occupancy of H3K36me3 on the flop and flip exons was detected between the wild type and Mecp2 KO brain (Fig 5d).
To determine whether LEDGF is functionally involved in the regulation of Gria2 flip/flop splicing, we tested the effect of knockdowning LEDGF on flip/flop ratio in a neuroblastoma cell line, Neuro-2A, using a Gria2 minigene. The Gria2 minigene spans the genomic region from exon 13 to exon 15 of Gria2 (S9a Fig, either the flip or flop exon can be included as exon 14). As a control, we co-transfected Mecp2 shRNA, Gria2 minigene along with a MeCP2 overexpression plasmid into Neuro-2a cells and found that flip/flop ratio is significantly reduced upon Mecp2 knockdown (Fig 5e and 5f), indicating that this artificial assay is capable of discovering factors that potentially affect flip/flop splicing. Next, we transfected a Ledgf shRNA in the cells and tested its effect on flip/flop splicing. Similar to Mecp2 knockdown, Ledgf knockdown also leads to a reduction of flip/flop ratio (Fig 5g and 5h), suggesting that LEDGF is functionally involved in the regulation of Gria2 flip/flop splicing.
Flip/flop exon encodes a 38 amino acids sequence in the ligand binding domain of AMPARs that controls desensitization rate. Compared to flip-containing receptors, flop-containing receptors desensitize with faster kinetics[50, 51]. To determine the functional consequence of altered flip/flop splicing in the cortex of Mecp2 KO mice, we performed outside-out patch clamp recording of glutamate-evoked current on layer 2/3 pyramidal neurons in acute brain slices. We found that the decay time constant τ was significantly reduced in Mecp2 KO mice (Fig 6a and 6b). In addition to evoked response, regular whole cell patch clamp recording of spontaneous synaptic events also detected a faster decay in miniature excitatory postsynaptic current (mEPSC) in layer 2/3 pyramidal neurons from the Mecp2 KO mice. Finally, bath application of cyclothiazide (CTZ), a positive allosteric modulator of AMPARs that inhibits desensitization of AMPARs[52], slowed down the decay kinetics in Mecp2 KO slice to a comparable level of wild type cells (Fig 6c and 6d). Together, these results uncover a previously unappreciated defect of faster desensitization kinetics of AMPAR-gated current in the Mecp2 KO mice, which correlates with altered flip/flop splicing and can be modulated by pharmacological reagents.
To causally link the change in flip/flop splicing and the altered AMPAR desensitization kinetics, we used engineered splicing factors (ESF)[53, 54] to specifically manipulate flip/flop splicing in the brain of Mecp2 KO mice. ESF is composed of a sequence-specific RNA-binding domain derived from human Pumilio1 (PUF domain) and a functional domain that suppresses (Gly domain) or enhances (SR domain) inclusion of a specific exon. We evaluated the effect of four ESFs (ESF-flop-Gly [flop suppressor], ESF-flop-SR [flop enhancer], ESF-flip-Gly [flip suppressor] and ESF-flip-SR [flip enhancer]) on flip/flop splicing using a Gria2 minigene (S9a Fig). We found that ESF-flop-Gly significantly increased the flip/flop ratio (Fig 6e), an effect opposite to the change we observed in the cortex of Mecp2 KO mice. Moreover, ESF-flop-Gly didn’t change the level of total Gria2 minigene (Fig 6f).
To further test the effect of ESF-flop-Gly on flip/flop splicing of the endogenous Gria2 transcript in neurons, we infected primary cortical neurons with adeno-associated virus (AAV) encoding either mCherry alone or ESF-flop-Gly and mCherry. As expected, AAV-ESF-flop-Gly-mCherry significantly altered the flip/flop splicing balance to favor the use of the flip exon (S9b and S9c Fig), suggesting that ESF-flop-Gly could be used in vivo to reverse the flip/flop splicing defect in Mecp2 KO mice. To that end, we injected lentivirus expressing ESF-flop-Gly into the cortex of Mecp2 KO mice, and measured the decay time constant τ of glutamate-evoked AMPAR-gated current in the outside-out patch clamp mode in acute brain slices 2 weeks post injection. Compared to neurons infected with control virus (KO+Ctrl), ESF-flop-Gly expressing neurons (KO+ESF) had a significant larger decay time constant τ, which was indistinguishable from that of WT cells (Fig 6g). These results strongly suggest that altered flip/flop splicing is required for a specific synaptic phenotype in the Mecp2 KO mice.
To further examine the effect of altered flip/flop splicing on synaptic transmission, we applied repetitive stimulation on the neurons in an interval of 100 ms and recorded the AMPAR-gated current. Upon repetitive stimulation, a fraction of AMPA receptors desensitizes and the short interval between stimulation does not allow full recovery. As a result, fewer AMPA receptors can respond to the subsequent stimulation and therefore current diminished. Comparing to WT neurons, KO neurons displayed even more drastic decrease in current amplitude over the course of five stimulations (Fig 6h and 6i). This difference could be partially due to a higher percentage of flop isoform that are more easily desensitized in the KO neurons. Consistent with this hypothesis, overexpressing ESF-flop-Gly in KO neurons partially rescued this phenotype (Fig 6h and 6i). These data suggest the altered flip/flop splicing ratio has important impact on synaptic transmission in the Mecp2 KO cortex, which can be reversed by ESF designed to specifically target flip/flop exons.
To determine the functional outcome of Ledgf knockdown-induced change in Gria2 flip/flop splicing, we injected lentivirus encoding shLedgf and control shRNA into the cortex of wild type mice. We found that Ledgf knockdown resulted in a significantly reduced decay time constant τ of glutamate-evoked AMPAR-gated current (Fig 6j), an effect similar to that caused by the loss of MeCP2 (Fig 6a and 6b). In addition, Ledgf knockdown led to significantly weaker response upon repetitive stimulations (Fig 6k and 6l), another phenotype caused by the loss of MeCP2 (Fig 6h and 6i). These data correlate well with the Gria2 minigene assay in Fig 5 and further support that both LEDGF and MeCP2 are required for the normal splicing of Gria2 flip/flop exons.
MeCP2 has been previously implicated in regulating alternative splicing of RNA in two studies. In 2005, Young et al reported RNA-dependent interaction between MeCP2 and YB1 in a neuroblastoma cell line forced to overexpress MeCP2 and some changes in alternative splicing in the Mecp2308/y brain[23]. In 2013, Maunakea et al reported intragenic DNA methylation-dependent MeCP2 binding to alternatively spliced exons in cancer cell lines[48]. Our work substantially extends these previous studies in several ways, and to our knowledge, this is the first report of the functional consequence for MeCP2-mediated splicing.
First, the physical interaction between MeCP2 and its interacting partners identified in our study are independent of any nucleic acid, suggesting that MeCP2 does not need to bind to RNA in order to regulate splicing. Additionally, these physical interactions have more physiological relevance, because they were identified in the mouse brain where MeCP2 is expressed from its endogenous locus. Furthermore, we identified multiple splicing factors as novel MeCP2-interacting partners in the brain. Since these factors are not part of the core splicing machinery but rather affect splicing as accessory splicing factors[55], the biochemical mechanism underlying their involvement in splicing regulation is not well known. Their interaction with MeCP2, a known chromatin protein, provides novel clues for studying how these factors regulate splicing. In addition, because we used a different RTT mouse model (Mecp2 KO mice in our study vs. Mecp2308/y mice in Young et al[23]) and a more sensitive method to profile alternative splicing (RNA-seq vs. microarray), the altered splicing events identified in our study were different from those previously identified by Young et al[23]. Nonetheless, combining results from three independent unbiased approaches (Co-IP mass spectrometry, RNA-seq and ChIP-seq), our study provides strong evidences for a significant involvement of MeCP2 in regulating RNA splicing.
Second, we discovered significant MeCP2 occupancy around exon/intron boundary and exons in the mouse brain, and characterized gene exon specific interaction between MeCP2 and two splicing regulators, providing a potential mechanism for MeCP2-dependent splicing regulation. Recent evidence suggests intragenic DNA methylation recruits MeCP2 and regulates pre-mRNA splicing through altering DNA polymerase II elongation rate[48]. However, our data suggests that it is not responsible for the altered flip/flop splicing in the cortex of Mecp2 KO mice. Instead, our results suggest a new model that co-occupancy of MeCP2 and LEDGF on the chromatin is required for the normal flip/flop splicing in the Gria2 gene.
Finally, and most importantly, we established a functional link between specific splicing changes caused by the loss of MeCP2 function to synaptic changes in RTT mice. The fact that a ESF specifically rescues the flip/flop splicing defect can reverse the corresponding synaptic changes in RTT brain strongly suggest that the specific change in synaptic property (AMPAR kinetics) is caused by altered flip/flop splicing. Given the central role of AMPARs in synaptic transmission, it is likely the altered AMPAR kinetics will lead to altered synaptic functions other than the repetitive stimulation paradigm employed in our study. Future study is needed to mechanistically link the altered AMPAR kinetics with specific neuronal defects in RTT symptoms, and to evaluate the effect of reversing flip/flop splicing on RTT disease progression. In addition to the flip/flop choice in AMPARs, alterative splicing of several other genes (e.g. Nrxn1, Dscam, lin7a) that play important roles in synaptic functions were changed in the RTT mouse cortex, indicating that additional synaptic changes may be caused by splicing deficits. Thus, altered RNA splicing appears to be a novel molecular mechanism underlying synaptic dysfunction in RTT.
Splicing misregulation has been increasingly recognized as a significant contributor to a number of neurological diseases, such as SMA[56], FTDP-17[27], ALS[57] and myotonic dystrophy[58]. The mechanistic study of how the genes mutated in neurological diseases can directly affect alternative splicing, as well as the functional consequences of splicing alteration in such diseases, will have important implications in human health. Our study adds to the growing list of studies on the novel links between specific events of altered splicing and neurological diseases.
All animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee at the University of Wisconsin-Madison. All mice in this study were euthanized by CO2 asphyxiation, according to the guidelines of the RARC at the University of Wisconsin-Madison and the recommendations of the Panel on Euthanasia of the American Veterinary Association.
The Mecp2-Flag mice have a Flag sequence inserted intermediately before the stop codon of the Mecp2 locus[34]. The Mecp2 KO mice used in this study are the Jaenisch strain (Mecp2tm1.1jae)[59]. Mice were housed in a facility with 12-hr light/12-hr dark cycle.
pRK5-HA-MeCP2-WT, Δ1–77, Δ1–162, Δ78–162 and Δ381–492 were a gift from Dr. Zilong Qiu[25]. DNA encoding Δ163–380, MeCP2R168X, MeCP2R255X, MeCP2R270X and MeCP2R290X were PCR amplified and inserted into pRK5-HA by replacing the sequence between SalI site and NotI site of pRK5-HA-MeCP2 using Gibson cloning (NEB). To construct Myc-tagged protein expression plasmid, cDNA of LEDGF/p52, LEDGF/p75, and DHX9 were amplified from a mouse cortex cDNA library using a Myc sequence-containing primer and inserted into pRK5 backbone. LEDGF is also known as Psip1. Engineered splicing factor (ESF) were designed to target the Flip (GCCAAGGA) and the Flop (GCAGCGGG) exons[38]. Gria2 minigene was constructed by amplifying Gria2 Exon13 to Exon15 from mouse genomic DNA and inserted into pEGFP-C1 backbone. The Ledgf shRNA (shLedgf) target sequence (5’-GCA GCT ACT GAA GTC AAG ATT C-3’) was adapted from a previous study[60] and cloned into pLL3.7 backbone. The Mecp2 shRNA (shMecp2) construct was used in our previous study[61]. A scrambled sequence (5’-GGA ATC TCA TTC GAT GCA TAC-3’) was used as negative control (shCtrl).
Nuclei were extracted from the whole brain of WT and Mecp2-Flag mice as previously described[62]. Purified nuclei were resuspended in lysis buffer containing 20mM Tris, 150mM NaCl, 1.5mM MgCl2, 1mM EDTA, 10% Glycerol, 0.2% NP-40 and 1X proteinase inhibitors cocktail (Roche) and sonicated using a Misonix 3000. After centrifuging at 20,000g for 20min at 4°C, supernatant was incubated with 50ul of Anti-Flag M2 Magnetic Beads (Sigma) overnight at 4°C. In the following day, beads were washed with lysis buffer for 6 times. Bound protein was eluted by competition with 100 mg/ml of Flag Peptide (Sigma F3290). Eluted proteins from 5 IPs per genotype were pooled together and precipitated by adding 8 volume of pre-chilled acetone. Pellet was resuspended in 100mM Ammonium Bicarbonate solution. After DTT and IOAA treatment, protein was digested into peptides using Trypsin Gold (Promega) and Proteinase Max (Promega) overnight at 37°C. Peptides were separated by a nano HPLC and analyzed by a Thermo LTQ mass spectrometer. MS/MS spectra data was analyzed using Bioworks software (Thermo). Only proteins identified in Flag IP eluate from Mecp2-Flag mice but not WT mice were considered to be potential MeCP2-interacting proteins.
Co-IP was performed as described above except using Dynabeads (Life Technologies). For co-IP with Benzonase treatment, lysate was treated with 250 Unit of Benzonase per mouse brain for 1hr at 4°C before incubating with beads. Proteins were eluted by adding 1X LDS sample buffer (Life Technologies) and heated at 95°C for 10min.
Proteins were resolved in a 10% SDS-PAGE gel and transferred into a nitrocellulose membrane. Membrane was blocked with 5% non-fat milk in PBS for 1 hour followed by incubating with primary antibody overnight at 4°C. Membrane was washed 3 times with PBST and incubated with DyLight Fluor Secondary Antibodies (Pierce) for one hour at room temperature. Membrane was imaged on a LI-COR Odyssey Imager. Western blot quantification was done using ImageJ. Primary antibodies used in this study were: anti-DHX9 (Abcam ab26271, 1:2000), anti-FLAG (Sigma M2, 1:1000), anti-FUS (Bethyl A300-293A, 1:10000), anti-HA (Covance MMS-101P, 1:5000), anti-hnRNP F+H (Abcam ab10689, 1:3000), anti-LEDGF (Bethyl A300-847A, 1:1500), anti-MeCP2 (Abcam ab50005, 1:2000), anti-Myc (Cell signaling 71D10, 1:1000), and anti-TDP-43 (ProteinTech 10782-2-AP, 1:1000).
HA-MeCP2 construct was co-transfected with Myc-LEDGF or Myc-DHX9 (1:1 ratio) into HEK293 cells using GenJet transfection reagent (Signagen). 24 hours after transfection, cells were washed with PBS twice and directly lysed with Pierce IP Lysis Buffer (Thermo Scientific) for 10min on ice. Lysate was centrifuged at 16,000g for 10min at 4°C and pellet was discarded. Six hours before lysate preparation, 30ul Dynabeads protein G was incubated with 3ug of anti-HA (Covance) or anti-Myc (Millipore) at 4°C to form the antibody-proteinG-bead complex. After washing off excess antibody, beads were incubated with lysate overnight at 4°C. Beads were washed with lysis buffer 6 times and then eluted by adding 1X LDS sample buffer (Life Technologies) and heated at 95°C for 10min.
Total RNA was extracted from cortices of 6-weeks-old WT and Mecp2 KO mice using Qiagen RNeasy Mini Plus kit. Genomic DNA was removed by a gDNA Eliminator column. 150ng total RNA was used to prepare sequencing library according to manufacturer’s instructions (Nugen Encore Complete). Each Library was subject to one lane of 100bp single end sequencing using Illumina Hi-Seq 2000. Reads were mapped to the mouse genome (mm9) using Tophat (2.0.8). Reads count for each gene was calculated using htseq-count function in the HTSeq package. Differential gene expression analysis was done using edgeR in R. Splicing analysis was performed using the Mixture of Isoforms pipeline (MISO 0.4.7). Considering the high similarity of the two replicates for each genotype (Correlation = 0.98 for each), reads from two replicates were combined for each genotype and processed with MISO. A stringent filter (total reads for the event ≥ 1000, reads supporting inclusion or exclusion isoform ≥ 50, total reads supporting inclusion and exclusion isoform ≥ 100, |ΔPSI| ≥ 0.20 and Bayes-factor ≥ 20) was used to generate a list of differential splicing events. Read density plot was generated using sashimi plot built in MISO.
RNA-Seq data from Chen et al[35] and Gabel et al[9] were processed as above for splicing analysis. A less stringent filter (total reads for the event ≥ 20, |ΔPSI| ≥ 0.05 and Bayes-factor ≥ 1) was applied to allow for generating more events for further overlap analysis.
Gene Ontology (GO) analysis was done using DAVID[63]. Briefly, official gene symbols were submitted to DAVID. We used our own RNA-seq data and applied a cutoff of RPKM ≥ 0.5 to generate a list of genes expressed in the mouse cortex (13846 genes). This set of genes expressed in the mouse cortex was used as background for all GO analysis in this manuscript. Terms with Benjamini adjusted P-value < = 0.05 was considered as significant.
Total RNA was extracted from cortices of 6-8-week-old wild type (WT) and Mecp2 KO male mice or 15-18-month-old WT and Mecp2 KO female mice using Qiagen RNeasy Mini Plus kit with on-column DNase treatment. RNA extraction from HEK293 or N2A cells was performed using TRIzol (Life Technology). RNA was reverse transcribed into cDNA using qScript cDNA SuperMix (Quanta Biosciences). qPCR was performed on an ABI Step-One plus machine using SYBR Green qPCR Master Mix (Biotool). Gapdh was used as endogenous control and 2−ΔCt method was used to calculate fold change. See S8 Table for primer sequence.
Chromatin immunoprecipitation (ChIP) was performed as previously reported[29]. Briefly, cortex tissue was dissected from 6-8-week-old mice, minced and crosslinked in 1% formaldehyde (wt/vol) and sonicated using a Misonix 3000. Antibody was first bound to Dynabeads and then incubated with sheared chromatin overnight at 4°C. After 4 washes with RIPA buffer and 1 wash with TE buffer, bound chromatin was eluted and reverse crosslinked at 65°C overnight. Eluted DNA was treated with RNase A (Thermo Scientific) and proteinase K (Promega), purified by phenol-chloroform extraction and dissolved in water. Antibodies used were: anti-Flag (Sigma M2), anti-LEDGF (Bethyl A300-847A), anti-H3K36me3 (Abcam ab9050), and anti-PolII (Abcam ab5408). Primer sequence for ChIP-qPCR is provided in S8 Table.
ChIP-Seq data were generated from two biological replicates (referred to as WT1 and WT2). Raw data was aligned to the mouse genome version mm9 with Bowtie (0.12.7). After excluding non-mapping reads, we had 72, 221, 924 reads for WT1 ChIP and 31, 333, 769 for its input and 84, 871, 157 reads for WT2 ChIP and 22, 412, 408 for its input. We firstly evaluated the quality of these data with respect to ENCODE’s ChIP-seq quality control metrics[64]. The Normalized Strand Cross Correlation (NSC) for WT1 ChIP and WT2 ChIP is 1.3 and 1.4, respectively. Another quality control measure is PCR Bottleneck Coefficient (PBC), which gives an estimate of the complexity of the ChIP-seq library[65]. PBC<0.5 indicates PCR bottlenecks are present in sequenced libraries. The PBC ranged within [0.63 0.83] across WT1 ChIP sample and [0.85, 0.94] for the WT1 input sample. Similarly, the PBC ranged within [0.63 0.83] across WT2 ChIP sample and was 0.93 for the WT2 input sample. These numbers suggest our libraries were of good quality.
We carried out peak calling using MOSAiCS package in R[66] using default parameters except for fdrRelaxed = 0.1 for WT1 and WT2 and fdrRelaxed = 0.2 for pooled replicates. Bin and fragment sizes were set to 200 bps for all the runs. We followed a conservative strategy and obtained peaks for individual replicates at false discovery rate of 0.1 and for pooled sample run at 0.2. Then, we identified the peaks in the intersection of the three peak lists and filtered them with mosaics parameters: logMinP > = -log10(0.05) & peakSize > = 150 & aveLog2Ratio > = log2(1.5). This resulted in a total of 20, 652 peaks with median size of 1731 bps. We performed location analysis using mm9 Refseq genes and the nomenclature in Blahnik et al[67].
The previously published independent datasets were used[42]. DNA methylation data in the frontal cortex of adult mouse (10-wk-old) were downloaded under accession number GSM 1173784. Each context of the cytosine methylation and the two following bases from the same strand was considered independently: CG, CHG or CHH (where H = A, C or T). To determine the frequency of each context, the frequency of the cytosine methylation of each context in MeCP2 ChIP-seq peaks was estimated as the average of ratio /(x) = nm(x)/ntot(x), where nm(x) is the number of reads supporting a methylated cytosine at position x and ntot(x) is the total number of reads at that position.
ESF construct was co-transfected into 293 cells with Gria2 minigene (8:2 ratio) using GenJet transfection reagent (Signagen). To test the effect of Mecp2 knockdown on Gria2 splicing, shMecp2, Mecp2 overexpression construct and Gria2 minigene (4.5:4.5:1 ratio) was co-transfected into N2A cells with using GenJet. To test the effect of Ledgf knockdown on Gria2 splicing, shLedgf construct and Gria2 minigene (9:1 ratio) was co-transfected into N2A cells using GenJet. Cells were lysed in TRIzol 48 hours after transfection for qRT-PCR analysis.
Male mice at 4–6 weeks postnatal were used. Coronal brain slices (400 μm) were prepared in ice-cold modified artificial cerebrospinal fluid (aCSF) (in mM: 124 NaCl, 2.5 KCl, 1 CaCl2, 2 MgSO4, 1.25 NaH2PO4, 26 NaHCO3, and 15 glucose) bubbled with 95%O2/5%CO2. Then the slices were incubated in normal aCSF (in mM: 124 NaCl, 2.5 KCl, 2.5 CaCl2, 1.2 MgSO4, 1.25 NaH2PO4, 25 NaHCO3, and 15 glucose) at room temperature for at least 1 hour and then transferred to a submerged recording chamber perfused with 95%O2/5%CO2 saturated aCSF for electrophysiological recordings. Whole-cell recording of mEPSCs and outside-out patch recording of glutamate-evoked currents was performed from the Layer 2/3 pyramidal neurons at room temperature. TTX (1 μM), D-APV (20 μM), bicuculline (50 μM) were added into the perfused aCSF to block voltage gated Na+ channels, NMDA receptors and GABA receptors respectively. The patch pipette (3–4 MΩ) solution contained (in mM): 140 Cs-Gluconate, 7.5 CsCl, 10 HEPES, 0.5 EGTA-Cs, 4 Mg-ATP, and 0.3 Li-GTP, pH 7.4. Raw data were amplified with a Multiclamp 700B amplifier and acquired with pClamp10.2 software (Molecular Devices). Neuronal currents were recorded under voltage clamp at the holding potential of -70 mV. An ALA fast perfusion system was used to perform application of glutamate (10 mM). In some experiments, CTZ (50 μM) was added. The detection of mEPSCs and exponential fitting were performed using Clampfit 10.2. The decay of glutamate evoked currents was fitted with double-exponential functions, and the fast- and slow- time constant were obtained. Signals were filtered at 2 Hz and sampled at 10 kHz by Digidata 1440A (Molecular Devices). mEPSCs were analyzed using the Template Search tool of the Clampfit10.2. To create the template, several well-shaped mEPSCs traces were picked and averaged to the template window. The mEPSCs events were accepted manually. Amplitude and the weighted time constant of decay phase of both mEPSCs and glutamate evoked currents were acquired.
To investigate whether enhanced depression of AMPAR responses to burst-type stimulations is expressed at synapses, we recorded excitatory postsynaptic potentials evoked through a bipolar stimulating electrode (FHC Inc.) placed in the white matter (eEPSCs, five pulses at 10 Hz). AMPAR-mediated eEPSCs were recorded in the presence of D-APV (20 μM) and bicuculline (50 μM) at a holding potential of -70 mV. The data was analyzed with Clampfit 10.
Lentivirus preparation was performed as described[68] except that we use minimum amount of media (leftover in the tubes) to resuspend the virus. Stereotaxic injection was done as previously described[61].
Custom AAV encoding ESF was generated by Vigene. DIV7 primary cortical neurons were infected with AAV at a MOI of 105. AAV was removed 48 hours after infection and cells were collected 7 days after infection for qRT-PCR analysis.
No statistical procedure was used to predetermine sample size. Student’s t-test was used to compare means between two groups. Multiple t-test comparisons were corrected using Benjamini-Hochberg procedure. One-way ANOVA followed by Tukey’s multiple comparison tests was used to test difference in experiments with multiple groups. Two-way ANOVA with repeated measure followed by Bonferroni's multiple comparisons test was used for analysis in the repetitive stimulation experiments. Statistical calculation was performed using Microsoft Excel and Graphpad Prism.
|
10.1371/journal.pgen.1004923 | ASAR15, A cis-Acting Locus that Controls Chromosome-Wide Replication Timing and Stability of Human Chromosome 15 | DNA replication initiates at multiple sites along each mammalian chromosome at different times during each S phase, following a temporal replication program. We have used a Cre/loxP-based strategy to identify cis-acting elements that control this replication-timing program on individual human chromosomes. In this report, we show that rearrangements at a complex locus at chromosome 15q24.3 result in delayed replication and structural instability of human chromosome 15. Characterization of this locus identified long, RNA transcripts that are retained in the nucleus and form a “cloud” on one homolog of chromosome 15. We also found that this locus displays asynchronous replication that is coordinated with other random monoallelic genes on chromosome 15. We have named this locus ASynchronous replication and Autosomal RNA on chromosome 15, or ASAR15. Previously, we found that disruption of the ASAR6 lincRNA gene results in delayed replication, delayed mitotic condensation and structural instability of human chromosome 6. Previous studies in the mouse found that deletion of the Xist gene, from the X chromosome in adult somatic cells, results in a delayed replication and instability phenotype that is indistinguishable from the phenotype caused by disruption of either ASAR6 or ASAR15. In addition, delayed replication and chromosome instability were detected following structural rearrangement of many different human or mouse chromosomes. These observations suggest that all mammalian chromosomes contain similar cis-acting loci. Thus, under this scenario, all mammalian chromosomes contain four distinct types of essential cis-acting elements: origins, telomeres, centromeres and “inactivation/stability centers”, all functioning to promote proper replication, segregation and structural stability of each chromosome.
| Mammalian cells replicate their DNA along each chromosome during a precise temporal replication program. In this report, we used a novel “chromosome-engineering” strategy to identify a DNA element that controls this replication-timing program of human chromosome 15. Characterization of this element indicated that it encodes large non-protein-coding RNAs that are retained in the nucleus and form a “cloud” on one copy of chromosome 15. Previously, we found that structural rearrangements of a similar element on human chromosome 6 causes delayed replication and structural instability of chromosome 6. Mammalian chromosomes are known to contain three distinct types of essential DNA elements that promote proper chromosome function. Thus, every chromosome contains: 1) origins of replication, which are responsible for proper initiation of DNA synthesis; 2) centromeres, which are responsible for proper chromosome separation during cell division; and 3) telomeres, which are responsible for replication and protection of the ends of linear chromosomes. Our work supports a model in which all mammalian chromosomes contain a fourth type of essential DNA element, the “inactivation/stability center”, which is responsible for proper DNA replication timing and structural stability of each chromosome.
| Morphological differences between chromosomes residing within the same cell were first observed in mammalian cells nearly fifty years ago (reviewed in [1]). The phenomenon of “chromosome pulverization” was first described in human cells, where infection with the measles virus causes a severe fragmentation that gives the appearance that the chromosomes have been “pulverized” [2], [3]. Subsequently, abnormal chromosome condensation was shown to occur when viruses cause mitotic cells to fuse with interphase cells [4]. This premature chromosome condensation (PCC) phenotype was found to affect one of the two complete sets of chromosomes present in heterokaryons when mitotic cells are fused with other cells in different phases of the cell cycle. The PCC is most dramatic when S-phase cells were fused to mitotic cells, suggesting that condensation of partially replicated chromosomes causes the pulverized appearance [5]. In contrast, a morphologically similar abnormal chromosome condensation phenotype was observed on one or a few chromosomes during mitoses of cancer cell lines [6]–[8], primary tumor cells [7], and in cells exposed to mitotic spindle poisons [9]–[11] or DNA damage [12]–[15]. Moreover, we found that certain tumor-derived rearranged chromosomes exhibit a delay in replication timing (DRT), which is characterized by a>2 hour delay in the initiation and completion of DNA synthesis along the entire length of the chromosome [7]. Chromosomes with DRT also display a delay in mitotic chromosome condensation (DMC), which is characterized by an under-condensed appearance and a concomitant delay in phosphorylation of serine 10 of histone H3 [7], [16]. DRT/DMC chromosomes were also detected in 5 of 7 tumor cell lines and in 5 of 13 primary tumor samples, indicating that DRT/DMC is common in human cancer cells in vitro and in vivo [7]. Subsequently, we found that chromosomes with DRT/DMC are present in as many as 25% of cells surviving exposure to ionizing radiation (IR), and that IR induced DRT/DMC on mouse chromosomes can persist for up to 2 years in vivo [15]. Importantly, DRT/DMC occurred predominantly on chromosome translocations, estimated to be ∼5% of all translocations induced by IR, indicating that structural rearrangement is responsible for the DRT/DMC phenotype [15].
We have developed a chromosome engineering system that allows for the systematic analysis of chromosomes with DRT/DMC [15]–[19]. Our strategy is based on the Cre/loxP system to generate precise chromosomal rearrangements. Using this system we previously identified five balanced translocations, affecting eight different autosomes, each displaying DRT/DMC on at least one of the two derivative chromosomes [17]. Subsequently, we found that translocations or deletions at a discrete cis-acting locus on human chromosome 6 result in DRT/DMC. Characterization of this locus identified a large intergenic non-coding RNA (lincRNA) gene, which we named ASynchronous replication and Autosomal RNA on chromosome 6, or ASAR6 [19]. The ASAR6 gene displays random monoallelic expression, and as the name implies displays asynchronous replication that is coordinated with other monoallelic genes on chromosome 6 [18], [19]. The ASAR6 RNA is a long (>200 kb), non-spliced, non-polyadenylated, RNA Polymerase II product that is retained within the nucleus at or near the site of transcription [18], [19]. In addition, disruption of the expressed allele of ASAR6 results in transcriptional activation of previously silent alleles of other monoallelic genes on chromosome 6 [19].
In the current report we show the characterization of a cis-acting locus present at the breakpoint of a second balanced translocation identified in our original screen for DRT/DMC chromosomes [17]. In this report, we found that deletions, or inversions at a complex locus on chromosome 15q24.3 cause DRT/DMC and structural instability of human chromosome 15. This locus encodes the large protein-coding gene SCAPER, the micro RNA gene MIR3713, and long RNA transcripts that are retained within the nucleus and form a “cloud” on one homolog of chromosome 15. Because this locus also displays asynchronous replication that is coordinated with other random monoallelic genes on chromosome 15, we have named this locus ASynchronous replication and Autosomal RNA on chromosome 15, or ASAR15.
Our chromosome engineering strategy is based on the Cre/loxP site-specific recombinase system to generate chromosomal rearrangements (reviewed in [20]). Because Cre/loxP is relatively inefficient at mediating inter-chromosomal events, we are using reconstitution of a selectable marker to isolate the cells that undergo Cre-mediated recombination (see Fig. 1A). Thus, following Cre-mediated recombination and selection for reconstitution of the Aprt gene, reciprocal exchanges are generated in recombinant clones (R-lines) of Aprt+ cells [17]. Fig. 1B illustrates the balanced translocation generated by Cre expression in the parental P268 cell line, which was identified during our original screen for DRT/DMC [17]. The balanced translocation generated in the resulting recombinant (R268) cells involves the long arms of chromosomes 15 and 16, and the chromosome 15 derivative displays the DRT/DMC phenotype (Fig. 1C and D).
Using our chromosome engineering system we previously found that Cre/loxP-mediated translocations and subsequently deletions affecting the ASAR6 gene result in DRT/DMC on chromosome 6 [17]–[19]. Using a similar strategy we next determined if deletions in either chromosome 15 or 16 could cause DRT/DMC in the P268 cell line. For this analysis we used a random Lentiviral integration approach to introduce new loxP cassettes into P268 cells. The rationale behind this set of experiments is based on previous studies showing that intra-chromosomal Cre events are more efficient than inter-chromosomal Cre events [21]. Thus, expression of Cre in pools of Lentiviral infected cells reconstitutes Aprt at a higher efficiency in cells containing Lentiviral integrations near the original loxP cassettes, and depending on the orientation of the newly integrated cassettes either deletions or inversions are generated following Cre transient expression [19]. Using this random integration approach we generated 14 different deletions in chromosome 15, and 9 different deletions in chromosome 16 (in Δ268 clones). Note that this approach resulted in nested deletions anchored at the original loxP cassettes in both chromosomes 15 and 16 (see Fig. 1E and Table 1). In addition, this approach allowed us to isolate multiple independent clones containing the same deletions. Thus, for the majority of the deletions characterized in this study we analyzed multiple independent clones representing the same Cre/loxP-mediated deletions (see S1 Table).
To characterize the Cre/loxP deletions in molecular detail, we first cloned and sequenced the original AP-loxP and loxP-RT cassette integration sites in P268 genomic DNA. The AP-loxP cassette is at position 54,407,235 base pairs of chromosome 16, and the loxP-RT cassette is at position 76,858,743 base pairs of chromosome 15 (NCBI Build 37/hg19). Subsequent analysis indicated that deletions in chromosome 15 displayed DRT/DMC (see below), while deletions in chromosome 16 did not. Therefore, we have concentrated on the chromosome 15 loxP-RT integration site in this report. In addition, because our random loxP integration approach generates deletions in only one direction (e.g. distal on chromosome 15; see Fig. 1E), we also modified the original loxP-RT integration site so that deletions could be generated proximal to the original loxP cassette (see [19] for details). Finally, we also characterized three different inversions in chromosome 15, which were generated by Lentiviral integrations of the loxP cassettes in the opposite orientation with respect to the original loxP-RT cassette integration site (see [21]).
We characterized the Cre-dependent rearrangements in chromosome 15 at the genetic level using multiple independent assays, including: Southern blot hybridizations (S1 Fig.); PCR designed to span the genome-loxP-RT cassette junctions (S2A Fig. and B); LAM-PCR [22] to identify the Lentiviral 5′-LTR integration sites (Table 1; see [19]); PCR designed to span the Lentiviral-genome junctions (S2C Fig.); SNP analysis for loss of heterozygosity (S2D Fig.); and DNA FISH using BACs or fosmids located within the deleted or inverted regions (see Figs. 1E, 2D–G, 3B–C, 4A–C, S3A and C, S6A, and S8B and D Figs.). The largest deletion distal to the original loxP-RT cassette was ∼12.8 megabases (mb), and the smallest distal deletion was only ∼2 kilobases (kb) (see Table 1). The deletions proximal to the loxP-RT integration site did not cause DRT/DMC, and therefore were characterized genetically for loss of heterozygosity and with DNA FISH with a BAC from the deleted region (CTD-2117F7; see Fig. 1E), which allowed us to only determine the approximate sizes of the proximal deletions. Thus, the largest proximal deletion was ∼18 mb, and we characterized two different relatively small, ∼67–126 kb, proximal deletions. The genomic locations of all of the deletions and inversions characterized in this study are shown in Table 1. A schematic illustration of the loxP integration sites that generated five different distal deletions and two different inversions, all inducing DRT/DMC on chromosome 15, is shown in Fig. 1E.
One complication in the analysis of chromosomes with DRT/DMC is structural instability of the affected chromosomes [7], [17], [18]. We previously used a “Luria-Delbruck fluctuation analysis” to characterize the instability associated with engineered chromosomes with DRT/DMC, and found that the t (15;16) generated in the P268 cell line results in an ∼80 fold increase in the rate of structural rearrangement of the affected chromosome [17]. In addition, once the DRT/DMC chromosomes become extensively rearranged they no longer display DRT/DMC [7], [16], [17]. Not surprisingly, the cells with DRT/DMC on chromosome 15, described in this report, acquired extensively rearranged chromosome 15 s and eventually lost the DRT/DMC phenotype during continued expansion in culture. Another complication in the analysis of DRT/DMC chromosomes is that they often experience non-disjunction events, resulting in cells that have either lost or gained the DRT/DMC chromosomes [16]. Furthermore, cells containing DRT/DMC chromosomes also experience endoreduplication at an increased frequency, resulting in an increase in the ploidy of the affected cells [16]. Consequently, during the characterization of the Cre/loxP-mediated rearrangements in chromosome 15, we observed additional copies of the Cre/loxP-rearranged chromosomes, and an increase in the ploidy of the cells with DRT/DMC chromosomes (see below).
To overcome the genome instability complications associated with DRT/DMC we employed a two-step process to characterize the stability and replication timing of the Cre/loxP-mediated rearrangements. First, we assayed low passage cultures for each clone, typically ≤5 passages, which limited the number of generations each clone was expanded through prior to the analysis. Second, we assayed each clone for rearrangements in chromosome 15 using DNA FISH with a whole-chromosome-paint (WCP) as probe, which allowed us to determine the frequency of secondary rearrangements in chromosome 15, and to identify clones with intact chromosome 15 s for detailed replication timing assays (see S1 Table).
An example of secondary rearrangements in chromosome 15 is shown in Fig. 2A. Mitotic spreads from Δ268-4f cells, which contain an ∼135 kb Cre/loxP-mediated distal deletion, were processed for DNA FISH using a chromosome 15 WCP. The chromosome 15 WCP detected 8 inter-chromosomal translocations involving chromosome 15 in this cell. In addition, the chromosome 15 WCP detected 4 non-rearranged chromosome 15s, indicating that this cell contained multiple copies of chromosome 15. Scoring additional metaphase cells with this WCP indicated that 25/50 (50%) cells in this clone contained structural rearrangements of chromosome 15, and the majority of the cells were also polyploid. Similar structural rearrangements affecting chromosome 15 were detected in clones containing Cre/loxP-mediated distal deletions≥124 kb (see below and S1 Table). In contrast, a similar DNA FISH analysis of two clones containing an ∼2 kb distal deletion did not reveal any new chromosome rearrangements involving chromosome 15, with 200 metaphase cells examined (S1 Table). Finally, an analysis of four independent sub-clones (generated by transfecting an Aprt expression vector, selection for Aprt+ cells, and clonal expansion) from parental P268 cells did not reveal any new rearrangements affecting chromosome 15 in any sub-clone, with 400 metaphase cells analyzed ([17]; and see S1 Table). This analysis indicates that transfection, selection for Aprt expression, and clonal expansion did not lead to structural instability of chromosome 15 in P268 cells.
We next assayed replication timing of the chromosome 15 s in clones containing the various Cre/loxP-mediated rearrangements using a BrdU incorporation assay in combination with DNA FISH (see [23]). One critical component of this replication assay is the timing of the BrdU pulse. We used two different “terminal label” pulses of BrdU prior to mitotic harvest (see Fig. 2B). The shorter pulse, 4.5 hours, detected extremely late replication, essentially DNA synthesis during the G2 phase. This relatively short pulse of BrdU detects DNA synthesis on DRT/DMC chromosomes, but not on non-DRT/DMC chromosomes, as non-DRT/DMC chromosomes have finished replication prior to the pulse. The longer pulse, 6 hours, detects DNA synthesis during G2, but also detects the latest replicating DNA of non-DRT/DMC chromosomes at the end of S phase (see Fig. 2B). This longer pulse of BrdU allows us to compare differences in BrdU banding patterns between homologous chromosomes within the same cell, which allows for an estimate of the length of the delay, and allows for a quantitative assessment of replication timing differences between homologous chromosomes (see [23]).
We carried out two independent replication-timing assays on the clones characterized in this report. The first assay involved BrdU incorporation combined with DNA FISH with a chromosome 15 WCP as probe. This BrdU-WCP assay allowed us to assess replication timing of every chromosome 15 within a given cell, including any secondary rearrangements involving chromosome 15. The second assay combined BrdU incorporation with DNA FISH using a chromosome 15 centromeric probe plus BACs located within the Cre/loxP-deletions (see Fig. 1E). This BrdU-BAC assay allowed us to distinguish between the deleted and non-deleted chromosomes. However, the BrdU-BAC assay was not informative in cells with highly rearranged chromosome 15 s, as the secondary rearrangements did not always hybridize to the chromosome 15 probes.
An example of the BrdU-WCP assay in cells with an ∼135 kb distal deletion is shown in Fig. 2C. Cells from clone Δ268-4f were exposed to BrdU for 4.5 hours, harvested for mitotic cells, and processed for BrdU incorporation and DNA FISH with a chromosome 15 WCP. This mitotic cell was chosen to illustrate several points: 1) a single chromosome displays extremely late replication, the only detectable BrdU incorporation in this cell was in a derivative chromosome 15; 2) this cell contains 2 different chromosome 15 rearrangements, indicating that secondary rearrangements have already occurred on chromosome 15; and 3) the delayed replication on the chromosome 15 derivative also occurred on the translocated DNA from an unknown chromosome. These observations illustrate how an initial ∼135 kb deletion event can induce DRT/DMC, which subsequently causes structural instability of the affected chromosome, and that delayed replication can subsequently occur on the translocated DNA from a second chromosome.
An example of the BrdU-BAC assay on a second independent clone containing the ∼135 kb distal deletion is shown in Fig. 2D–G. For this analysis, Δ268-4c cells were exposed to BrdU for 4.5 hours, harvested for mitotic cells, processed for BrdU incorporation and subjected to DNA FISH using a chromosome 15 centromeric probe plus a BAC from the deleted region. The FISH signal from the centromeric probe allowed us to identify the chromosome 15 s, and the presence or absence of the BAC allowed us to distinguish between the non-deleted and deleted chromosomes, respectively. The only two chromosomes showing any detectable BrdU incorporation in this cell hybridized to the chromosome 15 centromeric probe but not to the BAC from the deleted region, indicating that the two chromosomes with DRT/DMC represent deleted 15 s. In addition, DNA synthesis was detected on both arms of the deleted chromosomes (Fig. 2D–G), indicating that the affects of the ∼135 kb deletion extend across the centromere. Also note that two other non-deleted chromosome 15 s, which hybridized to both the centromeric and BAC probes, did not retain any detectable BrdU incorporation, indicating that the non-deleted 15 s had finished replication prior to the BrdU exposure.
Another example of structural instability and DRT/DMC on chromosome 15 is shown in Fig. 3. Mitotic spreads from Δ268-5d cells, which contain one of our larger distal deletions (∼5.6 mb), were processed for DNA FISH using a chromosome 15 WCP. This probe detected 12 rearrangements involving chromosome 15 in this cell (Fig. 3A). Scoring multiple cells with the WCP indicated that>95% of the cells in this clone contained numerous secondary rearrangements of chromosome 15, and the majority of the cells were polyploid. Due to the high frequency of secondary rearrangements the replication timing assays were not performed in this clone. However, we did detect DRT/DMC in a second independent clone containing the same ∼5.6 mb deletion. Δ268-5c cells were exposed to BrdU for 4.5 hours, harvested for mitotic cells, processed for BrdU incorporation, and analyzed by DNA FISH using a chromosome 15 centromeric probe plus a BAC from the deleted region (Fig. 3B and C). Note the extreme level of ‘pulverization’ and ‘G2 DNA synthesis’ on two chromosomes that hybridized to the chromosome 15 centromeric probe. Also note the complete lack of BrdU incorporation in every other chromosome within this cell.
The delayed replication timing and pulverized appearance of the Cre/loxP-mediated deleted chromosomes shown above represent extreme examples of the DRT/DMC phenotype. However, this extreme level of delayed replication and condensation was detected in ≤5% of the mitotic cells in any given clone. Therefore, to determine if the cells that don′t display an extreme phenotype, yet retain intact chromosome 15 s, also display delayed replication we used the longer pulse of BrdU during the replication-timing assay (see Fig. 2B). The 6-hour pulse of BrdU allowed us to detect late DNA synthesis on DRT/DMC and non-DRT/DMC chromosomes and to directly compare the replication timing of the deleted and non-deleted chromosomes within the same cells. An example of this analysis is shown for clone Δ268-4 g, which contains an ∼161 kb distal deletion. Fig. 4A-E shows this analysis for a cell with three copies of chromosome 15, one with the Cre/loxP-mediated deletion and 2 without. Note that the deleted 15 contains a more extensive BrdU banding pattern than either non-deleted 15 (Fig. 4C). Comparing the BrdU banding pattern of the deleted 15 s to the pattern of non-deleted 15 s in multiple cells indicated that the replication timing of the deleted 15 s is delayed by≥2 hours. In addition, quantification of the BrdU incorporation indicated that the deleted 15 s were delayed in their replication timing in all seven cells analyzed (Fig. 4F). Furthermore, DNA FISH analysis with a chromosome 15 WCP as probe on a second clone containing the same ∼161 kb deletion, detected numerous secondary rearrangements of chromosome 15 (Fig. 4G), indicating that the ∼161 kb deletion also results in secondary rearrangements of chromosome 15. In contrast, cells containing the ∼124 kb distal deletion displayed a high degree of variability in BrdU incorporation in chromosome 15 s containing the ∼124 kb deletion. S3 Fig. shows this analysis for cells from clone Δ26818a, and indicates that certain cells in the population displayed a high degree of replication asynchrony between the deleted and non-deleted 15 s (S3A, B, and E Fig.), while other cells showed very little if any asynchrony (S3C, D, and E Fig.). Therefore, the ∼124 kb deletion results in variable expression of the DRT phenotype.
In contrast, similar DNA FISH and replication timing assays on parental P268 cells and in cells containing an ∼2 kb distal deletion indicated that the chromosome 15 s were structurally stable and replicated synchronously (Figs. S4–S5; and see S1 Table). Furthermore, the Cre/loxP-mediated deletions proximal to the original loxP-RT integration site, including two different deletions of ∼66–127 kb and an ∼18 mb deletion (see Table 1), did not cause delayed replication (S6 Fig.).
While there is some variability in the replication timing between deleted and non-deleted chromosome 15 s in cells with deletions≥124 kb distal to the original loxP integration site, we did detect delayed replication timing and chromosome structure instability in clones from all of the distal deletions≥124 kb (Table 1 and S1 Table). In contrast, we did not detect delayed replication in P268 clones containg proximal deletions, an Aprt expression vector, nor in cells with the ∼2 kb distal deletion (see S1 Table). Therefore, the distal deletions define a relatively small critical region, ∼122 kb (see Fig. 5A), that when deleted results in delayed replication timing and instability of chromosome 15.
Next, to determine if chromosomal inversions at this same locus can also result in DRT/DMC we isolated P268 cells containing three different inversions generated at the original loxP-RT integration site in chromosome 15. For this analysis we used the random Lentiviral integration approach to isolate Aprt+ cells containing inversions (Inv268 clones) in both directions in chromosome 15 (see in Table 1). S7 Fig. shows BrdU-WCP assays on two different inversions in chromosome 15, and indicates that chromosomes containing inversions in either direction from the original loxP-RT cassette result in delayed replication and structural instability of chromosome 15. This result indicates that intra-chromosomal rearrangements without loss of DNA, in addition to deletions or translocations, can cause DRT/DMC and instability of chromosome 15.
The observations described above indicate that Cre/loxP-mediated rearrangements at a discrete cis-acting locus cause DRT/DMC and structural instability of human chromosome 15. Because the phenotype of the Cre/loxP-rearranged chromosome 15 s is similar to the phenotype induced by disruption of either Xist or ASAR6 [19], [24] and Xist and ASAR6 encode large monoallelically expressed non-coding RNAs, we next assayed expression of the chromosome 15 locus using RNA-DNA FISH. Because the smallest deletion that results in DRT/DMC on chromosome 15 is located entirely within the large protein-coding gene SCAPER (see Fig. 5A), we used multiple probes from within the SCAPER coding region (spanning ∼560 kb of genomic DNA) to detect RNA, plus either a chromosome 15 centromeric probe, BACs from nearby, or a chromosome 15 paint to detect the DNA. Fig. 5B–D shows an example of this analysis for parental HTD114 cells using a probe from within the deleted region (E4) to detect RNA, plus a chromosome 15 centromeric probe to detect the DNA. Note the large “cloud” of RNA detected on one copy of chromosome 15. An example of RNA-DNA FISH in P268 cells using the H1 probe to detect RNA is shown in Fig. 5E–G. Note that these two cells are tetraploid for chromosome 15 and that the H1 probe detects RNA expressed from only one or two of the chromosome 15 s. We also note that the size of the RNA hybridization signals detected by these probes was variable, ranging form large “clouds” to relatively small pinpoint sites of hybridization (see Fig. 5B–G). In addition, we detected similar RNA hybridization signals using the D1, H3, F5, C5, and C4 probes, and hybridization was detected on ∼50% of the chromosome 15 s in P268 cells (see Fig. 5A). Therefore, the D1, H3, H1, F5, E4, C5 and C4 probes, spanning ∼370 kb of genomic DNA, detect transcripts that are retained within the nucleus and are localized to one of the chromosome 15 homologs. In addition, to determine if the large clouds of RNA detected from within SCAPER are physically localilzed to chromosome 15 we carried out RNA-DNA FISH using a pool of fosmids (D1, H3, H1, F5, E4, C5 and C4) to detect RNA plus a chromosome 15 paint to detect DNA. Fig. 5H–M shows a panel of cells with large RNA signals that is localized to single chromosome 15 territories within the nuclei. We detected chromosome-sized RNA FISH signals in 82% (89/108) of P268 cells using this pool of fosmid probes.
In contrast to the D1, H3, H1, F5, E4, C5 and C4 probes, we were not able to detect RNA using the F8 or A4 probes in either HTD114 or P268 cells (Fig. 5A and see S8 Fig.). Thus, even though the F8 and A4 probes are within SCAPER, and are capable of detecting DNA as efficiently as the E4 probe using DNA FISH on metaphase chromosomes (see S8 Fig.), they were not able to detect expression of RNA in HTD114 or P268 cells. However, we did detect low-level expression of the 5′ region of properly spliced SCAPER mRNA using RT-PCR (see S11 Fig.). Therefore, the 5′ region of SCAPER distal to the C4 probe (∼150 kb) is expressed, but not at high enough levels to be detected by RNA FISH using the F8 or A4 probes.
Next, to determine if similar clouds of transcripts are present in human primary cells we used RNA-DNA FISH on human foreskin fibroblasts (HFFs). Fig. 5N–P shows an example of this analysis using the E4 probe to detect RNA, plus a BAC to detect the DNA from this locus. The E4 probe detected a single large RNA signal on one chromosome 15 in ∼76% (41/54) of cells, or on both copies of chromosome 15 in ∼24% (13/54) of cells (Fig. 5H–J). Similar results were obtained with the D1, H3, H1, F5, C5 and C4 probes (see Fig. 5A. In contrast, but similar to HTD114 and P268 cells, the F8 and A4 probes failed to detect RNA in HFFs (see Fig. 5A). In addition, to directly compare the size and appearance of the RNA signal detected from within SCAPER to XIST RNA on the inactive X chromosome we analyzed expression of the chromosome 15 locus plus XIST simultaneously in female primary dermal fibroblasts HDFs. Figs. 5Q and S9E and S9F show RNA detected by the H3 probe in relation to RNA detected by the XIST probe. We detected chromosome-sized RNA FISH signals in 72% (72/100) of primary HDF cells.
In addition, we also detected expression of RNA from ∼50% of the chromosome 15 s in human HepG2 cells (S9A–D Fig.). Furthermore, analysis of RNA-seq data, generated by the Encode Project, indicated that RNA expressed from the critical region for DRT/DMC, defined above, is enriched in the nuclear Poly A minus fraction in HepG2 cells (S10 Fig.). We previously found that ASAR6 RNA is present in the Poly A minus fraction and does not contain introns yet is transcribed by RNA Polymerase II [18]. Therefore, to determine if the RNA produced from within the critical region of SCAPER is the product of RNA Polymerase II, we analyzed expression of RNA in cells treated with α-amanitin, which is a selective inhibitor of RNA Polymerase II [25], using a semi-quantitative RT-PCR assay. S11 Fig. shows the results of this analysis, and indicates that the RNA derived from the critical region is indeed sensitive to α-amanitin. Similarly, RNA expressed from the protein-coding gene P300 is also sensitive to α-amanitin treatment. We also detected properly spliced SCAPER mRNA and found that it was also sensitive to α-amanitin treatment. In contrast, expression of 45S RNA (an RNA Polymerase I product) and a tRNA gene (an RNA Polymerase III product) were not inhibited by α-amanitin. We conclude that the RNA produced from the critical region for DRT/DMC is generated by RNA Polymerase II. In addition, this analysis indicated that the half-life of the RNA from within SCAPER introns 21, 23, and 24 is similar to the half-life of ASAR6 RNA, which is approximately 5 hours. In addition, the half-life of these intronic RNAs is much longer than the half-life of properly spliced P300 or SCAPER mRNAs (∼2 hours), which are spliced, polyadenylated, RNA Polymerase II products.
Taken together, our observations indicate that RNA transcripts from an ∼370 kb region from within the SCAPER gene are retained within the nucleus, and can be detected as chromosome sized clouds of RNA localized to either one homolog of human chromosome 15 in human cell lines and in primary skin fibroblasts. However, whether these transcripts represent transcription initiating within SCAPER, or represent transcripts that are processed from the larger SCAPER primary transcript and preferentially retained on chromosome 15 is currently not known. Regardless, because the locus that encodes these transcripts also displays asynchronous replication (see below), we have named these transcripts ASynchronous replication and Autosomal RNA on chromosome 15, or ASAR15.
Previous studies indicate that disruption of the expressed alleles of either Xist or ASAR6 result in extremely late replication and structural instability of their respective chromosomes [19], [24]. Therefore, to determine if the deletions that cause DRT/DMC and instability on chromosome 15 occurred on the chromosome retaining ASAR15 RNA we utilized an RNA-DNA FISH assay that allowed us to distinguish between the deleted and non-deleted chromosomes. For this analysis we used a probe from within the deletion (E4) plus a probe from outside the deletion (H1) simultaneously to detect RNA in cells containing the ∼161 kb distal deletion (see Fig. 5A). Thus, if the deletion occurred on the silent copy of chromosome 15, the H1 and E4 probes would detect ASAR15 RNA concomitantly. In contrast, if the deletion occurred on the expressed allele, the E4 probe would not detect RNA from the deleted 15, but the H1 probe would still detect RNA expressed from the deleted chromosome. First, to determine if the H1 and E4 probes detect the same sites of expression in non-deleted cells, we assayed expression of RNA in P268 cells using both probes simultaneously. Fig. 6A shows examples of this analysis and indicates that both probes detected RNA from the same chromosome 15. Note that the P268 cells in Fig. 6A are either diploid (cell #1 and #2) or tetraploid (cell #3) for chromosome 15, and hybridization with both probes was detected on either one or two chromosome 15 s, respectively. We detected simultaneous expression with the H1 and E4 probes in 100% (50/50) of P268 cells. Therefore, the H1 and E4 probes detect expression from the same chromosome 15 s.
Next, to determine if the rearrangements that cause DRT/DMC affect the expressed or non-expressed allele of ASAR15 we processed Δ268-4 g cells for RNA-DNA FISH using the H1 and E4 probes to detect RNA. Fig. 6B shows examples of this analysis and indicates that the H1 probe detected clouds of RNA, which are indistinguishable in number and appearance to the clouds expressed in parental P268 cells (Fig. 6A). Note that the Δ268-4 g cells in Fig. 6B are either triploid (cell #1) or tetraploid (cell #2 and #3) for chromosome 15, and the H1 probe detected RNA from one (cell #1) or two (cell #2 and #3) chromosome 15 s, respectively. In contrast, the E4 probe detected zero (cell #1) or one (cell #2 and #3) site of expression in these same cells, indicating that the RNA detected by the E4 probe was absent from one of the alleles detected by the H1 probe in 100% (50/50) of the cells. Therefore, the Cre/loxP-mediated deletion occurred on the expressed allele of ASAR15. In addition, because the H1 probe detected expression from the deleted chromosome, the ∼161 kb deletion did not disrupt expression nor retention of transcripts proximal to the original loxP-RT integration site (see Fig. 5A). Finally, because the E4 probe detected RNA in some of the cells in Δ268-4 g (cell #2 and #3), some of the Δ268-4 g cells retain an intact copy plus a deleted copy of the expressed allele of ASAR15, which was not surprising as the cells that still express RNA detected by the E4 probe were tetraploid for chromosome 15 (Fig. 6B). Regardless, these observations indicate that ASAR15 RNA is present on one chromosome 15 homolog in P268 cells, and that the Cre/loxP rearrangements that cause DRT/DMC occurred on the chromosome 15 that retained ASAR15 RNA.
Because all monoallelic genes display asynchronous replication between alleles (reviewed in [26]), we next tested whether the locus that encodes ASAR15 RNA also displays asynchronous replication. For this analysis we used a DNA FISH based assay [27] to determine the extent of asynchronous replication on chromosome 15. This assay uses probes to particular chromosomal locations hybridized to cells in S-phase. This assay also includes a methanol/acetic acid fixation step, which destroys the nuclear architecture and allows for a relatively accurate assessment of replication synchrony [28], [29]. The hybridization signals are present in nuclei in three distinct patterns, which are dependent on the replication status of each allele. The first pattern corresponds to two single hybridization dots (the SS pattern), indicating that neither allele has replicated. The second pattern corresponds to two double dots (the DD pattern), indicating that both alleles have replicated. The third pattern corresponds to one single dot and one double dot (the SD pattern), indicating that only one of the alleles had replicated. For asynchronously replicating loci the SD pattern is present in 30–50% of nuclei. In contrast, synchronously replicating loci show the SD pattern in only 10–20% of nuclei [29]–[31]. We used this “single dot-double dot” assay in primary HFFs to determine the degree of replication asynchrony on chromosome 15. We found that two different non-overlapping probes (BACs CTD-2299E17 and CTD-2117F7; see Fig. 1E) from within the ASAR15 region, as well as three additional chromosome 15 random monoallelic genes (MYO1E, PTPN9, and PEAK1; [32], [33]) located ∼1-20 mb from ASAR15, all display a high percentage of the SD pattern, which is indicative of asynchronous replication (Table 2).
Previous studies found that the asynchronous replication of random monoallelic genes on any given chromosome displays coordination either in cis or in trans [18], [19], [30], [31], [34]. Therefore, we next determined if the asynchronous replication of ASAR15 is coordinated with other monoallelic loci on chromosome 15. For this analysis we used a two-color DNA FISH assay to determine the SD pattern for two loci simultaneously. We used a probe representing ASAR15 in combination with BAC probes for MYO1E, PTPN9, and PEAK1 (see Table 2). This analysis indicated that the asynchronous replication of ASAR15 was coordinated in cis with MYO1E (72/100 cells, P<1×10−4), PTPN9 (71/100 cells P<1×10−4), and PEAK1 (74/100 cells, P<1×10−5).
One limitation of the DNA FISH assay described above is that probes on the same chromosome that are>50 mb apart are difficult to score, as the hybridization signals coming from one homolog may be closer to hybridization signals coming form the other homolog. To overcome this limitation, we utilized a second replication-timing assay known as Replication Timing-Specific Hybridization, or ReTiSH [34]. The ReTiSH assay involves BrdU incorporation for different times followed by the analysis of metaphase chromosomes using a modification of the Chromosome Orientation-Fluorescence In Situ Hybridization (CO-FISH) assay [35]. CO-FISH involves the conversion of BrdU incorporated chromosomal DNA into single stranded DNA followed by hybridization with specific probes without further denaturation of the chromosomal DNA. Because metaphase spreads are analyzed for hybridization signals along individual chromosomes, the distance between the loci is not a limitation of the ReTiSH assay [34]. In order to assay asynchronous loci along the length of chromosome 15 we used probes for ASAR15 (15q24), the rDNA cluster (15p11), and MYO1E (15q22). The rDNA clusters are located on five different human chromosomes (13, 14, 15, 21 and 22), and were recently shown to display random asynchronous replication [34]. Hybridization of the rDNA probe to chromosomes 13, 14, 15, 21, and 22 also served as an internal control for asynchronous replication on multiple chromosomes detected simultaneously by the ReTiSH assay. This assay also included a chromosome 15 centromeric probe, which allowed for the unambiguous identification of the chromosome 15 s. Centromeric heterochromatin is known to be late replicating, and the ReTiSH assay detects hybridization of centromeric probes to both homologs at both time points [34]. For this analysis we exposed human primary blood lymphocytes (PBLs) to BrdU for either 14 or 6 hours and then processed the cells for ReTiSH. Fig. 7A–C shows that the rDNA probe hybridized to both homologs of chromosomes 13, 14, 15, 21, and 22, at the 14 hour time point, but hybridization to single homologs of these same chromosomes at the 6 hour time point. Therefore, our ReTiSH assay detected asynchronous replication of the rDNA clusters on all five chromosomes. We detected hybridization of the rDNA probe to both copies of chromosome 15 in 100% (50/50) of cells at the 14 hour time point, and to a single copy of chromosome 15 in 100% (50/50) of cells at the 6 hour time point, which is consistent with previous observations that the rDNA cluster on chromosome 15 displays asynchronous replication [34]. Similarly, we found two sites of hybridization for ASAR15 or MYO1E at the 14 hour time point (50/50 cells for both probes), and single sites of hybridization for ASAR15 (47/50 cells), or MYO1E (47/50 cells) at the 6 hour time point, indicating asynchronous replication at both loci. Next, we used a two-color hybridization scheme to simultaneously detect ASAR15 plus rDNA or MYO1E. We found that the asynchronous replication of ASAR15 was coordinated in cis with the rDNA cluster on chromosome 15 (Fig. 7D; 44/50 cells, P<1×10−5), and with the MYO1E gene (Fig. 7E; 43/50 cells, P<1×10−5). Therefore, the asynchronous replication of ASAR15 is coordinated in cis with other monoallelic loci located>70 mb away and on both sides of the centromere.
Mammalian cells replicate their chromosomes every cell cycle during a defined temporal replication program (reviewed in [36]). Recent studies indicate that at least half of the genome is subject to changes in the replication timing of relatively large chromosomal domains along every chromosome during normal development [37], [38]. However, the determinants that control replication timing are poorly understood, and are not encoded within the DNA sequence of the origins. The current thinking is that the timing of origin firing is dictated by chromosomal location, and is directly linked to complex higher-order features of chromosome architecture [39], [40]. In addition, recent studies have identified DNA binding proteins that can dictate origin timing and implicate the spatial organization of origins within nuclear territories in the mechanism of replication timing control (reviewed in [41], [42]).
In this report we found that chromosomal rearrangements at a discrete cis-acting locus on human chromosome 15 cause delayed replication of the entire chromosome. This locus represents the third example of a cis-acting locus that, when disrupted, results in delayed replication of an entire mammalian chromosome, with mouse Xist and human ASAR6 representing the other two loci [19], [24]. The chromosome 15 locus described here is complex, and encodes the large protein-coding gene SCAPER, the micro RNA gene MIR3713, and large, preferentially retained, nuclear transcripts (ASAR15) that can be detected as chromosome-sized clouds on chromosome 15. Whether or not the ASAR15 transcripts are initiated within SCAPER, or represent transcripts that are processed from the larger SCAPER primary transcript and preferentially retained on chromosome 15 is currently not known. Regardless, the identification of loci, (Xist, ASAR6, and ASAR15) that control replication timing of three different chromosomes suggests that the replication-timing program of all mammalian chromosomes involves differentially expressed, asynchronously replicating, cis-acting, non-coding RNA genes.
Monoallelic gene expression in mammals occurs in two distinct patterns, random and imprinted (reviewed in [26]). The inequality of the two alleles in both patterns is characterized by differential DNA methylation, differential chromatin modifications, unequal nuclear localization, and expression of large non-coding RNAs [26], [43]–[45]. While parallels between X inactivation and the cis-coordinated replication asynchrony of autosomal random monoallelic genes have been made [26], [30], [31], [34], [46], [47], recent reports indicate that not all autosomal random monoallelic genes are expressed from the same homolog in some clonal cell lines, or that these same genes may be biallelically expressed in other clones [33], [48], [49]. Our RNA FISH analysis of ASAR15 RNA expression in HTD114, P268, and HepG2 is consistent with monoallelic expression of this locus in these clonal cell lines. However, a similar RNA FISh analysis of non-clonal primary fibroblasts revealed that at least some cells in the population contain biallelic expression of ASAR15. Therefore, ASAR15 appears to display similar characteristics as other random autosomal “monoallelic” genes, in that at least some cells display biallelic expression.
Another characteristic of all monoallelic genes, both random and imprinted, is that they display asynchronous replication between alleles located on homologous chromosomes [26], [29]–[31], [50]. Asynchronous replication with random choice represents an epigenetic state that is established early in development, is present in all tissues, and is independent of gene expression [29]–[31]. In this report, we used two different replication-timing assays, ‘single dot-double dot′ and ReTiSH, to assess replication synchrony along human chromosome 15 in two different primary cell types, HFFs and PBLs. We found that the asynchronous replication of ASAR15 is coordinated in cis with other random monoallelic loci separated by>70 mb of genomic DNA, and located on either side of the centromere. Furthermore, because the asynchronous replication of the rDNA locus on chromosome 15 is random and not imprinted [34] and the asynchronous replication of ASAR15 is coordinated in cis with the rDNA locus, the asynchronous replication of ASAR15 must also be random.
Another common feature of monoallelic genes is a relatively high density of LINE-1 elements [51]–[53]. Consistent with this observation, the ASAR6 and ASAR15 loci contain a high concentration of LINE-1 elements within the expressed regions of each locus, constituting>40% and>55% of the sequence, respectively. Note that in humans, autosomes contain on average ∼17.6% LINE-1 sequence and the X chromosome contains ∼31.0% LINE-1 sequence [53]. In addition, recent observations indicate that LINE1 RNA shares at least some characteristics with XIST RNA [54]. While LINE-1 elements have been implicated in monoallelic gene expression on the X chromosome and on autosomes for many years [51]–[53], [55]–[57], a mechanistic role for these abundant elements during transcriptional silencing and/or replication timing has not been established.
Previous studies indicate that ASAR6 and Xist share many important physical and functional characteristics [18], [19]. In this report we found that ASAR15 also shares many of these same characteristics, including: 1) differential expression of large RNA transcripts that are retained in the nucleus, 2) random asynchronous replication that is coordinated with other linked monoallelic genes, and 3) disruption of the expressed allele results in delayed replication timing and structural instability of an individual mammalian chromosome. In addition, ASAR15 RNA was detected as chromosome-sized clouds on one homolog of chromosome 15, and that these clouds are similar in appearance to XIST RNA localized to the inactive X chromosome.
The molecular mechanisms by which Xist, ASAR6 or ASAR15 control replication timing of entire chromosomes is currently not known. Interestingly, disruption of the transcriptionally silent mouse Xist gene, in male or female somatic cells, results in a small but significant delay in replication of the active X chromosome, indicating that Xist RNA expression is not involved in the delayed replication of the active X chromosome following Xist gene disruption [58]. In contrast, disruption of the expressed Xist allele, on the inactive X chromosome, results in extremely late replication, abnormal chromatin structure, and instability of the X chromosome [24]. Therefore, whether or not the Xist gene is transcribed has a dramatic affect on the severity of the replication timing delay and stability of the X chromosome, implicating the act of transcription or the Xist RNA in maintaining proper replication timing of the inactive X chromosome [24]. We have found that disruption of the expressed alleles of either ASAR6 or ASAR15 results in DRT/DMC and structural instability of their respective chromosomes ([18], [19]; and see above]. Whether or not the transcriptionally silent alleles of ASAR6 or ASAR15 control replication timing of their respective chromosomes is currently not known. Furthermore, ASAR15 RNA could be detected as a chromosome-sized cloud of RNA in human cell lines and in primary skin fibroblasts. In contrast, ASAR6 RNA does not form large clouds on chromosome 6 in clonal cell lines nor in primary blood lymphocytes [18], [19]. Therefore, the significance of the presence or absence of large clouds of RNA expressed from ASAR15 or ASAR6 remains an open question.
One complication in the analysis of chromosomes with DRT/DMC is the dramatic genomic instability that occurs in cells with these abnormal chromosomes. There are at least two distinct types of genomic instabilities that have been described in cells with DRT/DMC chromosomes. First, and the best characterized, is chromosome structure instability (CSI). The CSI associated with DRT/DMC is characterized by a 30–80 fold increase in the rate of chromosome rearrangements affecting the DRT/DMC chromosomes [17]. Because the secondary rearrangements that occur on the DRT/DMC chromosomes are often inter-chromosomal translocations, all of the chromosomes within a cell with DRT/DMC chromosomes can be affected [17]. The second type of genomic instability associated with DRT/DMC is chromosome instability (CIN). There are two distinct manifestations of CIN in cells with DRT/DMC chromosomes. First, DRT/DMC chromosomes activate the spindle assembly checkpoint, are often observed as lagging chromosomes during anaphase, and experience non-disjunction events at an increased frequency resulting in gains and losses of the DRT/DMC chromosomes [16], [18]. Second, cells with DRT/DMC chromosomes also experience failed cytokinesis, which results in endoreduplication and dramatic whole chromosome copy number alterations [16]. Therefore, the CIN that occurs in cells with DRT/DMC chromosomes can also affect the copy number of non-DRT/DMC chromosomes within the same cell. Therefore, the CSI and CIN observed in cells with DRT/DMC chromosomes affects the stability of the entire genome [1].
We previously used IR to generate chromosome rearrangements in mouse and human cells, and found that ∼5% of inter-chromosomal translocations involving many different autosomes displayed DRT/DMC [15]. In addition, we detected DRT/DMC in primary cells derived from skin fibroblasts, blood lymphocytes, and kidney epithelial cells, indicating that the DRT/DMC phenotype can affect chromosomes present in non-transformed cells from three different tissues. Furthermore, our original chromosome-engineering screen for DRT/DMC chromosomes identified five balanced translocations, affecting eight different autosomes, all displaying DRT/DMC on at least one of the two derivative chromosomes [17]. We previously characterized one of these balanced translocations in molecular detail and found that disruption of the ASAR6 gene results in DRT/DMC on chromosome 6 [19]. In this report, we characterized a second balanced translocation from this screen and identified a cis-acting locus that when disrupted results in DRT/DMC on a second human autosome. Our work, combined with the observation that disruption of Xist results in a similar replication and instability phenotype [24], suggests that all mammalian chromosomes contain cis-acting loci that function to synchronize chromosome-wide replication timing, promote monoallelic gene expression and help maintain structural stability of individual chromosomes [1], [18], [19]. We believe that these cis-acting loci are as fundamentally important to mammalian chromosome function as telomeres, centromeres, or origins of replication. Therefore, we propose that every mammalian chromosome contains four essential cis-acting elements: origins, telomeres, centromeres, and “inactivation/stability centers”, all functioning to ensure proper replication, segregation and stability of each chromosome.
The methods for culturing the cells in this study were carried out as previously described [18]. Low passage primary human (male) foreskin fibroblasts (HFFs) were obtained from ATCC and cultured in DMEM plus 10% fetal bovine serum (Hyclone). Primary, human (female) dermal skin fibroblasts (HDFs) were provided by Dr. Shoukhrat Mitalipov. Primary blood lymphocytes were isolated after venipuncture into a Vacutainer CPT (Becton Dickinson, Franklin Lakes, NJ) per the manufacturer's recommendations and grown in 5 mL RPMI 1640 (Life Technologies) supplemented with 10% fetal bovine serum (Hyclone) and 1% phytohemagglutinin (Life Technologies). HTD114 and P268 cells are APRT deficient cell lines derived from the HT1080 fibrosarcoma [59], and were grown in DMEM (Gibco) supplemented with 10% fetal bovine serum (Hyclone). P268 derivatives were grown as above with the addition of 500 mg/ml Geneticin (Gibco), 200 mg/ml Hygromycin B (Calbiochem), and/or 10 ug/ml Blasticidin S HCl (Invitrogen). The deletion-line derivatives were grown in DMEM supplemented with 10% dialyzed fetal bovine serum (Hyclone), 10 mg/ml azaserine (Sigma) and 10 mg/ml adenine (Sigma) to facilitate selection for Aprt-expressing cells. All cells were grown in a humidified incubator at 37°C in a 5% carbon dioxide atmosphere.
The methods for DNA FISH assays in this study were carried out as previously described [18]. Trypsinized cells were centrifuged at 1,000 rpm for 10 minutes in a swinging bucket rotor. The cell pellet was re-suspended in 75 mM potassium chloride for 15–30 minutes at 37°C, re-centrifuged at 1,000 rpm for 10 minutes and fixed in 3∶1 methanol∶acetic acid. Fixed cells were added drop-wise to microscope slides to generate mitotic chromosome spreads using standard methods [60]. Slides with mitotic spreads were baked at 85°C for 20 minutes and then treated with 0.1 mg/ml RNAase for 1 hour at 37°C. After RNAase treatment, the slides were washed in 2xSSC (1xSSC is 150 mM NaCl and 15 mM sodium citrate) with 3 changes for 3 minutes each and dehydrated in 70%, 90%, and 100% ethanol for 3 minutes each. The slides were denatured in 70% formamide in 2xSSC at 70°C for 3 min and whole chromosome paints were used according to the manufacturer's recommendations and hybridization solutions (American Laboratory Technologies and Vysis). Detection of digoxigenin-dUTP probes utilized a three-step incubation of slides with sheep FITC-conjugated anti-digoxigenin antibodies (Roche) followed by rabbit FITC-conjugated anti-sheep antibodies (Roche) followed by goat FITC-conjugated anti-rabbit antibodies (Jackson Laboratories). Slides were stained with DAPI (12.5 mg/ml) or propidium iodide (0.3 mg/ml), cover slipped, and viewed under UV fluorescence with appropriate filters (Olympus).
Centromeric, BAC, and fosmid probes: Mitotic chromosome spreads were prepared as described above. Slides were treated with RNase at 100 ug/ml for 1 h at 37°C and washed in 2xSSC and dehydrated in 70%, 90% and 100% ethanol. Chromosomal DNA was denatured at 75°C for 3 minutes in 70% formamaide/2XSSC, followed by dehydration in ice cold 70%, 90% and 100% ethanol. BAC and Fosmid DNAs were nick-translated using standard protocols to incorporate biotin-11-dUTP or digoxigenin-dUTP (Invitrogen). BAC and Fosmid DNAs were directly labeled with Cy3-dUTP, FITC-dUTP, Spectrum Orange-dUTP or Spectrum Green_dUTP (Vysis, Abbott Laboratories) using nick-translation or random priming using standard protocols. Final probe concentrations varied from 40–60 ng/µl. Centromeric probe cocktails (Vysis) plus BAC or Fosmid DNAs were denatured at 75°C for 10 minutes and prehybridized at 37°C for 30 minutes. Probes were applied to slides and incubated overnight at 37°C. Post-hybridization washes consisted of three 3-minute rinses in 50% formamide/2XSSC, three 3-minute rinses in 2XSSC, and finally three 3-minute rinses in PN buffer (0.1M Na2HPO4 + 0.0M NaH2PO4, ph 8.0, +2.5% Nonidet NP-40), all at 45°C. Signal detection was carried out as described [61]. Amplification of biotinylated probe signal utilized alternating incubations of slides with anti-avidin (Vector) and FITC-Extravidin (Sigma). Slides were then counterstained with either propidium iodide (2.5 ug/ml) or DAPI (15 ug/ml) and viewed under UV fluorescence (Olympus).
The methods for the RNA-DNA FISH assays in this study were carried out as previously described [18]. Cells were plated on microscope slides treated with concanavalin A (Sigma) at ∼50% confluence and incubated for 4 hours in complete media in a 37°C humidified CO2 incubator. Slides were rinsed 1 time with sterile RNase free PBS. Slides were incubated for 30 seconds in CSK buffer (100 mM NaCl, 300 mM Sucrose, 3 mM MgCl2,10 mM Pipes, ph 6.8), 5 minutes in CSK buffer plus 0.1% Triton X-100, and then for an addition 30 seconds in CSK buffer at room temperature. Cells were fixed in 4% paraformaldehyde in PBS for 10 minutes at room temperature. Slides were rinsed in 70% ETOH and stored in 70% ETOH at 4°C until use. Just prior to use, slides were dehydrated through an ETOH series (70%, 90% and 100%) and allowed to air dry. Denatured probes were prehybridized with Cot-1 DNA at 37°C for 30 min. Slides were hybridized at 37°C for 14–16 hours. Slides were washed as follows: 3 times in 50% formamide/2xSSC at 42°C for 5 minutes, 3 times in 2xSSC at 42°C for 5 minutes, 3 times in 4xSSC/0.1% Tween 20 at room temperature for 3 minutes. Slides were then counterstained with either propidium iodide (2.5 ug/ml) or DAPI (15 ug/ml) and viewed under UV fluorescence (Olympus). Z-stack images were generated using a Cytovision workstation. The hybridization signals on individual cells were captured and the slide coordinates recorded. The slides were then dehydrated in 70%, 90% and 100% ETOH, and then processed for DNA FISH, including the RNAase treatment step, as described above. The same cells that were imaged with the RNA FISH probes were located on the slides and then re-imaged for the DNA hybridization signals.
The methods for the replication timing assays in this study were carried out as previously described [19], [23]. The BrdU replication-timing assay was performed on exponentially dividing cultures as follows: asynchronously growing cells were exposed to 20 ug/ml of BrdU (Sigma) for 4.5 or 6 hours. Mitotic cells were harvested in the absence of colcemid, treated with 75 mM KCl for 15–30 minutes at 37°C, fixed in 3∶1 methanol∶acetic acid and dropped on wet ice cold slides. The chromosomes were denatured in 70% formamide in 2xSSC at 70°C for 3 minutes and processed for DNA FISH, as described above. The incorporated BrdU was then detected using a FITC-labeled anti-BrdU antibody (Becton Dickinson). Slides were stained with propidium iodide (0.3 mg/ml), cover slipped, and viewed under UV fluorescence.
All images were captured with an Olympus BX Fluorescent Microscope using a 100X objective, automatic filter-wheel and Cytovision workstation. Individual chromosomes were identified with either chromosome-specific paints or centromeric probes in combination with BACs from the deleted regions (see DNA FISH procedure above). Utilizing the Cytovision workstation, each chromosome was isolated from the metaphase spread and a line drawn along the middle of the entire length of the chromosome. The Cytovision software was used to calculate the pixel area and intensity along each chromosome for each fluorochrome occupied by the DAPI and BrdU (FITC) signals. The total amount of fluorescent signal was calculated by multiplying the average pixel intensity by the area occupied by those pixels.
We used the ReTiSH assay essentially as described [34]. Briefly, unsynchronized, exponentially growing cells were treated with 30 µM BrdU (Sigma) for 6 or 5 and 14 hours. Colcemid (Sigma) was added to a final concentration of 0.1 µg/mL for 1 h at 37°C. Cells were trypsinized, centrifuged at 1,000 rpm, and resuspended in prewarmed hypotonic KCl solution (0.075 M) for 40 min at 37°C. Cells were pelleted by centrifugation and fixed with methanol-glacial acetic acid (3∶1). Fixed cells were drop gently onto wet, cold slides and allowed to air-dry. Slides were treated with 100 µg/ml RNAse A at 37°C for 10 min. Slides were rinsed briefly in d2H20 followed by fixation in 4% formaldehyde at room temperature for 10 minutes. Slides were incubated with pepsin (1 mg/mL in 2N HCl) for 10 min at 37°C, and then rinsed again with d2H20 and stained with 0.5 µg/µL Hoechst 33258 (Sigma) for 15 minutes. Slides were flooded with 200 µl 2xSSC, coversliped and exposed to 365-nm UV light for 30 min using a UV Stratalinker 2400 transilluminator (Stratagene). Slides were rinsed with d2H20 and drained. Slides were incubated with 100 µl of 3U/µl of ExoIII (Fermentas) in ExoIII buffer for 15 min at 37°C. The slides were then processed directly for DNA FISH as described above, except with the absence of a denaturation step.
|
10.1371/journal.pntd.0006780 | Towards elimination of lymphatic filariasis in southeastern Madagascar: Successes and challenges for interrupting transmission | A global strategy of mass drug administration (MDA) has greatly reduced the burden of lymphatic filariasis (LF) in endemic countries. In Madagascar, the National Programme to eliminate LF has scaled-up annual MDA of albendazole and diethylcarbamazine across the country in the last decade, but its impact on LF transmission has never been reported. The objective of this study was to evaluate progress towards LF elimination in southeastern Madagascar.
Three different surveys were carried out in parallel in four health districts of the Vatovavy Fitovinany region in 2016: i) a school-based transmission assessment survey (TAS) in the districts of Manakara Atsimo, Mananjary, and Vohipeno (following a successful pre-TAS in 2013); ii) a district-representative community prevalence survey in Ifanadiana district; and iii) a community prevalence survey in sentinel and spot-check sites of these four districts. LF infection was assessed using the Alere Filariasis Test Strips, which detect circulating filarial antigens (CFA) of adult worms. A brief knowledge, attitudes and practices questionnaire was included in the community surveys.
None of the 1,825 children sampled in the TAS, and only one in 1,306 children from sentinel and spot-check sites, tested positive to CFA. However, CFA prevalence rate in individuals older than 15 years was still high in two of these three districts, at 3.5 and 9.7% in Mananjary and Vohipeno, respectively. Overall CFA prevalence in sentinel and spot-check sites of these three districts was 2.80% (N = 2,707), but only two individuals had detectable levels of microfilaraemia (0.06%). Prevalence rate estimates for Ifanadiana were substantially higher in the district-representative survey (15.8%; N = 545) than in sentinel and spot-check sites (0.8%; N = 618). Only 51.2% of individuals surveyed in these four districts reported taking MDA in the last year, and 42.2% reported knowing about LF.
Although TAS results suggest that MDA can be stopped in three districts of southeastern Madagascar, the adult population still presents high CFA prevalence levels. This discordance raises important questions about the TAS procedures and the interpretation of their results.
| Lymphatic filariasis is a neglected disease with chronic disabling consequences. Endemic countries have reduced lymphatic filariasis transmission through a strategy of annual rounds of mass drug administration (MDA), but the impact of such strategy has not yet been reported for Madagascar. In this study we conducted three different surveys and used rapid diagnostic tests to evaluate lymphatic filariasis transmission in four health districts of southeastern Madagascar. This included a school-based transmission assessment survey (TAS), the international gold standard to help national programmes confirm that they have interrupted lymphatic filariasis transmission, and two complementary community-based surveys. Our TAS results suggested that MDA could be stopped in three districts, confirming the consistent decline in lymphatic filariasis observed in recent years. However, the other two surveys revealed that the adult population still had high prevalence levels. This discordance raises questions about the TAS procedures and the interpretation of their results in contexts where, like in Madagascar, implementation of MDA is different for school age children than for the rest of the population.
| The world is on a path towards the elimination of lymphatic filariasis (LF), a major neglected tropical disease (NTD) and parasitic infection transmitted by mosquitoes and caused by Wuchereria bancrofti, Brugia malayi, and B. timori. LF can result in chronic disabling consequences (e.g. hydrocele, lymphoedema and elephantiasis), and is responsible for about 5.8 million Disability Adjusted Life Years (DALYs) [1], including a substantial burden on mental health [2]. The availability of cheap, safe and effective treatments against the filarial infection, with long-term effects on disease transmission, has driven this international effort and allowed for a strategy of mass drug administration (MDA) to populations where LF is endemic (a combination of albendazole plus ivermectin or diethylcarbamazine) [3]. Following the launch of the Global Programme to Eliminate Lymphatic Filariasis in 2000 to achieve elimination by 2020 [4], National Programmes, the cornerstones of this effort, have been set up in all endemic countries. By 2015 they had distributed more than 6 billion treatments, covering nearly 60% of the world population requiring MDA [5]. As a result, transmission of LF had been interrupted by 2014 in 18 out of 73 endemic countries, and many others had achieved partial elimination [5].
Lymphatic filariasis is endemic in Madagascar, where it has been known for more than a century [6]. Surveys conducted in the 1950s and 70s revealed stable prevalence levels of microfilaraemia in the eastern coast of the island at about 25% [7,8], higher than other regions of Madagascar [8]. A country-wide mapping study in 2004–2005, at baseline of the National Programme, revealed that LF was endemic in 98 health districts out of 114 and more than 18 million people required MDA [9]. Since then, annual rounds of albendazole and diethylcarbamazine have been administered in endemic districts to people of all ages, except for children under 2 years old and pregnant women. MDA was initiated in districts of the eastern coast and has been progressively scaled-up over the past 10 years, currently reaching 62 districts at different stages of MDA (3 to 10 rounds). As a result, the National Programme needed to evaluate whether transmission has been interrupted in these districts in order to decide on whether to continue with MDA.
WHO recommendations for the monitoring and evaluation of National Programmes involve surveillance of prevalence levels in sentinel and spot-check sites (individuals 5 years or older, including adults), followed by a confirmation that LF transmission has been interrupted through a transmission assessment survey (TAS) after certain criteria are met: a TAS is carried out in populations that have received MDA for at least 5 years (minimum coverage of 65%) and where prevalence of microfilaraemia, in sentinel and spot-check sites, has been reduced to less than 1% (or antigenaemia <2%) after the last effective round [10]. The survey is conducted in primary schools, targeting children between 6 and 7 years of age. In endemic areas for W. bancrofti, the surveys use point of care tests that detect circulating filarial antigens (CFA) of adult worms. In areas where Anopheles are the principal vectors, a proportion of positive CFA < 2% implies a recommendation for stopping MDA and beginning post-MDA surveillance [3]. Since 2013, these surveys use the Filariasis Test Strip (FTS) as the preferred CFA test, which has better sensitivity and stability than the previously used BinaxNOW Filariasis immunochromatographic card test (ICT; both produced by Alere, Scarborough, ME) [11,12].
The objective of this study was to evaluate LF transmission in four districts of the Vatovavy Fitovinany region (Eastern Madagascar) that had received 9–10 annual rounds of MDA, by comparing results obtained from three cross-sectional surveys, including a school-based TAS and two community-based prevalence surveys (Fig 1). All surveys took place in the fall of 2016 and used FTS to evaluate infection levels.
The region of Vatovavy Fitovinany, in southeastern Madagascar, has historically presented high levels of LF transmission [7,8]. In this region, the districts of Manakara Atsimo, Mananjary and Vohipeno had some of the highest CFA prevalence levels in the country (>40%, ICT) during the 2004 mapping survey conducted at baseline of the National Programme (Table 1). As a result, they were among the first districts to be targeted for MDA in 2006, and by 2016 had received 10 annual rounds of treatment. Adjacent to this area, the district of Ifanadiana, with a baseline mf prevalence estimated at 31% from an earlier study [13], started MDA one year later, in 2007. Coverage in all four districts has been consistently above 65% for the whole period, although MDA campaigns did not happen at exact one year intervals (Fig 2). Moreover, periodic surveillance in sentinel and spot-check sites revealed that, while all four districts have experienced dramatic prevalence reductions since the beginning of the programme, only Manakara Atsimo, Mananjary and Vohipeno had achieved levels of microfilaraemia below 1% (Fig 2).
This present study included three surveys carried out between November and December 2016 by separate teams in coordination with the National Programme (Fig 3). These include i) a school-based TAS carried out in the districts of Manakara, Mananjary, and Vohipeno (grouped as one evaluation unit, referred hereafter as “MAMAVO”); ii) a community survey in the adjacent district of Ifanadiana to obtain district-representative prevalence estimates; and iii) surveys in sentinel and spot-check sites of these four districts as part of the routine surveillance activities in the region. The two latter surveys were community-based and were targeted at children >5 years and adults of all ages. Despite the different designs, all surveys used the same diagnostic tests (FTS), and all teams received training and support from the National Programme for field survey execution.
The districts of Manakara, Mananjary and Vohipeno were merged into one single evaluation unit (EU “MAMAVO”) for the TAS evaluation, given their similarities in epidemiology and programme coverage over the past ten years (Fig 1). Survey design and execution was done strictly following WHO guidelines [10], with the support of “Survey Sample Builder” software (SSB, developed by the NTDs Support Centre) for sample size estimation and randomization. A school-based cluster survey was selected, with a total sample size of 1,692 children divided in 30 randomly selected schools. Detailed information on selected and replacement schools is available in S1 Table.
Before the beginning of the survey, authorities from the Ministry of Health and the Ministry of Education at the national, regional and district level participated in a three-day session of training and microplanning. School directors were informed of the survey in advance, and consent was obtained from the parents for inclusion of their children. Fieldwork was conducted between November 11–29 2016 by three teams of three people, including a lab technician, a district representative from the ministry of health, and a team member from the National Programme. At each school, all targeted children were organized in lines and were selected according to pre-existing lists produced by SSB, assisted by school teachers and staff. The TAS was continued until every randomly selected children in the target grades of all thirty schools had been tested. Thus, final sample size was larger than recommended (N = 1,825) due to differences in expected and actual class sizes and non-response rates.
A community survey to obtain district-representative estimates of FTS prevalence was conducted only in the district of Ifanadiana, which was not yet eligible for a TAS in 2016. To take into account spatial heterogeneity of LF, 545 individuals of ages 5 years or older were sampled using a cluster sampling scheme, with 30 random clusters at the fokontany level (smallest administrative unit consisting of several villages). Since villages in Ifanadiana vary greatly in terms of population size, one target village was selected and two additional villages from the same fokontany were used for replacement in case of insufficient sample sizes. Enumeration and selection of a random sample of 18 households within each village was done upon arrival by the field teams, in collaboration with the head of the village and community health workers. One member of each selected household was selected at random, excluding children under 5 years of age. Written consent was obtained from all survey participants, or their parents in the case of children. Fieldwork was coordinated by the National Institute of Public and Community Health (INSPC) and conducted between November 11–30 2016 by three teams of three people, each including a medical doctor, a lab technician, and a paramedic.
Besides the FTS prevalence rate, the survey included a brief questionnaire to gain a better understanding of the reasons why Ifanadiana was falling behind at LF elimination. Information in the questionnaire included demographic and socio-economic characteristics, knowledge related to MDA strategies, and previous practices to prevent LF such as preventive chemotherapy uptake and bed-net use.
As part of the LF monitoring and evaluation framework, WHO recommends National Programmes to routinely assess LF prevalence in both sentinel and spot-check sites. Sentinel sites are villages with a stable population of at least 500 inhabitants in an area of known high transmission or where achieving high coverage is difficult, and they are consistently surveyed since the beginning of the programme [10]. Spot-check sites have the same characteristics as sentinel sites, but are randomly chosen every year to avoid potential biases linked to program execution in sentinel sites. Selection of sites by the National Programme is done in collaboration with the district medical inspector, who provides a tentative list of villages that meet the above criteria. A lack of information about vector abundance or parasite prevalence means that site selection is based on symptomatic (chronic) LF cases known by the district officials or on locations where low MDA coverage rates have been reported. Following site selection, a convenience sample of 300 people is done, whereby the population is informed that LF testing will be carried out and volunteers meet the team in order to be tested.
As part of a World Bank project [14] that provided financial support for the fight against multiple NTDs in southeastern Madagascar, a prevalence survey was carried out in sentinel and spot-check sites of MAMAVO and Ifanadiana (among others) in 2016 to obtain updated prevalence estimates for the routine LF surveillance strategy of the National Programme. The sentinel site of each health district was included (same site as in previous surveys). Two spot-check sites per district (only one in Ifanadiana) were selected at random among villages of at least 500 inhabitants that were located further than 5 km from a health center. At each site, a convenience sample of 300 people was done among individuals of ages 5 years or older who agreed to participate in the study. While previous surveys in sentinel and spot-check sites had only assessed microfilaraemia, here CFA infection status of individuals was tested by FTS first, and then microfilaraemia was assessed for FTS positive cases. Similar to the community survey in Ifanadiana, this survey included a brief Knowledge, Attitudes and Practices (KAP) questionnaire administered to all participants. Fieldwork was coordinated by the National Malaria Control Program and conducted between November 23 and December 24 2016 by 4 teams of 3 people, each including a medical doctor, a lab technician or a paramedic, and 2 laboratory technicians.
In all three surveys, CFA was detected using FTS according to the manufacturer’s instructions. Briefly, 75μL of finger stick blood were collected between 8am and 5pm using the supplied micropipette, and the blood sample was added to the sample pad of the strip. Test results were read after 10 minutes, and any invalid tests (e.g. lack of a visible negative control line) were repeated. In addition, in the survey conducted at sentinel and spot-check sites, all positive FTS were examined for microfilaraemia. For these, a second sample of capillary blood was taken between 10pm-1am and was spread on a thin blood smear to evaluate the presence of microfilaraemia by microscopy.
For each survey, demographic characteristics of the study population were summarized and FTS prevalence rates were estimated for each age group and district. Levels of knowledge, attitudes and practices related to LF transmission were estimated for the overall population and for each age group. Differences between age groups were assessed using Pearson chi-square tests (two sided), and p-values were reported. Associations of FTS positivity with knowledge, attitudes and practices were studied. Odds ratios and p-values of these associations were estimated through univariate conditional logistic regressions, using the age group as matching variable. All data analyses were carried out using R software, version 3.4.1 and R package “survival”, version 2.41–3. Figures were created with R package “ggplot2”, version 2.2.1. Maps were developed using ArcMap, version 10.4.1.
Written consent was requested directly from all individuals (or their parents in the case of children) and obtained prior to participation in the community-based surveys. For the TAS, the national programme coordinated with the school directors to inform the parents in advance of the survey and seek permission to include their children in case they were selected via a written note. Permission for their children’s participation was written on the parent-teacher contact book and brought by the child to the school. All three surveys received the approval of the Madagascar National Ethics Committee. Individuals who tested positive were treated with a combination of albendazole and diethylcarbamazine, and those who presented signs of chronic infections were advised to seek treatment at the nearest health center. All statistical analyses were carried out on de-identified data in order to protect individuals’ confidentiality.
A total of 1,825 children participated in the TAS conducted in MAMAVO, 545 individuals of all ages participated in the Ifanadiana survey, and 3,325 individuals participated in sentinel and spot-check surveys in the four districts (Table 2). Most of the children included in the TAS (85.7%) had ages ranging between 5–7 years. Since the survey was targeted at first and second year classes at primary schools, rather than strict age ranges, some older children were also sampled (261 children ages 8–14 years). For sentinel and spot-check sites, children of ages 5–14 years represented about 35–50% of the population sampled. This proportion was significantly lower in the cluster survey at Ifanadiana, where they represented 20% of the sample. The sex ratio in the different surveys was balanced, although males made up for a smaller share in the sentinel and spot-check surveys (sex ratio 0.81). Information by district is available in S2 Table.
None of the 1,825 school-enrolled children that were sampled in the TAS tested positive to FTS (Table 3). This was consistent with results from sentinel and spot-check sites, which tested an additional 1,306 children of ages 5–14 years. None of the children in the 5–7 age group, and only one child in the age range 8–14 years in these routine surveys tested positive. Despite consistent negative results for children in these three districts, prevalence in adults was still high in two of them. Ten out of 431 people aged 15–45 (2.3%) and 9 out of 110 people aged 46–90 (8.2%) tested positive in Mananjary District. For Vohipeno, prevalence was 8.5% in ages 15–45 and 13.3% in ages 46–90, out of 421 and 135 people tested respectively. Two out of 79 adults in Mananjary and Vohipeno who tested positive to FTS had also detectable levels of microfilaria in their blood. Results for the adjacent district of Ifanadiana varied significantly depending on the survey. Results from sentinel and spot-check sites suggested that prevalence ranged from 1.5–2.0% in adults (ages 15–45), while all children under 15 years tested negative. In contrast, prevalence estimates from the randomized survey conducted in 30 clusters across the district were consistently higher for all age groups. Prevalence in this survey was below 4% for children under 15 years, increasing progressively with age to 15.5% in the 15–45 age group (95% CI 11.8–20.2) and up to 27.8% in adults aged 46–90 years (95% CI 20.4–36.6). Geographical distribution of positive cases was heterogeneous across Ifanadiana, with prevalence ranging from zero (in six of the fokontany) up to 30–40% (in five other fokontany), although sample sizes were too small and did not allow for robust estimations at the cluster level.
The KAP survey conducted at sentinel and spot-check sites in the four districts under study revealed that only 51.2% of the population surveyed reported taking MDA in the previous year, 63.8% in the previous 5 years (Table 4). Individuals aged 5–14 years reported lower uptake rates (45.6%) than those aged 15–90 years (54.8%). In contrast to the low MDA uptake rates, both bed net ownership and use were high in all age groups. More than 90% of people reported owning or sleeping under a bed net, and 88.2% reported sleeping under a bed net every day in the past year. Knowledge about LF was low in this population (42.2%) and varied significantly among age groups (Table 4). Less than one in ten individuals aged 5–14 years reported knowing about LF, compared with nearly two thirds of those aged 15–90 years.
Individuals who reported taking MDA in the past 5 years had significantly lower odds of infection (Table 5), as assessed through FTS (OR = 0.57; 95% CI 0.36–0.91), and this protective effect was stronger for individuals who reported taking MDA in the past 12 months (OR = 0.2; 95% CI 0.11–0.35). Moreover, individuals who had attended primary school or higher had significantly lower odds of infection than those who had never attended school (OR = 0.5; 95% CI 0.3–0.81). In contrast, knowledge about LF or bed net use was not significantly associated with FTS positivity. Despite higher FTS prevalence levels found in Ifanadiana district-representative survey, this population reported higher MDA uptake rates (66.2%) and knowledge about MDA (68.4%), although bed net use was lower (S3 Table). In this survey, MDA uptake and knowledge were not associated with FTS positivity (S4 Table). Only attendance to primary school or higher was significantly associated with a lower odds of infection (OR = 0.55; 95% CI 0.32–0.95).
At the turn of the 20th century, it was estimated that 120 million people were infected with LF globally and more than one billion were at risk of infection [3]. A strategy of MDA, following the 1997 World Health Assembly resolution to eliminate LF, has led to one of the most ambitious and successful interventions against a neglected tropical disease. Under sufficient levels of intervention coverage, transmission of LF can be interrupted within five years. To allow comparable monitoring and evaluation of national programmes, WHO designed an exhaustive plan that involves routine surveillance and the implementation of TAS to identify the areas where MDA can be stopped. Here we report on the results of the first TAS conducted in Madagascar (after the small district-island of Saint Marie), together with two other complementary prevalence surveys. We show that TAS results confirmed the consistent decline in LF observed in recent years and suggest that MDA can be stopped in our study area. However, the high FTS prevalence levels observed in the adult population, together with low reported MDA uptake rates, raise questions about a potential risk for recrudescence (e.g. if adult worms present in these populations are eventually able to produce microfilariae).
After ten rounds of MDA, we found no FTS positive children during the TAS conducted in three districts of southeastern Madagascar. This represents the first evidence for Madagascar that transmission is being halted in traditionally high endemic districts, and that MDA may be partially stopped in the country. These results are consistent with those reported in other endemic countries of Africa and Asia, where scale-up of TAS has allowed confirmation in multiple regions of a reduction in LF transmission below sustainable levels [15–18]. Despite progress, Madagascar still faces considerable challenges for LF elimination and its evaluation. Political instability together with a sudden drop in official development assistance between 2009 and 2014 contributed to the interruption of routine surveillance efforts in this period. Furthermore, a TAS scale-up has not recently been possible in the country due to financial constraints (a TAS has a median cost of over $20,000 [19]), despite having many districts with 5–10 rounds of MDA that fulfill the conditions for such evaluation. As a result, the National Programme continues to distribute preventive chemotherapy in potentially non-endemic districts, using resources that could be allocated to expanding the geographical reach of the MDA strategy, achieving nation-wide coverage.
The different surveys we conducted simultaneously in the same geographical area provided seemingly contradictory results and illustrate the complexity of evaluating lymphatic filariasis transmission. First, 2016 surveys in sentinel and spot-check sites of Mananjary and Vohipeno, with CFA prevalence levels over 2%, suggest that in retrospect a TAS should have not been conducted in these districts. However, the National Programme followed WHO guidelines regarding TAS eligibility [10] and the most recent surveys from 2013 had estimated levels of microfilaraemia lower than 1%, which was still the case in 2016. Second, prevalence estimates obtained from the district-representative survey in Ifanadiana were nearly twenty times higher than those from sentinel and spot-check sites in the district. It is worth noting that results were not adjusted for unequal probability of selection among individuals due to a lack of robust population estimates, which could have led to biases in the point estimates and confidence intervals. Yet, the gap is staggering and suggests that sentinel and spot-check sites were not representative of the district as a whole. WHO guidelines recommend that programmes select sites of known high transmission or low MDA coverage [10]. However, in low-resource settings where information about vectors or parasite prevalence is scarce, selection is frequently based on symptomatic LF cases, which do not necessarily reflect current transmission. In addition, the limited capacity of the National Programme to supervise community distribution of MDA could lead to inadequate administration (e.g. not directly observed uptake) or reporting, thus biasing coverage estimates and further undermining selection criteria for surveillance in sentinel and spot-check sites.
The high prevalence of antigenaemia observed in adults of three out of the four districts evaluated, after nine years or more of MDA, suggests that these populations are still infected with adult worms. A reason for including only children in TAS evaluations is that some treatments against LF (including diethylcarbamazine) have both partially macrofilaricidal and sterilizing effects on adult worms [20–23]. Thus, transmission may be stopped even when worms are still temporarily present in the population. However, the effectiveness and duration of these sterilizing effects are poorly understood. The presence of an infected pool of adults in the population could not only constitute a potential source of new infections, but can also increase the susceptibility for childhood infections mediated by in-utero immunological mechanisms [24,25]. Given the low number of microfilaremic individuals (2 out of 79 FTS positive in sentinel and spot-check sites), an immediate recrudescence of LF cases is unlikely. Yet, the longevity of the adult worms (5–7 years or more) coupled with high FTS prevalence, could indicate a risk for a recrudescence after a few years if MDA is stopped. This risk could be amplified by the nearby populations of Ifanadiana, where FTS district-prevalence was higher than 15%. The presence of such “hotspots” is consistent with previous studies in other African countries where some EUs have remained endemic even after 14 rounds of MDA, despite overall progress in nearby regions [26].
The results observed in our study, radically different for each age group, could be the result of distinct strategies implemented at schools and communities. A key component of the Madagascar MDA strategy is the school-based distribution of chemotherapy for all enrolled children 5–14 years of age, in partnership with other NTD programmes. For all other children and adults, MDA is facilitated through community health workers who distribute treatments on a door-to-door basis. However, uptake during community administration by health workers can be considerably more challenging, since household members may be absent, adults can decide not to take the drugs, or health workers may not be able to cover their assigned area [27]. Indeed, many studies have found very low uptake among adult populations (around 50%) due to a variety of context-specific factors [27–29]. Although results from KAP surveys of sentinel and spot-check sites did not support this hypothesis, the low level of knowledge about LF found in these children (less than one in ten) could have biased their responses. Since WHO guidelines promote an “integrated package” of NTD interventions that target children at schools [30–32], further studies should consider its potential impact on TAS results. Indeed, the current number of teachers participating in administration of preventive chemotherapy for NTDs globally is almost twice the number of health workers [30].
In conclusion, our study presents the recent progress in eliminating LF in Madagascar with the first successful TAS results in the island, but it suggests that evaluations of LF transmission may in some epidemiological contexts be more complex than generally recognized in international guidelines. Logistical local problems as well as ecological reasons may explain the heterogeneity of the results for different settings. In contexts where MDA for LF includes school-based administration, it may be appropriate to conduct prevalence surveys for the adult population or children not enrolled in school to complement the information gained from TAS results. As shown for Ifanadiana, cluster randomization for these complementary surveys may be important in order to gain a full geographical picture of transmission.
|
10.1371/journal.pgen.1003062 | Loss of the DNA Methyltransferase MET1 Induces H3K9 Hypermethylation at PcG Target Genes and Redistribution of H3K27 Trimethylation to Transposons in Arabidopsis thaliana | Dimethylation of histone H3 lysine 9 (H3K9m2) and trimethylation of histone H3 lysine 27 (H3K27m3) are two hallmarks of transcriptional repression in many organisms. In Arabidopsis thaliana, H3K27m3 is targeted by Polycomb Group (PcG) proteins and is associated with silent protein-coding genes, while H3K9m2 is correlated with DNA methylation and is associated with transposons and repetitive sequences. Recently, ectopic genic DNA methylation in the CHG context (where H is any base except G) has been observed in globally DNA hypomethylated mutants such as met1, but neither the nature of the hypermethylated loci nor the biological significance of this epigenetic phenomenon have been investigated. Here, we generated high-resolution, genome-wide maps of both H3K9m2 and H3K27m3 in wild-type and met1 plants, which we integrated with transcriptional data, to explore the relationships between these two marks. We found that ectopic H3K9m2 observed in met1 can be due to defects in IBM1-mediated H3K9m2 demethylation at some sites, but most importantly targets H3K27m3-marked genes, suggesting an interplay between these two silencing marks. Furthermore, H3K9m2/DNA-hypermethylation at these PcG targets in met1 is coupled with a decrease in H3K27m3 marks, whereas CG/H3K9m2 hypomethylated transposons become ectopically H3K27m3 hypermethylated. Our results bear interesting similarities with cancer cells, which show global losses of DNA methylation but ectopic hypermethylation of genes previously marked by H3K27m3.
| In plants and animals, repetitive DNA sequences and transposable elements are marked with DNA methylation, which is associated with methylation on lysine 9 of histone 3 (H3K9) and silencing. On the other hand, protein-coding genes, in particular the ones involved in differentiation processes, are targeted by Polycomb Group (PcG) proteins, which results in trimethylation of H3K27—another hallmark of transcriptional repression. These two systems of silencing are thought to be independent, but in this study we reveal an interplay between them. In the model plant Arabidopsis we show that, in a globally DNA–hypomethylated mutant, H3K27m3 marks can now be found at repeats and transposons; this is associated with a decrease of H3K27m3 at PcG targets, with some of them becoming targets of DNA and H3K9 methylation. Our data suggest that H3K27m3 prevents ectopic DNA/H3K9 methylation at cryptic DNA methylation targets, which could provide a novel significance for this mark with regard to genome integrity. In addition, this study reveals interesting similarities with cancer cells, which show global losses of DNA methylation but ectopic hypermethylation of genes previously marked by H3K27m3, and suggests the potential of Arabidopsis as a system for understanding mammalian developmental and cancer biology.
| Post-transcriptional modifications of histone tails—and combinations thereof—are thought to define specific chromatin structures and transcriptional states across eukaryotes [1], [2]. In both animals and plants, trimethylation of histone 3 lysine 27 (H3K27) and dimethylation of histone 3 lysine 9 (H3K9) (and/or trimethylation in metazoa) are two, generally alternative, hallmarks of transcriptional repression. In Arabidopsis thaliana, H3K27m3 is deposited by Polycomb group (PcG) proteins in euchromatic domains containing protein-coding genes—in particular, transcription factors and genes involved in developmental transitions [3], [4], [5]. H3K27m3 marks are largely non-overlapping with H3K9m2 and cytosine DNA methylation, which are targeted to repeated sequences throughout the genome and associated with silent, constitutive heterochromatin [4]. However, H3K27m3 was found to mark selected transposons and repeated sequences in some particular contexts when they are DNA hypomethylated such as in the met1 mutants [6] or in endosperms [7].
In mammals DNA methylation is usually observed exclusively in the CG-dinucleotide context, while in Arabidopsis thaliana cytosines are methylated in every context. At least three DNA methyltransferases control DNA methylation in Arabidopsis, each with its own sequence preference: CG, CHG, or CHH (where H is any base except G). Establishment of cytosine methylation in all sequence contexts is catalyzed by DOMAINS REARRANGED METHYLTRANSFERASE 2 (DRM2), the plant homolog of mammalian DNMT3a and DNMT3b. DRM2 is guided to chromatin by siRNAs in a pathway known as RNA-directed DNA methylation (RdDM). This pathway also maintains DNA methylation in the asymmetric CHH context. CG methylation is catalyzed by DNA METHYLTRANSFERASE 1 (MET1), the plant homolog of mammalian DNMT1, and this mark is passively maintained during replication. Finally, CHG methylation is mostly catalyzed by CHROMOMETHYLASE 3 (CMT3)—a plant-specific DNA methyltransferase that contains a chromodomain that recognizes dimethylated histone tails at lysine 9 (H3K9m2) [8]. In turn, CHG methylation is recognized by the SRA domain within the KRYPTONITE (KYP) H3K9m2 methyltransferase. Therefore, CHG methylation is largely maintained through a reinforcing loop of DNA and H3K9 methylation. This is consistent with genome-wide studies showing that CHG methylation and H3K9m2 are highly co-incidental [9], [10]. At some heterochromatic loci, H3K9m2 is also dependent on CG methylation, potentially through other SRA-domain containing proteins [11], [12].
Primary targets of DNA methylation in Arabidopsis include transposons, repetitive sequences, and occasionally genes when they contain repeats in their promoter [13], [14]. In these cases, DNA methylation is present in all three cytosine contexts and is associated with H3K9m2 marks and transcriptional silencing [9], [15]. However, at least 30% of expressed genes [15], [16], [17], [18], [19] show a significant amount of DNA methylation in their transcribed regions (or gene bodies). In this case, DNA methylation is dependent on MET1, is restricted to CG sites, is not associated with H3K9m2, and does not result in gene silencing [17], [18]. The function of this methylation remains unknown, although a recent study gave some insights into its regulation: hundreds of genes in Arabidopsis were shown to gain non-CG methylation (mainly at CHG sites) in plants mutated in the increase in bonsai methylation 1 (ibm1) gene [20]. IBM1 encodes a Jumonji C-domain protein with H3K9m2 demethylase activity [21], [22], and was initially identified in a genetic screen for mutants showing ectopic cytosine methylation of the BONSAI (BNS) gene. This discovery raised the idea that genes are actively being protected from acquiring H3K9m2 methylation. Interestingly, ectopic CHG methylation and associated H3K9m2 have been previously reported in the met1 background [16], [23], [24], [25]. However, neither the mechanism nor the biological significance of the ectopic DNA/H3K9m2 methylation in met1 is currently understood.
In order to gain a better understanding of the ectopic DNA methylation in met1, and to test the hypothesis that CHG hypermethylation in met1, like in ibm1, could be the result of crippled IBM1-mediated control of H3K9m2 at genes, we generated high-resolution, genome-wide maps of H3K9m2 methylation in met1 and ibm1 mutants. Our results revealed that hundreds of genes become H3K9m2 hypermethylated in both met1 and ibm1 backgrounds. However, the sets of genes most H3K9m2 hypermethylated in each mutant were largely non-overlapping, suggesting that MET1 and IBM1 regulate H3K9m2 at different subsets of genes. The genes most H3K9m2 hypermethylated in met1 tended to be either genes already marked with less extensive levels of H3K9m2 or, more surprisingly, genes marked with H3K27m3. To explore the relationship between the repressive H3K9m2 and H3K27m3 marks further, we mapped H3K27m3 levels using wild type and met1 plants. We observed a significant loss of H3K27m3 at PcG-targeted genes in met1, in particular at the ones that became DNA and H3K9m2 hypermethylated in met1. This phenomenon was accompanied by a massive redistribution of H3K27m3 marks to many H3K9m2 and/or CG hypomethylated loci in met1 such as transposons. Finally, to determine the effect of these changes in the epigenetic landscape on transcription, we conducted RNA-seq experiments using wild type and met1 plants. These analyses showed that the PcG targets remain relatively unexpressed upon replacement of H3K27m3 marks by H3K9m2 marks. Our results suggest that H3K27m3 and H3K9m2/DNA methylation are mutually exclusive, and can replace one another in a locus specific manner. In addition, these data bring important new insight into the biology of met1 mutants by showing an important role for MET1 in maintaining H3K27m3 patterning at PcG targets. Finally, our observations draw a striking parallel between the epigenetic phenomena displayed in the met1 mutant and the local DNA hypermethylation observed in cancer cells.
To examine the relationship between MET1 and IBM1 in negatively controlling H3K9m2 deposition throughout the genome, we generated high-resolution genome-wide maps of H3K9m2 in the two first inbred generations of met1 and ibm1 rosette-stage mutants by performing chromatin immunoprecipitation experiments coupled with whole-genome Roche Nimblegen microarray analyses (ChIP-chip). We observed hundreds of regions that became H3K9m2 hypermethylated in each of the mutants. However, while significant H3K9 hypermethylation was observed in the first generation of ibm1 mutants (ibm1-1st), this phenomenon only become clearly apparent in the second generation of met1 mutant (met1-2nd). These results are in contrast with previous immunofluorescence analyses showing appearance of H3K9m2 in the euchromatic, gene-rich regions only after three generations of the absence of a functional MET1 allele [25], but are consistent with genome-wide DNA methylation analyses where CHG ectopic methylation (and presumably H3K9 dimethylation) were evident in the flowers of met1 first generation homozygous mutants [16]. This suggests that immunofluorescence experiments may not be sensitive enough to detect de novo H3K9m2 patterns in the second inbred generation.
Regions of H3K9m2 hypermethylation were defined (using the BLOC algorithm [26]) and all the analyses performed in the first generation for ibm1 and the second generation for met1. The most hypermethylated genes were defined as genes that overlap with defined hypermethylated regions by at least 150 bases (the approximate length of DNA wrapped around one nucleosome). By this method, 1833 genes (6.49% of all genes, Table S1) were found to be H3K9m2 hypermethylated in met1. This set of genes included SUPERMAN (SUP) and AGAMOUS (AG), which have previously been reported to be DNA hypermethylated in met1 [23], [24], [27] (Figure S1). In addition, we examined published whole genome bisulfite sequencing data obtained from flowers of first generation met1 homozygous mutants, in which genic hypermethylation was readily detected [16]. We found that the set of genes that are H3K9 hypermethylated in met1 in our experiment also displayed increased levels of non-CG methylation, particularly in the CHG context, at individual loci and in a genome-wide manner (Figure 1A, Figure 1B, Figure S1). This indicates that the feed-forward loop between H3K9 and CHG methylation is also active at ectopically methylated loci in met1. However, we note that the comparison between the patterns of DNA and H3K9 methylation is limited by the use of different developmental stages in the two studies.
In ibm1 mutants, 1682 genes (5.96% of all genes, Table S2) were found to be H3K9m2-hypermethylated. Interestingly, this set of genes was largely distinct from the set observed for met1, with only 113 genes being significantly hypermethylated in both backgrounds (Figure 1C, 1D). These results imply that there are at least two mechanisms at play in the protection of genes from ectopic DNA and H3K9m2 methylation, one that depends on IBM1 and another that depends on MET1. One possibility is that the small overlap between the two sets is due to the reduced IBM1 mRNA levels in met1 mutant (Figure 1E). Consistent with this notion, we found that met1 usually had a smaller effect on H3K9m2 hypermethylation at these sites than ibm1 (Figure S2). A recent study reported that the re-establishment of IBM1 expression in met1 mutants restored the wild-type H3K9m2 patterns at selected loci and suggested that down-regulation of IBM1 could account for most of H3K9m2 relocation at genes [28]. However, the use of stringent thresholds to define H3K9m2 hypermethylated regions revealed that the most hypermethylated genes in met1 are usually not targets of IBM1.
One difference between the characteristics of genes most hypermethylated in ibm1 versus met1 is that most of the H3K9m2 hypermethylated genes in ibm1 are moderately expressed in wild-type, while many of the hypermethylated genes in met1 are lowly expressed (Figure 1F) or silent (Figure 1G).
By examination of individual loci (Figure 2A, Figure S3A) and genome-wide analysis (Figure 2B and 2C), we observed that many hypermethylated genes in met1 were enriched for genes pre-marked with H3K9m2 in wild-type (“Class I” genes), presumably due to the presence of repeats within their coding-region (Figure 2A) that do not necessarily correspond to transposable elements (Figure S3B). 531 such “Class I” genes were identified (28.9% of all H3K9m2 hypermethylated genes in met1). At these genes, siRNAs levels were increased in met1 (Figure 2D, Figure S3C), indicating that loss of CG DNA methylation at these genes, either upstream or in the coding-region, results in the stimulation of de novo methylation by the RdDM pathway, which in turn likely leads to increased H3K9m2 levels via the maintenance of CHG methylation which involves the H3K9m2 HMTase KRYPTONITE (KYP).
We also observed a second class of H3K9m2 hypermethylated genes in met1, including SUPERMAN or AGAMOUS, which consistently display H3K27m3 marks in wild type plants at this same developmental stage (“Class II” genes) (Figure 3A and 3C, Figure S4A and S4B). With our stringent parameters, this class comprised 515 genes (28.1% of all H3K9m2 hypermethylated genes). However, this number is an under-estimation as we noted a large number of genes that had the characteristics of Class II genes but were not retrieved by our conservative cutoffs for defining H3K27m3-marked genes in wild type (Figure 3D, Figure S5A). These genes seemed to largely account for the H3K9m2 hypermethylated genes in met1 that were not included in Class I or II (data not shown). In addition, genome-wide analyses revealed that genes H3K9m2 hypermethylated in met1 are significantly enriched in H3K27m3 marks relative to all genes (Figure 3B). Together, these observations indicate that the ectopic H3K9m2/DNA methylation in met1 is preferentially targeted to regions enriched in H3K27m3 in wild type. These observations are strikingly reminiscent of the phenomena displayed by human cancers cells where ectopic de novo DNA methylation occurs predominantly at genes specifically marked with H3K27m3 in either the corresponding normal adult cells or in the progenitor cells [29], [30], [31].
While regions that gained H3K9m2 in met1 were enriched for sites marked with H3K27m3 in wild type, only 7.8% of PcG-targeted genes (515 genes out of 6592) gain H3K9m2 in met1 (Figure S5B). While this number is likely an under-estimation (due to our conservative cutoffs), this nonetheless indicates that additional factors must contribute to the observed phenomenon. To identify such factors, we looked for additional features of H3K9m2 hypermethylated genes in the mutant background and found that they were enriched in sequences annotated as “dispersed repeats”, but were not annotated as transposable elements but rather as regions of homology between gene families (Figure 3E, Figure S4). Specifically, genes H3K9m2 hypermethylated in met1 were often found in tandem and had similar gene ontologies indicating recent gene duplication. In other cases, some H3K9m2 hypermethylated genes with paralogous domains were located on different chromosomes such as the transcriptions factors (At3g58780, At2g42830) related to AGAMOUS (At4g18960) by their MADS-box domain. Interestingly, tandemly repeated genes were also shown to be represented among H3K27m3-marked genes [3]. The homologous nature of the genes H3K9m2 hypermethylated in met1 suggests that a sequence-specific process—such as RdDM—may be involved in the formation of these ectopic methylation patterns. The presence of SUP DNA hypermethylated alleles (also known as clark kent or clk alleles) in a globally hypomethylated background was previously shown to depend on both the CMT3 and RdDM pathway [32]. Interestingly, in wild-type plants, we detected SUP-hybridizing small interfering RNAs, that did not originate from the SUP locus (since they were still detected in a strain with a deletion of the SUP gene) (Figure 3F). This shows that SUP-hybridizing siRNAs produced by another locus might potentially target SUP in trans. Consistent with this idea, the gene families H3K9m2 hypermethylated in met1 that we examined tended to match with at least one potential siRNA-generating locus (Figure S4A). Thus, our observations suggest that some paralogous genes might be cryptic targets of RdDM, and become methylated only in met1. This does not seem to be due to an increase of small RNAs at these sites in met1 (Figure S5D). One possibility is that a decrease in H3K27m3 marks could contribute to the onset of siRNA-directed DNA methylation (RdDM), after which methylation would then be maintained by H3K9m2-CMT3 feed-forward loop.
Since ectopic H3K9m2 and DNA hypermethylation in met1 occurs at PcG targets, and PcG targets are usually non-overlapping with siRNAs and H3K9m2 [4], we sought to test whether H3K27m3 levels are reduced in met1 at these loci. To this end, we generated high-resolution genome-wide maps of H3K27m3 in rosette-stage met1 mutant plants and observed a massive redistribution of H3K27m3 levels throughout the genome. H3K27m3 levels were significantly decreased at ectopically H3K9m2 hypermethylated genes (Figure 4, Figure S9) and were significantly increased at H3K9m2 hypomethylated regions (Figure 5A and 5B, Figure S9), namely transposons and other repetitive DNA elements. The extent of the H3K27m3 decrease at H3K9m2 hypermethylated regions was correlated with the extent of H3K9m2 ectopic methylation at the genes tested by quantitative PCR (Figure 4C). However, we also observed that many PcG-targeted genes with limited or no ectopic H3K9m2 were also partially depleted of H3K27m3 marks in met1, but on average the decreases in H3K27m3 at these genes seemed smaller than observed for PcG-targeted genes that gained H3K9m2 and DNA methylation (Figure 4D, Figure S6A). A possibility is that the smaller decreases in H3K27m3 at these loci could result from the partial relocation of H3K27m3 marks and/or PcG complexes to transposable elements and heterochromatic genes (Figure 5A, Figure 5B, Figure S8). In addition, the massive increase of H3K27m3 marks at transposons is likely contributed by a global increase of H3K27m3 marks (Figure S6B) and of PcG gene expressions. According to our RNA-seq data, in met1, FIE, CLF and SWN expression levels are increased by 33% (P = 0.005), 13% (P = 0.33), and 32% (P<0.001), respectively).
An increase in H3K27m3 marks was previously observed at discrete heterochromatic loci and at chromocenters in met1 [6] and our data show that this phenomenon can now be extended to hundreds of sites throughout the genome (Table S3). These findings suggest a model in which H3K9m2 and/or associated DNA methylation excludes H3K27m3 from heterochromatic loci in wild-type plants. To better understand the contributions of DNA and H3K9m2 methylation on H3K27m3 exclusion, we compared the pattern of H3K27m3 marks at several well-characterized transposable elements in various mutant backgrounds. Transposable elements such as ROMANIAT5, AtCOPIA28 lost H3K9m2 marks in the triple suvh4 suvh5 suvh6 (suvh456) mutant (in which both H3K9m2 and CHG methylation are reduced drastically, but not CG methylation) but H3K9m2 was not lost at these sites in the met1 mutant (Figure 5C). However, we observed that the met1 mutant exhibited a stronger increase in H3K27m3 marks at these sites compared to suvh456 (Figure 5C). In addition, at AtCOPIA4 (in its 3′half), H3K9m2 levels were reduced to the same extent in both suvh456 and met1, yet only met1 gained H3K27m3 at this locus (Figure 5C). Together, these results were consistent with previous data in the suvh4 mutant [6] and suggest that the loss of DNA methylation at heterochromatic loci in met1, rather than the loss of H3K9m2 marks, is associated with an increase in H3K27m3. This idea is further supported by the chromosomal distributions of H3K9m2 and H3K27m3 in met1 which show that H3K27m3 is targeted to centromeric sites that are free of CG methylation in this background but still contain similar levels of H3K9m2 or even increased levels of H3K9m2 (for example transposons behaving like Class I genes) (Figure S8).
Notably, there was no consistent gain of H3K27m3 at AtMU1 and at AtCOPIA4 (in its 5′half) (Figure S7) suggesting that loss of CG methylation and associated H3K9m2 alone was not sufficient to induce H3K27m3 deposition. A recent study suggested that a high density of unmethylated CpG sites could be sufficient for vertebrate Polycomb recruitment [33]. Consistent with this idea, we found that the density of CG sites was higher at transposable elements that gained H3K27m3 in met1 (Figure 5D) (including AtCOPIA28, ROMANIAT5, AtCOPIA4-3′half) than the ones that did not (including AtMu1 and AtCOPIA-5′half). Thus, CG density may contribute to the differential recruitment of PcG complexes to transposons in met1.
To gain insight into the biological significance of the relocation of H3K27m3 to heterochromatic loci and of H3K9m2/DNA methylation to PcG genes, we performed RNA-seq in wild type and met1 plants. Consistent with previous locus-specific analyses [6], transposable elements were usually reactivated in met1, despite the presence of ectopic H3K27m3 (Figure 6A). Therefore, H3K27m3 is not as competent as CG methylation and associated H3K9m2 in the silencing of transposons. Notably, the transposable elements targeted by H3K27m3 in met1 tend to be lowly expressed in wild type (Figure 6A, Figure 6B). Further analyses of transposon expression in a fie-met1 double mutant will be required to determine whether H3K27m3 marks can at least partially compensate for the loss of CG and H3K9m2 methylation in met1 and whether the increase of H3K27m3 at these sites is a back-up mechanism deployed by the plant cell to avoid massive transposon expression and transposition.
We found that H3K9m2 hypermethylated loci generally did not alter expression levels (Figure 6C). This was also true at PcG-genes despite a significant loss of H3K27m3 marks (Figure 6D). This finding suggests a functional redundancy between H3K9m2 and H3K27m3 at PcG-targets in leaves, even though the dynamics of these two silencing marks are thought to be quite distinct, with H3K27m3 marks acting in a reversible manner during the course of development to allow developmental switches [5] and H3K9m2 acting in a permanent manner to lock a gene or transposon into a silent heterochromatic state. With these differences in epigenetic plasticity in mind, we propose that the replacement of H3K27m3 by H3K9m2 at PcG-targets may contribute to array of developmental defects, including many floral defects, which are observed in the met1 mutant by locking PcG-target genes into a stably silent state, which is then unresponsive to developmental cues. Alternatively, the reactivation of key PcG-targets that lose H3K27m3 and become reactivated in met1 could also contribute the developmental defects, reminiscent of those displayed in the lhp1 mutant where H3K27m3-mediated silencing is impaired.
In Arabidopsis, loss of the maintenance CG DNA methyltransferase, MET1, results in DNA and H3K9 hypermethylation at some specific loci and in hypomethylation at other regions. Despite the strong parallels between this epigenetic state and those described in numerous cancers, how these patterns are established in met1 and their biological significance has remained unknown. In this study, we analyzed the genome-wide patterns of ectopic H3K9m2 methylation (which mirrors DNA methylation) in the otherwise globally DNA hypomethylated met1 mutant and provide significant insights into these questions. Based on our findings, we propose a model that accounts for the ectopic DNA and H3K9m2 methylation observed in met1 mutant. First, genes that gain DNA and H3K9 methylation in met1 fall into three categories: (1) a small class of genes that are affected by both met1 and ibm1 that are presumably IBM1 targets sensitive to the reduced levels of IBM1 expression in the met1 background, (2) genes that possess low levels of H3K9m2 in wild type plants and become H3K9 hypermethylated in met1 and (3) PcG-target genes (i.e. H3K27m3 pre-marked genes) that lose H3K27m3 marks in met1, but gain H3K9m2 marks. Interestingly, concomitant with the decrease in H3K27m3 at PcG targets, H3K27m3 levels increase at transposons and other heterochromatic loci where unmethylated CG sites may facilitate the recruitment of PcG complexes. The replacement of CG DNA methylation by H3K27m3 in met1 and vice-versa suggests that these two marks are mutually exclusive in Arabidopsis, as previously demonstrated in mammals at some imprinted loci [34] as well a in cancer cells [35]. While the exclusion of H3K27m3 by DNA methylation was previously proposed in Arabidopsis [6], [7], our findings add strength to this assertion and further suggest that it is the loss of H3K27m3 at PcG targets that contributes to the occurrence of H3K9m2 and DNA hypermethylation. This causative relationship is supported by the observation that the H3K27m3 decrease is not specific to H3K9m2 and DNA hypermethylated genes although it is stronger at these loci, consistent with the notion of mutual exclusion.
Interestingly, the PcG-target genes that become H3K9m2 and DNA hypermethylated may also represent cryptic targets of RdDM as they tend to have higher than average levels of sequence homology with other regions in the genome, many of which are known to generate siRNAs. Many genes such as transcription factors are part of large families and the presence of H3K27m3 at these loci may have the dual role of mediating transient, reversible repression and excluding RdDM and associated H3K9m2 methylation. How Polycomb-complexes are recruited to deposit H3K27m3 marks is still unknown in Arabidopsis [5]. However, at the FLC locus, there are cis-sequences that have been shown to be important for the recruitment of PRC2 (Polycomb Repressive Complex 2) through a long-coding RNA [36], [37]. Our data suggest that a high density of unmethylated CG sites, as previously observed in vertebrates, may be another factor facilitating PcG recruitment (Figure 5D). Further analyses may identify cis sequences in the transposons targeted by PRC2 in met1 and/or show a general role for non-coding RNAs in PRC2 recruitment. Alternatively, heavily methylated CG sites such as those seen in transposons, could recruit a H3K27m3 demethylase which would be inactive in met1. Future mechanistic exploration of these new epigenetic phenomena in met1 will likely bring insight into the recruitment of PcG complexes as well as RdDM components.
Our observations also provide a possible explanation for the drastic developmental phenotypes displayed by the met1 mutant: genes targeted by ectopic DNA and H3K9m2 methylation in met1 are PcG-targets in wild type plants, which are enriched in genes involved in transcriptional regulation and development. At specific developmental stages, for example during the vegetative phase where our analyses were performed, it appears that H3K9m2 marks are functionally redundant with H3K27m3 marks since the vast majority of genes with either mark in this study remained silent. However, the replacement of a transient repressive mark such as H3K27m3 by a stable silencing mark such as H3K9m2 may affect gene transcription during specific developmental windows where H3K27m3 marks are removed, thus impairing critical developmental switches and contributing to the myriad of developmental phenotypes observed in met1 mutants. Furthermore, the finding that hypomethylated regions of the genome induced by loss of MET1 in vegetative tissue can become targets of the Polycomb silencing machinery raises the question of whether hypomethylation of the genome caused by other processes also leads to PcG-targeting and gene silencing. Several examples of naturally occurring global hypomethylation have recently been described. These include the endosperm (plant extra-embryonic tissues), which is globally hypomethylated, due to the activity of the DNA demethylase DEMETER but also possibly due to down-regulation of MET1 in this context [38]. Interestingly, H3K27m3 was found at DNA hypomethylated transposable elements and genes that had less CG methylation than in the vegetative tissues [7]. In this respect, the strong endosperm phenotype observed after loss of polycomb function (proliferation and eventually seed abortion [39][40]) could indicate the crucial role of H3K27m3 marks at DNA hypomethylated sites. In addition, other studies revealed DNA hypermethylation (presumably associated with H3K9m2 hypermethylation) of specific sites in the endosperm [41]. However, the nature of these sites has not been investigated and it is possible that PcG-targets in the endosperm are similarly affected as in the vegetative tissues of met1 mutants.
Finally, our work in an Arabidopsis globally DNA hypomethylated mutant has uncovered striking similarities with epigenetic phenomena occurring in human cancer cells. First, H3K9m2 and DNA hypermethylated promoters in human cancer cells tend to be marked with H3K27m3 in the corresponding adult cells or in the progenitor cells they are derived from. Another point of convergence is the decrease of H3K27m3 marks associated with the ectopic gain of H3K9m2 in both contexts [42]. Finally, repressive chromatin formation, mediated in particular by H3K27m3, was observed at DNA hypomethylated regions in breast cancer cells [35]. The same study demonstrated genome-wide, mutual exclusivity of these two marks, which had been previously shown at one imprinted locus in mouse [34]. The striking similarities between the epigenetic landscapes of a globally hypomethylated mutant, the globally hypomethylated endosperm and human cancer cells suggest common underlying mechanisms, and suggests the potential of future Arabidopsis research as a framework for understanding developmental and cancer biology.
We used the met1-3 allele [43] and the SALK_006042 line for isolating ibm1 mutants. met1-1st generation homozygous mutants and ibm1-1st generation homozygous mutants were isolated from a segregating population by genotyping. met1-2nd generation second mutants are the progeny of a single met1-1st generation homozygous mutant that was partially fertile. The entire shoots of 3 weeks old Arabidopsis plants (Col-0 ecotype), grown for 3 weeks under continuous light, were harvested, cross-linked as described in [9], frozen under liquid nitrogen and grown to powder (2 g). Arabidopsis chromatin enriched for H3K9m2 and H3K27m3 was immunoprecipitated using an antibody that specifically recognizes H3K9m2 (Abcam 1220) and an antibody that specifically recognizes H3K27m3 (Upstate 07-449) respectively. Unmodified H3 was also immunoprecipitated (Abcam 1791-100). H3 ChIP and input DNA were used as controls. ChIP, DNA purification and amplification were performed as in [9]; Roche Nimblegen performed labelling and hybridization of the samples, washing and scanning.
All ChIP signals were normalized with either H3 ChIP or input genomic DNA by taking the log2 ratio and adjusted so that the average log2 ratio score across the genome was zero. For each mark, four independent ChIP experiments on four different biological sample replicates were performed. The two first independent ChIPs were pooled and used to generate libraries and for subsequent chip-hybridization. The two other independent ChIPs were used for validation of the ChIP-chip data by qPCR. qPCR was performed in duplicates or triplicates. H3K9m2 hypermethylated regions were defined by using BLOC [26]. The log2 ratio of mutant to wild-type scores were taken, Z-score transformed, and a cutoff of 0.75 was applied for all met1 datasets, and 0.8 was applied for all ibm1 datasets. The cutoffs were determined based on visually examining genome-wide data and validation experiments, to minimize false-positives but also to avoid missing truly hypermethylated genes. The choice of two different cutoffs for the ibm1 and met1 analyses presumably results from a slight difference in the ChIP efficiencies between the two experiments. All sets of defined regions were significant (FDR<0.01). To define H3K27m3 enriched genes, H3K27m3 enriched regions were identified. The genome was tiled into 200 bp bins (100 bp overlap) and z-scores were calculated. A Z>2 cutoff was applied, and regions within 200 bp were merged. Genes that overlapped by at least 150 bp were defined to be H3K27m3 enriched genes. Genes that became H3K27m3 hypermethylated in met1 were defined by calculating the log2 ratio of mutant to wild-type scores in 200 bp bins (100 bp overlap), Z-score transformed, and Z>2 cutoff was applied. Regions within 200 bp were merged, and finally only regions >500 bp in size were selected. Genes that overlapped by at least 150 bp were defined to be H3K27m3 hypermethylated genes in met1. Genes that became H3K27m3 hypomethylated in met1 were defined in a similar matter as hypermethylated genes, except that a Z<-3 cutoff was applied.
Plants grown in the same conditions were used for RNA extraction using a standard protocol. The RNA of two independent biological sample replicates was extracted and pooled for RNA-seq analysis, and a third replicate used to validate the data by RT-qPCR (performed in duplicates or triplicates). RNA-seq libraries were generated following the manufacturer instructions (Illumina). DNA methylation analyses were performed using published whole genome bisulfite sequencing data and RNA-seq data [16]. Tandem-repeats were defined in [17], dispersed repeats were defined in [16] and H3K9m2-regions were defined in [9].
Plants grown in the same conditions were used for Histone Western Blot experiments. Chromatin was extracted as was performed for ChIP. 30 ug and 10 ug of proteins were loaded to detect H3K27m3 and histone H3 respectively, and the antibodies used for ChIP experiments were used for detection.
http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=dvotficqwuacufg&acc=gse37075
|
10.1371/journal.ppat.1005915 | Vasodilator-Stimulated Phosphoprotein Activity Is Required for Coxiella burnetii Growth in Human Macrophages | Coxiella burnetii is an intracellular bacterial pathogen that causes human Q fever, an acute flu-like illness that can progress to chronic endocarditis and liver and bone infections. Humans are typically infected by aerosol-mediated transmission, and C. burnetii initially targets alveolar macrophages wherein the pathogen replicates in a phagolysosome-like niche known as the parasitophorous vacuole (PV). C. burnetii manipulates host cAMP-dependent protein kinase (PKA) signaling to promote PV formation, cell survival, and bacterial replication. In this study, we identified the actin regulatory protein vasodilator-stimulated phosphoprotein (VASP) as a PKA substrate that is increasingly phosphorylated at S157 and S239 during C. burnetii infection. Avirulent and virulent C. burnetii triggered increased levels of phosphorylated VASP in macrophage-like THP-1 cells and primary human alveolar macrophages, and this event required the Cα subunit of PKA. VASP phosphorylation also required bacterial protein synthesis and secretion of effector proteins via a type IV secretion system, indicating the pathogen actively triggers prolonged VASP phosphorylation. Optimal PV formation and intracellular bacterial replication required VASP activity, as siRNA-mediated depletion of VASP reduced PV size and bacterial growth. Interestingly, ectopic expression of a phospho-mimetic VASP (S239E) mutant protein prevented optimal PV formation, whereas VASP (S157E) mutant expression had no effect. VASP (S239E) expression also prevented trafficking of bead-containing phagosomes to the PV, indicating proper VASP activity is critical for heterotypic fusion events that control PV expansion in macrophages. Finally, expression of dominant negative VASP (S157A) in C. burnetii-infected cells impaired PV formation, confirming importance of the protein for proper infection. This study provides the first evidence of VASP manipulation by an intravacuolar bacterial pathogen via activation of PKA in human macrophages.
| Q fever, caused by the intracellular bacterial pathogen Coxiella burnetii, is an aerosol-transmitted infection that can develop into life-threatening chronic infections such as endocarditis. The pathogen preferentially grows within alveolar macrophages in a phagolysosome-like compartment termed the parasitophorous vacuole (PV). C. burnetii actively manipulates host cAMP-dependent protein kinase (PKA) signaling to promote PV formation and cell survival. Identification of bacterial effector proteins that manipulate PKA and downstream target proteins is critical to fully understand pathogen-mediated signaling circuits and develop new therapeutic strategies. Here, we found that PKA controls vasodilator-stimulated phosphoprotein (VASP) activity to promote PV formation and bacterial replication. VASP regulates actin-based motility used by a subset of intracellular bacteria for propulsion through the host cell cytosol and into bystander cells. However, C. burnetii does not use actin-based motility and replicates throughout its life cycle within a membrane bound vacuole. Thus, this study provides the first evidence of VASP manipulation by an intravacuolar bacterial pathogen. Characterization of VASP function in PV formation and identification of additional PKA substrates that promote infection will provide new insight into host-pathogen interactions during Q fever.
| Coxiella burnetii is an intracellular bacterial pathogen that causes the zoonosis human Q fever. C. burnetii infects domestic mammals and livestock, which serve as the primary reservoir for the pathogen in nature. C. burnetii is shed from infected animals in body fluids, particularly during parturition, resulting in human infection by inhalation of contaminated aerosols [1]. Q fever often manifests as a flu-like acute disease with atypical pneumonia, and most individuals recover without medical intervention. However, less than 5% of infected individuals develop chronic disease that largely manifests as endocarditis, and to a lesser extent as bone infection, vascular complications, and granulomatous hepatitis [2]. The fatality rate of patients with Q fever endocarditis approaches 60% if left untreated [1]. Chronic Q fever diagnosis is extremely difficult and treatment requires a prolonged course of antibiotic therapy that is not completely effective. Considering the global distribution of C. burnetii, and recent outbreaks in rural parts of the world, Q fever is now considered an emerging infectious disease [3,4].
In aerosol-acquired human infections, C. burnetii preferentially targets macrophages that reside in alveolar spaces. Virulent bacteria enter macrophages by αVβ3 integrin receptor-dependent phagocytosis [5]. Following invasion, organisms reside in tight-fitting phagosomes that decorate with the autophagosome marker LC3 and early endosomal Rab5 [6]. Maturation of nascent phagosomes into unique, replication-permissive parasitophorous vacuoles (PV) is achieved by continual heterotypic fusion with autophagosomes, endosomes, and lysosomes [7–9]. Although lysosome fusion creates an acidic, hydrolytic environment, C. burnetii has adapted to resist degradation, and low pH activates bacterial metabolism and subsequent replication [10,11]. C. burnetii actively controls infection by directing endosomes, vesicles, autophagosomes, and lysosomes to the PV, delivering nutrients, lipids, and proteins to the expanding vacuole that ultimately occupies most of the host cell cytosol [12,13]. Secretion of bacterial effector proteins into the host cytosol via a Dot/Icm type IV secretion system (T4SS) is essential for PV formation [14,15]. Some C. burnetii effectors contain sequences that resemble eukaryotic motifs and domains, bind host cell proteins, and manipulate host signaling to promote infection [16–18].
To support an intracellular lifestyle, C. burnetii manipulates several host cell signaling pathways. The pathogen activates Akt and Erk1/2 to promote host cell survival and allow completion of a lengthy infectious cycle [19]. C. burnetii also hijacks cyclic adenosine monophosphate (cAMP)-dependent protein kinase (PKA) signaling to support PV formation and prevent apoptotic cell death. When activated by cAMP, PKA binds and phosphorylates several downstream target proteins that modulate responses including cytokine production, apoptosis, and cytoskeletal remodeling. Differential phosphorylation of PKA substrates has been observed in C. burnetii-infected macrophages, and PKA signaling is indispensable for PV formation and intracellular bacterial replication [20]. Moreover, PKA phosphorylates Bcl-2-associated death promoter (Bad), resulting in sequestration of Bad to the PV and prevention of apoptosis [21]. Although most PKA substrates differentially phosphorylated during C. burnetii infection have not been characterized, they likely play unique roles in PV formation and Q fever pathogenesis.
PKA activity is critical for actin-related processes in eukaryotic cells and actin polymerization is involved in the intracellular lifestyle of multiple bacterial pathogens. A subset of intracellular bacteria recruit actin regulatory proteins to the bacterial cell surface to form actin tails that propel the pathogen through the cytosol and facilitate cell-to-cell movement without exposure to host immune cells. For example, Listeria monocytogenes produces ActA that recruits Arp2/3 and the PKA target protein vasodilator-stimulated phosphoprotein (VASP) to polymerize actin and form actin comets [22,23]. Manipulation of VASP by bacteria that do not use actin-based motility, such as C. burnetii, has not been reported. However, actin reorganization is essential for PV expansion in HeLa cells infected with avirulent C. burnetii [24]. In this study, we identified VASP as a PKA substrate that is preferentially activated during C. burnetii growth in human macrophages. Because PKA signaling is manipulated by C. burnetii [20,21] and VASP is a PKA substrate that regulates actin polymerization, we predicted that VASP is involved in PV formation and/or maintenance. Our findings indicate VASP is phosphorylated by PKA during infection in a T4SS-dependent fashion, and VASP activity is required for PV formation and C. burnetii replication in human macrophages. Additionally, we identified VASP residues (S157 and S239) required for PV formation and heterotypic fusion with other compartments. This study provides the first evidence of an intravacuolar bacterial pathogen usurping VASP function to promote replication vacuole formation.
We previously showed that PKA substrates are differentially phosphorylated during C. burnetii infection of human macrophages [20]. Moreover, PKA activation and phosphorylation of downstream targets occurs throughout intracellular growth (24–96 h post-infection (hpi)). To identify PKA target proteins with increased phosphorylation during infection, we immunoprecipitated PKA substrates from C. burnetii-infected THP-1 macrophage-like cells using a PKA phospho substrate-specific antibody. This antibody specifically recognizes proteins with phosphorylated Ser/Thr residues with arginine at the -3 and -2 positions (RRXS/T). At 24 and 96 hpi, immunoprecipitated proteins were analyzed by Coomassie blue staining. We consistently observed an increase in the levels of a ~47 kDa protein in C. burnetii-infected cells (Fig 1A). Mass spectrometry analysis identified this protein as eukaryotic vasodilator-stimulated phosphoprotein (VASP), a known PKA target [25]. VASP is involved in actin polymerization, contains distinct protein-protein interaction domains, and is regulated by phosphorylation at five residues (Fig 1B). Immunoblot analysis using a VASP-specific antibody confirmed increased VASP levels in proteins immunoprecipitated from infected cells (Fig 1C). PKA commonly phosphorylates VASP at S157 [26]. Therefore, we further assessed whether phosphorylated VASP (S157) was immunoprecipitated with the PKA phospho substrate-specific antibody. As shown in Fig 1C, immunoblot and densitometry analysis revealed an increase in phosphorylated VASP (S157) levels in immunoprecipitated samples collected from infected cells, indicating VASP is targeted by PKA during C. burnetii intracellular growth.
VASP is involved in remodeling actin and is regulated by phosphorylation at Y39, S157, S239, S322, and T278. PKA, which plays a significant role in PV biogenesis [20], is known to phosphorylate VASP at S157, S239, and T278. Actin remodeling is a dynamic process involved in PV formation and/or maintenance [24]. Therefore, we predicted the kinetics of VASP phosphorylation may differ during early and late stages of infection. To test this hypothesis, we infected THP-1 cells with C. burnetii and collected cell lysates at 2–96 hpi. Immunoblot analysis using a specific antibody directed against phosphorylated VASP (S157) revealed low levels of phosphorylated protein in uninfected cells (Fig 2A). Cells infected with C. burnetii for 2 and 6 h also did not show a significant increase in levels of phosphorylated VASP, suggesting VASP activity is not required for early infection events, such as cell adherence, phagocytosis, and nascent phagosome formation. However, VASP phosphorylation levels increased significantly at 24 hpi and were maintained through 96 hpi (Fig 2A). Levels of total VASP remained unaltered in C. burnetii-infected cells, indicating VASP expression and turnover were not substantially altered by the pathogen. Furthermore, VASP can be phosphorylated at S239 by PKA or protein kinase G, and at T278 by adenosine monophosphate-activated protein kinase [25, 26]. Although C. burnetii infection did not trigger increased levels of phosphorylation at T278 or S322 in THP-1 cells, significantly increased levels of VASP phosphorylated at S239 were evident with similar timing to S157 phosphorylation (Fig 2A).
Although VASP phosphorylation increased prior to vacuole expansion (24 hpi), we sought to confirm that phosphorylation was not simply a consequence of the presence of a large vacuole. Cells were infected with C. burnetii deficient in CpeD or CpeE, which are both T4SS-secreted proteins. CpeD and CpeE mutant bacteria entered cells and were maintained in a vacuole much smaller than a typical PV (Fig 2B). However, increased VASP phosphorylation was apparent at 72 hpi with either CpeD- or CpeE-deficient C. burnetii, demonstrating that phosphorylation does not require a pre-formed large vacuole. Relative to uninfected cells, CpeD mutant-infected macrophages showed an ~ 1.9 fold increase in S157 and S239 phosphorylation. CpeE mutant-infected macrophages showed an ~ 8-fold and ~ 2.1-fold increase in S157 and S239 phosphorylation, respectively.
C. burnetii secretes effector molecules into the host cell cytosol using a Dot/Icm T4SS. PKA activation is triggered by secreted bacterial effectors, suggesting T4SS-defective C. burnetii should not trigger VASP phosphorylation. Therefore, we anticipated that inhibiting C. burnetii protein synthesis or antagonizing PKA signaling would prevent increased VASP phosphorylation. To test these predictions, we examined the status of VASP phosphorylation in the presence or absence of chloramphenicol or the PKA inhibitor H89 at 72 hpi, a time when robust VASP phosphorylation occurs. As shown in Fig 3A, treatment with H89 significantly reduced VASP phosphorylation, confirming VASP is a downstream target of PKA during infection. Treatment with the bacterial protein synthesis inhibitor chloramphenicol also abrogated increased VASP phosphorylation, supporting previous observations that induction of PKA signaling requires metabolically active C. burnetii. Similar to cells treated with H89 or chloramphenicol, IcmD mutant C. burnetii, which lacks a functional T4SS and cannot secrete effectors, did not alter VASP phosphorylation, indicating a secreted effector(s) promotes this signaling event (Fig 3A). As expected, treatment with the PKA activator forskolin triggered significant VASP phosphorylation at S157 and S239. The abundance of total VASP was not altered during infection or any inhibitor treatment. The requirement of PKA for VASP phosphorylation was further confirmed by silencing expression of the Cα subunit of PKA using a siRNA approach. This method depleted > 90% of PKACα relative to THP-1 cells transfected with non-targeting siRNA (Fig 3B). As expected, C. burnetii triggered increased VASP S157 and S239 phosphorylation in non-targeting siRNA-transfected cells. However, decreased expression of PKACα abrogated increased VASP phosphorylation without altering expression of total VASP. Relative to non-targeting siRNA-transfected cells, PKACα-silenced, C. burnetii-infected cells showed an ~ 66% and ~ 52% decrease in S157 phosphorylation and S239 phosphorylation, respectively. Together, these results indicate that C. burnetii triggers PKA- and T4SS-dependent phosphorylation of VASP at S157 and S239 in macrophage-like cells.
Based on the results above, we predicted that VASP is required for C. burnetii PV expansion and intracellular growth. To assess the importance of VASP function during infection, THP-1 cells were transfected with VASP-specific siRNA or non-targeting siRNA. Cell lysates were collected from 24–144 h post-transfection and analyzed by immunoblot to confirm VASP knockdown. VASP production decreased >80% using this approach (Fig 4A) and phosphorylated VASP (S157) was barely detectable at 24 and 96 h post-transfection. Importantly, VASP silencing did not alter viability of transfected cells at any time point tested (Fig 4B). siRNA-transfected cells were then infected with mCherry-expressing C. burnetii and bacterial growth was monitored by measuring fluorescence for six days. In VASP-depleted cells, a significant reduction in C. burnetii growth was observed relative to non-targeting siRNA-transfected cells (Fig 4C). We next confirmed that reduction of mCherry fluorescence correlated with genome equivalents representing bacterial numbers. THP-1 cells were infected with wild type C. burnetii, total DNA isolated at 24 or 96 hpi, and genome equivalents determined as previously described [27]. As shown in Fig 4D, no significant difference in bacterial numbers was observed at 24 hpi between control and VASP-depleted cells, indicating VASP is not required for bacterial uptake by THP-1 cells. However, the number of bacterial genomes at 96 hpi was ~ 40% lower in VASP-depleted cells, corresponding to mCherry fluorescence results and confirming VASP is necessary for optimal bacterial growth in macrophages.
Next, we confirmed the growth curve analyses using confocal microscopy to monitor PV formation. The lysosomal marker CD63 was used to label the PV membrane. As shown in Fig 5, large (diameter > 6 μm), CD63-decorated PV were present in ~ 50% of non-targeting siRNA-transfected cells and only 7% of PV were smaller than 2 μm. In contrast, the number of vacuoles larger than 6 μm was reduced to 13% in VASP-silenced cells and 40% of PV were smaller than 2 μm (Fig 5). CD63 presence on small vacuoles in VASP-silenced cells indicated that VASP knockdown did not alter invasion events to prevent C. burnetii entry into the host cell or prevent phagolysosomal maturation. However, VASP silencing prevented vacuole expansion, potentially limiting available space and nutrients for replicating C. burnetii. Additionally, the impact of decreased VASP expression on PV formation was confirmed using a second set of VASP siRNA constructs. Together, these results indicate that VASP activity is required for optimal PV expansion and C. burnetii growth in human macrophages.
Avirulent and virulent C. burnetii produce different LPS structures, with virulent phase I LPS masking cell wall antigens that activate Toll-like receptors, preventing activation of the innate immune response [28]. Avirulent phase II C. burnetii is widely used for in vitro studies to characterize pathogen-host cell interactions [29]. However, results obtained from avirulent studies should be validated using virulent C. burnetii, as human disease is caused by phase I organisms. Similar to avirulent bacteria, THP-1 cells infected with virulent C. burnetii contained increased levels of phosphorylated VASP (S157 and S239) from 24–96 hpi without altering levels of total VASP (Fig 6A, left panel). siRNA-mediated knockdown of VASP expression substantially reduced PV expansion and resulted in smaller vacuoles (Fig 6A, middle and right panels). These results indicate that virulent C. burnetii uses host VASP for optimal PV formation. In addition to obtaining efficient silencing using individual siRNA, we observed similar PV formation results when cells were transfected with a mixture of VASP siRNA constructs. Although THP-1 macrophage-like cells are commonly used as an in vitro cellular model [17,30,31], C. burnetii preferentially infects and replicates within human alveolar macrophages (hAMs) during natural infection [32]. Additionally, alveolar macrophages express substantial levels of VASP [33]. Therefore, we isolated cells from human lungs post-mortem and infected primary hAMs with virulent C. burnetii. As shown in Fig 6B (left panel), infection triggered increased VASP phosphorylation (S157 and S239) from 24–96 hpi similar to THP-1 cells. Knockdown of VASP expression by siRNA interfered with PV expansion and cells contained smaller vacuoles relative to non-targeting siRNA-transfected cells (Fig 6B, middle and right panels). These results demonstrate the importance of VASP activity in a natural C. burnetii disease scenario.
Phosphorylation of VASP at S157 and S239 regulates protein localization and actin polymerization, respectively [26]. We specifically tested whether motifs containing S157 or S239 are required for PV formation by expressing GFP-VASP constructs with individual residues mutated (S157E and S239E) in C. burnetii-infected THP-1 cells. A serine to glutamic acid mutation mimics the conformation of a phosphorylated serine with respect to negative charge, maintaining the structure of a phosphorylated protein [34]. However, the ability to bind target proteins that directly recognize phospho-serine motifs is prevented by this mutation. To examine functional effects of individual VASP mutants, we assessed VASP localization, actin arrangement, and PV formation in infected cells. As shown in Fig 7, large typical PV were present at 72 hpi in GFP-VASP and GFP-VASP (S157E)-expressing cells. Additionally, wild type and VASP (S157E) co-localized with actin around the PV. In contrast, expression of GFP-VASP (S239E) resulted in substantially smaller PV containing fewer C. burnetii. Furthermore, actin levels were reduced around the PV in GFP-VASP (S239E)-expressing cells. These results indicate proper S239 phosphorylation and regulation of distinct downstream target proteins is critical for infection-specific VASP functions potentially through interactions with a host or bacterial protein(s). Reduced actin polymerization may also destabilize the PV, resulting in smaller vacuoles.
PV generation involves numerous heterotypic fusion events with phagosomes, autophagosomes, and lysosomes [29]. To determine if VASP function is required for phagosome trafficking to the PV, THP-1 cells were infected with C. burnetii for 72 h, then incubated with fluorescent beads, a common technique for assessing PV heterotypic fusion with cellular compartments [35]. As shown in Fig 8, beads were delivered to the PV in cells expressing GFP-VASP or GFP-VASP (S157E). In contrast, beads were sequestered away from PV in cells expressing GFP-VASP (S239E), corresponding to the PV formation results observed above. These results indicate proper VASP S239 activity is critical for heterotypic fusion events during C. burnetii infection.
To further assess the requirement of VASP phosphorylation for PV development, cells were transfected with constructs encoding phosphorylation-defective VASP mutants (S157A and S239A). In contrast to the phospho-mimetic mutants above, these proteins serve as dominant negative mutants and are not phosphorylated. As shown in Fig 9, cells expressing GFP-VASP or GFP-VASP (S239A) contained expanded PV while cells expressing GFP-VASP (S157A) harbored multiple small atypical PV. These results indicate that S157 phosphorylation is critical for PV expansion.
Here, we demonstrate, for the first time, that eukaryotic VASP can be exploited by an intracellular bacterial pathogen to promote replication vacuole formation. Successful PV formation by C. burnetii is necessary for development of Q fever and the pathogen actively manipulates host signaling using effector proteins secreted into the host cell cytosol via a Dot/Icm T4SS. Previous studies in our laboratory demonstrated that PKA is hijacked by C. burnetii to facilitate PV formation and prevent apoptotic cell death [20,21]. In this study, we identified VASP as a PKA substrate that is activated during infection by T4SS-proficient C. burnetii. VASP is an essential protein for actin remodeling; therefore, we hypothesize that VASP activity is required for actin-dependent PV expansion and/or maintenance in human macrophages.
As a member of the Ena/VASP protein family, VASP has an N-terminal EVH1 domain, C-terminal EVH2 domain, and proline-rich region (Fig 1B). The EVH1 domain binds to focal adhesion proteins that anchor VASP to the integrin complex of cell membranes [36,37]. The proline-rich region binds to SH3 and WW domain-containing proteins and profilin, which catalyze actin monomer formation and fast actin polymerization [38]. The EVH2 domain mediates F- and G-actin binding to regulate actin polymerization [39,40]. Anti-capping is one mechanism proposed for VASP-dependent actin polymerization where VASP binds to actin filament barbed ends and prevents recruitment of capping proteins, resulting in long actin filaments [41]. Additionally, VASP can inhibit Arp2/3-dependent actin filament branching; however, a molecular mechanism has not been characterized [23]. VASP dysfunction has been linked to many diseases including cancer, atherosclerosis and thrombosis [42].
VASP contains three major phosphorylation sites, S157, S239, and T278 that control numerous cellular functions. For example, phosphorylation by PKA at S157 facilitates localization to the cell periphery into focal adhesions [43], whereas phosphorylation at S239 and T278 regulates F-actin assembly [26]. During C. burnetii infection, prolonged VASP phosphorylation (24–96 hpi) occurs at S157 and S239. S157 is located adjacent to a proline-rich region and controls VASP translocation to the cell membrane, while exerting a minimal effect on F-actin polymerization [25,26,43–45]. S157 phosphorylation also blocks VASP interaction with Abl, SH3 domains, and Src proteins [26,46]. Phosphorylation of S239, which is located adjacent to a G-actin binding motif in the EVH2 domain, negatively regulates VASP anti-capping activity, resulting in shortened F-actin filaments, reduced F-actin accumulation, and filopodia formation [26,45,47,48]. Phospho-mimetic mutation of S239 reportedly decreases F-actin accumulation, does not efficiently co-localize with actin, and antagonizes adhesion and spreading of smooth muscle cells [49]. The timing of increased VASP phosphorylation during C. burnetii infection correlates with PV biogenesis and expansion, events that are regulated by T4SS effectors. Moreover, C. burnetii actively replicates during this time, and bacterial protein synthesis and T4SS activity are required to trigger increased VASP phosphorylation levels. Therefore, we predict a distinct secreted effector(s) hijacks PKA signaling to promote VASP phosphorylation.
PKA and protein kinase C (PKC) can phosphorylate VASP at S157 in human platelets [50] and protein kinase G preferentially phosphorylates the protein at S239 [45,50]. However, PKA phosphorylates S157 and S239 during C. burnetii infection as demonstrated by specific inhibitor and siRNA treatments. Four different PKA catalytic subunits, Cα, Cβ, Cγ, and X chromosome encoded protein kinase X (PRKX), have been identified in humans [51]. Using a confirmatory siRNA knockdown approach, it is clear that the PKACα subunit is required for S157 and S239 phosphorylation during infection. PKACα promotes breast cancer cell viability by inactivating the pro-apoptotic BCl-2-associated death promoter protein [52]. We previously showed that PKA promotes macrophage survival during C. burnetii infection, although the specific subunit responsible has not been defined. Considering an important role in cell survival, we anticipate PKACα is critical for PKA pro-survival signaling and VASP phosphorylation during infection.
A subset of intracellular pathogens, such as Listeria monocytogenes, recruit VASP to facilitate actin tail formation [53]. Resulting actin tails propel L. monocytogenes through the cytosol and to the cell surface, facilitating spread to other cells. Thus, hijacking VASP for actin-based motility is a critical part of the L. monocytogenes pathogenic life cycle. In contrast, some intracellular bacteria, such as Shigella spp., do not require VASP for actin-based mobility [54]. To our knowledge, the current study provides the first evidence of VASP manipulation by a non-motile, intravacuolar bacterium to control replication vacuole formation and promote intracellular replication. Indeed, siRNA-mediated VASP knockdown prevents typical C. burnetii growth and PV formation. Additionally, VASP activity is required during virulent C. burnetii infection of primary hAMs, suggesting the protein is necessary for optimal pathogen growth in the human lung. We predict that VASP-dependent actin rearrangements around the PV are required for vacuole stability and expansion. This prediction is supported by co-localization of GFP-VASP and actin at the PV membrane. Our macrophage infection results are consistent with reports suggesting actin is required for avirulent C. burnetii PV formation in HeLa cells [24]. Using F-actin depolymerizing chemical agents, Aguilera et al. showed that F-actin polymerization facilitates fusion between the PV and bead-containing phagosomes. Autophagosomes, lysosomes, and phagosomes continuously fuse with the PV, providing nutrients and membrane, and VASP may facilitate these actin-dependent vesicular fusion events. Additionally, actin may provide structural support for the maturing PV, allowing controlled expansion within the cytosol. A role for actin in structural support has been reported for Chlamydia trachomatis replication vacuole formation [55]. During growth in epithelial cells, actin cages form around the C. trachomatis inclusion, and disruption of actin polymerization diminishes vacuole membrane integrity. Therefore, it is possible that VASP is involved in growth of C. trachomatis and other intravacuolar pathogens by regulating formation of a nucleating complex at the replication vacuole membrane.
Phosphorylated VASP motifs recruit adaptor proteins that assemble protein complexes. For example, members of the 14-3-3 family, WD40 domain-containing F-box proteins, and WW domain-containing proteins form multimolecular signaling complexes through specific interactions with phospho-Ser/Thr motifs [56]. It is not known if similar proteins bind to VASP motifs to regulate macrophage responses to C. burnetii. However, ectopic expression of VASP (S239E) negatively impacts PV formation and reduces actin accumulation around the PV. In contrast, expression of VASP (S157E) does not alter PV formation or actin accumulation at the vacuole, consistent with reports that S157 does not significantly impact F-actin polymerization. During infection, VASP phosphorylated at S239 may localize to specific regions around the PV where controlled actin depolymerization could occur to provide space for vacuole expansion. Although VASP (S239E) has a negative charge, is functionally active, and negatively impacts actin polymerization, the mutant protein may not assemble phospho-Ser/Thr motif-binding protein complexes needed for PV formation. Moreover, VASP-mediated arrangement of actin ultrastructure around the PV may facilitate fusion of incoming vesicles and phagosomes with the PV. Indeed, bead trafficking results indicate S239 is critical for phagosome movement to the PV. These results do not prove that VASP plays a direct role in fusion events, but rather allows incoming phagosomes access to the PV along the cytoskeletal network.
Expression of VASP (S157A) severely impairs PV expansion, while VASP (S239A) expression allows normal vacuole development, demonstrating S157 phosphorylation is critical for infection. VASP phosphorylation at S157 prevents interactions with the SH3 domains of Abl, alpha II spectrin, and Src [46,57], and also regulates VASP cellular distribution. The dominant negative S157A mutant, which is not phosphorylated, may facilitate interaction with SH3 domain-containing proteins, preventing typical PV formation. The finding that expression of the phosphomimetic S157E mutant, which mimics phosphorylated VASP and provides conformational similarity, does not adversely impact PV formation supports the importance of S157 phosphorylation for optimal PV formation. Additionally, S157 phosphorylation minimally impacts actin filament formation [26]. Therefore, S157 phosphorylation may primarily act via regulation of other targets that contribute to optimal PV formation, and not by direct actin rearrangement. In contrast, expression of the dominant negative S239A mutant, has no significant impact on PV formation. Under physiological conditions, S239 is phosphorylated in small quantities and the level of phosphorylation is tightly regulated [26,47]. Therefore, PV-localized S239-phosphorylated VASP may reduce local F-actin levels, providing the necessary space for incoming cargo-laden vesicles and facilitating fusion events. It is also possible that infected macrophages regulate actin rearrangement via redundant mechanisms in the absence of S239-phosphorylated VASP.
In conclusion, C. burnetii hijacks host PKA signaling to phosphorylate VASP and facilitate PV formation. VASP phosphorylation requires bacterial protein synthesis and an active T4SS, indicating the pathogen actively targets this pathway. VASP may directly regulate actin polymerization dynamics, providing structural stability and physical space for the expanding PV needed for bacterial replication. Supporting this prediction, depletion of VASP negatively impacts intracellular bacterial growth and PV size. Phosphorylation at S157 and S239 is critical for VASP promotion of PV formation. Overall, this study provides the first evidence of a non-motile intravacuolar bacterial pathogen manipulating eukaryotic VASP to facilitate intracellular growth.
Avirulent C. burnetii Nine Mile phase II (NMII; RSA439, clone 4), virulent Nine Mile phase I (NMI; RSA493), ΔCpeD, and ΔCpeE bacteria were cultured in acidified citrate cysteine medium (ACCM) at 37°C with 5% CO2 and 2.5% O2 for 7 days. Cultures were then washed and resuspended in sucrose phosphate buffer (pH 7.4). mCherry-expressing C. burnetii was grown in ACCM supplemented with chloramphenicol (3 μg/ml; Sigma). IcmD mutant C. burnetii (icmD::Tn) [14] was grown in ACCM supplemented with kanamycin (350 μg/ml). Construction of CpeD- and CpeE-deficient strains is described in S1 Text. All work with virulent C. burnetii was performed in the Centers for Disease Control and Prevention-approved biosafety level-3 facility at the University of Arkansas for Medical Sciences.
Human THP-1 monocytes (TIB-202; ATCC) were cultured in RPMI1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS; Invitrogen). Before infection, THP-1 cells were treated with phorbol 12-myristate 13-acetate (PMA; 200 nM) overnight for differentiation into macrophage-like cells. After replacing PMA-containing medium with fresh medium, THP-1 cells were infected with C. burnetii at a multiplicity of infection (MOI) of 10. To facilitate infection, plates were centrifuged at 900 rpm for 5 minutes, then incubated for 2 h at 37°C. Cells were washed with medium to remove excess C. burnetii and supplemented with fresh medium for the duration of the infection. Primary human alveolar macrophages (hAMs) were collected from lung tissue obtained post-mortem from the National Disease Research Interchange and maintained in DMEM/F12 containing 10% FBS as previously described [36]. Use of primary hAMs was assessed by the University of Arkansas for Medical Sciences institutional review board and deemed to not be human subjects research (#87788).
THP-1 cells were infected with C. burnetii and harvested at 72 hpi in non-denaturing buffer (20 mM Tris-HCl pH 7.4, 0.1% Triton X-100, 150 mM NaCl, 2 mM NaF, 0.1% glycerol, and a protease and phosphatase inhibitor cocktail). Uninfected cells were processed as controls. Total protein was quantified using a DC protein assay (BioRad) and confirmed by immunoblot to detect β-tubulin. Immunoprecipitations (IPs) were performed using a Classical IP kit (Pierce). Lysates were pre-cleared by incubation with agarose resin for 1 h at 4°C. For immune complex preparation, 1 mg of total protein was incubated overnight with 10 μg of anti-PKA phospho substrate antibody at 4°C. Immune complexes were captured by incubation with protein A/G agarose beads for 1h at 4°C, then eluted in elution buffer (0.12 M Tris-HCl, 2% SDS, 20% glycerol, pH 6.8). Eluted PKA phospho substrate antibody was assessed by immunoblot and probing with anti-IgG to confirm equal amounts of capture antibody in control and infected samples.
THP-1 cells in 6-well plates were harvested in lysis buffer (50 mM Tris, 1% sodium dodecyl sulfate (SDS), 5 mM EDTA, 5 mM EGTA, 1 mM sodium vanadate, protease and phosphatase inhibitor cocktails). Lysates were passed through a 26 gauge needle 10 times and quantified using the DC protein assay. Equal amounts of total protein were separated using 4–15% Mini-PROTEAN TGX gels (BioRad). Proteins were then transferred onto a polyvinylidene fluoride membrane (BioRad), and membranes blocked with 5% milk in Tris-buffered saline containing 0.1% Tween 20 (TBS-T). Membranes were probed with mouse anti-human VASP (BD Biosciences), rabbit anti-human p-VASP S157 (Cell Signaling), rabbit anti-human p-VASP S239 (Cell Signaling), rabbit anti-human p-VASP T278 (Sigma), rabbit anti-human p-VASP S322, or mouse anti-human β-tubulin (Sigma) primary antibodies diluted in TBS-T with 5% BSA. Anti-mouse or anti-rabbit IgG secondary antibodies conjugated to horseradish peroxidase (Cell Signaling) were used for chemiluminescence-based detection. Secondary antibodies were diluted in 5% milk-containing TBS-T and membranes incubated 1 h at room temperature. Reacting proteins were visualized using a WesternBright ECL kit (Advansta) and exposure to film.
Previously optimized protocols were used for chemical inhibitor treatments [20]. THP-1 cells in 6-well plates were infected with C. burnetii as described above. After 2 h, infectious inoculum was replaced with fresh medium and cells were treated with specific inhibitors or inducers. Media was replaced daily and chemicals were present throughout. Whole cell lysates were collected at the indicated times post-infection and processed for immunoblot analysis. H89 (10 μM; Sigma) was used to inhibit PKA activity and forskolin (10 μM; Sigma) was used to trigger PKA signaling. To inhibit intracellular C. burnetii protein synthesis, infected cells were treated with chloramphenicol (10 μg/ml).
Validated human VASP siRNA (5′- GGACCUACAGAGGGUGAAAdTdT -3') was used for VASP silencing [58], and non-targeting siRNA (5′-UGGUUUACAUGUCGACUAA-3') was used as a control for transfection experiments. THP-1 cells were nucleofected with VASP or non-targeting siRNA as previously described [59] with some modifications. THP-1 monocytes (3 X 106) were resuspended in Human Monocyte Nucleofector Solution (Lonza). siRNA (1 μg) was mixed with cells, and cells were transfected using a Nucleofector 2b and program Y001 (Lonza). Following nucleofection, cells were transferred into fresh culture medium, then incubated 4 h at 37°C for recovery. Cells were treated overnight with PMA for differentiation into macrophage-like cells. After removing PMA-containing medium and replacing with fresh medium, monolayers were infected with C. burnetii as described above.
For siRNA-mediated knockdown of VASP expression in hAMs, DharmaFECT 1 transfection reagent (Thermo Scientific) was used according to the manufacturer’s instructions. hAMs were cultured on coverslips in 24 well plates. Transfection complex was formed using VASP siRNA (50 nM) and DharmaFECT 1 reagent, then added to cells drop wise and incubated overnight for siRNA uptake. Cells were washed once, supplemented with fresh medium, and infected with C. burnetii as indicated above.
For ectopic expression, full length human VASP, S157A, S239A, or S157E mutant cDNA cloned into pEGFP-N1, which were previously characterized for expression of GFP-VASP and GFP-VASP (S157E) [60], were used. Expression of GFP-VASP and GFP-VASP (S239E) was achieved using full length VASP or S239E mutant cDNA cloned into pCMV6-AC-GFP (OriGene). THP-1 cells were nucleofected with 1 μg of each plasmid as described above.
Viability was determined using a Cell Counting Kit-8 (Dojindo Laboratories) according to the manufacturer’s instructions. siRNA-transfected cells were left uninfected or infected for the indicated times in 96-well plates. At each time point, WST-8 (2-(2-methoxy-4-nitrophenyl)-3-(4-nitrophenyl)-5-(2,4-disulfophenyl)-2H-tetrazolium, monosodium salt) reagent was added to wells and incubation continued for 4 h at 37°C. Following measurement of the A450 of cultures, viability was calculated using the following formula: {(Atest-Abackground)/(Acontrol-Abackground) x 100}, where A = absorbance at A450, background = media alone, and control = non-transfected cells.
THP-1 cells were plated on 12 mm diameter circular cover glasses (Fisher) in 24-well plates. After treatments and infections, cells were fixed and permeabilized with 100% ice-cold methanol for 3 min and blocked with 0.5% bovine serum albumin (BSA) in PBS (pH 7.4) for 1 h at room temperature. For actin detection, cells were fixed with 4% formaldehyde for 15 min and blocked with 0.5% BSA containing 0.3% Triton X-100 for 1 h. Cells were incubated with mouse anti-CD63 (BD Biosciences) and rabbit anti-C. burnetii primary antibodies in 0.5% BSA for 1 h at room temperature. After washing with PBS, cells were incubated with Alexa Fluor 488-conjugated anti-mouse IgG and Alexa Fluor 594-conjugated anti-rabbit secondary antibodies (Invitrogen). Indicated samples were treated with Alexa Fluor 594-labeled phalloidin (Invitrogen) for 30 min at room temperature to detect actin. In phalloidin-treated samples, CD63 primary antibody was detected with Alexa Fluor 633-conjugated anti-rabbit secondary antibody. Cells were incubated with DAPI for 5 min at room temperature and mounted with MOWIOL (Sigma). Confocal imaging was performed with a Nikon C2si microscope and data were analyzed using NIS-Elements software (Nikon). PV diameter measurements were taken from at least 15 randomly selected fields. When multiple vacuoles were visible, the two largest vacuoles were measured in a single cell and average vacuole size calculated for all fields.
THP-1 cells were transfected with constructs encoding GFP-VASP or serine mutants (S157E or S239E), then infected with C. burnetii as described above. 0.5 million fluorescent 1.0 μm polystyrene microsphere beads (Invitrogen) were added to cells on coverslips and incubated overnight at 37°C. Cells were washed with media to remove excess beads. At 72 hpi, cells were fixed with 4% paraformaldehyde and processed for confocal microscopy as described above.
For fluorescence-based growth curves, THP-1 cells were transfected with non-targeting or VASP siRNA as described above. Cells were then cultured in glass flat bottom 96 well black plates. After PMA treatment and removal, cells were infected with mCherry-expressing C. burnetii overnight (MOI = 10). Medium was then replaced with fresh medium and intracellular C. burnetii growth monitored by fluorescence intensity using a BioTek Synergy H1 microplate reader (585 nm excitation/620 nm emission) for 6 days post-infection.
For determination of genome equivalents, infected THP-1 cells were harvested in media by centrifugation (10,000 x g). Cells were disrupted by vortexing with microbeads and total DNA was extracted using an UltraClean Microbial DNA kit (MoBio Laboratories). 10 ng of total DNA was used for quantitative PCR. Previously optimized primers [27] designed to amplify C. burnetii dotA were used with a Power SYBR Green PCR master mix (Applied Biosystems). pCR2.1-dotA was serially diluted and used to generate a standard curve and calculate C. burnetii genome copies.
Immunoblots were scanned in gray scale with a resolution of 300 dpi and band intensities quantified using ImageJ software (version 1.48v). Band intensities were normalized to β-tubulin levels. All experiments were performed at least in triplicate. Statistical significance between experimental and control groups was calculated using a Students t test and Prism 6 software (GraphPad). Results were considered statistically significant when p < 0.05.
|
10.1371/journal.pbio.0050228 | The Scent of the Waggle Dance | The waggle dance of honey bee (Apis mellifera L.) foragers communicates to nest mates the location of a profitable food source. We used solid-phase microextraction and gas chromatography coupled with mass spectrometry to show that waggle-dancing bees produce and release two alkanes, tricosane and pentacosane, and two alkenes, Z-(9)-tricosene and Z-(9)-pentacosene, onto their abdomens and into the air. Nondancing foragers returning from the same food source produce these substances in only minute quantities. Injection of the scent significantly affects worker behavior by increasing the number of bees that exit the hive. The results of this study suggest that these compounds are semiochemicals involved in worker recruitment. By showing that honey bee waggle dancers produce and release behaviorally active chemicals, this study reveals a new dimension in the organization of honey bee foraging.
| A honey bee colony consists of many thousands of individuals, all of which help to perform the work that allows their colony to thrive. To coordinate their efforts, honey bees have evolved a complex communication system, no part of which is more sophisticated than the waggle dance. The waggle dance is unique, because it exhibits several properties of true language, through which a forager communicates the location and profitability of a food source to other bees in the darkness of the hive. The information coded in the dance has been extensively researched, but we still do not understand how information is actually transferred from the dancing bee to the receivers of the message. Because information is often transferred by scent in honey bee colonies, we investigated whether waggle dancers produce a scent that distinguishes them from foragers that do not dance. We found that dancers produce four hydrocarbons that distinguish them from nondancing foragers, and that, when blown into the hive, increase foraging activity. These results show that waggle-dancing bees produce a unique scent that affects the behavior of their fellow foragers. We discuss likely meanings of this olfactory message and its potential role in waggle-dance communication.
| More than fifty years ago, Karl von Frisch demonstrated through a series of elegant experiments that the waggle dance of honey bees uses symbolic communication to convey information about a subject that is both spatially and temporally removed from the receiver of the signal [1]. The waggle dance is therefore a unique animal signal that exhibits several of the important properties of true language, which are generally attributed only to “advanced” organisms such as marine mammals, nonhuman primates, and humans. Honey bees are themselves quite advanced, however; a honey bee society consists of a large group of individuals of overlapping generations, living permanently together with cooperative care of young and a division of labor. To coordinate the complex interactions among the members of such a society, a sophisticated system of communication is necessary.
The role of the waggle dance in this sophisticated system of communication is primarily to direct the colony's foraging effort toward nectar- and pollen-producing flowers. Successful foragers perform the dance within the nest to recruit other bees to a profitable food source. By closely following a dancer, potential recruits acquire information about the location and richness of the advertised food source [1,2]. However, despite our considerable knowledge of the information contained in the dance, we still do not understand how dancers attract and convey information to recruits in the darkness of the hive [2]. Airborne sounds [3,4], substrate vibrations [5,6], and tactile cues [7] seem to play some role in attracting recruits to waggle dancers or conveying dance information, but each of these modalities appears to be neither necessary nor sufficient for recruitment [2].
Another modality that may be involved in waggle dance communication is olfaction. Evidence to date that olfaction plays a role in honey bee waggle-dance communication is limited to odors acquired from the environment at or en route to the floral food source [2,8,9]. (These odors are thought to serve as cues that allow foraging recruits to pinpoint the food source advertised by the waggle dancer.) However, the production of olfactory signals by the waggle dancers themselves could attract recruits and convey dance information. Pheromones and other semiochemicals are used frequently by honey bees and other social insects to coordinate the activities of colony members [10,11] and hence may be expected to help organize the vital task of foraging. Furthermore, evidence suggests that closely related Hymenopterans, such as bumblebees, use pheromones within the nest to organize foraging activity [12]. Whereas bumblebees do not communicate via waggle dances, olfactory-based foraging communication may be an ancestral trait and thus present in honey bees. A preliminary study suggested indeed that the scent of active honey bee foragers could encourage other bees to forage [13].
The goal of this study was to investigate whether waggle dancers produce and release into the air chemical compounds that distinguish them from other foragers. We addressed this first goal by using solid phase microextraction (SPME) and gas chromatography coupled with mass spectrometry (GC/MS). If these distinguishing compounds are semiochemicals or pheromones, then adding these compounds into the hive will affect the behavior of foraging bees. To test whether the compounds are behaviorally active, we measured foraging activity before and after we injected the volatilized compounds into the hive. Our results show that honey bee waggle dancers produce four characteristic volatile compounds that increase foraging activity.
We discovered four conspicuous compounds in the airspace surrounding dancing bees that were not present in the airspace surrounding nondancing bees (Figure1). We obtained similar results for the airspace surrounding waggle dancers on an artificial swarm as compared with the air over an area of the swarm with no waggle dancers. From extracts made of three intact waggle dancers immersed for 1 min into 250 μl of hexane, we identified the four substances that generated peaks 1, 2, 3, and 4 as marked in Figure1 to be Z-(9)-tricosene, tricosane, Z-(9)-pentacosene, and pentacosane, respectively.
All four compounds were present in significantly higher amounts on the abdomens of waggle dancers than in either (a) nondancing foragers that returned from the same unscented nectar source, or (b) nonforaging bees [Figure 2, one-way analysis of variance (ANOVA) for each compound, degrees of freedom (df) = 2, 49, p < 0.001 for all compounds; Tukey HSD for unequal n, p < 0.0005 for all compounds, experiment-wise α = 0.05]. Intriguingly, although only marginally significant, waggle dancers that danced more vigorously (i.e., appeared to perform waggle runs at higher rates and with more exaggerated movement of the abdomen as assessed subjectively by one person, n = 5 bees) tended to emit more of all four compounds than less vigorous dancers did (n = 13) (Figure 3, Mann-Whitney U test, p = 0.05, 0.07, 0.13, and 0.08 for peaks 1, 2, 3, and 4, respectively). Nondancing foragers and nonforaging bees did not emit different amounts of these compounds (Tukey HSD for unequal n, p = 0.076, 0.074, and 0.400, for peaks 2, 3, and 4, respectively) with the exception of Z-(9)-tricosene (p < 0.001).
To test whether the waggle-dance–specific substances affect behavior, we injected onto the dance floor a gaseous mixture of the three commercially available compounds dissolved in hexane [hereafter called TTP (Z-(9)-tricosene, tricosane, pentacosane) solution]. TTP trials lasted 32 min, during which we recorded the number of bees exiting the hive each minute, and during which we made two injections: Injection 1, which consisted of 50 μl of pure hexane, was made during minute 1, and Injection 2, which consisted of 50 μl of TTP solution, was made during minute 16. To control for solvent and treatment, we performed “Hexane trials”, in which Injection 2, like Injection 1, consisted of 50 μl of pure hexane. Experiments were done with two colonies, C1 and C2, using one colony at a time. We performed a total of 20 trials (10 TTP trials and 10 Hexane trials) with C1, and a total of 28 trials (15 TTP trials and 13 Hexane trials) with C2. Only one trial was performed per day, and all trials were performed at the same time of day. Because the compounds originate from waggle dancers under natural conditions, we ascertained that at least one dancer was present during each trial.
Injection of TTP increased the number of bees exiting the hive (Figure 4). The normalized mean number of bees exiting the hive during minutes 25–32 differed significantly between TTP trials (i.e., Injection 2 = TTP solution) and Hexane trials (i.e., Injection 2 = hexane) (two-sample T-test on normally distributed sample groups with equal variances; Colony 1: T = −3.29, p = 0.004, df = 18; Colony 2: T = −2.25, p = 0.033, df = 26). This difference between TTP trials and Hexane trials was not observed following Injection 1, which consisted of hexane for both types of trial (Colony 1: T = −0.61, p = 0.551, df = 18; Colony 2: T = −0.47, p = 0.643, df = 26).
We also observed within TTP trials an increase in the number of bees exiting the hive following Injection 2, as compared with the number exiting following Injection 1, but this was significant for only one colony (two-sample T-test; Colony 1: T = −1.63, p = 0.121, df = 18; Colony 2: T = −2.66, p = 0.013, df = 28). This increase could possibly be the result of circadian foraging patterns, but this would not explain the lack of such an effect during Hexane trials (two-sample T-test; Colony 1: T = 0.83, p = 0.415, df = 18; Colony 2: T = −0.68, p = 0.502, df = 24), which were conducted at the same time of day as TTP trials.
We have identified in this study four compounds that are unique to waggle-dancing bees and that are behaviorally active. This waggle-dance scent originates from the waggle dancers themselves; it is not acquired from the environment while foraging, nor is it a byproduct of a bee's age or task, because it is emitted in only minute quantities by nondancing foragers returning from the same food source.
The prominent compounds of the waggle-dance scent are the cuticular hydrocarbons Z-(9)-tricosene, tricosane, Z-(9)-pentacosene, and pentacosane. Hydrocarbons such as these are produced subcutaneously and are not stored within a gland [14]. The waggle-dance scent is therefore best classified as a semiochemical or nonglandular pheromone. Waggle dancers could raise the levels of the compounds through increased synthesis and/or increased release onto the epicuticle. The compounds could then be released passively into the air, perhaps assisted by the relatively high body temperature of the waggle dancers.
The chemical nature and source of the compounds of the waggle-dance scent differ from those of the bumblebee foraging recruitment pheromone [15]. The main compounds of the bumblebee recruitment pheromone were identified as eucalyptiol, ocimene, and farnesol, which are terpene derivatives. These compounds are produced in the bees' tergal glands. Their different chemistry and source suggest different evolutionary origins for the bumblebee foraging recruitment pheromone and the waggle-dance scent of honey bees. Hence, it is unlikely that the waggle-dance scent of honey bees has evolved from the bumblebee foraging recruitment pheromone.
The compounds of the waggle-dance scent have been identified in earlier studies on honey bees. Z-(9)-tricosene, tricosane, Z-(9)-pentacosene, and pentacosane have been previously identified in hexane washes of the cuticles of foraging-age worker bees [16] and have been shown to be perceptible to honey bee workers [17]. More recent work has shown that, among many others, the four compounds are present in the air surrounding foragers at a feeder station [18]. However, with the exception of demonstration of a possible kin-recognition function of Z-(9)-tricosene [19], past studies did not link specific compounds with specific behaviors of honey bees. In insects other than honey bees, however, the individual compounds of the waggle-dance scent have been linked to specific behaviors. For example, tricosane, pentacosane, and Z-(9)-pentacosene are compounds of a pheromone that foragers of the social wasp Vespula vulgaris use to lay and follow terrestrial trails [20]. Compounds of the waggle-dance scent are also well-known sex attractants in other insects, such as flies [14].
The waggle-dance scent may have a similar marking or attracting function, which the context and results of our experiments link to recruitment behavior. Honey bees produce the scent when they perform waggle dances, both in the hive and on swarms, and in both contexts, workers are recruited to the advertised site. In the more common context of foraging, a general measure of recruitment is the number of foragers that leave the hive. Our results show that injection of the waggle-dance scent onto the dance floor increased the number of bees that left the hive. These bees can be assumed to be foragers, because only foragers leave the hive without noticeably hesitating, performing orientation flights, or gathering at the hive entrance.
Our experiments show that the waggle-dance scent increases the number of foragers leaving the hive, but the exact mechanism underlying the effect is still unclear. Given the function of the waggle-dance scent in other hymenoptera, we propose that the waggle-dance scent, which in our experiments was fanned onto the dance floor, attracts potential recruits to the dance floor, thereby increasing the probability of encounters between potential recruits and dancers, and finally the number of recruits. Under natural conditions, the scent would originate from the dancers themselves, thus the odor plume would mark not only the dance floor, but the individual dancers. This should enable recruits to locate the dancers themselves, which could enhance recruitment efficiency. Recruits could even seek out more vigorous dancers, who typically advertise especially profitable food sources [21], and who seem to emit higher concentrations of the scent. Hence, the waggle-dance scent may mark and attract recruits to successful foragers, and thus help to rapidly focus the colony's foraging effort on the most profitable food sources. Besides marking successful dancers, it is feasible that the waggle-dance scent facilitates the transfer of information from the dancer to recruits. The spatio-temporal pattern of a dancer's odor plume could, for example, indicate the length of a waggle run, and thus provide information to a recruit, even if she lost contact with the dancer herself. This hypothesis is supported by the observation that a mechanical model of a waggle dancer recruits bees to a food source only after the model has touched a waggle dancer (H. E. Esch, personal observation).
Whereas the waggle-dance signal is likely a signal intended for new recruits, two other groups of bees, namely foragers already devoted to a food source and in-hive receiver bees, could glean cues from the waggle-dance scent. Foragers that are already devoted to a food source do not readily follow new dances if their source becomes unavailable, but rather wait for it to replenish [8]. Because the waggle-dance scent does not seem to identify specific food sources, it can provide only limited information to these foragers. However, high concentrations of the scent could alert them to generally good foraging conditions. This could be useful at the beginning of daily foraging or if foraging can be resumed after a spell of bad weather. This mechanism may have been responsible for the effect observed in an earlier preliminary study, in which the number of foragers visiting an empty feeder increased following exposure to the air from a foraging colony [13]. In our experiments, however, the increase in foragers was not likely caused by already-devoted foragers for three reasons. First, experiments were done well after the time that colonies started daily foraging. Second, external conditions such as the weather were remarkably stable, which made strong fluctuations in the numbers of already-devoted foragers unlikely. Third, we did not record a conspicuous drop in the number of bees that left the hive after the initial increase (Figure 4); if the increase would have been due mostly to already-devoted foragers, we would have expected a quick drop to original levels once these foragers found that there was no change in food-source profitability. However, it is possible that such a drop could be hidden by the more substantial numbers of newly recruited foragers. Receiver bees are the second group of bees that could glean cues from the waggle-dance scent. Receiver bees unload nectar from newly returned foragers, which relieves foragers of the time-consuming search for empty storage cells. A high concentration of the waggle-dance scent would indicate a high demand for receiver bees, and could help attract receiver bees to the dance floor. The tremble dance, which is performed by successful foragers that perceive a shortage in receiver bees [22], may help to additionally spread the scent to potential receiver bees.
Bees were kept indoors in four-frame observation hives at the Carl Hayden Bee Research Center in Tucson, Arizona, United States. To test whether waggle dancers produce specific chemical compounds, we performed experiments with two colonies, using first one, then the other colony. Foragers were trained to an artificial feeder [1] ∼100 m from the hive that offered nonscented sugar water. Foragers visiting the feeder were marked on the thorax with powdered paint and thus could be recognized in the hive.
We used a SPME fiber (65 μm polydimethylsiloxane/divinylbenzene; Supelco; http://www.sigmaaldrich.com/Brands/Supelco_Home.html) to sample chemicals. After sampling, the SPME fiber was injected into a GC, CP-3800 (Varian; http://www.varian.com) coupled to a MS (Saturn 2200 Ion Trap,Varian).
We compared the chemical profile of the air over dancing bees to that over nondancing bees, and we also compared the chemical profile of waggle dancers' abdomens to those of both nondancing foragers and nonforaging bees. For air samples, we exposed the fiber for 5 consecutive min approximately 2 cm above the surface of the comb. For abdomen profiles, we briefly touched the fiber to the tip of the abdomen of an individual bee [23–26]. Waggle dancers were sampled during a waggle run, shortly after they began dancing. Sampled bees were observed from the moment they entered the hive until the sample was taken. It is possible that we classified as nondancing foragers bees that danced before observation began (e.g., in the entrance tunnel to the hive) or that danced after SPME sampling, but this would bias our results in only a conservative direction.
After SPME sampling, the fiber was desorbed for 3 min in the GC/MS, and compounds separated on a Varian VF-5MS 30 m × 0.25 mm inner diameter (ID) column with an injector temperature of 250 °C, and a column temperature of 40 °C for 5 min, which was then ramped at 50 °C/min to 150 °C, followed by a ramp at 15 °C/min to 260 °C with a 4.5-min hold; flow rate was 1 ml/min. The MS was operated in electron ionization mode at 150 eV. Tentative identification of the peaks was made by comparing MS fragment patterns with spectra from the National Institute of Standards and Technology (NIST) 98 and Wiley library databases. Chemical ionization with acetonitrile was used to determine molecular weights and to assign double-bond position through derivatization of the double bond and formation of characteristic addition compounds [27,28]. The identities of peaks 1, 2, and 4 were further confirmed by the production of identical retention times and fragment patterns in both electron and chemical ionization modes when compared with chemical standards.
TTP trials consisted of 32 min during which we made two injections: Injection 1, 50 μl of pure hexane, during minute 1; Injection 2, 50 μl of TTP solution, during minute 16. Hexane trials were similar to TTP trials except that Injection 2, like Injection 1, consisted of 50 μl of pure hexane.
The TTP solution contained Z-(9)-tricosene, tricosane, and pentacosane, each diluted 1:100 in hexane and mixed at a ratio of 1:2:3, respectively (this ratio produced chromatograms with peak heights that approximately matched those from samples of waggle dancers), and then further diluted 1:10 in hexane. To volatilize the liquid TTP solution, we injected it into a heated glass tube with 0.5-cm ID. Immediately after injection, we inserted into the tube a fan to blow the vaporized solution through the tube and into a funnel (10-cm diameter) positioned 1 cm above the comb surface of a colony's dance floor. The identical method was used to volatilize and deliver hexane only during Hexane trials (Injection 1: hexane, Injection 2: hexane). To avoid contamination between TTP and Hexane trials, we used separate equipment for TTP solution and hexane. The temperature of the gaseous mixture arriving on the dance floor was maintained between 35 °C and 40 °C. At least one waggle dancer was present on the dance floor during each trial.
We conducted each trial on a different day between 14 July and 8 October 2004. To avoid seasonal effects, we randomized the type of trial (TTP or Hexane) performed each day. To reduce the effect of time of day, trials for a colony started during the same hour every day. Trials for the first colony started between 1200 and 1300, and for the second colony between 1030 and 1130. To further account for day and time effects, the data for each trial were normalized by dividing each 1-min count by the average number of bees exiting the hive per minute during the 10 min immediately preceding the trial.
To compare the presence of the four compounds on the abdomens of waggle dancers with either nondancing foragers that returned from the same unscented food source or nonforaging bees, we used a one-way ANOVA for each compound using Box-Cox transformed data, and a Tukey HSD for unequal n, with data for both colonies pooled.
To compare the effect of injection of TTP solution with injection of hexane, we used two-sample T-tests on normally distributed sample groups with equal variances. To account for day and time effects, the data for each trial were normalized by dividing each 1-min count by the average number of bees exiting the hive per minute during the 10 min immediately preceding the trial.
|
10.1371/journal.pgen.1003613 | The Conserved Intronic Cleavage and Polyadenylation Site of CstF-77 Gene Imparts Control of 3′ End Processing Activity through Feedback Autoregulation and by U1 snRNP | The human gene encoding the cleavage/polyadenylation (C/P) factor CstF-77 contains 21 exons. However, intron 3 (In3) accounts for nearly half of the gene region, and contains a C/P site (pA) with medium strength, leading to short mRNA isoforms with no apparent protein products. This intron contains a weak 5′ splice site (5′SS), opposite to the general trend for large introns in the human genome. Importantly, the intron size and strengths of 5′SS and pA are all highly conserved across vertebrates, and perturbation of these parameters drastically alters intronic C/P. We found that the usage of In3 pA is responsive to the expression level of CstF-77 as well as several other C/P factors, indicating it attenuates the expression of CstF-77 via a negative feedback mechanism. Significantly, intronic C/P of CstF-77 pre-mRNA correlates with global 3′UTR length across cells and tissues. In addition, inhibition of U1 snRNP also leads to regulation of the usage of In3 pA, suggesting that the C/P activity in the cell can be cross-regulated by splicing, leading to coordination between these two processes. Importantly, perturbation of CstF-77 expression leads to widespread alternative cleavage and polyadenylation (APA) and disturbance of cell proliferation and differentiation. Thus, the conserved intronic pA of the CstF-77 gene may function as a sensor for cellular C/P and splicing activities, controlling the homeostasis of CstF-77 and C/P activity and impacting cell proliferation and differentiation.
| Autoregulation is commonly used in biological systems to control the homeostasis of certain activity, and cross-regulation coordinates multiple processes. We show that vertebrate genes encoding the cleavage/polyadenylation (C/P) factor CstF-77 contain a conserved intronic C/P site (pA) which regulates CstF-77 expression through a negative feedback loop. Since the usage of this intronic pA is also responsive to the expression of other C/P factors, the pA can function as a sensor for the cellular C/P activity. Because the CstF-77 level is important for the usage of a large number of pAs in the genome and is particularly critical for expression of genes involved in cell cycle, this autoregulatory mechanism has far-reaching implications for cell proliferation and differentiation. The human intron harboring the pA is large and has a weak 5′ splice site, both of which are also highly conserved in other vertebrates. Inhibition of U1 snRNP, which recognizes the 5′ splice site of intron, leads to upregulation of the intronic pA isoform of CstF-77 gene, suggesting that the C/P activity in the cell can be cross-regulated by splicing, leading to coordination between these two processes.
| Pre-mRNA cleavage/polyadenylation (C/P) is a 3′ end processing mechanism employed by almost all protein-coding genes in eukaryotes [1], [2]. The site for C/P, commonly known as the polyA site or pA, is typically defined by both upstream and downstream cis elements [3], [4]. In metazoans, upstream elements include the polyadenylation signal (PAS), such as AAUAAA, AUUAAA, or close variants, located within ∼40 nucleotides (nt) from the pA; the UGUA element [5], typically located upstream of the PAS; and U-rich elements located around the PAS. Downstream elements include the U-rich and GU-rich elements, which are typically located within ∼100 nt downstream of the pA.
Most mammalian genes express alternative cleavage and polyadenylation (APA) isoforms [6], [7]. While the majority of alternative pAs are located in the 3′-most exon, leading to regulation of 3′ untranslated regions (3′UTRs), about half of the genes have pAs located in introns [8], leading to changes in coding sequences (CDSs) and 3′UTRs. Intronic pAs can be classified into two groups depending upon the splicing structure of the resultant terminal exon: composite terminal exon pA or skipped terminal exon pA. A composite terminal exon pA is located in a terminal exon which contains both exon and intron sequences. In this case, a 5′ splice site (5′SS) is located upstream of the pA. A skipped terminal exon pA is located in a terminal exon which can be entirely skipped in splicing. We previously found that composite terminal exon pAs in the human genome are typically located in large introns with weak 5′SS [9]. A classic model of composite terminal exon pA is the intronic pA of the immunoglobulin heavy chain M (IgM) gene [10]. IgM mRNAs switch from using a 3′-most exon pA to an intronic pA during activation of B cells, which results in a shift in protein production from a membrane-bound form to a secreted form.
In mammalian cells, over 20 proteins are directly involved in C/P [1], [11]. Some proteins form complexes, including the Cleavage and Polyadenylation Specificity Factor (CPSF), containing CPSF-160, CPSF-100, CPSF-73, CPSF-30, hFip1, and Wdr33; the Cleavage stimulation Factor (CstF), containing CstF-77, CstF-64, and CstF-50; Cleavage Factor I (CFI), containing CFI-68 or CFI-59 and CFI-25; and Cleavage Factor II (CFII), containing Pcf11 and Clp1. Single proteins involved in C/P include Symplekin, poly(A) polymerase (PAP), nuclear poly(A) binding protein (PABPN), and RNA Polymerase II (RNAPII). In addition, RBBP6, PP1α, PP1β are homologous to yeast C/P factors [12], whose functions in 3′ end processing are yet to be established in mammalian cells.
CstF-77 has been shown to interact with several proteins in the C/P complex, such as CstF-64 and CstF-50 in CstF [13], [14], [15], [16], CPSF-160 [17], and the carboxyl (C)-terminal domain (CTD) of RNAPII [18]. CstF-77 can dimerize through the second half of its amino (N)-terminal 12 HAT domains [15], [16], which is also responsible for dimerization of the CstF complex. Therefore, the role of CstF-77 in 3′ end processing appears to be bridging and/or positioning various factors for C/P. While CstF-77 has not been extensively studied in mammalian cells, its homolog in Drosophila, suppressor of forked or su(f), has been associated with a number of functions. First, su(f) regulates 3′ end processing of transposable elements, impacting their effects on cellular genes [14], [19], [20]; Second, su(f) regulates the usage of an intronic pA of its own pre-mRNA, creating an autoregulatory mechanism [21]; Third, su(f) is more expressed in mitotically active cells, which was suggested to be attributable to weak autoregulation in dividing cells compared to non-dividing ones [22], and the su(f) mutant strain showed a defect in cell proliferation [23]. However, not all parts of su(f) can be replaced by human CstF-77 for functional complementation, indicating structural and functional differences between these two proteins [24].
We previously identified a conserved pA in intron 3 (In3) of the human CstF-77 gene [25]. Here, we analyze the function of In3 pA and the significance of its flanking splicing and C/P features. We elucidate how In3 pA usage is related to global 3′UTR regulation across cells and tissues, and how In3 pA is regulated by C/P and splicing activities. We demonstrate that perturbation of CstF-77 expression leads to widespread APA and disturbance of cell proliferation and differentiation.
We previously found that vertebrate genes encoding the C/P factor CstF-77 contain a conserved intronic pA (Figure S1) [25]. To elucidate the function of this pA, we first focused on the human CstF-77 gene (CSTF3), which has 21 exons (Figure 1A) with the conserved intronic pA located in intron 3 (In3). Remarkably, the 5′ portion of the gene before exon 4 accounts for 69% of the gene region. Both introns 1 and 3 are very large, with intron 3 (33.2 kb) being larger than 96% of all introns in the human genome and accounting for 43% of the gene region, whereas intron 2 is small, below 8% of all introns in the genome (Figure 1B, left panel). In addition, introns 1–3 are highly conserved in size across vertebrate CstF-77 genes, both in absolute and relative values (Figure S2A), suggesting functional relevance. Using the maximum entropy (MaxEnt) method to examine splice site strength [26], we found that the 5′ splice site (5′SS) of intron 3 is very weak, at the 0.7th-percentile of all introns in the human genome (Figure 1B, middle panel), whereas the 3′SS of intron 3 is very strong, at the 94.3th-percentile of all human introns (Figure 1B, right panel). Notably, using PhastCons scores [27], we found that the surrounding sequences of the 5′SS and intronic pA are much more conserved than those of other 5′SSs and pAs, respectively, in the human genome (Figure S2B). By contrast, conservation of sequence around the 3′SS is modest (Figure S2B).
We next examined how unique is the combination of large intron, weak 5′SS and strong 3′SS in the human genome. Since splicing in higher species is typically governed by the exon-definition model [28], we also included 3′SS of exon 3 and 5′SS of exon 4 in the analysis (Figure 1C). Using the intron density map (see Materials and Methods for detail) to simultaneously interrogate intron size and 5′SS or 3′SS strength (Figures 1D and 1E), we found that large introns in the human genome in general are flanked by strong 5′SS and 3′SS of both upstream and downstream exons, as indicated by enrichment of introns with these features relative to introns with randomized size and 5′SS or 3′SS strengths (shown as observed (Obs)/expected (Exp)). This trend holds for intron 3 of CSTF3 except for its 5′SS. Indeed, the combination of large intron with weak 5′SS is significantly depleted in the human genome (Figure 1D). Therefore, the combination of large size and weak 5′SS of intron 3 is rather unique for human introns.
To examine the significance of various features surrounding In3 pA of CSTF3, we constructed reporter plasmids (called pRinG-77S) containing an intron using the 5′SS and 3′SS of intron 3 (Figure 2A). The 5′ region also contained the In3 pA, which can lead to a short, intronic pA isoform (isoform P) encoding the red fluorescence protein (RFP). If the intronic pA is not used, a long, splicing isoform (isoform S) is expressed, which encodes both RFP and enhanced green fluorescence protein (EGFP)(Figure 2A).
To examine the importance of intron size, we cloned 3′ regions of intron 3 with various sizes. As the insert size increased, the amount of intronic pA product also increased (Figure 2B). A linear correlation between the ratio of isoform P to isoform S (log2), representing the intronic pA usage, and the insert size can be discerned for inserts from 401–1,690 nt. However, the ratio did not change when the insert size was increased to 2,378 nt. This regulation of intronic pA usage is due to the change of intron size rather than the distance between the intronic pA and the SV40 pA at the 3′ end of reporter gene, because no significant difference could be discerned when we expanded the region after 3′SS by adding another EGFP sequence (Figure S3A). In addition, a linear decrease of intronic pA usage was also observed when the distance between 5′SS and pA was expanded by random sequences (Figure S3B). Taken together, these data indicate that there is a kinetic competition between C/P and splicing in the usage of In3 pA.
Several cis elements around In3 pA are highly conserved across vertebrates, including the upstream UGUA and AUUAAA elements and downstream U-rich and GU-rich elements (Figure S1). To examine the contributions of these cis elements to the pA strength, we mutated AUUAAA to AAUAAA, a stronger C/P signal [29], and/or deleted the downstream GU-rich elements. We found that mutation of AUUAAA to AAUAAA led to an ∼2-fold increase in pA usage, whereas deletion of the GU-rich elements led to an ∼10-fold decrease (Figure 2C). Thus, this analysis indicates that the strength of In3 pA of CSTF3 is at a suboptimal level. Interestingly, the slope of the curve for log2(P/S) vs. pA to 3′SS distance appeared different for constructs with GU-rich elements compared to those without (Figure 2C). By contrast, mutation of AAUAAA to AUUAAA did not lead to a slope change, suggesting that the contribution of GU-rich elements to pA strength may be different than that of PAS.
We next examined the importance of 5′SS and 3′SS strengths. We mutated the 5′SS sequence to two stronger sequences (mutants 1 and 2, Figure 2D) based on their MaxEnt scores. Mutants 1 and 2 would be at the 51.9th- and 95.5th-percentile, respectively, in the human genome. The relative strengths of the 5′SSs were also confirmed by comparison with the consensus sequence of all human 5′SSs, represented by position-specific scoring matrix (PSSM) scores, and by free energy values for base-pairing with U1 snRNA (Figure 2D). Strengthening 5′SS drastically inhibited intronic pA usage: ∼90% decrease for mutant 1 and no detectable intronic pA usage for mutant 2 even though AAUAAA was used as PAS (Figure 2D). We also weakened 3′SS strength from the 94.3th-percentile to the 3.8th-percentile based on the MaxEnt score. However, only a minor increase of intronic pA usage was detected (Figure 2E). Together, these data indicate 5′SS strength is a determining factor for the usage of In3 pA.
In both human and mouse cells, the In3 pA can lead to 2 short isoforms (isoforms 2 and 3 in Figure 1A), depending upon whether or not intron 2 is spliced. The isoform 2, which does not have retention of intron 2, is ∼2–3-fold more abundant than isoform 3 in HeLa and C2C12 cells based on the semi-quantitative reverse transcription (RT)-PCR analysis (Figure S4). According to the open reading frames, isoform 2 would encode a protein of 103 amino acids (aa), containing the N-terminal region of CstF-77 and some aa from the intronic region of intron 3 (Figure S5), whereas isoform 3 would encode a protein of 44 aa. However, we could not detect these protein products using various antibodies against the N-terminal region of CstF-77. In addition, several lines of evidence indicate that the coding region from intron 3 may cause rapid degradation of the protein encoded in isoform 2: 1) pRinG-77S constructs expressing different amounts of intronic pA isoforms showed the same red fluorescence to green fluorescence ratio when transfected into HeLa cells (Figure S6A); 2) immunoblot analysis using an antibody against RFP did not detect protein products of the intronic isoforms expressed from pRinG-77S constructs (Figure S6B); 3) a bicistronic mRNA containing RFP tagged with the intronic coding sequence between 5′SS and stop codon followed by IRES and EGFP resulted in green fluorescence only (Figure S6C). Thus, it appears that the protein products from intronic pA isoforms are expressed at very low levels at most.
Given that CstF-77 is a C/P factor, we next reasoned that it may regulate the usage of its own intronic pA, creating a feedback autoregulatory mechanism, similar to its fly homolog su(f) [21]. To this end, we used small interfering RNAs (siRNAs) to specifically knock down the expression of the CstF-77 transcripts encoding full length protein (named CstF-77.L mRNAs). The CstF-77.L mRNA level significantly decreased after 8 hr of siRNA transfection and its protein level started to decrease after 16 hr (Figure 3A). Interestingly, expression of isoforms 2 and 3, collectively named CstF-77.S mRNAs, also gradually decreased after 16 hr, indicating that the expression of CstF-77.S mRNAs can be controlled by the CstF-77 protein level. By contrast, knockdown of CstF-77.S mRNAs (by ∼50%, Figure 3B, left) did not affect CstF-77.L mRNAs (Figure 3B, right), suggesting that expression of CstF-77.S mRNA is not important for CstF-77.L expression. In accord with the autoregulatory mechanism, expression of exogenous CstF-77 led to increased expression of endogenous CstF-77.S mRNAs and decreased expression of endogenous CstF-77.L mRNAs (Figure 3C). Consistently, knockdown of CstF-77.L mRNAs inhibited intronic pA usage for the reporter construct pRinG-77S-831 (structure shown in Figure 2A), whereas knockdown of CstF-77.S mRNAs had no effect (Figure 3D); and overexpression of CstF-77 enhanced intronic pA usage for the reporter construct (Figure 3E).
We next reasoned that the negative feedback autoregulatory control may cause CstF-77.S and CstF-77.L isoforms to oscillate in their expression. To test this hypothesis, we examined expression of CstF-77.S and CstF-77.L mRNAs over time after plating cells. Indeed, as shown in Figure 4F, these two isoforms oscillated over time: when CstF77.L level was high CstF77.S level was low, and vice versa. Taken together, these data indicate that intronic pA usage is responsive to CstF-77 expression, creating a feedback autoregulatory mechanism.
We previously found that the expression of C/P factors negatively correlates with the 3′UTR length in development and cell differentiation [30]. We asked whether the CstF-77.S/CstF-77.L ratio is related to APA of 3′UTRs. To this end, we analyzed an exon array dataset for 11 mouse tissues and a deep sequencing dataset for 10 human tissues and 7 human cell lines. The CstF-77.S/CstF-77.L ratio was calculated by comparing the intensity of microarray probes or density of RNA-seq reads for CstF-77.S with those for CstF-77.L (Figure 4A). The global 3′UTR length changes were calculated by comparing the intensity of microarray probes or density of RNA-seq reads for the region upstream of first pA in 3′UTR (called constitutive 3′UTR or cUTR) with those for the downstream region (called alternative 3′UTR or aUTR)(Figure 4A). This value was also called Relative expression of isoforms Using Distal pAs (RUD, see Materials and Methods for detail). The median RUD of all genes reflects the relative global 3′UTR length. Interestingly, the CstF-77.S/CstF-77.L ratio generally correlated with the global 3′UTR length in both human (R2 = 0.61, Pearson Correlation) (Figure 4B) and mouse cells/tissues (R2 = 0.71, Pearson Correlation)(Figure 4C). This result indicates that the CstF-77.S/CstF-77.L ratio is associated with APA of 3′UTRs.
We next analyzed our previously published exon array data for differentiation of C2C12 myoblast cells [31], with which we reported general lengthening of 3′UTR during cell differentiation. A linear correlation (R2 = 0.61) between CstF-77.S/CstF-77.L and RUD was also detected (Figure 4D). To validate this finding, we examined expression of CstF-77.S and CstF-77.L mRNAs by RT-qPCR in proliferating C2C12 cells and cells after 1 day or 4 days of differentiation. CstF-77.S mRNAs showed increased expression by ∼20% after 1 day of differentiation but no significant change of expression after 4 days. By contrast, the expression of CstF-77.L mRNAs gradually decreased (Figure 4E). Consequently, the CstF-77.S/CstF-77.L ratio gradually increased in differentiation (Figure 4E). Consistently, the CstF-77 protein level decreased by 27% after 1 day and by 46% after 4 days (Figure 4F). Thus, the usage of In3 pA of CstF-77 gene inversely correlates with CstF-77 protein level in cell differentiation. Given CstF-77's role in C/P, this result suggests that CstF-77 protein level may be the underlying reason for the connection between the CstF-77.S/CstF-77.L ratio and global 3′UTR length.
The increased CstF-77.S/CstF-77.L ratio in differentiation could be due to activation of C/P at In3 pA, which, however, seems in discord with our previous finding that the C/P activity in general is weakened in C2C12 differentiation [31]. Notably, intronic pA of CstF-77 without flanking 5′SS and 3′SS was less used in differentiated cells compared to proliferating cells by reporter assays (Figure S7), suggesting that the pA usage per se is decreased in differentiation. To explore this issue further, we knocked down several factors in the C/P machinery, including CstF-64 in the CstF complex (Figure 5A), CFI-25, CFI-68, and CFI-59 in the CFI complex (Figures 5B), and CPSF-160 and CPSF-73 in the CPSF complex (Figure 5C). All the knockdowns led to significant decrease of the CstF-77.S/CstF-77.L ratio, indicating that the pA usage is responsive to perturbation of the C/P activity. Furthermore, changing the pA strength does not alter the trend of CstF-77.S/CstF-77.L ratio changes in differentiation (Figure 5D).
We next reasoned that since intron size and 5′SS strength can regulate the usage of intronic pA, change of splicing activity in differentiation may lead to change of CstF-77.S/CstF-77.L. Notably, mRNAs encoding several U1 snRNP and U2 snRNP factors are downregulated in differentiation based on microarray analysis (Figure 6A), suggesting weakening of their activities. To examine the effect of splicing regulation on CstF-77.S/CstF-77.L, we knocked down U1-70K, one of the components of U1 snRNP [32], SF3B1, a key component of U2 snRNP [33], and U2AF65, a factor involved in recognition of 3′SS and recruitment of U2 snRNP [34]. Knockdown of U1-70K led to a significant increase (∼50%) of the CstF-77.S/CstF-77.L ratio (P<0.05, Figure 6B), whereas knockdown of SF3B1 led to a marginal increase of the ratio (P>0.1, Figure 6C), and knockdown of U2AF65 led to a significant decrease of the ratio (P<0.05, Figure 6D). Thus, U1 snRNP may play a role in regulation of intronic pA usage in C2C12 differentiation.
To further explore the role of U1 snRNP in intronic C/P of CstF-77, we used an oligonucleotide which mimics the consensus sequence of 5′SS [35], termed U1 domain (U1D) oligo (Figure 6E). Presumably, U1D can sequester U1 snRNP in the cell, thereby inhibiting 5′SS recognition by U1 snRNP [35]. Upon treatment of U1D, the CstF-77.S/CstF-77.L ratio increased by 2-fold (Figure 6F). An even greater increase (∼9-fold) of intronic pA usage was observed from reporter assays using pRinG-77S-1690 (Figure 6G). Taken together, these results indicate that the intronic pA CstF-77 gene is under the control of U1 snRNP.
The effect of U1 snRNP regulation on intronic C/P is consistent with the critical role of 5′SS for pA usage (see above). To directly examine whether the 5′SS strength is important for intronic pA usage, we used reporter constructs with different 5′SS strengths in proliferating and differentiating C2C12 cells (Figure 6H). Interestingly, while the construct with wild type, weak 5′SS recapitulated the intronic pA usage of endogenous CstF-77 pre-mRNAs, the mutant 5′SS with medium strength showed the opposite trend, indicating that 5′SS strength is critical for the regulation of intronic pA usage. This result is in line with the general trend that intronic pAs activated in C2C12 differentiation tend to be in introns with weak 5′SS (Figure S8).
In order to understand how critical it is to control CstF-77 expression in cell proliferation and differentiation, we knocked down CstF-77 in proliferating C2C12 cells and examined APA and gene expression genome-wide using our newly developed method, 3′ region extraction and deep sequencing (3′READS)(Figure 7A) [8]. We found 1,068 genes that had significant APA changes in 3′UTRs (P<0.05, Fisher's Exact test, and >5% change in isoform abundance)(Figure 7A). However, there was no significant bias of expression to proximal or distal pA isoforms, indicating no global 3′UTR shortening or lengthening after CstF-77 knockdown. Gene Ontology analysis indicated that genes with different functions were affected differently (Table 1). For example, genes with functions in “protein localization”, “intracellular transport”, “RNA processing” were more likely to have 3′UTRs lengthened, whereas those with functions in “cell-cell adhesion”, “mitosis”, and “Ras protein signal transduction” were more likely to have 3′UTRs shortened.
We next compared this data with our recently published data for APA regulation in C2C12 differentiation [8]. Whereas only a small set of genes were found to be commonly regulated between CstF-77 knockdown and C2C12 differentiation, the number of consistently regulated genes was significantly greater than that of oppositely regulated genes (P = 4.1×10−9, Chi-squared test), suggesting downregulation of CstF-77 is involved in regulation of a subset of APA events in C2C12 differentiation (Figure 7B).
We next examined cis elements surrounding pAs of regulated isoforms. Remarkably, U-rich elements were significantly enriched for pAs whose isoforms were downregulated after CstF-77 knockdown, particularly in the −40 nt to −1 nt region relative to the pA (set to 0) (Figure 7C). This result suggests that pAs with U-rich elements are highly dependent on CstF-77 for C/P. Since CstF-77 is in the same complex as CstF-64, we next examined CstF-64 binding near regulated pAs using the CstF-64 CLIP-seq data we recently published [8]. Consistent with the interaction between CstF-77 and CstF-64, pAs of downregulated isoforms had significantly more CstF-64 binding in nearby regions than those of upregulated ones (P<0.05, bootstrap analysis), suggesting that usage of these pAs are also dependent on CstF-64 (Figure 7D).
Our data also indicated that a large number of genes (1,776 in total) had significant changes of expression (fold change >1.5 and P<0.01, Fisher's Exact test) after CstF-77 knockdown. By GO analysis, we found, to our surprise, that genes related to cell cycle were most significantly downregulated (Table 2 and Figure 8A). This result was validated by RT-qPCR for a set of cell cycle-related genes, such as Ccnb1 (cyclin B1), Cdca3 (cell division cycle associated 3), Cdk4 (cyclin-dependent kinase 4), Mcm6 (minichromosome maintenance complex component 6), and Tipin (timeless interacting protein)(Figure 8B). This result indicates that the CstF-77 level is important for expression of cell cycle genes, and suggests that downregulation of CstF-77 may help cells halt proliferation and launch differentiation. To explore this further, we overexpressed CstF-77 in proliferating C2C12 cells, induced differentiation, and examined marker genes that are normally upregulated during differentiation. All three marker gene mRNAs, including Myh3 (heavy polypeptide 3, skeletal muscle, embryonic), MyoG (myogenin), and Tpm2 (tropomyosin 2, beta), were significantly less upregulated in cells overexpressing CstF-77 (Figure 8C), further indicating that the CstF-77 level is important for cell proliferation/differentiation.
In this study, we examined the evolution and regulation of intronic C/P of human and mouse CstF-77 genes. The conservation of various features involved in pA usage across vertebrates underscores its importance. Notably, the Drosophila gene encoding the homologue of CstF-77, su(f), also contains an intronic pA [19]. Unlike the intronic pA isoforms of vertebrate CstF-77 genes, which have open reading frames, the su(f) intronic pA isoform does not have an in-frame stop codon. However, both vertebrate and fly intronic pAs appear to function to attenuate expression of the gene via feedback autoregulation. Remarkably, there is no conservation in surrounding sequences or adjacent intron/exon structures between the intronic pAs in vertebrates and in fly, indicating convergent evolution of this mechanism. Intriguingly, we could not find a similar mechanism in C. elegans after exhaustive search of all available public pA data. It remains to be seen whether or not the CstF-77 homolog in C. elegans is subject to another type of autoregulation.
In addition to autoregulation, we found that the intronic pA usage is regulated upon perturbation of several other C/P factors, including those in the CstF, CPSF and CFI complexes, suggesting it is responsive to the general C/P activity in the cell. Two key features of the intronic pA of the CstF-77 gene may make it particularly suitable for this function: first, its suboptimal strength can create a wide dynamic range of usage in response to change of C/P activity; second, its placement in an intron can allow rapid regulation because of competition of its usage with splicing.
Juge et al. proposed two modes of autoregulation for fly su(f), a strong mode in non-dividing cells and a weak mode in dividing cells [22]. Here, our study indicates that splicing plays a dominant role in the usage of intronic pA of CstF-77 gene. Consistently, inhibition of the U1 snRNP activity, not the C/P activity, recapitulates the intronic pA regulation in cell differentiation. Thus, we propose that the U1 snRNP activity sets the general level of intronic pA usage under different conditions, such as in cell proliferation and differentiation, and the C/P activity plays a fine tuning role to robustly control CstF-77 expression under a given condition (Figure 9). This model would readily explain the two modes of autoregulation proposed by Juge et al., i.e., the U1 snRNP activity is high in dividing cells and weak in non-dividing cells. Moreover, control of CstF-77 level by U1 snRNP suggests that the C/P activity in the cell is modulated by the splicing activity, leading to coordination between these two pre-mRNA processing steps. This coordination may ensure that the widespread cryptic pAs in introns are not activated when U1 snRNP is downregulated under conditions like cell differentiation [9]. Conversely, this result may explain, at least partially, that mild inhibition of U1 snRNP can lead to 3′UTR lengthening [36].
Regulation of intronic pA of CstF-77 is reminiscent of a similar mechanism for the IgM gene. Both intron size and 5′SS strength were found to be important for the usage of intronic pA in the IgM gene [37], [38]. A number of factors have been implicated in the regulation, including the C/P factor CstF-64 [10] , the U1 snRNP component U1A [39], and the RNAPII transcription elongation factor ELL2 [40]. Whereas we found U1 snRNP regulation correlates with the activation of intronic pA of CstF-77 gene, future studies are needed to examine whether additional mechanisms can contribute to this regulation. Of particular importance is whether other splicing factors also play a role in the regulation of intronic pA. Notably, splicing factors in general are downregulated in C2C12 differentiation (Figure S9). Here we observed only marginal activation of In3 pA after SF3B1 knockdown and, surprisingly, inhibition of the pA after U2AF65 knockdown. It remains to be seen how factors involved in different steps of splicing regulate the usage of In3 pA and intronic C/P in general.
Perturbation of CstF-77 expression led to widespread APA and expression changes of a large number of genes. Remarkably, the genes with functions in cell cycle are most significantly affected, indicating that they are highly dependent upon CstF-77 for expression. pAs surrounded with U-rich elements appeared to be more affected by CstF-77 knockdown. Indeed, we found downregulated genes with a single pA also tend to have U-rich elements surrounding the pA (Figure S10), suggesting that inefficient 3′ end processing may lead to their downregulation of expression. Intriguingly, cell cycle genes tend to have shortened 3′UTRs after CstF-77 knockdown (Table 1). Since genes with shortened 3′UTRs tend to be downregulated (Figure S11), it is possible that distal pAs of cell cycle genes are more responsive to the CstF-77 level. Future work is needed to fully unravel the mechanism by which CstF-77 regulates cell cycle genes.
We found pAs with more CstF-64 binding are more likely to be affected by CstF-77 knockdown indicating that some of the regulation is through the CstF complex. However, CstF-64 is also known to bind GU-rich elements [41], and our cross-linking immunoprecipitation and high-throughput sequencing (CLIP-seq) data from C2C12 cells showed that the top two most significant pentamers for CstF-64 binding are UGUGU and UUUUU [8]. But GU-rich elements are only modestly enriched in the downstream region of regulated pAs after CstF-77 knockdown. Whether pAs with a different number or placement of U-rich and UG-rich elements are differentially regulated by CstF-77 and CstF-64 needs to be examined in the future. Moreover, a recent genome-wide study of CstF-64 knockdown in HeLa cells indicated that only a small set of APA events in these cells are regulated by the factor [42]. However, co-depletion of CstF-64 and its paralog τCstF-64 leads to more APA changes, largely leading to 3′UTR lengthening. That APA pattern appears different than that observed in this study with CstF-77 knockdown. Whether the difference is due to different levels of knockdown or different cell types used in the studies needs to be further explored.
Construction of the pRinG vector and all plasmids derived from pRinG are described in Table S1. The pRiG vector and pRiG-77.AE containing the intronic pA of CstF-77 were described previously [31]. For pCMV-CstF-77, the open reading frame (ORF) of human CstF-77 was obtained from the IMAGE clone 5223351 (Invitrogen) by PCR using primers 5′-CGATGAATTCATGTC AGGAGACGGAGCC and 5′-GGCCCTCGAGCTACCGAATCCGCTTCTG. The fragment was digested by EcoR I and Xho I, and then inserted into the pcDNA3.1/His C vector (Invitrogen) digested with the same enzymes.
HeLa cells and C2C12 cells were maintained in Dulbecco's Modified Eagles Medium (DMEM) supplemented with 10% fetal bovine serum (FBS). Differentiation of C2C12 cells was induced by switching cell media to DMEM+ 2% horse serum (Sigma) when cells were ∼100% confluent. All media were also supplemented with 100 units/ml penicillin and 100 µg/ml streptomycin. Transfection with plasmids or siRNAs was carried out with Lipofectamine™ 2000 (Invitrogen) or jetPEI(polyplus) according to manufacturer's recommendations. Transfection was carried out for 48 hr unless described otherwise. siRNA sequences are shown in Table S2. The U1D oligo (5′-gCcAgGuAaGuau) and control oligo (5′-CAGAAATACACAATA), where locked nucleic acid (LNA) residues are in uppercase, 2′-OMe RNA residues are in lowercase, DNA nucleotides are in underlined uppercase, were previously described in Goraczniak et al. [35] (called UA17-13B-U1D and UA17-13B-TD, respectively). These oligos were transfected into C2C12 cells at 15 µM using Lipofectamine 2000 when the confluency of cells was about 50%. Cells were harvested 48 hr after transfection.
For fluorescent activated cell sorting (FACS) analysis, cells were released from culture dishes by Trypsin-EDTA 24 h after transfection and green and red fluorescence signals were read at 530 nm and 585 nm, respectively, in the BD FACScalibur system (BD Biosciences). For immunoblot, the RIPA buffer (1% NP-40, 0.1% SDS, 50 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5% Sodium Deoxycholate, and 1 mM EDTA) was used to extract proteins from the cell. Proteins were resolved by SDS-PAGE, followed by immunoblotting using antibodies. Antibodies used in this study and their sources are shown in Table S3.
Total cellular RNA was extracted using Trizol (Invitrogen) according to manufacturer's protocol. RNA was run in a 1.2% denaturing agarose gel, and was transferred to nylon membrane overnight. RNA was detected by hybridization with a radioactively labeled probe for the RFP sequence. The probe was made by PCR using pDsRED-Express-c1 as template and primers 5′-CGATGCTAGCATGGCCTCCTCCGAGGAC and 5′-GGCCCTCGAGCTACAGGAACAG GTGGTG with α-32P-dCTP. For RT-qPCR, mRNA was reverse-transcribed using the oligo-dT primer (Promega), and qPCR was carried out with Syber-Green I as dye. Primers are shown in Table S4.
To calculate the global 3′UTR length (RUD) score, we used exon array data for C2C12 cell differentiation [31], exon array data for mouse tissues (http://www.affymetrix.com/support/technical/sample_data/exon_array_data.affx), and RNA-seq data for human tissues and cell lines [7]. Exon array data were first normalized by the Robust Multichip Average (RMA) method. Expressed genes were selected by the Detection Above Background (DABG) method. The RUD score was based on the ratio of average probeset intensity of aUTR to that of cUTR, as previously described [31]. For RNA-seq data, the RUD score was based on the ratio of read density of aUTR to that of cUTR, as described previously [43]. For both analyses, aUTRs and cUTRs were defined by PolyA_DB 2 [44]. The relative expression of CstF-77.S vs. CstF-77.L was calculated based on the probes or RNA-seq reads specific for each isoform. For analysis of splicing factor expression, we examined mRNAs encoding splicing factors defined by Jurica and Moore [45].
5′SS and 3′SS were analyzed as previously described [9]. Briefly, we used all GT-AT type introns supported by human RefSeq sequences to build Position Specific Scoring Matrices (PSSMs) for 5′SS and 3′SS. For 5′SS, we used −3 to +6 nt surrounding the 5′SS, with 3 nt in the exon and 6 nt in the intron; for 3′SS, we used −22 to +2 nt surrounding the 3′SS, with 22 nt in the intron and 2 nt in the exon. The maximum entropy scores were calculated with MaxEnt [26]. 5′SS sequences were also scored by their ability to anneal with U1 snRNA. We used the sequence 5′-ACUUACCUG of U1 snRNA to form duplex structures with 5′SS sequences using the RNAduplex function of ViennaRNA [46]. For intron density map, we used all the RefSeq-supported introns as the observed set, and created an expected set using randomized pairs of intron size and splice site. The introns were divided into 20 fractions based on intron size and splice site strength, respectively, and distributed in a 20×20 grid. For each cell in the grid, the ratio of the number of introns in the observed set to that in the expected set was calculated and represented by color in a heatmap.
The 3′ region extraction and deep sequencing (3′READS) method used in this study is the same as previously described [8], except that 10 µg of total RNA was used, and poly(A)+ RNA was selected by the chimeric U5 and T45 (CU5T45) oligo conjugated on streptavidin beads and fragmented by 1 U RNase III at 37°C for 30 min. Poly(A)+ RNA fragments were subject to further processing as previously described [8]. pA identification was carried out as previously described [8]. Only poly(A) site supporting (PASS) reads, defined as having >2 non-genomic Ts at the beginning of read, were used for further analysis. The expression level of an APA isoform was calculated using the number of PASS reads assigned to the pA. To study APA in the 3′-most exon, we first selected the top two expressed isoforms and used the Fisher Exact test to examine their difference in abundance between CstF-77 knockdown and control. Significantly regulated isoforms were those with P<0.05 and change of abundance >5%. Gene expression was calculated using the total number of PASS reads assigned to a gene. Differentially regulated genes after CstF-77 knockdown were those with P<0.01 (Fisher's Exact test) and fold change >1.5 compared to control. The DAVID software was used to identify Gene Ontology terms enriched for genes with significant changes in gene expression or APA [47].
We examined four regions around the pA, i.e., −100 to −41 nt, −40 to −1 nt, +1 nt to +40 nt and +41 nt to +100 nt. For each region, the Fisher's Exact test was used to check whether a sequence was enriched for a set of pAs vs. another set, for example, those of upregulated isoforms vs. downregulated isoforms.
CLIP-seq reads of CstF-64 [8] were aligned to the mouse genome (mm9) using Novoalign (http://www.novocraft.com/). Reads with deletions caused by skipping of reverse transcriptase at the UV cross-linked bases were used for analyses. A bootstrapping method [8] was used to compare CstF-64 binding in the −10 to +40 nt region around the pA for pAs of upregulated isoforms vs. those of downregulated isoforms after CstF-77 knockdown.
|
10.1371/journal.pntd.0003423 | Reduction of Urogenital Schistosomiasis with an Integrated Control Project in Sudan | Schistosomiasis remains a major public health concern in Sudan, particularly Schistosoma haematobium infection. This study presents the disease-reduction outcomes of an integrated control program for schistosomiasis in Al Jabalain locality of White Nile State, Sudan from 2009 through 2011.
The total population of the project sites was 482,902, and the major target group for intervention among them was 78,615 primary school students. For the cross-sectional study of the prevalence, urine and stool specimens were examined using the urine sedimentation method and the Kato cellophane thick smear method, respectively. To assess the impacts of health education for students and a drinking water supply facility at Al Hidaib village, questionnaire survey was done.
The overall prevalence for S. haematobium and S. mansoni at baseline was 28.5% and 0.4%, respectively. At follow-up survey after 6–9 months post-treatment, the prevalence of S. haematobium infection was reduced to 13.5% (95% CI = 0.331–0.462). A higher reduction in prevalence was observed among girls, those with moderately infected status (around 20%), and residents in rural areas, than among boys, those with high prevalence (>40%), and residents in urban areas. After health education, increased awareness about schistosomiasis was checked by questionnaire survey. Also, a drinking water facility was constructed at Al Hidaib village, where infection rate was reduced more compared to that in a neighboring village within the same unit. However, we found no significant change in the prevalence of S. mansoni infection between baseline and follow-up survey (95% CI = 0.933–6.891).
At the end of the project, the prevalence of S. haematobium infection was reduced by more than 50% in comparison with the baseline rate. Approximately 200,000 subjects had received either praziquantel therapy, health education, or supply of clean water. To consolidate the achievements of this project, the integrated intervention should be adapted continuously.
| Schistosomiasis remains a major public health concern and is one of the major causes of morbidity among school-aged children in Sudan. To control schistosomiasis in White Nile State of Sudan, the Korea International Cooperation Agency (KOICA) implemented an integrated control program including mass chemotherapy with praziquantel and health education to school children and village residents, and construction of a drinking water supply facility at Al Hidaib village from 2009 to 2011. As a result of this project, the overall prevalence of S. haematobium infection was reduced by >50% in comparison with the baseline rates (95% CI = 0.331–0.462). The infection reduction rates were higher among girls, those with moderate infection status (around 20%), and residents of rural areas; than for those of boys, subjects with high infection status (>40%) and residents of urban areas. A supply of clean water at Al Hidaib village contributed significantly to the reduction in the prevalence of urogenital schistosomiasis in comparison to Khour Ajwal village, which is similar natural environment and lifestyle of Al Hidaib village. However, the prevalence of S. mansoni infection did not significantly change. Also, the awareness of knowledge about schistosomiasis and health improvement was apparently improved by the results of questionnaires survey. Through this project, approximately 200,000 individuals benefited from either drug treatment, health education, or a clean water supply. To consolidate the achievements of the project, sustainable integrated control activities should be implemented in the near future.
| Schistosomiasis is a parasitic trematodiasis caused by several species of the genus Schistosoma, of which S. mansoni, S. japonicum, S. mekongi, and S. haematobium are of public health importance. These worms live in the veins around the intestine or urinary bladder. Eggs are released in the stool or urine of the host and hatch in water [1]. Humans are usually infected when they come into contact with contaminated fresh water such as collecting water, washing, bathing, playing, fishing, or cultivating crops. In general, children, women, fishermen, and farmers are the high risk groups in schistosomiasis, also other people can infect in the irrigation channels or rivers and suffer from hematuria and anemia, enlargement of the liver and spleen, and growth retardation [1], [2].
In 2012, schistosomiasis was considered endemic in 78 countries [3]. In sub-Saharan Africa, 70 million individuals were estimated to have experienced hematuria in the previous 2 weeks, 32 million had dysuria associated with S. haematobium, 18 million had major bladder wall pathology, and 10 million had S. haematobium-related major hydronephrosis, resulting in an estimated mortality of 150,000 people per year [4]. In 2012, the World Health Organization (WHO) estimated that approximately 249 million people required preventive chemotherapy annually; this figure includes 114 million school-age children, 231 million of whom were in Africa. However, only approximately 42 million people, merely 14.4% of the infected population, received schistosomiasis treatment [3]. Sudan is one of the endemically infected sub-Saharan countries in which both the urogenital and intestinal forms of schistosomiasis are prevalent in populations along the Nile River [5]. S. haematobium infection is dominant and the prevalences range from 1.7% to 56.0% in different localities, whereas the prevalence for S. mansoni is very low [6]–[10].
The Blue and White Nile Rivers meet in Sudan to form the Nile River. Among the countries crossing the Nile River, Sudan has the widest river basin areas, and it also has large irrigated agricultural sectors along the banks of the Nile River. The development of such irrigation schemes has led to significant environmental modification, favoring the spread of vector-borne diseases, including schistosomiasis [11]. Residents of these geographical environments along the Nile River have faced the risk of schistosomiasis for many centuries.
During 1979–1990, the Blue Nile Health Project was implemented as a comprehensive plan to control malaria, schistosomiasis, and diarrheal diseases at Gezira, Managil, and Rahad irrigation schemes. As a result, the prevalence of schistosomiasis was reduced from 53% to less than 10% [10]. However, there have been no further integrated programs to control schistosomiasis in Sudan thereafter. In 2000, the Sudanese government established a national schistosomiasis control program to reduce the prevalence of S. haematobium to less than 10% by 2013. As part of this program, the Korea International Cooperation Agency (KOICA) supported a schistosomiasis control project at the Al Jabalain locality in the White Nile State of Sudan during 2009–2011. The KOICA project included a mass chemotherapy according to baseline survey, health education with material development, construction of a drinking water supply facility at high endemic village, provision of laboratory equipment, dispatch of Korean experts to Sudan and follow-up evaluation.
The present study focused on the outcomes of praziquantel administraion and health education for students and residents, as well as the impacts of a filtered drinking water supply facility at Al Hidaib village among integrated control project supported by the KOICA.
An agreement (Project Code No. 2008-01-0000-039) to conduct “The Project for Combating Schistosomiasis in Sudan" was officially signed by the KOICA of the Republic of Korea and the Federal Ministry of Health (MOH) of the Republic of Sudan in Khartoum, Sudan in June, 2009. According to the signed agreement, the KOICA provided full funding for this project as an official development aid (ODA) program for developing countries. The project was carried out between August 2009 and June 2011 by the Korea Association of Health Promotion (KAHP), and the KAHP provided management consultation for the project. The pilot area of this project was Al Jabalain locality of White Nile State, Sudan, which is composed of 6 units. In Sudan, the unit is a last step of local administrative organization, thus we used the unit as a basic structure of implementation activities. This site was recommended by Sudanese government, because the prevalence of S. haematobium infection of Al Jabalain locality was so high according to the results surveyed by White Nile State MOH, Sudan (25.5%–54.1%).
As shown in Fig. 1, the KOICA project included many activities; however, this study focused on the mass chemotherapy with baseline survey, the health education for primary school students and village residents, and the construction of a facility at Al Hidaib to supply drinking water. These 3 interventions were explained in detail below.
This study protocol was reviewed and approved by the institutional review board of KAHP, which was the implementing organization of this project (Approval No. 10-C-05). Data were collected at pilot areas according to the board-approved framework. Before conducting the survey at each school, informed verbal consent was first obtained from the chief of the villages and the parents of children during the village meeting and was also obtained from school teachers at schools prior to the recruitment of children. In villages, informed verbal consent was first obtained from the village leaders and/or the parents of children during a village meeting. As per traditional culture and due to the low literacy rates in village residents, verbal consent was deemed acceptable. The research program was approved by MOH, White Nile State, Sudan.
The implementation team for this project was composed of one Korean project manager (PM) in Sudan (team leader), one Sudanese PM, 2 Sudanese parasitology experts, 4 Sudanese laboratory technicians, 7 Sudanese health workers and 2 Sudanese drivers.
The Korean PM, who had stayed in Sudan during the project period, was an expert of parasitology and health administration. He managed the whole activities implemented in Sudan. The Sudanese PM and senior parasitology staff were public health experts, who worked more than 15 years at schistosomiasis control programs in Sudan. The Sudanese experts were responsible for the whole activities with the Korean PM, and implemented the health education, communication between Korean PM and Sudanese staff, contact with Sudanese administration officials and residents, drug delivery, and public relationships. Laboratory technicians have worked more than 5 years at parasitological examination using microscope, and worked as officials of MOH, White Nile State. They were responsible for performing the parasitological examination of urine and stools, and recording the results. Health workers were also officers of MOH, White Nile State, and learned about schistosomiasis control and information related with this project from Korean and Sudanese PM or parasitologists sufficiently. They were responsible for collecting the samples, questionnaire survey, assistance of health education and drug delivery.
In addition, two Korean parasitology experts and one administration officer visited project area in Sudan 5 times every 4–6 months for 2 weeks each time during the project periods. They managed the project and transferred the basic concept on parasite control. Also they monitored and evaluated the project progress with Sudanese experts and resolved the problems issued at implementation activities. Korean and Sudanese PM or parasitology experts provided training to health workers and laboratory technicians about the project implementation activities in case of necessary, from time to time, and individually or collectively.
The project site was Al Jabalain locality of White Nile State, Sudan (Fig. 2), which is composed of six units. We used the unit as a basic structure of implementation activities in Sudan. White Nile State is located along the White Nile River in the southeastern part of Sudan, and water canals are well developed in this region. As the land is almost flat, irrigation is based on a network of gravity-fed mud-lined canals. The Al Jabalain locality is situated on the east bank of the White Nile River in southern White Nile State. All villages included in the study were located adjacent to the White Nile River, and most houses were built with mud bricks. Most village residents used water from the river and earned their income from agriculture, livestock breeding, or fishing. The study areas included a total of 169 primary schools and 482,902 individuals, including 78,615 primary school students (Table 1). Rabak unit was an urbanized community, whereas the other Al Jabalain, Assalaya, Jazeera Aba, Joda and Kenana units were rural communities. Fig. 3 shows the adherence of participants to the present study.
According to the agreement, the total numbers for parasitological examination were approximately 10,000 among 78,615 students, and they were selected as follows. The numbers of total students (78,615) were approximately 7.8 times more than target number (approximately 10,000) for baseline survey. Assigned numbers of students of each unit were calculated by dividing of total student numbers of each unit by 7.8. Numbers of surveyed schools were calculated by dividing half of mean number of students of each school in each unit, because samples from the 1st, 3rd and 5th grade classes of each school were collected. Since there were some differences of students' number of each school, we substantially determined the number of schools to collect samples at each unit by a number of averages ± 20%.
The number of schools surveying at baseline were finally selected considering the location, total number of students, accessibility to school, and environmental condition of schools. To evaluate the prevalence of S. haematobium and S. mansoni infection, baseline survey was implemented at 61 primary schools in Al Jabalain locality from December 2009 to March 201 (Table 2, Fig. 4). In one school, we collected the samples around 80-250, but not more than 300. If the number of students of one school was less than 100, all students were subjected. The STROBE check is available as Supporting Information (S1 Checklist).
For evaluation of prevalence and intensity of schistosomiasis of the pilot areas, urine and fecal samples were collected from 1st, 3rd and 5th graders of the sentinel schools of project area. The sample collection team was composed of Korean PM in Sudan, Sudanese PM and Parasitology experts for health education, health workers, and driver. Health workers were responsible for sample collection basically, and receiving the oral consent from each school child.
Containers were delivered in the morning, and then stool and urine samples were collected at the same day (usually within 6 hours, or around 2–3 pm before leaving school), where the teachers helped to check who submitted the samples or not among enrolled students. After delivery the containers to students, Sudanese PM or experts educated students in the classes or ground of the school about prevention of schistosomiasis and public health. The collected samples were transferred to the Schistosomiasis Control Center established by the KOICA project in Rabak, White Nile State immediately.
The prevalence and infection intensity of S. haematobium and S. mansoni infection in students were examined at baseline and follow-up surveys. The samples were basically examined on the day or next day after collection.
Four laboratory technicians examined the samples parasitogically. They have career more than 5 years about urine and stool examination. For quality control, Sudanese or Korean PM or parasitology experts reexamined about 10% of the slides and confirmed the first results. In case of disagreement, the results were discussed with the concerned technician and re-read the discordant slides until agreement was reached. To increase the capability of technicians, Korean experts educated all laboratory technicians about microscope manipulation, identification of parasites, urine and stool processing methods repeatedly.
For the diagnosis of S. haematobium, urine samples were subjected to a urine sedimentation method, as described elsewhere [12]. Briefly, the amount of urine of each tube was adjusted into 10 mL. Urine samples (10 mL) were centrifuged at 1,500 rpm for 5 minutes at room temperature, and pellets were vigorously shaken and examined for S. haematobium or S. mansoni eggs by microscopy. To examine the whole pellets of the tube as possible, all sediments of each tube were transferred into 2–4 slide glasses. The presence of S. haematobium or S. mansoni eggs of each child was the results of a total of 2–4 smears.
For the diagnosis of S. mansoni, fecal samples were subjected to the Kato cellophane thick smear method [13]. Briefly, 1 g of fecal sample was obtained from each tube and prepared two smears using slide glasses, and allowed to clear for at least 30 minutes before examination. The presence of S. mansoni eggs of each child was the sum of 2 smears.
Four medical doctors participated in the KOICA project, namely, two Sudanese and two Korean parasitology experts. All drug treatment was under control of these doctors. Also Sudanese PM and health workers had much experience participating schistosomiasis control projects in Sudan before. The Korean and Sudanese PM received full education about chemotherapy many times. The treatment team included some or all of the followings: Korean PM in Sudan, Sudanese PM or parasitology expert, health workers for assistance of health education, reception and medication, and driver. Health workers delivered the drug actually under the control of medical doctors. All health workers are White Nile MOH officials, and they were educated drug treatment and schistosomiasis control from Sudanese and Korean parasitology experts. Within each school or village, treatment was provided on the basis of the baseline prevalence of schistosomiasis in sentinel primary schools.
The strategy for controlling morbidity due to schistosomiasis was based on chemotherapy using praziquantel. The chemotherapy strategy followed the schistosomiasis control protocol of Sudanese government, which was a modified treatment strategy of WHO recommendation [14]. In this project, the major target group was the primary school students. We treated all primary school children in Al Jabalain locality at the 1st year of the project regardless to the levels of infection. At the second year, mass chemotherapy was implemented in schools with more than 10% of urogenital schistosomiasis at baseline survey. The WHO dose pole was used to determine dosage (40 mg/kg) for school children [14]. In villages, mass treatment was conducted where the prevalence of urogenital schistosomiasis was more than 30% at baseline.
As shown in time bar of mass chemotherapy, drug administration was done over the year (Fig. 4). During treatment period, health workers went to schools in the morning, and they administered praziquantel for treatment of school children directly. If children have been registered but were absent on the day of treatment, the health workers visited again and treated them as long as it was within the treatment period. If they were absent because they were sick, the health workers should have waited until they got better.
In villages, residents were treated by health workers with assistance of village leaders. Treatment could be completed in a few days, but should take no longer than 1 week for schools or villages.
The total population of the pilot area was the subjects of health education (482,902), however the major target population was 78,615 primary school students. The health education team included one Korean PM in Sudan, one Sudanese PM or Parasitology experts, two health workers for assistance of health education, two health workers for questionnaire survey (not always) and one driver. Basically, Sudanese PM or parasitology expert educated teachers, students, and village residents about schistosomiasis and health improvement. However, trained health workers taught students when multiple programs of health education were held at the same time.
Korean and Sudanese PM, and Korean experts developed the materials for health education such as presentation file, leaflets, posters, and brochures to provide knowledge related to schistosomiasis and health improvement. The developed materials were used to educate school students and teachers as well as village residents. Also they developed the "Manual for Schistosomiasis Control" to standardize the schistosomiasis control activities and to improve the capacity building of Sudanese staffs participating this project (S1 Appendix).
For health education of school teachers, teachers were gathered at one school or community education center of each unit, and Sudanese PM or parasitology experts educated them about schistosomiasis and health improvements using beam project. Though this training, school teachers would be health education experts who can deliver sufficient knowledge of schistosomiasis to students. School teachers received the certificate of completeness of health education, and the education took 5 hours totally (S1 Video).
For education of school students and village residents, Sudanese PM or parasitology experts educated them in the project areas to provide sufficient knowledge on schistosomiasis and promote self-care health awareness for prevention of re-infection. Sudanese PM or parasitology experts educated the school children or village residents using education materials at the playground or classroom in the school for 1 hour, and sometimes beam projector was used for health education (S2 Video). In some cases, health education was combined with administration of praziquantel or questionnaire survey.
Every year, KOICA team planned the health education schedules for students through the attending school period. Basically, the team visited all primary schools at least one time during the project period. The schools of high prevalence (more than 30%) tried to visit 1–2 times per year for 2 years. Village residents were invited to school playground in case of more than 30% of prevalence of the sentinel school. School teachers were educated one time per year for 2 years (Fig. 4).
To evaluate the effects of health education, the semi-structured questionnaire survey was conducted at baseline and follow-up survey, which was consisted of information related to schistosomiasis. Teachers or Sudanese health workers thoroughly explained the questions to the students and obtained the consents, and then received the answer sheets.
Since schistosomiasis is one of the water-transmitted helminthiases, the supply of safe and clean water is very important for prevention of infection. This project included one large scale filtered drinking water supply facility at a high endemic village in the pilot area. We suggested selection criteria to determine the construction site; high prevalence of schistosomiasis, availability of drinking water supply facility, and feasibility of construction. According to the selection criteria, one drinking water supply facility was constructed at Al Hidaib village in Al Jabalain unit using the water from the main stream of the White Nile River.
The facility was constructed for 6 months from March to August 2010 (Fig. 4). It installed a slow sand filtration system using pipelines, pumps, and tanks for sedimentation, filtration and storage (Fig. 5). The gravel, sand and charcoals as a filter should be cleaned about 4 times annually depending on the degree of contamination, and maintenance manual should be prepared and be kept near the facility. The water storage capacity was up to 59,290 L/day, and 3,000 residents were lived in Al Hidaib village in Al Jabalain unit.
After starting the implementation of this project, the impacts and results of activities were monitored regularly by the project team itself or specific teams. During the follow-up evaluation, we conducted the parasitology examination, questionnaire survey from school students about health education, questionnaires survey from residents about the drinking water supply facility, questionnaires survey from school students about drug administration, and questionnaire survey from teachers about health education.
The evaluation team was composed of Korean and Sudanese PM or parasitology experts, Sudanese Federal and White Nile State MOH officials, and KOICA staffs. The schools for follow-up evaluation were basically selected by randomization procedure as follows. Simple random sampling was used to select primary schools per unit using the lottery method [15]. This involved the listing names of all baseline-surveyed primary schools in each unit on small slips of paper, then after a thorough mixing of the names, 1–3 schools were selected one by one. In one school, we collected the urine and stool samples around 80–250 from the 1st, 3rd and 5th graders, but not more than 300. If the sample size of one school was less than 100, all students were collected. Al Hidaib village in Al Jabalain unit was selected to evaluate the effects of drinking water supply facility from residents. The follow-up survey for evaluation of the prevalence of S. haematobium and S. mansoni infection was carried out 12 primary schools in Al Jabalain locality from September to December 2010 (Tables 3 and 4, Fig. 4).
They visited the selected schools and villages to measure the project indicators. The major indicators were the prevalence of S. haematobium infection, number of students and village residents participating in a health education activity (at least once), the percentage of participants with increased awareness of schistosomiasis after completing health education, number of village residents using the water supply facility, and its impacts.
SPSS version 15.0 software (SPSS Inc., San Diego, CA, USA) was used to analyze the experimental data. The differences in prevalence of S. haematobium and S. mansoni infection and questionnaire survey about health education between baseline and follow-up were tested using the logistic regression analysis. Odds ratio (OR) and 95% confidence interval (CI) were calculated. Categorical variables were tested using the χ2 test. A P-value of <0.05 was considered statistically significant.
For the baseline survey of schistosomiaiss in the Al Jabalain locality of White Nile State, we selected 61 primary schools according to the selection method (Fig. 3, Table 2). Three to twenty schools were selected each unit due to the total number of primary school students. A total of 10,671 urine samples and 9,107 stool samples were collected from students in grades 1, 3, and 5 at primary schools.
As a result of parasitological examination, the prevalence rates for S. haematobium and S. mansoni were 13.8% (1,473 of 10,671 cases) and 1.0% (92 of 9,107 cases), respectively. The prevalence of S. haematobium varied from 6.1% to 40.2% and S. mansoni 0% to 4.9% by unit of the Al Jabalain locality. The prevalence rate of S. haematobium was high at Joda (40.2%), Al Jabalain (21.5%), and Kenana (18.6%) units but low in Rabak (6.1%) and Jazeera Aba (7.2%) units. The S. mansoni prevalence rate was 0% at Joda and Kenana units, but 0.8% at Rabak unit and 4.9% at Assalaya unit.
In the 12 sentinel schools, the prevalence rate of S. haematobium and S. mansoni was 28.5% and 0.4% respectively at the baseline survey (Tables 3 and 4). Follow-up survey 6–9 months after the intervention detected 13.5% prevalence rate for S. haematobium and 0.9% for S. mansoni. The prevalence of S. haematobium infection after praziquantel treatment had decreased significantly in comparison to the baseline prevalence (OR = 0.391, 95% CI = 0.331–0.462, P<0.001). The S. haematobium infection rates in the Jazeera Aba (95% CI = 0.060–0.287), Assalaya (95% CI = 0.079–0.336), Joda (95% CI = 0.134–0.298), Rabak (95% CI = 0.284–0.644), and Kenana (95% CI = 0.385–0.769) units were significantly reduced, but no significant change in the prevalence was observed in the Al Jabalain unit (95% CI = 0.568–1.237).
To characterize the reduction patterns of urinary schistosomiasis, the changes in prevalence according to school were analyzed. The prevalence was prominently reduced at 11 schools among 12 examined schools after implementation of the project, whereas the prevalence was significantly increased at Khour Ajwal unit at the follow-up survey (OR = 2.232, 95% CI = 1.242–4.009, P = 0.007). As shown in Fig. 6, reduction in the prevalence of S. haematobium infection was greater at schools with 20–25% baseline infection rates (Tayba boys and Assalaya East schools) than at those with <15% baseline infection rates (Al Khansa girls and Al Wifak boys schools) or >40% baseline infection rates (Abu Baker Alsiddig and Al Hidaib schools). The prevalence reduction of the other schools was between 34.4% and 79.1% than that of Khour Ajwal school. According to the sex, the reduction rate of S. haematobium prevalence was greater in girls (61.8%) than that in boys (50.0%).
The overall prevalence of S. mansoni infection was unchanged significantly (OR = 2.536, 95% CI = 0.933–6.891, P = 0.068) (Table 4). The prevalence was not significantly changed by unit and sex.
Korean and Sudanese PM or parasitology experts developed the materials for health education such as leaflets, billboard, brochure, poster, and power point files to provide knowledge related to schistosomiasis and health improvement. The developed materials were used to educate school students and teachers as well as village residents. Also they developed the manual for schistosomiasis control (see supporting information) to improve the capacity building of Sudanese staffs participating this project and educated them in case of necessary, from time to time, and individually or collectively, as well as standardized the schistosomiasis control activities.
Education about public health and schistosomiasis control was provided to enhance awareness and self-protection. In case of primary schools in the pilot area, health education team visited one time per year basically. The cumulative number of students who received health education for >1 hour was 121,418; thus, this number was estimated 1.5 times of all students in the pilot area, and the coverage rate of each unit during the whole project period was ranged from 131% to 175% of total number of students of each unit (Table 5).
Health education for village residents depended on the baseline prevalence of S. haematobium infection. In high prevalence villages (more than 30% at baseline), health education was performed one time per year at the primary schools or village center independently during the project period, however the village residents who lived in less than 30% prevalence attended health education held in the regional primary schools. Thus, the cumulative number of 75,021 village residents received health education for more than 1 hour, which represented 19% of all village residents in the Al Jabalain locality (Table 5). Health education for primary school teachers were held every year for two years, and they were educated intensively for more than 5 hours for them to work as a public health educator to students about schistosomiasis.
The impact of health education was measured by questionnaires received from 244 students at the baseline and 217 students at follow-up assessments (Table 6). The percentages of correct answers to questions about the infection mode (by contact with contaminated water, 95% CI = 1.4323–3.292) and the impacts on health (blocking children learning well, 95% CI = 1.039–2.545; causing severe diseases if untreated for a long time, 95% CI = 2.554–6.495) in schistosomiasis significantly increased at the follow-up compared with at the baseline survey. Also the percentage of incorrect answerers to questions about the infection mode of schistosomiasis (by eating unwashed food, 95% CI = 0.331–0.932) was significantly decreased after health education. However, the other questions about clinical manifestations, infection model, preventive methods and the effects of body related to schitosomiasis were not changed significantly after health education intervention (Table 6).
We also checked the basal schistosomiasis-related knowledge of school health teachers before health education (Table 7). They answered higher correct responses about infection mode, health hazards, side effects, and preventive method of schistosomiasis in comparison to those of students. They thought that schools are good sites for deworming programs, and health education is the best method for prevention of schistosomiasis. However, teachers had some worries about the minor side effects after administration of praziquantel.
Praziquantel was administered to 98,586 students and 111,795 village residents in the Al Jabalain locality (Table 8). In case of school students, whole primary school children in the pilot area received adequate amount of praziquantel tablets at the first year of the project, except absentees, ill or refusing children taking medicine. At the second year, mass chemotherapy was done at schools more than 10% prevalence of urogenital schistosomiasis at baseline survey. About 60–70% of school children attended the mass treatment each year, thus cumulative coverage rates of drug administration were higher than 100% at all units, beside Al Jabalain unit (86%, Table 8).
We also delivered the drug to village residents to control the schistosomiasis. In villages, drug administration was performed at a school base, according to the baseline prevalence of S. haematobium infection. Basically, drugs were administered to the village residents who attended the health education. Mass chemotherapy for all village residents was implemented in cases that the prevalence of baseline at each school (village) was more than 30%. Thus, the overall cumulative coverage rate of drug administration of village residents was 28%. However, the coverage rates of each unit varied from 2% to 69%, although the number of visiting-village was different from the baseline prevalence of S. haematobium infection.
Questionnaires about the effects of drug treatment were received from 202 students. Fifty-nine percent of them answered improvement in their physical condition after praziquantel therapy. The improvement included feeling more active and healthy (72.7%) and the disappearance of hematuria (18.2%). After praziquantel medication, 10.9% of students complained of adverse effects, such as abdominal pain 50.0%, vomiting 18.2%, and dizziness 13.6% (Table 9).
To supply safe drinking water, which is free from cercariae of Schistosoma spp., a system of filtered drinking water supply facility was constructed at Al Hidaib village of Al Jabalain unit according to the selection criteria. Al Hidaib was one of the highest prevalence of schistosomiasis among Al Jabalain locality, White Nile State. Also there were no clean water supply facilities in Al Jabalain unit, and it is easy to obtain the water from the Nile River.
We checked the impacts and usage patterns of a drinking water supply facility by questionnaire survey from residents of Al Hidaib village 9 months after construction. As shown in Table 10, 94.3% of answerers used the water supply facility. They obtained water almost every day (86.3% of user), and they usually collected water less than 30 liters per person in a day. Thus, the coverage rate of a drinking water facility at Al Hidaib was about 66% (2000 among 3000 residents) in case of use 30 liter per day per person. The major use of collected water was drinking, making food, or washing clothes or body. The important one was that 81.4% of answerers were felt good health, and 55.9% of them answered saving the time to collect water. However, about 25% of user of water supply facility also collected water from the White Nile River.
The study was the partial results of comprehensive schistosomiasis control project implemented by KOICA in White Nile State, Sudan. The project reduced the prevalence of S. haematobium infection from 28.5% to 13.5% after 6–9 months and increased the awareness of students about the seriousness of schistosomiasis and the preventive measures by health education. This project also showed the importance of clean water supply facility for the control of schistosomiasis. Finally, approximately 200,000 students, teachers, or village residents have been benefited by providing either praziquantel medication, health education, or filtered water supply.
Mass chemotherapy with praziquantel has been employed by many national control programs for schistosomiasis [16]–[20]. Praziquantel treatment reduced the prevalence of S. haematobium infection by 87% [18] and reduced infection rates from 54.2% and 51.7% in high- and low- transmission seasons to 30.3% and 1.8%, respectively [19]. In this project, primary school students were the principal target group for treatment and education due to feasibility and the greater benefit of reducing the infection burden in children [21]. The overall S. haematobium prevalence of our data (28.5%) was similar to that reported previously for White Nile Province (21.4%) [6]. However, the S. mansoni prevalence (0.4%) was significantly lower than that reported previously (10.1%) [6]. It can be explained that detection of fecal S. mansoni by egg count may be affected by day-to-day and intra-specimen variation, resulting in the low sensitivity of single stool smear [22]. At 6–9 months post-treatment, the rate of S. haematobium infection had been reduced by 52.6% (from 28.5% to 13.5%), which was similar to that during the high-transmission seasons in Mozambique [19]. Surprisingly, the S. haematobium prevalence at Khour Ajwal, a village in the project area, increased from 27.3% to 45.6% during the post-intervention monitoring period. It may be due to low coverage of drug administration, poor environmental condition, low attendance rate of children at school and continuous reinfection. Collectively, the survey findings suggest that S. haematobium infection is more prevalent than S. mansoni infection in Sudan, and repeated exposure to schistosome. Therefore, wide-scale and sustainable chemotherapy is essential to successfully control schistosomiasis.
By analyzing the prevalence after intervention, we found a few interesting characteristics. First, the reduction rate of prevalence was higher in girls than in boys, which meant that boys were more likely to be infected and were reinfected more frequently due to more outdoor activity such as swimming [23]. Second, the highest reduction in prevalence was observed among the moderately infected group (20–25% at baseline) compared with the highly infected group (>40%). Generally, reduction rates in the >60% egg-positive groups were 69.4–87% at 3–24 weeks post-treatment [18], [20]. However, the reduction rates for the highly prevalent groups were around 50% in this study, lower than those previously reported [18], [20] and moderately infected group of this study. These data demonstrated that reinfection occurred more frequently in the highly endemic village than in the moderately endemic one. Third, praziquantel therapy produced little change in the S. mansoni prevalence compared with the reduction observed for S. haematobium. Similar findings were reported in Niger [20]. Previous reports and our data indicate that a single dose of praziquantel (40 mg/kg) is less effective in curing S. mansoni infection than that of S. haematobium. Development of resistance to praziquantel is also of concern [24] and should be studied by a well-designed investigation.
In addition to chemotherapy, encouraging self-protective health behaviors may reduce the transmission and reinfection rates of schistosomiasis [25]. Through this project, students received health education 1.8–2.8 times through the project period, thus cumulative coverage rates of health education were 131–175% at a unit level. Kenana unit showed the highest cumulative rate of health education (175%), the reduction rate of S. haematobium prevalence was 37.0%, which was below the mean reduction rate 52.6%. On the contrary, Joda unit revealed the lowest cumulative health education rate (13%), but the reduction rate was 69.6%. Thus, there was no correlation between health education coverage rates and reductions in prevalence, which was in agreement with a previous report [26]. In the present study, students showed a general increase in awareness of the health hazard of schistosomiasis and the preventive measures according to the questionnaire survey. However, this knowledge was not connected to prevent the infection in Khour Ajwal village in Al Jabalain unit, so continuous health education is needed at the school level. For this purpose, this project conducted intensive health education course for school health teachers. According to the questionnaire results from health teachers, they emphasized the health education and the role of schools to prevent the reinfection of schistosomiasis. Thus, the intensive program may contribute teachers to enforce the role as a health educator to students. To obtain better outcomes, health education models need to consider social representation and illness experience besides scientific knowledge in order to increase knowledge of schistosomiasis transmission and prevention [27].
The clean and safe water supply is important in reducing the rates of ascariasis, diarrhea, schistosomiasis, and trachoma [28,29]. Villages in the pilot area were settled along the White Nile River; however few facilities provided clean water and sanitation services. Therefore, village residents had no choice but to use water directly from the river. The KOICA project provided a large filtered water supply facility in Al Hidaib village in the Al Jabalain unit. The follow-up survey revealed a reduction in the prevalence at Al Hidaib, from 46.1% to 20.5% of urogenital schistosomiasis, whereas that in the neighboring village in the same unit, Khour Ajwal, increased from 27.3% to 45.6% during the same period. The natural environment and lifestyle of the two neighboring villages were very similar, and mass praziquantel therapy and health education were implemented in both. However, filtered clean water was supplied only to Al Hidaib village residents. This finding may reflect the importance of providing a supply of clean water to reduce reinfection rates as part of efforts to control schistosomiasis. According to the questionnaire survey among Al Hidaib residents, 71.2% of the village residents used only water collected from this water supply facility, and 81.4% of answerers felt better health condition after construction of a drinking water supply facility. Thus, clean drinking water supply facility in Al Hidaib village will contribute to improving the health and life quality of the residents and children by prevention of waterborne diseases including schistosomiasis.
Even though the project achieved the proposed objectives, it was also limited in several ways. First, since the project duration was not long enough to eliminate schistosomiasis, there was a potential for the resurgence of the disease. Second, molluscicides were not used for snail control due to the effects of environment, and it may be one of the factors of high reinfection. Third, this project targeted primarily students; thus, village residents and preschoolers in the villages were not covered fully. Fourth, a filtered water supply facility was provided to only one village, and this was not sufficient to affect the whole target area. Fifth, we used urine sedimentation method. This method is very good to detect the eggs of S. haematobium in urine, however the eggs might attach to the wall of tubes or dispensers to lower its sensitivity [30]. Due to these limitations, the results should be interpreted as reflective of the effects of mass praziquantel therapy and health education on the control of urinary schistosomiasis.
In conclusion, mass praziquantel therapy should be the core activity of efforts to control schistosomiasis in the field, which must be integrated with health education and clean water supply. The control activity should last long enough to reduce reinfection and keep low prevalence in the project area.
|
10.1371/journal.ppat.1006118 | Multiple Acid Sensors Control Helicobacter pylori Colonization of the Stomach | Helicobacter pylori’s ability to respond to environmental cues in the stomach is integral to its survival. By directly visualizing H. pylori swimming behavior when encountering a microscopic gradient consisting of the repellent acid and attractant urea, we found that H. pylori is able to simultaneously detect both signals, and its response depends on the magnitudes of the individual signals. By testing for the bacteria’s response to a pure acid gradient, we discovered that the chemoreceptors TlpA and TlpD are each independent acid sensors. They enable H. pylori to respond to and escape from increases in hydrogen ion concentration near 100 nanomolar. TlpD also mediates attraction to basic pH, a response dampened by another chemoreceptor TlpB. H. pylori mutants lacking both TlpA and TlpD (ΔtlpAD) are unable to sense acid and are defective in establishing colonization in the murine stomach. However, blocking acid production in the stomach with omeprazole rescues ΔtlpAD’s colonization defect. We used 3D confocal microscopy to determine how acid blockade affects the distribution of H. pylori in the stomach. We found that stomach acid controls not only the overall bacterial density, but also the microscopic distribution of bacteria that colonize the epithelium deep in the gastric glands. In omeprazole treated animals, bacterial abundance is increased in the antral glands, and gland colonization range is extended to the corpus. Our findings indicate that H. pylori has evolved at least two independent receptors capable of detecting acid gradients, allowing not only survival in the stomach, but also controlling the interaction of the bacteria with the epithelium.
| Helicobacter pylori is a bacterium that chronically infects the stomachs of 50% of humans, and infection can lead to serious diseases like peptic ulcers and stomach cancer. To survive, H. pylori rapidly senses acid and swims away to the protective mucus layer covering the stomach surface. The bacteria also burrow deep into the glands of the stomach through their narrow fissures and channels, and live in close contact with the cells lining the stomach. We report here that two H. pylori chemoreceptors, TlpA and TlpD, are the dominant acid sensors enabling H. pylori to discern and respond to minute changes in acid levels. H. pylori mutants lacking both TlpA and TlpD are unable to sense acid and are severely impaired in their survival in the murine stomach. By treating animals with omeprazole, a drug that blocks acid production, we restored the ability of the acid-sensor mutant to survive in the stomach. In addition, we found that blocking stomach acid production extended the range, distribution, and density of H. pylori living deep in the gastric glands. Our study provides new insights into H. pylori’s acid sensing machinery and how manipulation of acid gradients controls H. pylori’s localization and survival in the stomach.
| Helicobacter pylori is a bacterium that has co-evolved with humans since the origin of the human species [1, 2]. This intimate association with the human host has allowed it to develop a number of survival strategies to persist in one of the most challenging environments in the human body—the stomach. H. pylori’s survival relies on its ability to avoid the microbicidal effects of stomach acid. H. pylori can withstand acidic conditions for short periods of time due to its urease enzyme which degrades urea into ammonium and bicarbonate to buffer the cytoplasm and periplasm [3–6]. Another important strategy is to utilize chemotaxis to locate and swim to the gastric epithelium where the pH is near neutral due to the overlying protective mucus layer.
The bacteria colonize a narrow niche within 25 microns of the surface of the gastric epithelium where they are either found actively swimming in the mucus or directly adhered to epithelial cells [7, 8]. The attached bacteria utilize virulence factors to obtain essential nutrients from the host and replicate on the cell surface to form cell-associated microcolonies [9, 10]. We recently reported that a subpopulation of cell-associated H. pylori is found deep in the antral glands in direct contact with gastric progenitor cells and stem cells [11]. The gland-associated H. pylori induce the expansion and proliferation of stem cells, alter stem cell gene expression, and lead to gland hyperplasia [11]. We hypothesize that the factors that control H. pylori’s ability to colonize the gastric glands will help explain H. pylori’s ability to persist long-term in the stomach and to cause gastric diseases.
Despite living in the stomach, H. pylori is not an acidophile and swims away from hydrochloric acid (HCl). The acid secreted into the stomach lumen by parietal cells in the corpus forms gradients that keep the bacteria close to the gastric epithelium [8]. Previous studies have reported that the chemoreceptor responsible for sensing HCl as a repellent is TlpB [12, 13]. This chemoreceptor has been shown to detect auto-inducer 2 as a repellent as well [14]. We recently reported that TlpB can also sense chemoattractants, since it is a high affinity chemoreceptor for urea that allows H. pylori to sensitively detect and swim towards urea emanating from the gastric epithelium [15].
In this study, we initially proceeded to investigate how H. pylori may be sensing both a repellent and an attractant through TlpB. Using a previously developed videomicroscopy method that visualizes and films bacterial chemotactic responses to chemical gradients in real time [7, 15], we discovered that H. pylori mutants lacking TlpB (ΔtlpB) are not defective in detecting and swimming away from HCl gradients. Instead, we identified TlpA and TlpD as independent acid sensors with different sensitivities to HCl. We also found that TlpD allows H. pylori to chemotax towards less acidic and even basic pH environments, and this response is dampened by TlpB.
Using a murine model of infection in the stomach, we discovered that the double mutant lacking TlpA and TlpD (ΔtlpAD) is about 100-fold defective in its ability to colonize the stomach compared to wild-type H. pylori. However, treatment with the proton-pump inhibitor omeprazole raises the gastric pH and partially rescues the ΔtlpAD mutant’s defect, allowing it to reach significantly higher bacterial numbers in the stomach. We also observed that omeprazole treatment promotes wild-type H. pylori’s colonization of the gastric glands and extends its range of glandular colonization from the antrum into the glands of the corpus. Despite the higher loads of ΔtlpAD H. pylori in omeprazole-treated animals, the mutant is unable to colonize the gastric glands to the same levels as wild-type, suggesting that these two chemoreceptors are important in establishing colonization deep in the gastric glands.
Our study has identified two new roles for H. pylori’s chemoreceptors as acid sensors and demonstrate that H. pylori’s ability to detect and respond to the acid gradient is important for its localization within the stomach, its interaction with the glandular epithelium, and its survival in vivo.
H. pylori encounters many chemical gradients in the stomach that serve as cues to identify microniches that are safe for colonization. It must be able to integrate both repellents and attractants to optimize its chemotactic response. Since H. pylori’s TlpB chemoreceptor has been reported to detect HCl as a repellent [12, 13], and we found that it detects urea as a high-sensitivity attractant [15], we wondered how H. pylori would respond if it were simultaneously exposed to both urea and HCl gradients emanating from one point source.
We used our previously described microgradient chemotaxis assay [7, 15] to record the swimming responses of a live culture of H. pylori (strain PMSS1) exposed to a microscopic gradient of a mixture of 50 mM HCl and 5 mM urea emanating from the tip of a microinjection needle. Prior to the assay, H. pylori were grown in Brucella broth with 10% FBS (BB10), pH 6.7–6.8 to an OD600 of 0.3. During culture, bacterial urease depletes urea from the surrounding medium (S1A Fig) and the pH remains at a range of 6.6–6.74 (S1B Fig). A 270 μl volume of media with motile bacteria is placed onto a coverslip chamber and a microinjection needle is then rapidly inserted via a micromanipulator into the viewing field to produce a microscopic gradient of chemoeffectors. Using this method, we observed that H. pylori are attracted to the urea in the gradient until they reach a boundary approximately 60 micrometers away from the needle tip (Fig 1A). Within this boundary is a zone of clearance avoided by the bacteria, representing a threshold concentration of acid that acts as a chemorepellant. When bacteria swim into this zone of clearance, they quickly stop, reverse direction, and swim away (S1 Movie). As a negative control, we tested a chemotaxis null mutant ΔcheW H. pylori for its response to the same mixed microgradient. The density of swimming ΔcheW H. pylori remained constant throughout the field (Fig 1A) consistent with this mutant’s inability to respond to either the attractant or the repellent. These observations indicate that H. pylori is capable of simultaneously sensing and integrating multiple signals from one point source. To test whether the chemotactic response is dependent on the magnitude of these chemical gradients, we altered the concentration of acid within the needle while maintaining the same concentration of urea. We found that, indeed, the bacteria respond by increasing the distance of the boundary between repulsion and attraction and the needle tip as the concentration of HCl increases (S2 Fig).
Because H. pylori abundantly expresses a urease enzyme that degrades urea into ammonia to buffer its cytoplasmic and periplasmic pH, we wondered whether urease may play a role in acid sensing. We first tested a urease mutant (ΔureAB) for its response to a solution of urea and HCl. Unlike the WT H. pylori culture, which is depleted of urea due to urease activity, the ΔureAB culture contains levels of urea in the medium comparable to fresh media (S1A). We observed that ΔureAB cleared the field of view completely (S3A Fig), indicating that it was unable to sense urea because the surrounding urea in the medium interferes with chemotactic sensing as we had previously reported [15] but was able to sense acid. We further tested for ΔureAB’s response to a solution of 100 mM HCl in water (no urea). We found that it cleared the field of view like wild-type (S3B Fig). These results suggest that urease activity is not required for acid sensing.
Next, we tested ΔtlpB H. pylori’s response to this mixed solution. As the urea and acid sensor, we predicted that this mutant would respond to neither compound and that its swimming behavior would be like that of ΔcheW H. pylori. We were surprised to find instead that ΔtlpB H. pylori rapidly swims away and clears from the field of view like ΔureAB (Fig 1B, S3A Fig, and S1 Movie). This observation that the bacteria swim away from the urea and acid microgradient suggests that the ΔtlpB mutant is unable to detect urea as we had previously reported [15] but is able to detect HCl as a repellent.
We proceeded to test ΔtlpB’s response to a solution of HCl in water without urea. We found that, indeed, ΔtlpB efficiently swims away from an acid gradient (Fig 1C and S2 Movie). We plotted the bacterial density in the viewing field every fifth of a second before and after introduction of the needle tip injecting HCl, and by ten seconds post-exposure to the acid gradient, the ΔtlpB mutant has mostly cleared from the field of view (Fig 1C). Interestingly, we note from the clearance curves (Fig 1C) that ΔtlpB clears from the field of view faster than wild-type upon acid exposure. This result suggests that TlpB does affect H. pylori’s response to acid since ΔtlpB has response kinetics different from wild-type. To verify that the ΔtlpB response to acid is not specific to a particular H. pylori strain, we constructed ΔtlpB mutants in other strains of H. pylori and tested them in the same assay. We found that the ΔtlpB mutants in all strains tested (strains G27, SS1, PMSS1 (a second independent clone), and 7.13) are still able to respond to an acid gradient like their wild-type counterparts by swimming away upon exposure to HCl and maintain the fast clearing phenotype (S4 Fig). This result indicates that ΔtlpB’s response to HCl is not strain specific. For the rest of the experiments we use H. pylori PMSS1 as the strain background.
Our finding that ΔtlpB responds to an HCl gradient suggests that other chemoreceptors may be acid sensors. We tested the response of mutants lacking each of the other three chemoreceptors, ΔtlpA, ΔtlpC, ΔtlpD, by videomicroscopy in the same assay. To simplify the comparisons of the movies showing the escape from acid of different mutants, we graphed the percent of bacteria present in the viewing field 4 seconds before and 10 seconds after exposure to the acid gradient (Fig 2A). We found that each mutant was still able to respond and swim away from acid, indicating that either there is a novel unidentified chemoreceptor that senses acid or there are multiple chemoreceptors that can function redundantly in sensing acid.
To test the hypothesis that there may be redundancy in acid-sensing chemoreceptors, we made mutants lacking two of the four chemoreceptors in all possible combinations. We discovered that of all the six combinations of chemoreceptor knock-outs, only the mutant lacking both TlpA and TlpD (ΔtlpAD H. pylori) lost the ability to respond to HCl gradients (Fig 2B and S3 Movie). This result indicates that TlpA and TlpD function as acid sensors, each capable of compensating for the loss of the other in acid sensing.
Furthermore, when we tested for the response of ΔtlpAD H. pylori to the solution containing a mixture of urea and HCl, we observed that the bacteria are attracted towards the needle tip with no zone of clearance (Fig 2C). This result indicates that the ΔtlpAD mutant can detect and respond to urea as an attractant (through TlpB), but is unable to detect the HCl as a repellent. This result suggests that TlpB is not sufficient to sense acid in our assay, and TlpA and TlpD are acid sensors that detect acid gradients.
We next asked whether the response to acid is a response to low pH or a response specifically to HCl. To determine this, we tested the responses of WT, ΔtlpA, ΔtlpB, ΔtlpD, and ΔtlpAD to sulfuric acid (H2SO4) and phosphoric acid (H3PO4). As with HCl, we observed that H. pylori responds to both of these acids as repellents and requires either TlpA or TlpD for the response (S5 Fig). This result suggests that these two chemoreceptors allow H. pylori to sense and escape from conditions of low pH.
Several chemoeffectors have been described for TlpA and TlpD. The TlpA receptor was reported to mediate attraction to arginine and bicarbonate [16]. The TlpD chemoreceptor was reported to mediate repulsion from conditions that induce low-energy in the bacterium [17] or conditions that promote oxidative stress [18, 19]. These conditions may be triggered by low pH. One key difference between TlpA and TlpD is the location of these chemoreceptors within the bacterium. The TlpD chemoreceptor is cytoplasmic as it lacks a transmembrane domain [17] whereas the sensing domain of TlpA is periplasmic like that of the TlpB and TlpC chemoreceptors. Therefore, TlpD may be sensing changes in external pH indirectly. Given the difference in the location of these two chemoreceptors, we wondered if the two receptors may have distinguishable responses to the same HCl gradient.
To assess if there is a difference in sensitivity for detecting HCl between the two chemoreceptors, we determined the threshold concentration of HCl necessary to elicit a response (arbitrarily defined as a 60 microns clearance zone from the point source) for each mutant. We empirically determined that 25 mM HCl loaded in the microinjection needle is the minimum concentration required to repel wild-type H. pylori from the point source by 60 microns (Fig 3). ΔtlpA H. pylori, like wild-type, responds to a minimal effective concentration of 25 mM HCl while ΔtlpD H. pylori requires 50 mM HCl to respond (Fig 3). It is worth noting, however, that the actual minimal effective concentration of HCl that H. pylori is capable of detecting is markedly below 25 mM, since only a minute volume of HCl on the order of 1 picoliter/minute is injected into the solution, and the culture medium surrounding the bacteria consists of Brucella broth with 10% fetal bovine serum, which has a buffering capacity that we empirically determined to be about 4,000 fold greater than water (S6 Fig). Thus the gradient of free hydrogen ions would drop rapidly away from the needle tip. Since we observe the bacteria responding 60 microns away from the point source, the change in HCl concentration that the bacteria can sense is substantially less than the concentration of the solution in the needle. Indeed, when we expose the bacteria to a gradient starting from 15 mM HCl at the needle tip, we also see the bacteria being repelled but at approximately 30 microns away from the point source (S7 Fig).
Since H. pylori is exposed to both urea and HCl in vivo, we wondered whether urea sensing would affect the sensitivities of TlpA and TlpD in sensing acid. We exposed ΔtlpA and ΔtlpD to a gradient of a mixture of 5 mM urea and 50 mM HCl in the microinjection needle. We observed that both mutants are attracted to the urea but form a zone of clearance as they sense HCl like wild-type H. pylori. However, with urea present we could elucidate small differences in sensitivity because the attraction to urea highlights TlpD’s ability to detect lower concentrations of acid than TlpA. This is illustrated by the ΔtlpA mutant remaining at a farther distance away from the acid point source at the needle tip compared to ΔtlpD (S8 Fig). This result indicates that urea sensing does not alter the sensitivity hierarchy of TlpD and TlpA in acid sensing.
As chemotaxis is a result of detecting a change in the concentration of a chemical, we next sought to determine the smallest change in hydrogen ion concentration ([H+]) that H. pylori is able to detect and respond to. By comparing the pH of the Brucella broth culture medium in which the bacteria are grown with the pH of a buffer solution that elicits a repulsion response, we determined the smallest difference in [H+] that H. pylori is capable of detecting. To more accurately control the range of [H+] concentrations experienced by the bacteria, we loaded the microinjection needle with 1M phosphate buffer solutions with defined pHs made by combining different proportions of dibasic and monobasic phosphate solutions. We measured the pH of the Brucella broth medium that the bacteria are grown in to be pH 6.7 (S1B Fig). We then tested wild-type bacteria’s response to buffer solutions ranging from pH 4.2 to 7.1.
As shown in Fig 4A, buffers with pH 4.2, 6.0, 6.3, and 6.5 all elicited an escape response in the microgradient assay. The increase in [H+] between the solution released from the microinjection needle and the culture medium was approximately 63 μM, 800 nM, 360 nM, and 110 nM, respectively. For the lower pH measurements between pH 4 and pH 6 we also confirmed our results using citrate buffer, which buffers acidic pH more effectively than phosphate buffer (S9 Fig). Phosphate buffer solutions of pH 6.6 and 7.0, which locally changed the [H+] by + 35 nM or -100 nM, did not elicit a response, indicating that the differences were below the limit of detection of the receptors. Unexpectedly, a buffer solution with pH 7.1, (about -120nM change in [H+]) elicited an opposite response with bacteria attracted and forming a swarm at the needle tip (Fig 4B and S4 Movie). Thus, we discovered that H. pylori can sense and respond to small changes in both acidic and basic pH when the change in [H+] is greater than 100 nM.
To further characterize wild-type H. pylori’s response to more basic environments, we also tested phosphate buffers with pH 7.25 and 9.2. We found that H. pylori is attracted to basic pH solutions (S10 Fig). By testing the single chemoreceptor mutants, we identified TlpD as the necessary receptor for attraction to higher pH (Fig 5 and S5 Movie). The other receptors, including TlpA, were not necessary. To confirm that the TlpD-dependent attraction is a response to high pH rather than to a specific basic solution, we tested for H. pylori’s response to 40 mM sodium hydroxide (pH 12.6). We also observed H. pylori being attracted to sodium hydroxide, and the response was dependent on TlpD (S11 Fig). Taken together these results suggest that TlpD mediates both repulsion from lower pH and attraction to higher pH while TlpA only detects and mediates repulsion to lower pH.
We wondered whether the TlpB-dependent defect in acid sensing previously reported and our observation of faster clearance rates of ΔtlpB mutants compared to wild-type may be due to an effect of TlpB in modulating the sensitivity to pH of the other chemoreceptors. In order to see if lacking TlpB enhances the sensitivity to acid, we tested for the response of ΔtlpB H. pylori to low concentrations of HCl and found that the minimal effective concentration of HCl in the needle required to elicit an escape response is the same as that of wild-type at 25 mM HCl (Fig 3). We also tested the responses of the TlpA and TlpD chemoreceptors in the absence of TlpB by determining the sensitivities of the ΔtlpAB and ΔtlpBD double mutants to acid gradients. We found them to have the same sensitivity to HCl as ΔtlpA and ΔtlpD, respectively (S7 Fig).
However, when we tested for H. pylori’s response to higher pH, we noted that ΔtlpB’s attraction was always more pronounced than wild-type with each higher pH solution (S10 Fig); the attraction of ΔtlpB was faster, and the concentration of bacteria around the needle tip was denser than that of wild-type. This observation led us to hypothesize that TlpB may be modulating H. pylori's response to basic pH. We tested for the mutant’s response to a gradient formed by phosphate buffer at pH 7.0 where wild-type H. pylori does not show an attraction. The ΔtlpB mutant is still able to sense and respond to the solution at pH 7.0 (-100nM change in [H+]) (S10 Fig). This result suggests that lacking TlpB increases the bacteria's sensitivity and attraction to higher pH, and thus TlpB may be modulating the pH responses mediated through TlpD.
The mechanism by which TlpB influences pH sensing is unknown. We tested whether TlpB alters the expression levels of TlpA or TlpD, thereby resulting in a change in acid response kinetics. We performed a Western blot to assess expression levels of the chemoreceptors in the single knock-out mutants and ΔtlpAD H. pylori. We found that the expression levels of the remaining chemoreceptors were not affected by the loss of the TlpA, TlpB, or TlpD receptors (S12 Fig).
Our data show that TlpB is not sufficient for sensing acidic pH, but it does alter the kinetics of H. pylori’s response to changes in pH. Taken together, our data suggest that the mechanism of pH sensing in H. pylori is complex, involving multiple chemoreceptors. The difference in sensitivity between TlpA and TlpD and the role of TlpB in modulating pH responses allows H. pylori to discern even small changes in local pH gradients in the nanomolar range. This may have critical implications for H. pylori’s survival in the stomach.
Given the importance of avoiding the acidic lumen of the stomach environment, we next investigated whether ΔtlpAD H. pylori would be able to colonize the stomachs of mice. We infected C57Bl/6 mice with either wild-type H. pylori strain PMSS1 or the isogenic mutants ΔtlpA, ΔtlpD, or ΔtlpAD. After two weeks of infection, a time point in which wild-type H. pylori has established stable colonization [7, 20], we harvested the stomachs and assessed colonization densities by colony-forming units (CFU) per gram of stomach tissue. We found that the ΔtlpA mutant was not defective in colonization with similar CFU counts as wild-type bacteria. However, the ΔtlpD mutant had a 10-fold defect as had been previously reported [20], and the ΔtlpAD H. pylori mutant was about 100-fold defective in establishing colonization (Fig 6). We hypothesized that without the ability to respond to acid gradients in the stomach, the ΔtlpAD H. pylori is deficient in avoiding the microbicidal HCl. We wondered if pharmacologic inhibition of acid secretion in the stomach would improve ΔtlpAD H. pylori’s survival.
We maintained two experimental groups to test this hypothesis. One group of animals was treated with the proton-pump inhibitor omeprazole for three days prior to infection to raise the gastric pH before infection (S13 Fig), and the animals continued to receive omeprazole throughout the course of the 2-week infection. This experimental group allowed us to observe how the gastric pH experienced by the bacteria upon entering the stomach affects the bacteria’s survival in the stomach. The second group of animals was infected with H. pylori for a week before treatment with omeprazole throughout the final week of infection. This group represents the more common clinical scenario in which humans with an established H. pylori infection may take proton-pump inhibitors like omeprazole to treat conditions such as gastroesophageal reflux. We found that treatment of animals with omeprazole prior to or after the establishment of infection partially rescued ΔtlpAD H. pylori’s ability to colonize the stomach (Fig 6), while it did not affect the total number of wild-type bacteria.
We previously reported that in this murine model of infection and in humans the majority of the bacteria reside in the overlying mucus layer. However, a subpopulation of H. pylori can be found deep in the gastric glands adhered to epithelia cells that make up the mid-glandular proliferative zone [7, 11]. We wondered whether acid sensing would affect not only the overall fitness of the bacteria establishing colonization in the stomach, but also their ability to reach and colonize the epithelial surface of the gastric glands. To investigate gastric gland colonization, we used quantitative 3D confocal microscopy to determine the number of bacteria growing as microcolonies within the gastric glands in the antrum and corpus regions of the stomach. One of the main distinguishing features between antrum and corpus is the presence of acid-secreting parietal cell in the corpus glands.
In control animals with normal acid secretion, the wild-type strain is found mostly colonizing the antral glands and the transition zone between the antrum and corpus [7, 11] (S14A and S14B Fig) rather than the corpus glands (S14C and S14D Fig). We found that in only two out of seven animals infected with ΔtlpAD H. pylori, the mutant was able to colonize the antral glands but at significantly lower densities than wild-type H. pylori (Fig 7A, 7C and 7E). This result suggests that chemotaxis through TlpB and TlpC still allows some bacteria to reach the epithelium and colonize the glands. We know that one such signal is the chemoattractant urea sensed through TlpB [15]. The defect in gland colonization of the mutant, however, may be attributed to low bacterial numbers in both the mucus and the glands as it has a 100-fold defect in overall bacterial load compared to wild-type H. pylori (Fig 6).
We performed a similar analysis of gastric gland colonization of ΔtlpAD H. pylori in animals treated with omeprazole after infection, since in these conditions the bacterial load in the stomach was comparable to that of wild-type (Fig 6). When we analyzed the antral glands of infected animals treated with omeprazole one week post-infection, we noted that the bacterial density of both wild-type and ΔtlpAD H. pylori in the antral glands increased significantly compared to those in control animals (Fig 7B, 7D and 7E). Despite the increase in gland colonization, ΔtlpAD H. pylori does not reach the levels of wild-type in the antral glands of omeprazole-treated animals (Fig 7E). These results suggest that omeprazole promotes the colonization of H. pylori in the gastric glands of the antrum, and chemotaxis through TlpA and TlpD is important for proper colonization of the antral glands.
Using quantitative 3D-confocal microscopy, we also found that loss of stomach acidity through omeprazole treatment allowed H. pylori to extend its colonization to the corpus glands (Fig 8). Control animals normally do not have H. pylori in corpus glands (Fig 8A). In our analysis, only one out of the seven control animals had detectable levels of H. pylori in the corpus glands (Fig 8C). In the corpus, the gland-associated bacteria seen after omeprazole treatment were mainly concentrated in the neck region of the glands in the proliferative zone (Fig 8B), and also were seen in close proximity to parietal cells (Fig 8D). This result suggests that the acid in the stomach restricts H. pylori gland-colonization to the antral glands but that an increase in gastric pH allows H. pylori to extend its range to also colonize the epithelium of the corpus glands. We did not find ΔtlpAD H. pylori in the corpus glands of omeprazole-treated animals. These results from the omeprazole treatment experiments suggest that gland colonization is distinct from colonization of the mucus, and sensing through TlpA and TlpD may be necessary for localizing to and/or persisting in the corpus glands.
Our study has revealed that H. pylori evolved two independent chemoreceptors, TlpA and TlpD, capable of sensing and rapidly responding to acid gradients. The fact that H. pylori devotes at least half of its chemoreceptor repertoire towards acid sensing underscores the importance of this function for H. pylori’s survival in the stomach.
Despite the same overall function, we found that there is a difference in the sensitivity of TlpA versus TlpD in detecting HCl. The cytoplasmic TlpD chemoreceptor appears to be more sensitive than the periplasmic TlpA chemoreceptor, and it is able to sense both lower pH as a repellent and higher pH as an attractant. H. pylori may have evolved a more sensitive acid sensor in its cytoplasm as it would be crucial to detect even small changes in cytoplasmic pH to ensure homeostasis.
TlpD is the only chemoreceptor of the four that H. pylori possesses that is thought to be a soluble protein located in the bacterial cytoplasm and inner membrane [17]. It has been reported to be an energy sensor causing H. pylori to repel from electron transport inhibitors and other low energy-inducing environments [17]. Currently it is not known what are TlpD’s specific ligands and how TlpD may be detecting these ligands. TlpD contains a C-terminal chemoreceptor zinc binding domain (CZB) of unknown function [21] and does not contain a PAS domain commonly found in other chemoreceptors, making it challenging to identify the ligands that TlpD directly bind. It is possible that changes in intracellular pH may affect intracellular metabolism and the energy state of the bacterium, which would link TlpD’s ability to sense acid to its role as an “energy sensor,” and consequently, an indirect sensor of acidic pH. Interestingly, it was recently published that changes to particular protein interactions with TlpD or the metabolic state of the bacterium alters the localization of TlpD in H. pylori [19]. This and another study, also reported that the sensing mechanism of TlpD may be linked to oxidative stress and iron limitation [18, 19]. Specifically, upon metabolic stress or iron limitation, TlpD changes its localization from the poles of the bacterium to the cytoplasm. Also, in the absence of the interacting partners recently identified, TlpD changes its localization from the poles to the cytoplasm [19]. It is unclear, however, how TlpD’s localization to the poles or the cytoplasm affects its ability to sense and respond to particular chemoeffectors. There is no direct evidence that TlpD directly interacts or forms chemoreceptor arrays with TlpA, TlpB or TlpC to signal, and in fact TlpD has been shown to localize to the poles and be capable of transducing a chemotactic signal to the flagellar machinery in the absence of all other chemoreceptors [22]. Structural analysis of the chemoreceptor may reveal more insights into the mechanism of sensing through TlpD.
TlpA has been reported to sense arginine and bicarbonate as attractants in H. pylori strains 26695 [16] and 700392 [23], but its mechanism of sensing has not been further characterized. It has been proposed that chemotaxis towards a bicarbonate gradient in vivo may help H. pylori navigate towards a safe niche on the gastric epithelium where it is protected against the acid in the lumen. It is unclear whether TlpA is sensing bicarbonate directly or the basic pH that bicarbonate creates. While our studies here have shown that H. pylori is attracted to gradients of basic pH, we identify TlpD, not TlpA, as the chemoreceptor necessary for this attraction. It is possible that TlpA may be detecting bicarbonate specifically as a ligand rather than an increase in pH or we did not test the optimal basic pH that TlpA may be detecting. Further studies are needed to determine how TlpA is sensing acidic pH and whether TlpA may be involved in sensing basic pH as well.
Our data surprisingly revealed that loss of TlpB does not result in defects in escaping from an acid gradient in the microgradient assay. However, we noted that the response kinetics of ΔtlpB H. pylori differed from that of wild-type (Fig 1C). This suggests that lacking TlpB does have an effect on H. pylori’s ability to respond to acid even though TlpB is neither necessary nor sufficient to detect acid gradients. The difference between these recent results and previously published findings may be attributed to differences in the assays used to assess ΔtlpB’s responses to acid.
Our assay generates and maintains a constant microscopic gradient from a point source and records the chemotactic behavior immediately after exposure to a gradient [7, 15]. We observe bacterial responses within seconds, and the response is sustained for long periods of time (we have tested it for as long as 10 minutes). However, our assay does not change the overall pH of the medium containing the bacteria because we inject minute amounts of acid (with a flow rate on the order of picoliters per minute) at very low pressure through a femtotip needle. A previously described assay that places H. pylori in an acidic environment for several minutes describes the formation of a barrier of bacteria at a region where the pH has been altered [12–14]. In this barrier assay the bacterial culture is infused with a 100 mM solution of HCl, which exceeds the buffering capacity of Brucella broth, and when mixed would decrease the pH of the bacterial culture from about pH 6.7 to pH 4.76, as we determined empirically (S6 Fig). Thus, the media is likely acidified when the chemotactic behavior is observed at about 5 minutes after exposure.
Another assay used that has implicated TlpB’s role in acid sensing is a video chemotaxis assay where the bacteria are placed in chemical solutions of interest, such as acidic solutions [13, 14, 17, 24]. In this assay, the bacteria are not exposed to a chemical gradient, but the assay measures motility behavioral differences in the presence or absence of a chemical as stops/sec or reversals/sec. An increase in stops or reversals per second indicates the detection of a repellent, but since a gradient is absent, directed movement cannot be assayed. An increase in reversals may therefore also represent the loss of an attractant or a change in the functioning of the chemosensory signal transduction pathway. The conditions in these two assays differ drastically from that of the microgradient assay with regard to the shape and steepness of the acid gradient as well as the time scale in which chemotactic responses are assessed. These other two assays may better assess H. pylori’s response when immersed in a more homogenous low pH environment such as the stomach lumen as opposed to an environment where there is a steep acidic pH gradient, such as across the gastric mucus layer. Perhaps TlpB is important for acid sensing in the acidic lumen.
We do detect differences in the speed and sensitivity of acid sensing by the other receptors when TlpB is missing. Perhaps TlpB plays a role in acid sensing, for example, by changing the sensitivity of acid sensors at different baseline pH conditions or in other spatial and temporal conditions not replicated in the microgradient assay. The altered response kinetics of ΔtlpB compared to wild-type may be the resultant response from the integration of the remaining three receptors. Lacking TlpB may be changing the way the other receptors function in sensing acid as well as how other signals present in their environment alter responses to acid. While there is no evidence that TlpB and TlpD directly interact, chemoreceptors are known to form mixed arrays that transduce signaling responses to the flagellar motor [25, 26]. It is possible that TlpB may directly interact with TlpD under certain conditions to dampen its response to basic pH. One intriguing speculation is that sensation and responses to urea through TlpB may also be coupled to sensing cytoplasmic pH through TlpD, through a yet unknown mechanism. Our observation that lacking TlpB enhances H. pylori’s sensitivity to detecting higher pH suggests this possibility.
Our experiments elucidating acid sensing were all performed under conditions where urea was absent in the culture medium. Urea was only present when we deliberately introduced it in solution with HCl in the needle to investigate H. pylori’s response to multiple signals. This allowed us to pinpoint specifically the acid-sensing functions of the chemoreceptors. However, in vivo, H. pylori is exposed to both HCl and urea. While urea is a potent chemoattractant sensed through TlpB, which may have an effect on pH sensing, it is also the substrate for H. pylori’s highly expressed urease enzyme, whose activity certainly affects pH sensing. As a neutrophile, urease buffering activity is essential for H. pylori’s survival in vivo. In vitro studies have shown that H. pylori is able to survive under pH 1 conditions for several hours if the bacteria are in the presence of urea [27]. Many studies have been conducted to elucidate the intricate mechanism of how acidic conditions in the environment trigger a cascade of events resulting in an increase in cytoplasmic pH while the external pH remains acidic. Upon exposure to acidic pH conditions, urease assembles into a complex with a proton-gated urea channel embedded in the inner membrane [28–30]. Urea enters through this proton-gated channel to reach cytoplasmic urease where it optimally functions to degrade urea into ammonia and bicarbonate thereby raising the pH [31].
Based on the findings in these prior studies, we predict that in the presence of urea and urease H. pylori will be less responsive to an acid gradient in our microgradient assay. Future studies will need to integrate the role of urease and urea into acid sensing since it alters the intracellular and extracellular pH. It would be interesting to determine how urease activity might affect the sensitivities of TlpA and TlpD to acidic pH. This may explain why TlpD is more sensitive to pH since it detects changes in the cytoplasm.
Our newer data show that TlpA and TlpD are the primary acid sensors in H. pylori that allow immediate response to HCl gradients and are important for stomach colonization in the presence of gastric acid. We find that a mutant lacking both chemoreceptors has a severe defect in establishing colonization in the murine stomach. It has been previously reported that TlpD is important for H. pylori survival and proliferation in the antrum [20]. We also found that ΔtlpD H. pylori has an approximate 10-fold defect in overall colonization of the stomach at two weeks post-infection (Fig 6). This could be due to the loss of chemotaxis towards other signals important for survival in addition to acid sensing. Indeed, we report here that TlpD also mediates attraction to environments of higher pH. However, TlpD does contribute to acid sensing in vivo, since deletion of TlpA has no effect on colonization in vivo, yet double deletion of TlpA and TlpD markedly worsens the colonization defect of the ΔtlpD H. pylori mutant (Fig 6). Interestingly, it has also been reported that tlpD is one of the most upregulated genes when H. pylori is exposed to acidic conditions in vivo [32]. These results indicate that chemotaxis through TlpD is most crucial for establishing colonization of the stomach, but that TlpA can compensate for the loss of acid-sensing through TlpD.
Furthermore, we discovered that sensing through TlpA and TlpD is important for localizing properly to the gastric glands and that blocking acid secretion with omeprazole changes the distribution of gland-associated H. pylori in the stomach. We previously reported that bacteria growing as microcolonies in the gastric glands lead to inflammation in the regions of gland colonization. In addition, gland-associated H. pylori locally activate and induce proliferation of the stem cells in the infected glands leading to hyperplasia [11]. A previous investigation of the effects of omeprazole on H. felis distribution in the mouse stomach reported that acid suppression extends the colonization range of the bacteria into the corpus glands [33]. A similar finding was reported for mice infected with H. pylori and treated with omeprazole where the stomach glands were qualitatively found to contain less bacteria in the antrum than in the corpus [34]. Pulsed omeprazole dosing in gerbils has also been shown to alter the orientation of H. pylori within the mucus layer relative to the gastric epithelium, which could promote bacterial clearance by driving the bacteria closer to the lumen or perhaps improving the efficiency of antibiotics [35]. Our results support these previous findings that reducing gastric acidity through proton pump inhibitor treatment changes bacterial density and distribution in the gastric glands, extending their range within this niche. Our results further suggest that targeting chemosensation through TlpA and TlpD would interfere with overall stomach and gland colonization.
H. pylori infection in humans most often leads to chronic inflammation in the antrum (antral predominant gastritis). People with antral gastritis usually have no symptoms but may develop pyloric or duodenal ulcers as a consequence of the infection. People at risk for gastric adenocarcinoma are different in that they develop an anatomical pattern of gastritis that extends into the corpus. Corpus gastritis is associated with the loss of parietal cells, low acid secretion, high gastrin production and gastric atrophy (multifocal atrophic gastritis) [36]. These differences in the anatomical localization of the inflammatory changes in H. pylori infection has been ascribed to physiological differences in individuals, but could also reflect the anatomical site of infection of gland-associated H. pylori. That is, extension of the distribution of gland-associated bacteria from the antrum to the corpus glands may precede and contribute to the inflammatory and hyperplastic changes that lead to tissue pathology [11].
In our experimental system, gastric colonization of the corpus is promoted by proton pump inhibitor (PPI) treatment, suggesting that this could be a precursor step towards the development of multifocal atrophic gastritis and pre-neoplasia. Experiments in Mongolian gerbils, for example, have shown that PPI treatment can promote the development of adenocarcinoma [37]. In humans, the role of PPIs in atrophic gastritis remains controversial, but it is recommended that patients considered for long-term PPI therapy first be tested and treated for H. pylori infection [38, 39]. Our study, and others, supports this clinical recommendation and also suggests that targeting pH sensing through TlpA and TlpD may be an effective way of disrupting H. pylori colonization in the stomach.
The previously published wild-type H. pylori strain PMSS1 [40], ΔureAB and single knock-outs chemoreceptor mutants [15] were used for all experiments in this study except those strains noted in S4 Fig. H. pylori strains were either grown on Columbia blood agar plates or in Brucella broth supplemented with 10% fetal bovine serum (BB10) at 37°C, 10% CO2, as described previously [41]. The ΔtlpAB, ΔtlpAC, ΔtlpAD, ΔtlpBC, ΔtlpBD, and ΔtlpCD PMSS1 isogenic mutants were constructed by natural transformation with genomic DNA from the relevant single knock-out chemoreceptor mutant made in strain PMSS1. The PMSS1 chemoreceptor mutants were verified by immunoblotting using an antibody that recognizes a conserved domain in all four chemoreceptors, courtesy of K. Ottemann [42].
All solutions were diluted or dissolved in sterile, distilled, and deionized water. Urea and sodium hydroxide solutions were made by dissolving ultra-pure urea or sodium hydroxide tablets into sterile, distilled and deionized water. One molar hydrochloric acid (Sigma Aldrich, Saint Louis, MO) was diluted into sterile, distilled and deionized water to create various concentrations of hydrochloric acid solutions to test. To obtain phosphate buffered solutions with different pHs, 1 M monobasic sodium phosphate (NaH2PO4) pH 4.1 and 0.5 M dibasic sodium phosphate (Na2HPO4) pH 9.0 were combined in various proportions until desired pH is achieved and confirmed by pH meter. To obtain citrate buffered solutions with different pHs, 1 M citric acid (C6H8O7 • H2O) pH 2.0 and 1 M sodium citrate (C6H5O7Na3 • 2H2O) pH 8.6 were combined in various proportions until desired pH is achieved and confirmed by pH meter. The buffering capacity and corresponding pH of various concentrations of HCl in BB10 was determined empirically (S6 Fig).
H. pylori cultures used for the assay were made by subculturing from a 16-hour overnight Brucella broth + 10% FBS (BB10) culture with a starting OD600 of 0.15. The subcultures (also in BB10) were grown for 6 hours until they reach an OD600 of 0.3 before using in the microgradient assay. The pH of bacterial cultures prior to the microgradient assay (spent media) was measured at a mean of 6.65 +/- 0.04 SD (S1B Fig). These were not different from the pH of BB10 kept in the incubator without bacteria measured at 6.6, or freshly made BB10 (pH 6.6–6.8). The bacterial culture media for all strains tested, except ΔureAB, contain no detectable urea at the time of the assay due to bacterial urease activity during the subculture (S1A Fig). This was determined by measuring the concentration of urea using the QuantiChrom Urea Assay Kit (BioAssay Systems), a colorimetric assay that quantifies urea directly with a detection level of 13μM. Background was determined by quantifying urea in urease-treated BB10. Twenty five milliliters of fresh BB10 (predicted to contain 2–4 mM urea) were incubated with 300 units of Jack Bean urease for 2 hours at 25°C. This amount of urease is calculated to liberate 0.3 millimole of ammonia in 1 min at 25°C, and therefore is sufficient to hydrolyze all the urea in the media in about one minute. To ensure complete urea degradation, we allowed the reaction to proceed for 2 hours before boiling the urease-treated media for 10 minutes to inactivate urease and filtering the media through a 0.2 μm filter to remove particulates.
Two hundred seventy microliters of the subculture were placed into the center of a glass-bottom 35-mm dish (MatTek) contained by a ring of vacuum grease. The dish was placed above a 32x objective of a Zeiss Axiovert-35 inverted microscope equipped with phase-contrast optics and a heated stage (37°C). A Hammamatsu C2400 video charge-coupled-device (CCD) camera was used to record via an Argus-20 image processor onto Quicktime at 30 frames per second. A Femtotip II microinjection micropipette (Eppendorf) containing 8 μl of the test solution was inserted into or removed from the viewing field using a micromanipulator (Eppendorf 5171) [7]. To create a microscopic gradient (microgradient) in the bacterial culture, a compensation pressure of 30 hPa was applied via the Eppendorf transjector 5246 to maintain a constant flow at 0.372 picoliters/minute from the tip. This pressure was determined empirically and selected because it did not physically affect the bacteria swimming near the micropipette tip while generating a stable gradient within the viewing field.
H. pylori chemotactic responses were quantified from phase contrast video microscopy movies recorded at 30 frames per second using ImageJ software version 1.46r. Movies were analyzed starting from 4 to 10 seconds before micropipette insertion (pre-injection) to 10–30 seconds after micropipette insertion (post-injection). After background subtraction and contrast adjustment, movies were then subsequently quantified or bacterial traces were obtained for visual depiction of swimming behavior in response to the gradient. For quantification, 0.2 second segments of the movies (6 frames) were combined into Z-projections to generate traces of moving bacteria. Motile bacteria were detected by specifying a size between fifty to three hundred pixels and circularity values between 0.1 to 0.5 (non-motile particles will be more circular than motile bacteria, which will be elliptical traces). These parameters were set in the Analyze Particle function in ImageJ and the output provided the number of motile bacteria per frame. The number of motile bacteria per frame is normalized to the average number of motile bacteria prior to needle entry. Responses are shown as either scatter plots, which display the percent bacteria per field from 4–5 seconds pre-injection to 10–15 seconds post-injection, or bar graphs, which display the percent bacteria per field at 4 second pre-injection and 10 seconds post-injection. Statistical significance between the percent bacteria per field pre-injection versus percent bacteria per field post-injection in the bar graphs or of the overall response in the scatter plots was assessed using a 2-way repeated measures ANOVA. To produce the images that depict the bacterial swimming behavior, longer traces were generated to clearly show the response. The time stated in the pre-injection panel indicates the first frame until 0 seconds used to generate the trace (i.e. -1.5 sec means frames from 1.5 seconds prior to needle entry to 0 second when the needle enters were used to generate a 1.5 trace representing the pre-injection swimming behavior). The range of time stated in the post-injection panels indicates the frames used to generate the trace (i.e. 18.5–20 sec means the frames from 18.5 seconds to 20 seconds post-injection were used to generate a 1.5 trace representing the post-injection swimming behavior).
H. pylori cells were harvested from blood plates and lysed with 1x SDS sample buffer. Lysates were boiled for 5 minutes and then separated on a 10% SDS-PAGE gel. After transfer onto nitrocellulose membranes, H. pylori TlpA, TlpB, TlpC, TlpD were detected by blotting with rabbit anti-tlpA22, an antibody that recognizes a conserved domain in all 4 chemoreceptors, courtesy of K. Ottemann [42], followed by a goat anti-rabbit Alexa Fluor 660. UreA, the small subunit of the abundantly expressed urease protein, was used as a loading control and detected with a mouse anti-UreA antibody followed by a goat anti-mouse Alexa Fluor 800. The blot was then scanned with a Licor-Odyssey scanner at 700 and 800 nanometers.
Experiments involving animals were performed in accordance with NIH guidelines, the Animal Welfare Act, and US federal law. All animal experiments were approved by the Stanford University Administrative Panel on Laboratory Animal Care (APLAC) and overseen by the Institutional Animal Care and Use Committee (IACUC) under Protocol ID 9677. Animals were euthanized by CO2 asphyxiation followed by cervical dislocation.
6 week-old female C57BL/6J mice were purchased from the Jackson Laboratory (Bar Harbor, ME). Animals were infected intraorally by allowing the animals to drink a 5 microliter suspension containing 108 CFU of H. pylori grown in BB10 from a pipette tip [7]. One cohort of animals was treated with omeprazole at 400 μmol/kg/day via drinking water for seven days after seven days of infection. A second cohort of animals was treated with omeprazole (400 μmol/kg/day) for three days prior to infection and then maintained on omeprazole for the duration of the two-week infection. The control cohort was given untreated water throughout the course of the two week infection. Animals were sacrificed at 2 weeks post-infection by CO2 asphyxiation. The stomach was harvested with the forestomach removed and discarded, opened via the lesser curvature, and laid flat onto filter paper. Luminal content was removed and the stomach was divided into halves that spanned the corpus to the antrum. One half of the stomach was weighed and mechanically homogenized for 30 seconds in 1 ml Brucella broth. Homogenized stomachs were serially diluted and plated for CFU counts. The data represent the number of CFU/gram of stomach. Bars represent the geometric mean. Statistical significance in the recovered bacterial load between strains was assessed by a Mann Whitney test. The efficacy of the omeprazole treatment was determined via a separate experiment where 5 mice were administered omeprazole in their drinking water for 3 days prior to sacrifice and 5 mice were given water for 3 days prior to sacrifice (S13 Fig). To measure the pH of the stomach, the stomach was cut along the lesser curvature and splayed open. The stomach tissue along with its contents were placed at the bottom of a test tube (with opening wide enough to fit a pH probe) with the stomach lumen facing up. One milliliter of sterile water was added to the stomach in the tube. The pH was then measured using a pH meter.
The other half of the stomach was fixed in a 2% paraformaldehyde and stained for confocal microscopy as previously described [7, 11]. A custom made rabbit anti-H. pylori PMSS1 antibody and chicken anti-rabbit Alexa Fluor 488 antibody was used. DAPI (4 =, 6-diamidino-2-phenylindole) and Alexa Fluor 594-phalloidin (Molecular Probes) were used for visualization of the nuclei and actin cytoskeleton. Samples were imaged with a Zeiss LSM 700 confocal microscope and z-stacks were reconstructed into 3D images using Volocity software (Improvision). Number of bacteria per gland was determined with measurement functions in Volocity.
All microgradient assay data comparing percent bacteria per field at 4 seconds pre-injection to percent bacteria per field at 10 seconds post-injection as well as the scatter plots of percent bacteria per field over time were analyzed via a 2-way repeated measures ANOVA test in the GraphPad Prism 7 software program. The p-value for the scatter plots indicates the significance of the time by group interaction via a 2-way repeated measures ANOVA. For animal experiments, statistical significance was assessed via a Mann Whitney test. Center values are geometric means and error bars represent standard deviation (s.d.). n indicates the number of movies per condition or the number of animals used. NS indicates no statistical significance, * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.
|
10.1371/journal.ppat.1001329 | Dissection of the Influenza A Virus Endocytic Routes Reveals Macropinocytosis as an Alternative Entry Pathway | Influenza A virus (IAV) enters host cells upon binding of its hemagglutinin glycoprotein to sialylated host cell receptors. Whereas dynamin-dependent, clathrin-mediated endocytosis (CME) is generally considered as the IAV infection pathway, some observations suggest the occurrence of an as yet uncharacterized alternative entry route. By manipulating entry parameters we established experimental conditions that allow the separate analysis of dynamin-dependent and -independent entry of IAV. Whereas entry of IAV in phosphate-buffered saline could be completely inhibited by dynasore, a specific inhibitor of dynamin, a dynasore-insensitive entry pathway became functional in the presence of fetal calf serum. This finding was confirmed with the use of small interfering RNAs targeting dynamin-2. In the presence of serum, both IAV entry pathways were operational. Under these conditions entry could be fully blocked by combined treatment with dynasore and the amiloride derivative EIPA, the hallmark inhibitor of macropinocytosis, whereas either drug alone had no effect. The sensitivity of the dynamin-independent entry pathway to inhibitors or dominant-negative mutants affecting actomyosin dynamics as well as to a number of specific inhibitors of growth factor receptor tyrosine kinases and downstream effectors thereof all point to the involvement of macropinocytosis in IAV entry. Consistently, IAV particles and soluble FITC-dextran were shown to co-localize in cells in the same vesicles. Thus, in addition to the classical dynamin-dependent, clathrin-mediated endocytosis pathway, IAV enters host cells by a dynamin-independent route that has all the characteristics of macropinocytosis.
| Attachment to and entry into a host cell are the first crucial steps in establishing a successful virus infection and critical factors in determining host cell and species tropism. Influenza A virus (IAV) attaches to host cells by binding of its major surface protein, hemagglutinin, to sialic acids that are omnipresent on the glycolipids and glycoproteins exposed on the surfaces of cells. IAV subsequently enters cells of birds and a wide variety of mammals via receptor-mediated endocytosis using clathrin as well as via (an) alternative uncharacterized route(s). The elucidation of the endocytic pathways taken by IAV has been hampered by their apparent redundancy in establishing a productive infection. By manipulating the entry conditions we have established experimental settings that allow the separate analysis of dynamin-dependent (including clathrin-mediated endocytosis) and independent entry of IAV. Collectively, our results indicate macropinocytosis, the main route for the non-selective uptake of extracellular fluid by cells, as an alternative IAV entry route. As the dynamin-dependent and -independent IAV entry routes are redundant and independent, their separate manipulation was crucial for the identification and characterization of the alternative IAV entry route. A similar strategy might be applicable to the study of endocytic pathways taken by other viruses.
| Influenza A virus (IAV) is an enveloped, segmented negative-strand RNA virus infecting a wide variety of birds and mammals. As its first step in infection IAV attaches to host cells by the binding of its major surface protein, the hemagglutinin (HA), to sialic acids, which are omnipresent on the glycolipids and glycoproteins exposed on the surfaces of cells. Where the structural requirements for this interaction have been studied in great detail, much less is known about whether and how the attachment to specific sialylated receptors (e.g. to N-linked glycoproteins, O-linked glycoproteins or gangliosides or even to specific receptors within these groups) affects the subsequent endocytic steps. Obviously, knowledge about the repertoire of endocytic pathways that can successfully be used by IAV will increase our insights into cell and species tropism of IAV. In turn, this will contribute to our understanding of the requirements for the generation of novel viruses with pandemic potential that can arise by exchange of RNA segments between currently circulating human serotypes and an animal virus during occasional co-infection in a human or an animal host.
Clathrin mediated endocytosis (CME) has for long been identified and studied as the major route of IAV cell entry [1], [2] and is, by far, the best characterized endocytic pathway. Evidence obtained from live cell imaging has revealed the de novo formation of clathrin-coated pits at the site of virus attachment [3] and the requirement for the adapter protein epsin 1, but not eps15, in this process [4]. Still, specific transmembrane receptors linking viral entry to epsin 1 or to other adapters have not been identified although a recent study performed in CHO cells indicated the specific requirement for N-linked glycoproteins in IAV entry [5].
Some recent papers provided indications for the utilization of alternative entry pathways by IAV. Studies in which CME was obstructed by pharmacological or genetic intervention indicated the ability of IAV to enter host cells via alternative endocytic routes [4], [6], [7]. Also live cell imaging revealed the simultaneous availability of entry routes involving non-coated as well as clathrin-coated pits [4]. However, this alternative IAV entry route has not been characterized in any detail and requirements for any specificity in receptor usage apart from the need for the proper sialic acid moiety have not been established.
During the past decades quite a variety of endocytic pathways have been identified in eukaryotic cells [8], [9], [10]. Their occurrence, abundance and mechanistic details appear to vary between cell types, tissues and species and their utilization by viruses as a route of entry makes them an important factor in host and cell-type permissiveness for infection [11], [12]. Besides by CME, different viruses have been shown to enter cells via caveolae, macropinocytosis or other, less well described, routes [11], [12]. Most often, the selection of a specific endocytic route is linked to the utilization of a specific receptor that facilitates traveling via that particular route. Nevertheless, many receptors allow flexibility by their capacity to enter through multiple pathways. For IAV, an additional level of complexity to the dissection of potential entry routes is added by the apparent lack of an IAV-specific protein receptor.
A full experimental characterization of the IAV entry pathways will benefit from separation of the IAV entry pathways into routes that can be studied independently. Whereas co-localization with clathrin is an established marker for endocytosis via this route, the complete lack of unique markers for macropinosomes or most other endocytic compartments [13], [14] complicates such studies. Furthermore, crucial to any study concerning endocytic pathways is the abundantly documented fact that such pathways are highly dependent on experimental cell culture conditions [15]–[19]. Pathways that are constitutive in one cell type may be absent or inducible by specific experimental conditions in other cell types. Moreover, the manipulation of specific endocytic pathways may result in up or down regulation of other specific pathways. Here we have established entry assay conditions that allow dissecting cell entry of IAV into a dynamin-dependent (DYNA-DEP) and a dynamin-independent (DYNA-IND) component. Dynamin is a large GTPase forming multimeric assemblies around the neck of newly formed endocytic vesicles. GTP hydrolysis is required for pinching off of the vesicles [20]. Whereas CME is completely dependent on dynamin, several other endocytic routes do not require dynamin [21]. We performed an extensive characterization of the dynamin-independent IAV entry route using pharmacological inhibitors as well as by expressing dominant-negative mutants and applying siRNA induced gene silencing as tools. Taken together the results identify a pathway that closely resembles macropinocytosis as a novel entry pathway for IAV.
To identify and characterize potential non-CME entry routes taken by IAV, we adapted a luciferase reporter assay [22] to enable the quantitative determination of infection or entry by measuring the activity of secreted Gaussia luciferase. Twentyfour hours prior to infection HeLa cells were transfected with a plasmid (pHH-Gluc) allowing constitutive synthesis (driven by the human PolI promoter) of a negative strand viral RNA (vRNA) encoding a Gaussia luciferase under control of the untranslated regions (UTRs) of the NP segment of Influenza A/WSN/33 (H1N1) (hereafter called IAV-WSN) NP segment. Upon IAV infection, the combined expression of the viral polymerase subunits and NP will drive transcription of luciferase mRNA from the negative strand vRNA and subsequent synthesis of Gaussia luciferase.
A dose-response curve demonstrating the applicability of the assay to inhibitor screening (Fig. 1A) was obtained for Bafilomycin A1 (BafA1), a known inhibitor of IAV entry [23]. BafA1 acts upon the vacuolar-type H(+)-ATPase, thus preventing endosomal acidification and thereby trapping IAV in peri-nuclear immature endosomes with a lumenal pH that does not permit viral membrane fusion. Remarkably, dynasore, a small molecule inhibitor of the GTPase dynamin 2 that is crucial for endocytic vesicle formation in clathrin- and caveolin-mediated endocytosis [8] as well as in a poorly described clathrin- and caveolin-independent endocytic pathway [8], [19], did not give significant inhibition (Fig. 1B).
BafA1 specifically inhibits IAV during the entry phase as demonstrated in Fig. 1C. The continuous presence of 10 nM BafA1 (added to the cells 1 hr prior to infection) for 16 hrs completely prevents infection. In contrast the addition of BafA1 at 1 hr or 2 hrs post infection resulted in high levels of luciferase activity (again measured at 16 hrs p.i.) that were 63% or 90% respectively of the control to which no BafA1 was added, indicating that entry was essentially completed within 2 hrs. The last bar of Fig. 1C shows that the inhibition by BafA1 is reversible as withdrawal of the inhibitor after 2 hrs resulted in high levels of infection. The specific effect of BafA1 on IAV entry was confirmed by confocal microscopy demonstrating that BafA1, as expected, traps IAV particles in a peri-nuclear location, presumably in non-acidified endosomes (Fig. 1D). BafA1 was subsequently exploited to establish a specific IAV entry assay (hereafter further referred to as the Gluc-entry assay). HeLa cells transfected with pHH-Gluc were inoculated with IAV at a range of MOIs and incubated for 2 hrs after which the entry medium was replaced by complete growth medium containing 10% FCS and 10 nM BafA1 to prevent any further entry of virus. Entry was indirectly quantified by determination of luciferase activity after further incubation for 14 hrs demonstrating a quantitative correlation between infection dose and luciferase activity across a wide range of MOIs (Fig. 1E). The indirect Gluc-entry assay was next tested for its capacity to examine the effects of inhibitors on IAV entry. Dynasore or BafA1 (Fig. 1F) were included in the medium (DMEM containing 10% FCS) during entry (the first 2 h of infection) and were removed when the inoculum was replaced by growth medium containing BafA1. Concentrations up to 80 µM dynasore did not inhibit entry which is in agreement with the result shown in Fig. 1B. In contrast, 1.25 nM BafA1 already inhibited entry for more than 60% (Fig. 1F). As a control, dynasore was also added at 2 hrs post infection to analyze whether the drug affected IAV replication during the post entry phase. As expected, 80 µM dynasore did not significantly inhibit IAV replication when present from 2 to 16 hrs p.i. (Fig. 1F). Thus, with the Gluc-entry assay we can study the effect of specific inhibitors on IAV entry in a quantitative manner, at least as long as the inhibitors do not irreversibly affect IAV replication during the post entry phase.
Furthermore, the lack of inhibition of IAV entry by dynasore demonstrates that under these experimental conditions IAV is able to enter cells via a pathway that is fully redundant to any dynamin-dependent (DYNA-DEP) entry route, including the classical CME pathway. Also when IAV travels via this novel dynamin-independent (DYNA-IND) route, IAV apparently enters via low pH compartments as entry is fully sensitive to BafA1.
As factors present in serum are known for their potential to induce specific endocytic pathways, we further explored the conditions required for the novel DYNA-IND IAV entry pathway (using the Gluc-entry assay) by inoculating cells in PBS in the presence of increasing concentrations of fetal calf serum (FCS). Whereas dynasore completely inhibited entry in PBS, inclusion of 5% and 10% FCS resulted in increasing levels of dynasore resistant entry (Fig. 2A), suggesting the existence of a serum-inducible DYNA-IND IAV entry pathway. This effect was not caused by inactivation of dynasore during the experiment as vesicular stomatitis virus (VSV), which enters cells by CME [24], [25], was still sensitive to 80 µM dynasore in the presence of 10% FCS (Fig. 2B). In agreement herewith, the uptake of transferin, known to occur via CME, was inhibited by dynasore regardless of the presence of FCS (Fig. S2, panel A). As expected, both DYNA-DEP entry in PBS and DYNA-IND entry in the presence of 10% FCS and 80 µM dynasore required sialic acid receptors for efficient entry as pre-treatment of HeLa cells with neuraminidases almost completely abolished entry via either pathway (Fig. 2C). The kinetics of the DYNA-DEP and DYNA-IND entry pathways were compared by performing a time-course experiment in which IAV entry was terminated by the addition of 10 nM BafA1 at different time points (Fig. 2D). In comparison to entry via the DYNA-DEP pathway (the only pathway available in PBS) entry in the presence of FCS (when presumably both the DYNA-DEP and DYNA-IND entry pathways are available) showed similar kinetics. In contrast, entry via the DYNA-IND pathway (which is the only pathway that is active in the presence of 10% FCS and 80 µM dynasore) was slower. The difference was most prominent after 15 min, while after 4 hrs similar levels of entry were reached.
To validate and extend these results we visualized the reduction of the number of infected cells by immunoperoxidase staining using an antibody against NP (Fig. 3). A number of different cells of mammalian and avian origin were infected for 2 hours at an MOI of 1 in PBS with or without serum. After 2 hours the inoculum was replaced by growth medium containing 10% FCS and 10 nM BafA1 and the expression of NP was examined after 14 hours later. After incubation in PBS, staining was completely prohibited by the presence of 80 µM dynasore whereas in the presence of serum dynasore had no effect. A serum-inducible, DYNA-IND route of entry was thus functional in all five cell lines, including the human epithelial airway carcinoma cell line A549.
To confirm our results and to obtain further proof for the utilization of DYNA-DEP and DYNA-IND entry routes by IAV, we additionally used an IAV virus-like particle (VLP) direct entry assay [26]. These VLPs contain IAV HA and NA in their envelope and harbor a beta-lactamase reporter protein fused to the influenza matrix protein-1 (BlaM1), which allows the rapid and direct detection of entry, independent of virus replication. Upon fusion of viral and endosomal membrane, BlaM1 gains access to the cytoplasmically retained fluorigenic substrate CCF-2 that, after cleavage by BlaM1, shifts to a shorter fluorescent emission wavelength that can be detected by flow cytometry. Entry into HeLa cells was performed in the absence or presence of 10% FCS using VLPs containing HA and NA either from IAV-WSN (having a strict alpha 2–3 linked sialic acid binding specificity) or from the pandemic 1918 IAV (HA from A/NewYork/1/18, binding to alpha 2–3 and alpha 2–6 linked sialic acids; NA from A/BrevigMission/1/18). Entry of VLPs of both IAV strains was severely inhibited by dynasore when no serum was added to the inoculum (Fig. 4A, 4D), whereas the presence of 10% FCS rendered entry completely dynasore resistant. (Fig. 4B, 4E). Quantification of VLP entry is shown in Fig. 4C and F.
Importantly, to confirm the existence of the serum-inducible entry pathway by a method that is independent of dynasore, we used siRNA induced silencing of dynamin 2. Fig. 4G shows that two different siRNAs had a significant inhibitory effect (48 hrs after siRNA transfection) on entry of the Renilla luciferase-encoding pseudovirus WSN-Ren [27] in HeLa cells in the absence of FCS, whereas the presence of 10% serum no reduction in entry levels was observed, confirming the results obtained with dynasore. Knockdown of dynamin 2 protein levels (48 hrs after siRNA transfection) was analyzed by western blotting (Fig. 4H) and quantified in Fig. 4I which also shows the knockdown of dynamin 2 mRNA levels as determined by quantitative RT-PCR.
We conclude that a DYNA-IND entry pathway can be induced by serum in different cell types from several species. The evidence was obtained using both replication-dependent (Gluc-entry assay and immunodetection of infected cells) and replication-independent assays (entry of VLPs), the latter allowing immediate detection of the fusion-mediated delivery of viral M1 protein into the cytoplasm.
The DYNA-IND entry pathway was further characterized by inhibitor profiling using an 80-compound kinase inhibitor library. Serum-induced DYNA-IND entry was examined in 10% FCS using the Gluc-entry assay. 80 µM dynasore was added in order to block CME and any other potential DYNA-DEP entry pathways. This allowed the independent inhibitor profiling of the novel pathway by avoiding the potentially masking effect of the presence of redundant entry pathways. Cells were preincubated with the kinase inhibitors (10 µM) for 1 h at 37°C and then inoculated with virus (MOI 0.5) in the presence of 10% FCS and 80 µM dynasore for 2 h at 37°C (DYNA-IND entry). In parallel, inoculations were also done in PBS to compare the effects of the inhibitors on DYNA-DEP entry. After 2 hr the medium and inhibitor were replaced by full growth medium containing 10% FCS and 10 nM BafA1 to allow the subsequent expression of Gluc activity under identical conditions for the DYNA-IND and -dependent entry assay. Six kinase inhibitors appeared to act non-discriminatively, inhibiting both DYNA-DEP and DYNA-IND entry (Fig. 5A): the protein kinase C (PKC) inhibitors Ro 31-8220, rottlerin (both displaying moderate cytotoxicity, result not shown) and hypericin, which have all three been previously identified as IAV inhibitors [28], [29]; the highly cytotoxic pan-specific serine/threonine protease inhibitor staurosporine; the irreversible PI-3 kinase inhibitor wortmannin and the receptor tyrosine kinase inhibitor TYR9. In order to investigate whether some of these inhibitors affect IAV replication during the post-entry phase, we performed the same experiments but now adding the kinase inhibitors after viral entry. Four of the inhibitors thus appeared to induce significant inhibition of post-entry processes (Fig. 5A). Although unlikely, we cannot formally exclude that post-entry processes specific for only one of the two entry pathways are affected. Interestingly, whereas no specific DYNA-DEP entry inhibitors were identified, 15 inhibitors (none displaying cytotoxic effects, data not shown) caused significant (p<0.05) inhibition (>5-fold) of DYNA-IND entry (Fig. 5B). This included inhibitors of the calmodulin dependent kinases myosin light chain kinase (MLCK) and CaMKII and seven inhibitors of different growth factor receptor tyrosine kinases. In contrast to the three non-specific PKC inhibitors mentioned above, the PKC inhibitors BIM-1 and HBDDE appeared to have a specific inhibitory effect on DYNA-IND entry. The specific effect of these drugs on DYNA-IND entry is not only shown by the lack of inhibition of DYNA-DEP entry in PBS, but also by the observation that none of the fifteen compounds induced >2-fold inhibition when added post-entry (at t = 2 hr post infection). The kinase library screen was repeated on A549 human epithelial lung carcinoma cells in order to confirm the results in a potentially more natural host cell line. The inhibition profiles obtained were very similar to those found for HeLa cells with the exception of the strong effect of AG879 (99% inhibition) and moderate effects of AG825 (39% inhibition) and Tyr51 (68% inhibition) on DYNA-DEP entry. (Fig. 5C).
MLCK inhibitors ML-7 and ML-9 have been reported to be highly specific for their target kinase [30]. Phosphorylation by MLCK activates non-muscle myosin II light chain, indicating that a functional actomyosin network might be essential for DYNA-IND entry of IAV. This was further examined by testing the effect of Blebbistatin, an inhibitor of myosin II heavy chain activity, and of several inhibitors that affect actin dynamics by disrupting actin microfilaments (Cytochalasin B and D), by enhancing actin polymerization (Jasplakinolide) or by inhibiting actin polymerization (Latrunculin A). Actin inhibitors were used at the minimal concentration required to induce clearly visible changes in the actin cytoskeleton as pre-determined by staining with FITC-phalloidin (results not shown). Whereas the inhibitors did not affect DYNA-DEP entry (Fig. 6A) using Gluc-entry assay, all inhibitors as well as ML-7 and ML-9 significantly inhibited DYNA-IND entry (Fig. 6B).
Next, HeLa cells were transfected with plasmids encoding dominant negative or wildtype Rab5 fused to green fluorescent protein (Rab5 DN and Rab5 wt in Fig. 6) 24 h prior to infection with IAV. Rab5 is a small GTPase found in association with several endosomal compartments and crucial for the function and maturation of early endosomes. It is required for the trafficking of a wide range of endocytic cargo following different routes, including DYNA-DEP as well as DYNA-IND routes [31]. Entry of IAV has been shown to require Rab5 [32]. Consistently, we found that HeLa cells expressing Rab5 DN (as identified by GFP fluorescence, Fig. 6C) were much less susceptible to productive IAV infection (as judged by indirect immunofluorescence using Alexa-488 labeled NP antibodies) than cells transfected with Rab5 wt, both by DYNA-DEP (64% inhibition) and by DYNA-IND (47% inhibition) routes (Fig. 6D). In contrast, when examining the efficiency of DYNA-DEP and DYNA-IND entry routes in cells transfected with a dominant negative mutant of MLCK (MLCK DN; examined in comparison to MLCK wt), only IAV infection under DYNA-IND conditions was significantly reduced (Fig. 6E; 53% inhibition). Similarly, a construct encoding the N-terminal head domain of myosin II MyoII-head) also only significantly affected DYNA-IND entry (Fig. 6F; 92% inhibition). The combined results indicate that a dynamic actomyosin network requiring the activation of myosin II by MLCK is necessary for efficient entry of IAV via a DYNA-IND pathway.
Several dynamin-independent endocytic pathways have been described [8], [19]. Of these, macropinocytosis has been demonstrated to be stimulated by growth factors present in serum and to depend on actin dynamics [12]–[14]. Yet, studies on macropinocytosis are hampered by a lack of specific inhibitors, cargo, membrane markers and characteristic morphology. Amiloride and the more potent derivative EIPA are inhibitors of epithelial sodium channels (ENaC) as well as of several other Na+/H+ antiporters. EIPA has often been used as a hallmark inhibitor that specifically inhibits endocytosis via the macropinocytic pathway [14]. Whereas DYNA-DEP entry of IAV was not inhibited by EIPA (Fig. 7A), DYNA-IND entry was fully blocked EIPA (Fig. 7B). The existence of redundant entry pathways in the presence of 10% FCS is clearly demonstrated by the marginal inhibition by either EIPA or dynasore whereas the combination of EIPA and dynasore resulted in strong inhibition both in the Gluc-entry assay (Fig. 7C) and in the direct VLP entry assay (Fig. 7D and E). Supplementary Fig. S1 shows that other cell lines, including the human lung epithelial cell line A549, display similar IAV inhibition patterns for EIPA and dynasore. Consistently, virus production displayed a similar inhibitor sensitivity profile (Fig.7 F and G) as virus entry indicating that the entry pathways we characterized lead to a productive infection. Clearly, VLPs and viral particles follow similar redundant entry pathways, distinguishable in a DYNA-DEP and a DYNA-IND pathway, the latter being sensitive to EIPA and dependent on actomyosin function.
One characteristic of macropinocytosis is the nonselective uptake of large amounts of extracellular solutes [33]. Therefore, the uptake of soluble FITC labeled dextran (Fdx) into relatively large vesicles (0.3 to 5 µM) has often been applied as a morphological marker for macropinosomes. Using this marker we found that the addition of 10% FCS to the culture medium slightly increased the uptake of Fdx into HeLa cells (Fig. 8A). Notably, the distribution of Fdx changes in response to serum from a random distribution into a more granular pattern. At high magnification and at color settings adjusted to higher intensity it could be seen that these Fdx granules were free of actin staining (by phalloidin) indicating that they were in the lumen of vesicles (result not shown). Interestingly, in the presence of IAV (MOI of 10) the uptake of Fdx into vesicles was clearly enhanced. At a higher magnification viral particles could be found to co-localize in Fdx loaded vesicles as well as outside these vesicles (Fig. 8B). Phalloidin staining of actin was used to demonstrate that many virus particles localized to actin-rich protrusions at the periphery of the cell. The uptake of Fdx was studied in a quantitative manner by flow cytometry (Fig. 8 C). A moderate, but reproducible shift to higher Fdx fluorescence was observed at 37°C when virus was added in presence of 10% FCS whereas such a shift was absent when no serum or virus was added. This result confirms the observations by confocal microscopy (Fig. 8A) which showed that the combined presence of FCS and IAV increases the uptake of Fdx as compared to FCS alone. In a control experiment the uptake of Fdx in 10% FCS in presence of IAV was shown to be specifically inhibited by EIPA, but not by dynasore (Fig. S2, panel B). In contrast, transferrin uptake, which serves as a specific marker for CME, was affected by dynasore, but not by EIPA (Fig. S2, panel A).
In conclusion, serum induces the uptake of Fdx into large vesicles, which can be further enhanced by the addition of IAV particles that, after entry, co-localize in part with these vesicles. These results indicate the utilization of a macropinocytic pathway for entry of IAV, which is consistent with the observed sensitivity of the serum-inducible DYNA-IND entry of IAV and VLPs to EIPA.
Macropinocytosis has been implicated in the entry of several viruses [12], [14]. However, differences in susceptibility to inhibitors suggest that distinct forms of macropinocytosis might be used by different viruses [34], [35]. By screening specific inhibitors in the Gluc-entry assay using DYNA-IND entry conditions we evaluated the possible involvement of a few signaling cascades that have been implicated in the induction of macropinocytosis. Serum-inducible macropinocytosis has been shown to be activated via a myriad of signaling cascades initiated by growth factors binding to transmembrane tyrosine kinase receptors [14], [17], [36], [37], consistent with the results shown in Fig. 5. A prominent downstream effect of these signaling cascades is the activation of p21 associated kinase 1 (PAK1) which in turn can activate a number of different pathways leading to actin network rearrangements that can ultimately lead to the induction of macropinocytosis [38]. Fig. 9A–B shows that 20 µM IPA3, an inhibitor of PAK1 [39], specifically inhibits DYNA-IND entry of IAV. Activation of PAK1 in response to growth factor stimulation often involves upstream signal transduction by members of the Rho sub-family of small GTPases like CDC42 and/or Rac1 [34], [40], [41]. Alternatively, activated CDC42 and Rac1 can induce actin rearrangements independently of PAK1 [34], [40]–[43] by direct interaction with WASP or WAVE family proteins, respectively [44], [45]. However, inhibitors of CDC42 (Pirl1 [46]), Rac1 (NSC23766 [47]) or N-WASP (wiskostatin [48]) did not display inhibitory effects on DYNA-IND or DYNA-DEP entry of IAV (Fig. 9C–D). Instead, Pirl1 and wiskostatin induced a significant, concentration dependent increase of entry. This stimulatory effect was not observed for the control vaccinia virus strain WR, which enters cells via a Rac1-dependent, macropinocytotic pathway [43] (Fig. 9E), indicating that this effect is specific for IAV. The results suggest a requirement for PAK1 in DYNA-IND entry of IAV that does not require activation by either CDC42 or Rac1. Growth factor inducible activation of the tyrosine kinase src has also been linked to the induction of macropinocytosis [49]–[51]; consistent with this observation the src inhibitor PP2 [52] specifically inhibited DYNA-IND entry of IAV (Fig. 9A–B). Remarkably, 17-AA-geldanamycin, a specific inhibitor of the chaperone protein HSP90 [53], also caused specific inhibition of DYNA-IND entry (Fig. 9A–B). HSP90 affects the folding and activity of many proteins but the recent demonstration of direct activation of the catalytic activity of src by HSP90 [54] provides another indication of the involvement of src in DYNA-IND endocytosis of IAV.
In conclusion, like for other viruses utilizing a macropinocytic entry pathway, PAK1 seems to play a crucial role in DYNA-IND entry by IAV. However, this pathway is independent of Rac1 or cdc42 but may require src, either upstream and/or downstream of PAK1.
The data presented in this study demonstrate for the first time that IAV can enter cells via DYNA-IND macropinocytosis in addition to the previously described DYNA-DEP classical CME pathway [1], [2]. Several lines of evidence indicate that the DYNA-IND entry route of IAV that we identified corresponds with macropinocytosis. First of all, the entry pathway is dependent on the presence of serum, a well-known inducer of macropinocytosis. Second, IAV colocalized in vesicles with soluble FITC-dextran, a marker for macropinocytosis. Third, DYNA-IND IAV entry was sensitive to the amiloride-derivative EIPA, the hallmark inhibitor of macropinocytosis [14], [55]–[58]. Fourth, this IAV entry pathway is sensitive to inhibitors or dominant-negative mutants affecting actomyosin dynamics. Fifth, the specific inhibition of DYNA-IND IAV entry by a number of inhibitors of growth factor receptor tyrosine kinases as well as downstream effectors thereof also points at the involvement of macropinocytosis. Finally, macropinocytosis is independent of dynamin [12], [14], [19].
Despite this extensive list of arguments, viral entry by macropinocytosis needs to be considered with caution. The characteristics of the DYNA-IND route of cell entry by IAV are similar, but not identical to the macropinocytic entry routes taken by other viruses, like two different strains of vaccinia virus and by coxsackie virus B [34], [35]. As is shown in Table 1 and discussed in more detail below, the macropinocytic pathways used by each of these viruses have a few unique characteristics. This may very well reflect the growing notion that macropinocytosis represents a number of differentially induced and regulated processes, rather than being a single endocytic pathway [13], [14]. Macropinocytosis has collectively been described as an inducible form of endocytosis by which fluid-phase cargo travels via non-coated, relatively large and heterogeneous organelles that have emanated from extensive protrusions (e.g lamellar ruffles, circular ruffles or retracting blebs) of the plasma membrane [13]. In the case of DYNA-IND IAV entry more extensive studies using electron microscopy will be required to study the morphology of membrane protrusions with which IAV may associate. In addition, live cell imaging microscopy will be required to characterize the exact itinerary that is taken by IAV virions traveling via a macropinocytic process. This is especially important as different routes of IAV entry are likely to converge at some point in the endocytic pathway. Although unlikely, co-localization of IAV particles with fluid-phase dextran as shown in Fig. 8B may thus represent a situation occurring after convergence of several different routes. The use of microscopy to study macropinocytosis is however complicated by the lack of specific membrane-associated markers for any early step of this endocytic process.
A model (Fig. 10) based on our results explains the key steps involved in the macropinocytic entry pathway of IAV, which are described in more detail below.
By manipulating the inoculation conditions we were able to experimentally dissect IAV entry into a DYNA-DEP and DYNA-IND route. The DYNA-IND route required the presence of 10% FCS in the entry assay medium. Previously, a strict dependency on a DYNA-DEP entry route for IAV was concluded from experiments with a cell line expressing an inducible dominant-negative mutant of dynamin 2 [59]. In that study, as well as in other entry studies of IAV, entry was performed in DMEM containing 2% serum or BSA. Also in our hands 2.5% serum (Fig. 2A) or 0.2% BSA (result not shown) was not sufficient to allow DYNA-IND entry. We are currently investigating which serum component is responsible for the observed effects on IAV entry. Dialysis of FCS (MW cut off >10 kDa) did not affect its capacity to induce DYNA-IND endocytosis (result not shown), indicating that low molecular weight solutes are not responsible for the observed effect.
Our evidence for a DYNA-DEP and a serum inducible DYNA-IND entry route is based on the use of pharmacological (dynasore, a highly specific inhibitor of dynamin) as well as genetic (siRNA directed against dynamin 2) tools, ruling out the possibility that the inhibitory effect of dynasore was due for instance to absorption of the inhibitor by serum components. Whereas dynasore resulted in near 100% inhibition of DYNA-DEP entry, only 65% inhibition was observed upon siRNA induced silencing of dynamin 2 indicating that the residual levels of dynamin 2 that remain after 48 hrs of silencing still support a low level of DYNA-DEP entry (Fig. 4H). Reversible inhibitors like dynasore [60] offer a major advantage for characterization of IAV entry pathways. They can be applied for a limited period thus preventing the secondary adaptive effects of cells that may occur in response to long-term down regulation of a gene product by genetic methods like siRNA interference. Both entry routes were consistently identified by a viral entry assay quantified by virus induced expression of a luciferase reporter as well as by a VLP entry assay allowing direct analysis of the membrane fusion mediated entry step. The consistent performance of an HA with a strict preference for binding to α2-3 linked sialic acids (from IAV-WSN; our unpublished data) and an HA also binding to α2-6 linked sialic acids (from 1918 IAV [61]) in the VLP entry assay indicates that both pathways can be utilized by HAs of different specificity and may therefore be relevant to avian as well as human IAV infections. Consistently, serum-inducible DYNA-IND entry was observed both in avian DF1 cells and in a human lung epithelial carcinoma cell line A549 (Fig. 3).
The DYNA-DEP and DYNA-IND IAV entry pathways were found by our quantitative assays to be fully redundant. In the presence of serum, the combination of dynasore (inhibiting DYNA-DEP entry) and EIPA (inhibiting DYNA-IND entry) completely abolished entry whereas either drug alone had no effect. EIPA, an inhibitor of plasma membrane Na+/H+ exchangers, has been shown to invariably inhibit macropinocytosis [14], [55]–[58]. As other routes of endocytosis are generally not affected, EIPA is considered as a hallmark inhibitor of macropinocytosis [14], although results obtained with EIPA should be considered with care as long as a mechanistic explanation for its effect on macropinocytosis is not yet fully clear [62]. Occasionally, a moderate two- to three-fold inhibition by dynasore alone was observed (result not shown) indicating that the capacity of the serum-inducible entry pathway is somewhat variable, possibly depending on slight variations in serum quality and factors like cell distribution in the wells that have been reported to influence viral infection [63]. A redundancy in the utilization of CME as well as a clathrin-independent route for entry of IAV has been visualized previously by quantitative live cell imaging [4]. Both routes were operative simultaneously in the same sample and the specific down-regulation of CME did not affect the total number of entry events.
In response to specific extra-cellular signals (e.g. serum induction), changes in the actomyosin network occur that give rise to membrane protrusions required for macropinosome formation [13]. Compounds inhibiting actin polymerization (cytochalasin B and D), depolymerization (jasplakinolide) or sequestering soluble actin (latrunculin A) all specifically inhibited DYNA-IND IAV entry. In addition, the requirement for myosinII activity was established by a specific inhibitor (Blebbistatin) of myosin II ATPase activity and by the expression of a dominant negative mutant of myosinIIA heavy chain. Also, the regulation of myosinII activity by phosphorylation of myosin light chain through the action of MLCK is suggested by the inhibitory effect of MLCK inhibitors ML-7 and ML-9 as well as by the similar effect of an expressed MLCK dominant negative mutant. Recently, a function for the actin cytoskeleton in IAV entry was reported to be required for the entry into polarized epithelial cells but not for entry into non-polarized cells [64]. When using the low-serum conditions used in that paper (2% FCS), we only observed DYNA-DEP entry that was not affected by actin dynamics inhibitors. Perhaps, the polarized cells permit DYNA-IND entry at lower serum concentrations.
The changes in actin network dynamics that can lead to the formation of macropinosomes can be triggered by a number of signaling cascades. Actin dynamics are induced by the activation of growth factor receptor tyrosine kinases by their respective growth factor ligands that are normally present in serum [12]–[14], [17], [36], [37] The signal transduction cascades that link activation of growth factor receptor tyrosine kinases to actin remodeling and macropinocytosis are only beginning to be revealed. The specific inhibition of DYNA-IND entry of IAV by IPA3, an inhibitor of PAK1, provides proof for the involvement of these cascades. PAK1 is a key serine/threonine kinase regulating actin network dynamics but its crucial function in several pathways of endocytosis as well as numerous other cellular processes does not make it a very specific marker [65]. Even so, macropinocytosis has consistently been demonstrated to require PAK1 activation, both in the induction of the process and/or in further downstream trafficking events of macropinosomes [13], [14].
Growth factor dependent activation of PAK1 has most often been demonstrated to depend on upstream activation of small GTPases Rac1 or cdc42 [34], [40], [41]. Different strains of vaccinia virus were recently shown to induce their uptake by macropinocytosis via activation of either Rac1 or cdc42 [34]. Activation of Rac1 has been linked to the induction of macropinocytosis via actin network-mediated formation of lamellipodia and/or circular ruffles whereas cdc42 has most often been implied in the formation of filopodia [44]. An inhibitory effect of the Rac1 inhibitor NSC23766 or the cdc42 inhibitor pirl1 on IAV entry, however, could not be demonstrated. Remarkably, cdc42 inhibitor pirl1 enhanced IAV entry and a similar effect was observed by wiskostatin, an inhibitor of N-WASP which functions directly downstream of cdc42 as a scaffolding complex required for the activation of actin polymerization leading to filopodia formation. Similarly, the macropinocytosis-like entry pathway taken by Coxsackie B virus was also shown to require PAK1 activity that was independent of Rac1 activation [35]. Direct examination of the magnitude and timing of the activation of PAK1 will be required to obtain more insight in the involvement of this complex pathway. The induction of macropinocytosis by a PAK1-dependent mechanism has been associated with ruffling at the cell membrane [12], [14], [15], [37]. The identification of sub-membranous regions with increased actin staining by phalloidin has been interpreted as evidence for ruffling. This was not unambiguously identified by confocal microscopy in the experiments presented in Fig. 8 and Fig. S2 and needs to be investigated in depth by life cell imaging techniques.
In agreement with our observation that the DYNA-IND entry of IAV was inhibited by PP2, an inhibitor of src family kinases, the non-receptor tyrosine kinase c-src has been shown to function as a key signaling intermediate in the induction of macropinocytosis via a mechanism independent of Rac1 or cdc42 [49]–[51]. Downstream effects of c-src on actin networks proceed, amongst others, via phosphorylation of cortactin by c-src resulting in accelerated macropinosome formation [50]. C-src has been shown to associate with macropinosomes [49], [51], both during their formation and their trafficking, while c-src kinase activity is required for macropinocytosis following EGF stimulation of HeLa cells [49]. Interaction of HSP90 with c-src was recently shown to induce c-src kinase activity [54]. Also HSP90 has been demonstrated to associate with macropinosomes, while its specific inhibitor geldanamycin reduced the membrane ruffling that preceded macropinocytosis [66]. Thus, the inhibition of IAV entry via macropinocytosis by AA-geldanamcyin may very well involve the effects of HSP90 on c-src.
As detailed above, the DYNA-IND entry pathway of IAV shares many characteristics with the endocytic pathway macropinocytosis. This is corroborated by the observation that IAV particles and dextran colocalize in large vesicles in the presence of FCS. Several viruses have recently been reported to enter cells via macropinocytosis [12], [14]. Apart from common factors like the requirement for PAK1 activation, actin dynamics and independence of dynamin, virus specific details have been described [34], [35] (Table 1). In part these might be contributed to differences in experimental conditions (e.g. cell types tested) but diversity in the molecular mechanisms by which macropinocytosis can be induced and executed is likely to exist and to be exploited by viruses. Whereas vaccinia virus is able to trigger its own macropinocytic uptake [34], [43], we have described a macropinocytosis pathway that is operational under conditions that are activated by components in serum. Still, this does not exclude signaling induced by virus-host cell interactions, which are for instance suggested by the significant increase of FITC-dextran uptake in the presence of IAV. The possible requirement for co-stimulatory signals from serum components and virus imposes an additional layer of complexity on the analysis of IAV entry via DYNA-IND pathways.
Influenza viruses cause respiratory infections by targeting the epithelial cells lining the respiratory tract. These surfaces are covered by a mucous layer composed of a variety of small solutes and glycoproteins derived among others from goblet cells [67]. This semi-fluid layer in turn conditions the underlying cells and determines their physiological state, including the activities of their uptake and secretion pathways. It will be important to determine to what extent the DYNA-DEP and DYNA-IND IAV entry pathways are operational under the conditions prevailing along the respiratory tract. Current knowledge on the protein composition of the fluids covering the respiratory epithelium is rapidly expanding by the application of proteomic methods to determine the protein composition of bronchial alveolar lavage fluids (BALF). These studies have extended the previous notion that BALF is highly similar in composition to serum. For example, just as for the serum proteome more than 85% of the total protein mass of the BALF proteome is accounted for by albumin, immunoglobulins, transferring, α1-antitrypsin and haptoglobin. In addition, many other proteins have been identified both in serum and in BALF including growth factors that can bind to growth factor receptor tyrosine kinases [68]–[70]. Thus, BALF is likely to harbor, just as serum, the protein factors that can activate signaling pathways that are crucial for the induction of DYNA-IND entry of IAV. In agreement herewith, macropinocytosis has been described as a functional entry pathway of Haemophilus influenzae into primary human bronchial epithelial cells [71] although the factors involved in signaling the process have not been identified yet.
In addition to infecting the respiratory tract, IAV has been shown to be able to cause systemic infections involving multiple organs. This has mainly been studied in avian infections [72], [73] or by infection of mice with human-derived H1N1 or H3N2 IAVs [74] but is poorly documented for human infections and may have been underestimated thus far. Obviously, during potential systemic spreading of IAV, the serum-rich conditions that we have demonstrated here to enable the use of alternative entry pathways will be encountered and may contribute to such spreading.
MDCK, A549, DF-1 and HeLa cells were maintained in complete Dulbecco's Modified Eagle's Medium (DMEM) (Lonza, Biowittaker) containing 10% (v/v) fetal calf serum (FCS; Bodinco B.V.), 100 U/ml Penicillin, and 100 µg/ml Streptomycin. Chinese Hamster-E36 cells were maintained at 37°C in α-Minimal Essential Medium (Gibco) supplemented with 10% (v/v) FCS, 100 U/ml Penicillin, and 100 µg/ml Streptomycin. Cells were passaged twice weekly. Influenza A/WSN/33 (H1N1) (IAV-WSN) was grown in MDCK cells. Briefly, ∼70% confluent MDCK cells were infected with IAV-WSN at a MOI of 0.02. Supernatant was harvested after 48 hr of incubation at 37°C and cell debris was removed by centrifigutation (10 min at 2000 rpm). Virus was stored at −80°C and virus titers were determined by measuring the TCID50 on HeLa cells. The IAV-WSN luciferase pseudovirus (WSN-Ren) system has previously been described [27]. Briefly, WSN-Ren pseudovirus harbors a HA segment in which the HA coding region is replaced by Renilla luciferase. The pseudovirus is produced in a MDCK cell line that stably expresses the HA of IAV-WSN. WR-LUC, a firefly luciferase encoding vaccinia virus (strain WR) was previously described [75]. VSV-FL, a firefly luciferase encoding VSV virus was also previously described [76].
Stocks of bafilomycin A1 (BafA1), dynasore, cytochalasin D, cytochalasin B, Blebbistatin, 17-AA-geldanamycin, ML-7, ML-9, PP-2, 5-(N-ethyl-N-isopropyl)amiloride (EIPA), IPA-3 (all obtained from Sigma-Aldrich), Latrunculin A (Enzo), jasplakinolide, wiskostatin, NSC23766 (all obtained from Calbiochem) and pirl1 (Chembridge) were prepared in dimethylsulfoxide (DMSO). All stocks were stored at −20°C. A kinase inhibitor library composed of 80 kinase inhibitors was obtained from Biomol (2832A[V2.2]).
HeLa cells (10,000 cells/well in 96-well plates) were treated with 2 mUnits of Vibrio cholerae neuraminidase (Roche) in 50 µl phosphate-buffered saline (PBS) for 2 hr. After washing with PBS cells were infected with IAV as described.
Virus-like particles (VLPs) were produced as described [26]. Briefly, 293T cells were transfected using Lipofectamine 2000 (Invitrogen) with pCAGGS-BlaM1 (encoding a beta-lactamase reporter protein fused to the influenza matrix protein-1), pCAGGS-HA (encoding HA derived from either A/NewYork/1/1918 or IAV-WSN) and pCAGGS-NA (encoding IAV neuraminidase [NA] derived from either A/BrevigMission/1/18 or IAV-WSN) and maintained in OptiMEM. Supernatants were harvested 72 h after transfection and centrifuged to remove debris. VLPs were used for inoculation of cells without further concentration. VLPs were incubated for 30 min at 37°C with trypsin/TPCK for activation of HA. MDCK or HeLa cells grown to near confluency in 24-well plates were inoculated with 250 ul of VLPs after pre-treatment of the cells with inhibitors as indicated. Infection was synchronized by centrifugation at 1500 rpm for 90 min at 4°C and was performed by further incubation at 37°C for 2 h in the absence or presence of 10% FCS and inhibitors as indicated. Detection of beta-lactamase activity was performed as described [25] by loading cells with CCF2-AM substrate (InVitrogen) and subsequent analysis by flow cytometry on a LSRII flow cytometer (Becton Dickinson). Typically 10,000 events were collected and analyzed using FlowJo 8.5.2 software.
The reporter construct pHH-Gluc was derived from plasmid pHH-Fluc [22] by replacing the firefly luciferase coding region with the Gaussia luciferase coding region of pGluc-basic (New England Biolabs). Unique SpeI and XbaI restriction sites were introduced into pHH-Fluc using the Quikchange XL Site-directed mutagenesis kit (Stratagene) and oligonucleotides Spe4262 (5-′GCCTTTCTTTATGTTTTTGGCACTAGTCATTTTACCGATGTCACTCAG), Spe4263 (5′-CTGAGTGACATCGGTAAAATGACTAGTGCCAAAAACATAAAGAAAGGC), Xba4260 (5′-GTATTTTTCTTTACAATCTAGACTTTCCGCCCTTCTTGG) and Xba4261 (CCAAGAAGGGCGGAAAGTCTAGATTGTAAAGAAAAATAC). A SpeI site was introduced by site-directed mutagenesis in pGluc-basic directly following the start codon of the Gaussia luciferase coding sequence. The unique SpeI – XbaI fragment of pGluc-basic was subsequently cloned into the SpeI-XbaI site of pHH-Fluc resulting in plasmid pHH-Gluc.
Cells were seeded in 96-well plates at a density of 10,000 cells/well and transfected the next day with 10 ng pHH-Gluc using Lipofectamine 2000 (InVitrogen) according to the manufacturer's protocol. After 24 hrs the transfected cells were treated with inhibitors and infected as indicated. At 16 hr p.i. samples from the supernatant were assayed for luciferase activity using the Renilla Luciferase Assay system (Promega) according to the manufacturer's instructions, and the relative light units (RLU) were determined with a Berthold Centro LB 960 plate luminometer. WR-LUC and VSV-FL were used to inoculate HeLa cells (10,000 cells/well) at an MOI of 2, in complete Dulbecco's Modified Eagle's Medium (DMEM) (Lonza, Biowittaker). After 7 hr the luciferase activity was detected using the SteadyGlo assay kit (Promega). The addition of 10% (v/v) FCS did not change infection levels for both viruses.
Cells were fixed with 3.7% paraformaldehyde (PFA) in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. After blocking with normal goat serum IAV-infected cells were incubated for 1 h with a monoclonal antibody directed against the nucleoprotein (NP) (HB-65; kindly provided by Dr. Ben Peeters). After washing, the cells were incubated with a 1∶400 dilution of Alexa Fluor 488- or 568-labeled goat anti-mouse IgG (Molecular Probes) secondary antibody for 1 h. Nuclei were subsequently stained with TOPRO-3 and after three washing steps, the coverslips were mounted in FluorSave (Calbiochem). Actin was stained using phalloidin labeled with Alexa Fluor 633. The immunofluorescence staining was analyzed using a confocal laser-scanning microscope (Leica TCS SP2). FITC, GFP or Alexa Fluor 488 were excited at 488 nm, Alexa Fluor 568 at 568 nm, and TOPRO-3 at 633 nm.
HeLa cells were grown in 24-well plates on glass coverslips (50,000 cells/well). Prior to FITC-dextran uptake cells were serum-starved for 2 hr in PBS. FITC-dextran (MW70,000, Sigma-Aldrich) was incubated with HeLa cells (final concentration of 0.5 mg/ml) in 500 µl PBS or in PBS containing 10% FCS in the absence or presence of IAV (strain WSN; MOI 10; concentrated and purified by centrifugation through a 15 to 30% sucrose gradient with a 50% sucrose cushion at the bottom) at 37°C. After 15 min cells were washed 4 times with PBS at 4°C, fixed with 3.7% PFA in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. Slides were stained for examination by confocal microscopy as described above. For quantification of FITC-dextran uptake 1.5×105 HeLa cells were infected with IAV-WSN (MOI 10) in suspension in a volume of 1 ml in the presence of FITC-Dextran (1 mg/ml). Infections were performed for 15 min in PBS (containing 2% BSA to reduce unspecific binding of FITC-Dextran) or in PBS containing 10% FCS at 37°C or at 4°C (control for binding of FITC-Dextran to cells in the absence of endocytosis). Mock-infected samples were analysed in parallel. Infection was terminated by addition of 3 ml ice-cold PBS followed by three washes with cold PBS and fixation with 3.7% PFA. 20,000 cells were analyzed by FACS and results were represented as the mean fluorescence which was plotted relative to the uptake in the mock-infection in PBS (after subtraction of background fluorescence obtained at 4°C).
HeLa cells (grown on glass cover slips) were incubated at 4°C for 1 hr with 50 µg/ml Alexa633-labeled Transferin (InVitrogen) in PBS. After 1 hr the medium was replaced by PBS or PBS supplemented with 10% FCS containing IAV (strain WSN; MOI 10) and 0.5 mg/ml FITC-Dextran (Sigma; 70 kDa) and cells were transferred to 37°C for 15 min. After 15 min cells were fixed and stained as described above and examined by confocal microscopy.
Cells were fixed with 3.7% PFA in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. Peroxidase was visualized using an AEC substrate kit from Vector Laboratories. IAV-positive cells were detected using bright-field light microscopy.
Two siRNA duplexes targeting different sites within the coding sequences of dynamin 2 were obtained from Ambion Inc (15581 (Dynamin 2 siRNA 1) and 146559 (dynamin 2 siRNA2)). A scrambled siRNA (Ambion Inc.) was taken along as a control for non-specific effects of the transfection procedure and was used for normalization. One day after seeding in 96-well plates (6,000 cells/well), the HeLa cells were transfected with a final concentration of 10 nM siRNA using oligofectamine (Invitrogen). 48 h after transfection, the cells were inoculated with the WSN-Ren pseudovirus (MOI 0.5) in PBS or in PBS containing 10% FCS. After 2 h of infection the entry medium was replaced by complete growth medium containing 10 nM BafA1 to prevent further entry. At 16 h post infection intracellular Renilla luciferase expression was determined as described above. Each siRNA experiment was performed in triplicate. Cell viability was not affected as determined by performing a Wst-1 cell-viability assay (Roche). Functional knockdown of dynamin 2 mRNA levels was performed by quantitative RT-PCR. using a TaqMan Gene Expression Assay for DNM2 (Hs00191900_m1, Ambion) and using 18S RNA (Hs03928985_g1, Ambion) as a control for normalization. The comparative Ct-method was used for quantification of the results [77]. Reduction of dynamin 2 protein levels was determined by western blotting using polyclonal goat-anti-dynamin 2 C18 (Santa-Cruz SC-6400). A monoclonal against alpha-tubulin (DM1A, Sigma T9026) was used to detect tubulin for normalization. Results were quantified by Densitometric scanning of the dynamin 2 and tubulin signals displayed in Fig. 4H.
HeLa cells were grown in 24-well plates on glass coverslips (50,000 cells/well) for 24 hrs. Cells were then transfected (1 µg of DNA with lipofectamine 2000 as described above) with plasmids encoding wild-type or dominant-negative (DN) human MLCK fused to GFP [78], wild-type or DN Rab5 fused to GFP [79], or MyoII-tail or MyoII-head domain fused to GFP [80]. 24 hr after transfection cells were inoculated with IAV-WSN (MOI 1) in PBS or in PBS containing 10% FCS and 80 µM dynasore. 4 hr after infection cells were fixed and stained for examination by confocal microscopy as described above.
An unpaired Student's t-test was used for detemination of statistically significant differences. The use of the term significant in text refers to a comparison of values for which p<0.05.
|
10.1371/journal.pcbi.1003344 | Transformation of Stimulus Correlations by the Retina | Redundancies and correlations in the responses of sensory neurons may seem to waste neural resources, but they can also carry cues about structured stimuli and may help the brain to correct for response errors. To investigate the effect of stimulus structure on redundancy in retina, we measured simultaneous responses from populations of retinal ganglion cells presented with natural and artificial stimuli that varied greatly in correlation structure; these stimuli and recordings are publicly available online. Responding to spatio-temporally structured stimuli such as natural movies, pairs of ganglion cells were modestly more correlated than in response to white noise checkerboards, but they were much less correlated than predicted by a non-adapting functional model of retinal response. Meanwhile, responding to stimuli with purely spatial correlations, pairs of ganglion cells showed increased correlations consistent with a static, non-adapting receptive field and nonlinearity. We found that in response to spatio-temporally correlated stimuli, ganglion cells had faster temporal kernels and tended to have stronger surrounds. These properties of individual cells, along with gain changes that opposed changes in effective contrast at the ganglion cell input, largely explained the pattern of pairwise correlations across stimuli where receptive field measurements were possible.
| An influential theory of early sensory processing argues that sensory circuits should conserve scarce resources in their outputs by reducing correlations present in their inputs. Measuring simultaneous responses from large numbers of retinal ganglion cells responding to widely different classes of visual stimuli, we find that output correlations increase when we present stimuli with spatial, but not temporal, correlations. On the other hand, we find evidence that retina adjusts to spatio-temporal structure so that retinal output correlations change less than input correlations would predict. Changes in the receptive field properties of individual cells, along with gain changes, largely explain this relative constancy of correlations over the population.
| An influential theory of early sensory processing argues that sensory circuits should conserve scarce resources in their outputs by reducing correlations present in their inputs [1]–[3]. At the same time, recent work has clarified that some redundancy in the retinal output is useful for hedging against noise [4], [5]. Moreover, sensory outputs with varying amounts of correlation can engage cortical circuits differently and thus result in a different sensory “code” [6]. Thus, some degree of redundancy appears to be useful to the brain when dealing with response variability and making decisions based on probabilistic input [7]. Indeed, correlations between neurons in visual cortex are largely unchanged between unstructured and naturalistic visual stimuli [8]. In the absence of neural mechanisms supporting adaptation to the structure of sensory inputs, increased stimulus correlation would induce a corresponding change in neural correlation. Alternatively, the retina may dynamically adjust its coding strategy to represent the new stimulus class efficiently. To explore this possibility, we characterized the impact of stimulus structure on retinal output correlation. Previous studies have examined pairwise correlations amongst retinal ganglion cell spike trains in specific stimulus conditions [9]–[14] but did not report the changes in correlation for the same pairs across stimuli.
Are there mechanisms that might allow the retina to adjust its functional properties when stimulus correlations change? Traditionally, retinal ganglion cells (RGCs) have been described by a fixed linear receptive field followed by a static nonlinearity [15], where surround inhibition acts linearly to suppress pairwise correlations in natural visual input [2], [3]. In this view, the receptive field and nonlinearities might vary dynamically with stimulus correlations, possibly by changing the strength of lateral inhibition to maintain a fixed amount of output correlation. Indeed, correlation-induced changes in receptive fields have been observed in the LGN and visual cortex [16], [17].
To test these ideas, we performed a series of experiments in which we presented the retina with several stimuli with varying degrees of spatial and temporal correlations. The retina never fully decorrelated its input; even for the least correlated white noise checkerboard stimuli, some correlations were present between pairs of retinal ganglion cell spike trains. Responding to natural movies, however, output correlations were only moderately increased compared to correlations in responses to white noise checkerboards, despite the dramatic difference in input-induced correlations. Specifically, the differences in output correlations were much less than those predicted by a non-adapting linear-nonlinear functional model responding to these stimuli. We found a similar result for spatio-temporal exponentially correlated stimuli, with an even smaller change in output correlations. In this way, pairwise output correlations change by a relatively small amount as compared to the expected input-induced change in response to stimuli that span a broad range of spatio-temporal correlations. Stimuli with only spatial correlations, on the other hand, produced increases in output correlations similar to the input-induced changes predicted by a static, non-adapting functional model. In the extreme case, for temporally uncorrelated full-field flicker, the output correlation increased substantially. These results suggest a key role for temporal processing in maintaining the level of output correlations. Indeed, we observed a robustly faster response timecourse and a modest skew towards stronger inhibitory surrounds in response to spatio-temporally correlated stimuli. These changes were sufficient to largely explain the observed relative suppression of pairwise correlations in the retinal output for those experimental conditions where receptive field measurements could be made.
We used a multi-electrode array to measure simultaneous responses from groups of retinal ganglion cells in guinea pig; data and stimuli are available at [18]. Each recording interleaved 10-minute blocks of white noise checkerboard stimuli with 10-minute blocks of correlated stimuli. Example frames from each stimulus are shown in Fig. 1B, together with their respective spatial and temporal correlation functions. We probed retinal responses to natural movies, which allowed us to determine properties of ganglion cell population activity during natural vision. However, natural movies contain strong correlations in time (trace under “natural” stimulus in Fig. 1B) and space (Fig. 1A, B). There are challenges with reliably estimating receptive fields from natural stimuli due to these strong correlations and the highly skewed natural intensity distribution (see Methods). We therefore also assessed the effect of spatio-temporal correlations in a more controlled stimulus with short-range exponential correlations in time and space and a binary intensity distribution (Fig. 1B, “spat-temp exponential”). Additional stimuli allowed us to vary the spatial correlation over a broad range, without temporal structure, in order to test the hypothesis that surround strength adapts to remove correlations in nearby parts of an image. Thus, we examined spatial correlations, in the absence of temporal structure, of increasing extent: spatially exponential, a “multiscale” naturalistic stimulus featuring structure over many spatial scales, and full-field flicker (Fig. 1B, bottom row). The multiscale stimulus was designed to mimic the scale invariance of natural scenes in a controlled binary stimulus, featuring both small and large patches of correlated checks (such as the white area near the center). Its construction is detailed in Methods. In one experiment, we also compared responses to low-contrast white and multiscale stimuli to their high-contrast counterparts. Finally, to control for the effect of the skewed natural intensity distribution, we also conducted experiments presenting scrambled natural movies lacking spatial and temporal correlation while preserving the intensity distribution. The mean luminance and single-pixel variance were matched across all stimuli other than natural movies, scrambled natural movies, and low-contrast stimuli. Over 30 minutes of recording in each stimulus condition, the typical cell fired spikes. This was sufficient to assess spike train correlations and to measure receptive fields for the white and exponentially correlated stimuli.
For preliminary analyses, we measured the spike-triggered average (STA) from each ganglion cell's response to white noise. The resulting receptive fields typically gave good coverage of the sampled visual field (Fig. 2A) and clustered into classes on the basis of their response polarity and temporal properties (Fig. 2B; details in Methods). The four basic classes that we consistently identified across experiments were fast-ON and fast-OFF, distinguished by the transient and biphasic nature of their temporal filter, and slow-ON and slow-OFF, which had longer integration times and often less prominent biphasic filter lobes. It is possible that each of the functionally identified cell classes comprises multiple types of cells. Separating cells by class did not qualitatively change many of the results reported below; in these cases, we combined all cells to improve statistical power.
To probe the effect of stimulus correlation on ganglion cell response properties in detail, we applied a standard functional model, the linear-nonlinear (LN) model. In this model, the visual stimulus is filtered with a linear kernel that represents the spatio-temporal receptive field (STRF) of the cell. The filter output is then passed through a nonlinear transfer function to generate a predicted firing rate. The nonlinearity encompasses thresholding and saturation, as well as any gain on the linear response. For white noise stimuli, the STA is a good estimator of the STRF [19]. However, this simple property does not hold for correlated stimuli, and so we fit the STRFs and other LN model parameters by maximum likelihood estimation (see Methods). For the weakly correlated spatio-temporal exponential stimulus, this technique reliably extracted receptive fields (Fig. 2C).
We computed the correlation coefficient between spike trains (binned at 33 ms) for all pairs of simultaneously recorded neurons. In response to natural movies, correlations between most pairs of cells increased in magnitude when compared with the correlations between the same pairs when viewing white noise (Fig. 3A). We quantified the size of this increase by finding the least-squares best fit line (Fig. 3B, gray lines) and defining the “excess correlation” of a population as the slope of this line minus one (see Methods). If all cell pairs had, on average, the same correlation in both stimulus conditions, the excess correlation would be zero. Excess correlation was not strongly dependent on bin size (Fig. S2B). In the case of natural movies, the excess correlation was (95% confidence interval computed using bootstrap resampling, as explained in Methods; see Table 1), modestly different from zero (and significantly nonzero at the 95% confidence level).
Because the retinal ganglion cell output is a highly transformed representation of its input, it is not trivial to formulate a naïve expectation for the magnitude of output correlation given an input correlation. In particular, simply evaluating the input correlation between stimuli at the receptive field centers of a pair of cells provides a misleading picture, since it neglects the spatial extent and possible overlap of receptive fields. We therefore chose to quantify the output correlation expected for a given input in a simple null model: the LN model fit to the white noise responses. This model captures correlation due to receptive field overlap and simple nonlinear processing, while neglecting correlations due to shared circuitry and more complex nonlinear behavior, such as adaptation. For cells which had sufficiently well-estimated white noise LN model parameters (as described in Methods) we were able to compare the excess correlation predicted by the model to that observed in the data. In order to separate effects that might arise due solely to changes in firing rate between conditions (see Fig. S3A) from changes specifically in pairwise correlations between cells, we adjusted the threshold of each model neuron separately under each stimulus to match predicted average firing rates to their empirical values. All other parameters, namely the spatio-temporal receptive field and the gain, were unchanged between stimuli. This “non-adapting” model predicted a significantly larger excess correlation in response to natural movies (gray bars in Fig. 3D and Fig. 4A), suggesting that the low observed excess correlation value under natural stimulation is a consequence of nontrivial processing in the retina.
In addition to strong correlations, however, natural stimuli are also characterized by a skewed distribution with many dark pixels and a few extremely bright pixels, whereas our white noise stimulus, included equal numbers of bright and dark pixels. To disentangle effects of correlations from effects due to intensity distribution, we presented the same retinae with a scrambled natural movie. In this stimulus, we started with natural movies and randomly shuffled the pixels in space and time to maintain the intensity distribution but remove correlations. The excess correlation in response to this stimulus was consistent with zero in both the measured and simulated responses (Fig. 4A, left bars), suggesting that the skewed natural intensity distribution does not significantly affect output correlations. Moreover, comparing the natural movie and scrambled natural movie directly, we found a small excess correlation consistent with that in the natural movie vs. white noise case. The non-adapting model again predicted that the relative similarity of output correlations was nontrivial (Fig. 4A, right bars). Thus, the retina greatly suppresses changes in correlations of natural visual stimuli.
We found a similar set of results for the more weakly correlated spatio-temporal exponential stimulus (Fig. 3B). In particular, the excess correlation was low () compared to the increase predicted by the non-adapting model (excess correlation of ; Fig. 3D). We also examined the results of experiments in which we presented stimuli with varying degrees of spatial correlation. As shown in Fig. 3C, many stimuli produced only a modest increase in output correlations. Some stimuli with strong spatial correlations, particularly the multiscale and full-field flicker stimuli, resulted in a clear increase in output correlations when compared to white noise. When we varied the contrast of a white noise stimulus, output correlations decreased when the contrast was lowered while all other stimulus properties were kept fixed. Thus, the degree of correlation in the retinal output is not a reflection of stimulus correlations alone. On the other hand, decreasing the contrast of the multiscale stimulus did not significantly affect the output correlations, suggesting that stimulus correlation and contrast interact to shape output correlations.
For further analysis, we focused on the subset of stimuli shown in Fig. 3D, all of which were presented in experiments where we also obtained robust estimates of white noise receptive fields. Here we again simulated responses of an LN model using fixed receptive fields measured under white noise. For most stimuli, the model neurons showed changes in correlation at least as large as those observed in recordings. However, unlike the spatio-temporally correlated exponential and natural stimuli discussed above, the stimuli which had correlations in space only (spatial exponential and multiscale) or no correlations (scrambled natural movie) produced similar excess correlation values in recorded cells and in our non-adapting model. This suggests that a fixed linear filter, as in the non-adapting model, is largely sufficient to explain the effect of spatial correlations, whereas higher-order processing is necessary to suppress the impact of temporal stimulus correlations on output correlation. For the spatially uniform full-field stimulus, output correlations appear to increase more than expected from the non-adapting model. Note, however, that the full-field data were collected as part of other experiments in which we presented white noise checkerboards for a shorter time (10 minutes, as opposed to 30 minutes). Thus the receptive fields are less well estimated and further studies are needed to verify with confidence the predictive performance of a non-adapting model.
As discussed above, we were able to identify the cell classes for many of our recorded neurons. In response to spatio-temporally exponentially correlated noise and natural movies, cell class had a modest effect on output correlations (Fig. 3A, B). Cells with opposite ON- or OFF- polarities (gray points) tended to have negative correlations, whereas cells of the same polarity (black and colored points) generally had positive correlations. Several opposite-polarity pairs did have positive correlation; these tended to have non-overlapping receptive field centers (Fig. S5). Moreover, pairs with opposite polarity showed a greater-than-average excess correlation, particularly in response to natural movies. Under natural movies, opposite-polarity pairs had an excess correlation of 1.5; under the spatio-temporal exponential stimulus their excess correlation was 0.38 (See Fig. S4A, B). Within same- class pairs, slow-ON and slow-OFF pairs (blue and yellow) tended to show a greater excess correlation than fast-ON and fast-OFF pairs (red and green). Pairs of slow cells had an excess correlation of 0.29 in the natural stimulus and 0.28 in the spatio-temporal exponential, while fast pairs were measured as 0.01 and −0.02 for the two stimuli, respectively. All of these class-dependent excess correlations were small compared to the overall non-adapting model predictions (excess correlations of 4.33 and 0.67 for natural and spatio-temporal exponential stimuli). We also assessed the relationship between receptive field separation and output correlation (Fig. 4C). Pairwise correlations tended to decay with distance, but the average change in output correlation between the correlated and white stimuli did not depend on distance.
We next sought to determine whether receptive fields adapt to stimulus correlations and whether this adaptation can explain the observed pattern of output correlations. As noted above, we were able to obtain STAs from responses to white noise. STAs computed in response to correlated stimuli, however, will be artificially blurred by the stimulus correlations. To obtain a better estimate of the spatio-temporal receptive field (STRF), we used maximum likelihood estimation to fit a LN model separately for the white and exponentially correlated stimuli [20]. Examples of STRFs obtained in this way for one cell are shown in Fig. 2C. The strongly correlated structure of the multiscale stimulus and the natural movies precluded robust, unbiased STRF estimation with limited data (see Methods). For this reason, we restricted any STRF computations to white noise and exponentially correlated noise. The latter stimulus is only weakly correlated and thus we would expect at most weak changes in the receptive fields between the conditions; indeed, receptive fields are hard to distinguish by eye for many cells. Measuring such weak changes requires high-quality receptive fields whose locations can be unambiguously determined (see Methods), as was the case for 75 neurons ( of the neurons recorded under spatio-temporal exponential correlated conditions). Cells that did not meet this standard were likely to include types that do not respond as well to checkerboard stimuli, e.g., direction selective ganglion cells and uniformity detectors. We included such cells in the analysis of Fig. 3C in order to maximize the generality of our results and to allow for the possibility that these neurons had qualitatively different output correlations. For the neurons that did pass the quality threshold, we found that the parameters of the LN model (for each neuron, a linear filter and a nonlinearity gain and threshold) changed with the stimulus.
Spike trains with sparse, transient firing events tend to be more decorrelated [14]. Motivated by this finding, together with our observation that temporally correlated stimuli yielded excess correlation in the non-adapting model that was higher than in the data, we analyzed adaptation in the temporal filtering properties of retinal ganglion cells. To isolate changes in temporal processing, we examined each neuron's STRFs (estimated separately under the white and exponentially correlated stimulus conditions) and extracted the temporal components (see Methods). These temporal profiles were faster for the correlated stimulus than for white noise (Fig. 5A). To quantify this difference, we computed the power spectrum of each neuron's temporal filter under each stimulus (Fig. 5B and 5D, top) and found a systematic increase in high frequencies under the correlated stimulus, indicating a shift toward high-pass filtering (Fig. 5E). As the correlated stimulus had relatively more power at low frequencies compared to the white stimulus, this form of adaptation compensates for differences in the power spectrum and, hence, tends to equalize output auto-correlations. In contrast, a non-adapting model with a filter estimated from white noise acting on the correlated stimulus predicts large changes in the output power spectrum (Fig. 5C). Indeed, this compensation was nearly exact for many cells (Fig. 5C), though generally incomplete over the full population (Fig. 5F). These results, combined with those of [14], may indicate that the observed consistency of output correlations is produced by an increase in response transience when stimulus correlations increase.
Next, we found separate temporal profiles for the center and surround and computed the latency, measured as time to peak, of each. Surround latencies did not differ between white noise and spatio-temporally exponentially correlated noise (t-test, , ). However, center latencies were shorter for correlated noise. We quantified the shift in terms of an adaptation index . The histogram of the adaptation index (Fig. 6A; , ; t-test , ; Wilcoxon signed rank text ) showed a robust tail toward shorter center latency for correlated stimuli (). Moreover, almost every cell from which we obtained receptive fields had a longer latency for white noise than for correlated noise (Fig. 6B; ). This was true across cell classes.
To determine whether these changes in temporal filtering were due to the presence of temporal correlations in this particular stimulus (unlike many of the other stimuli we examined), we also measured receptive fields from a separate population of ganglion cells responding to white noise and to a stimulus that was exponentially correlated in space but not in time. In this case, filters did not show a systematic change in power spectra (Fig. 5D, bottom), but the center latencies were shorter for the correlated stimulus (Fig. 6C; ). Again, computing adaptation indices indicated that this effect was significant (, ; t-test , ; Wilcoxon signed rank test ). Thus, spatial correlations in the stimulus appear to produce a shift in the timing of the response without changing the shape of the filter (as measured by the power spectrum). The lack of an effect in the power spectrum may explain why output correlations for the spatially exponential stimulus are not reduced relative to the change in correlation for a non-adapting model (Fig. 3D).
The conventional view of retinal circuitry suggests that adaptive decorrelation arises from stronger or wider surround inhibition during viewing of correlated stimuli. We thus computed the amplitudes of the surround and center components of each neuron's STRFs in both white noise and spatio-temporally exponentially correlated noise. Defining the relative surround strength, , as the ratio of surround amplitude to center amplitude (details in Methods), we computed an adaptation index for each cell as . This adaptation index has a modestly positive mean (Fig. 6D; , ; two-tailed t-test, , ; Wilcoxon signed rank test ), as do the changes in surround strength themselves (Fig. 6E). In addition, the cells with the greatest degree of surround adaptation had a robust tendency to increase in surround strength (). There was no discernible dependence on cell class. Interestingly, the surround strength showed only a marginally significant change when spatial correlations (but not temporal correlations) were added to white noise (Fig. 6F; , ; two-tailed t-test, , ; Wilcoxon signed rank test ). Thus, while we do find some evidence for an increase in surround strength with stimulus correlation, the effect is subtle. This outcome is surprising given the common view since the work of Barlow [1], [2] that surround inhibition is primarily responsible for decorrelation of visual stimuli. However, it is possible that the exponential correlations that permitted us to estimate receptive fields are too weak to evoke strong lateral inhibition.
Finally, we examined the gain of each neuron, defined as the maximum slope of the logistic nonlinearity fit to each neurons' response (see Methods). Since the gain enters the nonlinearity after the stimulus passes through the linear filter, we normalized the filter to unit euclidean norm in order to obtain an unambiguous definition of . We found that the gains of individual neurons changed when the stimulus was more correlated, but there was not a systematic change between stimuli. Recall that the gain of many sensory neurons, including retinal ganglion cells, is known to change with the contrast of the stimulus [21], [22]. To test for a possibly related mechanism at work in our data we first defined “effective contrast,” and , as the standard deviation of the normalized linear filter output in each stimulus, respectively. This notion of effective contrast roughly captures the variability of the ganglion cells' input, taking presynaptic processing into account. Any nonlinear gain control mechanism in the ganglion cell layer should therefore be sensitive to this quantity. For some cells exceeded , while for others the reverse was true. Measuring the gains in both stimulus conditions ( and ), however, we found systematic adaptation opposing the change in effective contrast: gain tended to increase when effective contrast decreased and vice-versa. Specifically, the quantities and were significantly anticorrelated (Fig. 6G; Spearman's , , ).
Finally, we assessed whether the receptive field changes reported above could account for the observed modest increase in output correlations between white noise and the spatio-temporal exponential stimulus. For experiments using spatio-temporally exponential noise, as discussed above, we measured the adaptation in LN model parameters fit to each stimulus. We then separately examined the effect of adaptation in different parameters on the excess correlations predicted by the LN models. Including adaptation of the linear filters, but not the gain, produced a significantly improved match between the model and the data (Fig. 4B, “filter adaptation model”). Additionally allowing the gain to adapt produced output correlations consistent with the data (Fig. 4B, “filter+gain adaptation model”). The contribution of gain adaptation to decorrelation is interesting in light of our observation that output correlations are lower for stimuli with lower contrast (Fig. 3C). Low contrast stimuli generally evoke lower firing rates, which could result in decreased pairwise correlations absent any change in linear filtering properties. (See Text S1 for a derivation of this result and Fig. S3B for evidence that excess correlation tends to increase with, but is not fully determined by, average firing rate.) At the same time, changes in contrast lead to gain control, wherein gain is higher for lower stimulus contrast. This gain adaptation could also affect output correlations, as in Fig. 4B. It would be interesting to know how gain control interacts with changes in other properties, such as the nonlinearity threshold and the shape of the linear filter, to set the correlations in the retinal response. Note that the LN model is fit to each neuron independently, without taking correlations between neurons into account. Its successful prediction of the change in pairwise correlations, without explicit introduction of inter-neural interactions, is therefore noteworthy. We conclude that observed adaptation in receptive fields and gains is adequate to explain the output correlations in responses to a spatio-temporally correlated stimulus.
Our principal finding is that the mammalian retina maintains a moderate level of output correlation across a wide range of spatio-temporally correlated stimuli ranging from white noise checkerboards (with limited correlations) to natural movies (with wide spatial and temporal correlations). While the amount of output correlation varies between stimuli, the changes are much less than predicted by a non-adapting linear-nonlinear functional model. Our data also suggest a differential effect of spatial versus temporal correlations on the functional properties of the retinal output. We focused here on spatial variations in our stimuli, but it would be interesting to design future studies to explore the space-time differences more systematically. In response to spatio-temporal exponential noise, where the receptive fields could be estimated, we showed that the relative invariance of output correlations is largely accounted for by the observed changes in the linear receptive field (faster temporal kernels and slightly stronger surround inhibition for more correlated stimuli) and by changes in the nonlinear gain (anti-correlated to changes in effective contrast). The latter findings give an interpretation of the results in terms of a conventional functional model (here a linear-nonlinear cascade), but the measurement of output correlations is model-independent.
Classifying cells into classes revealed a slight dependence of excess correlation on cell class: most robustly, opposite polarity ON-OFF pairs showed the greatest increase in correlation magnitude when stimulus correlation increased. Indeed, if the retinal output is split across parallel functional channels, redundancy is likely to be highest within a channel due to shared circuit inputs. It may thus be advantageous, from an information encoding perspective, for decorrelation to act within a channel, with residual correlations across classes signaling to downstream areas relevant relationships between the information in different channels.
Pitkow and Meister [14] showed that salamander retina partly decorrelates naturalistic inputs but that the response to white noise is more correlated than the input, in part due to receptive field overlap between ganglion cells. Here we demonstrated a similar phenomenon in mammalian retina: consistent with their results, we found that changes in output correlations were often smaller than changes in input correlations. We also extended their findings by showing explicitly that this partial decorrelation occurs in individual pairs of neurons. In [14], it was also suggested that the linear receptive field measured from white noise was insufficient to explain the amount of decorrelation seen for naturalistic stimuli, and the bulk of the decorrelation was attributed to changes in the threshold of a functional model of ganglion cells. However, the authors did not directly measure the (possibly different) receptive fields of ganglion cells responding to correlated stimuli, nor did they follow particular cell pairs across different stimuli. Our measurements suggest that the nonlinear processing proposed in [14] can be described in terms of adaptation of the linear receptive field and nonlinear gain with the net effect that output correlations are reduced relative to the expected input-induced correlations, as was observed in visual cortex by [8]. Our results also recall those of [16], [23], [17], and [24], who showed that receptive fields in LGN and primary visual cortex differ in structure when probed with natural movies versus random stimuli.
We also found that the gain of retinal ganglion cells responding to correlated stimuli changes with “effective contrast” and , i.e. with the standard deviation of the input to the nonlinearity in a linear-nonlinear model of ganglion cells. In classical contrast gain control, firing rates and response kinetics adapt to temporal contrast and to the spatial scale of stimuli [21], [22]. As increased stimulus correlation may produce a qualitatively similar input to the inner plexiform layer as increased contrast, some of the cellular mechanisms underlying contrast adaptation might also contribute to the phenomena we have uncovered. This provides an avenue for future study of the functional mechanisms underlying adaptation to correlations.
We have focused in the present work on the failure of a non-adapting linear-nonlinear model to capture the relatively small scale of observed excess correlations and have seen that adaptation in the linear filter might remedy this discrepancy. Alternatively, shared circuitry in the population of neurons may be engaged by correlated inputs and require explicit inclusion in any functional model of retinal responses to different classes of correlated stimuli [9], [25]. Such shared circuitry leads to noise in one neuron being passed to multiple nearby neurons and is thus measured by “noise correlations.” While addition of fixed, stimulus-independent noise correlation would not greatly change our results, a change in noise correlation with stimulus would provide a different candidate mechanism to account for our data [26]. This is another avenue for future work.
We have focused here on the effects of spatial correlations in an experimental design where we could compare receptive fields computed from responses to two different stimuli. Thus, we used relatively weak exponential correlations to ensure that we were not measuring artifacts of the stimulus correlations themselves. Recovering receptive fields from strongly correlated stimuli can require long recording times. Because our experimental design involved comparisons between several different stimuli, we were only able to recover receptive fields for moderately correlated stimuli. Future work could simply present each stimulus for a longer duration to assess receptive field changes at a population level rather than analyzing multiple stimuli in one experiment. In such experiments with more data from each cell, alternative methods of receptive field estimation such as Maximally Informative Dimensions [17] could potentially be applied. Further work could also include parallel studies with stimuli including temporal correlations only to complement our findings on responses to spatial correlations.
Finally, it would be interesting to determine the timecourse of the adaptations observed here. Knowing whether a change in stimulus correlations induces changes in receptive fields and output correlations within seconds, tens of seconds, or longer would help to clarify the relationship between processing of correlations and adaptation to other stimulus features such as contrast. Again, the design of our experiments precluded making these measurements – we focused on long segments to measure steady-state processing of correlations, whereas assessing the timecourse of changes requires finer and more systematic sampling of transitions between stimuli.
Why would the retina need to adapt, in the behaving animal, to variations in spatial correlations? While natural scenes are scale-invariant on average, the specific correlations do vary depending on the scene and the viewing distance (see Fig. 1A). Barlow originally suggested that sensory systems should decorrelate their inputs to make efficient use of limited neural bandwidth [1]. Consistent with this idea, we found that retina removes redundancies in spatio-temporally correlated stimuli but also that the retinal output is not completely decorrelated. Rather, the output correlations are reduced to a lower level, roughly similar to correlations in responses to white noise checkerboards when considered relative to the much larger input-induced correlations predicted by a non-adapting functional model of neural response. What drives this tradeoff? Recall that redundancy can be useful to protect against noise, to facilitate downstream computations, or to enable separate modulation of information being routed to distinct cortical targets. Thus, it may be that a certain degree of output correlation between retinal ganglion cells represents a good balance between the benefits of decorrelation and the benefits of redundancy [5]. Sensory outputs with varying amounts of correlation may also be decoded differently by cortex [6], in which case maintaining a fixed visual code might require that retinal output correlations are within the range expected by downstream visual areas. In these interpretations, it makes sense that the retina adapts to maintain correlation within a relatively narrow range across stimulus conditions, as we have found.
This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, as well as the guidelines of the American Veterinary Medical Association. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Pennsylvania (Permit Number 803091). All surgery was performed under anesthesia, and every effort was made to minimize suffering.
We recorded retinal ganglion cells from Hartley guinea pig using a 30-electrode array (30 µm spacing, Multi Channel Systems MCS GmbH, Reutlingen, Germany). After anesthesia with ketamine/xylazine (100/20 mg/kg) and pentobarbital (100 mg/kg), the eye was enucleated and the animal was euthanized by pentobarbital overdose. The eye was hemisected and dark adapted. The retina was separated from the pigment epithelium, mounted on filter paper, and placed onto the electrode array, ganglion cells closest to the electrodes. Extracellular signals were recorded at 10 kHz. The retina was maintained in well-oxygenated bath of Ames' medium at a temperature of . The health of the preparation was monitored by tracking the average firing rates of active cells. Recording times were 2–4 hours, a typical duration over which the guinea pig retina preparation remains robustly responsive. We tested the consistency of responses offline by comparing activity levels near the beginning and end of the recording. We also measured the responses to a flash of light immediately before and after presentation of our main experimental stimuli. If any of these measures changed greatly, we took this as a sign of poor health and discarded the corresponding dataset. Spike times were extracted with the spike-sorting algorithm described in [27]; briefly, a subset of data was manually clustered to generate spike templates that were then fit to the remaining data using a Bayesian goodness-of-fit criterion.
Data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.246qg.
We displayed checkerboard stimuli (see Fig. 1B) at 30 Hz on a Lucivid monitor (MicroBrightField inc., Colchester, VT) and projected the image onto the retina. The mean luminance on the retina was (low photopic); each check occupied an area between 50 µm×67 µm and 100 µm×133 µm. To make white noise and exponentially correlated stimuli, we first produced random checkerboards with intensities drawn from a Gaussian distribution. Spatio-temporally correlated stimuli were produced by filtering the Gaussian stimulus with an exponential filter with a time constant of three stimulus frames (99 ms) and a space constant of one check to match the scale of typical receptive fields. Stimuli with only spatial exponential correlations were constructed similarly, but with a time constant set to zero. To create the multiscale stimulus, we first generated gaussian white noise checkerboards at several power-of-two scales. The largest scale consisted of a single check filling the entire stimulus field, the next largest was a 2×2 checkerboard (with check size equal to half the stimulus field), the third largest was a 4×4 checkerboard (check size one quarter of the stimulus field), and so on. The checkerboards at all scales were then summed and thresholded to obtain a binary stimulus qualitatively mimicking the scale-invariant structure of spatial correlations in natural scenes (Fig. 1B). This stimulus did not contain temporal correlations. Natural movies of leaves and grasses blowing in the wind were collected with a Prosilica GE 1050 high-speed digital camera with a 1/2″ sensor (Allied Vision Technologies GmbH, Stadtroda, Germany) connected to a laptop running StreamPix software (NorPix Inc., Montreal, Canada) to grab frames at 60 fps. The camera resolution was set to 512×512 pixels, and movies were filmed from a fixed tripod about 5 feet from the trees and grass. Natural light was used to illuminate our outdoor scenes, and exposure time was set ( µs) to capture variation in shadows and avoid saturation of our 8-bit luminance depth. Videos were collected for up to 30 minutes; 10 second to 5 minute segments with continuous motion were selected. Videos were downsampled to match the resolution and frame rate of our stimulus monitor. When we analyzed responses to movies taken from different settings, we did not see a sizable change in output correlation (Fig. S4C); thus, we combined all natural videos in our analysis. To produce a scrambled control for natural movies, pixels were randomly shuffled in space and time to remove all correlations. All stimuli other than natural movies (intact and scrambled) were thresholded at the median to fix the mean luminance and single-pixel variance and to maximize contrast. This binarization did not affect the power spectra significantly. For low-contrast stimuli, all deviations from the mean luminance were halved to give an overall contrast of 50%. Typically, we alternated 10-minute blocks of white noise with 10-minute blocks of a correlated stimulus.
We used reverse correlation to compute the spike-triggered average (STA) for each cell responding to white noise. We performed principal component analysis (PCA) on the best-fitting temporal kernels and split cells into two clusters based on the sign of the first component; the clusters were identified as ON and OFF classes based on the sign of their temporal kernels. (Our under-sampling of OFF cells [4], [28] may be due to electrode bias, as individual OFF cells are smaller and therefore less likely to be detected by our electrode array.) PCA was repeated for the ON and OFF groups separately. We manually identified clusters based on the projections onto the first three principal components; in this way we identified four functional classes, including slow-OFF, fast-OFF, fast-ON, and slow-ON (see Fig. 2B).
To obtain spatio-temporal receptive fields (STRFs) for both white and exponentially correlated stimuli, we used publicly available code (strflab.berkeley.edu; [20]) to carry out maximum likelihood estimation. We parameterized the model by a linear filter acting on the stimulus and a logistic nonlinearity, so that firing probability is , where represents the linear filter output, and and are gain and offset parameters. Gradient ascent with early stopping was used to compute a maximum likelihood estimate of the linear filter that best fit the data. We initialized the algorithm for each neuron using the spike-triggered average recorded in response to white noise. Many cells do not yield clear receptive fields when probed with white noise, either because this stimulus does not evoke a sufficiently strong response or because the response is not well modeled as a single linear filter. To avoid potential artifacts that could arise from including such cells in our receptive field and model analyses, we selected cells whose receptive fields had clearly visible centers. This classification of receptive fields as high-quality was done before any other data analysis in order to avoid biasing the selection. In datasets where we obtained receptive fields for both white noise and a correlated stimulus the designations of high-quality agreed between the two stimuli for 98% of cells. The subset of cells identified in this way also had center locations that were clearly delineated by our automated receptive field analysis, giving confirmation of our visual threshold.
The STRF baseline was poorly constrained by the maximum likelihood procedure, since an additive change in the STRF has a similar effect to a proportional shift in the offset parameter of the nonlinearity. We therefore normalized the STRFs by subtracting an estimate of the baseline: we allowed the fit to include components extending 100 ms after the spike — where the true filter must be zero by causality — and subtracted the mean of these frames. Inclusion of these post-spike frames also allowed us to verify that the temporal autocorrelations in our stimuli did not produce any acausal artifacts in the recovered STRFs. We normalized the estimated linear filters to have unit Euclidean norm (square root of the sum of squares of filter values) and then used gradient ascent to separately fit the gain and offset of a logistic nonlinearity. Since the likelihood function in this case is convex, there was no possibility of local maxima. While we were able to compute unbiased estimates of STRFs from responses to stimuli with exponential correlations, our multiscale and natural movie stimuli were too correlated to estimate unbiased receptive fields with the number of spikes we were able to obtain in a single recording. Maximally Informative Dimensions, an important alternative receptive field estimation method [29], would similarly be constrained by the number of spikes obtainable when exploring multiple stimulus conditions in a single recording session, as we have done.
Correlations were measured as the correlation coefficient between pairs of simultaneously recorded neurons. Spike trains were divided into 33 ms bins; we assigned a bin a 1 if it had one or more spikes and a zero otherwise. The results reported above did not change if we used spike counts in each bin rather than binarizing. Indeed, 98% of timebins had one or fewer spikes and less than 0.05% had more than three spikes. We summarized the results by finding the best fit line of the form , where and are the pairwise correlations under the white and correlated stimuli, respectively. We estimated the excess correlation, , by the total least squares regression method and computed 95% bootstrap confidence intervals from 50,000 bootstrap resamples of the set of ganglion cell pairs.
Such instantaneous correlations are thought to combine slow stimulus-driven effects with fast intrinsic effects due to shared noise [13]. To verify that this did not affect our results, we isolated the stimulus-driven component, by noting that our cross-correlation functions can feature a short-timescale peak riding on a slow component and extracting the latter. Specifically, we binned the spike trains into 1 ms bins and computed cross-covariance functions between pairs. To isolate the stimulus-induced component, we smoothed the cross-covariance functions by fitting a cubic B-spline curve with knots spaced at 20 ms to suppress the fast noise component. We then found the shift, within a 200 ms window, which maximized the absolute value of the smoothed cross-covariance and estimated the correlation coefficient as the cross-covariance at this shift normalized by the product of the standard deviations. This gave excess correlation values consistent with those reported above (not shown).
We also computed the power spectra of the stimuli, the best-fitting temporal kernels, and the filter outputs (i.e. stimulus power spectra multiplied by filter power spectra). We summarized each power spectrum by computing the total power above 5 Hz divided by the total power below 5 Hz.
Given a STRF estimated for one cell under one of the stimulus conditions, we first performed principal component analysis on the timecourses of the individual pixels. From the resulting set of “principal timecourses” we selected the one most similar to the timecourse of the pixel that achieves the peak value in the full STRF. The complete linear filter was collapsed into a single frame by finding the projection of each pixel onto this principal timecourse. This procedure is equivalent to finding the best (least squares) spatio-temporally separable approximation to the STRF: , where and are the spatial and temporal components of the approximation. From the single frame , we extracted the center and surround regions. To find the center, we began with the peak pixel and then recursively expanded the region in a contiguous patch to include any pixels whose timecourses had at least a 50% correlation with already included pixels. We ended the recursive process after the first pass in which no pixels were added to the center. At this point, all pixels not included in the center were considered part of the surround for the purpose of assessing the surround strength.
Taking the center defined in this way as a mask for the full STRFs, we summed all pixel values within the center at each time point to generate a temporal profile for the central receptive field. To obtain temporal kernels with greater precision than the 30 Hz time scale of our STRFs, we used cubic spline interpolation with knots spaced every 33 ms. From the interpolated timecourses, we measured the time to peak under each stimulus for the center. In addition, the peak value of this temporal profile was taken to be the center weight of the receptive field. Similar computations yielded the surround time to peak and surround weight. We then quantified the relative surround strength as the ratio of surround weight to center weight.
In addition, we measured the gain of each neuron under each stimulus condition. To test for contrast gain control, we defined “effective contrast,” , as the standard deviation of the linear filter output. To avoid ambiguity between linear filter amplitude and gain, we normalized each STRF to have unit Euclidean norm before computing the gain and the effective contrast.
We used the analysis method described here because it gave the most robustly unbiased results in our simulations (see below), but we also wanted to verify that our results did not change dramatically with slightly different methods (see details in Text S2 and Table S1). Briefly, we made a series of modifications to our receptive field extraction method and repeated the analyses described in the main text for each modification.
To validate our STRF analysis methods, we generated synthetic data using a linear-nonlinear (LN) model. We then applied STRF extraction and analysis methods identical to those applied to real data to check that the known LN parameters were extracted in an unbiased fashion. The linear filter was chosen to be spatio-temporally separable, with the temporal component taken from measured ganglion cell responses. The spatial filter was modeled as a difference-of-Gaussians, where the size and strength of the surround Gaussian relative to the center Gaussian were chosen to mimic receptive fields of real neurons. In each simulation, parameters for 100 model neurons were chosen independently. The results are summarized in Table 2.
In our first simulation, the surround radius (relative to center radius) was chosen from a Gaussian distribution with mean 2 and standard deviation 0.3, the relative surround strength from a Gaussian distribution with mean 1 and standard deviation 0.1, and the offset coordinates from Gaussian distributions with mean 0 and standard deviation 2 (“Standard model” in Table 2). For each model neuron, the same filter was applied to the spatio-temporal exponentially correlated and uncorrelated stimuli in order to simulate cases without adaptation. Across the population, our model neurons showed only a slight bias in center latency between the two stimuli (Fig. S1A). While this effect reaches significance (for ), the effect size is orders of magnitude smaller than that seen in the data and thus could not explain our experimental results. We also observed a tendency toward a slightly stronger relative surround strength under white noise than under correlated noise (Fig. S1B). Note that this is opposite the effect observed in our experimental results (Fig. 6D–F). Thus, if anything our results may be stronger than reported in the main text.
To further validate our analysis we ran simulations with an even wider range of model parameters. We first constructed model neurons with surround radii measured from Gaussian distributions with means of 1 (“Small surround radius” in Table 2) or 3 (“Large surround radius”), both with standard deviation 0.3, and all other parameters the same as in our original simulation. In separate simulations, we kept the mean surround radius at 2 but chose the relative surround strength from a Gaussian distribution with mean 0.5 (“Small surround weight”) or 2 (“Large surround weight”), both with standard deviation 0.1. As with our original set of parameters, the models recovered from STRF analysis had at most slight biases toward weaker surrounds and shorter center times to peak under correlated noise (see Table 2).
|
10.1371/journal.pntd.0003472 | The Influence of HIV and Schistosomiasis on Renal Function: A Cross-sectional Study among Children at a Hospital in Tanzania | Schistosomiasis and HIV are both associated with kidney disease. Prevalence and factors associated with abnormal renal function among HIV-infected children in Africa compared to uninfected controls have not been well described in a schistosomiasis endemic area.
This cross-sectional study was conducted at the Sekou Toure Regional Hospital HIV clinic in Mwanza, Tanzania. A total of 122 HIV-infected children and 122 HIV-uninfected siblings were consecutively enrolled. Fresh urine was obtained for measurement of albuminuria and Schistosoma circulating cathodic antigen. Blood was collected for measurement of serum creatinine. Estimated glomerular filtration rate (eGFR) was calculated using the modified Schwartz equation. Renal dysfunction was defined operationally as eGFR<60mL/min/1.73m2 and/or albuminuria>20mg/L in a single sample. Among 122 HIV-infected children, 61/122 (50.0%) met our criteria for renal dysfunction: 54/122 (44.3%) had albuminuria>20mg/L and 9/122 (7.4%) had eGFR<60. Among 122 HIV-uninfected children, 51/122 (41.8%) met our criteria for renal dysfunction: 48/122 (39.3%) had albuminuria>20mg/L and 6/122 (4.9%) had eGFR<60. Schistosomiasis was the only factor significantly associated with renal dysfunction by multivariable logistic regression (OR = 2.51, 95% CI 1.46–4.31, p = 0.001).
A high prevalence of renal dysfunction exists among both HIV-infected Tanzanian children and their HIV-uninfected siblings. Schistosomiasis was strongly associated with renal dysfunction.
| Ninety percent of schistosomiasis occurs in sub-Saharan Africa, where 91% of HIV-infected children reside. Both schistosomiasis and HIV affect the kidney, but their respective effects on kidney disease in children are not well described. Our prior work in HIV-infected adults demonstrated a high prevalence of kidney disease, possibly worsened by schistosomiasis, but adults are less commonly and less heavily infected with schistosomiasis than children. Therefore, we sought to describe the prevalence and factors associated with renal dysfunction (defined operationally as eGFR <60mL/min/1.73m2 and/or albuminuria >20mg/L in a single urine test) among HIV-infected children and their uninfected siblings living in a schistosomiasis endemic area. We found that half of HIV-infected children and more than one third of HIV-uninfected children had renal dysfunction. Schistosomiasis was the only factor significantly associated with renal dysfunction, increasing odds of renal dysfunction by 2.5-fold. Nearly 50% of the renal dysfunction we observed in both groups could be explained by schistosomiasis. The strong association between schistosomiasis and renal dysfunction among both HIV-infected and uninfected children should remind clinicians to screen for schistosomiasis. It also ought to spur further prospective research to assess for causality and outcomes in the relationship between S. mansoni and kidney disease in children.
| HIV remains common in sub-Saharan Africa (SSA) where 91% of HIV-infected children reside and 1 in every 20 adults is infected [1, 2]. Kidney disease is an important complication in HIV-infected individuals and is associated with an increased risk of morbidity and mortality [3, 4]. The prevalence of kidney disease among HIV-infected adults in high-income countries ranges from 5%–50%, and is most common in patients of African descent [4]. Among more than 300 HIV-infected adults starting ART at our own hospital in Tanzania, 70% had evidence of kidney disease [5, 6].
Kidney disease among children with HIV is less well described [7]. In studies from SSA, the prevalence of markers of kidney disease among children with HIV varied greatly, ranging from 0–31.6%, depending on the methods used to evaluate the kidney [8–18]. In the few studies that included HIV-uninfected control subjects for comparison, the controls had less markers of kidney disease (0–6% versus 0–20.5%) [8, 9, 13–15]. None of these studies determined both estimated glomerular filtration rate (eGFR) and albuminuria. For this reason, the total prevalence of kidney disease among HIV-infected children in SSA remains unknown, and it is difficult to know if the kidney disease observed among children in these studies was due to HIV infection itself, medication use, or other factors that might be common among African children.
Therefore we conducted a cross-sectional study among HIV-infected children and their HIV-uninfected siblings in an area where kidney disease is common among HIV-infected adults. The objectives of our study were to determine the prevalence and correlates of renal dysfunction (defined operationally as eGFR <60mL/min/1.73m2 and/or albuminuria >20mg/L in a single urine test) and to compare the prevalence of renal dysfunction among HIV-infected and HIV-uninfected children. Our hypothesis was that the prevalence of undiagnosed renal dysfunction would be 30% and 5% among HIV-infected and uninfected children, respectively, and that active schistosome infection would be associated with renal dysfunction.
This cross-sectional study was completed between August—December 2013 in the outpatient HIV clinic at Sekou Toure Regional Hospital. Sekou Toure Hospital is located in the city of Mwanza, along the shore of Lake Victoria in northwestern Tanzania, and it serves a population of approximately 2.7 million people. The outpatient HIV clinic of Sekou Toure follows approximately 475 children who are referred to the clinic from the surrounding community-based voluntary counseling and testing centers in Mwanza. In our study, we enrolled HIV-infected children ages 2–12 years old who were attending the Sekou Toure HIV clinic. All of the mothers of these HIV-infected children were also HIV-infected, so we assume that these children were perinatally infected. The caretakers of all enrolled HIV-infected children were invited to bring uninfected siblings between the ages of 2–12 years for enrollment as controls. We tested siblings for HIV using the Determine HIV-1/2 rapid antibody test (Alere Medical Co., Ltd, Chiba, Japan) as recommended by the Tanzanian National HIV Guidelines [19]. We excluded children with fever and those for whom urine samples could not be obtained.
At least one parent or guardian for each child was interviewed. Both HIV-infected and uninfected children were examined. A structured questionnaire was used to collect demographic information, past medical history, and clinical symptoms. Additional information collected from HIV-infected children included a history of opportunistic infections, antiretroviral therapy (ART) status, and WHO clinical stage. In both groups (HIV-infected and uninfected siblings), 4 milliliters of blood were drawn at the time of enrollment to measure the random blood glucose and serum creatinine as well as CD4+ T-cell counts for HIV-infected children. Clean, midstream urine samples were collected for most children, except for a small group of children ≤ 4 years old in whom urine bags were used for urine collection. Urine samples were collected between 8–10AM, after ≥2 hours of fasting. Children ≤4 years old who did not urinate within 1 hour were excluded in order to minimize the inconvenience to their mothers. In order to determine the species of Schistosoma and intensity of infection in this population, we additionally obtained 10mL of urine and fresh fecal samples on 40 consecutive children who were CCA positive.
Random blood glucose was measured using a OneTouch® glucometer (LifeScan, Inc., Milpitas, California, USA). Fresh urine samples were tested immediately for albumin using a dipstick (Micral B, Roche, Mannheim, Germany). Patients were considered to have albuminuria if the urine albumin concentration was above 20 mg/L, as per instructions provided by the manufacturer and according to our prior research [5, 6, 20], and since a concentration of >20 mg/L has been demonstrated to correlate well with elevated albumin excretion rates by standard laboratory methods [21]. A urine dipstick (Multistix 10SG, Siemens, USA) was used to test for leukocyte esterase, hematuria, nitrates, glucose and ketones. The fresh urine sample was also tested using a circulating cathodic antigen (CCA) cassette test (Rapid Medical Diagnostics, Pretoria, South Africa) to detect active Schistosoma infection. The CCA test indicates active schistosome infection and can be positive in the urine during infection with either species of schistosomes that are endemic in Tanzania (S. mansoni and S. haematobium), though its sensitivity is lower in S. haematobium [22–24]. CCA pointofcare testing is used widely and has been found to be more sensitive than the gold standard KatoKatz stool diagnosis of S. mansoni, particularly for lighter infections [25]. Following the manufacturer’s instructions, any visible line in the “test” area was considered positive. Line intensities were graded as “1” (test line very faintly visible), “2” (test line visible but lighter than control line), “3” (test line equal to control line), and “4” (test line darker than control line). For fecal samples, five slides were prepared using the Kato-Katz technique, as this has been shown to have a sensitivity comparable to examining stool specimens collected on different days [26]. Ten milliliters of fresh urine was filter concentrated and examined immediately by microscopy for both contamination and for S. haematobium. Intensity of infection was quantified as S. mansoni eggs per gram of stool and S. haematobium eggs per 10mL of urine. The laboratory personnel who performed these analyses were blinded to the HIV-status of the study subjects.
Serum creatinine was measured using a COBAS Integra 400 Plus clinical chemistry machine (Roche, Germany), calibrated by the Creatinine Jaffe 2 Method. An estimated glomerular filtration rate (eGFR) was calculated using the modified Schwartz equation as recommended by the Kidney Disease Improving Global Outcomes (KDIGO) guidelines [27].
Renal dysfunction was defined operationally as an eGFR ≤60 ml/min/1.73 m2 and/or albuminuria >20 mg/L on a single urine dipstick. The severity of renal dysfunction was classified as Stage 1 (albuminuria + eGFR≥90), Stage 2 (albuminuria + eGFR 60–89), Stage 3 (eGFR 30–59), Stage 4 (eGFR 15–29) or Stage 5 (eGFR<15). Given the age of our study population (2 to 12 years), we defined malnutrition according to the BMI-for-age child growth standards from the WHO, defining severe malnutrition as a z-score of ≤-3 [28]. We also reported weight-for-age and height-for-age for those children <5 years old.
The primary outcome was renal dysfunction as defined above. Based on the two-sample proportions Fisher’s exact test, we calculated that 122 children would be needed in each group to provide >95% power (at p = 0.05) to detect the difference in prevalence of renal dysfunction between the two groups that we hypothesized (30% and 5% among HIV-infected and uninfected children respectively) while also providing >80% power to show an association between schistosomiasis and renal dysfunction if the prevalence of schistosomiasis was 50% in the children with renal dysfunction and 25% in the children without renal dysfunction.
Data was entered into Microsoft Excel 2010 and analyzed using STATA version 12 (STATA Corporation, San Antonio, Texas). Categorical variables were described as proportions (%), and continuous variables were described as medians [interquartile range]. Univariable logistic regression analysis was performed to determine which baseline characteristics were associated with renal dysfunction. All variables significantly associated with renal dysfunction by univariable analysis were subjected to a predetermined multivariable logistic regression model, which also automatically included the variables for age, gender, and HIV status. The validity of the multivariable logistic regression model was assessed using the likelihood ratio test and by assessing for interactions. In addition, the linearity assumption was checked for continuous variables by comparing models with these variables represented continuously versus categorically. After schistosomal infection was found to be associated with renal dysfunction, we decided that we should perform an additional univariable logistic regression analysis to determine which specific markers of renal dysfunction were associated with schistosomiasis. P values of less than 0.05 were considered statistically significant.
Ethical approval for the study was obtained from the Research and Publications Committee of Bugando Medical Centre (under whose jurisdiction Sekou Toure Regional Hospital falls), as well as the Institutional Review Board of Weill Cornell Medical College. Informed written consent was obtained from the parents and assent was obtained from children ≥8 years. The results of all tests were reported immediately to the clinician caring for the child for the sake of further management. All children with schistosomiasis were treated according to the Tanzanian National Guidelines with 40 mg/kg of praziquantel.
During the study period, 139 HIV-infected children were seen at the Sekou Toure HIV clinic, and 17 were excluded for the following reasons: 8 were not able to provide urine samples, 5 had an acute, febrile illness, and 4 parents did not consent. In the end, 122 HIV-infected children were enrolled. For all enrolled HIV-infected children, parents were invited to bring HIV-uninfected siblings between the ages of 2–12 years for enrollment as controls. A total of 132 siblings were screened, and 10 were excluded for the following reasons: 6 were not able to provide urine samples, and 4 had an acute, febrile illness. A total of 122 siblings were enrolled. The characteristics of excluded children did not differ between the 2 groups.
Table 1 describes the baseline characteristics of the 2 groups. Among the 122 HIV-infected children, the median age was 8 years [4–11] and 63/122 (51.6%) were female. Among the 122 HIV-uninfected children, the median age was 8 years [5–10], and 55/122 (45.1%) were female.
Historical factors that were significantly different between the 2 groups included cough (26/122 [21.3%] vs. 12/122 [9.8%], p = 0.01), history of tuberculosis (10/122 [8.2%] vs. 1/122 [0.8%], p = 0.005), recurrent pneumonia (11/122 [9.0%] vs. 3/122 [2.5%], p = 0.03), and papular pruritic eruptions (12/122 [9.8%] vs. 2/122 [1.6%], p = 0.01). Physical exam factors that were significantly different between the groups were pallor (10/122 [8.2%] vs. 2/122 [1.6%], p = 0.02), thrush (4/122 [3.3%] vs. 0/122 [0%], p = 0.04) and lymphadenopathy (23/122 [18.9%] vs. 3/122 [2.5%] p=<0.0001). No children reported history of neurologic disease and none had signs of neurologic disease on physical examination. The BMI-for-age z-scores were similar in the 2 groups. For the children <5 years old, the weight-for-age and height-for-age z-scores were also similar in the 2 groups (-0.76 [-1.57–0.05] vs. -0.48 [-1.12–0.21], p = 0.47 and -1.58 [-2.16–0.13] vs. -1.33 [-2.54–0.49], p = 0.47 respectively).
Table 2 describes the renal dysfunction outcomes (defined operationally as eGFR <60mL/min/1.73m2 and/or albuminuria >20mg/L in a single urine test). Among 122 HIV-infected children, 61/122 (50.0%) met our criteria for renal dysfunction: 54 (44.3%) had albuminuria, and 9 (7.4%) had an eGFR <60. Among 122 HIV-uninfected children, 51 (41.8%) met our criteria for renal dysfunction: 48 (39.3%) had albuminuria and 6 (4.9%) had an eGFR < 60.
Table 3 shows the results of the univariable analysis for factors associated with renal dysfunction. In the univariable analysis only the presence of schistosomiasis was significantly associated with renal dysfunction (OR = 2.51, 95%CI 1.46–4.31, p = 0.001). Higher intensity of schistosomiasis was associated with higher prevalence of renal dysfunction (OR = 1.3 (1+) vs. OR≈4 (2+/3+), p for trend = 0.001). By multivariable logistic regression analysis including schistosomiasis, age, sex and HIV status, schistosomiasis remained the only factor associated with renal dysfunction (OR = 2.40, 95%CI 1.37–4.17, p = 0.002). HIV infection was not significantly associated with renal dysfunction by either univariable or multivariable analysis. There was more renal dysfunction among children with higher WHO clinical stage, but this relationship was not statistically significant (p = 0.23). Of note, though, the association between schistosomiasis and renal dysfunction was somewhat stronger among HIV-infected children than their HIV-uninfected siblings (OR = 3.04, 95%CI 1.39–6.68, p = 0.005 vs. OR = 2.08, 95%CI 0.97–4.46, p = 0.06).
Table 4 compares the prevalence of renal dysfunction among Schistosoma infected children and Schistosoma uninfected children regardless of HIV status. An eGFR less than 60 ml/min/1.73m2 was more common in the Schistosoma infected children (7.2% versus 5.6%, OR 2.61, 95% CI 1.35–5.04, p = 0.01). There was also a higher prevalence of albuminuria among Schistosoma infected children (55.9% versus 34.4%, OR 2.92, 95% CI 1.61–5.27, p = 0.001). The overall prevalence of renal dysfunction was 60.7% among Schistosoma infected children and 38.1% among uninfected children (OR 2.51, 95% CI 1.46–4.31, p = 0.001).
None of the 40 subjects who provided urine for microscopy had eggs of S. haematobium detectable by microscopy. Of the 7 subjects from whom stool samples were available, 6/7 (85.7%) had S. mansoni eggs detected, with concentrations ranging from 17 to 75 eggs per gram.
In our study, half of the HIV-infected children attending an HIV clinic in the Lake Zone of northwestern Tanzania had evidence of renal dysfunction (defined operationally as eGFR <60mL/min/1.73m2 and/or albuminuria >20mg/L in a single urine test): 44.3% had albuminuria >20mg/L and 7.4% had an eGFR <60 ml/min/1.73 m2. These rates are higher than those found in other studies from SSA. Four other countries in SSA (Burkina Faso, Democratic Republic of Congo, Nigeria, and Zimbabwe) have reported the prevalence of markers of kidney disease among HIV-infected children as ranging from 0–31.6% using methodologies similar to ours [8–18]. The reasons for the higher prevalence of renal dysfunction among HIV-infected children in the Lake Zone compared to prior studies in SSA is not known, but our findings are consistent with the findings among HIV-infected adults in our region [5, 6].
Surprisingly, the prevalence of renal dysfunction was equally high among HIV-uninfected siblings. More than one-third of HIV-uninfected siblings had evidence of renal dysfunction: 39.3% had albuminuria >20 mg/L and 4.9% had eGFR<60 ml/min/1.73 m2. In other studies among children in SSA that have included HIV-uninfected controls, the prevalence of markers of kidney disease in the control group has been low (0–6%) [8, 9, 13–15]. The higher prevalence of renal dysfunction among HIV-uninfected children in our study is likely related to the unique nature of our control group. In order to minimize differences in socioeconomic factors, household exposures and genetics, we chose HIV-uninfected siblings as our control group, whereas prior studies have all used children from the pediatric outpatient clinics of the hospital in which the study was conducted as their controls [8, 9, 13–15]. The high prevalence of renal dysfunction that we observed among HIV-uninfected controls could therefore be related to either in utero HIV exposure or a high population prevalence of renal dysfunction. Because the dates of maternal HIV infection and viral suppression were not known, we could not confirm the HIV-exposure status among HIV-uninfected control siblings, but we suspect that most if not all were exposed since they were born within ~5 years of their HIV-infected sibling. HIV exposure, even in the absence of infection, has been associated with multiple abnormalities that may affect the development and function of the kidney in childhood [29–31]. On the other hand, the high prevalence of renal dysfunction among controls could also reflect a high community prevalence of renal dysfunction among children, possibly related to the known high prevalence of schistosomiasis. In order to investigate this possibility, we are currently planning a longitudinal study to examine a cohort of HIV-uninfected, unexposed school children in the Lake Zone.
Schistosomiasis was strongly associated with renal dysfunction among HIV-infected and uninfected children (OR = 2.51, 95% CI 1.46–4.31), and a higher intensity schistosome infection was associated with even more renal dysfunction (p = 0.001 for trend). Approximately 46% of children with renal dysfunction had schistosomiasis by CCA testing compared to 25% of children without renal dysfunction. Multiple studies and epidemiological evidence have shown that S. haematobium infection may cause kidney disease (particularly albuminuria) in both adults and children [32–34], but all of the 40 consecutive urine samples we tested were negative for S. haematobium, and <20% of CCA-positive subjects had the hematuria that is typical for this infection. The dominant species of schistosomiasis in our region, S. mansoni [35, 36], is also known to cause glomerulonephritis and kidney disease [37]. The prevalence of proteinuria among adult subjects infected with S. mansoni has been reported to be as high as 20% in Egypt [38, 39] and 15% in Brazil [40, 41], and the severity and irreversibility of the disease has been demonstrated in several clinicopathologic and experimental studies [42–44]. The association between markers of kidney disease and S. mansoni in children, by contrast, has only been investigated in two small studies in SSA [45, 46] which did not find an association between S. mansoni and overt proteinuria, but neither one of them investigated both eGFR and albuminuria as we have done. If further studies confirm the relationship between S. mansoni infection and kidney disease among children, schistosomiasis may become an important target for prevention of kidney disease in our population.
Although schistosomiasis could explain a large proportion of the renal dysfunction in our study, 55% of children with renal dysfunction did not have evidence of current schistosomiasis. In these subjects multiple other factors may be contributing to their renal dysfunction such as acute glomerulonephritis, infections (malaria, recurrent diarrhea), or genetic factors. The APOL1 gene mutation, for example, is known to be associated with renal disease in African populations [47], younger age of onset of kidney disease [48] and faster decline in kidney function [49]. Further studies are needed to determine the factors other than schistosomiasis that might be contributing to the high prevalence of renal dysfunction that we observed. Kidney disease occurring at a young age may lead to end stage renal disease or other complications (e.g. hypertension) in adulthood; therefore targeted screening for early detection of kidney disease and treatment of reversible factors are high priorities. Many simple diagnostic tools such as light microscopy of the urine sediment are currently underutilized in screening efforts. In addition, the implementation of proven strategies for prevention and treatment, such as early antimicrobial therapy for severe infections and rapid correction of hypovolemic shock, must be accelerated.
Our study has several limitations. First and foremost, this was a cross-sectional study and, therefore, neither confirmation of chronic kidney disease nor causality in the relationship between schistosomiasis and kidney disease could be examined. In addition, several gold standard investigations for HIV and kidney disease, such as quantitative HIV viral load testing, urine albumin-to-creatinine ratio, and kidney biopsies were not performed since they were not available in our region at the time of the study. Our operational definition of renal dysfunction may have resulted in information bias, with over diagnosis of kidney disease, and we are currently planning to assess our findings with a study using the standard KDIGO definition of chronic kidney disease. Also, the exclusion of children who could not produce urine samples, as well as the possibility that HIV-infected patients may have received previous praziquantel therapy could have caused some selection bias and underestimation of the population prevalences of renal dysfunction and schistosomiasis, respectively.
In conclusion, our study identified a high prevalence of renal dysfunction (defined operationally as eGFR <60mL/min/1.73m2 and/or albuminuria >20mg/L in a single urine test) among HIV-infected Tanzanian children attending our pediatric HIV clinics. Almost 50% of both HIV-infected children and their siblings had renal dysfunction, and 6% had an eGFR <60 mL/min/1.73 m2. Surprisingly, the prevalence of renal dysfunction among HIV-uninfected siblings was similar to the HIV-infected children. This may be related to either in utero HIV exposure or a high community-wide prevalence of renal dysfunction in children, and further studies are urgently needed to distinguish these possibilities. Schistosomiasis was strongly associated renal dysfunction in this population, and the predominant species of schistosomes in our region is S. mansoni (not S. haematobium). Schistosomiasis may be an important target for prevention of kidney disease in children in sub-Saharan Africa.
|
10.1371/journal.pbio.1002070 | Cofilin1 Controls Transcolumnar Plasticity in Dendritic Spines in Adult Barrel Cortex | During sensory deprivation, the barrel cortex undergoes expansion of a functional column representing spared inputs (spared column), into the neighboring deprived columns (representing deprived inputs) which are in turn shrunk. As a result, the neurons in a deprived column simultaneously increase and decrease their responses to spared and deprived inputs, respectively. Previous studies revealed that dendritic spines are remodeled during this barrel map plasticity. Because cofilin1, a predominant regulator of actin filament turnover, governs both the expansion and shrinkage of the dendritic spine structure in vitro, it hypothetically regulates both responses in barrel map plasticity. However, this hypothesis remains untested. Using lentiviral vectors, we knocked down cofilin1 locally within layer 2/3 neurons in a deprived column. Cofilin1-knocked-down neurons were optogenetically labeled using channelrhodopsin-2, and electrophysiological recordings were targeted to these knocked-down neurons. We showed that cofilin1 knockdown impaired response increases to spared inputs but preserved response decreases to deprived inputs, indicating that cofilin1 dependency is dissociated in these two types of barrel map plasticity. To explore the structural basis of this dissociation, we then analyzed spine densities on deprived column dendritic branches, which were supposed to receive dense horizontal transcolumnar projections from the spared column. We found that spine number increased in a cofilin1-dependent manner selectively in the distal part of the supragranular layer, where most of the transcolumnar projections existed. Our findings suggest that cofilin1-mediated actin dynamics regulate functional map plasticity in an input-specific manner through the dendritic spine remodeling that occurs in the horizontal transcolumnar circuits. These new mechanistic insights into transcolumnar plasticity in adult rats may have a general significance for understanding reorganization of neocortical circuits that have more sophisticated columnar organization than the rodent neocortex, such as the primate neocortex.
| Plasticity in the adult neocortex is the basis of our learning and memory. However, its molecular mechanisms are still unclear. In the sensory barrel cortex of rodents, a well-characterized model for neocortical plasticity, neurons directly code for whisker displacement—neurons within a given barrel will fire when the whisker that that barrel represents is moved. Strikingly, the deprivation of all but a single whisker alters the original representations—cortical columns representing the deprived inputs shrink and that representing the spared inputs expands, intruding into the surrounding deprived columns. Because single-neuron-level structural changes are suggested to be involved in this plasticity, here we focused on cofilin1, a protein that is known to modulate the cytoskeleton and to regulate the structure of dendritic spines. We induced experience-dependent plasticity in the D1 column by sparing only the D1 whisker, and knocked down the expression of cofilin1 in the D2 column. Cofilin1 knockdown differentially affected plasticity, such that experience-dependent increases in spared input representation were impaired, whereas decreases in deprived input representation were intact. We then found that during these plastic changes, the density of dendritic spines increased in a cofilin1-dependent manner around the connections between the D1 and D2 columns. Cofilin1-dependent density increase was observed only in the most superficial part of the cortex but not in deeper parts, consistent with the distribution patterns of axons that transmit spared and deprived information, respectively. These results suggest that cofilin1 regulates neocortical functional plasticity through the remodeling of dendritic spines within circuits that connect columns.
| Experience-dependent plasticity (EDP) in adult neuronal circuits is considered to form the basis of learning and memory [1–3]. EDP is also related to the recovery of cortical responses after the disruption of peripheral inputs [4–6]. The rodent barrel cortex provides a key model system for studying adult EDP in which response field patterns (or functional maps) can change in a sensory experience-dependent fashion [7–9]. In this regard, the response field corresponding to a spared whisker (spared cortical barrel column) expands during sensory deprivation into neighboring cortical columns that correspond to deprived whiskers (deprived columns) [10]. The deprived inputs associated with these columns in turn shrink [11]. The primary locus of these plastic changes is layer 2/3 (L2/3) in adult barrel cortex [12,13]. An L2/3 neuron increases its responses to spared inputs and simultaneously decreases its responses to deprived inputs [14].
These plastic changes have been suggested to be mediated by different neuronal circuits. Horizontal transcolumnar projections from neighboring column neurons (spared column → deprived column L2/3) [15,16] are important for map expansion because electrolytic lesioning of this pathway prevents plasticity [17] and synaptic transmission in this pathway is potentiated after sensory deprivation [18]. In contrast, ascending intracolumnar projections from layer 4 (deprived column L4 → deprived column L2/3) are depressed during shrinkage of deprived whisker representation [19].
Morphological studies indicate that sensory deprivation promotes the turnover of dendritic spines in the rodent barrel cortex [20]. Within dendritic spines, actin filaments are highly concentrated [21,22] and provide the structural foundation for synaptic plasticity [23–25]. Actin depolymerizing factor (ADF)/cofilin regulate dendritic spine structure through their actin filament-severing and monomer-binding activities [26]. ADF/cofilin are thus predominant regulators of dendritic spine structure and synaptic plasticity [27–32]. Indeed, ADF/cofilin govern both the expansion and shrinkage of the spine structure of the hippocampal neuronal dendrite in vitro [28,33], and postnatal knockout impairs both stimulus-induced long-term potentiation (LTP) and depression (LTD) in hippocampus [31]. However, it remains untested whether ADF/cofilin regulate both directions of response changes during adult EDP. In the present study, therefore, we investigated causal impact of perturbation of ADF/cofilin function on two different components of barrel map plasticity: spared input expansion and deprived input shrinkage.
For this purpose, we knocked down cofilin1 (CFL1) within excitatory neuron in an L2/3- and deprived (D2) column-restricted manner (Fig. 1A). This strategy enabled us to examine the impact of CFL1 knockdown (KD) only in the direct postsynaptic neurons (in this case, the L2/3 excitatory neurons of the deprived column) involved in the horizontal transcolumnar and ascending intracolumnar connections. We found that the response field expansion of spared whisker input was impaired by CFL1 KD in the deprived column, while the response field shrinkage of deprived whisker input was preserved. We then explored the mechanistic insights for this dissociation in the effects of CFL1 KD, and found that spine densities increased in a CFL1-dependent manner at the dendritic branch segments around spines connecting with transcolumnar projections, selectively in a part of the supragranular layer where dense transcolumnar projections were observed. These results provide the first direct evidence that a CFL1-mediated change in synaptic connectivity underlies the EDP in a circuit-specific manner.
To manipulate CFL1 gene expression, we used a microRNA (miRNA)-based gene KD system in which the polymerase II promoter was available for driving miRNA expression. Because several weeks are typically required for the induction of EDP [10,14], we employed a lentiviral vector for the stable in vivo expression of miRNA for CFL1 KD (Fig. 1B). The vectors co-expressed channelrhodopsin-2 (ChR2)-enhanced yellow fluorescent protein (eYFP), with ChR2 as a light-activatable tag for extracellular single-unit recordings [34–38] and eYFP as a fluorescent marker of CFL1 KD neurons, under the control of the excitatory neuron specific Ca2+/calmodulin-dependent protein kinase II alpha (CaMKIIα) promoter [39].
Targeted injection of the lentiviral vector into the right D2 barrel column of rats was achieved by functionally identifying the column center via intrinsic signal optical imaging (Fig. 1C-1F) [40]. Single column-restricted expression of eYFP was confirmed in tangential sections that were processed via cytochrome oxidase staining (Fig. 1G and 1H). By targeting the vector injection to a shallow depth within the cerebral cortex (~300 μm from the pial surface), viral infection was restricted to L2/3 (Fig. 1I), and strong eYFP expression was confined to L2/3 (Fig. 1J). Owing to the existence of axonal projections from L2/3 to L5 [16], weak fluorescence derived from eYFP-positive axons originating from L2/3 neurons was observed in L5 (Fig. 1J). Immunostaining of a neuronal marker, microtubule-associated protein 2 (MAP2), confirmed that most of the eYFP+ neurons resided within L2/3, whereas eYFP+ neurons were rarely found in L5 (90.1% in L2/3 versus 0.2% in L5) (Fig. 1J and 1K). These results demonstrate that viral expression was mostly restricted to the L2/3 neurons in the D2 barrel column.
In the present study, the CFL1 KD experiments made use of a negative control miRNA (miR-Neg) and two miRNAs (miR-CFL1_1 and miR-CFL1_2) with different target sequences within the CFL1 gene. The KD efficiencies of miR-CFL1_1 and miR-CFL1_2 were first assessed in vitro. Both miRNAs showed high KD efficiency at the mRNA level in rat CFL1-overexpressing human embryonic kidney (HEK) 293T cells (miR-CFL1_1, 95.0 ± 0.9%; miR-CFL1_2, 89.3 ± 0.3%; n = 3: miR-CFL1_1, p = 2.7 × 10-8; miR-CFL1_2, p = 2.8 × 10-8 versus miR-Neg, Dunnett’s multiple comparison test) (Fig. 2A). Similar results were observed at the protein level in rat pheochromocytoma-12 (PC-12) cells (miR-CFL1_1, 96.8 ± 1.5%; miR-CFL1_2, 83.8 ± 5.7%; n = 3: miR-CFL1_1, p = 0.0036; miR-CFL1_2, p = 0.0073 versus miR-Neg) (Figs. 2B and S1A). We next confirmed CFL1 KD in vivo. CFL1 protein expression is observed in neuronal somata, dendritic spines, and astrocytes in the normal cortex [21]. Immunostaining of CFL1 in a miR-CFL1_1- or miR-CFL1_2-expressing rat showed that CFL1 expression decreased locally in an L2/3 subregion corresponding to the region where eYFP expression was observed in the adjacent section (Figs. 2C, 2D, and S1B). This finding is in clear contrast with the miR-Neg-expressing control section, where no decrease in CFL1 protein expression was observed (Fig. 2E and 2F). Double staining of CFL1 and NeuN in a miR-CFL1_1-expressing rat revealed that CFL1 immunoreactivity decreased in neurons within eYFP+ region compared to eYFP− region (Fig. 2G and 2H). Indeed, percentage of CFL1-positive cells in NeuN-positive cells significantly decreased in eYFP+ region (eYFP− region, 81.6 ± 1.2%; eYFP+ region, 17.8 ± 4.4%; n = 3; p = 0.0003, t-test with Bonferroni’s correction) (Fig. 2I). This percentage did not decrease in miR-Neg-expressing rats (eYFP− region, 75.2 ± 3.8%; eYFP+ region, 71.7 ± 3.0%; p = 0.51, t-test) (Fig. 2I). These results demonstrate that expression of miR-CFL1 knocked down CFL1 in neurons and this effect was not due to overexpression of ChR2-eYFP or miRNA.
To ensure the specificity of miR-CFL1, we next examined mRNA expression levels of three genes related to CFL1 (ADF, Twinfilin 1 and 2) [41] in PC-12 cells. Expression of miR-CFL1 through the infection of Lenti-CMV-ChR2-eYFP-miR-CFL1_1 or-CFL1_2 did not affect the mRNA levels of these genes (ADF; miR-CFL1_1, p = 0.11; miR-CFL1_2, p = 0.50: Twinfilin1; miR-CFL1_1, p = 0.66; miR-CFL1_2, p = 0.99: Twinfilin2; miR-CFL1_1, p = 0.86; miR-CFL1_2, p = 0.69: versus miR-Neg, Dunnett’s multiple comparison test), although CFL1 expression significantly decreased (miR-CFL1_1, p = 2.6 × 10-7; miR-CFL1_2, p = 5.2 × 10-7 versus miR-Neg) (Fig. 2J). As for ADF, which is closely related to CFL1 in terms of structure and function [26], we examined its expression at the protein level both in PC-12 cells and rats expressing miR-CFL1. ADF protein expression was not significantly affected by CFL1 KD (miR-CFL1_1, p = 0.094; miR-CFL1_2, p = 0.078: versus miR-Neg, Dunnett’s multiple comparison test), although there was a slight increase (Figs. 2K, 2L, and S1C). This tendency of increase is consistent with the previous observations in CFL1 knockout mice [31]. These data clearly demonstrate the specificity of genetic manipulation mediated by miR-CFL1.
In the present study, extracellular single-unit recordings were only taken from regular-spiking neurons (S2A–S2C Fig.) [42]. For efficient recording from CFL1 KD neurons that existed only within a small cortical region (approximately, a sphere with a radius of 200–250 μm) (Fig. 1H and 1I), we searched CFL1 KD neurons that co-expressed ChR2 (Fig. 1B) with illuminating blue laser (peak wavelength: 473 nm). Furthermore, to exclude neurons that weakly expressed or did not express ChR2, we used only the data of light-responsive L2/3 neurons that showed high response reliability [37] to repetitive light (20 Hz) stimulation (S2D–S2K Fig.) (for details, see Materials and Methods).
To examine the effects of CFL1 KD on EDP, lentiviral vectors were injected at 2 weeks before the onset of sensory deprivation (Fig. 3A). Sensory deprivation was induced by using the single whisker experience protocol [7,10], in which all whiskers but the D1 whisker were trimmed on the left side of the face. We first confirmed that the L2/3 neurons in the D2 column of the right hemisphere showed increased responses to spared D1 whisker stimulation after sensory deprivation in wild-type (WT) rats (WT non-deprived versus WT deprived, p = 2.7 × 10-6, Tukey-Kramer’s multiple comparison test) (Fig. 3B-3E). This observation indicates that the cortical representation of spared whisker inputs expanded into surrounding deprived columns (Fig. 1A). In contrast, the neuronal response increase was almost completely absent in the putative ChR2+ neurons from which recordings were taken in the miR-CFL1_1-expressing deprived rats (WT deprived versus miR-CFL1_1 deprived, p = 3.3 × 10-7) (Fig. 3C-3E). On the other hand, expression of miR-CFL1_1 in non-deprived rats did not affect neuronal responses to D1 stimulation (WT non-deprived versus miR-CFL1_1 non-deprived, p = 0.91) (Fig. 3C-3E), indicating that CFL1 KD in and of itself does not decrease responses to the D1 whisker.
We performed three lines of control experiments. The first control experiment demonstrated that the enhanced spared whisker response was not altered in neurons in which miR-Neg and ChR2-eYFP were co-expressed (WT deprived versus miR-Neg deprived, p = 0.98) (S3A Fig.). This finding suggests that the overexpression of miRNA and ChR2-eYFP itself does not affect the expansion of spared input representation. The second control experiment demonstrated that CFL1 KD with miR-CFL1_2 also impaired the increase in the spared whisker response as well as miR-CFL1_1 (WT deprived versus miR-CFL1_2 deprived, p = 4.1 × 10-6) (S3A Fig.). This finding suggests that the observed effects were not due to “off-target” actions of miR-CFL1 [43]. Finally, the effect of CFL1 KD on experience-dependent response increase was weaker in neurons determined as putative ChR2− than those determined as putative ChR2+ both in miR-CFL1_1 and miR-CFL1_2 (F1, 90 = 9.01, p = 0.0035, main effect of factor 1; factor 1, neuron type; factor 2, miR type; two-way ANOVA: miR-CFL1_1, p = 0.017; miR-CFL1_2, p = 0.15; ChR2+ versus ChR2−, t-test with Bonferroni’s correction) (S2L Fig.), validating further that experience-dependent potentiation to D1 deflections was impaired in miR-CFL1-expressing D2 neurons.
To exclude the possibility that biased recording locations within the D2 column affected our results, we performed another analysis. Recording locations in the D2 column were reconstructed based on lesion marks (Fig. 3F and 3G). Responses recorded from CFL1 KD neurons of deprived rats were lower than that recorded from WT deprived rats, regardless of the distance from the D1 column (Fig. 3H) and the cortical depth (Fig. 3I) (comparison of distance distribution, WT deprived versus miR-CFL1_1 deprived, p = 0.37; comparison of depth distribution, WT deprived versus miR-CFL1_deprived, p = 0.19, Mann-Whitney’s U-test). Accordingly, the observed effects of CFL1 KD in this study were not due to recording location bias.
To further confirm the specificity of the effects of miR-CFL1, we next examined whether the impaired experience-dependent increase in neuronal responses (Fig. 3C-3E) recovers by expression of a mutant CFL1 resistant to miRNA. We first designed three resistant CFL1s (resCFL1s) that had seven or eight point mutations, which did not change the amino acid sequences, within the miR-CFL1_1 target sequence (21 bp) (Fig. 4A). Impaired CFL1 expression by miR-CFL1_1 was indeed rescued by resCFL1 expression in vitro, and the efficiency of expression recovery was highest in resCFL1_1 (resCFL1_1, 36.3 ± 3.4%; resCFL1_2, 32.8 ± 1.7%; resCFL1_3, 28.4 ± 2.8%; n = 4: resCFL1_1, p = 5.8 × 10-6; resCFL1_2, p = 2.2 × 10-5; resCFL1_3, p = 1.3 × 10-4 versus miR-CFL1_1 group, Tukey-Kramer’s multiple comparison test) (Fig. 4B). Therefore, resCFL1_1 (henceforth “resCFL1”) was selected for in vivo experiments. We also confirmed that the expression level of resCFL1 was not affected by miR-CFL1_1 in vitro (p = 0.30, t-test) (Fig. 4C). We injected the lentivirus which co-expressed resCFL1 and mCherry into the same cortical region (D2 barrel column) where the lentivirus expressing ChR2-eYFP/miR-CFL1_1 was also injected (Fig. 4D and 4E). After inducing EDP by whisker deprivation, the responses of putative ChR2+ neurons were selectively recorded. This procedure assured miR-CFL1 was expressed in the recorded neurons. Co-expression of ChR2-eYFP and mCherry in the infected cortical area was confirmed by histological analysis (Fig. 4F and 4G). We showed that responses to D1 whisker deflections were significantly larger in neurons expressing both miR-CFL1_1 and resCFL1 than in neurons expressing only miR-CFL1 (miR-CFL1+resCFL1 deprived versus miR-CFL1 deprived, p = 0.0069, Tukey-Kramer’s multiple comparison test) (Fig. 4H-4J). These data clearly demonstrate that the effects of miR-CFL1 expression on experience-dependent potentiation were not due to non-specific effects of miR-CFL1 on genes other than CFL1, but due to an impairment of CFL1 function.
The same set of cells from which we recorded responses to D1 stimulation was also tested for D2 stimulation (S1 Data). We first confirmed that the L2/3 neurons in the D2 column showed decreased responses to deprived D2 whisker stimulation after sensory deprivation in WT rats (WT non-deprived versus WT deprived, p = 0.0009, Tukey-Kramer multiple comparison test) (Fig. 5A-5C). This process is indicative of the shrinkage of deprived whisker input representation (Fig. 1A). We then examined the effects of CFL1 KD on this process, and found that experience-dependent response decrease to the deprived D2 whisker was preserved in either miR-CFL1_1-expressing neurons (WT deprived versus miR-CFL1_1 deprived, p = 0.89) (Fig. 5B and 5C) or miR-CFL1_2-expressing neurons (WT deprived versus miR-CFL1_2 deprived, p = 0.89) (S3B Fig.). Expression of miR-CFL1_1 in non-deprived rats did not affect L2/3 neuronal responses to D2 stimulation (WT non-deprived versus miR-CFL1_1 non-deprived, p = 0.81) (S3B Fig.), suggesting that CFL1 KD itself does not affect L2/3 neuronal responses to whisker stimulation.
Taken together, these results suggest that CFL1-mediated actin dynamics is necessary for the experience-dependent expansion of spared input representation in L2/3, but not for the experience-dependent shrinkage of deprived input representation.
To simultaneously visualize the transcolumnar (D1 → D2) projections and dendritic spines of D2 neurons, we next constructed a new set of lentiviral vectors expressing either tdTomato or enhanced green fluorescent protein (eGFP)/miRNA (Fig. 6A and 6B). Injection of these vector solutions was targeted to the D1 column (tdTomato) or the D2 column (eGFP/miRNA) (Fig. 6C). To avoid dense neuronal expression of eGFP, a low-titer solution of the eGFP/miRNA vector (3.0 × 108 − 1.0 × 109 gc·ml−1) was employed (Fig. 6D and 6E). Because the eGFP expression level was low under this low-titer condition and eGFP fluorescence was unendurable for repeated confocal imaging, we used sections stained with an antibody to eGFP for morphological experiments (Fig. 6E). The eGFP/miRNA vector effectively knocked down CFL1 even in the low-titer condition in vivo (S4A and S4B Fig.) and in the low multiplicity-of-infection condition in vitro (S4C Fig.).
Our findings so far indicate that CFL1 dependency dissociates between two types of barrel map plasticity. Together with the fact that CFL1 is a predominant regulator of dendritic spine structure [29,32], it can be hypothesized that CFL1-mediated structural modifications underlie this dissociation; modification of spine structure occurs selectively in cortical regions where horizontal transcolumnar axons emanating from the spared barrel column exist densely. We thus performed the step-by-step test of this hypothesis.
We first found that the tdTomato intensity emanating from the D1 axons peaked at 150–200 μm from the cortical surface in the D2 column, and decreased with cortical depth (Figs. 6F, 6G, and S5A–S5C). This was in clear contrast to the vertical profile of L4 input strength [44], which was fairly weak at shallow depths and stronger at deeper zones, forming a complementary pattern with that of the D1 axonal intensity profile (S5C Fig.). There was a significant difference in tdTomato intensity between the distal (0–200 μm from the cortical surface) and proximal (200–500 μm) portion (p = 0.022, paired t-test) (S5D Fig.). These results suggest that a greater number of transcolumnar synaptic connections from the D1 column are generated in the distal portion of the D2 column than in the proximal portion at which the ascending deprived axonal inputs from L4 are thought to predominate.
We next tested the effects of sensory deprivation on dendritic spine number in the miR-Neg (non-deprived and deprived) groups. Learning/experience-driven dendritic spine formation and synaptic plasticity spatially cluster on dendritic branches in cortical pyramidal neurons [45,46] as well as hippocampal neurons [47], and thus spine densities were measured in dendritic branch segments that are supposed to receive dense transcolumnar inputs. For this purpose, we measured spines around (<15 μm) the identified putative transcolumnar connections (Fig. 6H). In non-deprived rats, spine densities were relatively low at the cortical region just below the surface and increased with depth, while densities were nearly constant throughout the supragranular layer in deprived rats (comparison of slopes of regression lines, miR-Neg non-deprived versus deprived, F1,63 = 9.33, p = 0.0033, F-test) (S5E Fig.).
These results demonstrate that sensory deprivation affects dendritic spine numbers in a cortical depth-dependent manner. We thus separately compiled dendritic spine densities measured at distal and proximal portions of the D2 column supragranular layer, and examined the impact of CFL1 KD on these values. In the distal portion, spine densities significantly increased with sensory deprivation in the control miR-Neg-expressing neurons (miR-Neg non-deprived versus miR-Neg deprived, p = 1.1 × 10-4, Tukey-Kramer’s multiple comparison test), but this increase was impaired in the CFL1 KD neurons (miR-Neg deprived versus miR-CFL1_1 deprived, p = 0.0027) (Fig. 6I-6K). MiR-CFL1 expression under non-deprived condition did not affect baseline spine densities (miR-Neg non-deprived versus miR-CFL1_1 non-deprived, p = 0.97; miR-CFL1_1 deprived versus miR-CFL1_1 non-deprived, p = 0.46), suggesting that the effects of CFL1 KD are specific for the deprivation and that the absence of an increase in spine density in miR-CFL1_1 deprived rats was not due to a general reduction in spine density caused by miR-CFL1_1 expression. In contrast, sensory deprivation did not affect dendritic spine densities in the proximal portion of the supragranular layer (miR-Neg non-deprived versus miR-Neg deprived, p = 0.34; Fig. 6L-6N). Furthermore, the same conclusion was reproduced even if dendritic spine densities were measured within 5 μm around transcolumnar connections (distal; 1.06 ± 0.07, 1.44 ± 0.07, 1.06 ± 0.06, and 1.09 ± 0.05 spines·μm−1 for miR-Neg non-deprived, miR-Neg deprived, miR-CFL1_1 non-deprived, and miR-CFL1_1 deprived groups, respectively; mean ± standard error of the mean [SEM]; miR-Neg non-deprived, p = 0.00056; miR-CFL1_1 non-deprived, p = 0.0005; miR-CFL1_1 deprived, p = 0.0015; versus miR-Neg deprived, Tukey-Kramer’s multiple comparison test) (proximal; 1.18 ± 0.08, 1.31 ± 0.07, 1.24 ± 0.06, and 1.21 ± 0.07 spines·μm−1 for Neg non-deprived, Neg deprived, CFL1_1 non-deprived, and CFL1_1 deprived). These observations suggest that during sensory deprivation, CFL1-mediated actin dynamics causally regulate the spine numbers around dendritic spines receiving horizontal transcolumnar synaptic inputs from the spared D1 column. Moreover, these events take place in the distal portion of the D2 column supragranular layer, where dense horizontal transcolumnar projections reside (Fig. 7).
In addition to the dendritic spine density, we also analyzed the sizes of the D2 spines that made putative synaptic connections with horizontally projecting D1 axons. Sensory deprivation did not influence spine sizes in miR-Neg-expressing neurons in either the distal or the proximal portion (miR-Neg ND versus miR-Neg D; distal, p = 0.99; proximal, p = 0.97; Tukey-Kramer’s multiple comparison test) (S6A and S6B Fig.). In agreement with a previous observation regarding the hippocampal neurons of CFL1 knockout mice [31], spine sizes increased in CFL1 KD neurons in the distal portion (F1, 181 = 10.7, p = 0.0013, main effect of factor 1; factor 1, miR type; factor 2, deprivation type; two-way ANOVA) (S6A and S6B Fig.). However, this increase did not correlate with the neuronal response changes described above (Fig. 3D and 3E), suggesting that the observed spine size changes might not contribute to the functional barrel map plasticity induced by sensory deprivation.
In the present study, CFL1 was locally knocked down in L2/3 excitatory neurons of a deprived column (D2 column) using a lentiviral vector-based RNAi approach. In rats injected with miR-CFL1-expressing vectors, the experience-dependent expansion of the spared input representation was prevented in CFL1 KD neurons, whereas the shrinkage of the deprived input representation was preserved. Furthermore, the spine density around the dendritic spines receiving transcolumnar axonal projections from the spared D1 column was increased in L2/3 neurons of the deprived D2 column, and this increase was impaired by CFL1 KD. These results provide, to the best of our knowledge, the first direct evidence that CFL1-mediated actin dynamics are necessary for plasticity in horizontal transcolumnar circuits during adult cortical EDP.
In rodents, the ADF/cofilin family consists of three genes, namely, ADF, CFL1, and cofilin2. Among their gene products, only ADF and CFL1 proteins are found within the neuronal dendritic spine [21,31]. Importantly, CFL1 knockout mice exhibit impaired synaptic plasticity at hippocampal synapses [31], while ADF knockout mice do not show such deficits [48]. The current investigation therefore focused on CFL1. In the present study, we validated the specificity of the effects of CFL1 KD based on three different lines of evidence. Firstly, we showed that miR-CFL1 expression did not affect the expression levels of genes related to CFL1 including ADF (Figs. 2J-2L and S2C). Secondly, we showed that the effect of CFL1 KD was consistent between the two miR-CFL1s with different target sequences within the CFL1 gene (Figs. 3D and S3A). Because miRNAs can downregulate genes bearing sequences complementary to their seed sequences (positions approximately 2–7 of the guide strand) [43], evidence of the same outcome with different RNAi sequences reduces concerns about the target specificity of RNAi experiments [43,49]. Finally, we showed that impairments of EDP caused by miR-CFL1_1 could be rescued by expression of resCFL1 (Fig. 4). Therefore, our results suggest that the effects observed in the CFL1 KD rats were not due to off-target actions of the miRNAs.
The effects of CFL1 KD on experience-dependent increase of responses to D1 deflections were larger in putative ChR2+ neurons than those in light-responsive putative ChR2—neurons (S2L Fig.). This fact confirmed that experience-dependent potentiation to D1 deflections was impaired in miR-CFL1-expressing D2 neurons. We also showed that responses of light-responsive ChR2—neurons were significantly lower than neurons in WT deprived rats (p = 0.0063, t-test) (S2M Fig.). This fact suggests that the potentiation of putative ChR2—neurons was also slightly impaired. There are two possible explanations for this observation (which are not mutually exclusive): (1) false negative categorization of ChR2+ neuron as ChR2—neurons by our criteria, and (2) an indirect decrease of responses within the D2 column resulting from a response decrease in ChR2+ neurons surrounding (and potentially connected with) light-responsive ChR2—neurons.
MiR-CFL1 expression did not affect spine density in the proximal part of the L2/3 where ascending axons from L4 dominate, whereas it impaired experience-dependent increase in densities in the distal part where transcolumnar axons dominate (Fig. 6I-6N). In addition to these facts, it is also important to note that CFL1 KD itself does not affect basal spine density in non-deprived rats expressing miR-CFL1 (Fig. 6I-6N). These data suggest that effects of CFL1 KD were specific for dendritic branches receiving “potentiated” transcolumnar inputs.
We found that sensory deprivation increases dendritic spine density selectively in a part of the supragranular layer where dense transcolumnar projections were observed (Fig. 6I-6K). This finding is consistent with those from previous studies showing that sensory deprivation promotes the formation of stable dendritic spines in a deprived column located adjacent to a spared column [20] and also increases the density of the horizontal projections from a spared column to the adjacent deprived columns [50]. By contrast, the absence of dendritic spine structural plasticity in the proximal portion of the D2 column supragranular layer (Figs. 6L-6N and S6B) is consistent with the idea that CFL1-mediated actin dynamics do not involve a decrease in deprived input representation because intracolumnar deprived inputs from L4 are thought to prevail in the proximal portion (S5C Fig.) [15,44]. Taken together, our results suggest that dendritic spines receiving horizontal transcolumnar inputs from the spared column are selectively generated, whereas those receiving ascending intracolumnar inputs from L4 remain constant during sensory deprivation.
The previous report demonstrated that ADF could compensate CFL1 function in the presynapse but not in the postsynapse [51]. Therefore, the absence in functional compensation in deprived rats expressing miR-CFL1 (Fig. 3D and 3E) may suggest that the experience-dependent potentiation in transcolumnar circuits is, at least in part, postsynaptic origin. Interestingly, CFL1 is under the control of calcineurin, a regulator of LTD, and is necessary for the dendritic spine shrinkage associated with hippocampal LTD [28]. Given that LTD contributes to the response decrease to deprived inputs in the barrel cortex [19], it is expected that CFL1 also participates in this process. In hippocampal neurons, however, it is also known that spine shrinkage and the decrease in synaptic transmission efficacy during LTD are dissociated processes at the level of molecular pathways [28]. The present data together with these previous observations suggest that the functional depression of the deprived input representation is likewise dissociated from the CFL1-dependent structural changes in the dendritic spines in the rat barrel cortex. However, this view is inconsistent with the observation reported by Rust and colleagues that postnatal CFL1 knockout impairs hippocampal LTD as well as LTP in mice [31]. Although the reason for this inconsistency is unclear, the function of CFL1 in the experience-dependent depression of the deprived input representation in adult cortex may differ from that in stimulus-induced LTD in hippocampal slices.
We labeled the D1 and D2 columns simultaneously (Fig. 6E). The tdTomato-positive structure observed within the D2 column mostly consisted of axons (Fig. 6G), which validates that our measurements for tdTomato intensity (S5A–S5D Fig.) did not include D1 neuron dendrites travelled from the neighboring column and also that tdTomato signals within the D2 column were not derived from ectopic tdTomato expression outside the D1 column caused by a horizontal spillover of the injection. We accomplished the localized cortical labeling by utilizing the property of lentiviral vectors that have relatively large particle size (~100 nm; [52]) and are thus restricted in its diffusion in vivo compared to other vectors such as adeno-associated viral vectors [53]. This labeling technique allowed the demonstration of CFL1-dependent changes in dendritic spine density that could not have been previously accomplished (Fig. 6I-6K). In this analysis, we measured dendritic spines only around the identified putative transcolumnar connections (Fig. 6H). Because tdTomato-expressing area was almost limited within the L2/3 of the D1 column (S7 Fig.), the putative synaptic connections that we identified should consist of monosynaptic connections between D1 L2/3 neurons and D2 L2/3 neurons. On the other hand, it is possible that polysynaptic connections from D1 L2/3 to D2 L2/3 (e.g., D1 L2/3 → D1 L5 → D2 L2/3 [54]) could also have contributed to the dendritic spines that increased around the detected connection, although they are considered to be relatively weak due to the low efficacy of L2/3 → L5 connections [55]. With regard to changes in numbers of synaptic connections, we found it difficult to measure it because the number of D1 neurons labeled with tdTomato (and thus the number of D1 axons that could be detected on D2 dendrites) varied between animals in our small-volume (200 nl) injection method. Indeed, 2-D measurements of the infected areas (on histological sections) in all animals used for spine morphological analysis showed that the sizes of tdTomato-expressing areas varied from the entire supragranular layer of D1 column (~500 μm × 500 μm) to a portion of it (S7 Fig.). This variability in labeled D1 neuron numbers was unavoidable to limit tdTomato labeling within D1 column, and it directly influenced the numbers of synaptic connections that could be detected by this labeling method. It is important to note that it did not affect the spine morphological measurements that were confined to D2 spines connected with D1 axons and the spines around them. Therefore, our strategy may be currently optimal for examining structural modifications in transcolumnar circuits while maintaining between-animal variance at a minimum. However, alternative approaches (e.g., within-animal chronic monitoring of dendritic spines [56]) may also reveal similar conclusions and will be the focus of future studies.
Spine areas measured in the L2/3 distal portion of miR-CFL1 ND and miR-CFL1 D groups showed similar tendency to increase compared to those measured in miR-Neg groups (S6A Fig.). This fact indicates that, regardless of animal’s sensory experience, CFL1 KD itself slightly expands spine sizes. The slight spine expansion was also observed in the hippocampal neurons of CFL1-knockout mice reared under normal environment [31]. Therefore, it is suggested that blocking or decreasing CFL1 action expands dendritic spine sizes independently of each spine’s involvement in sensory EDP. This finding is consistent with the typical role of cofilin as an actin-depolymerizing factor [26]. On the other hand, CFL1 action also promotes actin polymerization [57] and active turnover of actin filaments [29]. Considering these factors together, KD-mediated decrease in CFL1 activity might make dendritic spines more “fixed” state with decreased turnovers and slightly increased sizes. Because active spine turnover is involved in cortical EDP [20,58], our results might suggest that CFL1 regulates EDP by controlling the turnover rates of dendritic spines/actin filaments.
In conclusion, we demonstrate here that CFL1-mediated actin dynamics function in a horizontal connection-specific manner during EDP induced by the single whisker experience protocol. In addition to the rodent vibrissal system, the primate brain is characterized by a highly sophisticated neocortical columnar organization [2]. Cortical circuit reorganization across functional columns through horizontal connections is necessary for functional cortical recovery after peripheral deficits, such as focal retinal lesions [5]. Moreover, plasticity in horizontal transcolumnar or interareal connections is also critical for learning [5]; thus CFL1-mediated circuit reorganization may possibly be a general mechanism for the flexible nature of the human brain. However, this proposal will require further study.
All procedures were performed in accordance with a protocol approved by the University of Tokyo Animal Care Committee (permit number, MED: P11–050). Surgical procedures for lentiviral injection were performed with isoflurane induction (3%) and under maintenance with isoflurane (1%) or ketamine/xylazine (90 mg·kg−1 and 10 mg·kg−1, respectively) anesthesia. Surgical procedures for electrophysiology experiments were performed under ethyl carbamate (1.2 g·kg−1) anesthesia. All efforts were made to minimize suffering and the number of animals employed. Fifty-four male Wistar rats (Nihon SLC) were used for the study (ten were used as WT and 44 were used for viral injection).
Oligonucleotides encoding miRNAs that target the CFL1 gene were designed with BLOCK-iT Pol II miR RNAi expression vector kits and the associated software (Invitrogen). miR-CFL1_1, miR-CFL1_2, and a negative control miRNA (miR-Neg), which is predicted to not target any known vertebrate gene, were purchased from Invitrogen. miR-CFL1_1 and miR-CFL1_2 target two different regions within the CFL1 gene (target sequences: miR-CFL1_1, 5′-AGGAATCAAGCACGAATTACA-3′; miR-CFL1_2, 5′-GTTCGCAAGTCTTCAACGCCA-3′).
HEK293T cells (for miRNA screening) or rat PC-12 cells (for examining effects on endogenous gene expression) were used for the in vitro experiments. For miRNA screening (Fig. 2A), plasmid vectors expressing miR-CFL1 (pcDNA6.2-GW/EmGFP-miR vector; Invitrogen) and the rat CFL1 gene (pCAG-rCFL1, the kind gift of H. Kasai, University of Tokyo) were co-transfected into the HEK293T cells. Three days after transfection, total RNA was prepared and used as a template for real-time reverse-transcriptase PCR with the StepOne Real-Time PCR system (Applied Biosystems).
For resCFL1 screening (Fig. 4B), pcDNA6.2-GW/EmGFP-miR-CFL1, pCAG-rCFL1, and the plasmid vector expressing resCFL1 (pCMV-resCFL1) were co-transfected. For generation of the pCMV-resCFL1, PCR fragments encoding each of the N-terminal and C-terminal side of CFL1 and an annealed oligonucleotide encoding mutated portion of CFL1 (resCFL1_1, 5′-CAAGAAGAAACTGACTGGCATTAAACATGAGCTCCAAGCTAACTGCTACGA-3′; resCFL1_2, 5′-CAAGAAGAAACTGACGGGTATCAAACATGAGCTCCAAGCTAACTGCTACGA-3′; resCFL1_3, 5′-CAAGAAGAAACTGACCGGGATAAAACATGAGCTCCAAGCTAACTGCTACGA-3′) were simultaneously fused to EcoRI/NotI-digested pIRES2-AcGFP1 (Clontech) by using InFusion Cloning kit (Clontech). Primer sequences were as follows: CFL1-N-F, 5′-CTCAAGCTTCGAATTACCGGTATGGCCTCTGGTGTGGCT-3′; CFL1-N-R, 5′-GTCAGTTTCTTCTTGATGGCATCC-3′; CFL1-C-F, 5′-AGCTAACTGCTACGAGGAGGTCAA-3′; CFL1-C-R, 5′-TCTAGAGTCGCGGCCGCTCACAAAGGCTTGCCCTC-3′;
To examine the effects of miR-CFL1 expression on endogenous gene expression (Figs. 2B, 2J, 2K, S1A, and S1C), the PC-12 cells (1 × 105) were transfected with the Lenti-CMV-hChR2-eYFP-miR-CFL1_1 or-miR-CFL1_2 vector (2.0 × 107 gc). Four days later, total RNA was prepared from these cells or the cells were solubilized and the cell extracts were obtained. For protein expression analysis, the extracts were immunoblotted [59] using the following antibody combinations: rabbit antibody to cofilin (1:250; Cytoskeleton) or to destrin (1:1,000; Sigma-Aldrich)/horseradish peroxidase-conjugated antibody to rabbit IgG (1:2,000; Rockland Immunochemicals Inc.), and mouse antibody to β-actin (1:5,000; Sigma-Aldrich)/horseradish peroxidase-conjugated antibody to mouse IgG (1:1,000; Vector Laboratories, Inc.). The band intensities were quantified by using ImageJ software (National Institutes of Health).
The primer sets used for reverse-transcriptase PCR experiments were as follows: CFL1-F, 5′- GCTCTTTTGCCTGAGTGAGG-3′; CFL1-R, 5′-CTTAAGGGGTGCACTCTCG-3′; ADF-F, 5′-GTGCATAGTCGTTGAAGAAGG-3′; ADF-R, 5′-CCTTCGAGCTTGCATAGATC-3′; Twf1-F, 5′-CTGAGTAAGAGACAGCTCAACTATG-3′; Twf1-R, 5′-GCTCTCTTATGCTGCATGTG-3′; Twf2-F, 5′-CTGAAGATGCTGTATGCAGC-3′; Twf2-R, 5′-CTGGTGCTTACTCTCCACAC-3′; GAPDH-F, 5′-TGAACGGGAAGCTCACTGG-3′; GAPDH-R, 5′-TCCACCACCCTGTTGCTGTA-3′.
For generation of the lentiviral transfer vector, pCL20c CaMKIIα-hChR2-eYFP-miR, the PacI/BamHI-digested mouse CaMKIIα promoter (1.3 kb) from pLenti-CaMKIIα-hChR2-mCherry-WPRE (the kind gift of K. Deisseroth) was inserted into MluI/EcoRI-digested pCL20c MSCV-hChR2-eYFP [37] by blunt-end ligation to replace the MSCV promoter. A PCR fragment encoding miR-CFL1 or miR-Neg was inserted into the ClaI site of pCL20c CaMKIIα-hChR2-EYFP using an InFusion Cloning kit. For generation of pCL20c CaMKIIα-eGFP-miR, a PCR fragment encoding miRNA was inserted into the ClaI site of pCL20c CaMKIIα-eGFP [39]. For generation of pCL20c CaMKIIα-mCherry-P2A-resCFL1, PCR fragments encoding each of mCherry (derived from pmCherry-N1 vector, Clontech) and resCFL1 were simultaneously fused to AgeI/NotI-digested pCL20c CaMKIIα-eGFP by using InFusion kit. The P2A sequence was separately added to each of the primers used to amplify mCherry and resCFL1 (primer sequences: mCherry-F, 5′-CCCGGGATCCACCGGCGCCACCATGGTGAGCAA-3′; mCherry-P2A-R, 5′-CTGCTTGCTTTAACAGAGAGAAGTTCGTGGCTCCGGAGCCCTTGTACAGCTCGTCCATGCC-3′; P2A-resCFL1-F; 5′-TGTTAAAGCAAGCAGGAGACGTGGAAGAAAACCCCGGTCCCATGGCCTCTGGTGTGGCTGTC-3′; resCFL1-R, 5′-ATTATCGATGCGGCCTCACAAAGGCTTGCCCTC-3′). The lentiviral vectors were produced and titrated by using the DNA titration method, as described previously [35].
All whiskers, except for the D1 whisker, on the left side of the face were trimmed by cutting the whiskers to fur level (<1 mm) under brief isoflurane anesthesia (3%) by using an anesthetizer (MK-AT200D; Muromachi Kikai). The ipsilateral whiskers were not trimmed. Rats were 10 weeks old at the onset of whisker trimming. Subsequently, the whiskers were re-trimmed every 2 days. Trimming was continued for more than 3 weeks (range 23–45 days) and ceased at 1 week before electrophysiological recordings to stimulate the regrowth of the trimmed whiskers.
Each whisker was inserted into a glass capillary (inner diameter, 0.5 mm) glued to a piezoelectric bending element. A stereoscope was used to insert the whisker to a distance of 10 mm from the whisker pad. For electrophysiological recording, 200 μm ventral-dorsal deflections (10 ms at 1 Hz, repeated 50 times) were applied, resulting in an angular deflection of 1.14°. For intrinsic signal optical imaging, 1–2° amplitude ventral-dorsal deflections (50 ms at 10 Hz, repeated 50 times) were applied.
Rats were 8 weeks old at the time of lentiviral vector injection. Anesthesia was induced with 3% isoflurane and anesthesia was maintained by either 1% isoflurane, ketamine (90 mg∕kg IP)∕xylazine (10 mg∕kg IP). Each rat was positioned in a stereotaxic apparatus (SR-6R; Narishige). The skull over the barrel cortex was carefully thinned to create a cranial window.
Functional maps of the barrel cortex were determined by using intrinsic signal optical imaging. The cortical surface was illuminated with a red light (wavelength, 705 nm) while stimulating a single whisker. Images were collected with a charge-coupled device (CCD) camera (Tokyo Electric Industry) and digitized with an IBM/PC-compatible video system equipped with a video frame grabber board (Matrox Imaging). The imaged area was a 4.2 × 3.1 mm region with a spatial resolution of 320 × 240 pixels. The surface blood vessels were imaged by using a green light (wavelength, 540 nm). The focusing depth was adjusted to 500 μm below the cortical surface. For each recording trial, data were collected for 8 s with a frame length of 0.5 s (16 frames per trial). Reflectance changes in response to whisker stimulation were estimated by subtracting a 3 s averaged frame taken before the onset of whisker stimulation from a 3 s averaged frame taken at the time of whisker stimulation.
To prepare the rats for the electrophysiology experiments, the D2 barrel column was first identified, and a glass pipette (tip diameter, ~40 μm) (sharply [>45°] grinded so as to make depth control easier and to mitigate the spillover of cerebrospinal fluid) attached to a 1 μl Neuros syringe (7001 KH; Hamilton Company) was then vertically inserted into the center of the D2 barrel column to a depth of 300 μm from the cortical surface. Before the pipette insertion, a mannitol solution (25% in saline) was intraperitoneally injected to mitigate the spillover of cerebrospinal fluid. Next, a solution of the lentiviral vector (Lenti-CaMKIIα-hChR2-eYFP-miR vector containing miR-CFL1_1, miR-CFL1_2, or miR-Neg; 200 nl of 1.0 × 1010 gc·ml−1 solution; n = 10, 5, and 6 rats, respectively) or a mixed solution (1:1) of two vectors (Lenti-CaMKIIα-hChR2-eYFP-miR-CFL1_1 [1.0 × 1010 gc·ml−1] and Lenti-CaMKIIα-mCherry-P2A-resCFL1 [1.0 × 1010 gc·ml−1], 200 nl, n = 4 rats) was injected at a flow rate of 25–50 nl·min−1 with the aid of a micropump (UltramicroPump III; World Precision Instruments [WPI]) and a microprocessor-based controller (Micro4; WPI). The needle was left in place for additional 15 min. The scalp incision was carefully sutured, and the rat was returned to a standard cage after recovering from anesthesia.
To prepare the rats for the morphology experiments, both D1 and D2 barrel columns were functionally identified prior to lentiviral vector injection. Solutions containing either the Lenti-CMV-tdTomato-WPRE vector (200 nl; 1.0 × 1010 gc·ml−1) or the Lenti-CaMKIIα-eGFP-miR vector (200 nl; 3.0 × 108 − 1.0 × 109 gc·ml−1; miR-CFL1_1, n = 7 rats; miR-Neg, n = 7 rats) were then injected into the center of the D1 or D2 barrel column, respectively.
A fiber enclosed in the glass coated optrode was coupled to a diode laser (peak wavelength at 473 nm, Omicron Laserage Laserprodukte GmbH). The timing of the stimulation was managed with an electrically controlled mechanical shutter (UNIBLITZ). The light power was controlled with a Neutral Density (ND) filter (Thorlabs). For each neuron, 10 × 10 light pulse trains, each with a duration of 5 ms, were delivered at 1 and 20 Hz. The interval between each train was 15 s [37]. The light intensity was adjusted after observing the neuronal responses so as to avoid the skew of light-evoked spike waveforms from spontaneous spike waveforms [36]. More specifically, the maximum light intensity that did not skew waveforms under visual inspection was employed. The light power at fiber input was in the range of 0.1 − 5 mW.
Each rat was anesthetized with ethyl carbamate (1.2 g·kg−1). The body temperature was maintained at 37.5°C throughout the experiment. A catheter (Natsume Seisakusho) was surgically inserted into the left femoral vein [35]. Ringer’s solution and additional doses of anesthesia (urethane, 0.2 − 0.4 g·kg−1) were then administered through the catheter. The skull over the barrel cortex was exposed and carefully removed. In vivo eYFP fluorescence was identified by using a cooled CCD camera (VB-7000; Keyence) attached to a fluorescence stereoscopic microscope (VB-G05; Keyence).
The activities of single neurons were extracellularly recorded using a glass-coated tungsten microelectrode (impedance < 1 MΩ) in WT rats, or a glass-coated tungsten optrode (impedance < 1 MΩ) [37] in ChR2-expressing rats. The electrode was vertically inserted into the cortex via a hydraulic micromanipulator (MO-10; Narishige). Neuronal signals were amplified with an AB651J amplifier (Nihon Kohden), band-pass filtered (0.4–5 kHz filter; Nihon Kohden), digitized at 25 kHz, and stored by using the Recorder Software (Neural Data Acquisition System). Single units were obtained in the off-line analysis with Offline Sorter Software (Plexon) [60,61]. Briefly, the SD was first calculated to estimate the variance of the baseline noise. Spikes were then extracted using a threshold of >5 × SD from the baseline mean. The information encoded in spike waveforms was compressed using principal component analysis. If waveforms with shapes uncharacteristic of neuronal action potentials were existed, they were excluded before the calculation of principal components. A cluster was selected in 2-D or 3-D feature (typically using the first three principal components) space by drawing a contour manually. The presence of a refractory period was confirmed in the autocorrelogram. If the number of spikes with interspike intervals < 2 ms exceeded 1% of the total for a given unit, the unit was discarded or additional feature combinations were examined to subdivide the cluster further until meeting the criteria in the autocorrelogram [60,61]. Single-unit data were analyzed with MATLAB software (MathWorks).
Recordings were performed from both L2/3 (depth from the pial surface, 0–500 μm) and L5 (800–1250 μm) [62] neurons. In some cases, electrolytic lesions (1 μA, 5 s, tip negative) were applied at a depth corresponding to L4 (750 μm) to map recording locations onto the barrel pattern.
In parallel with the single-unit recordings, cortical electroencephalograms were also recorded to monitor the cortical state. A stainless steel screw was threaded into the bone above the occipital cortex. For reference, another screw was threaded into the bone above the cerebellum. Signals were amplified with an AB-610J amplifier, band-pass filtered (0.5–100 Hz), and stored. During whisker stimulation trials, anesthesia was maintained to a depth equivalent to stage III slow-wave sleep, as described previously [63].
Rats were perfused with saline, followed by 4% paraformaldehyde in phosphate buffer. The brains were post-fixed in 4% paraformaldehyde for 2–4 h and immersed in a solution of 20% sucrose in PBS. To recover the location of the electrophysiological recordings and virus expression, the cortex was flattened between two glass slides, sectioned at 50 μm, and processed for cytochrome oxidase staining (2–3 h at room temperature [RT] in 20 ml phosphate buffer containing 10 mg diaminobenzidine [DAB], 10 mg cytochrome c, and 0.8 g sucrose) [64]. Blood vessels were used as a reference for projecting the barrel patterns in L4 onto the eYFP-expressing L2/3 sections.
Although ChR2-eYFP expression is mostly restricted to the cell membrane [65], weak fluorescence was also observed at the cytoplasm (Fig. 1J). No eYFP fluorescence was observed at the cell nuclei. To confirm L2/3-restricted expression of eYFP, we thus stained ChR2-eYFP-expressing brain sections with an antibody against MAP2, which stains neuronal cytoplasm as well as dendrites [66], rather than with an antibody against neuron-specific nuclear protein (NeuN), which mainly stains nuclei [67].
For the detection of MAP2 or NeuN in selected neuronal populations, coronal sections (25 μm thick) were immunostained with either mouse antibody to MAP2 (1:2,000; Sigma-Aldrich) or mouse antibody to NeuN (1:1,000; Millipore). The sections were reacted with an Alexa Fluor 647-conjugated antibody to mouse IgG (1:500; Invitrogen). Sections were counterstained with the DNA-specific fluorescent dye, Hoechst 33342 (Invitrogen).
To visualize expression of CFL1 or ADF, sections were immunostained with a rabbit antibody to cofilin (1:250) or a rabbit antibody to ADF (1:100), followed by either an Alexa Fluor 647-conjugated antibody to rabbit IgG (1:500; Invitrogen) or a horseradish peroxidase-conjugated antibody to rabbit IgG (Dako Corporation) and DAB. The stained images were obtained by using a BZ-9000 fluorescence microscope (Keyence) and a TCS-SPE confocal microscope (Leica).
Confocal images were used for cell counting. For estimating layer distribution of eYFP+ neurons, sections double-stained with MAP2 and Hoechst were used. Layers were manually identified based on differences in cell density and size. Three non-adjacent sections were chosen that encompassed each injection point, and all eYFP+ neurons were counted in each section. For counting the percentages of CFL1+ neurons, CFL1- and NeuN-stained sections were used. Three region-of-interests (ROIs, 183.3 × 183.3 μm) were randomly selected from both eYFP-expressing region and neighboring normal cortical region (where eYFP was not expressed), and all CFL1+ and NeuN+ neurons were counted within each ROI.
For all morphological analyses, the parasagittal floating sections (50 μm thick) were prepared from the rat cortex that included both the D1 column and the D2 column. Immunohistochemistry was performed as described previously [25]. Briefly, sections were immunostained with chicken antibody to GFP (1:500; Abcam) and rabbit antibody to DsRed (1:500; Clontech), followed by Alexa Fluor 488-conjugated anti-chicken (1:200; Invitrogen) and Alexa Fluor 546-conjugated anti-rabbit (1:200; Invitrogen) secondary antibodies.
For morphometric analysis of dendritic spine density and area on immunostained section [24], confocal immunofluorescence images (voxel size, 0.1 × 0.1 × 0.5 μm3) were acquired with a CSU-22 spinning-disc confocal unit (Yokogawa Electric) coupled to an Axiovert 200M microscope through a Plan Apochromat 63× objective (NA 1.4; Carl Zeiss). The acquired images were then analyzed using MetaMorph software (Universal Imaging Corporation). Given sets of tdTomato+ D1 axons and eGFP+ D2 dendritic spines were defined as synaptically connected if both fluorescent signals were found within the same voxel.
The density of the dendritic spines (including all types of spines, e.g., thin, mushroom-shaped, and stubby spines) was measured at dendritic branch segments that met the following criteria: (1) the segment was nearly parallel to the xy plane (1,280 × 1,010 pixels) in a stacked image of consecutive focal planes (typically, 10–50 stacked planes) taken at 0.5 μm intervals in the z direction; (2) at least one putative synaptic connection with a tdTomato+ axon was identified; (3) no bifurcations were present within the segment; (4) the segment demonstrated no crossing with other branches. For identification of putative synaptic connections, we did not consider overlaps between dendritic shaft and axons. Therefore, all putative synapses were identified on the basis of overlaps between dendritic spines and axons. Branch segments in which the total measured length was less than 4 μm were discarded. Spines within 15 μm of the identified putative synaptic connection were counted, along with dendritic length. In some cases (eight of 51 dendrites in the distal portion and five of 41 dendrites in the proximal portion), two or more connections were identified on a single dendritic branch. If distance between a given connection pair was less than 2 × 15 μm (five of eight in distal and four of five in proximal), two connections were regarded as forming a single branch segment. If this was not the case (three of eight dendrites in distal and one of five in proximal), two connections were regarded as forming different branch segments with each other. We also measured spine densities within 5 μm from connections using the same criteria.
For measurement of the dendritic spine area on immunostained section, the fluorescence images were first thresholded, where the threshold was more than the mean plus 5 × the standard deviation (SD) of the background intensity distribution. We did not normalize intensity levels in each image. Spines that made putative synaptic connections with D1 axons were then identified. Next, a 2-D projection image was reconstructed from the z plane that included the identified spine. Spine area was estimated in the 2-D image by manually enclosing the spine head, followed by measurement of the total pixel number included within the enclosure. All types of spine morphologies were included in the analysis, and all measurements of spine density and spine area were performed by investigators who were blind to the animal’s sensory experience and the identity of the injected virus.
For analysis of axonal density on immunostained section, fluorescence images were acquired with a BZ-9000 microscope through a Plan Apochromat 20 × objective (NA 0.75; Nikon). The intensity of the tdTomato fluorescence attributable to D1 column axons was measured within a region of interest (100 × 50 μm2) localized within the D2 column where the eGFP signal was observed. The region of interest was vertically scanned from the surface of the cortex to the superior end of L4 (depth, 500 μm). Measurements were performed in three sections for each rat. The background intensity (measured at L4 for each section) was first subtracted from the determined intensity of the tdTomato fluorescence, and then the subtracted intensity was normalized to the value of the baseline-subtracted maximum intensity, and then averaged across the three sections. For measurement of tdTomato-positive area (S7 Fig.), fluorescent images were thresholded at 10× background intensity of each section (measured within L4) and binarized. TdTomato-positive area was estimated by manually enclosing the signal-positive area in the binarized image.
Data analysis was performed with MATLAB software. Light-responsive neurons were identified using 1 Hz stimulation data, by comparing firing rates as a function of stimulation latency during the first 100 ms after each light pulse with the firing rates obtained for similar time blocks after shuffling the spike times of each cell within an interval (−100, +100 ms) around stimulation onset [38]. After each shuffling, spikes were counted in 1 ms bins, and the three successive bins that showed the maximum spike numbers during the 100 ms period after stimulation onset were identified. The spike times were shuffled 10,000 times for each cell. Three successive bins with a maximum spike numbers were also identified for the real data. Cells were classified as light-responsive if the number of spikes in the three-bin block with the maximal spike numbers in the real data exceeded the 99.9th percentile value of the distribution of the maximum spike numbers in the shuffled data. The latency of the response was defined as the mean latency of all spikes contributing to this block.
Light-responsive ChR2− neurons were previously reported as showing a lower spike probability for high frequency repetitive light stimulation compared with ChR2+ neurons [34,36–38]. Therefore, repetitive light pulses (20 Hz) were applied to each neuron to examine the probability of spike. The first light pulse in each 10-pulse train was excluded from the analysis. The number of spikes evoked by 90 light pulses was estimated as the spike number detected during the 25 ms after light onset subtracted by the spike number detected during the 25 ms before light onset. In our experimental preparation, ChR2+ somata were almost completely absent in L5, but ChR2+ axons originating from ChR2+ L2/3 neurons were abundant in L5. Thus, light-responsive L5 neurons were considered to be a pure population of “indirectly” activated neurons. Indeed, neurons that reliably responded to repetitive optical stimulation were observed in L2/3, while the firing probability was lower in L5 (S2D–S2G Fig.), consistent with previous reports; thus putative ChR2+ neurons were defined as those neurons exhibiting a higher reliability than the neuron that showed the highest reliability in L5 (S2F and S2G Fig.).
Recording locations were reconstructed based on the relative location of the lesion marks within the barrel patterns, as visualized by cytochrome oxidase staining in tangential sections. To allocate each recording location within the D2 barrel column, the center of the mass was calculated for D1 and D2 columns. A line passing through the center of both D1 and D2 was drawn, and each recording location was then vertically projected onto the line. Distance of each recording location from the D1 column center was measured and normalized to the distance between D1 and D2 centers. In one out of 32 rats (miR-CFL1_1 deprived rats, seven putative ChR2+ units from four recording tracks), we failed to reconstruct the barrel pattern histologically, and thus estimated column centers and recorded locations based on intrinsic signal optical imaging data.
All statistical tests for the in vivo study were performed with MATLAB and freely available R software. Data are given as the means, and error bars denote the standard error of the mean, except when indicated otherwise.
|
10.1371/journal.pcbi.1006022 | Stoichiometric balance of protein copy numbers is measurable and functionally significant in a protein-protein interaction network for yeast endocytosis | Stoichiometric balance, or dosage balance, implies that proteins that are subunits of obligate complexes (e.g. the ribosome) should have copy numbers expressed to match their stoichiometry in that complex. Establishing balance (or imbalance) is an important tool for inferring subunit function and assembly bottlenecks. We show here that these correlations in protein copy numbers can extend beyond complex subunits to larger protein-protein interactions networks (PPIN) involving a range of reversible binding interactions. We develop a simple method for quantifying balance in any interface-resolved PPINs based on network structure and experimentally observed protein copy numbers. By analyzing such a network for the clathrin-mediated endocytosis (CME) system in yeast, we found that the real protein copy numbers were significantly more balanced in relation to their binding partners compared to randomly sampled sets of yeast copy numbers. The observed balance is not perfect, highlighting both under and overexpressed proteins. We evaluate the potential cost and benefits of imbalance using two criteria. First, a potential cost to imbalance is that ‘leftover’ proteins without remaining functional partners are free to misinteract. We systematically quantify how this misinteraction cost is most dangerous for strong-binding protein interactions and for network topologies observed in biological PPINs. Second, a more direct consequence of imbalance is that the formation of specific functional complexes depends on relative copy numbers. We therefore construct simple kinetic models of two sub-networks in the CME network to assess multi-protein assembly of the ARP2/3 complex and a minimal, nine-protein clathrin-coated vesicle forming module. We find that the observed, imperfectly balanced copy numbers are less effective than balanced copy numbers in producing fast and complete multi-protein assemblies. However, we speculate that strategic imbalance in the vesicle forming module allows cells to tune where endocytosis occurs, providing sensitive control over cargo uptake via clathrin-coated vesicles.
| Protein copy numbers are often found to be stoichiometrically balanced for subunits of multi-protein complexes. Imbalance is believed to be deleterious because it lowers complex yield (the dosage balance hypothesis) and increases the risk of misinteractions, but imbalance may also provide unexplored functional benefits. We show here that the benefits of stoichiometric balance can extend to larger networks of interacting proteins. We develop a method to quantify to what degree protein networks are balanced, and apply it to two networks. We find that the clathrin-mediated endocytosis system in yeast is statistically balanced, but not perfectly so, and explore the consequences of imbalance in the form of misinteractions and endocytic function. We also show that biological networks are more robust to misinteractions than random networks when balanced, but are more sensitive to misinteractions under imbalance. This suggests evolutionary pressure for proteins to be balanced and that any conserved imbalance should occur for functional reasons. We explore one such reason in the form of bottlenecking the endocytosis process. Our method can be generalized to other networks and used to identify out-of-balance proteins. Our results provide insight into how network design, expression level regulation, and cell fitness are intertwined.
| Protein copy numbers in yeast vary from a few to well over a million[1, 2]. Expression levels, along with a protein’s binding partners and corresponding affinities, are critical determinants of a protein’s function within the cell. In the context of multiprotein complexes–especially obligate complexes such as the ribosome–it is thought that protein concentrations are balanced according to the stoichiometry of the complex. This is referred to as the dosage balance hypothesis (DBH)[3–5]. Here, we expand this hypothesis to a network wide level, where proteins participate in multiple distinct complexes as well as transient interactions. In these more complex networks (Fig 1A), balance can be defined as having just enough copies of each protein to construct a target vector of complex abundances, with no proteins (or protein binding sites) in significant deficiency or excess. This generalized definition of balance reproduces the expected result for obligate complexes, where, for example, the ARP2/3 obligate complex (Fig 1B) would be balanced if all subunits had equal copy numbers.
For obligate complexes, dosage balance means that there are no leftover subunits, as these would be a waste of cell resources. However, even for proteins in non-obligate complexes a number of deleterious effects could be caused by imbalance. An overexpressed core or “bridge” subunit may sequester periphery subunits, paradoxically lowering the final number of complete complexes[5, 6]. Excess proteins may be prone to misinteractions, also called interaction promiscuity, with nonfunctional partners. Numerous studies have identified proteins with high intrinsic disorder as sensitive to overexpression[7–9], and these proteins have low, tightly regulated native expression levels[10, 11] indicating that misinteraction propensity and abundance are related. Underexpression carries its own dangers: a single underexpressed subunit will become a bottleneck for the whole complex. In addition, weakly expressed proteins are noisier[12] and thus less reliable for the cell. Male (XY) animal cells are known to employ “dosage compensation” mechanisms to increase the expression of X-chromosomal genes to be on par with female cells[13, 14], though for other genes it is the female cell that cuts expression levels in half[15], indicating that the cell preserves an optimized set of expression levels.
But optimized does not necessarily mean balanced. Imbalance may be necessary for functional reasons: signaling networks utilize underexpressed hubs to regulate which pathways are active as a given time[16]. Recent models show imbalance can be beneficial to complex assembly when affinity and kinetics are taken into account[17, 18]. A study of over 5,400 human proteins by Hein et al. found that strong interactions forming stable complexes are correlated with balance, but weak interactions are not, which may mean that the network as a whole is not balanced [19]. Finally, the concept of dosage balance being an optimal set of protein copy numbers generally relies on the assumption that proteins reach an equilibrium state of complex yield. Most processes in the cell do not occur at equilibrium and therefore deviations from balance could be beneficial in non-equilibrium models.
Here, we test the hypothesis that protein expression levels are significantly biased towards balance, even for complex PPINs that include weak and transient interactions. This first required us to develop a method to quantify stoichiometric balance in any arbitrary PPIN, given known binding interfaces and some observed copy numbers, which we call Stoichiometric Balance Optimization of Protein Networks (SBOPN). Copy number correlations thus are evaluated beyond direct binding partners to the more global network of interactors. We then can quantify the consequences of imbalance relative to perfect balance according to two criteria: 1) the deleterious consequences and cost of forming misinteractions, and 2) the potentially beneficial control of specific functional outcomes by modulating which complexes, given known binding affinities, actually assemble. Applied to the 56-protein, manually curated, interface-resolved CME PPIN [20], two of its sub-networks, as well as the ErbB PPIN[16], we find that stoichiometric balance in observed copy numbers is often significant, and observed imbalances, particularly of underexpressed proteins, could provide tuning knobs for functional outcomes.
The first consequence of imbalance we evaluate, misinteractions cost, has an indirect effect on function by allowing unbound proteins to bind to non-functional partners, sequestering components and thus affecting formation of specific complexes. They are believed to play a role in dosage sensitivity[7, 8, 21], and avoiding them has been shown to be an evolutionary force limiting protein diversity[22, 23], expression levels[24, 25], binding strengths[26], and protein network structure[23, 27]. Misinteractions, not being selected for by evolution, are weak and generally unstable, but there are far more ways for N proteins to misinteract (order N2) than bind to their few functional partners (order N) [22, 23]. Cells have evolved a variety of mechanisms to increase specificity, such as allostery[28, 29], negative design[30, 31], compartmentalization[22], and temporal regulation of expression[32]. Copy number balance would be another such mechanism, as protein binding sites would saturate their stronger-binding functional partners.
The second and ultimately more direct consequence of imbalance we evaluate is that changes to copy numbers control which specific and functionally necessary complexes can form. When the central clathrin protein is knocked out in cells, for example, clathrin-mediated endocytosis (CME) is terminated, as clathrin is functionally irreplaceable[33]. The plasma membrane lipid PI(4,5)P2 is also essential for CME, as it is required for recruiting the diverse cytosolic clathrin-coat proteins to the membrane to assemble vesicles[34]. Many clathrin-coat proteins, however, can be knocked out without fully terminating CME[35]. As the CME network illustrates (Fig 1), most of these proteins have multiple domains mediating interactions involving both competitive and non-competitive interactions. Adaptor proteins (proteins that bind to the membrane, to transmembrane cargo, and often to clathrin as well) exhibit redundancy in their binding partners that can partially explain how knock-outs to one protein can be rescued by the activity of related proteins. With simulation of simple kinetic models, we can then test these hypotheses, including for the non-equilibrium production of vesicles at the membrane. Although these models are far too simple to recapitulate the complexities of CME in vivo, they are nonetheless useful in highlighting potential bottlenecks in assembly due to copy numbers or binding affinities.
Quantifying balance in protein networks can thus lead to new insights, as unbalanced proteins may serve as assembly bottlenecks, or maintain alternate cellular functions outside of the network module being analyzed[18]. Dosage balance is also important for understanding dosage sensitivity[4, 21], a phenomenon where overexpression of a gene is detrimental or even lethal to cell growth. Studies estimate ~15% of genes in S. cerevisiae to be dosage sensitive[9, 36], but the negative effects of gene overexpression have been observed in several eukaryotic species including maize[4], flies[37], and humans[38–40]. Studying balance at a network-wide level is challenging because it requires resolved information about the interfaces proteins use to bind. A protein that binds noncompetitively with two partners requires equal abundance to its partners. But if the binding is competitive–i.e. the same interface is used to bind two different partners–the protein’s abundance must equal the sum of that of its partners to have no leftovers (Fig 2). Classic protein-protein interactions networks (PPINs) lack this resolution, but recent studies have begun to add this information, creating what we refer to as interface-interaction networks (IINs)[16, 20, 41]. An IIN tracks not just protein partners but also the binding sites that proteins use to bind.
Our study of stoichiometric balance in larger, interface resolved PPINs is organized in the Results section in three parts. In the first part, we define a metric for quantifying stoichiometric balance and how noise in protein expression levels can be approximately accounted for. We apply our algorithm SBOPN to the CME PPIN [20, 41] and the ErbB PPIN [16], highlighting which proteins are over- and underexpressed relative to perfect balance. Although this analysis excludes temporal expression and binding affinity, it provides a starting point for the analysis of these features in the subsequent parts. In the second part, we switch to generalized interface-interaction network (IIN) topologies and network motifs to focus exclusively on how our first evaluation criteria, the cost of misinteractions under imbalance, is worse for strong binding proteins and for network topologies that resemble biological networks. In the third part, we return to the interface-resolved CME PPIN to evaluate the observed degree of stoichiometric balance in two smaller sub-networks of the CME network: the 7-subunit ARP2/3 complex and a simplified, nine protein, clathrin-coat forming module. In these sub-modules, we now can also evaluate our second criteria and assess how observed copy numbers influence proper multi-protein assembly given known binding affinities of interactions. Our simulations of (non-spatial) kinetic models demonstrate that stoichiometric balance does, in fact, improve multi-protein assembly relative to observed copy numbers, even for the nonequilibrium clathrin-coat assembly module. We speculate that the observed imbalances in clathrin adaptor proteins could offer a mechanism for making the vesicle formation process more tunable, since adaptor proteins are responsible for selecting cargo for endocytic uptake, which is the ultimate purpose of CME.
For a multi-subunit complex such as the ribosome or ARP2/3 complex (Fig 1B), all subunits bind together non-competitively to assemble a functional complex. Stoichiometric balance is simply having enough of each subunit to form complete complexes, with no subunit in excess. But quantifying balance in a general protein-protein interaction network is more challenging because some proteins will bind competitively, using the same interface for multiple interactions. Such proteins will need a higher concentration in order to saturate their functional partners (Fig 2). Thus, to establish stoichiometric balance in a PPIN the binding interfaces must be known. In previous work we analyzed several interface-resolved PPINs, including the 56-protein clathrin-mediated endocytosis (CME) network in yeast [20, 41] (Fig 1A), and the 127-protein ErbB signaling network in human cells[16].
To balance a network, a number of desired complexes may be assigned to each edge and then the number of required interface copies directly solved for. This is constrained with a starting set of copy numbers, C0, otherwise the solution would be arbitrary. However, the inclusion of multiple interfaces per protein introduces a new constraint: interfaces on the same protein should have the same copy number. This constraint often makes nontrivial solutions (i.e. when none of the proteins are set to zero) impossible (see Methods). Therefore, we treat it as a soft constraint, using a parameter “α” to balance its influence. A high α allows more variation of interface copy numbers on the same protein (Fig 2C). We constructed and minimized an objective function using quadratic programming (Methods), which produces a new, optimally balanced set of copy numbers, Cbalanced. For any given interface-resolved PPIN, there can be multiple locally optimized solutions of balanced copy numbers. In Fig 2C we illustrate solutions found by our algorithm SBOPN using the copy numbers of Fig 2B as C0. If we apply our algorithm to Fig 2A, which is an already balanced network, it simply recovers the input copy numbers, such that Cbalanced = C0, regardless of α. Because our algorithm minimizes distance from C0 to Cbalanced, the optimal solutions produce both under and overexpressed proteins.
The benefit of this method is that the distance between C0 and Cbalanced gives you a relative estimate of how “balanced” C0 already is, and thus a metric from which to evaluate the significance of balance in the observed copy numbers. Using real copy numbers taken from Kulak et al.[2], Creal, as C0, we calculated both chi-square distance (CSD) and Jensen-Shannon distance (JSD) between Creal and Cbalanced (Methods). The former metric looks at differences between absolute values and penalizes high deviations more strongly than low deviations, whereas the latter converts both vectors to distributions and measures the similarity between them. We do not expect any networks to have Creal that is already perfectly optimized, such that Creal = Cbalanced. To establish the significance of both distance metrics, we generated 5,000 sets of random C0 vectors, sampled from a yeast concentration distribution. We then measured the CSD and JSD from C0 to Cbalanced for each of these random copy number vectors. If Creal is balanced, its distance metrics should have a significant p-value relative to yeast copy numbers selected randomly from the yeast distribution. The C++ code for our SBOPN algorithm and example input and output files may be downloaded at https://github.com/mjohn218/StoichiometricBalance.
In this second part, we investigate how the cost of imbalance, measured solely in terms of misinteractions, depends on general properties of proteins, including binding affinity and number of binary partners. In a stoichiometrically balanced network, proteins will be driven to saturate their stronger-binding functional partners. Any “leftover” proteins, however, may misinteract, or form non-functional complexes that, while weak, are combinatorically numerous.
In the second part, “Imbalance increases misinteractions dependent on the network topology and binding affinities of proteins”, we only studied binary, competitive interactions. But proteins often bind noncompetitively into higher complexes, and they may interact weakly and thus form few complexes, in which case imbalance may have functional benefits [17, 18]. Furthermore, the above models looked at equilibrium results, whereas many biological systems exhibit non-equilibrium dynamics. We created kinetic models of two modules from the CME network with observed imbalances: the ARP2/3 complex and a simplified vesicle forming protein subset. Simulating higher complex formation is challenging because of the exponentially large number of possible species, so we used NFSim[48], a stochastic solver of chemical kinetics that is rule-based, enabling an efficient tracking of higher-order complexes as they appear in time.
The metrics and SBOPN algorithm we have developed objectively determine whether a protein is under or overexpressed relative to not only its direct binding partners, but to a larger network including partners of partners. This global evaluation is thus sensitive to the size of the network, but directly captures how the multiple binding interfaces of a protein can control its competition for binding partners. In the interface-resolved CME network, we have shown evidence of imperfect, but statistically significant stoichiometric balance. However, the original 56-protein network was overall unbalanced due to the high overexpression of the actin binding protein cofilin. The size of the network clearly matters, in the small modules, we are statistically out-of-balance, but on a larger scale, still in balance. Outliers are emphasized in smaller networks. At the same time, leaving out additional partners can provide some explanation for the observed imbalance. Imbalance may also indicate possible missing interactions in the network. Despite the simplicity of our metric, our method was still able to highlight both correlated concentrations and proteins that violate balance for functional reasons, such as the kinase PRK1. Furthermore, the observed balances can suggest possible mechanisms of assembly, for example, that can then be studied using kinetic modeling, as we did here. What our results emphasize is that correlations are highly important: functionality can be obliterated with significant imbalance, and misinteractions can also be overwhelming with significant imbalance.
Although we only applied our stoichiometric balance analysis to the 56 protein CME network, two smaller modules of this network, and the 127-protein ErbB network, these networks are significantly larger than the obligate complexes previous studied for copy number balance[5, 6]. Our networks also contain a much larger variety of binding interaction strengths and competitive and non-competitive interactions. As we showed above, balance depended on the protein network’s underlying IIN. While it would be beneficial to repeat this analysis on a larger network, there is a paucity of manually curated IINs in the literature. There are various larger automatically constructed IINs, constructed with homology modeling[77, 78], but our previous work found these automatic IINs suffer from various inaccuracies and differ significantly from manually curated IINs in topology[41].
The SBOPN method only accounts for the binding interface network structure and observed copy numbers. A missing feature of our stoichiometric balance metric is that proteins within a network can be expressed with both spatial and temporal variation. For a small binding network this is not a major concern, since proteins in the same complex tend to be co-expressed[79] and co-localized so they may bind. But as network size is scaled up, the probability of all proteins being equally present reduces. Such temporal and spatial variations could be taken into account in the construction of the network, leaving out proteins that are not functional at the same time.
A natural extension to our measure of stoichiometric balance would be to also account for binding affinities of interactions in addition to the binding interface network structure and observed copy numbers. Our results here and previous studies[19] indicate that balance should be more tightly constrained for strong binding proteins. However, one benefit to leaving affinities out of the measurement is that biochemical data is in even more limited availability than binding interface data. Our existing metric can thus be much more easily applied to a variety of networks. Furthermore, by picking out highly correlated expression levels, our method can then indicate which interactions might be quite strong, or vice-versa, which may be transient or weak.
In this study we used yeast copy numbers from Kulak et al. because it was the most comprehensive. The other three studies we used for comparison did not cover all 56 proteins in our network. However, for the proteins we could compare, we found significant discrepancies between relative abundances. Light chains are weakly expressed in other studies, for example[1, 49, 50]. A few possible reasons for this exist. The first is that fluorescence data is inherently noisy. Experimentalists must deal with background noise, interference with protein localization due to the large fluorescent tags, and cross interactions with other proteins[80]. The second is that cell lines can accrue mutations over time that decrease or increase gene expression, a phenomenon observed with HeLa cells[81]. Finally, cells may alter gene expression for regulatory reasons, so the environment in which cells are grown may alter gene expression.
We do not expect the cell to perfectly optimize the yield of all of its many assemblies. Each network we have evaluated here is ultimately part of a larger, global cellular network. Perfectly optimizing isolated, local modules does not appear to be a significant pressure for the cell, particularly when a sufficient balance, such as we observe for the vesicle-forming module, maintains functionality. Additionally, these processes, such as in the vesicle forming model discussed below, typically do not occur at equilibrium. Therefore, the concept of minimizing ‘leftover’ proteins based on expected equilibrium complexes formed is a simplification. Correlations in copy numbers are nonetheless often significant relative to randomly assigned copy numbers.
We found that copy number imbalance can lead to misinteractions and the features of biological IINs (power-law-like degree distribution, square and hub motifs, sparseness) typically have less misinteractions under balance copy numbers but more misinteractions under imbalance. These networks thus should require more tightly controlled balance to avoid misinteractions. But misinteractions are of course not the only pressure on copy numbers. For multi-protein assembly in an obligate complex (ARP2/3) and in a minimal model of vesicle formation for CME, we found that the functional cost of imbalance was dominated more by its impact on determining specific functional complexes than avoiding misinteractions. Nonetheless, the fact that misinteractions can decrease vesicle formation, by sequestering away adaptor proteins into large aggregates, shows that misinteractions are worse than simply having an excess of free proteins. If this result can be generalized, it may have important implications for mechanistic modeling of biological systems, as misinteractions or system error is rarely taken into account.
Although the functional effects of copy number balance are usually discussed in the context of number of complete complexes at equilibrium, we have shown that non-equilibrium dynamics can be affected as well. While the clathrin heavy chains and light chains were balanced with each other, they were overexpressed compared to their adaptor proteins, and this limited the frequency of vesicle formation. Although we found that perfectly balanced copy numbers therefore improved vesicle formation frequency compared to observed copy numbers, we speculate that specific imbalances could still be selected for evolutionarily. There are various possible reasons for this imbalance: the function of endocytosis is cargo uptake, and there is a cargo loading process before endocytosis occurs.[75, 76] Hence to maximize function, controlled endocytosis around high-cargo areas of the membrane may be preferably to frequent, spontaneous endocytosis, and the adaptor proteins can serve as an intentional bottleneck in the process. Clathrin, which cannot directly bind to the membrane, may be kept at a high expression in the cytosol so that there are enough triskelia to quickly form a vesicle no matter where the endocytic site occurs. However, the observed underexpression could also be because there are other adaptor proteins not included in our model, or because clathrin interactions have weaker affinities than interactions between adaptor proteins and must saturate them.
Finally, the predictions of our minimal vesicle-forming model are ultimately limited by the approximations we made to simulate the clathrin coat assembly and vesicle formation. Our model vesicles formed about 10 times faster than is observed in vivo. To fully capture the dynamics of this complex process, an ideal model would include all the proteins in our CME network (Fig 1), and include both the known biochemistry of binding interactions and the physics and biomechanics of membrane bending and scission. In yeast, the cytoskeleton is needed to help induce membrane budding, after which energy-consuming proteins such as dynamin scission off the vesicle from the plasma membrane for transport into the cell [76, 82]. However, such a modeling approach does not exist, due to the computational limitations of simulating such large complexes and membrane remodeling, and the lack of biochemical data.
Based on the model we did construct, however, there are some more specific limitations. The first is that while rule-based modeling is a convenient way to model complex formation, some theoretical aggregates may be impossible due to steric hindrance. Our model predicted that a vesicle of 100 triskelia could contain ~1900 additional proteins. Assuming each vesicle is a sphere with 100nm diameter, the allowable surface area per adaptor/scaffold protein would only be ~17nm2, which is too small to accommodate the excluded volume of the large, disordered regions of proteins such as ENT1 and 2[83]. Second, we did not include cooperatively in our model. Molecules localized in the same aggregate do not interact at a faster rate in conventional rule-based modeling. Clathrin triskelia weakly polymerize, as noted above, but the aggregation effect of the adaptor proteins–especially the SYP1/EDE1 complex–localizes triskelia close together, allowing them to bind strongly. In future work we will consider effects of cooperativity on assembly, as well as construct more detailed spatial and structural models of the vesicle forming process.
A stoichiometrically balanced network has the copy numbers of each interface matched to the copy numbers of all pairwise complexes it participates in (Fig 2). Balanced copy numbers are obtained by assigning a number of desired complexes to each edge in the interface binding network. The balanced copy numbers of each interface can then be calculated from the equation:
Ax=C
(2)
Where “A” is a binary matrix with Nint rows (one for each interface) and Medge columns (one for each pairwise interaction). Ai,j = 1 if the interface i is used in the interaction j, or 2 if a self-interaction, and 0 otherwise. “x” is the vector of desired pairwise complexes (Medge x 1), and “C” is the number of interface copy numbers (Nint x 1). In Fig 10 we illustrate this procedure for a small toy network.
If desired pairwise complexes, x, is specified, interface copy numbers, C, can directly be solved for using Eq 1, but if interface copy numbers, C, are specified, x will not, in general, have an exact or nontrivial solution unless C is balanced. This is because all entries of x must be >0 or some other minimum value, as negative copies cannot exist. This produces a hard constraint on x. Given a vector C, an optimal solution to x must be solved for using quadratic programming rather than linear least-squares.
Our goal is to select for an optimal x given an input set of copy numbers “C0”. This is a soft constraint on the optimal x, because the input C0 may not be balanced. Once an optimal x is found, forward solving Eq 1 will in general not perfectly recover C0. C0 can constrain all interfaces or a subset of them. To constrain a protein is to constrain all interfaces on it. We introduce a third constraint on the optimal x: the copy numbers of interfaces on the same protein should be equal. This often makes nontrivial solutions impossible (Fig 10), so it is also a soft constraint. Combining all of these constraints, the optimal desired number of complexes “x” can be found by minimizing the equation:
minx[α(Ax‑C0)TZ(Ax‑C0)+(Ax)TH(Ax)],x≥0
(3)
Where each variable is defined as follows:
A: Nint x Medge matrix defining which interfaces are used in which interaction, i.e. pairwise complex.
x: Medge x 1 vector of desired pairwise complex copy numbers
C0: Nint x 1 vector of constrained copy numbers.
Z: Nint x Nint diagonal matrix that selects which interfaces are constrained. Entries = 1 if the interface is constrained and = 0 otherwise. If all interfaces are constrained, Z equals the identity matrix.
H: Nint x Nint permutated block diagonal matrix with positive and negative entries such that H*C = 0 if interfaces on the same protein have equal copy numbers. Each block corresponds to a protein (Fig 10).
α: 1x1 scaling parameter which determines the relative weight of the C0 soft constraint vs the equal interfaces soft constraint.
For any vector x, Eq 2 produces a positive scalar value. The equation was minimized using the OOQP (object-oriented quadratic programming) 0.99.26 package for C++[84]. Quadratic programming is necessary due to the constraint of x≥0. Eq 2 can be converted into a quadratic equation of the form
12xTQx+dTx+r
(4)
Using
Q = 2αATZA + 2ATHA
dT = -2αC0TZTA
r = αC0TZC0
“r” can be ignored by the solver when minimizing the equation since it is a constant term.
Once xmin is found via Eq 3, the optimized interface copy numbers can obtained by forward solving A*xmin = Cbalanced,int. Interfaces on the same protein will not necessarily have equal copy numbers due to the competing constraints of Eq 2 (Fig 2C). We can assign a single copy number to each protein by averaging over all interface copy numbers on that protein to give Cbalanced, a vector of protein copy numbers. These values were used when calculating which proteins were over or underexpressed in the networks. Distance from C0 to Cbalanced was used as a metric to determine relative balance (see below).
For the yeast CME network, C0 was used to constrain all 56 proteins (Z = Identity matrix) because copy numbers from Kulak et al. were available[2]. For the ErbB signaling network, only 115 out of 127 proteins with available expression level data were constrained. 100 of these proteins were constrained with HeLa copy number estimations from Kulak et al. [2], while estimated copy numbers for 15 additional proteins were added from four additional studies[19, 51–53], leaving 12 proteins with unknown expression data. See S2 Table for all values.
Using the optimized copy numbers, Cbalanced, we can then ask, how close are the original, biologically observed copy numbers to these optimally balanced values? If the original copy numbers are already perfectly balanced, then they will match the optimal copy numbers. If they are imperfect, then the two distributions will differ. We use two metrics to quantify the distance between the observed and optimized concentrations: chi-square distance (CSD)
∑i(Xi−Yi)2(Xi+Yi)
(5)
and Jensen-Shannon Distance (JSD) after converting both vectors (X and Y) to distributions (x and y)
12(DKL(x‖z)+DKL(y‖z))
(6)
Where z = (x+y)/2 and DKL is the Kullback-Leibler divergence
DKL(x‖y)=∑ixilogxiyi
(7)
For cases where Z≠I (i.e. not all interfaces were constrained) only distance between constrained interfaces was measured.
Binding for the five 3- or 4-node network motifs; triangle, chain, square, 4-node hub, and flag; was simulated using the Gillespie algorithm[47]. Besides the specific binary interactions, nonspecific interactions were allowed at a strength determined by an “energy gap” between binding energies, though in practice we defined the ratio nonspecific KD to specific KD by factors of 10. This corresponded to a linear difference in free energies via the equations:
KD,specific=c0e−ΔE1KBT
KD,nonspecific=c0e−ΔE2KBT
KD,specificKD,nonspecific=e−(ΔE1−ΔE2)KBT
The networks were simulated under various initial concentrations. The steady-state ratio of Eq 1 was recorded, where Nnonspecific is the number of nonspecific binary complexes, Nspecific is the number of specific binary complexes, and Nfree is the number of free proteins. Ratios were averaged across 5,000 runs.
To generate surface plots, two proteins were chosen to be variable while the remaining proteins were given fixed copy numbers. Because the flag motif produced asymmetric plots, two different choices of variable proteins were used. (S3 Fig) Surface plots were generated using Matlab.
We calculated sensitivity by determining the principal component of the surface plot data (i.e. the vector of greatest variance) and measuring the percent change in ratio from the optimum along this vector. For better comparison, we normalized distance along the surface plots via dividing the abundance of the variable proteins by the abundance of the fixed proteins.
Motifs with purely noncompetitive interactions were not considered, because the interface network would then consist entirely of pairs, such as the IIN for Fig 1B. The balance is simple for pairs: all interfaces have the same copy numbers. We limited our analysis of Results part 2, “Imbalance increases misinteractions dependent on the network topology and binding affinities of proteins”, to small competitive motifs where we could enumerate all possible complexes and study effects of concentration variation systematically.
For the large network analysis we used the 500 networks from Johnson et al, J Phys Chem B 2013[27]. 25 sets of 10 networks each were randomly generated using two parameters: number of nodes (90, 110, 125, 150, 200), keeping the number of edges fixed at 150; and the preferential attachment exponent “γ” from Goh, 2001[85]. γ = 0 corresponds to a binomial, Erdos-Renyi network, whereas γ = 1 corresponds to a power-law or “scale-free” network. Values of 0, 0.2, 0.4, 0.6, and 0.8 were used. Finally, a local topology optimization algorithm that decreased the frequency of chain and triangle motifs and increased hub motifs was applied to each network, for 500 networks in total. All networks assume competitive (binary) binding.
Rather than assign an arbitrary specific and nonspecific KD for the networks, we used the relative binding energies determined for each network in the source paper. This was determined by a physics-based Monte Carlo optimization scheme of amino acid residues, as described in Johnson, 2011[23]. The minimum energy gap between specific and nonspecific interactions could be measured as a relative metric of the network’s propensity for misinteractions. Because the binding strengths were relative, we could alter the average binding strength to determine the effects on misinteractions. This was varied between 7 values of 1 nM to 1 mM, using factors of 10. Finally, to obtain results more comparable to the simple networks, we also ran simulations where each specific interaction had KD = 100 nM and each nonspecific interaction had KD = 100 μM.
Networks were simulated to steady state using the Gillespie algorithm[47] under five differing sets of copy numbers (CNs) for free proteins: equal CNs for each protein, random CNs sampled from a yeast protein concentration distribution (performed 20 times) and three forms of balanced CNs using the network architecture. Any set of CNs without leftovers–i.e. having exactly enough proteins to create a certain number of specific complexes–is considered “balanced”, and thus there are infinite solutions. The first balanced set assumed an equal number of each type of specific complex, which results in protein CNs proportional to the protein’s number of partners. The remaining balanced CNs were determined by finding “x” to minimize a simplified form of Eq 2:
minx(A*x‑C0)T(A*x‑C0)
(8)
Here there is only one interface on each protein, and all the proteins are constrained, so there is no need for a Z matrix, the α scaling parameter, or the second term. C0 is either equal copy numbers or randomly sampled copy numbers. After xmin is found via quadratic programming (see above), the balanced CNs are obtained by forward solving Cbalanced = A*xmin.
To measure nonspecific complex formation, a modified ratio was used:
Cost(C0)=2Nnonspecific(C0)2Nspecific(C0)+Nfree(C0)
(9)
to compare total individual proteins in each bound or unbound state, rather than number of unbound or bound states. To measure sensitivity, the ratio under unbalanced CNs (C0) divided by the ratio under balanced CNs (Cbalanced) was calculated. A higher ratio indicates higher sensitivity to CN balancing.
The kinetic model was simulated using the stochastic simulation method (the Gillespie algorithm). Binding interactions were encoded via the rule-based language BioNetGen and simulated via the Network Free Simulation (NFSim) software [48]. Trimer cooperativity was modeled by increasing the rate of the third reaction if three members of a correct trimer were held together by two reactions. For example, if A is bound to B is bound to C, and a binding between A and C is possible, that reaction rate was set to be arbitrarily high. Reaction rates were arbitrary, but interactions with the core subunit ARC19 were set to be ~10 fold stronger than interactions between periphery subunits, as this increased yield. Yield was measured via the equation
Yield=NdesiredNdesired+Nundesired
(10)
Where Ndesired is the number of proteins in complete complexes (equal to seven times the number of complex complexes) and Nundesired is the number of proteins in incomplete or misbound complexes. Completely free proteins were ignored.
A subnetwork of nine proteins–clathrin heavy chain (CHC1), clathrin light chain (CLC1), SLA2, ENT1/2, EDE1, SYP1, and YAP1801/2 –was defined based on known binding interactions (Table 1). Because the existence of multiple interfaces, allowing noncompetitive binding, results in a large number of possible species we simulated our model using the Network Free Simulator (NFSim)[48]. Binding dissociation constants were obtained from the literature, including for protein-lipid binding. For simplicity, the heavy chains were already assumed to be in trimer form, and ENT1/2 was combined into a single protein as the binding partners were the same. Binding constants were pulled from the literature. (Table 1)
The cell membrane and the cell cytoplasm function as different compartments with different volumes, but NFSim is not integrated with BioNetGen’s compartment language. We bypassed this problem by doubling the number of rules: besides the main rule for each reaction, an additional rule stated that if both proteins are on the cell membrane then the kon rate should be increased according to the membrane volume. Cell membrane ‘volume’ was determined by multiplying the membrane surface area by a factor 2σ = 2 nm to capture the change in binding affinities between 3D and 2D (see S1 Text).
Since our primary goal was to measure clathrin recruitment to the membrane, any complex on the membrane with at least 100 triskelia (a complex of three CHC1 and three CLC1) was considered a “vesicle” and deleted at a high rate kdump. Proteins in the vesicle were then added back to the cytoplasmic pool at a rate krecyc, which was set to be equal to kdump to indicate fast recycling. However, we clarify that even fast recycling is not instantaneous, and that proteins are added back one at a time rather than all at once. Fast vesicle formation thus could still drain the pool of adaptor proteins.
Misinteraction strengths were determined by calculating the geometric mean of the dissociation constants of each interface, as this provided a KD based on the arithmetic mean of the binding energies.
KD,mean=KD,1KD,2…KD,nn=e−ΔE1KBT∙e−ΔE1KBT∙…e−ΔE1KBTn=e−(ΔE1+ΔE2+…ΔEn)KBTn=e−(ΔE1+ΔE2+…ΔEn)nKBT
The KD of a misinteraction between two interfaces was set to be:
fKD,mean,1KD,mean,2
(11)
where f = 10,000 (weak misinteractions, corresponding to an energy gap of ~9.21) or 1,000 (stronger misinteractions, energy gap of ~6.91)
Network maps were generated using Cytoscape[86] and RuleBender[87]. Plots were generated in MATLAB. C++ code for the network balancing algorithm SBOPN is available at https://github.com/mjohn218/StoichiometricBalance, and may be applied to any interface-resolved network. The CME and ErbB networks are provided as example inputs.
|
10.1371/journal.ppat.1004334 | P. aeruginosa SGNH Hydrolase-Like Proteins AlgJ and AlgX Have Similar Topology but Separate and Distinct Roles in Alginate Acetylation | The O-acetylation of polysaccharides is a common modification used by pathogenic organisms to protect against external forces. Pseudomonas aeruginosa secretes the anionic, O-acetylated exopolysaccharide alginate during chronic infection in the lungs of cystic fibrosis patients to form the major constituent of a protective biofilm matrix. Four proteins have been implicated in the O-acetylation of alginate, AlgIJF and AlgX. To probe the biological function of AlgJ, we determined its structure to 1.83 Å resolution. AlgJ is a SGNH hydrolase-like protein, which while structurally similar to the N-terminal domain of AlgX exhibits a distinctly different electrostatic surface potential. Consistent with other SGNH hydrolases, we identified a conserved catalytic triad composed of D190, H192 and S288 and demonstrated that AlgJ exhibits acetylesterase activity in vitro. Residues in the AlgJ signature motifs were found to form an extensive network of interactions that are critical for O-acetylation of alginate in vivo. Using two different electrospray ionization mass spectrometry (ESI-MS) assays we compared the abilities of AlgJ and AlgX to bind and acetylate alginate. Binding studies using defined length polymannuronic acid revealed that AlgJ exhibits either weak or no detectable polymer binding while AlgX binds polymannuronic acid specifically in a length-dependent manner. Additionally, AlgX was capable of utilizing the surrogate acetyl-donor 4-nitrophenyl acetate to catalyze the O-acetylation of polymannuronic acid. Our results, combined with previously published in vivo data, suggest that the annotated O-acetyltransferases AlgJ and AlgX have separate and distinct roles in O-acetylation. Our refined model for alginate acetylation places AlgX as the terminal acetlytransferase and provides a rationale for the variability in the number of proteins required for polysaccharide O-acetylation.
| Bacteria utilize many defense strategies to protect themselves against external forces. One mechanism used by the bacterium Pseudomonas aeruginosa is the production of the long sugar polymer alginate. The bacteria use this polymer to form a biofilm – a barrier to protect against antibiotics and the host immune response. During its biosynthesis alginate undergoes a chemical modification whereby acetate is added to the polymer. Acetylation of alginate is important as this modification makes the bacterial biofilm less susceptible to recognition and clearance by the host immune system. In this paper we present the atomic structure of AlgJ; one of four proteins required for O-acetylation of the polymer. AlgJ is structurally similar to AlgX, which we have shown previously is also required for alginate acetylation. To understand why both enzymes are required for O-acetylation we functionally characterized the proteins and found that although AlgJ exhibits acetylesterase activity – catalyzing the removal of acetyl groups from a surrogate substrate – it does not bind to short mannuornic acid polymers. In contrast, AlgX bound alginate in a length-dependent manner and was capable of transfering acetate from a surrogate substrate onto alginate. This has allowed us to not only understand how acetate is added to alginate, but increases our understanding of how acetate is added to other bacterial sugar polymers.
| Pseudomonas aeruginosa is an opportunistic, Gram-negative pathogen that can cause acute and chronic infections. The bacterium is the dominant bacterial species in the lungs of cystic fibrosis (CF) patients and if left untreated, is the leading cause of morbidity and mortality among these individuals [1]. P. aeruginosa is able to persist through the formation of a biofilm where communities of surface attached bacteria are encapsulated in a matrix composed primarily of secreted extracellular polysaccharides. Bacteria embedded within a biofilm are more resistant than their planktonic counterparts to environmental stresses such as antibiotics and disinfectants, and are able to evade the defense mechanism(s) of the host [2]–[7]. P. aeruginosa has the genetic capability to produce at least three different biofilm exopolysaccharides: Pel, Psl, and alginate [8], [9]. P. aeruginosa clinical isolates obtained from CF patients with chronic pulmonary infections secrete large amounts of alginate [10], [11]. This exopolysaccharide is synthesized in the cytoplasm and is translocated across the inner membrane as a linear homopolymer of d-mannuronic acid [12], [13]. The polymer is subsequently modified in the periplasm through O-acetylation and epimerization to form a β1–4 linked non-repeating chain of d-mannuronic acid and its C5 epimer l-guluronic acid [14], [15].
Modification of polysaccharides through the addition or removal of acetate is an important biological process for survival and virulence in many bacterial species. For example, biofilm formation by the human pathogens Escherichia coli, Staphylococcus aureus and Staphylococcus epidermidis, requires the partial de-N-acetylation of the exopolysaccharide poly-β-1,6-N-acetyl-d-glucosamine (PNAG) [16]–[19]. Similarly, deacetylation of Pel from P. aeruginosa is required for Pel-dependent biofilm formation [20] while deacetylation of the holdfast polysaccharide synthesized by Caulobacter crescentus is required for adhesion and cohesion [21]. In each case the deacetylation or removal of acetate from the acetylated polysaccharide requires a single enzyme, which has been shown to be a member of carbohydrate esterase family 4 (CE4). In comparison, O-acetylation of polysaccharides is a complex process requiring two enzyme functionalities. The first functionality is the transfer of an acetyl-donor into the periplasm, which is hypothesized to be catalyzed by a membrane bound O-acetyltransferase (MBOAT) [22]. The second functionality, catalyzed by a periplasmic O-acetyltransferase, transfers acetate onto the polysaccharide. There are currently three distinct systems that are differentiated by the number of proteins required for O-acetylation; a single protein system whereby both functionalities are encoded on one polypeptide, a two-protein system with one MBOAT and one periplasmic O-acetyltransferase and, in the case of alginate and cellulose, a four protein system comprised of an MBOAT protein, two periplasmic O-acetyltransferase and a protein of unknown function [22], [23].
The O-acetylation of the C6 hydroxyl of muramoyl residues in peptidoglycan is critical in many human bacterial pathogens including; Methicillin-resistant S. aureus (MSRA), Bacillus anthracis, Neisseria meningitides and Neisseria gonorrhoeae [24], [25] as it confers resistance to degradation by endogenous autolysins and the host immune system during infection [26]. A single integral membrane O-acetylpeptidoglycan transferase (Oat) is utilized in Gram-positive bacteria, while Gram-negative bacteria utilize a two-protein system composed of PatA, the MBOAT, and PatB, the periplasmic O-acetyltransferase [27]–[29]. The O-acetylation of alginate in P. aeruginosa, is an important modification as acetylated alginate is less susceptible to recognition and clearance by the host immune system than its non-O-acetylated counterpart [30]. Similar to the acetylation of cellulose by Pseudomonas fluorescens through the combined action of WssG, WssH, WssI and WssF, the O-acetylation of alginate requires four proteins; AlgF, AlgI, AlgJ and AlgX [12], [13], [31]–[33]. Acetylation of alginate can occur at the C2 and C3 hydroxyl groups of mannuronic acid residues. AlgI is predicted to be a member of the MBOAT family, while AlgJ and AlgX are required in the O-acetylation of alginate as the polymer passages through the periplasm [13], [31], [33]–[35]. The function of AlgF, which is not predicted to have a catalytic domain, is currently unknown. It is also unclear why alginate acetylation requires two active O-acetyltransferases.
To probe the role of AlgX in alginate acetylation, we recently determined its structure and found that it is a two-domain protein with an N-terminal SGNH hydrolase-like domain and a C-terminal carbohydrate-binding module (CBM) [34]. Our in vivo functional characterization of AlgX demonstrated that three catalytic residues, D174, H176 and S269, located in the active site are required for alginate O-acetylation. In the present study, we sought to delineate the role of AlgJ, and examine why both AlgJ and AlgX are required for alginate acetylation [13]. To this end, we have determined the structure of Pseudomonas putida AlgJ75–370 (PpAlgJ75–370) to 1.83 Å resolution and have functionally characterized the protein.
The TMHMM server v2.0 indicates that AlgJ from P. aeruginosa PAO1 possesses a transmembrane helix from residues 12–29 that tethers the periplasmic domain to the cytoplasmic membrane [36]; a prediction that is supported by in vivo localization studies [13]. To probe the function of AlgJ in alginate biosynthesis, we attempted to crystallize a soluble domain, residues 79–379, of P. aeruginosa AlgJ (PaAlgJ79–379). Although this protein was recalcitrant to crystallization, we were able to crystallize the orthologous domain from P. putida, PpAlgJ75–370, which shares 83% and 53% sequence similarity and identity to PaAlgJ79–379, respectively. Selenomethionine-incorporated (SeMet) protein was expressed, purified, crystallized, and the structure determined to 1.83 Å using the single-wavelength anomalous dispersion (SAD) technique (Table 1). PpAlgJ75–370 crystallized in space group C2 with two molecules in the asymmetric unit. After iterative rounds of model building and refinement, the structure yielded models with an Rwork and Rfree of 17.5% and 21.2%, respectively (Table 1). Analytical size exclusion chromatography suggests that AlgJ is a monomer in solution (data not shown). The dimer observed in the crystal structure is therefore not believed to be biologically relevant. The two molecules in the asymmetric unit superimpose well with a root mean square deviations (RMSD) of 0.19 Å over the 906 aligned backbone atoms. Due to the poor quality of the electron density we were unable to build residues N75 to G77 and A266 to K278 in both molecules, and residues R78 and L242 in molecule A, and P244 and L245 in molecule B. Given the absence of two residues in the loop between L242 and F246 in molecule B, all structural analyses were performed using molecule A. All structural features defined using molecule A were also present in molecule B with no significant deviations.
The SGNH hydrolase superfamily consists of enzymes with varying hydrolytic activities (e.g., proteases and lipases) [37]. SGNH hydrolases have an α/β/α fold with each of the four conserved active site residues responsible for catalysis residing in one of four conserved blocks. The structure of PpAlgJ75–370 reveals that the protein has an α/β/α fold with a core of four parallel β-strands (Figure 1: β3, β6–8) and an isolated β-bridge at β3, which are comparable to the five parallel β-strands found in canonical SGNH hydrolases [37], [38]. The core β-strands are surrounded by nine α-helices (Figure 1: α1, α3–10) that complete the α/β/α fold. A surface representation of PpAlgJ75–370 reveals a shallow groove that crosses the face of the protein (Figure 2A, left, centre). Examination of the electrostatic surface potential shows a distinct region of electronegativity on the face of the protein located within and below the shallow groove (Figure 2B, centre).
Previous in vivo complementation experiments have suggested that AlgJ functions as an O-acetyltransferase and that this activity is dependent on conserved residues D193 and H195 (P. aeruginosa numbering) [33]. In PpAlgJ the corresponding residues D190 and H192 are located in the shallow electronegative groove (Figure 2B, centre). Residues within the groove are well conserved in AlgJ homologs from six Pseudomonas sp. (P. putida, P. aeruginosa, P. syringae, P. protegens, P. entomophila, and P. alkylphenolia) and Azobacter vinelandii (Figure 2A, centre). A 90° rotation along the horizontal axis relative to the groove depicts a generally electroneutral surface (Figure 2B, right) that also contains highly conserved residues (Figure 2A, right).
A search for structurally similar proteins using DALI reveals that the core of AlgJ is most similar to the N-terminal domain of P. aeruginosa AlgX, residues 44–344 (PaAlgX44–344), with an RMSD of 2.06 Å over 165 aligned Cα-atoms (Figure 3A). The active site of PpAlgJ75–370 and the orientation of the putative catalytic triad D190, H192 and S288, is analogous to PaAlgX27–474 and other serine esterases and proteases (Figure 3B) [39], [40]. Similar to PaAlgX27–474, the structure of PpAlgJ75–370 reveals several differences relative to canonical SGNH hydrolases [32]. Firstly, block III that contains the conserved asparagine residue that forms part of the oxyanion hole is absent (Figure 3C). Examination of the PpAlgJ75–370 structure suggests that Y348 is in a position to act as a hydrogen bond donor in place of the conserved asparagine (Figure 3B). In PaAlgX27–474 a tyrosine residue, Y328, also occupies this position. PpAlgJ75–370 also deviates from the canonical GDSL(S) and DxxH motifs in blocks I and V, respectively (Figure 3C). AlgJ has a GTSYS motif in block I which contains the catalytic serine (S288), while a single spacer residue in block V separates the two remaining catalytic triad residues (D190 and H192) to form a DxH motif. AlgJ appears to be a circularly permuted member of SGNH hydrolases as the order of conserved residues in primary sequence (H-S-G-Y) is different than that of SGNH hydrolases (S-G-N-H). However, the overall three dimensional shape and fold are structurally similar despite the rearrangement of conserved residues. PpAlgJ75–370 also contains secondary structure features not present across SGNH superfamily members. Two long anti-parallel β-strands are present on one side of the protein (β4 and β5, Figure 1), along with eight 310 helices (t1–8) and a small cap domain above the proposed active site which consists of two short anti-parallel β-strands (β1–2), five 310 helices (t1–3, t5–6) and one α-helix (α2) (Figure 1). Despite these differences, PpAlgJ75–370 is structurally comparable to other SGNH hydrolases including E. coli thioesterase I (TAP), and Aspergillus aculeatus Rha [38] as the SGNH hydrolase domains align with RMSDs of 3.2 Å and 3.8 Å over 101 and 104 equivalent Cα residues, respectively.
Two conserved sequence motifs, termed the AlgJ signature motifs, have been defined in P. aeruginosa AlgJ and its homologs [33]. These motifs are characterized by conserved regions of ΦΦΦPxK (Φ represents any hydrophobic residue), and (R/K)TDTHW. Mutation of residues within these signature motifs leads to impairment or ablation of alginate O-acetylation [33]. Utilizing the structure of PpAlgJ75–370, we identified the location of the signature motif residues (Figure 4A). Two distinct types of intramolecular interaction networks are observed within the motifs, which localize to the cap domain. The cap domain sits atop the SGNH hydrolase-like core that typically contains the active site and catalytic residues of canonical SGNH hydrolases [37], [38]. The first network composed of residues; K134, T189, D190 and H192 form a hydrogen-bonding network with L187, D254 and S288 (Figure 4B) in PpAlgJ75–370. It was previously observed that alginate O-acetylation was abrogated in vivo for the PaAlgJ variants K137A, D193A and H195A [33]. Superposition of PpAlgJ75–370 with AlgX44–344 suggests that D190 and H192 (D193 and H195 in PaAlgJ79–379) form part of the catalytic triad (Figure 3). Thus, two of the three catalytic triad residues (aspartic acid and histidine) in AlgJ reside in the separate cap domain distinct from the α/β/α fold, yet occupy an equivalent spatial position in the active site to their SGNH counterparts. The location of D193 and H195 in PaAlgJ (D190 and H192 in PpAlgJ) and the ablation of acetylation in vivo for the D193A and H195A variants provide further evidence of their crucial role in the catalytic mechanism of the protein. The second intramolecular interaction network is composed of a series of hydrophobic interactions centered around the conserved W193, with V131, Y289, W295 and F297 (Figure 4C). W193 is completely buried and comprises part of the hydrophobic core of AlgJ.
To determine whether AlgJ is catalytically active we examined the ability of the enzyme to exhibit O-acetylesterase activity, a commonality among SGNH hydrolases and the first half of the acetyltransferase reaction. PaAlgJ79–379 and PpAlgJ75–370 were both assayed to demonstrate that the enzymes are functionally equivalent. Using the substrate 3-carboxyumbelliferyl acetate, the kinetic parameters for PaAlgJ79–379 and PpAlgJ75–370 were observed to be comparable (Table 2), with a 2-fold difference in Km. The kcat/Km obtained for AlgX, which has been previously demonstrated to catalyze this reaction, differed by only 3-fold compared to the AlgJ orthologs. To assess the function of putative catalytic residues; D190, H192 and S288 in PpAlgJ and D193, H195 and S297 in PaAlgJ were substituted with alanine. Interestingly, while catalytic alanine variants in AlgX result in the abrogation of 3-carboxyumbellifyl acetate hydrolysis [32], mutation of the catalytic triad in both AlgJ orthologs only reduced the catalytic activity by ∼80% (Figure 5). Circular dichroism spectroscopy of the AlgJ orthologs and their respective variants exhibited no significant difference in spectra (data not shown). This indicates that the protein variants are properly folded and that the differences in catalytic activity are not due to large structural perturbations.
Comparison of PpAlgJ75–370 and AlgX44–344 indicates a significant difference in electrostatic surface potential between the active site regions of the enzymes (Figure 2B and C). PpAlgJ75–370 contains a shallow electronegative groove architecture around the active site whereas AlgX44–344 contains a deep electropositive groove compatible with binding the anionic alginate orpolymannuronic acid polymer. In addition, AlgX has a C-terminal carbohydrate-binding module, which presumably aids in binding and guiding alginate either to or from the active site. To examine whether PaAlgJ79–379 and AlgX27–474 interact with mannuronic acid oligomers, a direct electrospray ionization mass spectrometry (ESI-MS) binding assay was carried out using nine mannuronic acid oligosaccharides ranging from 4–12 sugar units in length (ManA4 – ManA12). Representative ESI mass spectra acquired for aqueous ammonium acetate solutions of AlgX and the protein reference scFv (Pref) with ManA6 or ManA12 are shown in Figures 6A and 6B, respectively. In the mass spectrum shown in Figure 6A, ion signals corresponding to protonated AlgX monomer and protonated 1∶1 (AlgX+ManA6) complex, the +13 to +16 charge states, are observed. Signals corresponding to protonated Pref and (Pref+ManA6 or ManA12) are also present, indicating that nonspecific carbohydrate-protein binding took place during the ESI process. The mass spectrum shown in Figure 6B is qualitatively similar to that shown in Figure 6A, although a visibly larger fraction of the protein is in the bound form. Similar results were obtained for the other alginate ligands tested (data not shown). ESI-MS measurements were also performed on solutions of AlgX and undeca- and pentadeca-hyaluronic acid (HA11 and HA15). Importantly, these negative controls revealed no evidence of specific binding between AlgX and these acidic oligosaccharides, thereby confirming the specificity of AlgX for the alginate oligomers rather than any acidic oligosaccharide. Listed in Table 3 are the association constants (Ka) determined by ESI-MS, following correction for nonspecific carbohydrate-protein binding. Notably, the Ka values for AlgX are seen to clearly increase with the length of the mannuronic acid oligosaccharides. It must be noted that the location of ligand binding site cannot be explicitly identified using this methodology and the observed interaction for AlgX may include both ligand-active site and ligand-CBM interactions.
The same assay was utilized to quantify binding of PaAlgJ79–379 to the mannuronic acid oligomers. Representative ESI mass spectra are shown in Figures 6C and D for ManA6 and ManA12, respectively. Analysis of the ESI mass spectra reveals two distinct charge state distributions (+10 to +13 and +14 to +26) for the protonated ions of AlgJ. Ions corresponding to 1∶1 complexes of AlgJ with alginate ligand were only observed at the lower charge states. Taken together, these results suggest that a fraction of AlgJ is unfolded in solution (corresponds to the +14 to +26 charge state distribution) and does not bind to the oligosaccharides [41]–[45]. Only the lower charge state AlgJ ions were considered for the Ka determinations (Table 3). Because the relative protein ion abundances measured by ESI-MS do not necessarily reflect relative concentrations in solution, the concentration of folded AlgJ used for the Ka calculations was not accurately known. Nevertheless, the ESI-MS results suggest that under the conditions tested binding of AlgJ to the mannuronic acid oligomers is extremely weak, with Ka values of less than 500 M−1.
Since AlgX interacts with mannuronic acid oligomers under the conditions tested, we examined the ability of AlgX to transfer an acetyl group from the pseudo-acetyl donors, 4-nitrophenyl acetate and 3-carboxyumbelliferyl acetate, to a decamer of polymannuronic acid (ManA10). Reactions analyzed by ESI-MS revealed the production of both mono- and di-acetylated mannuronic acid oligomers in the AlgX containing reactions as observed by an m/z shift of 42.01 for the sodiated singly acetylated and 42.01 for the protonated doubly acetylated. The mass increase of the protonated singly acetylated species was 42.00 (Figure 7). Control reactions containing all of the reaction components except AlgX, did not result in acetylated alginate even though 4-nitrophenyl acetate exhibited low spontaneous hydrolysis at pH 7.0 resulting in the production of 4-nitrophenyl and acetate. Neutral, commercially available sugars, cellohexose, xylohexose and maltotriose were not O-acetylated in the presence of AlgX. While both AlgJ and AlgX are necessary for alginate O-acetylation in vivo [13], [33] our binding and acetyltransferase data suggest AlgJ and AlgX do not have overlapping functions. Our current in vitro data provide additional evidence to support previous in vivo studies that there is no redundancy in the alginate acetylation machinery [31]–[33].
The O-acetylation of alginate requires the concerted action of four proteins: the putative MBOAT protein, AlgI; a protein of unknown function, AlgF; and two annotated O-acetyltransferases AlgJ and AlgX. To shed light on the role of each O-acetyltransferase, we determined the structure of PpAlgJ75–370 and compared it with the recently solved structure of PaAlgX27–474.Additionally, we kinetically characterized the acetylesterase activity of AlgJ and tested the ability of AlgJ and AlgX to bind and O-acetylate short polymannuronic acid oligomers.
The structure of PpAlgJ75–370 reveals a fold that is structurally comparable to AlgX44–344 and other SGNH hydrolases. PpAlgJ, like AlgX, is best described as an SGNH hydrolase-like protein as they both exhibit several key differences to canonical SGNH hydrolases. Not only is the order of the catalytic residues circularly permuted, but both proteins contain a cap domain and two long anti-parallel β-strands on one side of the protein that are not observed in other SGNH members. Although the function of the two long anti-parallel β-strands is currently unknown, given the involvement of other proteins in the O-acetylation system it is tempting to speculate that these regions may be involved in protein-protein interactions. A prediction for potential interaction surfaces on AlgJ was made using the consensus Protein-Protein Interaction Site Predictor (cons-PPISP) [46], [47]. Two predicted clusters of residues had positive scores for a possible interface. One cluster is comprised of residues 353–363 and is located on a loop and the beginning of α10 on the C-terminal end of the protein. A second cluster comprised of residues P79, G80, V81, D235 and F239 in PpAlgJ is localized to the cap domain, located above the SGNH core. This cluster contains the conserved AlgJ signature motif residues that are imperative for alginate O-acetylation. The structure of PpAlgJ has allowed us to confidently define the location of these residues, and propose roles for their function in O-acetylation. Signature residue variants in PaAlgJ; P135A, K137A, D193A, H195A and W196F (PpAlgJ equivalents; P132, K134, D190, H192 and W193) ablate O-acetylation [33]. The structure of PpAlgJ, in addition to our kinetic data, indicates that D190 and H192 are part of the catalytic triad. Therefore, alanine variants D190A and H192A would disrupt the proposed catalytic serine by increasing the pKa of the nucleophile and altering its orientation in the active site. The relatively rare internal lysine, K134 uses hydrogen bonding to stabilize the backbone oxygens of L187, R188 and D190 that form a loop between helices α3 and α4. It is expected that the loss of hydrogen bonding in the K134A variant disrupts the proper positioning of D190 in the DxH motif required for catalysis. The P132A variant located proximal to K134 would alter the secondary structure of, and impede proper function of K134. Lastly, the W193A variant is anticipated to disrupt the hydrophobic interactions with its neighbouring residues. Given that W193 is located proximal to catalytic H192, structural perturbations caused by disruption to the hydrophobic interactions are expected to have a negative impact on the catalytic triad.
Our in vitro enzymatic analysis probed the ability of PaAlgJ and PpAlgJ to perform the initial acetylesterase step of the overall acetyltransferase reaction. The results indicate that both AlgJ enzymes exhibit comparable catalytic parameters to AlgX for the hydrolysis of acetate (acetylesterase) from 3-carboxyumbelliferyl acetate [32]. Catalytic variants reduced acetylesterase activity by >80%. Although, the complete loss of activity was observed when similar catalytic residues were replaced in AlgX, in vitro residual activity in catalytic variants has been reported in at least one SGNH esterase [40]. The precise reason for residual activity is not clear since there are limited in vitro studies characterizing catalytic triad variants in SGNH hydrolase superfamily members. However, one possibility is the surrogate acetyl-donor tested does not optimally mimic the native acetyl donor of AlgJ. Residual hydrolysis of esterase substrates may also be attributed to surface amino acid residues that form microenvironments that promote spontaneous, non-enzymatic hydrolysis. For example, non-enzymatic hydrolysis has been reported for the SGNH superfamily member glutathione-S-transferase [48] and in the non-enzymatic protein albumin [49], [50]. The observation that the catalytic triad variants ablate O-acetylation in vivo further supports the notion that non-specific hydrolysis occurs in vitro.
The electrostatic surfaces of PpAlgJ75–370 and the SGNH hydrolase-like domain of PaAlgX27–474 are distinctly different with respect to the active site region. PaAlgX contains an electropositive groove that stretches from the active site to the C-terminal CBM [32]. We previously proposed that this highly conserved electropositive region would be compatible for binding the anionic alginate polymer. A direct ESI-MS binding assay confirmed that AlgX is able to bind mannuronic acid oligomers, with longer, more physiologically relevant polymers of mannuronic acid exhibiting higher affinity for the protein. The addition of a surrogate acetyl-donor in the presence of mannuronic acid oligomers led to the first demonstration in vitro that AlgX is an O-acetyltransferase capable of transferring acetyl groups to mannuronic acid oligomers. While ESI-MS cannot explicitly identify the location of carbohydrate-protein interaction, the ability for AlgX to O-acetylate mannuronic acid oligomers demonstrates that these oligosaccharides must, in part, bind to the active site. In comparison, PpAlgJ demonstrated very weak or no affinity toward mannuronic acid oligosaccharides under the conditions tested. Taken together, these data suggest that AlgX may be the only enzyme that O-acetylates alginate and that it functions in a non-redundant, successive mechanism with the other proteins in the acetylation complex machinery. Our data allow us to propose an updated model for alginate O-acetylation (Figure 8). Briefly, AlgI interacts with an unknown acetyl donor molecule in the cytoplasm and transfers acetate or acetyl donor across the inner membrane to either AlgJ and/or AlgF in the periplasm. The transmembrane domain of AlgJ (residues 12–29) tethers the enzyme to the membrane and potentially closer to AlgI than either AlgF or AlgX which lack this spatial constraint. Given the current state of knowledge, the function and mechanism of the intermediate step(s) involving AlgJ or AlgF can only be speculated. The order of transfer of the acetyl donor is uncertain, but the fact that AlgJ can perform the initial esterase stage of transferase reaction suggests that, given the right donor, recipient, and environment, this enzyme is directly involved in the O-acetylation process. This possibility cannot be ruled out for AlgF either, but the lack of identifiable catalytic residues in this protein suggests that it may serve a more accessory role. However, it is clear that AlgX can catalyze the direct O-acetylation of alginate. Thus, it is conceivable that AlgX receives the acetate or acetyl donor molecule from either AlgJ or AlgF, and then subsequently catalyzes the transfer of this substrate to O-acetylate alginate. This model is supported by the observation that active site variants of either AlgJ or AlgX are sufficient to abolish acetylation in vivo [32], [33]. In addition, the data presented herein exclude the possibility that both enzymes can O-acetylate alginate.
Although the O-acetylation of polysaccharides requires a minimum of two functionalities, O-acetylation of alginate and cellulose requires four distinct proteins including a membrane associated and non-membrane associated periplasmic O-acetyltransferase. In contrast, the O-acetylation of peptidoglycan utilizes a single O-acetyltransferase that may not be constrained to the inner membrane depending on the presence of an N-terminal transmembrane domain. Regardless of the number of proteins involved in the system, the catalytic mechanism of acetate transfers across the inner membrane and the acetylation of the polymer is undoubtedly highly conserved. Additionally, the role of O-acetylation of these polysaccharides in both Gram-negative and Gram-positive bacteria is functionally similar, allowing for protection against external agents [16]–[21], [26], [27]. Intuitively, the requirement of four proteins involved in relaying acetate from the cytosol to the polysaccharide appears inefficient when compared to the O-acetylation of peptidoglycan. While additional proteins could facilitate increased regulation, the extent of O-acetylation of peptidoglycan and alginate between species, strain, and culture conditions, varies between 20–70% (relative to muramic acid content) and 4–57%, respectively [51], [52]. This indicates that the amount of O-acetylation cannot be simply categorized based on the number of proteins involved or their localization.
A defining characteristic of cellulose and alginate biosynthesis compared to peptidoglycan is that these polysaccharides must traverse two membranes for export before reaching their final location, which in turn requires the involvement of several additional proteins in polymer modification and export. Therefore, four O-acetylation proteins may be an inherent requirement to adapt the O-acetylation machinery to the biosynthetic export machinery. In support of this, AlgX has been demonstrated to interact with both outer membrane alginate export proteins and periplasmic proteins required in alginate modification [53]–[57]. Such interactions have not been observed in the O-acetylation of peptidoglycan. The proteins WssFGHI are compulsory for cellulose acetylation [58] and WssGHI are homologous to AlgFIJ, with amino acid sequence identities of 24, 46, and 33%, respectively. We have previously suggested that WssF, which is predicted to belong to the SGNH hydrolase superfamily may be analogous to AlgX although it lacks the CBM present in AlgX [32]. Since proteins involved in cellulose acetylation have not been studied at either the structural or functional level, our present studies on AlgJ and AlgX are significant as they provide further data to support the roles of these proteins beyond sequence homology and phenotypic analysis.
We have successfully determined the structure of AlgJ, a protein involved in the alginate O-acetylation pathway. AlgJ exhibits acetylesterase activity that is mediated through the Ser-His-Asp catalytic triad similar to that of the SGNH hydrolase-like enzyme AlgX. ESI-MS confirmed that AlgX but not AlgJ binds polymannuronic acid and we have conclusively demonstrated that AlgX is an O-acetyltransferase, and that it is the only annotated O-acetyltransferase in the pathway that can both interact with, and O-acetylate alginate. Refining the model for alginate O-acetylation provides new avenues for further studies into the mechanism of O-acetylation of polysaccharides in the microbial kingdom.
Superflow Ni2+ NTA-agarose resin was obtained from Qiagen (Mississauga, ON). Graphitized carbon solid phase extraction columns (Carbograph SPE) are products of Grace Canada, Inc (Ajax, ON). All other chemicals and reagents, unless otherwise stated, were supplied by Sigma-Aldrich Canada Ltd. (Oakville, ON). All growth media was obtained from Bio Basic (Markham, ON). DNA manipulations were performed in E. coli DH5α and protein expression of the SeMet protein was carried out using E. coli B834 (DE3) Met-auxotroph cells and grown in media supplemented with kanamycin at 50 µg mL−1. Protein expression for binding and enzyme assays was carried out in E. coli BL21-CodonPlus cells.
The nucleotide sequences of algJ from P. putida KT2440 and P. aeruginosa PAO1 were acquired from the Pseudomonas Genome Database [59]. The boundaries of the putative O-acetyltransferase domain of P. aeruginosa AlgJ were predicted to range from amino acids 79 to 379 (PaAlgJ79–379) based on Phyre2 [60] structural alignments using the N-terminal domain of P. aeruginosa AlgX (PaAlgX44–344) [32]. Sequence alignment revealed that these boundaries correspond to amino acids 75 to 370 in the P. putida AlgJ homolog (PpAlgJ75–370). PCR amplification was carried out using the high fidelity DNA polymerase, PfuTurbo (Stratagene). FastDigest restriction enzymes were obtained from Fermentas. Plasmid DNA was extracted from E. coli using the PureLink Quick plasmid miniprep kit from Invitrogen (Burlington, ON). Primer sequences used in the generation of wild type (WT) PaAlgJ79–379 and PaAlgX44–344, as well as proposed active site mutants are summarized in Supplementary Table S1. The PCR reaction conditions were as follows: Pfu buffer with 2 mM MgSO4 (Thermo Scientific), 10 ng template DNA, 10 ng forward and reverse primer each, 25 µM dNTPs, 2 mM MgSO4, 2.5 U Pfu polymerase (Thermo Scientific) in a total reaction volume of 25 µL. The PCR product was digested with NdeI and XhoI and ligated into a pET28a vector backbone to generate (i) a thrombin cleavable N-terminal hexahistidine tag (His6) construct; and (ii) a second construct used for crystallography that contained both an N and C-terminal His6-tag. Site directed mutagenesis was performed using the QuikChange Lightning kit according to the prescribed protocol (Agilent Technologies). Constructs generated were verified by sequencing performed by ACGT DNA Technologies Corporation (Toronto, ON).
The expression and purification of native wild type and mutant PaAlgJ79–379 and PpAlgJ75–370 constructs were identical. SeMet PpAlgJ75–370 was used for structure determination and was expressed using the protocol described by Lee et al [61]. The expression and purification protocol was as follows: Starter cultures were grown overnight in 50 mL Luria-Bertani (LB) broth containing 50 µg mL−1 kanamycin at 310 K in a shaking incubator, with E. coli BL21-CodonPlus cells transformed with the appropriate plasmid. The cells were subsequently inoculated into 1 L LB broth containing 50 µg mL−1 kanamycin at 310 K in a shaking incubator. Upon reaching an OD600 of 0.7, the cells were induced with isopropyl β-D-1-thiogalactopyranoside (IPTG) to a final concentration of 1 mM. The induced cells were allowed to grow for an additional 18 h at 291 K. The cells were harvested via centrifugation at 5000× g for 20 min at 277 K. The cell pellet was stored at 253 K until needed. Frozen cell pellet was thawed over ice and re-suspended in 50 mL cold lysis buffer (500 mM NaCl, 20 mM Tris-HCl pH 8.0) containing one SIGMAFAST EDTA-free protease-inhibitor cocktail tablet (Sigma). The cells were homogenized at 10,000 psi through an Emulsiflex C3 (Avestin Inc.) with at least 3 passes until uniform in consistency. The resultant cell lysate was centrifuged at 25000× g for 25 minutes at 278 K to pellet cell debris and insoluble material. The soluble cell lysate was loaded onto a 5 mL Ni2+-NTA gravity column equilibrated with Ni-NTA buffer (500 mM NaCl, 20 mM Tris-HCl pH 8.0, 5 mM imidazole). The column was washed with 10 column volumes of Ni-NTA buffer containing 30 mM imidazole. Protein bound to the column was eluted with 4 column volumes of Ni-NTA buffer containing 150 mM imidazole. The eluent was dialyzed against 4 L of S200 buffer (150 mM NaCl, 20 mM Tris-HCl pH 8.0). Protein concentration was measured using the Pierce BCA Protein Assay Kit from Thermo Scientific (Rockford, IL). The His6-tag was cleaved from the protein via incubation with thrombin (Novagen) at 0.5 U per mg of protein at 298 K for at least 2 h. The thrombin treated protein was loaded onto a 1 mL Ni2+-NTA column equilibrated with S200 buffer containing 5 mM imidazole. The column was washed with 10 column volumes of S200 buffer containing 30 mM imidazole. The initial flow through and wash were pooled together and contained the un-tagged protein. The untagged protein was concentrated to a 1–2 mL volume using an Amicon Ultra centrifugation filter device (Milipore) with a 30 kDa cutoff. Approximately 20 mg of purified protein could be obtained per litre of bacterial culture. The concentrated protein was further purified via size exclusion chromatography on a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare). Fractions containing protein were pooled and analyzed via SDS-PAGE to be >95% pure. PaAlgJ79–379 and PpAlgJ75–370 protein could be stored at 277 K for up to 2 or 4 weeks, respectively, before significant degradation was observed by SDS-PAGE.
SeMet PpAlgJ75–370 was concentrated to ∼6–8 mg mL−1 by an Amicon Ultrafiltration device (30 kDa MWCO, Milipore) for crystallization trials. Sparse-matrix screens were setup by hand using MCSG suites 1–4 (Microlytic) in 48-well VDX plates (Hampton Research). The drops consisted of a 1∶1 ratio of protein to well solution at a final volume of 4 µL equilibrated over 250 µL of well solution, and stored at 293 K. Numerous hits were obtained after 1 week and were primarily found in conditions containing a divalent cation (Mg2+ or Ca2+) and polyethylene glycol (PEG) solutions between 3350–6000 Da. In most cases the crystals nucleated from a single point and radiated outwards forming a cluster with individual crystals estimated to be a maximum of 500 µm, and were of diffraction quality directly from the sparse matrix screens.
Crystals used for data collection were found in MCSG-1, condition 7 (0.2 M MgCl2, 0.1 M Bis-Tris∶HCl pH 5.5, 25% (w/v) PEG3350). Prior to data collection, crystals were cryo-protected by exchanging the drop solution with cryo-protectant solution (0.2 M MgCl2, 0.1 M Bis-Tris∶HCl pH 5.5, 25% (w/v) PEG3350, 20% (v/v) ethylene glycol). The exchange was performed through the addition of cryo-protectant solution directly to the drop and the removal of the added volume until complete exchange had occurred. The crystal clusters were disrupted via physical contact with a fine needle until an isolated single crystal could be looped. Crystals were vitrified in liquid nitrogen and stored. Selenium single-wavelength anomalous dispersion (Se-SAD) X-ray diffraction data were collected on beamline X29A at the National Synchrotron Light Source (NSLS) at Brookhaven National Laboratory. 90 images at 94% beam attenuation with 2° Δφ oscillation and 360 images without beam attenuation with 1° Δφ oscillation were collected on an ADSC Q315 CCD detector at a 260 mm crystal-to-detector distance and 0.4 s exposure time per image. The Se-SAD data was indexed, integrated, scaled and merged using HKL-2000 (Table 1) [62], and used in conjunction with HKL2MAP to locate 12 (of 14) selenium sites, with density modified phases calculated using SOLVE/RESOLVE [63]. The electron density maps were of sufficient quality for automatic model building using PHENIX AutoBuild [64] and subsequent manual model building using COOT [65], [66]. Model refinement was performed using PHENIX.REFINE [67] and progress was monitored as a function of the reduction and convergence of Rwork and Rfree (Table 1) [66]. TLS groups were added to the refinement in PHENIX through the use of the TLSMD server [68].
All figures that display the structure of PpAlgJ75–370 and/or PaAlgX27–474 were generated using PyMol (The PyMol Molecular Graphics System, version 1.6.0.0, Schrödinger, LLC). The secondary structure was determined using the STRIDE web server [69]. Surface representations demonstrating either electrostatics or surface residue conservation were depicted with all side chains present even if they could not be accurately modeled in the structure. Electrostatic surfaces were generated using the ABPS Tools 2.1 plugin that is integrated into PyMol [70]. Surface residue conservation was determined using the ConSurf web server [71] and visualized using the provided color scheme. Solvent accessible surface area was calculated using the PDBePISA server [72]. Surface residue conservation was determined using the ConSurf web server [71] and visualized using the provided color scheme. Solvent accessible surface area was calculated using the PDBePISA server [72].
Sequence alignments to determine AlgJ homologs were determined using BLAST [73]. Multiple sequence alignments were generated using Multiple Sequence Comparison by Log-Expectation (MUSCLE) [74], [75]. Prediction of potential protein-protein interaction surfaces was performed using cons-PPISP [46], [47].
All enzyme assays were performed at least in triplicate, in a 96-well microtiter plate, using a SpectraMax M2 from Molecular Devices (Sunnyvale, CA). Standard reactions contained 3.0 mM 3-carboxyumbelliferyl acetate (ACC), dissolved in DMSO for specific activity assays and variable concentrations ranging between 0.1 Km and ≥2 Km and 30 µg each wild-type protein and 100 µg for each variant in a total volume of 100 µL in 50 mM sodium HEPES buffer (pH 7.0 for AlgJ, pH 8.0 for AlgX) at 298 K. The final DMSO concentration did not exceed 10% (v/v). Due to low substrate solubility, reactions with higher substrate concentrations could not be obtained. Reactions were initiated by the addition of substrate and reactions were monitored in real time for a duration of 10 min using an excitation of 386 nm and an emission of 447 nm as previously described [32]. The hydrolysis and release of acetate results in an increase in the fluorescence signal. Background hydrolysis rates, in the absence of enzyme, were monitored and subtracted from enzyme-catalyzed reactions. A calibration curve for 7-hydroxycourmarin-3-carboxylic acid, the fluorescent hydrolysis product of 3-carboxyumbelliferyl acetate, was obtained under the reaction conditions and used to calculate reaction rate. The protein concentration of each enzyme variant was determined using the Pierce BCA Protein Assay Kit from Thermo Scientific (Rockford, IL). Data were fit by nonlinear regression to the Michealis-Menten equation using GraphPad Prism 6.0c for Mac, (GraphPad Software, La Jolla California USA, www.graphpad.com).
Immediately prior to ESI-MS analysis, PaAlgX27–474 and PaAlgJ79–379 were each dialyzed against aqueous 100 mM ammonium acetate (pH 7.0) using microconcentrators (Millipore Corp., Bedford, MA) with a MW cut-off of 30 kDa (for AlgX) and 10 kDa (for AlgJ). Two different reference proteins (Pref) were used to correct ESI mass spectra for the occurrence of nonspecific carbohydrate-protein binding (during the ESI process): a single chain fragment (scFv, MW 26 539 Da) of the monoclonal antibody (mAb) Se155-4, which was produced and purified as described previously [76], and lysozyme (Lyz, MW 14 300 Da), which was obtained from Sigma-Aldrich Canada (Oakville, Canada) and used without further purification. Each protein was concentrated and dialyzed against 50 mM ammonium acetate using microconcentrators (Millipore Corp., Bedford, MA) with a MW cut-off of 10 kDa and stored at −20°C until needed. Stock solutions of each of the polymannurnic acid oligomers, tetramer through dodecamer (ManA4 – ManA12), and undeca- and pentadeca-hyaluronic acid saccharides (HA11 and HA15), synthesized as previously described [77], [78] were prepared by dissolving known amounts of solid compound in ultrafiltered water (Milli-Q, Millipore, Bedford, MA) to give a final concentration of ∼1 mM. The ligand solutions were stored at 253 K until needed.
Affinity measurements were carried out on a Synapt G2 quadrupole-ion mobility separation-time of flight (Q-IMS-TOF) mass spectrometer (Waters, UK) equipped with a modified nanoflow ESI (nanoESI) source. Complete details of the instrumental and experimental conditions used for the ESI-MS binding measurements along with descriptions of how the data were analyzed to establish association constants (Ka) for protein-ligand interactions have been described previously [79]–[81].
O-Acetyltransferase assays were completed in a similar fashion as those previously described for the O-acetylation of peptidoglycan with modifications [51]. Briefly, 3 mM 4-nitrophenyl acetate dissolved in ethanol (3% final concentration), and 1 mM of ManA10 [78], was incubated with 60 µg of AlgX in 50 mM sodium phosphate buffer, pH 7.0 for a duration of 1 h at 298 K. Control reactions containing 1 mM of cellohexose, xylohexose and maltotriose in place of alginate were also completed. All reactions were repeated in the absence of AlgX as a negative control. The total reaction volume was 150 µL and the reaction was initiated by the addition of protein. The reactions were quenched by applying the entire reaction to a 4 mL graphitized carbon solid phase extraction column previously washed with three column volumes of 100% acetonitrile containing 0.1% (v/v) trifluoroacetic acid (TFA) and then equilibrated with three column volumes of water. Following application of the reaction, the column was washed successively with three column volumes of 100% acetonitrile, 100% acetonitrile containing 0.1% (v/v) TFA, 3∶1 isopropyl alcohol: acetonitrile (v/v). Polymannuronic acids were eluted with 6 mL of 50% tetrohydrofuran containing 0.1% TFA (v/v) and dried under vacuum in a Speed Vac at 298 K. The dried samples were resuspended in 100 µL of water and stored at 253 K prior to analysis.
Liquid chromatography–mass spectrometry analyses were performed on an Agilent 1200 HPLC liquid chromatograph interfaced with an Agilent UHD 6530 Q-Tof mass spectrometer at the Mass Spectrometry Facility of the Advanced Analysis Centre, University of Guelph. A C18 column (Agilent Poroshell 120, 150 mm×4.6 mm 2.7 µm) was used for chromatographic separation with solution A (0.1% formic acid) and solution B (100% acetonitrile with 0.1% formic acid). The mobile phase gradient was as follows: initial conditions, 2% solution B increasing to 98% solution B in 30 min followed by a column wash at 98% solution B and 10 minute re-equilibration. The first 2 and last 5 minutes of gradient were sent to waste. The flow rate was maintained at 0.1 mL/min. The mass spectrometer electrospray capillary voltage was maintained at 4.0 kV and the drying gas temperature at 523 K with a flow rate of 8 L/min. Nebulizer pressure was 30 psi and the fragmentor was set to 160. Nitrogen was used as both the nebulizing and drying gas, and the collision-induced gas. The mass-to-charge ratio was scanned across the m/z range of 100–3000 m/z in 4 GHz (extended dynamic range positive-ion auto MS/MS mode). Three precursor ions per cycle were selected for fragmentation. The instrument was externally calibrated with the ESI TuneMix (Agilent). The sample injection volume was 10 µl. The resultant spectra were analyzed using mMass software (http://www.mmass.org/).
The coordinates and structure factors for PpAlgJ75–370 have been deposited in the PDB, ID code 4O8V.
|
10.1371/journal.pbio.1000434 | The Major Yolk Protein Vitellogenin Interferes with the Anti-Plasmodium Response in the Malaria Mosquito Anopheles gambiae | When taking a blood meal on a person infected with malaria, female Anopheles gambiae mosquitoes, the major vector of human malaria, acquire nutrients that will activate egg development (oogenesis) in their ovaries. Simultaneously, they infect themselves with the malaria parasite. On traversing the mosquito midgut epithelium, invading Plasmodium ookinetes are met with a potent innate immune response predominantly controlled by mosquito blood cells. Whether the concomitant processes of mosquito reproduction and immunity affect each other remains controversial. Here, we show that proteins that deliver nutrients to maturing mosquito oocytes interfere with the antiparasitic response. Lipophorin (Lp) and vitellogenin (Vg), two nutrient transport proteins, reduce the parasite-killing efficiency of the antiparasitic factor TEP1. In the absence of either nutrient transport protein, TEP1 binding to the ookinete surface becomes more efficient. We also show that Lp is required for the normal expression of Vg, and for later Plasmodium development at the oocyst stage. Furthermore, our results uncover an inhibitory role of the Cactus/REL1/REL2 signaling cassette in the expression of Vg, but not of Lp. We reveal molecular links that connect reproduction and immunity at several levels and provide a molecular basis for a long-suspected trade-off between these two processes.
| Malaria annually claims the lives of almost 1 million infants and imposes a major socio-economic burden on Africa and other tropical regions. Meanwhile, the detailed biological interactions between the malaria parasite and its Anopheles mosquito vector remain largely enigmatic. What we do know is that the majority of malaria parasites are normally eliminated by the mosquito's immune response. Mosquitoes accidentally acquire an infection by sucking parasite-laden blood, but this belies the primary function of the blood in the provisioning of nutrients for egg development in the insect's ovaries. We have found that the molecular processes involved in delivering blood-acquired nutrients to maturing eggs diminish the efficiency of parasite killing by the mosquito immune system. Conversely, molecular pathways that set the immune system on its maximal capacity for parasite killing preclude the efficient development of the mosquito's eggs. Our results reveal some of the molecules that underpin this example of the trade-offs between reproduction and immunity, a concept that has long intrigued biologists.
| Malaria is a mosquito-borne parasitic disease affecting annually an estimated 250 million people, of which close to 1 million (mostly children in sub-Saharan Africa) succumb to the disease (World Health Organization fact sheet #94, April 2010; http://www.who.int/mediacentre/factsheets/fs094/en/index.html). Several Plasmodium species cause malaria, the most deadly being P. falciparum transmitted mainly by the Anopheles gambiae mosquito. As mosquito females require a blood meal to produce eggs, feeding on a malaria-infected host simultaneously activates oogenesis and triggers immune responses to malaria parasites. In the midgut, ingested Plasmodium gametocytes differentiate within minutes into gametes. After fertilization, zygotes rapidly transform into ookinetes, i.e. motile cells that traverse the midgut epithelium between 16 and 48 h post infection (hpi). Once they reach the hemolymph-bathed basal side of the midgut, ookinetes round up and transform into oocysts, protected capsules within which asexual multiplication of the parasite takes place. Previous studies have established that the ookinete is the parasite stage most vulnerable to the mosquito immune response [1],[2]. As a consequence of this response, most mosquito species efficiently eliminate all the invading ookinetes, thereby aborting the parasite cycle [3]. In a few parasite/mosquito combinations, up to 20% of ookinetes survive and the disease can be further transmitted. A number of mosquito humoral antiparasitic proteins have been characterized (reviewed in [4]). The molecularly best characterized and phenotypically most prominent defense pathway mediating the killing of Plasmodium berghei in A. gambiae involves a thioester-containing protein (TEP1) homologous to vertebrate complement factor C3 [2],[5],[6]. Depletion of TEP1 by RNA interference (RNAi) renders mosquitoes hypersusceptible to Plasmodium infections, resulting in abnormally high infection levels. Two leucine-rich repeat (LRR) proteins, LRIM1 and APL1C, act as TEP1 control proteins to stabilize the mature form of TEP1 in the hemolymph [7],[8] and show the same RNAi phenotype as TEP1 in P. berghei infections [9]–[12]. The depletion of either protein results in precocious deposition of TEP1 on self tissues and completely aborts its binding to the ookinetes [7]. Therefore, it appears that LRR proteins regulate maintenance of mature TEP1 in circulation; however, the factors that control TEP1 targeting to the parasite surface remain unknown.
Simultaneously to the midgut crossing by ookinetes, the physiology of the mosquito is profoundly modified by a blood meal in preparation for the laying of a clutch of eggs. Within 2 to 3 d after a blood meal, the massive ovary growth allows maturation of 50–150 oocytes, a process called vitellogenesis (reviewed in [13]). The blood meal provides the mosquito with amino acids and lipids that are transferred through midgut cells to the hemolymph and signal via the Target of Rapamycin (TOR) pathway to initiate massive synthesis of nutrient transport proteins in the mosquito fat body [14]. These transport proteins include the lipid transporter lipophorin (Lp, AGAP001826) (also known as apolipoprotein II/I or retinoic and fatty acid binding protein, RFABG/P) and vitellogenin (Vg, AGAP004203), a precursor of the yolk storage protein vitellin. Both proteins are secreted into the hemolymph and transported to the ovaries. Vg is a large phospholipoglycoprotein encoded in A. gambiae by a small family of nearly-identical genes. Insect Vg harbors potential sites for lipidation, glycosylation, and phosphorylation and is internalized by developing oocytes where it is proteolytically cleaved to generate vitellin, a nutrient source for the developing embryo (reviewed in [15],[16]). Lp, encoded by a single transcript and post-translationally cleaved, is composed of two subunits of 250 and 80 kDa that together scaffold a lipidic particle. Similar to vertebrate low- and high-density lipoproteins (LDL and HDL, respectively), mosquito Lp particles contain a core of fatty acids and sterols, surrounded by an outer leaflet of phospholipids [17],[18]. These particles function to deliver lipids and fatty acids to energy-consuming tissues, including rapidly growing imaginal discs in larvae, muscles, and the ovary in adult females [19]. In addition to lipids, Lp particles serve as a vehicle for morphogen proteins in the imaginal discs of Drosophila larvae [20]. Interestingly, human HDL has been shown to host a fraction of complement factor C3 [21] as well as trypanosome-killing protein complexes [22]. In mosquitoes, recent studies [23]–[25] have implicated Lp in both mosquito reproduction and Plasmodium survival. In particular, experimental depletion of Lp by RNAi inhibited oogenesis and also reduced the number of developing Plasmodium oocysts in the mosquito midgut [23]. This could point to a nutritional requirement for Lp in the early stages of parasite development. Indeed, Lp has recently been detected by in vitro approaches inside developing P. gallinaceum oocysts, suggesting that it provides parasites with a source of lipids [26]. An intriguing alternative explanation is that the increasing levels of Lp following a blood meal may negatively impact mosquito immunity against parasites. Artificially blocking the physiological rise in Lp levels would then allow the immune system to exert its full strength against the parasite.
In the mosquito fat body, two distinct pathways are required for optimal expression of proteins involved in vitellogenesis: (i) the nutrient-sensing TOR pathway and (ii) a hormonal cascade that oversees production of 20-hydroxyecdysone [14],[27],[28]. Furthermore, in Ae. aegypti mosquitoes infected with microbes and Plasmodium, the NF-κB factor REL1 positively regulates expression of Lp and its receptor [24], suggesting that the NF-κB pathway may also contribute to the regulation of oogenesis in addition to its known role in mosquito immunity [29]–[31]. However, our understanding of how oogenesis and immunity impact each other remains incomplete: on one hand depletion of Lp strongly inhibits development of P. gallinaceum; on the other hand over-expression of Lp resulting from the depletion of the REL1 inhibitor Cactus in Ae. aegypti is insufficient to rescue the complete block in parasite development [24].
Here, we investigated the role of the two major nutrient transport proteins Lp and Vg in mosquito antiparasitic responses using a common laboratory model of malaria transmission: A. gambiae mosquitoes infected with the GFP-expressing rodent parasite P. berghei [32]. We show that similarly to Lp, Vg depletion reduces parasite survival in mosquito tissues. Strikingly however, Lp and Vg are no longer required for parasite survival if TEP1 is depleted, suggesting that the low parasite survival phenotype associated with the Lp/Vg knockdowns requires TEP1 function. We propose that Lp and Vg exert distinct non-redundant roles in reproduction and immunity: Lp is crucial for oogenesis and is required for normal Vg expression after an infectious blood meal, whereas Vg contributes to oogenesis and negatively impacts TEP1 binding to the ookinetes. We suggest that the reported negative impact of Lp depletion on ookinete survival is indirect and is mediated by reduced levels of Vg. We further demonstrate that the NF-κB factors REL1 and REL2 limit the expression of Vg after an infectious blood meal. These results reveal an unexpected network of interactions whereby Plasmodium killing in mosquitoes is potentiated by NF-κB pathways at two levels: (i) activation of anti-Plasmodium genes and (ii) inhibition of the expression of the nutrient transport protein Vg.
Lp knockdown causes a decrease in parasite loads and simultaneously arrests oogenesis [23]. We examined whether the Lp knockdown phenotype requires the antiparasitic factor TEP1. To this end, we compared the numbers of surviving parasites in single TEP1 or Lp knockdown mosquitoes and in double TEP1/Lp knockdowns by injecting double-stranded RNA (dsRNA) resulting in RNAi. Four days after dsRNA injection, mosquitoes were fed on a mouse infected with GFP-expressing parasites. Mosquitoes were dissected 8 to 10 d later to gauge prevalence of infection and mean oocyst numbers per midgut (Figure 1A, Figure S3A). As reported earlier, Lp silencing strongly reduced the number of developing oocysts. Strikingly, silencing TEP1 at the same time as Lp annihilated the effect of Lp silencing, i.e. yielded the high oocyst numbers typically observed upon silencing of TEP1 alone. Therefore, the low oocyst counts observed in Lp-depleted mosquitoes are not due to a nutritional dependence of ookinetes on Lp-derived lipids but are a consequence of TEP1 activity. This result also suggests that the increased parasite killing in Lp-depleted mosquitoes takes place at the ookinete stage, since TEP1 binding does not kill oocysts. Further, these results imply that the loss of Lp renders ookinetes more vulnerable to TEP1-dependent killing.
To explain these data, we initially hypothesized that Lp particles might physically sequester components of the TEP1 machinery in an inactive state, but a search for Lp-associated immune factors was unsuccessful (with the notable exception of prophenoloxidase), suggesting that TEP1-containing complexes are not carried in the hemolymph by Lp particles (see Text S1 and Figure S1).
To investigate whether the adverse effect on immunity is a specific property of Lp or may be manifested as well by other nutrient transport factors, we injected mosquitoes with dsVg and compared parasite development with dsLacZ and dsTEP1-injected mosquitoes. A 4-fold reduction in mean parasite numbers was observed in the dsVg group compared to dsLacZ controls (p<0.001, p<0.001, p<0.05, and p<0.05 depending on the replicate of this experiment; Figure 1B and Figure S3B). This effect was more profound than the effect of dsLp (Figure 1A and 1E). We then examined whether depletion of the major yolk protein would compromise oogenesis. In contrast to Lp silencing, which resulted in total abortion of ovary development, roughly 50% of mosquito females still developed eggs after silencing of Vg compared to 80% in dsLacZ control mosquitoes (Figure 1C), though ovaries that did mature usually contained only a few eggs bearing melanotic spots (unpublished data). When given a chance to lay, Vg-silenced females did lay a few eggs, the majority of which never hatched (unpublished data). The difference in strength between the Lp and Vg silencing phenotypes regarding egg development suggests either that Lp is more crucial than Vg for egg development or that the efficiency of Lp silencing is greater than the efficiency of Vg silencing. Residual Vg protein may allow the development of a few eggs in dsVg-treated mosquitoes. It is interesting to note that the strengths of the silencing phenotypes are reversed when considering parasite survival. To verify the efficiency of RNAi-mediated depletion of Lp and Vg, we used specific antibodies directed against the large and small subunits of Lp, and against Vg. RNAi silencing caused Lp and Vg protein amounts to drop below 10% of control levels (Figure S2). Subsequently, we systematically controlled for Lp and Vg silencing efficiency and noted that Vg depletion was somewhat more variable than Lp depletion, residual Vg protein sometimes approaching 20% of control levels (unpublished data). Strikingly, this analysis revealed that the major protein bands detected in hemolymph samples by Coomassie staining of SDS-PAGE gels (or of PVDF membranes after protein transfer) correspond to the Vg and Lp signals detected by specific antibodies (Figure S2). We excised these easily visualized bands from Coomassie-stained protein gels and submitted them to MALDI mass spectrometry. The peptide mass spectra were searched against the NCBInr database. Each band from a triplet running between 160 and 200 kDa was unequivocally identified as Vg, and the bands running at ∼250 and 80 kDa were unequivocally identified as the large and small subunits of Lp, respectively. In addition, a protein running at ∼70 kD and showing an expression pattern identical to that of the ∼200 kD Vg band (including after RNAi silencing) was identified as the N-terminal fragment of the polypeptide encoded by Vg mRNA (visible in Figures 3C, 4C, and S1). This fragment was not recognized by our Vg antibody, raised against a C-terminal Vg fragment. Its existence is consistent with the cleavage of Ae. aegypti Vg prior to secretion [33]–[35]. No contaminating proteins were detected at these sizes in the mass spectrometry analysis. Therefore, Lp and Vg proteins can be readily visualized after hemolymph electrophoresis and Coomassie staining of SDS-PAGE gels even without immunoblotting. The efficiency of TEP1 silencing was also confirmed by immunoblotting (Figure 1D).
We next investigated whether Vg and Lp cooperate to sustain oogenesis and parasite development or are involved in independent processes. We performed double-knockdown experiments by simultaneously injecting dsVg-dsLp to compare to single injections of dsVg and dsLp as controls. As expected, dsLp completely blocked oogenesis and the same was observed in concomitant dsLp-dsVg knockdowns (Figure 1F). Moreover, single dsVg (p = 0.0001) and double dsLp-dsVg (p<0.0001) knockdowns caused comparable reductions in oocyst counts; these reductions in oocyst numbers were stronger than in the single dsLp knockdown (p = 0.024) (Figure 1E). These results suggest that the influences of Lp and Vg on reproduction and immunity are balanced differently. Lp may be more crucial for oogenesis than Vg, whereas Vg influences Plasmodium survival more strongly than does Lp. In most experiments, the effect of Vg and Lp knockdowns on parasite counts did not appear to be additive (Figures 1E, 2A, and unpublished data). Although this observation is not supported by strong statistical significance, it raises the possibility that the two proteins may be involved in a single process benefiting ookinete survival in the physiological situation.
To determine whether similarly to Lp, the effect of Vg on parasite development required TEP1 function, we performed triple knockdown experiments by injecting combinations of dsTEP1, dsVg, dsLp, or control dsLacZ. Again, total inhibition of oogenesis was observed in all dsRNA combinations that included dsLp, suggesting that oogenesis is not influenced by TEP1 function but absolutely requires Lp (Figure 2B). In striking contrast, high parasite loads similar to that detected in the dsTEP1 single knockdown were obtained when TEP1 was depleted simultaneously to Vg (unpublished data) or to both Vg and Lp (Figure 2A, Figure S3C). These findings imply that blocking the transport of lipids and vitellogenin-derived nutrients does not limit parasite survival when the immune defense is suppressed; instead, the observed reduction in parasite numbers in dsLp and dsVg knockdowns is dependent on TEP1. We conclude that TEP1-dependent parasite killing is more efficient when Lp and/or Vg levels are low and that the TEP1-mediated immune pressure exerted by the vector is a bigger impediment to the establishment of a Plasmodium infection than nutrient availability. If this constraint is removed via TEP1 depletion, Plasmodium parasites can effectively exploit even reduced vector resources and proceed with the formation of viable oocysts.
We next examined at which level Vg and Lp genetically interact with TEP1. Binding of mature TEP1 to the parasite surface is one of the first steps leading to parasite killing; either increasing or reducing this event greatly influences the outcome of infection [31],[36]. Therefore, we gauged the efficiency of TEP1 binding to ookinetes in dsLp- and dsLp-Vg-injected mosquitoes. At early time points (24 hpi) TEP1 binding to ookinetes did not differ in the Lp or Lp-Vg -depleted versus control mosquitoes; but at 48 hpi 70% to 86% of ookinetes were TEP1 positive (i.e., either dead or moribund) in dsLp- or dsVg-Lp-injected mosquitoes versus only 41% to 68% in dsLacZ controls (Figure 2C and Table S1, p = 0.005 or less by chi-square analysis). Thus, TEP1 binding to parasites is more efficient in the absence of Lp/Vg. This strongly suggests that physiological levels of Vg and Lp interfere with the efficient binding of TEP1 to ookinetes once the invasion phase is completed.
To see if we could also detect an effect of Lp and Vg depletion at a later stage of parasite development, we examined oocyst growth. Strikingly, oocyst size 9 d after infection was markedly reduced when Lp, but not Vg, was depleted (Figure 2D). In contrast to oocyst numbers, silencing TEP1 at the same time as Lp did not rescue oocyst growth (unpublished data), indicating that the small oocyst size does not result from TEP1 activity in Lp-deficient mosquitoes. This supports the hypothesis that Lp contributes nutrients to oocyst development [26]. Therefore, Lp benefits Plasmodium development at two independent levels: an early effect favoring ookinete survival by protecting against TEP1-dependent killing, and a later effect favoring normal oocyst growth. The latter effect does not require Vg or TEP1 function.
Previous work [7],[31] has demonstrated that boosting mosquito basal immunity via depletion of the inhibitory IκB protein Cactus up-regulates components of the TEP1 pathway (including TEP1, LRIM1, and APL1C) and completely blocks parasite development. Therefore, we asked whether the knockdown of Vg and Lp could mimic the effect of Cactus depletion and elevate TEP1 expression levels, providing an explanation to the above observations. We silenced Lp and/or Vg and examined the transcript levels of TEP1 before and after blood feeding using quantitative real-time polymerase chain reaction (qRT-PCR). Silencing of the two nutrient transport genes did not alter TEP1 expression (Figure 3A). We then evaluated the effect of Lp and Vg silencing on TEP1 protein amounts and TEP1 cleavage in the hemolymph by immunoblotting using polyclonal anti-TEP1 antibodies. This analysis did not reveal any marked increase in the amounts of full-length or mature TEP1 protein (Figures 3C and 1D).
Surprisingly, silencing of Lp reproducibly lowered the expression of Vg mRNA (Figure 3B and unpublished data). At the protein level, Lp depletion strongly reduced Vg levels at 47 h (but not 24 h) post-infectious feeding compared with the controls (Figure 3C), confirming that Lp is indeed required for full Vg expression between day 1 and day 2 post-infectious blood-feeding. In contrast, the depletion of Vg had no effect on Lp expression (Figure 3B) or protein levels (Figure 3C).
The unexpected observation that Lp and Vg knockdown simultaneously arrests oogenesis and facilitates TEP1 binding to ookinetes led us to re-examine the previously observed striking phenotype of dsCactus, which boosts basal immunity while arresting oogenesis ([31] and unpublished data). Depleting the IκB-like repressor protein Cactus increases the activity of NF-κB factors REL1 and REL2, leading to elevated expression of TEP1 and other immune factors. Therefore, we investigated whether REL1, REL2, and Cactus influence the expression of Vg and/or Lp. To this end, mosquitoes were injected with either dsRel1, dsRel2, dsCactus, or co-injected with dsRel1-dsRel2, dsRel1-dsCactus, dsRel2-dsCactus, and dsLacZ control. Mosquitoes were fed on an infected mouse, and subsequently, the expression of Vg and Lp was monitored by qRT-PCR. Strikingly, Vg expression was almost abolished in dsCactus mosquitoes at 24 hpi; conversely, the depletion of REL1 or REL2 at this time point elevated Vg expression above the levels in the dsLacZ control (Figure 4A). Interestingly, concomitant silencing of Cactus/Rel1 and Cactus/Rel2 restored Vg expression to physiological levels (Figure 4B), indicating that REL1 and REL2 contribute to the regulation of Vg expression. At the protein level, Vg amounts were unchanged at 24 h but strongly reduced 43 h after infectious blood feeding specifically in dsCactus-injected-mosquitoes (Figure 4C), confirming the qPCR data and revealing a clear delay between mRNA and protein fluctuations. Thus, in the dsCactus background, while TEP1 expression is upregulated, Vg expression is directly or indirectly repressed by REL1/2. Therefore, the Cactus protein affects TEP1 and Vg levels in opposite directions. We extended our analysis to Lp, but in contrast to the situation reported for Ae. aegypti [24], its expression was unaffected by the knockdown of the NF-κB-like factors (Figure 4A). Since Vg silencing alone, unlike Cactus silencing, is not sufficient to completely block oogenesis, other molecules required by developing mosquito oocytes may be regulated by Cactus in the same manner as Vg.
Taken together, our findings uncover the complex phenotype of Cactus depletion. It leads to a lower level of Vg expression after a blood meal, thereby contributing to the arrest in oogenesis seen in Cactus knockdown mosquitoes. On the other hand, it stimulates the mosquito antiparasitic defense at least at two different levels: (i) by lowering the level of Vg, rendering TEP1-mediated killing more efficient, and (ii) by elevating the levels of TEP1 pathway proteins.
The first indication that nutrient transport after a blood meal influences mosquito susceptibility to P. berghei was provided by Vlachou et al. [23], who demonstrated that experimental depletion of the lipid carrier protein Lp by RNAi reduces the number of developing oocysts in the mosquito midgut. Recently, these results were extended to P. falciparum [25]. However, how and at which stage of development the parasites were eliminated in Lp-deficient mosquitoes remained to be determined. We show here that the major yolk protein Vg shows a similar but more drastic knockdown phenotype than Lp on Plasmodium survival and that the Lp and Vg depletion phenotypes require the function of the immune factor TEP1, which targets ookinetes for killing. Further, high numbers of parasites actually survive and turn into oocysts even in the context of Lp and/or Vg depletion, as long as TEP1 is also experimentally depleted. From these observations, we infer that physiological levels of both nutrient transport proteins following a blood meal somehow dampen the strength of the immune defense and protect ookinetes against destruction by the TEP1 pathway. The effects of Lp and Vg depletion on TEP1-mediated parasite killing are similar, and we find that Lp is required for the full induction of Vg expression on day 2 following an infectious blood meal. We therefore propose that Lp may indirectly affect ookinete survival by influencing Vg expression, while Vg impinges either directly or more closely than Lp on the TEP1-killing mechanism.
The induction of Vg expression after a blood meal requires both the TOR pathway and ecdysone signaling [14]. It is unclear why Lp depletion reduces the expression of Vg after an infectious blood meal. One possible explanation is that an Lp shortage precludes ovarian follicle development, preventing the normal secretion of ecdysone by follicle cells; thus leading to the reduction in Vg expression. However, attempts to rescue the Lp silencing effect on Vg expression with exogenously provided 20-hydroxyecdysone were unsuccessful. As the lower level of Vg expression in Lp-deficient A. gambiae is reminiscent of the situation observed in adult Ae. aegypti mosquitoes malnourished during larval life [37], it would be interesting to determine if Plasmodium survival is compromised in such malnourished mosquitoes in laboratory and field settings.
The GFP-tagged P. berghei strain used in this study provides a good model and enables analyses of vectorial capacity that are much more demanding with wild malaria parasites. However, recent studies indicate that the mosquito response to P. berghei and to P. falciparum differ in important ways [10],[38]. In addition, the P. berghei–A. gambiae model is an unnatural host-parasite association. Therefore, it will be important to see whether our observations hold true in the A. gambiae–P. falciparum relationship. Importantly though, the TEP1 pathway does limit P. falciparum survival in A. gambiae natural infections ([39] and Levashina et al., unpublished results) and the Lp knockdown was shown to have similar effects in both systems [25].
What is the molecular basis of the negative effect of the two nutrient transport proteins on the TEP1 pathway? We initially hypothesized that Lp-scaffolded lipidic particles could sequester components of the TEP1 pathway in an inactive state. However, TEP1 and its interacting partners LRIM1 and APL1C were not detectable in Lp extracts, suggesting that the Plasmodium-killing machinery is not carried by Lp particles. Instead, RNAi-mediated depletion of Lp and, more strikingly, of Vg resulted in more efficient TEP1 binding to the surface of ookinetes at 48 hpi, promoting their killing. One explanation could be that Vg (and perhaps Lp, to a lesser extent) are recruited to the parasite surface, where they might mask TEP1 binding sites. Consistent with this idea, fish vitellogenin has recently been found to bind microorganisms and to opsonize them for phagocytosis [40]. Mosquito Vg may behave non-productively in a similar manner and outcompete TEP1 from the ookinete surface. Alternatively, a physical interaction between TEP1 and Vg could inhibit TEP1 activity, a hypothesis that should be further investigated. Yet another possible explanation is that transient interactions of ookinetes with Vg might alter the lipid composition in the ookinetes' membrane, rendering them less visible to the TEP1 machinery. The parasite molecules to which TEP1 covalently attaches are currently unknown, but hydroxyl residues on surface lipids could be good targets for thioester-dependent TEP1 covalent binding.
We further observed a retarded oocyst growth in Lp-deficient mosquitoes 9 d post infection. This phenotype was specific to Lp, as parasites developed normally in Vg-deficient mosquitoes. Therefore, Lp is a probable lipid source for developing oocysts. Indeed, Lp was detected inside P. gallinaceum oocysts in vitro, suggesting that oocysts tap some of the host's Lp for their development [26]. Taken together, Lp appears to regulate parasite development at two distinct stages by two independent mechanisms: (i) providing an indirect protection to ookinetes via regulation of Vg levels after a blood meal and thereby dampening TEP1 binding to ookinetes, and (ii) exerting a direct nutritional role by supplying lipids to growing oocysts.
The quantitative RT-PCR and protein expression results reported here added the IκB/NF-κB-like factors Cactus/REL1 and REL2, previously known to control immunity [29]–[31], to the list of factors that influence Vg expression. We propose that Cactus depletion boosts TEP1 parasite killing by simultaneously increasing TEP1 expression [31] and decreasing the expression of Vg, in the absence of which TEP1-mediated killing is more efficient. Previously, the reason why Cactus depletion blocked oogenesis while boosting anti-Plasmodium immunity was unknown. Our results shed new light on this phenomenon by suggesting that Cactus activity is necessary for the expression of Vg, and probably of additional factors involved in vitellogenesis.
Although many mosquito genes showing antiparasitic activity are induced by the NF-κB-like factors REL1 and REL2 [12],[29]–[31],[41], it is currently unclear whether parasite invasion of mosquito tissues actually activates the NF-κB pathways. However, the expression of nutrient transport molecules is affected by signals arising from the parasite's invasion, in addition to being influenced by hormone signaling, the TOR pathway, and NF-κB factors. Indeed, ookinete invasion of the midgut induces Lp mRNA expression further than does an uninfected blood meal in A. gambiae and Ae. aegypti [23],[24]. At the protein level, we did not observe a corresponding increase in Lp amounts using specific antibodies (unpublished data), which may reflect consumption of the additionally produced Lp by parasites and/or by the midgut wound healing response to parasite invasion. This implies that Lp protein homeostasis is under tight physiological regulation. Conversely, Ahmed et al. [42] reported that parasite invasion reduces the abundance of the Vg transcript in A. gambiae, while Vg protein levels were only transiently reduced before accumulating in the hemolymph. Therefore, the production of both proteins is subjected to multiple physiological switches. The reported changes in Vg levels correlated with apoptosis of patches of ovarian follicular cells, which was prominent following infections and immune stimulation. Dying ovarian follicles stop secreting ecdysteroids and taking up Vg protein, which may explain both the drop in Vg transcription and the accumulation of Vg protein in the hemolymph [43],[44]. It would be interesting to identify infection-dependent signals arising at the midgut and triggering ovarian follicle apoptosis. In Drosophila, pathogenesis is also reported to trigger cell death in ovaries [45]. In the presence or absence of an infection, activation of the Immune deficiency (Imd) pathway (e.g., by injection of dead bacteria) negatively impacted oogenesis. This effect depended on the immune status, as oogenesis remained normal in Imd pathway mutants injected with dead bacteria [46]. The mosquito Cactus/REL1/REL2 NF-κB pathway is related to the Drosophila Toll and Imd immune pathways; its targets would therefore represent attractive candidates as modulators of mosquito reproduction. A full understanding of the interactions between reproductive and immune functions in mosquitoes will require a thorough study of the molecular pathways influencing the transcription of immune and vitellogenic factors, and how these pathways are affected by blood meals, immune defense, and parasite invasion. To our knowledge, Vg and Cactus are the first molecules reported to occupy a central position at the interface between reproduction and immunity, providing a molecular handle to further explore the long-suspected trade-off between these two processes.
Approximately 0.5 g of mosquito adults (ca. 330 mosquitoes) were roughly ground with a Polytron electric homogenizer in 2 ml ice-cold TNE buffer (100 mM Tris-HCl pH 7.5, 0.2 mM EGTA, 150 mM NaCl) + Complete protease inhibitors (Roche). Debris were centrifuged at 4°C in a tabletop centrifuge. The supernatant was transferred to 2.2 ml ultracentrifuge tubes and spun for 3 h at 120,000 g at 4°C in a Sorvall ultracentrifuge equipped with an S55-S rotor. The cleared supernatant was recovered, completed with solid potassium bromide to a final concentration of 0.34 g/ml, overlayed with 0.5 ml TNE buffer+0.33 g/ml KBr, and centrifuged in 2.2 ml PET ultracentrifuge tubes (Hitachi Koki) at 250,000 g, 10°C, for at least 36 h. The top layer of fat was discarded and 5 or 6 fractions of 0.5 ml were carefully collected starting from the top. Lipophorin particles were present in the top fraction, while the majority of other proteins fractionated into the fourth.
The top fraction of a potassium bromide gradient prepared using a scale-up of the above method was desalted on a Pharmacia PD-10 column according to the manufacturer's instructions. The two subunits of Lp were the predominant proteins in the extract according to Coomassie staining of an SDS-PAGE gel. Protein amount was quantified with a Bradford assay. Six-week-old female BALB/c mice were injected intraperitoneally with 40 µg of these lipophorin particles and 100 µg of poly I/C as adjuvant. Three injections were performed at 2-wk intervals. Four days prior to hybridoma fusion, mice with positively reacting sera were reinjected. Spleen cells were fused with Sp2/0.Agl4 myeloma cells as described [47]. Hybridoma culture supernatants were tested at day 10 by ELISA for cross-reaction with purified Lp particles. Positive supernatants were then tested by Western blot on mosquito extracts. All ELISA-positive supernatants recognized peptides corresponding in size to either the large (250 kDa) or the small (80 kDa) Lp subunit. Specific cultures were cloned twice on soft agar. A hybridoma clone (2H5, immunoglobulin subclass IgG2aκ) recognizing the 80 kDa Lp subunit was selected and ascites fluid was prepared by injection of 2×106 hybridoma cells into pristane-primed BALB/c mice. The resulting antibody efficiently immuno-precipitated the 80 kDa Lp subunit and co-immunoprecipated the 250 kDa subunit. The identity of both immuno-precipitated subunits, excised from Coomassie-stained protein gels, was confirmed by mass spectrometry. Similarly, we prepared a monoclonal antibody (2C6) recognizing the large Lp subunit. Rabbit polyclonal antibodies specific to Vg were obtained by immunizing rabbits with a purified recombinant Vg fragment fused to GST. The Vg gene fragment used for protein production was amplified from mosquito cDNA using attB-site (capital letters)-containing primers GGGGACAAGTTTGTACAAAAAAGCAGGCTtcaagtttgtgctgcagcacaagcag and GGGGACCACTTTGTACAAGAAAGCTGGGTCCTAagcgcaagatggatggtagtttc. The PCR product was cloned into pDEST15 (Invitrogen) using the Gateway technology. Protein was produced in E. coli BL21-AI.
120 adult mosquitoes were severed by opening the thorax and abdomen cuticles with fine forceps and bled on ice in 1 ml IP buffer (TRIS pH 7.9 50mM, NaCl 100 mM, EDTA 2 mM, BSA 0.1 µg/ml) + Complete protease inhibitors (Roche). Carcasses and cellular debris were removed by two successive 2,500 g centrifugation steps (for 2 min at 4°C); the extract was further cleared by three 16,500 g centrifugations (2 min each). The sample was pre-cleared for 1 h at 4°C under gentle rocking with 2 µg of an irrelevant mouse IgG2aκ antibody that was removed by incubation at 4°C with 35 µl protein A-sepharose slurry (Pharmacia) for 1 h followed by centrifugation. Supernatant was split in two aliquots, one subjected to a 1 h incubation with specific antibody and the other with a non-specific antibody of the same immunoglobulin class. 35 µl of protein A-Sepharose were added to each sample, further rocked at 4°C for 1 h, centrifuged. The supernatant was saved (post-IP supernatant sample). Sepharose beads were washed 5×10 min in TE buffer with or without 500 mM KCl, successively. Lipophorin and associated proteins were eluted from the beads using SDS-PAGE sample buffer and submitted to Western blotting.
At least 8 anesthetized mosquitoes were aligned on ice under the binocular microscope. Their proboscis was clipped with dissection scissors. Each mosquito was gently pressed on the thorax with forceps and the hemolymph droplet forming at the tip of the cut proboscis was collected into 1× sample (Laemmli) buffer. An hemolymph amount equivalent to that collected from 4 mosquitoes was loaded in each lane of SDS-PAGE gels.
The 741 bp long HincII fragment of Vg1 (AGAP004203) and the 431 bp long BspHI/BsgI fragment of Lp (AGAP001826) were cloned from cDNA library clones into the pLL10 vector. RNAi constructs for TEP1 and NF-κB factors have been described (Frolet et al. 2006) [31]. Potential cross-silencing effects of the chosen sequences were analyzed using the Deqor software ([48]; http://deqor.mpi-cbg.de/) with the predicted A. gambiae transcriptome ENSEMBL database. DsRNA was synthesized as previously described [36]. A. gambiae susceptible G3 strain were maintained at 28°C, 75%–80% humidity, and a 12/12 h light/dark cycle. Two-day-emerged adult female mosquitoes from the same cohort were injected with 0.2 µg of dsRNA using a Nanoject II injector (Drummond, http://www.drummondsci.com). Co-injection experiments were performed by injecting a double volume of 1∶1 mixtures of 3 µg/µl solutions of dsRNAs. Four days after dsRNA injection mosquitoes were fed on a mouse carrying P. berghei GFP-con 259cl2 as previously described [36],[37]. Statistical significance was determined with a Kruskall-Wallis test for non-parametric data followed by Dunn's post-test. The indicated p values are those obtained with Dunn's test.
The ovaries of dissected females were observed under the binocular microscope. Ovaries containing 3 fully grown eggs or more were scored as positive. Ovaries with only undeveloped oocytes or less than 3 fully grown eggs were scored negative.
Total RNA from 10 mosquitoes was extracted with Trizol reagent (Invitrogen) before and after dsRNA injection or after blood feeding. 2–8 µg of RNA was reverse transcribed using M-MLV enzyme and random primers (Invitrogen). Specific primers (Table 1) were used at 300 nM for qRT-PCR reactions. Ribosomal protein L19 (RPL19) served as an internal control to normalize gene expression. The reactions were run on an Applied Biosystems 7500 Fast Real-Time PCR System using Power SYBR Green Mastermix (http://www.appliedbiosystems.com).
In order to count the surviving GFP-expressing parasites, mosquito midguts were dissected between 7 and 10 dpi and prepared as previously described [36],[37] and observed under a fluorescence microscope. To assess TEP1 binding to ookinetes, mosquito midguts were dissected at 18, 24, and 48 hpi, fixed in 4% formaldehyde at room temperature for 45 min, then washed with phosphate buffered saline, and stained with anti-TEP1 antibodies as previously described [31],[36]. Parasite numbers and TEP1 labeling were scored using a Zeiss fluorescence microscope (Axiovert 200M) equipped with a Zeiss Apotome module (http://www.zeiss.com). GFP-expressing parasites were considered live while dead parasites were GFP negative. Differential TEP1 staining on ookinete were gauged at 18, 24, and 48 hpi. At least three independent experiments were conducted per treatment group with a minimum of five mosquito midguts per treatment. For each midgut, all ookinetes visible in 4 fields covering most of the midgut were scored. Table S1 summarizes the ookinete counts from three independent experiments.
Coomassie-stained protein bands excised from SDS-PAGE gels were digested with trypsin. Tryptic peptides eluted from the gel slices were subjected to MALDI mass measurement on an Autoflex III Smartbeam (Bruker-Daltonik GmbH, Bremen, Germany) matrix-assisted laser desorption/ionization time-of-flight mass spectrometer (MALDI-TOF TOF) used in reflector positive mode. The resulting peptide mass fingerprinting data and peptide fragment fingerprinting data were combined by Biotools 3 software (Bruker Daltonik) and transferred to the search engine MASCOT (Matrix Science, London, UK). Peptide mass error was limited to 50 ppm. Proteins were identified by searching data against NCBI non-redundant protein sequence database.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.